Developments in Healthcare Information Systems and Technologies: Models and Methods

Developments in Healthcare Information Systems and Technologies: Models and Methods Joseph Tan McMaster University, Canada

Medical inforMation science reference Hershey • New York

Director of Editorial Content: Director of Book Publications: Acquisitions Editor: Development Editor: Publishing Assistant: Typesetter: Production Editor: Cover Design:

Kristin Klinger Julia Mosemann Lindsay Johnston Julia Mosemann Casey Conapitski Travis Gundrum Jamie Snavely Lisa Tosheff

Published in the United States of America by Medical Information Science Reference (an imprint of IGI Global) 701 E. Chocolate Avenue Hershey PA 17033 Tel: 717-533-8845 Fax: 717-533-8661 E-mail: [email protected] Web site: http://www.igi-global.com Copyright © 2011 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher. Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark. Library of Congress Cataloging-in-Publication Data Developments in healthcare information systems and technologies : models and methods / Joseph Tan, editor. p. ; cm. Includes bibliographical references and index. Summary: "This book presents the latest research in healthcare information systems design, development, and deployment,investigating topics such as clinical education, electronic medical records, clinical decision support systems, and IT adoption in healthcare"--Provided by publisher. ISBN 978-1-61692-002-9 (hardcover) -- ISBN 978-1-61692-003-6 (ebook) 1. Medical informatics. 2. Information storage and retrieval systems--Medicine. I. Tan, Joseph K. H. [DNLM: 1. Medical Informatics. W 26.5] R858.D48 2011 610.285--dc22 2010027162 British Cataloguing in Publication Data A Cataloguing in Publication record for this book is available from the British Library. All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the authors, but not necessarily of the publisher.

Table of Contents

Preface ................................................................................................................................................. xvi Chapter 1 Evaluating Health Information Services: A Patient Perspective Analysis ............................................. 1 Umit Topacan, Bogazici University, Turkey A. Nuri Basoglu, Bogazici University, Turkey Tugrul U. Daim, Portland State University, USA Chapter 2 Gastrointestinal Motility Online Educational Endeavor ...................................................................... 14 Shiu-Chung Au, State University of New York Upstate Medical University, USA Amar Gupta, University of Arizona, USA Chapter 3 Envisioning a National e-Medicine Network Architecture in a Developing Country: A Case Study ........................................................................................................................................ 35 Fikreyohannes Lemma, Addis Ababa University, Ethiopia Mieso K. Denko, University of Guelph, Canada Joseph K. Tan, Wayne State University, USA Samuel Kinde Kassegne, San Diego State University, USA Chapter 4 Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR: One Facility’s Approach ...................................................................................................................... 54 Karen A. Wager, Medical University of South Carolina, USA James S. Zoller, Medical University of South Carolina, USA David E. Soper, Medical University of South Carolina, USA James B. Smith, Medical University of South Carolina, USA John L. Waller, Medical University of South Carolina, USA Frank C. Clark, Medical University of South Carolina, USA

Chapter 5 Information Technology (IT) and the Healthcare Industry: A SWOT Analysis ................................. 65 Marilyn Helms, Dalton State College, USA Rita Moore, Dalton State College, USA Mohammad Ahmadi, University of Tennessee at Chattanooga, USA Chapter 6 Using a Neural Network to Predict Participation in a Maternity Care Coordination Program ............ 84 George E. Heilman, Winston-Salem State University, USA Monica Cain, Winston-Salem State University, USA Russell S. Morton, Winston-Salem State University, USA Chapter 7 Can IT Act as a Catalyst for Change in Hospitals? Some New Evidence ............................................ 94 Teemu Paavola, LifeIT Plc, Finland Chapter 8 Informatics Application Challenges for Managed Care Organizations: The Three Faces of Population Segmentation and a Proposed Classification System .................................................. 102 Stephan Kudyba, New Jersey Institute of Technology, USA Theodore L. Perry, Health Research Corporation, USA Jeffrey J. Rice, Independent Scholar, USA Chapter 9 Scrutinizing the Rule: Privacy Realization in HIPAA ....................................................................... 112 S. Al-Fedaghi, Kuwait University, Kuwait Chapter 10 In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care? A Comparison among Norway, Denmark, and Sweden .................................................................... 126 Agneta Ranerup, Göteborg University, Sweden Chapter 11 Characteristics of Good Clinical Educators from Medical Students’ Perspectives: A Qualitative Inquiry Using a Web-Based Survey System ................................................................ 145 Gary Sutkin, University of Pittsburgh School of Medicine, USA Hansel Burley, Texas Tech University, USA Ke Zhang, Wayne State University, USA Neetu Arora, Texas Tech University, USA Chapter 12 Open Source Software: A Key Component of E-Health in Developing Nations ............................... 162 David Parry, Auckland University of Technology, New Zealand Emma Parry, National Women’s Health, Auckland District Health Board, New Zealand Phurb Dorji, Jigme Dorji Wanchuck National Referral Hospital, Bhutan Peter Stone, University of Auckland, New Zealand

Chapter 13 An Empirical Investigation into the Adoption of Open Source Software in Hospitals ...................... 175 Gilberto Munoz-Cornejo, University of Maryland Baltimore County, USA Carolyn B. Seaman, University of Maryland Baltimore County, USA A. Güneş Koru, University of Maryland Baltimore County, USA Chapter 14 Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching ................. 195 Masoud Mohammadian, University of Canberra, Australia Ric Jentzsch, Compucat Research Pty Limited, Australia Chapter 15 Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care ......................................................................................................................................... 214 Jongtae Yu, Mississippi State University, USA Chengqi Guo, James Madison University, USA Mincheol Kim, Jeju National University, South Korea Chapter 16 E-Patients Empower Healthcare: Discovery of Adverse Events in Online Communities .................. 232 Roy Rada, University of Maryland Baltimore County, USA Chapter 17 Towards Process-of-Care Aware Emergency Department Information Systems: A Clustering Approach to Activity Views Elicitation ......................................................................... 241 Andrzej S. Ceglowski, Monash University, Australia Leonid Churilov, The University of Melbourne, Australia Chapter 18 Applying Dynamic Causal Mining in Health Service Management .................................................. 255 Yi Wang, Nottingham Trent University, UK Chapter 19 Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems: An Introduction and Literature Survey ............................................................................................... 275 Christos Vasilakis, University College London, UK Dorota Lecnzarowicz, University of Westminster, UK Chooi Lee, Kingston Hospital, UK Chapter 20 TreeWorks: Advances in Scalable Decision Trees .............................................................................. 288 Paul Harper, Cardiff University, UK Evandro Leite Jr., University of Southampton, UK

Chapter 21 Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor ........................................................................................................................ 302 Hussein Atoui, Université de Lyon and INSERM, France David Télisson, Université de Lyon and INSERM, France Jocelyne Fyan, Université de Lyon and INSERM, France Paul Rubel, Université de Lyon and INSERM, France Compilation of References .............................................................................................................. 312 About the Contributors ................................................................................................................... 343 Index ................................................................................................................................................... 348

Detailed Table of Contents

Preface ................................................................................................................................................. xvi Chapter 1 Evaluating Health Information Services: A Patient Perspective Analysis ............................................. 1 Umit Topacan, Bogazici University, Turkey A. Nuri Basoglu, Bogazici University, Turkey Tugrul U. Daim, Portland State University, USA The objective of the chapter is to explore the factors that affect users’ preferences in the health service selection process. In the study, 4 hypothetical health services were designed by randomly selecting levels of 16 attributes and these services was evaluated by the potential users. Analytical Hierarchy Process (AHP), one of the decision making methods, was used to assess and select the best alternative. Chapter 2 Gastrointestinal Motility Online Educational Endeavor ...................................................................... 14 Shiu-Chung Au, State University of New York Upstate Medical University, USA Amar Gupta, University of Arizona, USA Medical information has been traditionally maintained in books, journals, and specialty periodicals. A growing subset of patients and caregivers are now turning to diverse sources on the internet to retrieve healthcare related information. The next area of growth will be sites that serve specialty fields of medicine, characterized by high quality of data culled from scholarly publications and operated by eminent domain specialists. One such site being developed for the field of Gastrointestinal Motility provides authoritative and current information to a diverse user base that includes patients and student doctors. Gastrointestinal Motility Online leverages the strengths of online textbooks, which have a high degree of organization, in conjunction with the strengths of online journal collections, which are more comprehensive and focused. Gastrointestinal Motility Online also utilizes existing Web technologies such as Wiki-editing and Amazon-style commenting, to automatically assemble information from heterogeneous data sources.

Chapter 3 Envisioning a National e-Medicine Network Architecture in a Developing Country: A Case Study ........................................................................................................................................ 35 Fikreyohannes Lemma, Addis Ababa University, Ethiopia Mieso K. Denko, University of Guelph, Canada Joseph K. Tan, Wayne State University, USA Samuel Kinde Kassegne, San Diego State University, USA Poor infrastructures in developing countries such as Ethiopia and much of Sub-Saharan Africa have caused these nations to suffer from lack of efficient and effective delivery of basic and extended medical and healthcare services. Often, such limitation is further accompanied by low patient-doctor ratios, resulting in unwarranted rationing of services. Apparently, e-medicine awareness among both governmental policy makers and private health professionals is motivating the gradual adoption of technological innovations in these countries. It is argued, however, that there still is a gap between current e-medicine efforts in developing countries and the existing connectivity infrastructure leading to faulty, inefficient and expensive designs. The particular case of Ethiopia, one such developing country where e-medicine continues to carry significant promises, is investigated and reported in this article. Chapter 4 Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR: One Facility’s Approach ...................................................................................................................... 54 Karen A. Wager, Medical University of South Carolina, USA James S. Zoller, Medical University of South Carolina, USA David E. Soper, Medical University of South Carolina, USA James B. Smith, Medical University of South Carolina, USA John L. Waller, Medical University of South Carolina, USA Frank C. Clark, Medical University of South Carolina, USA Evaluating clinician satisfaction with an electronic medical record (EMR) system is an important dimension to overall acceptance and use, yet project managers often lack the time and resources to formally assess user satisfaction and solicit feedback. This article describes the methods used to assess clinician satisfaction with an EMR and identify opportunities for improving its use at a 300-physician academic practice setting. We administered an online survey to physicians and nurses; 244 (44%) responded. We compared physician and nurse mean ratings across 5 domains, and found physicians’ satisfactions scores were statistically lower than nurses in several areas (p<.001). Participants identify EMR benefits and limitations, and offered specific recommendations for improving EMR use at this facility. Methods used in this study may be particularly useful to other organizations seeking a practical approach to evaluating EMR satisfaction and use. Chapter 5 Information Technology (IT) and the Healthcare Industry: A SWOT Analysis ................................. 65 Marilyn Helms, Dalton State College, USA Rita Moore, Dalton State College, USA Mohammad Ahmadi, University of Tennessee at Chattanooga, USA

The healthcare industry is under pressure to improve patient safety, operate more efficiently, reduce medical errors, and provide secure access to timely information while controlling costs, protecting patient privacy, and complying with legal guidelines. Analysts, practitioners, patients and others have concerns for the industry. Using the popular strategic analysis tool of strengths, weaknesses, opportunities, and threats analysis (SWOT), facing the healthcare industry and its adoption of information technologies (IT) are presented. Internal strengths supporting further industry investment in IT include improved patient safety, greater operational efficiency, and current investments in IT infrastructure. Internal weaknesses, however, include a lack of information system integration, user resistance to new technologies and processes, and slow adoption of IT. External opportunities including increased use of the Internet, a favorable national environment, and a growing call for industry standards are pressured by threats of legal compliance, loss of patient trust, and high cost of IT. Chapter 6 Using a Neural Network to Predict Participation in a Maternity Care Coordination Program ............ 84 George E. Heilman, Winston-Salem State University, USA Monica Cain, Winston-Salem State University, USA Russell S. Morton, Winston-Salem State University, USA Researchers increasingly use Artificial Neural Networks (ANNs) to predict outcomes across a broad range of applications. They frequently find the predictive power of ANNs to be as good as or better than conventional discrete choice models. This paper demonstrates the use of an ANN to model a consumer’s choice to participate in North Carolina’s Maternity Care Coordination (MCC) program, a state sponsored voluntary public health service initiative. Maternal and infant Medicaid claims data and birth certificate data were collected for 59,999 births in North Carolina during the years 2000-2002. Part of this sample was used to train and test an ANN that predicts voluntary enrollment in MCC. When tested against a hold-out production sample, the ANN model correctly predicted 99.69% of those choosing to participant and 100% of those choosing not to participant in the MCC program. Chapter 7 Can IT Act as a Catalyst for Change in Hospitals? Some New Evidence ............................................ 94 Teemu Paavola, LifeIT Plc, Finland This chapter presents a succesful reorganization of a patient care process that was carried out in a middle sized Finnish hospital. The reorganization of the patient care process for joint replacement surgery succeeded in achieving a 50 per cent increase in operations. This study proposes that IT may have an indirect influence on the achievement of goals, such as productivity, as soon as the IT investment has been decided upon; in other words, IT benefits start accruing before the IT component is even in place. This is a new feature to add to the previous definitions, because this particular benefit cannot be logically derived from any of the features of the actual IT system. Paying enough attention to this phenomen at the planning stage can be vital to the success of new IT system investment.

Chapter 8 Informatics Application Challenges for Managed Care Organizations: The Three Faces of Population Segmentation and a Proposed Classification System .................................................. 102 Stephan Kudyba, New Jersey Institute of Technology, USA Theodore L. Perry, Health Research Corporation, USA Jeffrey J. Rice, Independent Scholar, USA Organizations across industry sectors continue to develop data resources and utilize analytic techniques to enhance efficiencies in their operations. One example of this is evident as Managed Care Organizations (MCOs) enhance their care and disease management initiatives through the utilization of population segmentation techniques. This article proposes a classification system for population segmentation techniques for care and disease management and provides an evaluation process for each. The three proposed operational areas for Managed Care Organizations are: 1) Risk Status: early identification of high-risk patients, 2) Treatment Status: compliance with treatment protocols, and 3) Health Status: severity of illness or episodes of care groupings, all of which require particular analytic methodologies to leverage data resources. By applying this classification system an MCO can improve its ability to clarify internal goals for population segmentation, more accurately apply existing analytic methodologies, and produce more appropriate solutions. Chapter 9 Scrutinizing the Rule: Privacy Realization in HIPAA ....................................................................... 112 S. Al-Fedaghi, Kuwait University, Kuwait Privacy policies, laws, and guidelines have been cultivated based on overly verbose specifications. This article claims that privacy regulations lend themselves to a firmer language based on a model of flow of personal identifiable information. The model specifies a limited number of situations and acts on personal identifiable information. As an application of the model, the model is applied to portions of the Privacy Rule of Health Insurance Portability and Accountability Act (HIPAA). Chapter 10 In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care? A Comparison among Norway, Denmark, and Sweden .................................................................... 126 Agneta Ranerup, Göteborg University, Sweden The aim of this article is to evaluate the provision of Web support in choice reforms in health care in Norway, Denmark, and Sweden. Two main issues are investigated: (1) What institutional frameworks for choice in health care exist, and how is the exercise of choice supported by Web technology in these countries? (2) As a consequence of this, what roles of the individual are mediated by this technology? The present study provides a critical analysis of current technologies for providing information about health care. It is concluded that in Norway the individual is equipped to be a reasonably informed consumer, customer, and citizen. A similar situation exists in Denmark, but here the consumer role is even more prominent. In Sweden, there has been little technological support for these roles, but recently national actors have initiated a project aimed at creating a national portal for public health care.

Chapter 11 Characteristics of Good Clinical Educators from Medical Students’ Perspectives: A Qualitative Inquiry Using a Web-Based Survey System ................................................................ 145 Gary Sutkin, University of Pittsburgh School of Medicine, USA Hansel Burley, Texas Tech University, USA Ke Zhang, Wayne State University, USA Neetu Arora, Texas Tech University, USA Medical educators have a unique role in teaching students how to save lives and give comfort during illness. This article reports a qualitative inquiry into medical students’ perspectives on the key qualities which differentiate excellent and poor clinical teachers, using a Web-based questionnaire with a purposeful sample of third- and fourth-year medical students. Thirty-seven medical students responded with 465 characteristics and supportive anecdotes. All participants’ responses were analyzed through reviewing, coding, member checking, recoding and content analysis, which yielded 12 codes. Responses from 5 randomly chosen participants were recoded by two authors with an inter-rater reliability coefficient of 0.72, implying agreement. Finally, 3 larger categories emerged from the data: Content Competence, Teaching Mechanics, and Teaching Dynamics. We incorporate these codes into a diagrammatic model of a good clinical teacher, discuss the relationships and interactions between the codes and categories, and suggest further areas of research. Chapter 12 Open Source Software: A Key Component of E-Health in Developing Nations ............................... 162 David Parry, Auckland University of Technology, New Zealand Emma Parry, National Women’s Health, Auckland District Health Board, New Zealand Phurb Dorji, Jigme Dorji Wanchuck National Referral Hospital, Bhutan Peter Stone, University of Auckland, New Zealand The global burden of disease falls most heavily on people in developing countries. Few resources for healthcare, geographical and infrastructure issues, lack of trained staff, language and cultural diversity and political instability all affect the ability of health providers to support effective and efficient healthcare. Health information systems are a key aspect of improving healthcare, but existing systems are often expensive and unsuitable. Open source software appears to be a promising avenue for quickly and cheaply introducing health information systems that are appropriate for developing nations. This paper describes some aspects of open-source e-health software that are particularly relevant to developing nations, issues and problems that may arise and suggests some future areas for research and action. Suggestions for critical success factors are included. Much of the discussion will be related to a case study of a training and E-health project, currently running in the Himalayan kingdom of Bhutan. Chapter 13 An Empirical Investigation into the Adoption of Open Source Software in Hospitals ...................... 175 Gilberto Munoz-Cornejo, University of Maryland Baltimore County, USA Carolyn B. Seaman, University of Maryland Baltimore County, USA A. Güneş Koru, University of Maryland Baltimore County, USA

Open source software (OSS) has gained considerable attention recently in healthcare. Yet, how and why OSS is being adopted within hospitals in particular remains a poorly understood issue. This research attempts to further this understanding. A mixed-method research approach was used to explore the extent of OSS adoption in hospitals as well as the factors facilitating and inhibiting adoption. The findings suggest a very limited adoption of OSS in hospitals. Hospitals tend to adopt general-purpose instead of domain-specific OSS. We found that software vendors are the critical factor facilitating the adoption of OSS in hospitals. Conversely, lack of in-house development as well as a perceived lack of security, quality, and accountability of OSS products were factors inhibiting adoption. An empirical model is presented to illustrate the factors facilitating and inhibiting the adoption of OSS in hospitals Chapter 14 Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching ................. 195 Masoud Mohammadian, University of Canberra, Australia Ric Jentzsch, Compucat Research Pty Limited, Australia Radio frequency identification (RFID) is a promising technology for improving services and reduction of cost in health care. Accurate almost real time data acquisition and analysis of patient data and the ability to update such a data is a way to improve patient’s care and reduce cost in health care systems. This article employs wireless radio frequency identification technology to acquire patient data and integrates wireless technology for fast data acquisition and transmission, while maintaining the security and privacy issues. An intelligent agent framework is proposed to assist in managing patients’ health care data in a hospital environment. A data classification method based on fuzzy logic is proposed and developed to improve the data security and privacy of data collected and propagated. Chapter 15 Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care ......................................................................................................................................... 214 Jongtae Yu, Mississippi State University, USA Chengqi Guo, James Madison University, USA Mincheol Kim, Jeju National University, South Korea In the advent of pervasive computing technologies, the ubiquitous healthcare information system, or Uhealth system, has emerged as an innovative avenue for many healthcare management issues. Drawing upon practices in healthcare industry and conceptual developments in information systems research, this paper aims to explain the latent relationships amongst user-oriented factors that lead to individual’s adoption of the new technology. Specifically, this study focuses on the introduction of chronic disease U-health system. Using the Ordinary Line Square (OLS) regression analysis, we are able to discover the insights concerning which constructs affect service subscriber’s behavioral intention of use. Based on the data collected from over 440 respondents, empirical evidences are presented to support that factors such as medical conditions, perceived need, consumer behavior, and effort expectancy significantly influence the formation of usage intention.

Chapter 16 E-Patients Empower Healthcare: Discovery of Adverse Events in Online Communities .................. 232 Roy Rada, University of Maryland Baltimore County, USA E-patients can empower themselves and improve healthcare. In online communities, patients may discuss adverse events that are inadequately addressed in the literature. The author as a patient joined various online patient discussion groups and identified several such adverse events. For each such adverse event, the patient findings, the medical literature, and the implications are noted. Extracts from the literature that were provided to the patients were welcomed by the patients. Possible approaches to financially supporting such activities are sketched. Chapter 17 Towards Process-of-Care Aware Emergency Department Information Systems: A Clustering Approach to Activity Views Elicitation ......................................................................... 241 Andrzej S. Ceglowski, Monash University, Australia Leonid Churilov, The University of Melbourne, Australia The critical role of emergency departments (EDs) as the first point of contact for ill and injured patients has presented significant challenges for the elicitation of detailed process models. Patient complexity has limited the ability of ED information systems (EDIS) in prediction of patient treatment and patient movement. This article formulates a novel approach to building EDIS Activity Views that paves the way for EDIS that can predict patient workflow. The resulting Activity View pertains to “what is being done,” rather than “what experts think is being done.” The approach is based on analysis of data that is routinely recorded during patient treatment. The practical significance of the proposed approach is clinically acceptable, verifiable, and statistically valid process-oriented clusters of ED activities that can be used for targeted process elicitation, thus informing the design of EDIS. Its theoretical significance is in providing the new “middle ground” between existing “soft” and “computational” process elicitation methods. Chapter 18 Applying Dynamic Causal Mining in Health Service Management .................................................. 255 Yi Wang, Nottingham Trent University, UK This article describes an application that illustrates the role of data mining technology in identifying hidden causal knoledge from health and medical data repositories. Across the health care and medical enterprises, a wide variety of data is being generated at a rapid rate. Current information technologies tends to focus on a more statical side of causal knowledge and do not address the dynamic causal knowledge. This article shows that the dynamic causal relation data can be captured for treatment, payment, operations purposes and administrative directed insights. Accessing this currently unrealized knowledge potential would enable the delivery of actionable knowledge to medical practitioners, healthcare system managers, policy planners and even patients to make a significant difference in overall healthcare.

Chapter 19 Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems: An Introduction and Literature Survey ............................................................................................... 275 Christos Vasilakis, University College London, UK Dorota Lecnzarowicz, University of Westminster, UK Chooi Lee, Kingston Hospital, UK The unified modelling language (UML) comprises a set of tools for documenting the analysis of a system. Although UML is generally used to describe and evaluate the functioning of complex systems, the extent of its application to the health care domain is unknown. The purpose of this article is to survey the literature on the application of UML tools to the analysis and modelling of health care systems. We first introduce four of the most common UML diagrammatic tools, namely use case, activity, state, and class diagrams. We use a simplified surgical care service as an example to illustrate the concepts and notation of each diagrammatic tool. We then present the results of the literature survey on the application of UML tools in health care. The survey revealed that although UML tools have been employed in modelling different aspects of health care systems, there is little systematic evidence of the benefits Chapter 20 TreeWorks: Advances in Scalable Decision Trees .............................................................................. 288 Paul Harper, Cardiff University, UK Evandro Leite Jr., University of Southampton, UK Decision trees are hierarchical, sequential classification structures that recursively partition the set of observations (data) and are used to represent rules underlying the observations. This article describes the development of TreeWorks, a tool that enhances existing decision tree theory and overcomes some of the common limitations such as scalability and the ability to handle large databases. We present a heuristic that allows TreeWorks to cope with observation sets that contain several distinct values of categorical data, as well as the ability to handle very large datasets by overcoming issues with computer main memory. Furthermore, our tool incorporates a number of useful features such as the ability to move data across terminal nodes, allowing for the construction of trees combining statistical accuracy with expert opinion. Finally, we discuss ways that decision trees can be combined with Operational Research health care models, for more effective and efficient planning and management of health care processes Chapter 21 Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor ........................................................................................................................ 302 Hussein Atoui, Université de Lyon and INSERM, France David Télisson, Université de Lyon and INSERM, France Jocelyne Fyan, Université de Lyon and INSERM, France Paul Rubel, Université de Lyon and INSERM, France Recent years have witnessed a growing interest in developing personalized and nonhospital based care systems to improve the management of cardiac care. The EPI-MEDICS project has designed an intel-

ligent, portable Personal ECG Monitor (PEM) embedding an advanced decision making system. We present two of the ambient intelligence models embedded in the PEM: the neural-network based ischemia detection module and the Bayesian-network risk stratification module. Ischemia detection was expanded to take into account the patient ECG, clinical data, and medical history. The neural-network ECG interpretation module and the Bayesian-network risk factors module collaborate through a fuzzylogic-based layer. We also present two telemedicine solutions that we have designed and in which the PEM is integrated. The first telemedical architecture was created to allow the collection of medical data and their transmission between healthcare providers to get an expert opinion. The second one is intended for improving healthcare in old people’s homes. Compilation of References .............................................................................................................. 312 About the Contributors ................................................................................................................... 343 Index ................................................................................................................................................... 348

xvi

Preface

A State-of-the-Art Review of Developments in Health Information System & Technology Models and Methods: The MEDIA Paradigm

IntroductIon For years, misguided health organizational information technology (IT) leadership, and the lack of available expertise and skills in the application of health IT (HIT) models and methods have somehow predisposed administrators of large-scale health maintenance organizations (HMOs) to become generally reluctant to migrate away from legacy health information systems (HISs). For many health provider organizations, the more recent decline in economic activities over the last few years has also further led to new debates on allocating limited public and private resources so gravely needed to build and sustain large-scale interoperable systems infrastructure in order to achieve such a system migration efficiently and effectively. Not surprisingly, progress towards adopting new health IT initiatives such as innovative e-technologies for the health care industry in the United States (US) has been slower than most other industries. Inadvertently, a major gap in the strategic and opportunistic use of emerging health IT models and methods now exists to significantly transform the apparently fragmented nature of US health services delivery system. Aside from the fear of costly systems failure, many key stakeholders in the US health care system have, admittedly, been resistant to invest in new and contemporary enterprisewide systems due to the lack of a well-focused national health IT vision and strategy. On the one hand, attempts to successfully diffuse interoperable, integrative HIT applications throughout the US health care system must now depend on the strength of the health IT leadership shouldered by the current Administration. On the other hand, progress in health IT implementation and diffusion will also depend on how quickly many of these health key stakeholder groups who have been technological laggards for one reason or another can be motivated, challenged, and appropriately enticed to design, develop, and deploy large-scale, complex and interoperable computer-based enterprisewide systems. Such enterprisewide systems include, but are not limited to, those that are designed to incorporate integrative health data management models, new biomedical informatic methods, web-based semantic search capabilities, and emerging clinical decision support methodologies. In the context of today’s complex and largely fragmented US health services

xvii

systems, specific applications of enterprisewide, interoperable systems can range from patient-centric records and information services systems such as electronic medical records (EMRs), electronic health records (EHRs), personal health records (PHRs), payor-based health records (PBHRs) and computer-based physician order entry (CPOE) systems, to health administrative-aided transaction and health information exchange (HIE) systems such as e-prescribing systems (EPS), supply chain management (SCM), customer relationship management (CRM), enterprise resource planning (ERP), and e-payment systems. Briefly, key reasons why the diffusion of interoperable, integrative HIT models and methods is needed in the US health care services sector entail: a. b.

c.

an urgency, as a whole, to contain escalating health care cost - the growing health care cost has become an increasingly unsustainable burden on US taxpayers over the years; the growing complexities and uncertainties in the health information processing and exchange environment of large, medium, and even smaller health organizations populating the US health care system - the expansion of stakeholder groups and increased federal, state, and municipal regulatory oversight mechanisms surrounding health services delivery have and will continue to add to the already intricate health information management (HIM) and services delivery system; and the potential of interoperable, integrative HIT models and methods to aid complex data analysis and semi-structured decision making - such analysis will not only empower care providers, but also enable critical information sharing to occur among referring physicians in consultation with patients, and especially when specialists and a team of caregivers are involved in key administrative and clinical decision making.

With increasing attention paid to the critical role that HIT models and methods can and will play in reforming the US health care system, we are finally seeing a trend in increased projections on health IT spending, which is currently anticipated to exceed 15 billion dollars annually for the US (Lipowicz, 2009). According to the National Coalition on Health Care, in 2008 alone, the US has spent well over 17% of its Gross Domestic Product (GDP) on health care - a percentage that clearly exceeded those spent by many other OCED countries (National Coalition on Health Care, 2009). Sadly, the fact that the US has outspent almost every other country on health care has not translated into better health or even more convenient, accessible, available or affordable health care services delivery for Americans. In fact, findings abstracted from 2007-2008 data in a study championed by the consumer health advocacy group Families USA has revealed that one out of every three Americans under 65 may still have to live, at one point in time or another, without health insurance coverage (Parisi and Bailey, 2009). Clearly, many of the issues raised here have now taken central stage in the debate raised by the Obama Administration for championing health care reform in the US. As well, various solutions have been considered, chief among them, using and adopting interoperable, integrative HIT applications. Such a strategy is not without merit, as various forms of health technologies have indeed risen over the years to similar challenges; for instance, online claims processing and e-prescribing have been successfully deployed by various HIT vendors to serve as tools to combat rapidly escalating health administrative costs while simultaneously leveraged to reduce wastes, increase efficiencies, eliminate redundancies, and improve the overall quality of clinical care and services. Raghupathi and Tan (2002, 2008) noted that various strategic applications of HIT models and methods can evidently improve the efficiency and effectiveness of US health care services delivery, adding value to existing, legacy-based HISs, and helping to integrate the islands of health services management

xviii

systems. Their arguments not only cited the power of e-technologies to streamline increasingly complex routine HIM processes that may require multi-provider, cross-organizational collaboration, but also the ability of interoperable, integrative HIT capabilities to augment enterprisewide efficiencies and care provider network connectivity. Accordingly, the key underlying argument for migrating from legacy HISs is that both data and processes linked to diverse functioning information systems must now be shared on an increasingly real-time basis between both on-site and off-site caregivers if we want to improve the quality of patient care. These interoperable, integrative systems can also streamline complex administrative and clinical workload, easing the communication needs among health care administrators, HIT personnel, health engineers, health informatic researchers, and HIT consultants, all of whom need to work closely together in today’s health services delivery systems if major systems bottlenecks are to be effectively managed over time. Figure 1 depicts the Model and Evidence Driven Integrated Analysis (MEDIA) paradigm, an integrated model management framework that is applied in this review to provide an integrative conceptualization of the evolving, state-of-the-art developments in HIT models and methods. Against this background, existing HIT models and gathered evidence will not simply aggregate but will be seen as complementing each other alongside the MEDIA integration and analysis process. Take the case of the influenza propagation - a disease model - and imagine how it can intermingle with a patient care process model, such as that of a primary clinic. Given the separate evidence about the flu outbreak and patient arrival pattern, running both models in parallel within the MEDIA framework produces either a more realistic health care supply-demand scenario in the context of a real-world community health setting, and/or offers insights into other related and potential issues needed to be addressed when challenged with health hazard preparation such as when an apparent discrepancy is observed to be operative within the overall system. All such information analysis and its integration can be further quantified and addressed consistently. Such multi-attribute analytic capability is crucial to the success of any future applications and developments of HIT models and methodologies. The rest of this review on health IT models and methods is organized as follows. In Section 2, a high-level systems perspective of HIS information and process flow in the context of applying HIT models and methods is outlined. Fundamentally, the classical information system input-process-output

Figure 1. The model and evidence integrated analysis (MEDIA) paradigm Model Integration Ontology modeling

Probabilistic dependency modeling

Hybrid computational analysis

Evidence Fusion Service Assurance a. Delivery improvement b. Risk management

xix

triad that underlies all systems engineering conceptualization is revisited. In Section 3, the focus will shift to the complexity-uncertainty challenge encountered in the current US health care sector. Following this, in Section 4, the emerging engineering approaches in health services applications at two levels are highlighted. These levels include: (1) the general health engineering level; and (2) the more focused health services modeling level. In Section 5, the MEDIA paradigm, including its three key facets are briefly overviewed: (a) domain ontology modeling and management that lays the foundation for model representation and infrastructural connectivity; (b) hybrid probabilistic modeling that computationally integrates systems and subsystems models; and (c) adaptive knowledge fusion to generate quality assurance that support active information collection and continuing operations management. Essentially, the MEDIA approach is based on a linked series of health engineering concepts, which deviates fundamentally from the traditional health care best practices in which quality is heavily dependent upon the hands-on skills of expert clinical practitioners as well as the deployment of progressively specialized medical instrumentation. The application of the MEDIA paradigm is also illustrated in a case study on risk management. Finally, in Section 6, we conclude this review by speculating on future research directions and practical implications in the field of health IT models and methods.

VIewIng HIt Models & MetHods In tHe classIcal HIs Flow cycle Model Owing to the need to deal with increasingly complex health data processing routines and the growing number of participating stakeholders who may need to share information and provide their particular interpretation of such data taken from the very same electronic health databases, an exciting playground for health researchers and practitioners now exists to incorporate efficient medical information processing models and emerging health decision support systems (HDSS) methodologies into interoperable systems linked to existing health databases, model management bases, and knowledge management bases. Despite recent investments in medical information packages such as the Vista medical information system and the GoogleHealth patient record management system, the lack of system interoperability has made it difficult, if not impossible, to integrate isolated health records stored at varying functional levels. Moreover, a key challenge is the integration of HIT models and methods through a knowledge fusion process so as to enhance the capability of viewing and analyzing any chosen set of discrete and continuous events in a health services delivery system such as relating patient-physician encounters and the progression of different disease stages from an enterprisewide and trajectory perspective, for example, the interactions and optimal interventional strategies for a particular patient cohort within a HMO, built up from its connected elements, including the affiliated hospitals, clinics, practicing physicians, nurses, information and personnel resources, labs, and departments linked to its health services delivery system. Figure 2 depicts a generic health information flow process model in which health administrative, clinical, and service delivery decision and policymaking must necessarily transpire at an enterprisewide level while aided with the use of relevant and applicable HIT models and methods within the traditionally defined input-process-output system triad.In the initial data collection, and information-knowledge gathering stage, a huge amount of raw data, guided protocols, and knowledge elements are often sourced from various input sensors and devices as well as recorded answers to questions asked of the individual patient and/or groups of patients, including underlying reasons for these patients to seek care. All of the

xx

Figure 2. A general health information flow process model with Its input-process-output system triad Data Collection

Intelligent Information Processing

Health IT Models & Methods Knowledge Formation & Feedback

Health Care Services Decisions & Policymaking

collected information is now ready to be pre-processed into some form of meaningful and intelligent datasets. These datasets are then stored appropriately either in a centrally located or in multiple electronic locations, typically via databases and data warehouses, waiting for further processing to serve a variety of clinical, research and administrative goals and purposes. Such purposes could range from clinical testing and follow up, to research, administrative billing, and/or mere reporting for managerial decision making. As a case in point, imagine the gathering of demographical data as well as specific vaccination records for a population of young adult patients to guide the clinical management of an ongoing epidemic, such as the “swine flu”. The data gathered should not only be accurately coded but they should also be securely stored in a more or less structured format to be communicated to the attending physicians for further clinical evaluations, testing, and/or diagnosis. In fact, the same information may be aggregated with other data for research analysis or it may be used to establish insurance co-payments to the various care provider groups. Finally, the information may also be critical to plan health vaccination programming by the different workplace or educational institutional settings tied in with the population of young adult patients. At the data manipulation-information mining-task processing stage, embedded patterns are often sought and hidden knowledge uncovered within the captured datasets to be translated intelligently into meaningful clusters and to provide additional advice and insightful guidance to the end-users, in this illustrative case, the various caregivers. In other words, the same data-information-knowledge collected of the young adult patients may now be further fused with other high-level information, such as expert clinical knowledge about the state of the patient immunization types and dates as well as the state of ongoing epidemics or pandemics, if any. The combined information can then be used to determine the health status of the individual young adults following the immunization, for example, their resistance to prevailing illnesses, their showing signs of some sort of common allergies, and/or their susceptibility to some other atypical reactions. In fact, it is here that the HIT models and methods may be most relevant to aid in the follow-up of crucial clinical and administrative decisions.

xxi

In operationally organizing, aggregating, mixing, dicing, and mining the stored patient data, various HIT models and methods may be combined in a complex fashion to generate a meaningful and “best” fit for clustering and/or grouping the stored data based on some known statistical patterns. The resulting data pattern classification are in turn routed back, on the one hand, to the clinicians to guide them in making prognostic, diagnostic and other therapeutic decisions, and, on the other hand, to the administrative staff to assist them in determining patient diagnostic codes, and thereby, generate accurate and appropriate third-party billings. At the end of the data processing-data mining stage, feedback will also typically occur in terms of second opinions, reviews and/or expert evaluations on model validity to further enhance systems finetuning and performance outcomes. Such enhancements recognize the possibility for more intelligent decision making in the face of missing or incomplete data and/or the fact that inadequate tools were available to appropriately process such data so that adjustments to these decisions could still be made to further improve accuracy or timeliness of key decisions and outcomes. In our case example, the nursing unit may, at this point, requests for feedback from the attending physicians as to who among these young adults would be due for specific and further vaccination visits, be sent for more clinical testing and evaluations, such as generating referrals to other departments or physician specialists, or they should simply be discharged. This commonly encountered health information flow cycle is characteristic of an open health information processing system that follows the input-process-output triad. Hence, despite the complexity of HIT models and methods, the potential for these methodologies to contribute significantly to future health services management is evident and apparent. While their applications represent mostly a part of the wider and far reaching field of HIS, it should not and cannot exist in and of itself. It must go hand in hand with the many different types of decisions on health services management that have to be executed routinely alongside the processing of initially gathered data, stored information, and captured knowledge. With a basic understanding on the flow of information and processes through a generic HIS flow cycle model, it is clear that multiple disciplines – including computer and library sciences, medical and/or health informatics, sub-fields in nursing informatics, teleradiology, telemedicine, telehome care, telematics, biomedical informatics, as well as domains of health IT, health engineering, and health operations research (OR) and management science, HDSS, and even the integration of cognitive sciences, information sciences, and health sciences – must all come together to enrich our ability in using HIT models and methods to meet a diversity and variety of clinical and health administrative decision needs. Ultimately, in order to achieve high quality health services management and related decision making, not only will we need medical informatics tools such as computers, clinical protocols, formal and informal medical terminologies, but also the integration of various informatic resources, tools, models, devices, techniques, methods, decision aids, and methodologies to optimize the collection, storage, retrieval, analysis, design, and use of health information in health services delivery (Tan and Payton, 2010).

ratIonale For IntegratIng HIt Models & MetHods: tHe us HealtH care coMplexIty-uncertaInty landscape While the interplay of multiple disciplines and sub-disciplines in managing a growing body of health services information and processes makes the health care system one of the most complex systems to study, the growing intricacies of the US health services sector may also be seen in many real-world events.

xxii

The 2007 drug-resistant tuberculosis case, for example, shows the high degree of interdependency that exists in the US health care infrastructure, and uncovers the many hidden flaws, previously ignored, in the management of patients, hospitals, the center for disease control and prevention (CDC), as well as the complex sociopolitical forces that link the US government with other national governments (U.S. House of Representatives, 2007)

Key challenges & risk Management in the us Health services sector For the US health care sector to achieve greater accessibility, affordability, and accountability, two key determinants for its continuing growth and future developmental success include: (1) ensuring a 24/7 service availability; and, (2) having a focus on service quality. In the coming years, as the US health care sector experiences sweeping reform, daunting risk and unpredicted challenges will likely arise. Hence, dealing with risk and uncertainty is a critical skill in running any modern-day health services enterprise, and this is precisely where HIT models and methods can and will play a central role. The major threats and challenges facing the US health care sector are depicted in Figure 3. As shown, these challenges include systems complexity, event uncertainty, and the need to share information. Owing to the fact that such threats and challenges will often emerge unpredictably and their diverse manifestations will also typically blur the line of most standard classifications, both health administrators and clinicians will want specialized tools to aid them in their routine policymaking and decision execution, as we moved forward with the US health care reform. First, with regard to systems complexity, there is the overwhelming intricacy of diverse evolving structures alongside the rapidly expanding health services supply networks. With hospital expansions and alliances formed among various stakeholders, increasing complexities have also followed. Again, with stakeholders trying to leverage on increasingly competitive global health supply chains, the resulting system relations are even more complex. Moreover, stakeholder collaboration across diverse industrial sectors may often follow different policy paths and practices. These factors, and the fact that most such collaboration is developed independently, compound the intricacy. Hence, there is the need for integrated HIT models and methods to guide future decisions. Second, in regard to event uncertainty, there is the need for systems configuration flexibility. A health service is subject to risk whenever and wherever the stakeholder cannot predict changes to the typical Figure 3. Major threats & challenges facing the US health care sector industry

• • • •

Complexity Healthcare alliance; Global supply chains; Regional, national, and international coordination; …

Staff

• • • •

Challenges Uncertainty New, rare, and epidemic diseases; Natural disaster and man-made emergency; Terrorist attacks; …

• • • • • •

Information Sharing Incompatibility; Inoperability; Integrity and accuracy; Cyber attack; Privacy regulation; …

Service Availability and Quality Resources Information/Knowledge

xxiii

environmental situation. As event uncertainty can arise, without prior warning, from diverse occurrences, such as (a) variations in staff and patient flow, (b) the onset of new, rare or epidemic diseases, and (c) emergencies, whether natural and/or man-made, all such uncertain events will require appropriate and timely adaptation in health services management. These challenges are further convoluted by evolving structures, novel technologies, the constant threats of natural disasters and/or man-made events, including the unpredictable growing number of participating stakeholder groups. Traditionally, stakeholders including patients, care providers, insurance companies, and the government do not see eye-to-eye. As many health administrators and clinicians tend to focus largely on the details, and likely also to stubbornly stick to their personal heuristic business decision-making biases, challenges of prevailing complexity and uncertainty will become even more evident over time. In light of this, it is not uncommon to have conflicting views shared among the multiple stakeholders. Moreover, health care debates often hinge on highly sensitive issues, and it is unlikely that any lopsided discussions or outcomes will contribute to the overall satisfaction of certain key stakeholder groups. Therefore, given that current management practices rely largely on personal heuristics, qualitative comparisons, and subjective guidelines, without precise accounting and science-based quantitative analysis, the significance of HIT models and methods cannot be overly emphasized. As today’s health services delivery systems often encompass huge numbers of interacting elements, significant interdependency, and enormous uncertainty and risk, it is very difficult to expect quality and outcomes of the decisions to be appropriately calibrated and/or justified quantitatively. Hence, the attempts to incorporate the latest science and methodological advances in HIT models are some of the ways to ensure better health decision outcomes and policymaking. Third, in regard to information sharing, it is in the interest of health provider organizations, at the minimum, to ensure that their patients receive prompt and accurate medical care services by increasing the accessibility and availability of health information services on the one side, and ensuring the secured, confidential, and private storage of personal medical records for sharing among affiliated caregivers on the other side. Technologically speaking, effective information sharing has to do chiefly with systems interoperability and the availability of an interoperable enterprisewide infrastructure. This is why a move towards integrative, interoperable HIT models and methods is key to achieving US health care reform. In summary, the challenges of complexities, uncertainties, and sharing of health data have deterred key stakeholders within the US health care industry sector to work collaboratively, effectively, and productively. But it has also highlighted the importance of having a national health IT strategy and vision as well as building a cumulative effort to apply and diffuse integrative, interoperable HIT models and methods.

Inadequacy of Health services Modeling practice & research Despite the wide range of quantitative models that has emerged to address various performance optimization risks with regards to different parts and/or aspects of health services systems over the years, there is still the need for a system-wide perspective on how these different HIT models and methods may be combined and integrated to aid health care decisions and policymaking. Roberts (2007), for instance, emphasized models to represent the progression of certain disease categories as well as to predict best treatment timing and cost. Denton, Fowler, Schaefer, Batun, Erdogan, and Gul (2009) detailed scheduling systems that model patient arrival and how they may be queued efficiently and effectively to be served by doctors in operating rooms. Finally, Shehad, Bertino, and

xxiv

Ghafoor (2005) discussed the application of computational models to quantify information leakage and availability based on the analysis of user communication via a computer infrastructure. Notwithstanding, each of these isolated attempts represented a series of uncoordinated efforts to address only a small, specific set of health services elements, without adjusting to benefit the greater whole – that is, the limited impact of the individual studies on the total system do not account for the intricate interdependencies among the systems elements. In other words, confined within a limited scope of a system component or sub-system, the performance measures adopted by these studies are typically defined in terms of cost, quality, time, or other specific but narrowly defined criterion, which, without exception, would still impact, one way or another, on other systems elements and components on an unknown basis and scale. Thus, isolated models and methodologies, working in and by themselves, simply cannot address the larger challenges presented in the US health services management system. With health engineering as an emergent field, researchers interested in HIT models and methods are beginning to redefine the boundary of traditional health care management philosophy. The modeling capability to guide flexible reconfiguration of health care structures, such as new investment and/ or rearrangement of provider organizations and facilities, is critical if we are to handle the complexity of systems modeling and analysis appropriately. Arguably, the different levels of local, regional, and national coordination that are actually interconnected (Schnase & Cunnius, 1995) will dictate how the national health infrastructure is to be reconfigured to manage existing resources efficiently, effectively and productively. All in all, for enterprisewide systems modeling, not only should external disturbance be taken into account, but also losses incurred due, simultaneously, to rising health care cost and the need to minimize poor quality medical (patient) care (Hick, Hanfling, Burstein, DeAtley, Barbisch, Bogdan, & Cantril, 2004) In this sense, many industry leaders and practicing engineers have now recognized the significance of integrating the operation, audit and control management of health supply systems (Barbera & Macintyre, 2002). Unfortunately, most, if not all, of these initiatives are still in their infancy. For instance, many health services institutions still lack the required level of modeling capability and organizational preparedness for dealing with external interfering events. Indeed, as we attempt to introduce the latest enterprise paradigms such as the system-of-systems and the reconfigurable enterprises across institutions where a large number of participants interact and evolve on a continuous basis, the lag in health services reengineering practices and health performance management research becomes clearly evident. As early as 2003, Woolhandler, Campbell, and Himmelstein (2003) have noted that, by 1999, 31% of the US health expenditures and 16.7% of the same expenditures in Canada were largely health administrative overhead expenses; clearly, this points to the high potential for HIT models and methods to achieve a significant system performance improvement largely through a reduction in health administrative cost. Moreover, just as we set to data mine and understand the interactive behavior of a complex adaptive system (CAS) (Kiel, and Elliott, 1996; Woolhandler et al, 2003) emerging states can also be generated under various systems configuration to be studied so as to achieve across-the-system improvements.

comparison to service enterprise risk Management From a service supply chain management perspective, the dynamics of the US health care system may be conceived as comprising, with special structures and performance requirements, service supply networks of suppliers on the one hand in terms of caregivers and medical resources, and patient demands on the

xxv

other hand in terms of patient needs and services. As we know, health systems are, without exception, multi-layered and closely interconnected; these systems must necessarily provide high value to their customers in terms of patient benefits and services; as well as the availability of services, security, and privacy of captured and stored patient information. Many of these factors entail the highest priority in risk management. Whereas traditional OR techniques such as game theory (Shubik, 2005) 18 derived from artificial intelligence (AI) and economics, and optimization models 18 taken from quantitative analysis have been extensively used for enterprise modeling, current research focuses largely on core strategic and operational issues such as location configuration, demand response analysis, dynamic pricing and contract management, as well as e-commerce challenges (Graves, Kletter, & Hetzel, 1998; Lambert, Cooper, & Pagh, 1998).19,20 Risk assessment and risk management cover essentially many forms of risks, including, but not limited to, business risk, financial risk, technological risk, and physical risk (Anupindi & Akella, 1993; Argrawal & Nahmias, 1997; Kouvelis & Milner, 2002; Simchi-Levi, Kaminsky, a&nd Simchi-Levi, 2003;) 22, 23, 24, 21. Lately, research into the applications of HIT models and methods has also taken on a broader perspective. Thomas (2002), for example, analyzed the reliability of a supply chain under contingency when it is being impacted by unexpected disasters. Using Bayesian networks, Pai, Kallepalli, Caudill, and Zhou (2003) provided a conceptual business risk assessment framework. Based on industry practitioners’ input, Blackhurst, Craighead, Elkins, and Handfield (2005) summarized the common themes and issues surrounding supply chain disruption. Hale and Moberg (2005) presented a set cover location model that is used in disaster preparation for identifying the minimum number and possible locations of off-site storage facilities for supplies. Lee and Whang (2005) compared the inspection of hazardous goods, such as for transporting explosives, with total quality management (TQM) in manufacturing. A RAND Corporation (2004) report focused on the impact of terrorist attacks on global container supply chain performance and advocated the importance of fault-tolerance or resilience. Despite the stream of well-funded research in the area of service supply chain and enterprise management, the emerging body of knowledge has not adequately addressed the growing complexity and uncertainty-risk management dimensions of the US health care sector. Little, if any, of the past scattered research has provided a comprehensive analysis or given enough attention to systems and systems modeling integration, which has resulted in the following significant challenges: a.

b.

c.

Current research is sparse and isolated, lacking a cumulative agenda. Even so, common terminologies or protocols are not often shared to ease communications among future researchers. Ultimately, there is the lack of a bridging theme to aid the systemic interfacing among heterogeneous models and documented evidence; Past researchers generally hold a narrow view, focusing solely on a small subset of traditional risks. In this sense, most studies deal primarily with individual system components or aspects thereof, without an overall context to account for all interlaced challenges and subsystem interactions; and Enterprisewide modeling and analysis is still in its infancy. Moreover, past studies tend to be mostly conceptual and qualitative, with limited applicability to modern enterprises in addressing current challenges.

Put together, the need for a systemic health services management framework, a representation that would exhibit three fundamental characteristics. First, it would encapsulate a family of heterogeneous

xxvi

models at different levels of detail and for different subsystems, scalable also to the growth of the enterprise and its captured knowledge cannot be overly emphasized. Second, such a representation would bridge high-level qualitative knowledge with quantitative computation by automatic conversion and instantiation, a capability that is crucial to practical deployments of HIT models and methods. Finally, we would also expect such a representation to incorporate quality assurance with confidence and clarity, measured to handle uncertainty in analysis and to guide decision making in all facets of health services management. The MEDIA paradigm, which we will discuss in Section 5, is such an integrative paradigm exhibiting all of the three fundamental characteristics noted above.

state-oF-tHe-art HIt Models & MetHods Before overviewing the MEDIA paradigm, however, we will first review the state-of-the-art HIT models and methods by categorizing and highlighting the various bodies of literature with respect to the emerging engineering approaches in health services applications. Improving a health services delivery system can be achieved through one of two means: (1) relying on medical equipment, technological and/or procedural advances; and, (2) focusing on improved medical, business, clinical and administrative processes. Modern health engineering technologies such as nano-scale materials, genetic informatics, robotics and applications of virtual health have gone a long way to provide effective and reliable prognoses and treatments to a variety of diseases. Such scientific progression foresees an accelerating pattern in light of technological advances. Yet, management theories and practices have taught us that the quality of health services delivery is also dependent on achieving better health information flow and system processes that best fit the type of organizational structure and culture inherent to particular health institutions. In other words, a solid foundation for health care reform may be achieved only through the appropriate balancing of the health resource configuration and process control. In light of this, a large body of models to address medical and health services delivery issues in health system management has evolved through best practices over the years. This approach, which concentrates on the applications of health IT models and methods, is precisely the focus of this review. It has, in the last decade, gained increasing attention among researchers interested in contributing to enhancing the performance of the US health care sector.

Health engineering Today, a new body of health engineering models has emerged. These models are applicable at different system levels, including, for example, at the low level of disease progression of individual patients, that of a clinical session process, or at a higher level involving a health facility of interconnected subsystems and components, or even at the level of a complete life cycle of health services delivery of a particular region or country. In the following section, we survey the extant literature on the new “health engineering” paradigm, following which we shift focus then to the more specific health care modeling and simulation domains. Evidently, health engineering methods and IT models will not only help hospital personnel to reduce medical errors and risk, but aid also in reducing health care costs, improving health services timeliness, and increasing patient satisfaction. In some cases, it is of course possible that the traditional culture and

xxvii

strict legislation imposed on the health care sector, which uniquely distinguishing it from many other industries, may limit the transferability and diffusion of new concepts and methods. Today, opportunities and challenges in the health care sector exist for testing and adopting concepts, best practices, and tools from diverse engineering domains to improve the efficiency of systems processes and the effectiveness of health services delivery in terms of quality, safety, and productivity. In this sense, both the goals and challenges of the health care sector match those of other industries where knowledge engineering has provided a long-term basis for resolving bottleneck issues. Hence, the combination of knowledge engineering and health IT applications promises to provide many more reengineering opportunities for health services delivery, for example, the installation of bedside terminals, the adoption of new informatic methods for infectious disease control, and therapies, and the use of innovative data and DSS technologies in medical and clinical settings, including hospital laboratories and pharmacies. This section reviews emerging operational systems engineering (OSE) research categorized into several different dimensions. Specifically, the focus is on OSE methods abstracted from a mix of engineering disciplines, including human factors, factory and product design, and security engineering. As the managerial processes and services in the health industry are often similar to those of other manufacturing and servicing industries, we argue that the field of health services management can be enhanced through the intelligent applications of traditional health IT OR models, emerging engineering philosophy and methodology, and mature OSE methods.

Traditional HIT OR Models Pierskalla and Brailer (1994) discussed the application of OR models and methods in a broad range of health services management tasks. Major applications of health IT OR models include: 1.

2.

3.

4.

5.

Demand Forecasting – A fundamental input to many other analyses in health engineering, demand forecasting, such as predicting daily census for the different types of resource needed, is important to improve the efficiency of health resources allocation. Capacity Planning - The allocation of bed capacity within the hospital is a critical factor in operational efficiency. Thus, health services capacity planning typically focuses on total bed capacity, bed capacity allocation to different services, surgical system capacity, capital equipment capacity, and ancillary service capacity. Discrete event simulations and semi-Markov process models have been used to examine bed allocation and related capacity questions. Patient Screening - Screening of patients for particular disease can improve medical diagnosis on the one hand and disease detection on the other. It may also be applied individually (individual screening) and/or population-wise (mass screening). Specifically, for individuals, the objective is often to prolong a patient’s life, whereas, in mass screening, the objective may be to minimize the cost at the societal level, thereby lowering the prevalence of a specific contagious disease. Clearly, when attempting to achieve any such objective function in OR modeling, there still may be resource constraints and compliance levels to be factored into the solution. Patient Scheduling - Scheduling is critical for matching demand with the supply of available but limited resources. Most scheduling systems attempt to optimize the combined objectives of patient and worker satisfaction, as well as the utilization of facilities. Clinical Decision Making - Clinical decisions can be aided through OR models that incorporate mathematics and structural analysis. Not only can such analytic models assist in the formulation

xxviii

6.

7.

of health care policies, but they can also be applied to the structuring of critical medical decisions, and the fine-tuning of health systems performance. Workforce Planning and Scheduling - Human resources management is one of the most costly and intensely unpredictable activities that is to be managed intelligently across health organizations. For example, Hershey, Pierskalla, and Wandel (1981) conceptualized the nurse staffing process as that of a hierarchy of three decision levels over different time horizons and with different precision – that of corrective allocations, shift scheduling, and workforce planning. Cost Cutting – Kumar, Ozdamar, and Zhang (2008) have developed several reengineering conceptual and simulation models, which were used for cost containment within the Singapore’s health industry in the domain of supply chain management process reengineering.

Emerging Engineering Philosophy and Methodology Over the years, various streams of engineering philosophy and methodology have been applied to improve health systems performance. Process orientation and patient focus are the essential concepts embedded in these methodological philosophies. Key approaches that have been discussed in the extant literature include: (1) Lean Thinking; (2) Six Sigma; and (3) Theory of Constraints. Young (2005) argues how a clinical session for a patient’s treatment would serve as a good analogy for explaining how these different streams of engineering philosophy and methodology can be combined to address various system bottlenecks encountered in the health services delivery industry. 1.

2.

3.

Lean thinking – Kollberg, Dahlgaard, and Brehmer (2007) believed that the idea of lean thinking is applicable specifically to health care systems in a number of ways. For example, just-in-time (JIT), level scheduling, and multi-skilled teams are generic techniques that can create a smooth operation process flow through the matching of the supply-demand level of health care resources. When applying lean thinking to health care, a measurement framework for lean initiatives that reflects both efficiency and effectiveness of health systems performance, such as patient satisfaction, referral management, process mapping, and fulfillment of targets and policies, is needed in order to fully capture lean changes. Six Sigma - This methodology involves standardized data collection and informed reporting protocols based upon a well-controlled quality feedback cycle to minimize variations and quantitatively align production or service quality to a predetermined standard. “Bridges to Excellence” is an example of a national case initiative based on Six Sigma quality feedback methodology launched to improve clinical care quality. This initiative targeted on physicians and their practices to enhance patient care quality (Brantes, Galvin, & Lee, 2003) and had further been incorporated as part of a collaborative product commerce (CPC) approach to health supply chain purchasing (Ford & Hughes, 2007), as highlighted in the next section. Theory of Constraints (ToC) – Similar to Six Sigma, ToC applies the root cause thinking processes for analyzing system bottlenecks. Unlike Six Sigma, however, ToC attempts to deal with managing constraints in CAS not from a technical limitation perspective, but from a “qualitative” and philosophical perspective. The methodology is first applied to identify the most vulnerable constraint, then exploiting and increasing flow through that constraint, working from the weakest link upward to other links between that constraint and the overall system.

xxix

Mature OSE Discipline A maturing OSE discipline focuses on examining, analyzing, and further understanding the operating elements and systems dynamic processes in complex systems to achieve efficient and effective systems performance. This greatly supports the view of health care as a CAS. In this sense, OSE tools and techniques may be applied to achieve a balance for meeting multiple goals, for example, quality patient safety, accessibility, availability, comprehensiveness, and affordability of care. 1.

2.

3.

4.

Supply chain management (SCM) – In recent years, SCM topics have gained significance as health services organizations vie to lower the cost and improve the ease of accessibility and delivery of health services and their associated resource supplies (Brantes et al, 2003; Ford & Hughes, 2007). 38,39 . Ford & Hughes (2007) identified potential barriers that health services organizations must overcome in order to apply SCM principles successfully within the health care sector. In their study, they inferred that physician services are the primary channels in a health care supply chain to provide the relevant expertise and services to group practices, hospitals and pharmacies. These practices, in turn, behave as secondary channels, acting as refineries and production facilities to serve health insurance distributors and purchasing programs in the supply chain. Therefore, the starting point for cost containment of health services in SCM will be determined by how physician services are being managed. A specific form of SCM used by US employers is the collaborative product commerce (CPC), which is discussed next. Collaborative product commerce (CPC) – CPC attempts to extend the limited boundaries of enterprise collaboration for product design by leveraging innovative e-technologies to engage members from internal as well as external constituencies. CPC approach differs from other SCM tools in two significant aspects (Swinehart & Smith, 2005): (1) it permits inter-organizational collaboration feeding on a common supply chain or meeting similar consumer product or services needs; and (2) its processes are transparent to all stakeholders. CPC models, therefore, allow health provider organizations and/or third-party payers, who are located in different health care markets, to share in innovative product life cycle designs. Business process reengineering (BPR) – BPR is a process-driven technique to improve the efficiency and effectiveness of a business through meaningful process redesign, change management, and system reorganization. As an example, Kumar, Swanson, and Tran (2009) conducted BPR using simulation modeling on the complex operating theatre (OT) system in a Singapore Hospital. His case simulation produces two recommendations: (1) a need to redesign the OT process in order to maximize its productivity without altering the current workload of surgeons and anaesthetists; and (2) reviewing the OT utilization data periodically so as to derive a meaningful productivity index and accurately gauge its utilization. Health management information systems (HMIS) – HMIS, which has to do with all aspects of business information systems functions in health care, plays an integral part in any modern health services system.43 As noted earlier, a national health IT strategy to support and ensure systems interoperability to link health services networks throughout America is believed to be a necessary step towards realizing US health care reform (IEEE-USA, 2005). Today, many traditional HMIS functions can be easily and logically augmented to encompass database, model-based and knowledge-based HDSS technologies when trying to build OR models to aid in the management

xxx

of health services systems, or more specifically, to improve administrative productivity, increase clinical decision-making responsiveness, and enhance patient care quality.

Health services Modeling Serving as the foundation for our proposed MEDIA framework to be discussed later, the extant literature in health services models may be further subdivided along the following three dimensions: 1. 2. 3.

the scale (level) of health system problems being studied; the goal being sought or performance measure being evaluated; and the modeling method being applied.

The intelligent applications of health IT models and methods depend largely on understanding how each of these dimensions will impact on a specific health services modeling study. Many variations and types of models exist such as policy models, procedural models, intervention models, graphical models, deterministic as well as stochastic models. As a case in point, Eldabi and Young (2007) indicated that quantitative models, methods or tools can be studied across different organizational levels, from the physical and mechanical design level through the services and policy design level.

The Scale (Level) of Health System Problems This dimension focuses mainly on the extent and/or scale of health services problems to be modeled. Four sub-levels include: (a) Individual patient disease models; (b) Operational process models; (c) Organizational system-level models; and (d) National sociopolitical-level models Individual patient disease models focus mainly on the biological disease processes occurring in individual patients, that is, the infection processes among either healthy or infected person/population. These models can range from microbiological or cellular, to organ, as well as person-to-person transmission level. O’Leary (2004), for instance, simulated person-to-person infection by building mathematical models to predict the epidemic path of disease transmission, to assess possible infectious outcomes of events over time, and to test the effectiveness of different intervention measures such as vaccination strategies and quarantine policy. In general, such disease progression models are applicable to evaluating clinical effectiveness or cost effectiveness of different interventions to particular disease (Brailsford, 2007). The operational process models are devoted to observing and simulating individual patients residing in a ward, a clinic, or a hospital department such as the Emergency Department (ED) or Intensive Care Unit (ICU). Often, such models are used to facilitate business process reengineering, resource allocation, capacity planning, staffing, and scheduling challenges. Tan, Gubaras, and Phojanamongkolkij (2002) for instance, employed a discrete event simulation model for studying capacity planning, staffing, and scheduling of Dreyer Urgent Care Center. Organizational system-level models combine different departments within a large institution or enterprise, and attempt to study the interactions among the different departments. Typically, such models address longer-term and broader issues, similar to the conceptualization of enterprisewide models. System dynamic (SD) is a common example of organizational system-level model that is often applied at a

xxxi

strategic level where the stakeholder is interested to look at the forest more than just the trees (Brailsford, Lattimer, Tarnaras, & Turnbull, 2004). Finally, national sociopolitical-level models encompass and address a wide range of very high-level issues. Having a view at a national sociopolitical level on the state of emergency, for instance, will aid government policymakers and decision makers design a reliable health services delivery policy that will coordinate hospitals, police, and emergency units for a rapid and more readied response. For example, the BioSense Real-Time Clinical Connection Program of the United States (Laudon& Laudon, 2007) builds a national surveillance system model to continuously summarize and analyze the disease and health information by source, day, and syndrome for each ZIP code, state, and metropolitan area. This model is designed to improve the national capabilities for disease detection and monitoring, as well as awareness of real-time health situations.

Goal Sought & Performance Measures In health systems, goals sought are often different from many other for-profit industries, for example, appropriateness of care, safety, and patient satisfaction are critical relative to cost and resource utilization, not just profits. These goals are usually measured by a set of key performance indicators such as patient throughput as measured in system wait time and wait line; length of stay (LOS); system capacity; utilization of staff, equipment, and space; cost; and various other related measures such as service quality in terms of error and patient satisfaction. Major groupings of studies on various health systems performance measures include the following: 1.

2.

3.

4.

Length of Stay (LOS) and Patient Throughput – Ramis, Palma, Estrada, and Coscolla, (2002) created a generic simulator within a network of clinics to reduce patient LOS in the system. The simulator also facilitated the appropriation of resources and reallocated these resources based on the number of patients and how quickly they go through the system. Bosire, Wang, Gandi, and Srihari, (2007) modeled a computer tomography (CT) scan facility in a hospital to study how patient wait time can be minimized and how staffs can be used efficiently to increase patient satisfaction. Resource Allocation and Capacity Planning – Rico, Salari and Centeno (2007) built a model using ARENA simulation software and OptQuest heuristic optimization to increase system capacity, improving nursing staffing allocation, and augmenting the utilization of equipment and space. They suggested multiple ways in which the number of nurses needed for health services delivery during a pandemic influenza outbreak may be combined. Cahill and Render (1999) created a model to assess ICU bed availability. Their model was applicable for excess capacity rebalancing that would otherwise lead increasingly and unjustifiably to wasting limited health resources such as bed space and personnel in the Cincinnati VA Medical Center. Cost Containment - In the Cahill-Render (1999) study cited previously, an additional application of their model was to contain costs by drawing from outside resources in meeting patient needs. Stahl, Roberts, and Gazelle (2003) also investigated the strategy for setting appropriate preceptorto-trainee ratio in the context of a teaching ambulatory care clinic. The key purpose of their study was to achieve an optimal financial feasibility and to cut system operational cost. Policymaking - Policymaking is fueling a resurgence of interest in modeling and simulation to improve health services delivery performance. This is especially applicable in countries such as Canada and the UK, where the system is based largely on a single payer (the government). More

xxxii

specifically, these systems employ a governmental body of agencies to manage the entire system through a set of national performance measures. A specific example would be the star rating used for measuring wait time targets for emergencies such as how long it takes a patient to be served upon admission into the emergency department (ED) of a hospital. Gunal and Pidd (2005) provided an example in their policy-oriented simulation study that aimed to improve the accuracy of the rating system in the UK National Health Service (NHS) through uncovering performance irregularities.

Modeling Method As well, a number of modeling methods and techniques have been identified for health care services implementation. Among these, the most influential, useful and widely applied methods are simulation techniques, which include the mainstream methods as well as simulation algorithms based on artificial intelligence as discussed in Cooper, Brailsford, and Davies (2007) and Kuljis, Paul, and Stergioulas (2007) `s study. Table 1 summarizes the current prevalent modeling technologies and simulation methodologies, as well as their potential application domains in health services analysis. 1.

2.

3.

4.

Discrete-Event Simulation (DES) - Among the most widely used simulation techniques in health care as evidenced by many previously cited studies, DES appears to be tailor-made for hospital systems to study queuing behaviors of patients waiting for appointments, investigations and treatments. In particular, DES allows the modeler to construct more complex, dynamic and interactive systems. Nonetheless, as Cooper et al. (2007) pointed out in their choice of modeling technique for evaluating health care interventions, it may take more time and money to develop DES models. System Dynamics (SD) – To resolve systems bottlenecks and understand emerging systems states, SD modeled patient flow behaviors in complex health services systems by capturing the feedback loops and inventory control rules for patient arrival, discharge and follow-up visits. This is similar to studying how water flows through a heating system. SD has gained popularity in recent years and some researchers who first applied the DES had switched to SD to deal with the dynamic changes in patient flow. For example, after using DES to redesign the phlebotomy and specimen collection centers in Calgary Laboratory Services (Alberta, Canada), Rohleder, Bischak, and Baskin (2007) have later on decided also to implement a SD model to handle the unexpected performance discrepancies found due to the dynamic interactions within these service centers. Markov process models - Markov models can describe changes in the state of patient health over time such as the case of benign vs. malignant tumors. In this sense, Cooper et al. (2007) argued that Markov models are suitable for chronic disease interventions. Accordingly, they employed a Markov model to evaluate the effectiveness of statins, one of the cholesterol lowering drugs, over time for at risk patients of coronary heart disease (CHD), based on a so-called Southampton CHD model (Cooper, 2005). Monte Carlo Simulation – Essentially, Monte Carlo uses repetitive sampling process to make estimates about key performance variables of interests under uncertainty conditions just like throwing a dice repeatedly to predict the chance of picking a winning stock in the Dow Jones Industrial market. For example, Jacobson, Lindberg, Lindberg, Segerstad, Wallgren, Fellstrom, Hulten, and Jensen-Waern (2001) used the Monte Carlo methodology to sample various potential vaccine price distributions determined by the mix of care providers vis-à-vis parent-guardian acceptance so as to assess the economic values of variously combined vaccines for pediatric immunization.

xxxiii

Table 1. Current modeling and simulation techniques and their applications in health Technique

Potential Applications

Discrete-Event Simulation (DES)

Process reengineering, ward layout design, patient pathway design, scheduling, queuing management

System Dynamics (SD)

Strategic and operations management, alteration management, resource and asset management, patient pathway management

Continuous Simulation

Physical/ biological laboratory processes control

Monte Carlo Simulation

Decision making under uncertainty conditions, risk analysis in the long run

Agent-Based Simulation

Demand and supply management, health economics, risk management

Decision tree

Acute interventions but not for disease recurrence modeling

Markov process

Cohorts of patients between health states over time, chronic disease intervention

Operations research

Patient arrival patterns, LOS management, waiting time management, cost management

Human factors and ergonomic models

Workload analysis, safety monitoring, productivity increasing, error reducing

5.

6.

Continuous Simulation - Continuous simulation is primarily employed to accommodate continuous systems variables and therefore has limitation in its applicability for health care studies. Specifically, it is limited predominantly to physical or biological laboratory processes control simulation such as modeling the trajectory of a missile launch. In health care, it can, for example, be used to assess the design of equipment to further enhance the production volumes and manufacturing process efficiencies of pharmaceutical products (Kuljis et al., 2007) 67 Decision tree – Based a hierarchical tree-like structure, decision trees aid decision making through an assessment of various probabilities in terms of possible consequences corresponding to different options and alternatives. Cooper et al (2007) showed that a decision tree can facilitate comparative decisions on mean life expectancy of CHD patients. Here, the current response states for CHD patients are compared to more efficient ambulance delivery services and thrombolysis (“clot busting” drug) intake that may result in the death and/or the survival of the patients.

In summary, whenever a health system problem is encountered, a critical issue then is how to make choose intelligently among various available modeling and simulation techniques. Schriber and Brunner (2007) proposed to explore the nature and logical foundations in every method and software and thus gain a detailed understanding of “how simulation works”. Cooper et al. (2007) stated in their study that the choice of modeling technique depends on several aspects including modeling technique acceptance, model appropriateness, dimensionality, and ease and speed of model development. Generally a decision can be made based on the complexity and dynamics of the system to be modeled in terms of interaction

xxxiv

of systems elements, model size or resource constraints. The introduction of OSE philosophies also helps making a better combination of choices. For instance, Young (2005) studied the philosophies driving changes in health care services delivery such as Lean Thinking and the Theory of Constraints. He proposed that a strategic agenda could be created out of a three-way fusion of health care delivery, industrial process and simulation capacity, on which basis, an appropriate modeling and simulation techniques can then be selected.

tHe Model and eVIdence drIVen Integrated analysIs (MedIa) paradIgM & Its applIcatIon The MEDIA paradigm, which has previously been introduced at the beginning of this review in Figure 1, offers a handle to unraveling today’s highly complex and environmentally uncertain health services enterprises by focusing on the effective integration of operational models and evidence. Three key phases underline the MEDIA paradigm: (a) Referential ontology modeling; (b) Hybrid probabilistic model integration; and (c) Adaptive knowledge fusion for quality assurance. Referential ontology modeling relates to the qualitative representation of model structure that is generally constrained by the health services systems complexity and uncertainty. Nodes, representative of crucial system artifacts, and the interdependencies among nodes, that represent their relations, typically encapsulate the model structure. Figure 4, for example, shows the nodes and the interdependency links among these nodes for a partial risk assessment model of operational and information risk representation in the context of a generic model structure. As the purpose of referential ontology modeling is essentially to support the generation of hybrid probabilistic models, probabilistic modeling requirements should first be examined in terms of the domain of enterprise services and elements, environment risks, and their relations (dependency). The emergent systems complexity is then represented by referential knowledge in formal ontology based on description logic domain and the uncertainty among systems relations is often addressed through probabilistic representation for ontology modeling such as applying the Bayes approach. Accordingly, a critical step is the conversion of ontology to probabilistic models. The next phase, hybrid probabilistic model integration, entails bridging the qualitative ontology representations with computational analysis to quantify the relations among referential nodes and to achieve an integration of heterogeneous models. This quantification process comprises fundamentally an attempt to assign probabilistic distribution to nodes and/or conditional probability to relations among the nodes. It is important to note that boundary nodes are those nodes shared by multiple models and the same quantification process is applied to these nodes to connect among the models. Special procedures such as those algorithms dealing with virtual and soft evidence may have to be instantiated to adjust the associated probabilities to satisfy all of the probabilistic constraints across each of the model to be integrated (Kim, Valtorta, & Vomlel, 2004; Xiang, 2002). In this second phase, the core migration procedures can utilize semantic web techniques such as ByesOWL framework (Ding, Peng, & Pan, 2006) to convert the ontology to probabilistic networks. The final phase, knowledge fusion, aims at improving the accuracy and confidence of hypothetical estimations about key systems variables based on discrete evidence gathered from multiple locations throughout the systems being studied. Furthermore, adaptive knowledge fusion attempts to achieve a high quality assurance of information collection and operations management decisions in the knowledge

xxxv

Figure 4. A model structure represented by nodes and interdependencies among nodes

vaccination

backup

epidemic

HMM Power Model (1)

Markov Cancer Model (2)

power surge computer

staff

patient

OR

Bayesian Parameter Model (3)

Staff B

Staff A Patient arrival

OR

Patient exit

Computer network topology

staff OR Simulation Model

Staff C

Probabilistic Communication Model (5)

(4)

Decision Model (6) node dependency

evidence control

Resource allocation information

utility

resource flow link

fusion process. Appropriate quality assurance strategies are used iteratively to guide the knowledge fusion process and are typically aided by dynamic quality scores or indices calculated from information entropy of evidence and effectiveness of different operations. Essentially, these strategies will attempt to provide insights to key questions regarding the knowledge fusion process, including: (1) how good the resulting diagnosis or prediction may be; (2) when to engage and/or stop the knowledge integration procedures; (3) where and what information is to be collected; and (4) finally, what resource allocation decisions could be best implemented (Li & Chandra, 2006; Li & Ji, 2005).

MedIa procedural Framework Figure 5 overviews a MEDIA procedural framework that can be used to guide the implementation of the different phases involved in the construction of relevant health IT models and methods for a generic risk management system problem. In the context of the MEDIA paradigm, any kind of a management task such as a risk assessment scenario for a specific system like a hospital in a chosen ontology domain, specifically, health care, will therefore pass through a set of generic procedures. •

Phase 1a: Physical constructs and security components are modeled in ontology bases for the chosen domain. These class templates have common attributes and relations sufficient to support the modeling of most management tasks;

xxxvi

Figure 5. Flowchart of the model and evidence driven integrated analysis (MEDIA) Domain Specific Physical Constructs

Security Components

System and Task Specific System Information

Relations

Security Tasks

Relation Instances

Existing Models

Model ontology

Ontology

Instantiate system models

Identify boundary interfaces

Referential Dependency Models

Translate ontology and fit parameters

modify System Specific Evidence

•

•

•

Link models

Hybrid Probabilistic Computation Models Adaptive knowledge integration

Phase 1b: Aided by the domain templates above, two different types of inputs are needed for referential dependency models to be instantiated for the specific system and the risk management task: (i) for an existing model, boundary nodes and their dependency, and (ii) for a new model, complete knowledge about the model structures of nodes and dependencies; Phase 2: Probabilistic computational models are then generated through ontology translation and parameter fitting to capture the interdependency and their associated uncertainty. This is helped by additional descriptions specific to the system and task, for example, hypothesis nodes of interests and available evidence collection positions designated for risk management; Phase 3: When evidence can be found, probability distributions for nodes connected within a network as well as quality scores are updated accordingly through probabilistic inferences. The probabilistic inferences that are carried out here should be specific to the different types of models, for example, Markovian process, Bayesian, or DES models.

If a model needs to be modified in terms of both its structure and parameters due to changes in the underlying systems dynamics, or a new model needs to be introduced, the process goes back to Phase 1b to repeat or iterate on the instantiation process (Cooper & Herskovits, 1992; Heckerman, Geiger, & Chickering, 1995). The rest of this section will turn to discussing the application of MEDIA paradigm in a multi-tier capacity planning system.

xxxvii

MedIa application for a Multi-tier capacity planning system In the past, mathematical models for the optimal timing prediction of individual disease treatments such as cancers and for the best health services scheduling such as emergency and operation rooms scheduling have been developed respectively. Yet it is important for these models to work together as may be necessary for setting up new global optimization strategies. In the real world, for example, the service appointment and operation room scheduling do not always guarantee that the patient gets the screening test or the surgery at the best chosen timing as dictated by the individual disease progression model. Suboptimal timing for treatment may therefore have to be chosen and analyzed again for potentially undesired consequences to patients. As this might further delay the treatments for patients and jeopardize their life quality while increasing medical cost, these two types of models may have to be reconnected and integrated through their interdependency to optimize the two processes of the overall system. In addition, the challenging issue of uncertainty arises when we consider the dynamic changes such as an epidemic or more disturbing natural and man-made emergencies. In a normal operating environment, the discussed risk in the abovementioned case may not be very serious. However, if epidemic cases such as SARS or events such as the 9/11 terrorist attack occurring simultaneously, for example, excessive demand on medical aids in responding to these events may easily deprive regular patients of “optimized” treatments. The strategy here then is to factor in the uncertainty by integrating and incorporating additional appropriate models. Indeed, this integration can even be expanded either horizontally or vertically into as many channels of the health care system as needed such as the sub-units within the same health care enterprise, or the various business processes and institutional structures of a regional health alliance. Naturally, instead of building a new, but highly sophisticated model from scratch by applying the same algorithmic approach or method, applicable existing models and/or the most suitable type of model for each of these processes can now be reused or variously combined to satisfy the overall systems model optimization requirements. Such an application shows the dominant complexity in terms of a large number of participating systems, subsystems, and basic units, and more significantly, their interdependency. Figure 4, which was also discussed earlier, illustrates a partial risk assessment model of operational and information risk representation in the patient care process. This hybrid probabilistic modeling solution employs a variety of models. A hidden Markov process model (1) predicts the state of power supply for potential power surge. A heterogeneous Markov process model (2) captures the progression of a given type of cancer and estimates the best treatment timings. The predictions from these models enter a Bayesian parameter model (3) that estimates the states of corresponding variables of staff, patient, operation room, and computer network, with the help of other available evidence. Then a simulation model (4) can run to collect statistics on current operating room capacity, using these parameters. Another model in the form of a probabilistic network graph (5) accounts for computer network topology and user configuration. It calculates the risk of information leaking and availability for the different collaboration scenarios based on staffing and computer network situations. Actually this availability prediction can be feedback to the operating room simulation model, if applicable. Moreover, a computational decision model (6) will enable the computation on the utility based on performance information for different allocation actions for given resource constraints. This decision model can also be distributed to more nodes across the system, if and as needed.

xxxviii

Imagine how a comprehensive model of a medical center can be generated through the application of the MEDIA framework. Various sources of evidence can be introduced into this model on a continuing basis, in terms of errors, delays, and clinic starvation alarms from physical, software or human sensory channels, signaling emerging or potential problems. The administrators or doctors, facing an uncertain and complex operational environment, have to ask first, given the evidence, what the big picture really is and where the weakest links may be situated. Then the best decision (where and how) to invest resources (capital and manpower) to avoid further deterioration of the system state and to mitigate the problems can be made intelligently. Moreover, this process will be carried out, not haphazardly and without knowledge, but in an active and timely manner to handle existing incomplete and uncertain information. Imagine also that this model can further be updated with real-time system status by monitoring facilities/sensors in an “online” fashion. The key decision makers and policymakers are not asked to just (always) rely on the “average” profile about user/disease from collected historical data. For example, patient arrival may fluctuate from day to day; different treatment results of a special disease may require different actions to be taken; and different caregivers as well as health administrators may also be motivated to stick stubbornly to their personal information processing styles and biases in making certain decisions. Therefore the latest evidence, collected on a continuing basis to update information relating to individual “special” patient, at a prescheduled pace and/or for significant events, will all be made available as feedback to the existing model structure. Over time, the model structure or parameters may, of course, be no longer appropriate, relevant or accurate enough for current or future predictions (so called “concept drift”). Then, model learning and knowledge fusion will again be executed at these times as well. Put together, the MEDIA paradigm is clearly a useful and practical framework that can be applied intelligently across any type of a real-world scenario to enable the mixture and integration of various health IT models and methods to facilitate the making of short-term key health services delivery and management decisions as well as to aid longer-term health systems planning and policymaking.

conclusIon Essentially, our review on existing bodies of knowledge about the state-of-the-art health IT models and methods has led us to become increasingly aware of the need for and significance of having an integrative platform for leveraging existing models and methodologies intelligently such as the MEDIA paradigm. This paradigm aims to offer an appropriate framework for conjugating heterogeneous probabilistic models including the Bayesian network (Pearl, 1998), Markov process models (Doob, 1953), and Monte-Carlo simulation (Fishman, 2001), through virtual or soft evidence update of distributions over boundary nodes. The rationale for achieving such an integration is that: (1) inherent probabilistic representation and inference is able to deal with uncertainty; (2) many such probabilistic models have built-in mechanisms, such as in dynamic Bayesian networks and Markov process models to deal with the temporal dependency; and (3) many of these models support localization of information processing and can be easily implemented into agent-based architectures, aided by the interdependency present in enterprise constructs and model components. More specifically, just as languages spoken by people, we have seen that ontology representation is necessary for computer models to communicate and interact with each other. Ontology defines the concepts and the relations of generic elements in a domain and provides a common “language” shared

xxxix

by components and subsystems of MEDIA. It is more than just a classification or a dictionary because dynamic behaviors of the systems elements are also captured using description logic and language such as RDF and OWL (Baader, Calvanese, McGuinness, Nardi, & Patel-Schneider, 2003; Barwise, 1997). We have also seen how the integration of health IT models and methods can benefit through the application of such a shared “language” for accurate and practical representation of various constructs in information theoretic terms to represent risk and uncertainty in complex systems environments, thereby permitting the accurate and rapid assessments of emergency situations and allowing efficient and effective health services to be delivered productively. Currently, there is a dire shortage of major studies in the applications of interoperable, integrative health IT models and methods. As well, inadequate knowledge has yet to be derived from other industrial and system engineering sectors such as manufacturing enterprises into aiding comparative processes in health services delivery and management. For instance, the surgery department is always at the core of many hospitals. Managing operation rooms requires a variety of expensive resources including surgical space and facility, equipment and material supply, nurse, technician and doctor, to provide timely appointments to patients. It is a challenging task in and by itself. Yet, the people manning the surgery department has still to interact with all the inpatient-outpatient departments and clinics, as well as to be linked to a huge number of medically interconnected components that will affect their own behaviors and dynamics at any moment. Reports and requests are largely the only primary information artifacts interfacing between the operation rooms and other care providers in the same hospital, and mostly for discrete interactions updated at a fixed time interval. The surgeons and surgical administrator will often not be familiar or have knowledge of the status of other systems and organizational components, and without the aids provided to them from an integrated platform of health IT models and methods, they cannot possibly make the best decisions, including the operation room assignments. Ironically, most hospital departments and clinics nowadays have management models, policies, and supporting software packages in place, but all running individually, to “efficiently” schedule patient appointments and make medical decisions in an isolated manner. With the MEDIA framework application, decisions on such localized “best” operations can now be optimized for the entire system. Feasible assignments satisfying all of the required health policies for the various connected health provider institutions can then be assured. Our case example may now be further generalized to many other equally complex scenarios that commonly occur in health services delivery. For example, even for a specialty clinic such as the urology or cardiology unit located within a hospital, it has to deal with many other elements including primary care, inpatient, outpatient, emergency, surgery, lab testing, logistics, and many other areas in terms of the different nodes and types of information flow, including patient, provider and material flows. Similarly, an insurance company has to navigate through patients, medical providers, pharmaceutical companies, and other vendors or suppliers in order to keep routine financial claim transactions in order. For all of these health provider organizations, if we do not attempt to empower the different health services managers with the necessary, interoperable health IT capabilities, there will be no assurance that a well-coordinated effort in patient care delivery will be sustained. The need for such a well-coordinated effort and enterprisewide collaboration across the spectrum of health services delivery is even greater when we consider health alliances, which may be formed at varying scales, levels and sizes. Frequently emerging out of the cooperation among multiple health provider institutions of primary clinics, hospitals, research institutes, insurance management organiza-

xl

tions, and even government health agencies, health alliances are dependent on the separate units to work together in order to achieve high quality patient care. Piece-by-piece examination of related management models and systems does not and will not offer much help when local operation decisions have to be made based upon disparate evidence. In such networked systems and sub-systems, a medical provider or manager needs to know far more than his or her own world in order to make better, if not the “best,” decisions for the patients and the health alliance organization as a whole. Often, the critical need is not more models, or even more sophisticated yet isolated models – where abundant, if still not adequate, models have already been developed, implemented, and studied – instead, the primary need here may just be how to integrate existing models productively and intelligently so as to allow interoperable health IT applications to be shared. Then, and if necessary, new models and methods can be added to enhance those aspects of decisions that may still be lacking or have yet to be addressed. A wonderful developing story for applying the MEDIA paradigm is the 2009 case of the “swine flu” (H1N1) pandemic that is ongoing at this point in time (Esterl, 2009). It is just impossible to have one model or system to capture everything relating or reported about this pandemic, from the private clinics to the US governmental health services sector plan for controlling H1N1 to the different magnitude of H1N1 developments spreading throughout the globe. In other words, it is just not enough to harbor the narrow perspective of a tiny component within the entire global health service system, but having an interconnected worldview of even just how the H1N1 is affecting the different nations will significantly improve the patient survival chance and H1N1 prevention for a particular or entire country. Even so, the demand to integrate models and evidence for seamless analysis and management of a pandemic like H1N1 may be attributed largely to the following facts: 1.

2.

3.

The complexity and uncertainty are greatly increasing in the system of systems context of health services supply chains and global health information exchange networks to which any analysis and management systems have to appropriately respond. The rapidly changing medical and health technology landscape and innovative practices in medicine, health management, industry and government policy, as well as the evolving field of health IT models and methods require the constant incorporation of new systems perspective and integrative model approaches. Unfortunately, as noted previously, vast legacy HISs have typically been built piece by piece, resulting in the preservation only of disparate health IT models without the capability to leverage intelligently from evidence aggregated over time from diverse sources embedded in the overall global health care system.

Development of novel technology and unconventional business model has inspired complex enterprise structures to evolve as in the case of the US health services delivery system. These enterprises usually form close partnerships in order to survive and grow under increasing competition, for which decision making and policymaking relying on existing and past discrete models for isolated system segments is no longer sufficient. Another obstacle many of these health enterprises encountered in across-the-organization management is the complicated systems dynamics. This needs to be dealt with by systematically increasing the connectivity of existing models, resulting in the constant understanding and monitoring of the evolving systems states comprising rapid changes in participants and configurations along both temporal and spatial dimensions. As characterizing these different systems states efficiently can simply become overwhelming, health services quality assurance is undoubtedly a very demanding task. Interoperable,

xli

integrative health IT models and methods via the MEDIA paradigm aim at reducing such complexity to meet the demands of high quality health services. The MEDIA paradigm also addresses these systems challenges through referential ontology modeling, hybrid probabilistic model integration, and adaptive knowledge fusion that adopt a real system-wide and process oriented perspective. Thus, the introduction of the MEDIA framework to emerging health IT models and methods research represents a key step to unifying our knowledge for modern health services enterprise management. The future of health IT models and methods is dependent on the development of even more powerful integrative frameworks beyond the MEDIA paradigm. Such a paradigm will seek to provide a consistent approach to reusing and integrating available knowledge, dealing with complexity and uncertainty dimensions in multiple ways, and encouraging the sharing and exchanging of health information privately and securely across organizational boundaries that do not currently exist. This systematic methodological conjugation represents a departure in traditional integrated modeling and analysis from fitting all systems into just “one type of model” to an emphasis on knowledge interpretation, interaction and fusion among models. This makes it much easier to achieve balance between the local and global representation of different models. In other words, the hybrid probabilistic model integration can actually make use of any available models and/or choose the most suitable ones for each subsystem/task, rather than having to start the modeling process from scratch each and every time one or more new systems problem(s) is encountered. In closure, the MEDIA methodology or an expansion of such a methodology can further be applied to create novel solutions to new health systems challenges faced in a wide range of practical contexts that vie to achieve the convergence of enterprise systems and processes. The trend toward preventive health, alternative and integrative medicine, e-health and healthy lifestyle promotion, for example, are all virgin grounds for the applications of integrative, interoperable health IT models and methods. Imagine how a self-monitoring community health care system that can be equipped with a regional center that uses interoperable and integrated health IT models and methods to aid in guiding individual residents throughout the community to cope with undue stresses arising from risky driving behaviors on a daily basis, become quickly alerted and fully prepared for emergencies by deploying the community health resources efficiently and effectively on an as needed basis, and channeling any and all unused health resources to evolve the community in multiple ways to adopt a healthy, low risk and more active lifestyle environment within the community. Joseph Tan McMaster University, Canada Xiangyang Li University of Michigan - Dearborn, USA Yung-wen Liu University of Michigan - Dearborn, USA

xlii

reFerences Agrawal, N., & Nahmias, S. (1997). Rationalization of the supplier base in the presence of yield uncertainty. Production and Operations Management, 6, 291–308. Anupindi, R., & Akella, R. (1993). Diversification under supply uncertainty. Management Science, 39, 944–963. Baader, F., Calvanese, D., McGuinness, D. L., Nardi, D., & Patel-Schneider, P. F. (2003). The Description Logic Handbook: Theory, Implementation, Applications. Cambridge, UK: Cambridge University Press. Barbera, J., & Macintyre, A. (2002). Medical and health incident management system: A comprehensive functional system description for mass casualty medical and health incident management, Retrieved from http://www.gwu.edu/~icdrm. Barwise, J. (1997). An Introduction to First-Order Logic. Handbook of Mathematical Logic (pp. 5-46). Amsterdam. Blackhurst, J., Craighead, C.W., Elkins, D., & Handfield, R. B. (2005). An empirically derived agenda of critical research issues for managing supply-chain disruptions. International Journal of Production Research, 43(19), 4067-4081. Bosire J., Wang, S., Gandhi, T., & Srihari, K. (2007). Comparing simulation alternatives based on quality expectations. In Proceedings of the Winter Simulation Conference (pp. 1579-1585). Brailsford, S.C. (2007). Tutorial: advances and challenges in healthcare simulation modeling. In Proceedings of the 39th Conference on Winter Simulation (pp. 1436-1448). Brailsford, S., Lattimer, V., Tarnaras, P., & Turnbull, J. (2004). Emergency and on-demand health care: modeling a large complex system. The Journal of the Operational Research Society, 55(1), 34-42. Brantes, F.D., Galvin, R.S., & Lee, T.H. (2003). Bridges to Excellence: building a business case for quality care, Journal of Clinical Outcomes Management, 10, 439. Cahill, W., & Render, M. (1999). Dynamic simulation modeling of ICU bed availability. Winter Simulation Conference, 2, 1573-1576. Cooper, G.F., & Herskovits, E. (1992). A Bayesian method for the induction of probabilistic networks from data. Machine Learning, 9, 309-347. Cooper, K (2005). A comparative study of modeling approaches for the evaluation of healthcare intervention with reference to coronary heart disease. PhD Dissertation (unpublished), University of Southampton. Cooper, K., Brailsford, S., & Davies, R. (2007). Choice of modeling technique for evaluating health care interventions. Journal of the Operational Research Society, 58(2), 168-176. Denton, B., Fowler, J., Schaefer, A., Batun, S., Erdogan, E., & Gul, S. (2009). Optimization of the design and operation of surgery delivery systems. In Proceedings of the 2009 NSF engineering research and innovation conference.

xliii

Ding, Z., Peng, Y., & Pan, R. (2006). BayesOWL: Uncertainty modeling in semantic web ontologies. Soft Computing in Ontologies and Semantic Web. Springer-Verlag. Doob, J. L. (1953). Stochastic Processes. New York: John Wiley and Sons. Eldabi, T., & Young, T. (2007). Towards a framework for healthcare simulation. In Proceedings of the 39th Conference on Winter Simulation. 1454-1460. Esterl, M. (2009, June 11) Swine Flu pandemic is closer, says WHO. Wall Street Journal. Fishman, S. (2001) Discrete-Event Simulation: Modeling, Programming, and Analysis. New York: Springer-Verlag. Ford, E.W., & Hughes, J.A. (2007). A collaborative product commerce approach to value-based health plan purchasing. Supply Chain Management: An International Journal, 12(1), 32-41. Graves, S.C., Kletter, D.B., & Hetzel, W.B. (1998). A dynamic model for requirements planning with application to supply chain optimization. Operations Research, 46(3), 35-49. Gunal, M., & Pidd, M. (2005). Simulation modeling for performance measurement in healthcare. In Proceedings of the 37th Conference on Winter Simulation (pp. 2663-2668). Hale, T., & Moberg, C.R. (2005). Improving supply chain disaster preparedness: A decision process for secure site location. International Journal of Physical Distribution & Logistics Management, 35(3), 195-207. Heckerman, D., Geiger, D., & Chickering, D. (1995). Learning Bayesian networks: The combination of knowledge and statistical data. Machine Learning, 20(3), 197-243. Hershey, J., Pierskalla, W., & Wandel, S. (1981). Nurse staffing management. In D. Boldy (Eds.), Operational Research Applied to Health Services, (pp. 189-220). London: Croon Helm. Hick, J. L., Hanfling, D., Burstein, J. L., DeAtley, C., Barbisch, D., Bogdan, G. M., & Cantrill, S. (2004). Health care facility and community strategies for patient care surge capacity. Annals of Emergency Medicine, 44(3), 253-261. IEEE-USA, (2005) Interoperability for the National Health Information Network, Whitepaper by the Medical Technology Policy Committee (MTPC) of The Institute of Electrical and Electronics EngineersUnited States of America (IEEE-USA), November. IT Governance Institute (ITGI). (2007). COBIT4.1-Control Objectives for Information and Related Technology. Retrieved January 20, 2010, from http://www.itgi.org/ Jacobson, M., Lindberg, J. E., Lindberg, R., Segerstad, C.H., Wallgren, P., Fellstrom, C., Hulten, C., & Jensen-Waern, M. (2001). Intestinal cannulation: model for study of the midgut of the pig. Comp Med., 51(2), 163–170. Kiel, L. D., & Elliott, E. (Eds.). (1996) Chaos theory in the social sciences: Foundations and applications. Ann Arbor, MI: University of Michigan Press.

xliv

Kim, Y.G., Valtorta, M., & Vomlel, J. (2004). A prototype system for soft evidential update. Applied Intelligence, 21(1), 81-97. Kouvelis, P., & Milner, J.M. (2002). Supply chain capacity and outsourcing decisions: The dynamic interplay of demand and supply uncertainty. IIE Transactions, 34, 717–728. Kollberg, B., Dahlgaard J. J., & Brehmer, P. (2007). Measuring lean initiatives in health care services: issues and findings. International Journal of Productivity and Performance Management, 56(1), 7-24. Kuljis, J., Paul, J., & Stergioulas, L. (2007). Can health care benefit from modeling and simulation methods in the same way as business and manufacturing has? In Proceedings of the 2007 Winter Simulation Conference (pp. 1449-1453). Kumar, A., & Ozdamar L. (2004). Business process reengineering at the hospitals: A case study at Singapore hospital. In Proceedings of the 18th European Simulation Multiconference (CD version). Kumar, A., Ozdamar L., & Zhang C. (2008). Supply chain redesign in the healthcare industry of Singapore. Supply Chain Management: An international Journal, 13(2), 95-103. Kumar, S., Swanson E., & Tran, T. (2009). RFID in the healthcare supply chain: usage and application. International Journal of Health Care Quality Assurance, 22(1), 67-81. Lambert, D.M., Cooper, M.C., & Pagh, J.D. (1998). Supply chain management: Implementation issues and research opportunities. The International Journal of Logistics Management, 9(2), 1–19. Laudon, K. and Laudon, J. (2007). Can information systems help prevent a public health crisis? In Management Information Systems (10th ed., pp. 77-78). Lee, H. L., & Whang, S. (2005). Higher supply chain security with lower cost: Lessons from total quality management. International Journal Production Economics, 96, 289-300. Li, X., & Chandra, C. (2006). Efficient knowledge integration to support complex supply network management. International Journal of Manufacturing Technology and Management, 10(1), 1-18. Li, X., & Ji, Q. (2005). Active affective state detection and user assistance with dynamic Bayesian networks. IEEE Transactions on Systems, Man, and Cybernetics-Part A, 35(1), 93- 105. Lipowicz, A. (2009, October 23). Government agencies to spend $15 billion on health IT by 2014. Federal Computer Week, Retrieved January 20, 2010, from http://fcw.com/Articles/2009/10/23/Governmentagencies-to-spend-15-billion-on-health-IT-by-2014.aspx?p=1 National Coalition on Health Care. (2009). Facts on Health Insurance Costs, Retrieved from http://www. nchc.org/facts/cost.shtml O’Leary, D. (2004). Models of infection: person to person, Computing in Science & Engineering, 6(1), 68-70. Pai, R.R., Kallepalli, V.R., Caudill, R.J., & Zhou, M. (2003). Methods toward supply chain risk analysis. In Proceedings of the IEEE International Conference on Systems, Man and Cybernetics, 5, 4560-4565.

xlv

Parisi, L., & Bailey, K. (2009, April). Too Great a Burden: Americans Face Rising Health Care Costs. Retrieved from http://hdl.handle.net/10207/18507. Parlar, M., & Perry, D. (1996). Inventory models of future supply uncertainty with single and multiple suppliers. Naval Research Logistics, 43, 191-210. Pearl, J. (1998). Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. San Mateo, CA: Morgan Kaufmann Publishers. Pierskalla, W.P., & Brailer, D.J. (1994). Applications of operations research in health care. Handbooks in OR & MS, 6, 469-505. Raghupathi, W., & Tan, J. (2002). Strategic IT Applications in Health Care. Communications of the ACM, 45(12), 56-61. Raguphati, W., & Tan, J. (2008). Information Systems and Healthcare XXX: Charting a Strategic Path for Health Information Technology. The Communications of the Association for Information Systems, 23(1), 500-525. Ramis, F., Palma, J., Estrada, V., & Coscolla, G. (2002). A simulator to improve patient’s service in a network of clinic laboratories. In Proceedings of the Winter Simulation Conference (Vol. 2, pp. 1444-1447). RAND Corporation (2004). Evaluating the Security of the Global Containerized Supply Chain, Technical Report, Santa Monica, CA Willis, H.H., & Ortiz, D.S. Rico, F., Salari, E., & Centeno, G. (2007). Emergency departments nurse allocation to face a pandemic influenza outbreak. In Proceedings of the Winter Simulation Conference (pp. 1292-1298). Roberts, S. D. (2007). Simulation of medical decisions, IERC 2007, Nashville, TN. Retrieved January 20, 2010, from http://74.125.95.132/search?q=cache:GYvFAgiR6rIJ:ccehub.org/resources/121/download/ steve-roberts-simulationofmedicaldecisions-crc.ppt+Simulation+of+medical+decisions,+IERC+2007& cd=1&hl=en&ct=clnk&client=safari Rohleder, R., Bischak, P. & Baskin, B. (2007). Modeling patient service centers with simulation and system dynamics, Health Care Management Science, 10(1), 1-12. Schnase, J. L., & Cunnius, E. L. (Eds.). (1995). Proceedings from CSCL ‘95: The First International Conference on Computer Support for Collaborative Learning. Mahwah, NJ: Erlbaum. Schriber T. J. & Brunner, D. T. (2007). Inside discrete-event simulation software: how it works and why it matters. In Proceedings of the Winter Simulation Conference (pp. 113-123). Shehad, M., Bertino, E., & Ghafoor, A. (2005). Secure collaboration in mediator-free environment. In Proceedings from CCS’05: ACM Conference on Computer and Communications Security, Alexandria, VA. Shubik, M. (2002). Game theory and operations research: Some musings 50 years later. Operations Research, 50, 192-196. Simchi-Levi, D., Kaminsky, P., & Simchi-Levi, E. (2003). Designing and Managing the Supply Chain: Concepts, Strategies, and Case Studies (2nd ed.). McGraw-Hill Irwin.

xlvi

Stahl, J., Roberts, M., & Gazelle, S. (2003). Optimizing management and financial performance of the teaching ambulatory care clinic. Journal Gen Intern Med, 18(4), 266-274. Swinehart, K. D., & Smith, A. E. (2005). Internal supply chain performance measurement: a health care continuous improvement implementation International Journal of Health Care, 18(7), 533-542. Tan, B., Gubaras, A., & Phojanamongkolkij, N. (2002). Simulation study of Dreyer urgent care facility. In Proceedings of the Winter Simulation Conference (Vol. 2, pp. 1922-1927. Tan, J., & Payton, F. C. (2010). Adaptive health management information systems: concepts, cases, & practical applications (3rd ed.). Sudbury, MA: Jones & Bartlett Publishers, Inc. Tan, J., Wen, J.H., & Awad, N. (2005, May). Health care and services delivery systems as complex adaptive systems. Communications of the ACM, 48(5), 36-44 Thomas, M.U. (2002). Supply chain reliability for contingency operations In Annual Proceedings of Annual Reliability and Maintainability Symposium (pp. 61-67). U.S. Bureau of Economic Analysis. (2009). Manufacturing Industries Led Slowdown in 2008. Retrieved from http://www.bea.gov/newsreleases/industry/gdpindustry/2009/gdpind08.htm. U.S. House of Representatives, (2007). The XDR tuberculosis incident: a poorly coordinated federal response to an incident with homeland security implications. Hearing of U.S. House of Representatives Homeland Security Committee, June 6, 2007. Woolhandler, S., Campbell, T., & Himmelstein, D. U. (2003). Costs of health care administration in the United States and Canada. The New England Journal of Medicine, 349(8), 768-775. Xiang, Y. (2002). Probabilistic Reasoning in Multiagent Systems: A Graphical Models Approach. Cambridge University Press. Young, T. (2005). An agenda for healthcare and information simulation. Health Care Management Science, 8, 189-196.

1

Chapter 1

Evaluating Health Information Services:

A Patient Perspective Analysis1 Umit Topacan Bogazici University, Turkey Nuri Basoglu Bogazici University, Turkey Tugrul U. Daim Portland State University, USA

aBstract The objective of the chapter is to explore the factors that affect users’ preferences in the health service selection process. In the study, 4 hypothetical health services were designed by randomly selecting levels of 16 attributes and these services was evaluated by the potential users. Analytical Hierarchy Process (AHP), one of the decision making methods, was used to assess and select the best alternative.

IntroductIon Healthcare service providers benefit from different technologies so as to reduce cost and improve quality of the medical procedures (Gagnona, Godinb, Gagnéb, Fortina, Lamothec, Reinharza, & Cloutierd, 2003). In particular, telemedicine resides on the center of these technologies. The American Telemedicine Association defines telemedicine as “the use of medical information exchanged from one site to another via electronic communications to improve patients’ health status” (ATA, 2009). Telemedicine applications were used in a broad

range including consultation (Berghout, Eminovic, de Keizer, & Birnie, 2007), education and training (Chen, Yang, & Tang, 2008), and home care (Biermann, Dietrich, Rihl, & Standl, 2002). Selecting the best telemedicine service among given ones is a complex task. The process needs considerations of trade-offs between cost and benefits of the service. Analytic Hierarchy Process (AHP) (Saaty, 1977; Saaty, 1996) is an outstanding method that can be used in multifactor decisionmaking environments. It presents a structured approach to determine individual weights of multiple attributes of a product or service so that

DOI: 10.4018/978-1-61692-002-9.ch001

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Evaluating Health Information Services

they can be compared in a simple way. Then, it simplify decision-making in the selection process. Recent technological developments enable advancements in delivery of medical services, appropriate healthcare at a reasonable cost, and access to quality healthcare in underserved areas in the medical sector. Telemedicine is one of these developments that “enable remote medical procedures and examinations between patients and medical providers via telecommunication technologies like the Internet, or telephone” (AlQirim, 2007). Moreover, many previous researches show that compared to traditional medical care, telemedicine services present many benefits to the patients and physicians (van den Brink, Moorman, de Boer, Pruyn, Verwoerd, & van Bemmel, 2005; Chae, Lee, Ho, Kim, Jun, & Won, 2001). Diffusion of intelligent monitoring systems in the medical industry has gather speed with the help of recent developments in the information and communication technologies. “Smart homes’ for telecare by means of movement detector, oxymeter, tansiometer and various other devices (Rialle, Lamy, Noury, & Bajolle, 2003), a ringsensor that monitors patient’s blood oxygen saturation (Yang & Rhee, 2000) and a web based electrocardiogram monitoring application facilitating collect, analyze and storage of patient data (Magrabi, Lovell, & Celle, 1999) were designed by researchers to follow-up of patients’ health status at home. Evaluating health technologies is a complicated procedure because people face with some difficulties while evaluating trade-offs between alternatives. Analytic Hierarchy Process (AHP) is a potential decision making method to deal with complex decisions. The aim of AHP is to qualify relative priorities for a given set of alternatives on a ratio scale. In the literature, many applications of AHP have published in different fields including planning, resource allocations, and selecting a best alternative, etc. (Magrabi, Lovell, & Celle, 1999). AHP method also widely used in vendor selection problems (Nydick & Hill, 1992; Tam & Tummala, 2001).

2

There are many applications of AHP in the medical field. In one of the researches, AHP is used to develop an human resource planning model for hospital laboratory personnel (Kwak, McCarthy, & Parker, 1997). Turri applied AHP approach to select a magnetic resonance imaging vendor by using the criteria like price, service, and technology (Turri, 1988). Another application of AHP was designed by Kahen and Sayers for selection of medical expert systems (Kahen & Sayers, 1997). In the study, assessment of a health service prototype have done in the design phase of the service development so as to make clear the factors affects users attitude toward health services. Usability tests of the health services (Kaufman, Patel, Hilliman, Morin, Pevzner, Weinstock, Goland, Shea, & Starrena, 2003), affects of ergonomics on the medical device development (Martin, J.L., Norris, Murphy, & Crowe, 2008) and self-measurement satisfaction of hypertension patients (Bobrie, Postel-Vinay, Delonca, & Corvol, 2007) are just some of them. These studies have applied different techniques in the analyses phase of the research. However, it is difficult to find a study that uses AHP method. Topacan et al (2008) identified 37 different criteria for health information service (HIS) adoption including cost, time factor, content, language, security, customizability, output quality, menu items, input type, sound quality and availability of face-to-face communication. The studies of Karahanna et al (1999), Simon et al (2007), Chang et al (2007) identified a number of criteria with respect to triability, medical provider and vendor support that are applicable for selecting health information service. Tung et al (2008) define financial cost as “the extent to which a person believes that using the electronic logistics information system will cost money”. It was found that financial cost has an influence on adoption of electronic logistic information system (Tung, Chang, & Chou, 2008). Chang et al also examined security protection and

Evaluating Health Information Services

vendor support while studying electronic signature adoption. They found that vendor support affects adoption, and there is no relation between security and e-signature adoption (Chang, Hwang, Hung, Lin, & Yen, 2007).

eValuatIon process

•

•

•

evaluation Model

•

Based on the previous studies, 21 different health service selection criteria were selected for the research. We conducted a mini survey involving randomly selected 3 male and 3 female potential users. The purpose of this mini survey is to assess and identify critical factors and reduce the number of criteria. In this phase, potential users ranked these criteria based on importance level of them. These participants also made suggestions about the levels of the attributes. Finally, 16 attributes were selected and used in the AHP model. Attributes used in the study is explained below;

•

•

•

•

•

•

Implementation Cost of the service is the purchase price of it. This price is paid only once while buying the service. Operating Cost is the price that was paid by the user in every month to be able to use the service. Usage Time is the amount of time required to use the service. 10 minutes usage time means that the user spends 10 minutes in a day while entering meal information and other data. Response Time is the doctor’s response time to the patients’ requests. As an example, in the alternative 1, the service provider guaranteed that doctor could respond the patients’ questions in 45 minutes. Language is the set of visual and auditory signs of communication.

•

•

• • •

•

Security means whether the service provides secure communication between patients and doctors, or not. Triability is the degree to which patient may be experimented with the service on a limited time or functionality. Sound Type is the voice style of the service. It can be one of computerized, male or female voices. Input Type refers the data entry procedure of the patient. Availability of Face-to-Face Communication means whether the service enable the patient to visit the doctor in the hospital or not. Technical Support is how to assist the user in order to solve the specific problems with the service. User Training means the demonstration the features and functionality of the service to the potential users. Medical Provider is the hospital that supports the medical related operations. Customizability is the ability of the service to be changed by the user preferences. Clarity of Menu Items is the degree to which the user may understand the function of the menu item. Output Quality refers how well the system performs the jobs (Venkatesh & David, 2000).

These attributes can be grouped into three major categories of cost, service characteristics and service provider characteristics. The cost factors include implementation and operating cost. The service characteristics consist of mobile device, triability, content, availability of face-toface communication and time. Similarly, service provider characteristics include technical provider and medical provider. Moreover, some of these sub criteria include one more level called sub criteria 2. Mobile device consists of input type, customizability, clarity of menu items and sound

3

Evaluating Health Information Services

Figure 1. AHP Model

type. Content attribute, on the other hand, includes output quality, language and security. Time factor contains usage time and response time. Technical provider consists of technical support and user training. The hierarchy can be visualized as a diagram shown in Figure 1. The diagram contains an overall goal and criteria broken down into two levels of sub-criteria. AHP model was applied on 4 different hypothetical home based telemonitoring services. Each of the services has 16 different attributes. Levels of these attributes were proposed by the potential users in the mini research phase of the study. Based on these levels, 4 different alternative services were designed by randomly selecting among levels. All of the alternatives and their levels are shown in Appendix A.

4

participants profile 6 male and 8 female potential users from different age groups were selected to conduct the research. Age of the participants was in a range between 23 and 60. Average age of the male and female group was 37. Table 1 contains details about the profile of participants.

FIndIngs Each of the participants have compared the attributes and their levels two by two in order to incorporate judgments about them. They responded 69 different pair wise comparison questions. Sample questions shown in Table 2 and Table 3. In these tables, scale number “9” refers that the attribute has extreme importance with respect to counter attribute. Intensity of importance decreases while

Evaluating Health Information Services

Table 1. Participants Profile Age Groups

Average Age

20-30

30-40

40-50

50+

Total

Male

2

2

1

1

6

36

Female

2

2

2

2

8

38

Total

4

4

3

3

14

37

Table 2. Sample Questionnaire Item for Attributes Implementation Cost

9

8

7

6

5

4

3

2

1

2

3

4

5

6

7

8

9

Operating Cost

Table 3. Sample Questionnaire Item for Attribute Levels Implementation Cost 200 YTL

9

8

7

6

5

4

3

2

1

2

3

4

5

6

7

8

9

500 YTL

200 YTL

9

8

7

6

5

4

3

2

1

2

3

4

5

6

7

8

9

800 YTL

500 YTL

9

8

7

6

5

4

3

2

1

2

3

4

5

6

7

8

9

800 YTL

Table 4. Sample Pair-wise Comparison Matrix for Cost Attribute Implementation Cost

Operating Cost

Weight

Implementation Cost

1

1/4

0.20

Operating Cost

4

1

0.80

moving down to number “2” in the scale. Moreover, scale number “1” refers that both of the attributes has equal importance for the participant. The following steps illustrate the AHP method applied in this research. 1.

2.

Participants responded questionnaire form. In the first part of the questionnaire (a sample pair-wise comparison question shown in Table 2), they compared attributes against each other to calculate weight of them. In the second part of the questionnaire (sample pair-wise comparison questions shown in Table 3), the participants compared level of attributes.

3.

4.

5.

Weight of attributes (Table 4) and priority vectors of alternatives (Table 5) was calculated for each of the participants separately. A table that shows the preferences of a single participant was produced for each of participants separately (Table 6). As an example, Table 6 shows that operating cost is the most important factor for the selected participant. In order to calculate total affects of 14 participants, average of the weight and priority vector values calculated in a separate table. Tables 7, 8 and 9 shows the overall results that is the average values of 14 participants.

5

Evaluating Health Information Services

Table 5. Sample Pair-wise Comparison Matrix for Implementation Cost Levels A1 (200 YTL)

A1 (200 YTL)

A2 (500 YTL)

A3 (800 YTL)

A4 (500 YTL)

Priority Vector

1

4

8

4

0.57

A2 (500 YTL)

1/4

1

6

1

0.19

A3 (800 YTL)

1/8

1/6

1

1/6

0.04

A4 (500 YTL)

1/4

1

6

1

0.19

Table 6. Selection Matrix of a Participant Attribute

Weight

A1

A2

A3

A4

Implementation Cost

0.040

0.573

0.191

0.045

0.191

Operating Cost

0.161

0.050

0.138

0.138

0.673

Usage Time

0.020

0.654

0.062

0.142

0.142

Response Time

0.156

0.046

0.121

0.713

0.121

Language

0.014

0.658

0.106

0.118

0.118

Security

0.099

0.250

0.250

0.250

0.250

Triability

0.033

0.188

0.154

0.188

0.470

Sound Type

0.007

0.250

0.250

0.250

0.250

Input Type

0.087

0.538

0.174

0.115

0.174

Availability of Face-to-Face Comm

0.130

0.375

0.125

0.125

0.375

Technical Support

0.013

0.713

0.121

0.121

0.046

User Training

0.003

0.228

0.291

0.384

0.097

Hospital

0.076

0.429

0.071

0.071

0.429

Customizability

0.043

0.313

0.313

0.063

0.313

Clarity of Menu Items

0.014

0.232

0.117

0.418

0.234

Output Quality

0.104

0.083

0.417

0.083

0.417

Sum

1.000

0.246

0.185

0.226

0.343

In Table 7, the local weights represent the relative weight of the nodes within a group of siblings regarding their parent. The global weights are calculated by multiplying the local weights of the siblings by their parents’ local weights. It can be seen from the table that local weights of each group add up to 1 and the global weights of all the 16 attributes add up 1. According to final calculations, Table 7 shows that potential users pay more attention on the characteristics of the service like face-to-face communication, content, and time in the health service selection process. Moreover, operating cost

6

of the service is more important than purchasing cost. Also, content and availability of face-toface communication are two of the significant service characteristics. Medical provider affects decision of potential user more than technical provider. Clarity of menu items, security and response time are other important factors. And, as a characteristic of the technical provider, users attach slightly more importance to user training than technical support. Compared to usage time, short response time has more positive effects on patients selection decision of HIS.

Evaluating Health Information Services

Table 7. Overall Local and Global Weights of Attributes Criteria

Local Weight

Sub Criteria 1

Local Weight

Service Chars

0.39

Face to Face Comm.

0.27

Content

0.26

Ser. Prov. Chars

Cost

0.37

0.24

Time

0.22

Triability

0.15

Mobile Device

0.10

Medical Provider

0.78

Technical Provider

0.22

Local Weight

Global Weight 0.105

Security

0.48

0.049

Output Quality

0.35

0.035

Language

0.17

0.017

Usage Time

0.23

0.020

Response Time

0.77

0.066 0.058

Clarity of Menu Item

0.40

0.016

Customizability

0.28

0.011

Input Type

0.25

0.010

Sound Type

0.07

0.003 0.289

Technical Support

0.57

0.046

User Training

0.43

0.035

Operating Costs

0.65

0.156

Implementation Costs

0.35

0.084

Table 8. Overall Global Weights of Attributes Attribute

Sub Criteria 2

Weight

Medical Provider

0.289

Operating Cost

0.156

Face-to-Face Communication

0.105

Implementation Cost

0.084

Response Time

0.066

Trialability

0.058

Security

0.049

Technical Support

0.046

Output Quality

0.035

User Training

0.035

Usage Time

0.020

Language

0.017

Clarity of Menu Items

0.016

Customizability

0.011

Input Type

0.010

Sound Type

0.003

Sum

1.000

Table 8 presents weight of the 16 attributes in the decreasing order. According to Table 8, medical provider is the most important factor for health service selection. Secondly, people give importance on the operating cost of the medical service. Thirdly, they choose services that make available face-to-face communication with the physician. On the other hand, input type, customizability and sound type was found three of least important attributes according to overall values. Table 9 presents overall priority vector values of alternatives. According to the table, people significantly choose to pay least amount of money and not spare much usage time for the health information services. Moreover, alternatives that provide secure data communication and storage environment, offer one week unlimited trial period, and include seven days twenty-four hours technical support were highly preferred by the participants. People prefer consulting about their health status to private hospitals instead of public ones and

7

Evaluating Health Information Services

Table 9. Overall Priority Vectors of Alternatives Attribute

A1

A2

A3

A4

Implementation Cost

0,60

0,17

0,06

0,17

Operating Cost

0,05

0,17

0,17

0,61

Usage Time

0,62

0,06

0,16

0,16

Response Time

0,09

0,16

0,58

0,16

Language

0,31

0,10

0,30

0,30

Security

0,41

0,09

0,41

0,09

Triability

0,07

0,25

0,07

0,61

Sound Type

0,19

0,25

0,37

0,19

Input Type

0,29

0,12

0,47

0,12

Availability of Face-to-Face Comm

0,41

0,09

0,09

0,41

Technical Support

0,66

0,15

0,15

0,05

User Training

0,21

0,46

0,28

0,05

Hospital

0,36

0,14

0,14

0,36

Customizability

0,36

0,22

0,06

0,36

Clarity of Menu Items

0,23

0,16

0,23

0,38

Output Quality

0,12

0,38

0,12

0,38

receiving guidance conducted by professionals instead of online education and manual. Table 10 shows the total decision weights of the alternatives. Total decision weight of a specific alternative was calculated by multiplying weight of the attribute with overall priority vector value of the attribute for that alternative and summed up these values calculated for 16 attributes. According to the table, people have a tendency on selecting alternative 4 which is followed by alternative 1. Alternatives 2 and 3 are the least preferred services by the potential users. Total Decision Weight = Σ Global Weight of the Attribute* Priority Vector of the Attribute Alternative 4 has the finest attribute values for the most important three attributes listed in Table 8. Medical provider of the alternative is a private hospital. Although its implementation cost is higher than alternative 1 and equals to alternative 2, it has cheapest operating cost that has valuable

8

impact on potential users’ health service selection decisions. Moreover, it provides patients to face-to-face communication with the physician.

conclusIon As explained in section 1, electronic health information service selection is an important and difficult problem to a medical company and patients. We first identified 16 criteria and then formulated an AHP-based model to select health information service as shown in Appendix B. As for HIS selection, all potential users agreed that a service characteristic was the most important factor. Service providers and cost followed as the second and third most important consideration. The availability of face-to-face communication of the HIS was ranked first among service factors, followed by content, time, triability and mobile device. Moreover, designers should pay more attention on medical provider and cost of the service while developing a telemedicine service. Instead of public hospital, private hospital should be selected as a medical provider, because people take into confidence in private hospitals. Also, patients prefer the services that enable doctor visits. The most important attributes, found in the research, enable the service designers and providers to know the characteristics of the most preferred e-health service. Furthermore, governments, especially in developing countries, face with some difficulties while providing quality healthcare in underserved and rural areas. Even if they achieve to serve, they suffer from high cost of the health service. With the rapid development of information and communication technologies, they benefit from internet to serve the healthcare service. But, at this time user adoption problems becomes one of the obstacles in spreading of the electronic health services. The proposed AHP framework will help the authorities to overcome adoption problems so as to reduce health care

Evaluating Health Information Services

Table 10. Final Decision Matrix Total Decision Weights

costs, to improve health care and access to health care, to improve patience obedience. For a future study, a comprehensive qualitative research with large sample size can be conducted to test validity of the results. Moreover, attribute values of proposed alternatives can be changed and examined the affect of them on the health service selection process.

reFerences Al-Qirim, N. (2007). Championing telemedicine adoption and utilization healthcare organizations in New Zealand. International Journal of Medical Informatics, 76, 42–54. doi:10.1016/j. ijmedinf.2006.02.001 ATA. (2009). Americal Telemedicine Association. Retrieved from http://www.atmeda.org Berghout, B. M., Eminovic, N., de Keizer, N. F., & Birnie, E. (2007). Evaluation of general practitioner’s time investment during a store-and-forward teledermatology consultation. International Journal of Medical Informatics, 76, 384–391. doi:10.1016/j.ijmedinf.2007.04.004 Biermann, E., Dietrich, W., Rihl, J., & Standl, E. (2002). Are there time and cost savings by using telemanagement for patients on intensified insulin therapy: A randomised, controlled trial. Computer Methods and Programs in Biomedicine, 69, 137–146. doi:10.1016/S0169-2607(02)00037-8 Bobrie, G., Postel-Vinay, N., Delonca, J., & Corvol, P. (2007). Self-Measurement and SelfTitration in Hypertension. American Journal of Hypertension, 20, 1314–1320. doi:10.1016/j. amjhyper.2007.08.011

A1

A2

A3

A4

0,30

0,17

0,18

0,35

Chae, Y. M., Lee, J. H., Ho, S. H., Kim, H. J., Jun, K. H., & Won, J. U. (2001). Patient satisfaction with telemedicine in home health services for the elderly. International Journal of Medical Informatics, 61, 167–173. doi:10.1016/S13865056(01)00139-3 Chang, I. C., Hwang, H. G., Hung, M. C., Lin, M. H., & Yen, D. C. (2007). Factors affecting the adoption of electronic signature: Executives’ perspective of hospital information department. Decision Support Systems, 44(1), 350–359. doi:10.1016/j.dss.2007.04.006 Chen, I. J., Yang, K. F., & Tang, F. I. (2008). Applying the technology acceptance model to explore public health nurses’intentions towards web-based learning: A cross-sectional questionnaire survey. International Journal of Nursing Studies, 45, 869–878. doi:10.1016/j.ijnurstu.2006.11.011 Gagnona, M. P., Godinb, G., Gagnéb, C., Fortina, J. P., Lamothec, L., Reinharza, D., & Cloutierd, A. (2003). An adaptation of the theory of interpersonal behaviour to the study of telemedicine adoption by physicians. International Journal of Medical Informatics, 71, 103–115. doi:10.1016/ S1386-5056(03)00094-7 Kahen, G., & Sayers, B. M. (1997). Health-care technology transfer: Expert and information systems for developing countries. Methods of Information in Medicine, 36, 69–78. Karahanna, E., Straub, D. W., & Chervany, N. L. (1999). Information technology adoption across time: a cross-sectional comparison of preadoption and post-adoption beliefs. Management Information Systems Quarterly, 23, 183–213. doi:10.2307/249751 9

Evaluating Health Information Services

Kaufman, D. R., Patel, V. L., Hilliman, C., Morin, P. C., Pevzner, J., & Weinstock, R. S. (2003). Usability in the real world: assessing medical information technologies in patients’ homes. Journal of Biomedical Informatics, 36, 45–60. doi:10.1016/S1532-0464(03)00056-X

Simon, J. S., Rundall, T. G., & Shortell, S. M. (2007). Adoption of Order Entry with Decision Support for Chronic Care by Physician Organizations. Journal of the American Medical Informatics Association, 14, 432–439. doi:10.1197/jamia. M2271

Kwak, N. K., McCarthy, K. J., & Parker, G. E. (1997). A human resource planning model for hospital/medical technologists: An analytic hierarchy process approach. Journal of Medical Systems, 21(3), 173–187. doi:10.1023/A:1022812322966

Tam, M. C. Y., & Tummala, V. M. R. (2001). An Application of the AHP in vendor selection of a telecommunications system. Omega, 29(2), 171–182. doi:10.1016/S0305-0483(00)00039-6

Magrabi, F., Lovell, N. H., & Celle, B. G. (1999). A web-based approach for electrocardiogram monitoring in the home. International Journal of Medical Informatics, 54, 1145–1153. doi:10.1016/ S1386-5056(98)00177-4 Martin, J. L., Norris, B. J., Murphy, E., & Crowe, J. A. (2008). Medical device development: The challenge for ergonomics. Applied Ergonomics, 39, 271–283. doi:10.1016/j.apergo.2007.10.002 Nydick, R. L., & Hill, R. P. (1992). Using the analytic hierarchy process to structure the supplier selection procedure. Journal of Purchasing and Materials Management, 25(2), 31–36. Rialle, V., Lamy, J. B., Noury, N., & Bajolle, L. (2003). Telemonitoring of patients at home: a software agent approach. Computer Methods and Programs in Biomedicine, 72, 257–268. doi:10.1016/S0169-2607(02)00161-X Saaty, T. L. (1977). A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology, 15, 234–281. doi:10.1016/00222496(77)90033-5 Saaty, T. L. (1996). The analytic hierarchy process. Pittsburgh: RWS Publications.

10

Topacan, U., Basoglu, A. N., & Daim, T. U. “Exploring the Success Factors of Health Information Service Adoption. In Proceedings of Portland International Conference for Management of Engineering and Technology’08 (pp. 2453-2461). Tung, F. C., Chang, S. C., & Chou, C. M. (2008). An extension of trust and TAM model with IDT in the adoption of the electronic logistics information system in HIS in the medical industry. International Journal of Medical Informatics, 77, 324–335. doi:10.1016/j.ijmedinf.2007.06.006 Turri, J. J. (1988). Program eases decision making. Health Progress (Saint Louis, Mo.), 69, 40–44. Vaidya, O. S., & Kumar, S. (2006). Analytic hierarchy process: An overview of applications. European Journal of Operational Research, 169, 1–29. doi:10.1016/j.ejor.2004.04.028 van den Brink, J. L., Moorman, P. W., de Boer, M. F., Pruyn, J. F. A., Verwoerd, C. D. A., & van Bemmel, J. H. (2005). Involving the patient: A prospective study on use, appreciation and effectiveness of an information system in head and neck cancer care. International Journal of Medical Informatics, 74, 839–849. doi:10.1016/j. ijmedinf.2005.03.021

Evaluating Health Information Services

Venkatesh, V., & David, F. D. (2000). A theoretical extension of the technology acceptance model- Four longitudinal field studies. Management Science, 46, 186–204. doi:10.1287/ mnsc.46.2.186.11926 Yang, B. H., & Rhee, S. (2000). Development of the ring sensor for healthcare automation. Robotics and Autonomous Systems, 30, 273–281. doi:10.1016/S0921-8890(99)00092-5

endnote 1

This chapter is based on a paper presented at Portland International Conference on Management of Engineering and Technology 2009 in Portland Oregon USA by the same authors

11

Evaluating Health Information Services

appendIx a Table 11. Proposed Health Service Alternatives Attribute

Alternative 1

Alternative 2

Alternative 3

Alternative 4

Implementation Cost

200 YTL

500 YTL

800 YTL

500 YTL

Operating Cost

100 YTL / month

50 YTL / month

50 YTL / month

25 YTL / month

Usage Time

10 min / day

50 min / day

30 min / day

30 min / day

Response Time

45 min

15 min

5 min

15 min

Language

Turkish

English

Turkish / English

Turkish / English

Security

Available

Not Available

Available

Not Available

Triability

Not Available

1 week limited functionality

Not Available

1 week unlimited functionality

Sound Type

Computerized

Male

Female

Computerized

Input Type

Sound

Text

Selection among alternatives

Text

Availability of FacetoFace Communcation

Available

Not Available

Not Available

Available

Technical Support

7 / 24

09:00 – 17:00

09:00 – 17:00

Not Available

User Training

Operating manual is given to the user

Training is conducted by a professional

Online Education

Not Available

Medical Provider

Private Hospital

Public Hospital

Public Hospital

Private Hospital

Customizability

Frequently used menu items are automatically shown on top of the menu

Users are selecting frequently used menu items manually

Menu items are shown in the same order

Frequently used menu items are automatically shown on top of the menu

Clarity of Menu Items

Service does not contain menu

Textual menu items

Graphical menu items

Both of the graphics and text are used in the menu items

Output Quality

Meal list is given as a output

User preference are taken into consideration in the meal list

Meal list is given as a output

User preference are taken into consideration in the meal list

12

Evaluating Health Information Services

appendIx B Figure 2. AHP Model

13

14

Chapter 2

Gastrointestinal Motility Online Educational Endeavor Shiu-chung Au State University of New York Upstate Medical University, USA Amar Gupta University of Arizona, USA

aBstract Medical information has been traditionally maintained in books, journals, and specialty periodicals. A growing subset of patients and caregivers are now turning to diverse sources on the internet to retrieve healthcare related information. The next area of growth will be sites that serve specialty fields of medicine, characterized by high quality of data culled from scholarly publications and operated by eminent domain specialists. One such site being developed for the field of Gastrointestinal Motility provides authoritative and current information to a diverse user base that includes patients and student doctors. Gastrointestinal Motility Online leverages the strengths of online textbooks, which have a high degree of organization, in conjunction with the strengths of online journal collections, which are more comprehensive and focused. Gastrointestinal Motility Online also utilizes existing Web technologies such as Wiki-editing and Amazon-style commenting, to automatically assemble information from heterogeneous data sources.

IntroductIon For the last several decades, Harrison’s Principles of Internal Medicine, published by McGraw Hill, has served as a major source of information in the field of Gastrointestinal Motility. This book and its online presentation have been, and continue to be, used by many medical colleges to train the

next generation of medical doctors; practitioners in this field also frequently refer to them. Traditionally, papers and articles in specialty medical journals supplemented the material in textbooks like Harrison. The latter book would itself be updated periodically to reflect the state of the art in medicine and the various specialties, providing a consensus opinion of the standard of care.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Gastrointestinal Motility Online Educational Endeavor

The advent of computers and Internet has given rise to online sources of information such as UpToDate (http://www.uptodate.com/) and WebMD (http://www.webmd.com/). While gaining tremendous following and being updated frequently, these sources of online information relate to the medical field as a whole and not to particular specialties. Furthermore, the information on these sites is generally maintained by personnel of the respective organizations, not by specialists in specific disciplines of medical science. These organizations are usually set up as commercial entities, rather than not-for-profit ones. The progressive transformation of information has seen many journals that were previously in paper format opting to use new electronic technologies; most of them now come out both in paper and electronic formats. Searchable electronic archives, such as PubMed (http://www. pubmedcentral.nih.gov/), now place a plethora of information into the hands of researchers and physicians. However, such searches are very time consuming and often produce irrelevant or poorly supported articles. Sites like Harrison’s Online (http://www.accessmedicine.com/) serve as information directories that can be searched, hoping to place the most suitable information on a medical topic in a user’s hand. Students have gradually come to expect information in quick and readily available forms without having to bother about inter-library loans or even hardcopy versions at all. The goal of the endeavor described in this article was to adapt emerging technologies to improve methods of teaching gastrointestinal material to students and to serve as a more effective source of relevant and accurate information for medical practitioners and specialists.

evidence-Based Medicine A study from the School of Information Management and Systems at UC Berkeley estimates that, in 2003, the World Wide Web contained

about 170 terabytes of information on its surface alone, equivalent to seventeen times the size of the information in the Library of Congress (Lyman & Varian, 2003). With this increasingly information-rich society, the most precious ability for students and learners is no longer to find the information, but to discern the most relevant pieces of information and to integrate them into practice. The American Library Association describes “information literacy” as the ability of individuals to “recognize when information is needed and have the ability to locate, evaluate, and use effectively the needed information” (American Library Association, 1989). The medical domain version of information literacy is evidence-based medicine. Evidence-based medicine (EBM) is the integration of best research evidence with clinical expertise and patient values (Guyatt et al., 1992). The Centre For Evidence-Based Medicine in Toronto, Canada, states that the origins of evidencebased medicine date back to post-revolution Paris (CEBM, 2007), but that the current growth is most closely attributed to the work of a group lead by Gordon Guyatt at McMaster University in Canada in 1992. EBM publications, reflecting interest in this field, have grown from a lone publication in 1992 to thousands in 2007. Studies have become increasingly critical of the value of textbook sources (Antman et al., 1992). Didactic continuing medical information may be ineffective at changing physician performance (Davis et al., 1997), and clinical journals may lack practical application (Haynes, 1993). In addition, physicians are faced with an increasing burden on their time, forced to diagnose patient findings within a matter of minutes (Sackett & Straus, 1998), and can only afford to set aside half an hour or less per week for general medical reading (Sackett, 1997). The staggering mass of information being discovered is also daunting: 500,000 articles are added to the commonly used Medline medical journal database every year, and “if a physician read 2 articles each day, every day

15

Gastrointestinal Motility Online Educational Endeavor

for a year, (s)he would still find herself or himself 648 years behind” (Lindberg, 2003). As research increases the quantity of information available, medical practitioners are compelled to find efficient methods to educate themselves. The Centre for Evidence Based Medicine has cited several examples of strategic, educational, and technical improvements in medicine that have enabled the current explosion in interest in this field. These include the emergence of new strategies for evaluating information; the creation of systematic reviews; the growing emphasis on continuing medical education and lifetime learning; and the advent of online journals, meta analysis of multiple studies, and ready access to such resources through electronic archives. Rapid dissemination of accurate and comprehensive compilations of research results enables medical practitioners to make informed decisions that are supported by the latest research results, and not by outdated trials. In light of the increasing number of medical journals, especially journals that focus on specialties, the sheer quantity of information threatens to overwhelm medical practitioners. The concept of information mastery has been coined to describe the set of skills that physicians must nurture in approaching, analyzing and incorporating or rejecting new information. The issues mentioned above are not limited to the medical arena alone: Former Vice-President Al Gore described the state of information management as “resembling the worst aspects of our agricultural policy, which left grain rotting in thousands of storage files while people were starving” (Gore, 1992). The Center for Information Mastery at the University of Virginia asserts that the usefulness of medical information is dependent upon

its relevance, validity, and the work required to obtain it, as specified in Equation 1 (Slawson et al., 2007). Further, the increasing quantity of research being performed by commercial enterprises, as well as other organizations with potential conflicts of interest, requires information be filtered for validity before incorporation into medical the canon. Finally, increased effort involved in accessing the relevant pieces of information reduces the accessibility of such information. In addition, healthcare organizations are siloed, gaining the advantage of sub-optimizing local departments, possibly at the cost of the whole (Senge, 1980); this complicates the problem further.

Information retrieval and decision-Making As Stephen Hawking observes in The Universe in a Nutshell, the rate of growth of new knowledge is exponential. While 9,000 articles were published annually in 1900 and around 90,000 in 1950, there were 900,000 scientific articles published per annum in 2000 (Hawking, 2001). The explosive growth of information is challenging both the information repositories designed to hold it, and the ability of users to access relevant information. At the time of this writing, Wikipedia serves as the de-facto standard for online general encyclopedias, and is among the top ten most-visited Web sites (Alexa Internet, 2007a). Its open-source, volunteer-without-accountability approach led to initial concerns about information validity, but these concerns have been largely addressed. Nature magazine studied Wikipedia and Encyclopedia Brittanica and found that the two of them

Equation 1. Usefulness (Slawson et al., 2007)

Usefulness of Medical Information =

16

(Relevance)(Validity) Work

Gastrointestinal Motility Online Educational Endeavor

were largely similar in accuracy (Giles, 2005). The growth of Wikipedia’s information base further enhances the quality and breadth of coverage, and supports the possible future use of Wikipedia or Wiki-style architecture as an academically respectable source of reference. In contrast to the indexed and contributed semi-structured format of Wikipedia, Google relies on search-keyword phrases. The usefulness of Internet-crawling indexers, like Google, is based upon the ability to retrieve and capture information from many sites, and to retrieve relevant pages on query. Google’s initial strength and rise to stardom was achieved through its superior PageRank algorithm, which still remains a carefully guarded trade secret; this algorithm provides an uncannily relevant list of matches to any user query, ranging from commonplace query phrases to obscure esoteric trivia and even misspellings. In a small supermarket today, shoppers are bombarded with a selection of 285 varieties of cookies and 95 varieties of chips, leading the consumer to a state of decision overload (Schwartz, 2004). There is a growing need to restructure data to meet the informational and management requirements of an organization or group of people (Carlson, 2003).

Medical Information repositories For the medical arena, PubMed is the most widely used information database in the world, accounting for 1.3 million daily queries by 220,000 unique users (Lindberg, 2003). It is a free access search engine, provided by the U.S. National Library of Medicine as the main access point to the Medline database, a cataloged repository of medical

literature classified using the descriptors known as Medical Subject Headings (MeSH). A broad range of search features are offered, including combined searches, exclusions, classification by type of article (original research versus review) and related articles. Another feature of the Medline database, known as MedlinePLUS, provides generalized information on health topics, and is aimed at the public or at practitioners outside their specialty domain. At the current time, PubMed serves as the gold standard in comprehensive medical information, despite its dated interface. HighWire Press, a Stanford-originated endeavor, distributes thousands of journals, and provides its own search engine. In a recent study, the relevance of articles retrieved from HighWire was found to be greater than that of PubMed, but with the disadvantage of a slower retrieval Speed (Vanhecke et al., 2006). In order to make a comparative evaluation between different approaches, it is appropriate to characterize the information recall ability using three parameters: precision, recall, and fall-out. Precision (Equation 2) can be defined as the proportion of all retrieved documents that are relevant. Recall (Equation 3) captures the concept of complete retrieval of all relevant documents. Fall-out (Equation 4) is a measure of the number of documents that are retrieved but are unrelated to the issue being searched. Medical research, while generally emphasizing maximal precision and minimal fall-out, occasionally requires increased recall, in the case of obscure diseases, or unusual side effects of medications. Medline serves as a canonical list for such purposes, but at the cost of significantly lower precision.

Equation 2. Precision (Wikipedia, 2007) precision =

{relevant documents}∩ {retrieved documents} {retrieved documents}

17

Gastrointestinal Motility Online Educational Endeavor

Equation 3. Recall (Wikipedia, 2007)

recall =

{relevant documents}∩ {retrieved documents} {relevant documents}

Equation 4. Fall-out (Wikipedia, 2007) fall-out =

{non-relevant documents}∩ {retrieved documents} {non-relevant documents}

Medical practitioners using Google for their searches will often find themselves frustrated at the large quantity of articles on obscure and irrelevant topics. A researcher searching for a pharmacologic treatment of a syndrome will turn up with thousands of articles dealing with various sub-types, biochemical-signaling processes involved, and even support groups, before finding a therapeutic treatment. Due to the nature of the search engine and the storage methods, there are concerns about Google’s or any search engine’s ability to maintain a collection of such information. Carlson (2003) showed that due to the relatively small collection of documents indexed by an average search engine, a significant amount of relevant information would not be returned even in the presence of a perfectly formulated search phrase. In order to accommodate domain-specific areas, Google has introduced the concept of “Refine Your Search” (http://www.google.com/coop). Without altering its main core search methodology, Google allows users to more quickly locate the type of information desired (i.e., treatment or symptoms). These refinement tools, provided by vendors and other private individuals or agencies that are deemed authoritative, subsequently label Web sites with appropriate descriptor tags. The potential conflict of interest created by these corporate associations is a matter of concern, due to omissions or maliciousness of the labeling.

18

The range of challenges and issues that characterize the medical domain include: •

•

•

The presence of an extensive array of synonyms for various drugs and diseases that require semantic knowledge to be encoded into the search engine in order to link concepts that are not lexically related; The naming of disease subtypes (often after a major contributor or discoverer), requires that hierarchies be constructed to allow users looking for the subclasses of disease to find information on the main umbrella disease, and vice versa; The growth in the understanding of generalized syndromes results in a corresponding need for reclassification based on new etiologies of disease, thereby suggesting a dynamic organizational structure for the online medical information systems.

In view of the growing difficulty in locating desired pieces of information, individuals performing research are in increasing danger of information overload. As such, the next generation of medical information access tools must aim to improve the ability to retrieve the right chunks of information quickly, with zero or minimal extraneous information; this concept is termed as increasing the signal to noise ratio in the field of electrical engineering.

Gastrointestinal Motility Online Educational Endeavor

VIsIon and goals For gastroIntestInal MotIlIty onlIne Gastrointestinal (GI) Motility Online is an example of a medical information system that seeks to provide access to high quality medical information online related to a particular medical specialty by centralizing the information and presenting relevant information that is customized to the user’s information requirements. The field of Gastrointestinal Motility is complex and interdisciplinary, involving a variety of experts. The possible user base includes laypersons, patients, medical students, biomedical scientists, physiologists, pathologists, pharmacologists, biomedical students, researchers, pharmaceutical staff, house staff, specialty fellows, internists, surgeons, and gastroenterologists. Each role requires a different approach to depth, scholastic relevance, and clinical direction in terms of information presentation. For example, students are interested in innovative research or review papers; researchers would like to know the most recent developments, and practitioners might be more interested in using the information for differential diagnosis purposes. GI Motility aims to serve as a collaboration of medical professionals, approaching diseases and patients from different angles. In a library, a user interacts with the data in books very differently from the way that she or he interacts with data in an online presentation. The user expects the book to be focused and to address the topic in a linear fashion. Online, the same user navigates quickly, using hyperlinks, to explore secondary topics. In fact, the user expects a different presentation and a different style of information; as a result, the nature of the interactions will differ even with the same content. One of the aims of Gastrointestinal Motility Online is to address these different styles and to present the desired subset of information in the manner that the user might expect. For example, a user

might be interested in viewing articles from the perspective of case-based, symptom-based, or test result-based diagnosis in order to apply the information to a particular problem at hand. In essence, the vision of Gastrointestinal Motility Online is to present information as framed by the interaction with the particular user at the particular point in time. The first step in the vision of Gastrointestinal Motility Online was to collect the information in a manner consistent with the goal to acquire the reputation for the highest quality of knowledge. The information base is assembled entirely from material provided by internationally acknowledged experts. All chapters including synopses, articles, and reviews are written by reputed authorities. The pool of information is envisaged to be shared between different types of users and for different purposes. The design of the system emphasizes a one-stop information approach that enables the users to derive information at various depths. This applies to onsite information, as well as to information at offsite locations.

details of effort A two-phase approach is being utilized for the creation of the information system: the first involves full leveraging of commercial technology as it exists today, and the second involves further research on aspects that can be incorporated in future versions of our system. In the absence of a better term, the term gastrointestinal knowledge repository is used for the final system, as well as for the initial concept-demonstration prototype system. For the first phase, the acquisition of knowledge proceeded with the establishment of titles and themes for chapters, as determined through discussions involving the concerned authors and the editors for this project (Dr. Raj Goyal of Harvard Medical School and Dr. Reza Shaker of the Medical College of Wisconsin). The creation of the gastrointestinal motility knowledge repository

19

Gastrointestinal Motility Online Educational Endeavor

began with calls to key gastrointestinal experts inviting them to submit a chapter, in electronic form, for inclusion in this knowledge repository. The inputs from the contributing authors were reviewed by these two editors and by others, on an anonymous peer review basis. Under the aegis of an unrestricted grant from Novartis Corporation, the two editors worked closely with the staff of Nature Publishing Group on a number of tasks that ultimately led to the creation of the following Web site: http://gimotilityonline.com. The creation of the site involved an automated conversion process to adapt MS-Word and rich text documents into a Web presentable format, with special emphasis placed on images, tables, and video. The majority of the investment for development lay in formatting and typesetting; the design itself was of less concern as it followed existing Web branding and style guidelines of Nature Publishing Group. Rights for images needed to be obtained; further, images, tables and video needed to be edited to fit a standard look and feel. After receiving author contributions, the project required approximately 18-24 months to complete, with 9-12 months required for the editorial process itself. The current site consists of 1,000 HTML pages, 1,000 images, 500 Powerpoint presentations and 40 videos. Articles are cross-linked by topics, and the current volume is equivalent of about 700 pages of text. This volume is increasing as this endeavor continues to progress.

site purpose The ostensible purpose of the site is to disseminate information on the specialty of gastrointestinal medicine, primarily to gastrointestinal medical practitioners and to other interested parties. The ambition for the site is for it to become the central hub of information on this specialty, aggregating information from many sources, authors, and journals into one central location with the objective of becoming a de facto standard for online gastrointestinal information.

20

A secondary role is the development of a community of gastrointestinal specialists and other interested parties, who can form an online collaborative, expand upon the information repository, and facilitate peer communications and discussions. In addition, the site aims to explore and to expand the concept of online information repositories, especially the optimal integration of the current generation of electronic journals and online textbooks.

site audience The primary site audience is the set of gastrointestinal specialists, associated medical staff, and staff under training. Members of this primary user base would initially visit the site because of a recommendation from a colleague or because another site (search or advertisement) directed the user to it. Occasional patients are expected, but are unlikely to become the primary users of this site. Initially, the core attractions for users to this site are the quality and depth of information coupled with the ease of access and the low cost; these same factors will also help to retain the user base. Most physicians and healthcare specialists spend relatively little time online (six hours per week on average); in order to impress the user base, the signal to noise ratio of the site must be high, along with the ease of locating a desired piece of information (through the information architecture) (Friedman, 2000). The primary user of the site is unlikely to demand high interactivity; instead, the interest will be on locating and extracting the information as quickly as possible. As such, the site must be efficiently organized, streamlined, and equipped with powerful search and index functions. After drawing a user initially, repeat visitors would use the site to browse new topics of interest, as well as to entrust the editorial staff to select articles that represent innovative research in the relevant field. An interacting community base would evolve over time and eventually cause a

Gastrointestinal Motility Online Educational Endeavor

change in the workflow. Specialists would then visit the site to explore the comments from their colleagues on the topics expressed and to leave their own authoritative inputs on different articles, eventually contributing entire articles as much as a good electronic journal aspires to do today. While relatively low in terms of being technically savvy, a typical user of Gastrointestinal Motility is likely to be very comfortable with using the Internet for retrieval of scholarly information through PubMed, online textbooks, and ready references such as UpToDate. A sleek, uncluttered design is likely to be the most attractive, even though it may not support applications involving high load times, such as Flash or embedded videos. Users are comfortable reading large articles online, but they also expect printable versions to be available, on an as-needed basis. The user base, while highly intelligent, is relatively small in numeric terms, and is characterized by a small presence on the Web. The process of attracting a critical mass to build and to maintain a user community is a high priority task; this involves contacting a high percentage of all members of this specialist community.

site content Gastrointestinal Motility Online is a site that provides information on both medical practice and the fundamentals of the gastrointestinal tract. The information must provide both breadth and depth on the topic, and should ultimately serve as an encyclopedia of the domain-specific knowledge. Information Architecture: As in a library, information must be properly accessible in order to have value. The architecture of the information is partly determined by the methods that the users will use to query the evolving knowledge base. In a library, the name of the author and the title of the book are important. In a journal, the age of the article or the issue in which it was published may be essential. In Gastrointestinal Motility Online, the most likely user scenarios involve searching for

information by anatomical section or syndrome. The timing is also important; recent material is favored over older material. Since relevance of the article is also important, the name of the journal and the name of the author are also used prominently in searches. Furthermore, classification schemes or hierarchies to group-related topics are essential to narrow the branching process used by the search technique. The determination of related information is a complex topic. In order to address the latter issue, medical organizations are creating medical ontologies, such as unified medical language system (UMLS), to quantify and to impose structure on conceptual relationships. With this classification, the selection criteria can be hierarchically built up or finely specified to obtain desired results. A lengthier discussion of this topic appears later in this article. Look and Feel: The look and feel of the site needs to convey the sense of scholarly authority, but with a sleek technological approach. Medical personnel have high standards concerning the accuracy of articles, and a professional presentation aids in supporting this perception. A site that is gaudy or shows too many bells and whistles, or involves a long load time, reflects poorly on the content, as do garish colors or lack of color. Images and video must be consistent and should be available to download as needed for reference and for closer examination. The level of interactivity available should be low, as most gastrointestinal specialists have small online presence. Natural language queries, such as those used with AskJeeves (http://www.ask.com/), are unnecessary, as long as the search and browse features are precise and efficient. Pages may be presented as either textbook or journal articles, based on the preference of the user. Extracting Content, Metadata, and Cross Referencing: One of the most powerful functions of the Internet is the ability for Web users to span several related articles quickly because of cross-referenced links. A user interested in

21

Gastrointestinal Motility Online Educational Endeavor

the preparation of a particular chicken recipe can quickly reference how to sauté and with what form of pan, moving quickly to sources to buy the appropriate cast-iron skillet or wok to benefits and comparisons of different brands and retailers. A Wiki of only gastrointestinal specialists, with limited control from an editor, would be appropriate for the collection and dissemination of information: Gastrointestinal Motility Online is striving towards that goal. A particularly useful feature in Gastrointestinal Motility Online, not available in online journals, is its cross-referencing tool. Articles that are closely related or broach a topic in greater depth can be quickly accessed by a cross reference within the Gastrointestinal Motility Online domain. These links are established by content extraction tools that create metadata and relate that data between different documents. Footnotes are available at the bottom and allow for a broader topical search; however, the inline linking of articles is particularly appealing for tracking particular items or syndromes of interest.

auxiliary technologies Search features are incorporated in Gastrointestinal Motility Online, but are considered secondary to the organization of the information. A dynamic keyword search for anatomical sections of the gastrointestinal tract is less likely to reveal useful information on general function than a manual perusal of the literature through the prepared subject browse option. The efficient organization of the information, based on the anticipated needs and access patterns of users, is an essential feature for building a knowledge repository. By acknowledging that specialists would be more likely to query by anatomy or syndrome, one needs mechanisms for structured order, as compared to mechanisms that order by article size, author name, or recent usage. An additional feature, incorporated within the community-building module, is a user rating system that allows users

22

to rank the importance of articles so that users browsing information can be directed to the most useful and informative articles. One difficulty with searches is the likelihood that a particular phrase will appear in nearly all documents unless the search is very specific. In cases where the gastrointestinal tract is analyzed as an interactive system, the phrases for anatomical locations may occur multiple times as reference points, but the central theme of the article may not be easy to determine by word frequency. As a result, a number of semantic tools, described later in this article, are used to analyze articles and to classify them appropriately. Experts of the Nature Publishing Group (NPG), who possess prior experience in online information presentation, based on the online version of Nature magazine, helped to develop the initial vision of the online knowledge repository. The base site is hosted by Nature Publishing Group. While the site was being developed, commercial tools were available to handle both the production of electronic journals and static textbook efforts like AccessMedicine/Harrison’s Online (http://www. accessmedicine.com/) and WebMD. However, the gastrointestinal knowledge repository falls somewhere in between these two cases; accordingly, few over-the-shelf tools and algorithms were available for immediate use. As such, a significant fraction of the interface and architecture had to be innovated and refined, through experimentation. Many of the existing tools for creating electronic journals are geared towards collation of articles, graphics, and layout work. These tools reduce the time needed by the authors and editors for the processes of uploading, formatting, and editing. In the development of Gastrointestinal Motility Online, the use of a software suite facilitated handling of images and consistency of look and feel. One area where tools are lacking is the ability to organize information into a coherent topical fashion, as in a textbook. Searching by keyword is especially difficult on a physician specialty site, where the dialect is limited and

Gastrointestinal Motility Online Educational Endeavor

the concepts are reused multiple times. As such, editorial staff must impose additional control to prevent the site from becoming a write-only knowledge repository. Collections of electronic journals, such as Ovid (http://www.ovid.com/site/index.jsp), have been primarily targeted for libraries and research centers. The primary purpose of Ovid is to serve as a warehouse of information, albeit uncategorized. Gastrointestinal Motility Online’s current state differs from that of Ovid in terms of the presentation of the material: the former system is specifically formatted to provide an online view as well as a hardcopy output. The long-term goal of Gastrointestinal Motility Online is to collect a comprehensive knowledge on subjects (as Ovid does), as well as to add more intelligent search tools or information utilities. While Ovid does not organize information except into broad categories, Gastrointestinal Motility Online refines classifications, provides responses to search queries that are more accurate, and supports tools that use the knowledge in compelling ways, such as for differential diagnosis. The advantages of sites built in a textbook style, such as AccessMedicine, is the hierarchical organization and ready access to information. The evolution of online textbooks has generated significant activity on the sites, as teaching tools. Gastrointestinal Motility Online uses over 40 illustrative videos, which are not typical of a journal, but fit the online textbook paradigm. Online textbooks are excellent repositories of information, except that updating the sites to incorporate new information is generally cumbersome because of the level of interactivity involved. “Gastrointestinal Motility Online” is a hybrid, adopting the best qualities of online textbooks, such as Harrison’s Online, and journal collections, such as PubMed. Organized, Well Edited, Frequently Updated and Comprehensive Information are self-explanatory. Search features specifically refer to the ability to search for a keyword or phrase. Differential diagnosis refers to the abil-

ity to integrate clinical patient presentations and create a list of possible problem diagnoses. At a broader technological level, eBooks represent a technology that has been adapted to deliver information electronically. Medical eBooks can be argued to be a natural outgrowth of the eBook movement to electronic media: volume and space requirements are reduced, key phrase searching can be performed, and portability is enhanced. Nevertheless, few medical texts are adapted as eBooks. Although eBooks have grown in popularity, they have not grown as rapidly as projected by consultants; this could be because of the following reasons: 1. 2.

3.

4.

Most readers see no need to replace print books. Due to the limited screen size, limited battery life, and navigation interface issues with eBooks, many people still find paper books easier to handle. Digital rights management causes compatibility and portability problems when attempting to move the eBook from desktop to PDA or laptop. Current pricing of eBooks does not account for the reduced value relative to paper books. When readers finish reading a paper book, they can give it to a friend or sell it to a used bookstore; neither is possible with most eBooks. (Crawford, 2006)

Additionally, delivery of information may not be simply online, but online and to a mobile user using a PDA or other portable device. The constraints involved in transmitting and displaying information on a limited display panel create a new set of challenges. In most markets other than healthcare, the primary applications for PDAs are for scheduling and contact management (as a busy executive might use in lieu of a pen-and-paper daily planner), or as a portable browser or e-mail client (as in the case of technologists and engi-

23

Gastrointestinal Motility Online Educational Endeavor

neers). In such cases, the application of the PDA works within the bounds of the limited display and the modern constraint of minimal bandwidth, often serving as a surrogate cell phone of sorts. However, in medicine, the PDA is often stretched beyond its limits. The current trend is the delivery of detailed information pages into a portable format, downloadable to PDA. Since medical practitioners can no longer maintain a complete mental catalog of all drugs and particularly obscure clinical symptoms or diseases, PDAs assist physicians in their duties without requiring a quick trip to a computer terminal or a large paper binder archive. Harrisons and UpToDate have both moved rapidly in this area, and the list of available drug databases is already large. PDAs fill the role of drug lookup very well, as well as serve as a primer for obscure diseases. The difficulty lies with more graphically intense data that may not display properly, or may need to be downloaded on the fly. In such cases, medical information systems are pushing the technological limits of PDAs. PDA sales as a whole, however, are in decline, except as a niche application. Analysts at organizations such as IDC and Gartner have predicted downtrend trend in sales of PDA. Dell has withdrawn its PDA line from production (Mechaca, 2007). In the long term, the PDA may carefully constrain its niche to feature more of the portable planner features and less multimedia and display power, rendering it less useful to medical practitioners. Until the advent of revolutionary new

display technologies, such as holographic displays or direct-to-eye projection technologies, the ability for PDAs to contribute to medical reference appears to be technologically limited. As shown in Table 1, Gastrointestinal Motility Online is a hybrid that lies between the two models of electronic journals and online textbooks, exclusively focused on providing authoritative information in an organized fashion within a specific domain. Using this system, a gastrointestinal researcher can find the most recent journal articles because of the frequent updates, and a gastrointestinal clinician can easily locate a detailed diagram of the lower esophageal sphincter. In addition, topical information can be cross-linked between papers—a typical feature of textbooks and very pertinent for teaching and presentations.

concepts and InnoVatIons The exploration of the technological space between electronic journals and online textbooks is a relatively new idea. All new ideas face challenges in terms of deployment and adoption. Consider the fax number. As the number of fax machines increases, the value of the fax increases—this illustrates the fact that networks attain greater value with larger number of users. The difficulty faced by Gastrointestinal Motility Online and other specialty interest sites is in terms of initial growth and development of specific communities of interest. These sites must be aesthetically

Table 1. The online endeavor Online Textbook

Hybrid – GI Motility Online

X

X

Well Edited

X

X

Frequently Updated

X

X

Comprehensive Information

X

X

X

X

Search Features Differential Diagnosis

24

Journal Collections

Organized

X

X

Gastrointestinal Motility Online Educational Endeavor

attractive, informative, efficient, and up-to-date. The case of Gastrointestinal Motility Online illustrates one form of evolution of online journals and textbooks into an active online scientific community. Site loyalty is achieved and maintained by the reputation of the authors and contributors. Gastrointestinal Motility Online needs only to achieve a critical user mass before gaining the benefits of Web sites like e-Bay (http://www.ebay. com/) or Amazon (http://www.amazon.com/) in terms of de facto authority and brand recognition. For sites which are less commercially oriented, the loyalty of the user base is perhaps even more heavily emphasized. Two notable sites, which have grown rapidly without such a strong commercial bias, are Wikipedia and Imdb. Wikipedia has been both maligned and praised for its loyal community and efforts to create a free encyclopedia that is accurate and up to date without any commercial affiliations. Wikipedia began in 2000 as a complementary project for Nupedia, in which articles were written by experts and reviewed by a formal process (Wikipedia, 2007). In 2001, Larry Sanger proposed on the Nupedia mailing list to create a wiki as a “feeder” project for Nupedia (Sanger, 2001); this spawned rapid growth. By 2001, Wikipedia contained approximately 20,000 articles and 18 language editions. As of 2007, English Wikipedia contains over 1.7 million articles, making it the largest en¬cyclopedia ever assembled. The site relies on the goodwill of its members to write, update and contest articles, and has thus far proven that the Internet community as a whole is willing to contribute towards the database, albeit haphazardly (recent events are more likely to be covered in detail, while significant historical figures languish). Given the size and relatively stable growth of the project, the prospect of a peer-reviewed and written information repository might not be so cynically doomed. The International Movie Database (IMDB), a site in the top 20 (Alexa Internet, 2007b) U.S. Web sites visited, is the largest Internet compilation of

movie information (approximately 900,000 titles and 2.3 million names) (IMDB, 2007). IMDB draws a significant portion of its information from the participation of its user base. Beyond subjective reviews, users are also asked to supply cast and crew lists, production details, and actor biographies. IMDB grew from two lists that started as independent projects in early 1989 by participants in the Usenet newsgroup rec.arts.movies. Each list was maintained by a single person, recording items e-mailed by newsgroup readers, and posting updated versions of his list from time to time. The lists were eventually combined, and by late 1990, the lists included almost 10,000 movies and television series. As the contributions continued to grow rapidly, the IMDB formed as an independent company in 1995 and was later purchased by Amazon Inc. (Wikipedia, 2007b). The approach of the IMDB system closely resembles the envisioned system for GI Motility Online, with the user population submitting proposed changes, followed by an editorial process. Database content is generally provided and updated by a vast collection of volunteer contributors. There are only 17 members of the IMDB who are dedicated to monitoring received data, although 70% of IMDB’s staff serve as editors (IMDB, 2007), reviewing changes, verifying the information before posting the changes, and policing the forums. Peer review is considered to be one of the most even-handed and least biased methods of scrutinizing articles for publication. The development of a community, as well as the ability for members of that community to voice their opinions about professional papers is pivotal to the dissemination of accurate information. The model of Amazon or eBay is to allow users to comment, and thereby to signify reliability and approval. Gastrointestinal Motility Online allows users to provide feedback, both critical and supportive, in order to enhance the relevance of articles. According to Harris Interactive poll, only one-fourth of physicians use the Internet to

25

Gastrointestinal Motility Online Educational Endeavor

communicate with their patients (Computing in the Physician’s Practice, 2000). In the same poll, although 89% of physicians use the Internet in their practice in search of information, they spend only six hours per week to browse medical developments. Accordingly, Gastrointestinal Motility Online has been configured to serve as an encyclopedic source, as well as a high-value news feed. AccessMedicine uses a Podcast update model, with 10-minute broadcasts generated daily for use in family practice. Gastrointestinal Motility Online caters to a smaller specialty group with correspondingly less frequent pace of developments, and is therefore less time sensitive.

automation Automated content generation or extraction from other publications is not feasible, due to the stringent need to maintain relevance and quality. Thus, deployment of a pure peer-reviewed wiki-style community is precluded by the need to maintain quality. If membership of the wiki is restricted to those users whose credentials are accepted, or if changes must be approved at an editorial level, then a wiki would facilitate the rapid exchange of information as well. In the second part of the endeavor, the goal is to enable machine-assisted updating of the material in the gastrointestinal knowledge repository. Currently, all updates must be initiated manually. The pressure to update, but to update accurately applies to many situations: addition of new material, editing of existing material, and deletion of parts of existing material as new study results become available. Initial thoughts and test results are documented in Sharma (2005). As the site evolves, the need to keep the site relevant to a particular group of specialist physicians may conflict with the preferences of another significant category of users: the researchers. Gastrointestinal Motility Online intends to use user-based customized layout, as presented by Sarnikar et al. (2005). The filtering of details from

26

the main knowledge repository is not intended to block or hide information, but to provide a more relevant source. Classification and weighting are accomplished by a rules-based system that can assess whether an article is clinical- or basic science-related. Login will also provide both the ability to contribute and comment on articles, and to obtain a specialized view depending on the type of user. The application of user-based site layouts is not innovative, but is important to the domain of medical informatics because physicians attach such a high priority on relevance. Given the large number of articles published weekly and the difficulty in ascertaining relevance and quality, a number of automated tools will be used to optimize the updating process. Sarnikar et al. (2005) present one technique that will assist in filtering journal results and maintaining and updating the site. Their method selects articles ranked by relevance using a combination of both rule-based and content-based methods, using the following principles: 1. 2.

3. 4. 5. 6.

Profiles are modeled in the form of rules. The purpose of the rule-based profile is to identify a sub-set of documents of interest to a user. Each role has a set of predefined rules associated with it. Rules specify knowledge sources to access (e.g., nursing journals for Nurses). Rules can specify knowledge depth and knowledge breadth. Rules can specify semantic types of primary importance to roles.

Profiles are used in the gastrointestinal motility context to separate information into categories: for example, new clinical findings versus basic science. Articles may be assigned a category and a relevance weight, given categorization rules based on Unified Medical Language System (UMLS) synonym lists and the categories sign or symptom, diagnostic procedure therapeutic or preventive

Gastrointestinal Motility Online Educational Endeavor

procedure and disease or syndrome semantic types. In addition to a text search in the abstract, Sarnikar and Gupta (2007) also assign weights to the type of journal. These tools can select and filter relevant articles for presentation as an RSS XML news feed to editors or automatically assemble relevant articles for use by the editors or Web site administrators. While these tools will aid the editor, there is no replacement for the role of humans in decisively selecting and classifying information. Ontologies and semantic networks are prerequisites to the development and classification of information repositories. Ontologies serve many purposes (Kumar, 2005) including: • • •

Reuse and sharing domain knowledge Establishing classification schemes and structure Making assumptions explicit

They further enable analysis of information and complement the stricter terminology that is used in straightforward text searches. Examples of ontologies in use today include the National Library of Medicine’s Medical Subject Heading (MeSH), disease specific terminologies such as the National Cancer Institute’s PDQ vocabulary, drug terminologies such as the national drug data file, and medical sociality vocabularies such as the classification of nursing diagnoses and the current dental terminology. In Gastrointestinal Motility Online, the ontological hierarchy will be used to distinguish between sections of the gastrointestinal tract, from the stomach and esophagus to the large intestine and colon (Kumar, 2005). A key enabler for development of automatic information processing is the set of ontologies presented in the UMLS semantic network that rely upon the concepts built in the UMLS concept hierarchy. Hierarchical and clustered ontologies allow software to construct knowledge trees, conglomerating relevant knowledge. An overview of the UMLS is available at the unified

medical language system: What is it and how to use it? (http://www.nlm.nih.gov/research/umls/ presentations/2004-medinfo_tut.pdf). UMLS is an aggregate of over 134 source vocabularies, including the classifications from such lists as ICD-10 and DSM: IIIR-IV. It represents a hierarchy of medical phrases that can be used to classify most medical articles and textbook entries. For example, using UMLS, the following phrases are grouped similarly: Deglutition Disorders, Difficulty in swallowing, Difficulty in swallowing (context-dependent category), Dysphagia NOS, Dysphagia NOS (context-dependent category), Can’t get food down, Cannot get food down, Difficulty swallowing, Difficulty swallowing (finding), Dysphagia, Dysphagia (Disorder), Swallowing difficult, Swallowing Disorders (Aronson, 2001). The system described in Sharma (2005) uses techniques of Natural Language Processing (NLP) to construct a semantic understanding that surpasses text searching. Using the automated integration of text documents in the medical domain (ATIMED) system, the content and order of phrases are related lexically using a concept called Word-Net. Word-Net operates on the verbs, subjects and objects of the sentences, comparing sets and subsets of subject-verb-objects collections in order to determine topic relatedness. Word-Net further uses a lexical dictionary to determine similarity in all verb pairs and then subject-verb-action pairs. Sharma (2005) uses the following two sentences as examples: Dysphagia is a disease and defined as a sensation of sticking or obstruction of the passage of food. Dysphagia is related to obstruction of passage of food. Since both sentences contain similar objects and subjects, and use the verb “is”, the sentences are deemed similar. However, within the current mechanism, the phrase, “Dysphagia relates to obstruction of passage of food” would result in a poorly scored correlation or low match because the action verb is not similar (Sharma, 2005). Finally, the same technique allows the creation of new documents by collating sentences and

27

Gastrointestinal Motility Online Educational Endeavor

Figure 1. Integration of semantic understanding to generate new information (Adapted from Sharma, 2005) Sentence Syntax Definition Tables

Feature Selection

Feature 1

Feature 2

Feature 3

Feature 4

WordNet

Concept Formulation and Creation

paragraphs from various documents. An initial method of grouping sentences uses the quantity of concepts expressed. This method is further refined by evaluating the sentences based on the following criteria: similar-subjects, similar-objects, similarsubjects-objects, and similar-objects-subjects. Based on the structure of English grammar, these techniques have been reliably shown to collate relevant data into a readable format. A diagram of the method is shown in Figure 1. A sample output, based on the use of this technique, is provided below: REGURGITATION is defined as the spontaneous appearance of gastric or esophageal contents in the back of the throat or in the mouth. In distal esophageal obstruction and stasis, as in achalasia or the presence of a large diverticulum, the regurgitated material consists of tasteless mucoid fluid or undigested food (Sharma, 2005). The context interchange of heterogeneous sources of information being collated for different classes of users leverages tools from an additional

28

branch of information technology. One level of development is the creation of maps between inputs and outputs, in much the same way that a dictionary might map between languages. The conversion of n sources to m outputs can grow to be an exponentially difficult problem; this can be addressed with the use of intelligent heuristics and protocols to selectively prune the data. Difficulties become particularly apparent with changes in any client schema that cause a cascade of changes in the mappings (Sarnikar, 2005). Using an independently developed predefined mediating schema would restrict the amount of information that could be exchanged. Apart from relational schema, a client schema could also be specified as a hierarchical schema or as an XMLbased message (Sarnikar & Gupta, 2007). In the context of Gastrointestinal Motility Online, the specific source contexts are innovative journal papers, reviews, and textbook articles. The output contexts are specialist clinicians, researchers, students, and other health professionals. Developing a schema to accurately represent journal abstracts and determine the relevance of those abstracts is

Gastrointestinal Motility Online Educational Endeavor

another method of exchanging contexts. Innovation in this domain will allow Gastrointestinal Motility Online to maintain updated, consistentquality references without requiring an editor to read every journal article published immediately.

lessons learned The GI Motility Online site benefited from a mature development environment for Web-based information retrieval. Interface: In the case of GI Motility Online, the user base is a readily identifiable group, trained in a similar fashion, with specific needs and expectations of organization and formatting. The following two aspects influenced the design process: •

•

Pagination is a critical issue for online didactic materials. As people do not like the interface to online books (Crawford, 2006), the development of the site must reflect the reading style and needs of its users. In the case of GI Motility Online, there are currently no page delimiters; this is encouraging users to print the material to be used as a reference. A solution of delivering a formatted print-ready PDF could be applied, but the cost begins to climb with the number of different formats and delivery options supported. Tables must be handled intelligently. Tables are used in many medical publications to rapidly and clearly present information. The trend in the mid-1990s for netiquette was to inline the tables with the text, but this can create awkward gaps or poor formatting choices. GI Motility Online chooses the path of Harrison’s Online, in linking to the tables outside the main document to preserve readability.

Based on their respective training, many physicians expect a particular language and lay-

out of medical information. The highest-value information for each physician may vary based on specialty, and user-customization of graphical widgets may enhance information value. Since GI Motility Online aims to support a broad base of users with many different roles and information needs, reusable widgets, of the type available in customizable Web-portals, may potentially solve these varying needs. Workflow: As stated previously, over nine months was required for simply editing the site information, moving documents back and forth between editors, experts and developers. The method involved several inefficient technologies, such as mailing CDs and sending large files through mail servers. The advantage of such hands on interaction and communication is the result: the authors of the site are particularly pleased by the polished appearance of the delivered product. Electronic collaboration between different sites was difficult due to the nature of document formats, which does not produce a consistent print layout on different computers, and layout formats, which do not allow easy markup or revision to the document. The use of a standardized input format and efficient conversion of the RTF-formatted documents into Web viewable formats was crucial to the development of the Web site. Though conversion of documents is a minimally complex task, the production of a site that allows conversation on the fly to support concurrent editing and proofing between multiple users is a challenge. Online publishing companies, such as Atypon, offer suites of software to facilitate publishing workflow, allowing multiple authors to upload articles, multiple editors to revise articles, and art editors to manage graphics, all in an organized fashion. Meta-Information: The World Wide Web provides a tremendous asset in connecting related articles seamlessly. In determining related pages, an active agent—a human, machine or some combination thereof—who must isolate the crucial elements of a page and capture that as

29

Gastrointestinal Motility Online Educational Endeavor

meta-information, preferably categorized. Organized meta-information allows automated tools to develop connections and links to potentially relevant information. Extracting crucial elements of a page is best handled by encouraging authors to define keywords, as with scientific articles. Although many automated agents have historically been unsuccessful in information extraction, the medical domain provides some assistance with standardized ontologies, which allow agents to categorize information in a framework, reducing some identification errors. Maintenance: Many articles in mainstream sites, such as ESPN.com, offer the opportunity for users to comment and leave feedback. This increases user participation and value to the site, but raises additional issues that need to be addressed, such as: • • •

Profanity or other inappropriate comments must be censored Sites are more open to attack and security leaks due to their increased functionality Bandwidth usage increases, and may debilitate the site

In addition, a version control framework must be established for sites that allow updates, in order to rollback unwanted changes. Furthermore, hardware resources must be allocated to store site changes.

current usage and Future dIrectIons Anecdotal evidence provided by the lead creator of GI Motility Online, Dr. Raj Goyal, highlights that the representatives of Nature Publishing Group describe the site as being popular, and that site visits to GI Motility Online are increasing. Due to the delicate nature of conversation with his peers in gastroenterology, Dr. Goyal cannot be

30

certain, but he states that initial impressions are unanimously positive and enthusiastic. Dr. Goyal notes that the majority of the traffic to GI Motility Online currently arrives from Google, and indeed, Google’s first returned site in response to “GI Motility” is GI Motility Online (Google, 2007). GI Motility Online does not currently permit any advertising; if this policy is revised, the usage patterns may change. Further, the numbers for site visits are expected to increase if the site were to be more supported by Nature Publishing Group. The stakeholders are satisfied with the quality of the product, and are ambitiously pursuing an extension of the product to broaden coverage to other parts of the GI tract. Dr. Goyal has received inquiries to publish the material in a hardcopy format. Custom-built books or full-page colored slides offer additional utility to medical educators, specialists, nurses, and students, and may also offer a revenue stream. These custom-built printable books are a logical extension of the current “print what you want” photo fulfillment services that are popular on photo sharing sites or album hosting services. Given that Harrison’s Principles of Internal Medicine contains over 2,000 colored pages in book form, there may be a future for customized books in medicine. Retrieval: New algorithms for search and ranking are not merely lexically-based, but also combine closely related concepts and relationships between ideas. This facilitates the creation of a richer search language that can account for relationships, such as causation, consequence, association, treatment, or opposite. For example, hypothermia is directly opposite hyperthermia, and may be caused by thalamic alterations. Such basic chains of relationships could be easily captured if the language and storage of meta-information about articles could contain the necessary underlying details. Presentation: Using Tufte’s principles (Tufte, 2001), the real estate of a screen needs to be more efficiently used. Currently, most search-engine

Gastrointestinal Motility Online Educational Endeavor

results do not often give the sense of the relevance of the article, of the correct sections, or the tone of the article. The information density of the return pages is low, causing users to scroll through potentially hundreds of articles to find relevant information. Graphical interfaces are likely to be a solution, as the ability of the Internet to handle higher bandwidth applications grows. In medical research in particular, an interface, that allows users to quickly see search results of the primary concept and even related concepts, may dramatically affect usability. Online Sharing: Bulletin board systems, which flourished during pre-Internet days, have re-emerged on the Internet as forums, and are a popular source of information. Corporations with significant Internet presence, and especially gaming companies like Nintendo (http://forums. nintendo.com/nintendo/), have begun adding and using these forums as a method of improving public relations and offering support. Intranet and Internet file sharing systems have also flourished, as bandwidth rates increase. The rise of Youtube is one phenomenon, but the comfort of using the Internet to disseminate multimedia (such as GI endoscope video) appears to be more solidified. These high bandwidth applications have grown in acceptance, and despite the threat of viruses in downloaded files, file-sharing traffic is increasing daily. Megaupload and Rapidshare, two prominent file sharing services, are now ranked #18 and #27 in the reliable Alexa ranking (Alexa Internet, 2007b) of most visited Internet sites in the U.S. However, these services are predominantly used by noncommercial entities. Commercial enterprises may be hesitant due to slow adoption, security issues, bandwidth maximums, or unprofessional presentation. Such file sharing sites will certainly cater in the future to commercial entities, perhaps by providing specially developed sites, or providing branded services. The future of the Internet, and especially the ability to deliver high bandwidth, will increasingly rely upon specialized sites with

high-capacity and high-bandwidth connections close to Internet backbones. Collaboration: Collaboration in the domain of GI motility will drive improvements in language tools, with many GI specialists throughout the world who may need to collaborate via the Internet or phone services. Currently, Altavista and Google offer Web-based text translation, but more intelligence or domain-specific knowledge may be required. One common example in GI translation is translating the world “oral” to mean “verbal”, when in fact, the correct reference is to the mouth cavity. Overall, however, text translation between languages is generally adequate for initial communication. Oral translation is a high value direction to pursue, but at the same time, is a very difficult task. Psycholinguistics research is still unraveling the complexities of language parsing, and the ability of current artificial intelligence to understand language is severely limited. Nevertheless, translators in this area will prove essential and highly desirable for the next generation of online collaborators. Similarly, one of the holy grails of artificial intelligence development is the creation of an artificial system capable of interpreting human language. Within specialized domains with limited vocabularies, artificial readers become more feasible, but the medical domain is particularly difficult, due to its large specialized vocabulary. Development in this area would provide rewards for medical researchers, allowing the creation of agents, which would allow researchers to process more information by selecting and even summarizing articles. Organizationally, scientific research in domains such as medicine would benefit tremendously from the creation of a centralized authority to monitor, synthesize and rate research. The current system of research funding in America places research at the whim of special-interest private funding and sometimes misdirected public funding in overly popular or extremely esoteric

31

Gastrointestinal Motility Online Educational Endeavor

areas. Regulating and directing research might also help avoid repeating inconclusive research, which does not get published (and thus may be repeated).

conclusIon Gastrointestinal Motility Online is an evolving knowledge base related to Gastrointestinal Motility disorders. The current phase of the endeavor focuses on the collection and organization of knowledge from many different sources. Knowledge-mining tools are being developed to utilize this information as it becomes available to add fast, relevant access and other utilities to the information repository. The continuous change in standards of care and knowledge due to rapid discoveries in the basic and clinical sciences prompts for a system that is more flexible than a textbook, while demanding thoroughness and accuracy. Knowledge mining tools and other advanced technologies to aid in the conversion and integration of articles and research into the mainstream science are being integrated into Gastrointestinal Motility Online, and look to impact the breadth and speed of knowledge-base upgrades. Gastrointestinal Motility Online serves to balance the needs of its user base while embracing the academic rigor in a novel application of technology to the science of medicine.

acKnowledgMent The authors would like to thank Dr. Raj Goyal for his invaluable input and contribution to GI Motility Online and this article. The authors also thank Richard Martin for his helpful comments and proofreading of the text.

reFerences Alexa Internet. (2007a). Three-month traffic statistics for wikipedia.org. Retrieved from http://www. alexa.com/data/details/main?q=&url=wikipedia. org. Alexa Internet. (2007b). Top Sites United States. Retrieved from http://www.alexa.com/site/ds/top_ sites?cc=US&ts_mode=country&lang=none. American Library Association. (1989). Presidential Committee on Information Literacy - Final Report. Retrieved from http://www.ala.org/ala/ acrl/acrlpubs/whitepapers/presidential.htm. Antman, E., Lau, J., Kupelnick, B., Mosteller, F., & Chalmers, T. (1992). A comparison of results of meta-analyses of randomized control trials and recommendations of clinical experts. Journal of the American Medical Association, 268, 240-248. Aronson A. (2001). Effective mapping of biomedical text to the UMLS Metathesaurus: The MetaMap program. Proceedings of the 2001 AMIA Symposium, pp. 17-21. Carlson, C. (2003). Information overload, retrieval strategies and Internet user empowerment. In: L., Haddon (Ed.), The Good, the Bad and the Irrelevant (COST 269) 1(1) (pp. 169-173). Helsinki, Finland. CEBM. (2007). Why the sudden interest in EBM?. Retrieved from http://www.cebm.utoronto.ca/ intro/interest.htm. Computing in the Physician’s Practice. (2000). Retrieved from http://www.harrisinteractive.com/ harris_poll/index.asp?PID=58. Crawford, W. (2006). Why aren’t ebooks more successful?. Retrieved from http:// www.econtentmag.com/Articles/ArticleReader. aspx?ArticleID=18144. Davis, D., Thomson, M., Oxman, A., & Haynes, R. (1997). Changing physician performance: A

32

Gastrointestinal Motility Online Educational Endeavor

systematic review of the effect of continuing medical education strategies. Journal of the American Medical Association, 274, 700-705. Fridman, S. (2000). Doctors lag when it comes to computer use - Industry trend or event. Retrieved from http://www.findarticles.com/p/articles/ mi_m0HDN/is_2000_March_28/ai_60907019. Giles, J. (2005). Internet encyclopedias go head to head. Retrieved from http://www.nature.com/ news/2005/051212/full/438900a.html. Google. (2007). Search returned http://www. google.com/search?hl=en&q=GI+Motility. Gore, A. (1992). Infrastructure for the global village. Scientific American, 265,150-153. Guyatt, G. Cairns, J., Churchill, D., Cook, D. Haynes, B., Hirsch, J., & et al. (1992). EvidenceBased Medicine Working Group: Evidence-based medicine. A new approach to teaching the practice of medicine. Journal of the American Medical Association, 268, 2420-2425. Hawking, S. (2001). The universe in a nutshell. New York, NY: Bantam Books. Haynes, R. (1993). Where’s the meat in clinical journals [editorial]? ACP Journal Club, 119, A22-A23. IMDB. (2007). How/where do you get your information? How accurate/reliable is it? Retrieved from http://imdb.com/help/show_leaf?infosource. Kasper, D., Fauci, A., Longo, D., Braunwald, E., Hauser, S., & Jameson, J. (2005). Harrison’s principles of internal medicine, 16th ed. New York, NY: McGraw-Hill. Kumar, A. (2005). Ontology-driven access to biomedical information (ODABI). Undergraduate Thesis, University of Arizona. Lindberg, D. (2003). NIH: Moving research from the bench to the bedside. Presentation to the Subcommittee on Health.

Lyman, P. & Varian, H. (2003). How much information?. Retrieved from http://www2.sims.berkeley.edu/research/projects/how-much-info-2003/ index.htm. Mechaca, L. (2007). Goodbye, Axim. Message posted to http://direct2dell.com/one2one/ archive/2007/04/11/11397.aspx. Sackett, D. & Straus, S. (1998). Finding and applying evidence during clinical rounds: The evidence cart. Journal of the American Medical Association, 280, 1336-1338. Sackett, D. (1997). Using evidence-based medicine to help physicians keep up-to-date. The Journal for the Serials Community, 9, 178-181. Sanger, L. (2001). Let’s make a wiki. Message posted to http://web.archive.org/web/20030414014355/ http://www.nupedia.com/pipermail/nupedial/2001-January/000676.html. Sarnikar, S., Zhao, J., & Gupta, A. (2005). Medical information filtering using content-based and rulebased profiles. Proceedings of the AIS Americas Conference on Information Systems (AMCIS 2005), Omaha, NE. Sarnikar, S. & Gupta, A. (2007). A context-specific mediating schema approach for information exchange between heterogeneous hospital systems. Forthcoming in International Journal of Healthcare Technology and Management. Schwartz, B. (2004). The paradox of choice. New York, NY: HarperCollins Publishers. Senge, P. (1994). The fifth discipline: The art and practice of the learning organization. New York, NY: Doubleday/Currency. Sharma, R. (2005). Automatic integration of text documents in the medical domain. Undergraduate Thesis, University of Arizona. Slawson, D., Hauck, F., Strayer, S., & Rollins, L. (2007). What is information master?. Retrieved from http://www.healthsystem.vir-

33

Gastrointestinal Motility Online Educational Endeavor

ginia.edu/internet/familymed/docs/info_mastery. cfm#Information. Tufte, E. (2001). The visual display of quantitative information. Cheshire, CT: Graphics Press. Vanhecke, T., Barnes, M., Zimmerman, J., & Shoichet, S. (2006). PubMed vs. HighWire Press: A head-to-head comparison of two medical literature search engines. [Epub ahead of print] Computers in Biology and Medicine.

Wikipedia. (2007a). Information retrieval. Retrieved from http://en.wikipedia.org/wiki/Information_retrieval. Wikipedia. (2007b). Wikipedia. Retrieved from http://en.wikipedia.org/wiki/Wikipedia.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 1, edited by J. Tan, pp. 24-43, copyright 2008 by IGI Publishing (an imprint of IGI Global).

34

35

Chapter 3

Envisioning a National e-Medicine Network Architecture in a Developing Country: A Case Study

Fikreyohannes Lemma Addis Ababa University, Ethiopia Mieso K. Denko University of Guelph, Canada Joseph K. Tan Wayne State University, USA Samuel Kinde Kassegne San Diego State University, USA

aBstract Poor infrastructures in developing countries such as Ethiopia and much of Sub-Saharan Africa have caused these nations to suffer from lack of efficient and effective delivery of basic and extended medical and healthcare services. Often, such limitation is further accompanied by low patient-doctor ratios, resulting in unwarranted rationing of services. Apparently, e-medicine awareness among both governmental policy makers and private health professionals is motivating the gradual adoption of technological innovations in these countries. It is argued, however, that there still is a gap between current e-medicine efforts in developing countries and the existing connectivity infrastructure leading to faulty, inefficient and expensive designs. The particular case of Ethiopia, one such developing country where e-medicine continues to carry significant promises, is investigated and reported in this article.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Envisioning a National e-Medicine Network Architecture in a Developing Country

IntroductIon Healthcare consumers in general tend to seek access to affordable health services that will meet their needs. From an ethical standpoint, healthcare has to be available when and where consumers need it; physical separation between consumers and healthcare facilities must not pose severe limitations on the delivery of efficient healthcare, even if patients are located in remote areas. In this sense, information and communications technology (ICT) has been demonstrated to offer a competitive choice for accessing affordable and effective health services, especially when access is difficult and limited (Horsch & Balbach, 1999; Kirigia, Seddoh, Gatwiri, Muthuri, & Seddoh, 2005; Tan, Kifle, Mbarika, & Okoli, 2005). More recently, with the continued maturity of network such as Integrated Services Digital Network (ISDN) and Asynchronous Transfer Mode (ATM) networks and related technologies (Perednia & Allen, 1995; Tan, 2001), e-medicine implementation has entered a stage where both the health providers and consumers can now benefit significantly. IT-based horizontal and vertical communications among the healthcare facilities within the organizational structure of the healthcare system are essential. Such communications facilitate efficient information exchange and help the delivery of essential health services to underserved rural areas. These communications can be supported through a nationwide e-medicine network that is based on affordable telecommunications infrastructure. The network should connect all regional clinics to urban area hospitals. The benefits of such a network include: (a) establishing reliable horizontal-vertical communications and information sharing among facilities, thereby driving up quality, improving efficiency, and enhancing cost-effectiveness of services; (b) achieving e-health commitments and bringing healthcare closer to underserved and un-served rural areas; (c) strengthening collaboration among hospitals

36

within a multi-provider care management context; (d) minimizing long distance travels among rural people in need of proper medical care to urban areas or the capital city; and (e) providing medical information to clinical practitioners that will help them keep abreast of clinical breakthroughs as well as new technological advances. For Ethiopia, a lesser developing country with significant challenges in meeting basic healthcare needs, it is argued that e-medicine development is emerging and can be fruitfully cultivated over the coming years if a vision and long-term strategy for this technology can be used to help increase the number of citizens receiving care and decrease the subsequent healthcare costs. This article lays out such a vision and strategy for a nationwide e-medicine infrastructure to be designed. It is organized as follows. First, the background and various design considerations for a nationwide e-medicine network is presented. Next is an overview of the requirements for the network design followed by a more in-depth description of the local and wide area network (LAN/WAN) architecture envisioned. The focus of the discussion will then shift to the existing Broadband Multimedia Network (BMN) and Very Small Aperture Terminal (VSAT) infrastructures and how these networks may be integrated into the nationwide e-medicine infrastructure. Finally, the article concludes with insights into potential future work in e-medicine for developing countries.

e-MedIcIne networK desIgn consIderatIons E-medicine refers to the electronic delivery of healthcare and sharing of medical knowledge over a distance employing ICT. A national e-medicine network allows sharing and exchanging of clinical data among physicians, administrators, even patients or other participating health professionals regardless of physical distance separation or geographical terrain of the whereabouts of these

Envisioning a National e-Medicine Network Architecture in a Developing Country

network participants within the national boundaries. The network also facilitates communications among physicians and academics across diverse cultures, affiliated healthcare organizations, and publicly or privately funded research institutions. Since there is lack of transportation and communication infrastructure in developing countries, medical and clinical data exchanges can be further secured and facilitated through an existing e-medicine network. In developed nations, e-medicine services (Wright, 1998) can benefit remote locations that may not be easily accessed due to unpredictable or harsh weather conditions found during certain times of the year, for example, parts of North America and Scandinavian countries are often heavily affected by snow and other natural hazards such as avalanches, falling boulders and closure of highways due to multi-vehicle accidents or other calamities. Mountainous terrain in certain parts of North American regions such as Alaska, British Columbia, Alberta and New Territories implies the need for viable distance healthcare delivery solutions. E-medicine allows health professionals around the world to establish faster communication and exchange information with clients and regional authorities irrespective of geographical locations. It may also support rural dwellers to get healthcare services delivery similar to their urban counterparts. A mobile e-medicine system, for instance, provides a convenient platform for acquisition, transmission and delivery of health-related data to healthcare providers through 2G/3G-based wireless networks (Wootton, 2001). Recognizing these benefits, the International Telecommunications Union (ITU) has set a global agenda to promote e-medicine applications in developing countries. Ethiopia, one of the beneficiaries of such an initiative, has commissioned some ICT projects such as SchoolNet, WoredaNet and BMN to enable fully-fledged connectivity to make better use of the ICT in the health and education sectors.

network architecture The design of suitable national e-medicine network architecture in a lesser developing country such as Ethiopia requires several key components to be integrated into a dynamic and enterprising communication infrastructure. Major architectural components include: (1) LAN architecture for local networking and sharing of health-related information; here, communications may be established using wireless cellular or ordinary fixed telephone lines; (2) WAN architecture for national networking and sharing of health-related data; this will allow communications among local and national physicians, healthcare workers and clients covering urban and rural communities; and (3) designing a suitable back-end database and front-end user interface applications that integrate seamlessly with the (prototype) implementation of the proposed architecture. The overall goal of the nationwide e-medicine network architecture is then to provide an affordable and a low-cost system that facilitates uninterrupted communications among physicians and health professionals across the country. The system bolsters connectivity among rural clinics and urban area hospitals to support primarily clinical e-consultation and maintenance of stored patient records. This network should also be cost-effective, expandable, and secure. It must support a stateof-the-art ICT access schema and connectivity to rural area clinics. Existing ICT infrastructure will be given priority to minimize the cost of implementing the nationwide network. In the Ethiopia design, expandability is a concern. First, few hospitals are built in the country while more clinics are being added every year. Moreover, there is a chance to incorporate private hospitals in the nationwide e-medicine network as and when necessary, which will further increases the number of future connected sites. As well, the area of e-medicine applications will not be limited to just some specific diseases, but will be expected to increase in type and number over

37

Envisioning a National e-Medicine Network Architecture in a Developing Country

the long haul. In fact, the network should also support advanced applications, which require real-time connectivity such as videoconferencing capabilities for future use. During e-consultation or patient referral, most of the data exchanged over the network are sensitive patient information. Confidentiality of patient information must therefore be respected. For secure communications, protocols such as Secure Socket Layer (SSL) could be used. SSL ensures secured communications over web-based applications and provides the ability to safely exchange patient information across the network (Elmasri, 2000). When doctors exchange patient information, they could adhere to medical protocol that defines the rules to be followed during this process. In addition, the network and accompanying servers could be protected by firewall against hacking from external parties. Firewalls are software or hardware for the sole purpose of keeping digital pests such as viruses, worms, and hackers out of the network (http://www.cisco.com, Tanenbaum, 2004).

networK desIgn reQuIreMents As cost must be one of the driving factors for choosing among existing or emerging ICT infrastructures in the country, implementing nationwide e-medicine network infrastructure may seem at first to be more expensive than building clinics or supplying existing regional clinics with medical personnel and equipment. Yet, a cost-benefit analysis comparing various IT investment approaches will provide best directions to achieving a lower cost solution to the problem of delivering adequate and proper healthcare and disseminating confidential health information to and from various connecting points throughout the country. With today’s oil prices at a premium, network connectivity among the healthcare facilities, both in the urban and rural areas over an existing ICT

38

infrastructure is now considered a cost-effective solution. Of course, set-up costs depend on the type of WAN to be used—to ensure low installation cost, it is proposed that the network design will incorporate an existing WAN provided by the Ethiopia Telecommunication Corporation (ETC). In Ethiopia, most of the inter-hospital communications are traditionally dependent on telephone and hand-delivered referral messages. During referrals patients have to travel afar to one of the urban referral hospitals, carrying the written messages of the referring physician. Clinics located in the telephone coverage areas communicate using telephone to exchange information about availability of specialist(s) or bed in another hospital. Yet, the communication needs of hospitals have grown over the years ahead of its technological capabilities. Geographically dispersed clinics lack modern telecommunication technology access. Among them are instantaneous access to patient information, access to electronic medical records, and access to the Internet. These and other communication needs of health providers also require the development of e-medicine application software backed by electronic patient record systems. Design of such communication networks will also require the understanding of organizational structure of the clinics involved in the network. Since the government/public clinics are owned and organized under their respective regions, the WAN design should follow the organizational structure of the administrative regions in the country. A detailed study about the inclusion of various clinics, their locations relative to the nearest access point to existing ICT infrastructure, traffic load and its characteristics, security, LAN/WAN protocol, topology and bandwidth requirements and utilization, and allocation of bandwidth, among other issues, have to be considered while trying to design a nationwide e-medicine network architecture. For example, issues of communicating patient information electronically may further raise question on medical

Envisioning a National e-Medicine Network Architecture in a Developing Country

ethics, the need for developing standard medical protocols, and detailing policies for use in routine activity via the e-medicine network.

lan architecture To design the LAN for each hospital, we consider the central site, Tikur Anbassa Specialized Hospital located in the capital Addis Ababa, as a model. The hospital is organized into 16 departments with each department further divided into smaller units as necessary. For instance, the Internal Medicine department has several units such as the Renal Unit, the Cardiology Unit, the Neurology Unit, and other units. Physicians in these departments and units need to communicate whenever a patient visits more than one of the units. It is proposed that the LAN follows the hierarchical structure of the hospital. The decision to make the selection between various LAN technologies was based on: (a) expected application to run on the network and their traffic patterns; (b) physical locations of the offices and users to be connected in campus; (c) the rate of network growth; (d) the abundance of the network technology in the market; and (e) simplicity of installation and maintenance. Each of these criteria will now be explored in more depth.

Expected Application to Run on the Network and their Traffic Patterns Currently we expect a Web-based e-medicine application to run on the network. The application will use a central database server where all the user and patient information will be stored. The type of data to be transmitted on the network should accommodate both text and image formats. Since all communications are to be channeled through the server, the traffic pattern around the center is expected to be heavy. Higher speed devices should be installed at the center of the LAN where servers will be located.

Physical Locations of Offices and Users to be Connected on Campus The sample hospital (Tikur Anbassa Specialized Hospital) is housed in a series of five buildings (Blocks A-E). These blocks are not physically separated. Even though precise figures were unavailable, these five buildings are built on roughly 8,000 to 10,000 square meters. While the main offices and departments in the hospital are located in one of the respective blocks, most of these offices are in either of the first two stories of the block they belong. Having routers switches in each of the departments is ideal to design a high-speed and expandable LAN, but it will also make the design expensive to install, support and maintain. A more cost-effective approach is to put switches per building and then get the departments to be connected into various groups by using Virtual LAN technology.

The Rate of Network Growth The rate of the hospital LAN growth depends on the level of computerization in the hospital. Currently in this central hospital site, there is a LAN that connects a few offices and a computer room. The network employs star topology, using a centrally located hub and Unshielded Twisted Pair (UTP) cables forming a peer-to-peer LAN. The purpose of this LAN was to enable offices to share printer and students to get access to research documents. In this design, it is anticipated that as the use of Web-based applications becomes commonplace, there will be opportunities to add more applications and connect more computers and offices to the LAN. The switches-routers selected in this design should therefore have many free ports to help cascade the growing number of anticipated future connections.

39

Envisioning a National e-Medicine Network Architecture in a Developing Country

The Abundance of Network Technology in the Market

layer (core layer); (b) second layer (distribution layer); and (c) third layer (access layer).

Capitalizing on the abundance of emerging network technology in the Ethiopia market, we gathered data from existing network technology vendors and organizations that implement computer networks in the capital city, Addis Ababa. Ethernet technology is common in organizations that implement computer networks, such as Addis Ababa University (AAU Net). AAU Net is a network backed by triangular shape fiber optic cable connecting the three main campuses. The topology is an extended star topology that fastens together fiber optic cables for vertical cabling (backbone cabling) between buildings that house various faculties and departments. These backbone cabling provide interconnection between wiring closets and Point of Present (POP). The zones that fall within a departmental area are served by internetworking devices such as hubs and UTP cables. Interestingly, what is described so far appears to be the dominant design of the small number of networks existing in the capital Addis Ababa. As such, it is not surprising to find that suppliers of network technology devices and support in Ethiopia are restricted to only a limited number of vendors.

Core Layer

Simplicity of Installation and Maintenance To design the LAN architecture we have therefore selected the hierarchical model. This enables us to design and arrange the inter-network devices in layers. Figure 1 depicts the hospital-based LAN architecture. It is a model preferred by most of network design experts for its ease of understanding, expandability and improved fault isolation characteristics (http://www.cisco.com). The model encompasses the following three layers: (a) first

40

Core layer high performance switches, capable of switching packets as quickly as possible, are to be deployed. Essentially, this layer connects the LAN backbone media as well as connects to the outside world via a firewall through WAN. In this design, the devices in the core layer will be placed at a central location in the hospital. The devices will then be connected with high-speed cables such as fiber optics, or fast Ethernet cables. The servers will also be connected to switches, shielded by a firewall.

Distribution Layer Distribution layer will contain switches and routers capable of Virtual LAN (VLAN) switching and allow defining departmental workgroups and multicast domains. The devices should also support connectivity of different LAN technologies since they also serve as the demarcation point between the backbone connections in the core layer and the access layer. In this hospital-based LAN design, the distribution layer represents switches/routers at each building connected to the core layer on the one end and to the access layer on the other. Redundant links will be used for maximum availability and the departments could be grouped forming their own Virtual LAN.

Access Layer Access Layer is where the end-users are allowed into the network. This layer contains switches/ hubs from which PCs in each department gain access to the hospital-based LAN. Each department will have at least one switch/hub, which will in turn have redundant links to more than two of the switches in the distribution layer.

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 1. Hospital LAN design

WAN architecture Designing the WAN architecture for a nationwide e-medicine network raises the issue of WAN service provider. Unlike LAN, WAN connectivity depends on the availability of WAN infrastructure in the country. The sole WAN service provider is the Ethiopian Telecommunications Corporation (ETC). ETC provides a number of services (http:// www.telecom.net.et) from which the WAN infrastructure suitable for the e-medicine network may be derived. Existing WAN services include: (a) Internet Services, or, providing basic Internet services over dial-up or leased lines; (b) Digital Data Network (DDN), supporting dedicated Internet, ISDN and frame relay services; (c) SchoolNet VSAT, covering services for secondary schools and institutes of higher learning; (d) WoredaNet VSAT, covering services for districts (Woreda) administrations; and (e) Broadband Multimedia Network (BMN), offering high-speed optical communications to major cities.

To choose among these possible infrastructures for nationwide e-medicine network, the parameters to be considered include the geographical coverage, bandwidth, mode of communication, rental cost of WAN connection and capacity to add more LANs. Table 1 summarizes the data comparing among the available ICT infrastructures in Ethiopia. Based on the data, it appears that WoredaNet is best suited to the national e-medicine network, as long as the existing infrastructures are functioning efficiently and effectively. However, as noted in Table 1, there may be a tradeoff between coverage and capacity, that is, when the coverage is acceptable the capacity may be somewhat limited. For example, BMN coverage is ideal as it represents state-of-the-art service and higher bandwidth. However, it is centered primarily in the urban areas. It is also under development and we have thus considered it as a potential option to be used when integrated with the VSAT-based networks to enhance nationwide e-medicine network. Finally, the SchoolNet needs to be upgraded to support two-way interactivity.

41

Envisioning a National e-Medicine Network Architecture in a Developing Country

Table 1. Summarized comparison of existing ICT infrastructure Internet

Coverage

Bandwidth

Interactivity

Cost

Capacity to scale

DDN

SchoolNet

Telephone coverage areas only

The capital and regional Urban areas only

About 500 schools covered. There are Woredas that do not have schools

571 Woredas out of 594 are covered

The Capital city and 13 regional towns.

Maximum of 56k dialup and 1Mbps in Leased line

Maximum of 1Mbps

Can be upgraded to 384k upstream

Downstream/ upstream 45Mbps/ 256k downlink

ADSL Services: Variable bandwidth Downstream/ upstream 512k/128k and 1024k/256k

Two-way

Two-way

One-way broadcasting

Two-way

Two-way

Free for schools

Free For Woredas

Not yet determined, under development

Will have more than 10 ports free at each Woreda

Can be expanded

0.11 birr/min dialup 1000 birr/ month leased line

Not scalable

Not scalable enough

Thus, one alternative approach is using a combination of VSAT networks and terrestrial BMN. VSAT-based connectivity is believed to be cost-effective and in the case of WoredaNet and SchoolNet, it enables connectivity to the public, even in the rural areas. In addition to serving the rural areas, it also covers urban areas. Together, this will provide modern, convenient as well as economical connectivity to hospitals. For improving state-of-the-art applications such as videoconferencing, connectivity via the emerging BMN is proposed to connect urban area hospitals in the capital city and in the regions where the network can be easily accessible. Figure 2a shows the e-medicine network as a first alternative. Note that this approach requires that urban hospitals maintain two WAN connections. Having more than one WAN connection, however, may become expensive in the long run. If the two WAN infrastructures could be integrated,

42

WoredaNet

BMN

an improved WAN design will result with only one WAN connection to the urban hospitals through which the hospitals will be connected to BMN and the rural area clinics through the WoredaNet. Figure 2b depicts the second alternative solution of the WAN design.

current etc InItIatIVes The recent development in the ETC in providing multimedia network infrastructure is the integration of the VSAT-based networks (SchoolNet and WoredaNet) with the BMN (Tiruneh, 2006). In other words, these VSAT-based networks can now be used as a point of access to the BMN. As part of a longer term vision and mission of broadband initiatives for socioeconomic development in Ethiopia, ETC has also planned an e-health setting that tries to cover rural areas,

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 2a. Logical WAN design based on BMN and VSAT networks (1st Alternative)

schools, clinics, hospitals, prisons, and nursing homes, including assisted living with several requirements: (a) high quality patient data, video and images to be exchanged between different medical institutions; (b) ICT infrastructure to connect geographically dispersed institutions,

nationwide or worldwide; (c) infrastructure that supports data, video and voice/audio (multimedia) services; (d) high quality, secured and fast delivery of medical information; and (e) high speed (BW) connectivity or the deployment of broadband infrastructure.

Figure 2b. Logical WAN design based on BMN and VSAT networks (2nd Alternative)

43

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 3. ETC’s broadband-integrated infrastructure for e-health applications

Key challenges faced are the need to encompass multiple locations, to use multiple access technologies, to deliver multiple services and to address multiple user markets.

Current BMN Development ETC has already completed building the Core Terrestrial Broadband Infrastructure, which is capable of providing data, video, and voice services with 24 Points of Presence. This infrastructure services key business sites (urban areas) and supports Multiple Broadband Access via ADSL, FWA, WIFI, and fiber networks.

Current VSAT Development In Ethiopia, a Broadband VSAT Network platform, which supports integrated services such as video, data, Internet, and voice on a single infrastructure, is currently in place. It has countrywide coverage (450+ schools and 550+ woredas) as part of SchoolNet and WoredaNet deployments. It is integrated with the core multimedia network,

44

also serving as broadband access means. Figure 3 shows the two recent developments. Apparently, the new developments in WAN infrastructure support the 2nd alternative emedicine WAN architecture discussed here as depicted in Figure 3.

tHe prototype Based on the specified network requirements and architecture, a working prototype for the national e-medicine network is now presented and its operations highlighted. The prototype is a basic e-Medicine service (BEMS), which provides a Web-based graphical user interface (GUI) for health providers. BEMS facilitates the information exchange between remotely located health providers for the purpose of e-consultation, as well as for maintaining electronic patient information. The traditional paper-based forms and patient cards used in the hospitals will be digitized and reproduced electronically. Web-based technology is chosen for its

Envisioning a National e-Medicine Network Architecture in a Developing Country

ubiquity. Using Web-based technology constitutes not only a network that can be used universally, but the technology also supports system-independent platforms, thus providing access to many different computer systems at client sites (http://java.sun. com/products/jsp). Key requirements in these client sites are simply the availability of Web browser software and network connectivity. For a secured network, password protected system ensures that user login is needed to access capabilities of the system. In addition, user types are defined so that there will be a role-based access to database and system functions in BEMS. To ensure compatibility with most legacy systems, a relational database is advocated for storing user and patient information. Beyond e-mails, this approach allows the users to mobilize structured information exchange among the communicating health providers.

Major Features of BeMs Basically the BEMS prototype may be conceptualized as a database-driven Web site with the following main features and functions: (a) providing user management services where administrator can register users, assigning username and password, and defining user type, as well as searching and editing user information; (b) providing patient management services where health providers can register patients, search patients and view patient information on a Rolodex-like interface, as and when necessary; (c) providing, on the one hand, referral systems where physicians can write referral messages to a particular department and hospital, and, on the other, a system whereby a physician can retrieve and study the list of referrals forwarded to the department s/he is working in and allowing the physician to write feedback instantaneously after examining the referral message and patient information; (d) providing a system by which physicians can request and schedule lab test at any hospital laboratories so that patients can get tested in the clinic/hospital

they are being treated; and (e) providing a list of lab test requests to laboratory technicians and allowing them to input lab test results.

BeMs architecture The BEMS architecture is built on three-tiered, client-server architecture. The first layer is where the client machines run Web-browser software. This layer is used to display the user interface (Web pages) of the system and send secure HTTP request to the Web server in the second layer. Along with the Web server, application server resides in the second layer. This application server manages the clinical business logic. The bottom layer contains the persistent data of the system. All data of patients, physicians and other communicated messages will be stored and maintained in this third layer. This layer runs the database management system (DBMS) software. Put simply, BEMS functions as a Web-based application connected to a Web server to provide all the interfaces of the system and that of a database server to manage all the knowledge and information elements stored in the system. Figure 4 charts the BEMS system architecture. The BEMS prototype is constructed with a combination of open source products and freely available software components. The Web server suggested is the Apache Jakarta’s Tomcat Web server (http://jakarta.apache.org) with the func-

Figure 4. BEMS prototype architecture Web Browser HTTP

Web Server

Application Server

Database

Mysql:JDBC

45

Envisioning a National e-Medicine Network Architecture in a Developing Country

tionalities as well as the mandated business rules programmed in Java (Haile-Mariam, 2002). Java Server Pages (JSP) is used to capture the user interface and the text of Web pages (http://java. sun.com/products/jsp; http://www.coreservlets. com). Some scripting is included on the Web pages in JavaScript. JSP has a capability to import java classes and run them from the Web pages when the pages are downloaded to the client machine (http:// www.coreservlets.com). Unlike other server side languages such as Active Server Pages (ASP), JSP makes the system platform independent. It also allows users to take advantage of the full power of java programming language which overcome some of the limitation of other scripting languages such as PHP (http://www.coreservlets.com). The database conceptualized is the open source MySql to back up the database driven application. MySql works on many different operating system platforms and is known for its speed of data retrieval (http://www.mysql.com). It provides Application Program Interfaces (API) for many programming languages including Java. Passwords are secure because all password traffic is encrypted when connecting to the MySql server. For database connectivity, we use mm.mysql driver, which is a Java Database Connectivity (JDBC) driver, from MySQL AB, implemented in 100% native Java (http://www.mysql.com/ products/connector-j).

database design Issues BEMS needs to keep track of information about patient and related medical records, user’s information, and messages for both medical referral requests and feedbacks. A well-designed minimal database is needed to manage this information. A relational database model is selected to store the persistent data of the system, as it could be easier to manage, and provides better management for complex query of such data (Amenssisa & Dabi, 2003). This database is expected to maintain and

46

manipulate basic entities such as users, patients, and medical records. Each component of the medical record of a patient is an aggregation of different types of data, which are stored in the database. In the traditional paper-based system, the medical record of a patient is identified by an Out Patient Card (OPCard) Number, which is usually called patient record number. OPCard is a four-page hard-paper card, which contains patient’s generic information, such as name, sex, age, address on the first page and a table of two columns for date and clinical note so as to record chronologically the compilation of health providers’ notes. All other components such as laboratory test results and x-ray reports, among other pieces of information, are stored inside the hard-paper card referenced by the card number or name of the patient. The lab test results may contain zero or more test request forms along with the results for Urine, Parasitology, Blood Chemistry, Hematology, Serology, Bacteriology, Fine Needle Aspiration Cytology and Biopsy. When a patient is admitted to the hospital, admission and social services information is stored. The admission data include identification information and name and address of next-of-kin, marital status, and number of siblings (children) information, besides occupational information and other demographics, as and when provided by the patient. Subsequently, follow-up data such as vital sign measurements, fluid balance information and other measures will be collected and recorded. Order sheet, which contains a list of treatments to be ordered following admission, is also part of the inpatient medical record. In addition to these, information about the hospitals, departments and laboratories are also stored and captured in respective entities. To minimize connectivity cost and increase system performance, a distributed database is recommended. Horizontal partitioning that splits tables along rows, based on the location of patient and healthcare facility, is seen to be an ideal choice in the e-medicine application that tries to create

Envisioning a National e-Medicine Network Architecture in a Developing Country

nationwide connectivity. Finally, to use the database, transparent data access schemes must be defined for applications that run over the network.

Figure 5a. Administrator’s main page within BEMS

BeMs Interfaces In this part of the discussion, the design of various BEMS interfaces is presented. BEMS is accessed when opening the initial Web page where user authentication is first performed. The initial page contains a typical login screen for specifying username with authenticated password. There is no need of menu or different buttons to be submitted based on the user types. Since the user types are defined in the database when the user registered, the page corresponding to the specific user type will automatically be opened upon successful login. Currently, administrators, physicians, and lab technician user types are defined and all of these user types will have their own main pages as described below.

Figure 5b. User registration page

Administrator’s Main Page The administrator’s main page is used for managing users. The functionalities accessible from this page include: register new user and search user by a combination of name, father’s name, and user name. Figure 5a shows the administrator’s main page whereas Figure 5b depicts the user registration page. The other function provided to administrators is the search user function. It is possible to search users by keying in any combinations of name, including father’s name and/or user name. Note that username is treated as “unique” in the user table with the search result being quickly displayed. Although not shown here, the full name of the search result is captured as a link. This link then leads to a page containing the relevant user information from which the administrator can edit a particular user.

Physician’s Main Page The physician’s main page contains a button to open the patient manager page and has the capability of retrieving a list of referrals forwarded to the department where the physician is located. To open a new page, the physician is free to click on the manage patients button or link to one of the referrals. Figure 6a illustrates that the patient, Ato Andualem Lemma, is one of Dr. Aman’s referrals. If Dr. Aman chooses to treat this patient in the hospital where he is privileged, he can simply

47

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 6a. Physician’s main page within BEMS

Figure 6b. Patient manager page within BEMS

open the patient manager page by clicking on the manage patients button. Figure 6b shows that the patient manager page has two options, that is, physicians can “register new patient” or “search patient” for those who were previously registered. When selecting “register new patient,” a patient registration form, similar to the user registration page, will be opened. In contrast, if the physician wants to look for a patient, s/he can input one of the search criteria such as name or record number of the patient, resulting in the display of a matching record number or name and the hospital where the patient was first

48

registered. The patient full name is a link that leads to the patient information page similar to the traditional patient card used in the hospitals. An example of the patient card, which opens up when the full name link in the previous interface is selected, can be seen in Figure 6c. The patient card contains patient’s general information, address information and clinical notes that are ordered in descending order. In addition to the information displayed on the patient card, laboratory test results and medical images related to the patient are accessible by clicking corresponding buttons from the patient

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 6c. Patient card page within BEMS

card interface. The physician can add clinical notes, refer, and/or admit the patient. From the physician’s main page, the other option available to the physician is to see referrals forwarded to his or her department. This is possible by clicking the link that opens a referral page corresponding to the patient. The patient referral page contains the referral messages and buttons that will lead the user to patient information as well as a button that can lead to the feedback input page. If the physician user wants to view the patient information, the view patient card button will support such a function. Otherwise, if the physician would like to give feedback to the referral using the feedback slip, the open feedback slip button will serve this purpose. Basically, the feedback slip is represented as an input form similar to a traditional form for capturing feedback information related to the current referral. Put simply, the idea here is to mimic within a virtual environment what the physicians have already become accustomed to routinely when working within a

paper-based environment, which will only hasten the processing and matching of the stored computerized information for them.

Graphical User Interface Apparently, human computer interface (HCI) design issues are critical in determining the successful deployment and continuing use of the BEMS prototype. As indicated, the current approach attempts to optimize the interface design in mimicking more or less the traditional forms, documentation formats and paper-based patient record system that the physicians have grown accustomed to using over the years. In other words, this ensures that the BEMS supports the habits of the physician users. It will also serve to preserve physician user satisfaction and promote a high rate of acceptance among physician users with the new system implementation on the one hand while reducing disruption to the care processes on the other.

49

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 7. Laboratory information and parasitological test request page of BEMS

Nonetheless, new system development such as the BEMS typically provides new opportunities for revisiting the care processes that have been put in place over the years. Elimination of redundant processes as well as the need to streamline certain administrative and clinical processes may be warranted to improve quality, cost, efficiency and effectiveness of the care provided. Online requests of patient and referral information and querying of databases are expectedly translatable into more efficient, effective, appropriate and quality care. Use of GUI also permits substantial amount of data to be viewed together, improving the communication, exchange and sharing of patient data. Feedback from physician users should ultimately be channeled to an even more enhanced user interface design. In this environment, clinical test results and specialist reports can also be captured quickly and shared collaboratively among all relevant health providers. As an example, the physician can be empowered to request laboratory test results in BEMS by viewing the patient laboratory information page, which is accessible from the patient card page by a button called “Laboratory Tests.” Figure 7 provides illustrative screenshots for the laboratory information page and a parasitological test request pages. When a physician wants to request lab tests, instead of writing a prescription to the patient and having the patient wait for further

50

scheduling information from the nursing clerk, all that is needed now is a mere click of the button corresponding to the type of “test” required from the laboratory information page. The specific lab test request page can be designed such as to provide the physician user with a dropdown list from which the appropriate lab test can be immediately selected and performed as scheduled. This was found to be important in order to forward the lab test request on a real-time basis to the other user types called, the Lab technicians.

Lab Technician’s Main Page The third type of user, laboratory technician, sees a list of laboratory requests to the department s/he is practicing, on the lab technician’s main page. The list contains a link to open lab test result input form where the lab technician can enter his or her report following the test as shown in Figure 8.

ConclusIon The key contribution of this effort lies in envisioning the planning and detailing of a Web-interface for a hierarchical model-based LAN architecture that enables the integration of the inter-network devices in layers and a WAN architecture, both

Envisioning a National e-Medicine Network Architecture in a Developing Country

Figure 8. Lab technician’s main page of BEMS

fine-tuned for a developing country such as Ethiopia. The hierarchical model adopted for the LAN is a preferred model due to its ease of expandability and improved fault isolation characteristics. The WAN design considers the existing VSAT-based WAN infrastructure in the country, the WoredaNet. Even if urban areas are relatively better equipped with adequate ICT technologies such as Internet access and digital telephone networks, the communication infrastructure is not well developed in many rural areas. These regions have to be equipped with an access to urban areas. In this context, the newly emerging state-owned, low-cost VSAT networks such as SchoolNet and WoredaNet provide the rural areas with suitable means of communication with urban areas and beyond. VSAT, an earthbound station used in satellite communications of data, voice and video signals, excluding broadcast television, comprises two parts: (a) a transceiver that is placed outdoors in direct line of sight to the satellite; and (b) a device that is placed indoors to interface the transceiver with the end user’s communications device, such

as a PC. The transceiver receives or sends a signal to a satellite transponder in the sky. The satellite sends and receives signals from a ground station computer that acts as a hub for the system. Each end user is interconnected with the hub station via the satellite, forming a star topology. The hub controls the entire network operation. For one end user to communicate with another, each transmission has to first go to the hub station that then gets retransmitted via the satellite to the other end user’s VSAT. VSAT can handle up to 56 Kbps. More importantly, the BEMS architecture discussed here is designed to integrate with a large part of the existing LAN and WAN infrastructure designs. The system can then be used to facilitate both intra- and inter-hospital communications and for all forms of information exchange. The alternative design selected will not only improve quality of healthcare services while protecting the privacy, confidentiality and integrity of sensitive patient information, but its interfaces have been set up to mimic the physician routines working in a paper-based environment. Moreover, this will also yield opportunities for further review of the paper flow and work processes to cut down on

51

Envisioning a National e-Medicine Network Architecture in a Developing Country

redundancies and errors while simultaneously boosting both administrative and clinical efficiencies and effectiveness of care. As future work in the area of developing nationwide e-medicine networks, we recommend the following considerations: (a) the intended network should support real-time e-consultations via video and audio conferencing, advocate doctor-topatient interactions, and facilitate remote training for health professionals; (b) it should also support a distributed database structure, where individual hospitals should keep their own databases, which can be further treated as one “huge” database; (c) the definition of standards is essential to facilitate information exchange among private and government hospitals as well as overseas; (d) the integration of expert systems such as case-based system where doctors can query the database to get experience from previously stored similar cases should also be considered—such a system will aid future physicians and residents working anywhere in the country to learn from past successes and/or failures of the attending specialist(s), especially for non-trivial and complex patient cases; and (e) the infrastructure should be independent of chosen platform and operating systems (e.g., Windows vs. Apple) and be able to support physicians needing to remotely monitor their patients over heterogeneous networks, including handheld devices in 2G/3G mobile networks and wirelessly. Beyond the design of a nationwide e-medicine infrastructure, there will be a host of potential e-medicine applications that may be supported, including, but not limited to a series of healthy lifestyle promotion programs such as e-consultation and tracking of participation in smoking cessation, weight reduction, dental health, stress reduction, exercise and nutritional programs and many more. In this regard, one of the contributing authors is actively and precisely engaged with a growing network of researchers in generating such a series of educational modules intended for seniors and other population groups that will eventually be

52

serviced on a network such as BEMS discussed in this article (Tan, 2005).

reFerences Amenssisa, J. & Dabi, S. (2003). District-based telemedicine project in Ethiopia. Ministry of Health. Addis Ababa, Ethiopia. Apache Jakarta Project. (2007). Retrieved from http://jakarta.apache.org Cisco Documentation. (2007). Retrieved from www.cisco.com Elmasri, R. (2000). Fundamentals of database systems, 3rd edition. Addison Wesley. Ethiopian Telecommunications Corporation. (2007). www.telecom.net.et Haile-Mariam, A. (2002). Renaissance: Strategies for ICT Development in Ethiopia. M.Sc Thesis, School of Engineering Postgraduate Engineering Program. Hall, M. (2007). Core servlet and Java server pages. Sun Microsystems Press, Retrieved from http://www.coreservlets.com. Horsch, A. & Balbach, T. (1999). Telemedicine information systems. IEEE Trans. Inform., Technol. Biomed., 3, 166-175. Java Server Pages Documentation. (2007). Retrieved from http://java.sun.com/products/jsp. Kirigia, J., Seddoh, A., Gatwiri, D., Muthuri, L., & Seddoh, J. (2005). E-health: Determinants, opportunities, challenges and the way forward for countries in the WHO African region. BMC Public Health, 5, 137. MySQL. (2007). http://www.mysql.com. MySQL Connector/J. (2007). Retrieved from http://www.mysql.com/products/connector-j/.

Envisioning a National e-Medicine Network Architecture in a Developing Country

Perednia, D. & Allen, A. (1995). E-medicine technology and clinical applications. Journal of the American Medical Association, 273(6), 483-488. Tan, J. (2001). Health management information systems: Methods and practical applications, 2nd edition. Gaithersburg, MD: Aspen Publishers, Inc. Tan, J., Kifle, M., Mbarika, V., Okoli, C. (2005). E-Medicine in developed and developing countries. In J. Tan (Ed.), E-healthcare information systems. Jossey-Bass (A Wiley Imprint).

Tanenbaum, A. (2004). Computer networks, 4th edition. Prentice Hall, Inc. Tiruneh, M. (2006). ETC, broadband network infrastructure for e-health. ICT-H-2006 Workshop, Addis Ababa, Ethiopia. Wootton, R. (2001). Recent Advances: Telemedicine. British Medical Journal, 323, 557-560 Wright, D. (1998). Telemedicine and developing countries. A report of study group 2 of the ITU development sector. J Telemedicine Telecare.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 1, edited by J. Tan, pp. 44-62, copyright 2008 by IGI Publishing (an imprint of IGI Global).

53

54

Chapter 4

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR: One Facility’s Approach Karen A. Wager Medical University of South Carolina, USA James S. Zoller Medical University of South Carolina, USA David E. Soper Medical University of South Carolina, USA James B. Smith Medical University of South Carolina, USA John L. Waller Medical University of South Carolina, USA Frank C. Clark Medical University of South Carolina, USA

aBstract Evaluating clinician satisfaction with an electronic medical record (EMR) system is an important dimension to overall acceptance and use, yet project managers often lack the time and resources to formally assess user satisfaction and solicit feedback. This article describes the methods used to assess clinician satisfaction with an EMR and identify opportunities for improving its use at a 300-physician academic practice setting. We administered an online survey to physicians and nurses; 244 (44%) responded. We compared physician and nurse mean ratings across 5 domains, and found physicians' satisfactions scores were statistically lower than nurses in several areas (p<.001). Participants identify EMR benefits and limitations, and offered specific recommendations for improving EMR use at this facility. Methods used in this study may be particularly useful to other organizations seeking a practical approach to evaluating EMR satisfaction and use. Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

IntroductIon The degree of interest and momentum in furthering the widespread adoption and use of electronic medical record (EMR) (or electronic health record systems) is at an all time high in the United States. Healthcare providers, purchasers, payers and suppliers are all looking to the EMR as a tool to help promote quality, enhance patient safety, and reduce costs. Despite this energy, recent estimates indicate EMR adoption rates in ambulatory care remain in the 15-20% range (Hillestad et al., 2005). Cost, lack of uniform interoperability standards, limited evidence showing use improves patient outcomes and clinician acceptance are among the barriers to widespread EMR adoption (Bates, 2005). Those who have overcome these initial hurdles and made the transition from a paper-based medical record system to an EMR often lack the time, resources and expertise to evaluate the system’s impact on the organization, including clinician use and satisfaction with the system (Anderson & Aydin, 2005; Wager, Lee, & Glaser, 2005). Use and satisfaction are two key measures of the success of any information system (DeLone & McLean, 2003), including EMR system success (Anderson & Aydin, 2005). Various researchers have assessed physician use and satisfaction with the EMR (Sittig, Kuperman, & Fiskio, 1999; Gadd & Penrod, 2001; Penrod & Gadd, 2001; Likourezos et al., 2004; Joos, Chen, Jirjis, & Johnson, 2006) and some have found that user group perspectives can differ even within the same institution (Wager, Lee, White, Ward, & Ornstein, 2000; O’Connell, Cho, Shah, Brown, & Shiffman, 2004b; Hier, Rothschild, LeMaistre, & Keeler, 2005). Assessing user reaction to the human-computer interface is also an important dimension (Sittig, Kuperman, & Fiskio, 1999; Despont-Gros, Mueller, & Lovis, 2005) in evaluating EMR satisfaction. Likewise, the timing of the evaluation study is important. Conducting formative evaluation studies during enterprise EMR implementation can be particu-

larly useful in identifying perceived problems and making adjustments to the implementation plan or reallocating resources as needed (Friedman & Wyatt, 1997; Burkle, Ammenwerth, Prokosch, & Dudeck, 2001; Anderson & Aydin, 2005; Brender, 2006). Our study was designed to assess physician and nurse use and satisfaction with an enterprisewide ambulatory care EMR. We incorporated into the evaluation a means of assessing physician and nurses’ reactions to the human-computer interface and also solicited their input and suggestions on how to improve the system’s usefulness to our organization. This article summarizes the methods used, findings, and relevance to other organizations in the throes of implementing and evaluating EMR acceptance.

BacKground The Medical University of South Carolina (MUSC) in Charleston, South Carolina has implemented an electronic medical record (EMR) system, known as Practice Partner Patient Record® in the majority of its ambulatory clinics over the past few years. Although EMR use is not new to MUSC (family medicine has used the system since the early 1990s and internal medicine since the mid-1990s), it was not until February 2004 that we secured funding to deploy the EMR throughout the ambulatory care enterprise. When the system is fully implemented, paper medical records will have been replaced, and both primary care and specialty providers will share a single electronic medical record for their patients. This message was conveyed from the top down, starting with senior leadership and the dean of our medical school. The EMR product itself has many of the attributes of a typical EMR system, including electronic health data capture, results management, decision-support, and electronic communications. However, MUSC has not yet installed enterprise-wide direct order entry nor activated

55

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

the preventive care reminder functions. Diseasespecific progress note templates are available for facilitating program note entry; however, direct data entry is not a requirement for using the system, and transcribed notes can also be incorporated. The system also includes a prescription writer, which allows prescriptions to be entered and tracked in the patient record while automatically checking for interactions with other medications and documented allergies. Patients receive printed prescriptions at the time of the visit; the prescriptions are not yet sent electronically to the pharmacy. We provided formal training sessions to physicians, nurses and administrative support; however, nurses and administrative staff received at least four hours of initial training, while most physicians only had 30-45 minutes of training. Nurses and administrative staff were trained in small groups in a classroom setting with computer workstations for each individual. Physicians were offered several options for training, including small group sessions, but the far majority opted for one-on-one training with a clinical analyst. Knowing that physicians’ schedules were harried, we decided to provide more intensive training to nurses and administrative staff. We felt if nurses and administrative staff were comfortable with the system, they could assist physicians in their respective departments as needed. The purpose of the study was to gain insight into user satisfaction and determine if there were differences in user group satisfaction, and solicit formal feedback on how to improve the system’s use at our organization. We defined ‘user’ as any MUSC employee or student who had been issued an EMR login and password. This article describes the methods used to conduct the survey and the results from our attending physician and nurse user community.

56

MetHods survey Instrument We conducted preliminary interviews with physicians and nurses regarding their views and experiences with the EMR, but were interested in soliciting input from all EMR users at MUSC. Given the size of the user group, we decided to adopt a survey approach. We adapted the Questionnaire for User Interaction Satisfaction (QUIS), developed by researchers at the University of Maryland, for use in this study after reviewing the literature, consulting with medical informatics professionals and soliciting input from EMR physician leaders at MUSC. The QUIS has been tested as a valid and reliable instrument in settings similar to MUSC (Chin, Diehl, & Norman, 1988) and was used recently by researchers affiliated with Brigham and Women’s Hospital to assess physician satisfaction with its internally developed outpatient EMR (Sittig, Kuperman, & Fiskio, 1999). The QUIS has also validated for online administration (Slaughter, Harper, and Norman, 1994). The instrument assesses user satisfaction in five major domains, with four to six questions in each: (1) overall user reactions, (2) screen design and layout, (3) terminology and system messages, (4) learning, and (5) system capabilities. Recognizing that participants might respond differently depending upon whether they were assessing how long it takes to open the application or navigate within it, we separated the item “system speed” survey component into two statements—system launch speed and navigation speed. We asked participants to rate each question on a scale from 0 (the lowest) to 9 (the highest) level of satisfaction. We included several additional items in the survey: (a) demographic data (e.g., position, age, gender), (b) use of the EMR (whether provider dictated notes or directly entered them into EMR), and (c) three open-ended questions, in which participants were asked to identify the three greatest

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

benefits or advantages and three greatest limitations or disadvantages to using the EMR and any recommendations they have for improving the use and effectiveness of the EMR within their clinic or at MUSC. The items and scale from the QUIS were kept intact. A draft survey was sent to a pilot group of physicians, nurses and administrative staff. Suggested changes were incorporated into the final version of the survey. Institutional Review Board approval was obtained prior to conducting the study.

user population and survey administration We sent an e-mail message from one of the authors (who happened to serve as chair of the Physician Information Council) to all attending physicians and nurses listed in the EMR user database. Our sampling frame included all individuals who had been issued an EMR username and password—for a total population of attending physicians (245) and nurse EMR users (304) of 549. In the e-mail message, participants were asked to click on an embedded hyperlink to complete the online questionnaire. We gave participants the option of printing and faxing the completed survey to us (48 surveys were returned by fax). Two reminder notices were sent—one week and two weeks following the initial mailing. All participants were assured confidentiality of their responses.

Data Analysis Survey responses were downloaded from the survey provider to an electronic database and then imported into SPSS for analysis (Windows, 2005). Responses to open-ended questions were grouped and assigned to categories by the primary author, experienced in qualitative research. These were then tabulated within each of the categories and discussed among the participating authors.

results demographics of participants and their use of eMr system We received 244 completed surveys from attending physicians and nurses for an overall response rate of 44%. Forty-seven (47%) of physicians and 38% of nurses responded. 32% of participating physicians and 98% nurses are women; 85% are between the ages of 30-60, equally distributed within each decade of life. Nearly 37% of physicians and 19% of nurses reported having prior experience with other EMR systems, and 62% have used the EMR at MUSC for at least one year. When physician participants were asked how often they dictate patient information that is eventually transcribed into the EMR, 39% reported frequently (defined as more than 50% of the time), 4% sometimes (25-50% of the time) and 58% reported rarely or never (defined as less than 25% of the time) (n=109). Almost 40% of participants reported having paper medical records pulled daily for patient visits, 14% sometimes, and 44% rarely or never (n=108). 70% reported that they access the EMR remotely on at least a weekly or daily basis.

Satisfaction with EMR We assessed the internal consistency reliability of the five domains of the QUIS part of the survey (Table 1) and examined mean satisfaction scores for attending physicians and nurses for each item Table 1. Reliability of domains (Crohbach’s Alpha) Domain

Alpha

Overall user satisfaction Screen design and layout Terms and system information Learning System capabilities Overall

.911 .859 .891 .876 .835 .961

57

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

Table 2. Physician and nurse mean satisfaction scores by domain/item

System Capabilities

Learning

Terms and System Information

Screen Design and Layout

Overall satisfaction

Domain

Item (0-9 scale)

Physician

Nurse

Terrible…………………………………….……………………....................….Wonderful

5.07

5.92*

Difficult……………….…………………………………….……………….……….Easy . Frustrating……………………………………..……………………………...…Satisfying

5.17

6.21*

4.33

5.09*

Inadequate Power............………………………………...………………..Adequate Power

4.45

5.02

Dull…………………..……………………………………………………….…Stimulating

4.65

5.39*

Rigid…………………………………………………………...……………..…….Flexible

4.42

5.41*

Overall satisfaction composite score

4.67

5.50*

Characters on the screen: Hard to read……….………………………………Easy to read

6.67

7.16

Highlighting on the screen simplifies task: Not at all…………………….…....Very much

5.49

6.54*

Organization of information on screen: Confusing………….……………….…Very clear

5.45

6.17*

Sequence of screens: Confusing………………………………………………...Very clear

5.56

5.81

Use of terms throughout system: Inconsistent…………………………….……..Consistent

6.09

6.58

Computer terminology is related to the task you are doing: Never………………...Always

5.89

6.31

Position of messages on screen: Inconsistent………..………………….……….Consistent

6.05

6.48

Messages on screen which prompt user for input: Confusing………………….Very clear

5.42

6.14*

Computer keeps you informed about what it is doing: Never……..……………….Always

4.90

5.42

Error messages: Unhelpful……………………………………………….….……...Helpful

3.22

4.19

Learning to operate the system: Difficult………………………………….………….Easy

5.23

5.98*

Exploring new features by trial and error: Difficult……………………………….….Easy

4.73

5.76*

Remembering names and use of commands: Difficult…………….……………..…..Easy

4.93

5.51

Tasks can be performed in a straight forward manner: Never………………….….Always

5.06

5.74*

Help messages on the screen: Confusing…………………………….………………Clear

4.14

5.12*

Supplemental reference materials: Confusing………………………………..………Clear

4.10

5.45*

System speed (to open or launch program): Too Slow………………………Fast Enough

2.65

2.85

System speed (to navigate within EMR, e.g. open a note): Too slow………...Fast enough

4.41

4.13

System reliability: Unreliable…………………………………………………..….Reliable

4.72

5.01

System tends to be: Noisy…………………………………………..……………….Quiet

7.23

7.87*

Correcting your mistakes: Difficult…………………………………………………..Easy

5.12

4.96

Experienced and inexperienced users’ needs are taken into consideration: Never………………………………………………………………………………Always

4.68

5.31

*significant at the p>.05 level

58

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

within the five domains. Overall Crohbach’s alpha was 0.961. Attending physician and nurse mean scores are shown in Table 2. Items were rated on a scale of 0 to 9, with 0 being the most negative, 9 the most positive. Both groups gave low ratings to system launch speed and clarity of error messages. Items rated most positive by both groups included system noise, ease in reading characters, consistency of terms, and the clarity of messages appearing onscreen. We calculated overall satisfaction scores for physicians and nurses using the average of the six items in the overall user reaction domain as a proxy. Overall satisfactions scores were 4.67 and 5.50 (t-test, p < .001) for attending physicians and nurses, respectively. We compared attending physician ratings of the other 22 individual items with those of nurses and found that the physicians’ mean ratings of the EMR were statistically significantly lower in nine of the 22 items (p<.05). Using one-way ANOVA, we found no difference in overall satisfaction between experience groups, less than 6 months, 6-12 months, 1-2 years, and more than 2 years. We also ran a standard multivariate regression with overall satisfaction score as the dependent variable and the 22 remaining items as independent variables and found two items were significant predictors of overall satisfaction—(1) task can be performed in a straightforward manner (p<.001) and (2) clarity of help messages on screen (p<.01).

perceived eMr Benefits, limitations and recommendations for Improved use Participants were asked three open-ended questions: • •

•

What have you found to be the three greatest benefits or advantages to using the EMR? What have you found to be the three greatest limitations or disadvantages to using the EMR? What recommendations do you have for improving the use and effectiveness of the EMR within their clinic/area at MUSC?

Benefits/Advantages to the EMR System Nearly 82% of physicians and 56% of nurses identified availability and accessibility of the patient’s record as a major benefit or advantage to using the EMR. See Table 3. Included in this category were comments related to having access to other clinic notes and records and having remote access to the EMR. One physician commented, “Everyone has access to the entirety of the record across all disciplines.” Another noted, “The EMR greatly facilitates communication among primary care providers and specialists by having all patient information in one place.” Other frequently cited benefits included (a) quality of record/documen-

Table 3.Most frequently cited benefits/advantages to using MUSC’s EMR system Physicians n=114

Nurses n=130

Availability and access to patient record

81.6%

56.2%

Quality of record/documentation

36.0%

14.6%

Comprehensiveness and completeness of record

33.3%

25.4%

Positive impact on efficiency

13.2%

10.0%

Continuity of care and standards of care

13.2%

10.8%

Ease of use

12.3%

15.4%

Avoidance of perils of paper

9.6%

20.0%

EMR Benefit

59

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

tation (legibility, data consistency and accuracy, organization of record, coding/compliance); (b) comprehensiveness and completeness of patient record (e.g., medications, results from diagnostic tests); (c) impact on efficiency (e.g., speed, timeliness in completing notes);(d) continuity of care and standards of care (e.g., templates, continuous outpatient record); and (e) avoidance of perils inherent with paper records (e.g., no lost charts, no moving charts).

Limitations/Disadvantages to the EMR System Limitations or disadvantages identified by at least ten participants are listed in Table 4. The two most-cited limitations or disadvantages of the EMR were speed and cumbersome user interface. Both physicians and nurses described the system as being too slow. Approximately 20 participants mentioned that the start time to open the application was particularly problematic. System downtime, including the moments when the “system crashes”, “locks up or freezes” and requires user to “reboot” were described as frustrating by 14% of physicians and 22% of nurses. One physician commented that system crashes are frustrating: “The program froze three times today while I was editing the same note, losing the edits each time. I spent 30 minutes trying

to edit a single note without success.” A nurse noted, “When the system is down, you are put completely on hold.” Several participants also expressed frustration with the amount of typing required and the time-consuming nature of entering data into the system. Understanding error messages and correcting mistakes also emerged as points of frustration. Comments reflecting concerns including statements such as: “too much trial and error”, “if you don’t already know how to do something, you’ll never figure it out”, and “some error messages don’t tell you how to solve the problem.”

Suggestions for Improving EMR Use at MUSC Participants gave numerous written suggestions for improving EMR use at MUSC—73% of the physicians and 49% of nurses wrote comments. Suggestions ran the gamut from “provide more training”, “fix bugs in system”, “address speed and reliability of system”, to “make it more user friendly.” Several common themes emerged from the written suggestions. Some of these relate to how the EMR is structured and used at MUSC and others relate more directly to the Practice Partner application itself. General suggestions included (1) address the speed/performance issues, (2)

Table 4. Most frequently cited limitations/disadvantages to using MUSC’s EMR system EMR Limitation Cumbersome user interface Speed/too slow

60

Physicians n=114

Nurses n=130

38.6%

24.6%

37.3%

32.3%

Having to access multiple clinical systems or poor integration with other systems

17.5%

4.6%

Problems with templates

11.4%

4.6%

Insufficient training and support staff

11.4%

3.8%

Downtime (including system crashes or system locking/freezing up)

14.0%

21.5%

Time-consuming to use

13.2%

14.6%

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

expand training and support personnel/resources, (3) improve the user interface/templates and simplify data entry, and (4) integrate more fully with other MUSC applications (e.g., scheduling, lab, radiology).

dIscussIon Nearly 250 physician and nurse EMR users at MUSC participated in this study. Nurses are more satisfied overall with the EMR system than are attending physicians. The greater system acceptance by nurses may be due to the approach MUSC took in implementing the EMR enterprise-wide. Initial efforts were focused on nurses and administrative staff, because these individuals were expected to lead in the effort. We felt if nurses were comfortable with the system, they could assist physicians. The downside of this approach is it has been time-consuming and resource-intensive to roll out the EMR throughout all the ambulatory care areas. Consequently, some physician users have not been fully trained on all of the system capabilities—yet, they have been using the application for two or more years. Additionally, 58% of participating physicians report that they rarely, if ever dictate. Earlier studies have shown that data entry can negatively impact physicians’ perceptions of their time, particularly if physicians are not proficient typists or comfortable entering information in the examination room with the patient (Gadd & Penrod, 2001; Penrod & Gadd, 2001; O’Connell, Cho, Shah, Brown, & Shiffman, 2004a; Hier, Rothschild, LeMaistre, & Keeler, 2005; Scott, Rundall, Vogt, & Hsu, 2005; Linder et al., 2006). Researchers at Partners HealthCare System conducted a time-motion study and found no differences in primary care physician time utilization before and after EMR implementation, yet the majority of their physicians still perceived that the EMR required more time than paper records to document patient information (Pizziferri et al., 2005). Nurses in an earlier study reported that

the EMR enabled them to finish their work much faster than before implementation (Likourezos et al., 2004); although we did not ask this specific question, it may help explain why MUSC nurses generally viewed the system more positively than physicians. Nurses also tended to value the “avoidance of the perils of paper” more often than the physicians did, yet they did not mention availability as a major benefit as often as physicians did. We categorize these measures of availability/accessibility and avoidance of periods with paper separately, yet they are related. Nurses are responsible for ensuring that the patient’s record is available at the time of the visit, thus, they experience the “grief” when the record is not available. Physicians are not the ones searching for the paper record, so they may not describe the benefit in quite the same manner. On the other hand, physicians were more likely to mention the benefits associated with having access to clinic notes from other physicians and more likely to comment on the overall quality of the documentation. Having a more “complete” picture of the patient’s care (e.g., other visit notes, ancillary test results) stood out as particularly important to physicians. The EMR benefits identified by MUSC physicians and nurses are consistent with earlier studies (Sittig, Kuperman, & Fiskio, 1999; Gadd & Penrod, 2001; Penrod & Gadd, 2001; Likourezos et al., 2004; Joos, Chen, Jirjis, & Johnson, 2006). Physicians and nurses had similar concerns when asked about limitations of the system or opportunities for improving its use at MUSC. In fact, speed and performance-related concerns seemed to cast a cloud over the benefits. Physicians and nurses identified speed and the slowness of the system as a major concern—launching the system was particularly problematic. Yet, in our multivariate regression analysis, system speed did not show up as a significant predictor of overall satisfaction. Although this finding surprised us, an earlier study at Brigham and Women’s also found that system “response time” was not correlated with overall user satisfaction (Sittig, Kuperman, 61

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

& Fiskio, 1999). Factors associated with physicians’ ability to effectively use the system were more likely to be predictors of EMR satisfaction than speed alone. In addition to their concerns with speed and performance, nearly 40% of the physicians felt the user interface was cumbersome, took too many clicks, and made it difficult to make corrections or changes. A common complaint was the “system requires too many steps or clicks to perform a simple task.” Other institutions using different EMR applications have found their clinicians share these same concerns (Sittig, Kuperman, & Fiskio, 1999; Wager, Lee, White, Ward, & Ornstein, 2000; Gadd & Penrod, 2001; Penrod & Gadd, 2001; Miller & Sim, 2004; Scott, Rundall, Vogt, & Hsu, 2005). Some of the difficulty or frustration at MUSC may stem from a lack of sufficient training. For example, we observed from the open-ended comments that participants are frustrated with aspects of the system they may not fully understand how to use (e.g., writing prescriptions, modifying templates). Other frustrations may stem from using an EMR whereby primary care providers and specialists share and record their notes in a single patient record. Historically, each set of providers had specialty-specific records that only those in the specific group could view and modify. Interestingly, the results of this survey are remarkably similar to those reported by researchers at Brigham and Women’s in their ambulatory care division, despite this group’s use of an inhouse-developed EMR system. Using the same QUIS instrument, researchers there discovered that attending physicians scored their EMR lowest in the areas of system speed (although they did not distinguish between launch speed and speed of navigating within the application), helpfulness of error messages, and flexibility of system (Sittig, Kuperman, & Fiskio, 1999). The pattern of responses are quite similar, although MUSC physician reactions were less positive in terms of screen design and layout (p<.01), terms

62

and system information (p<.01) and learning (p<.001). Brigham and Women’s may have a more advanced training environment, and because they developed the EMR in-house, they have more flexibility in customizing the screens and layout to accommodate their physicians’ preferences. No differences were found among overall user reaction and system capabilities. The physicians participating in the study at Brigham and Women had two or more years EMR experience. Our study evaluated responses from both relatively new and experienced users. One might assume that due to the learning curve in moving from a paper-based system to an EMR, more experienced EMR users would be more satisfied with the system than less experienced users as Gamm et al. suggest (Gamm, Barsukiewicz, Dansky, & Vasey). We did not find that to be the case in our study. EMR users with less than one year of experience were equally satisfied overall with the system as were those who had used the application for two or more years. Participants in this study provided a host of important suggestions for improving the EMR’s use at MUSC. Other institutions would do well to address these issues and concerns up front to avoid problems later. Our study has two primary limitations. First, the study is limited to a single healthcare organization and one EMR product. Thus, the results may not be generalizable to other healthcare organizations or those who use a different EMR system. Second, response bias is a concern. We compared the demographics of our physician and nurse participant populations with the general physician and nurse EMR population at MUSC and found no differences, yet response rate remains a limitation. The real value of this study to other healthcare organizations may not be the results per se found at MUSC, but rather the methods and process used. Using the QUIS instrument, we were able to identify the aspects of our EMR system that stood out as problematic to clinician users—

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

launch speed, navigation speed, and the clarity and helpfulness of error messages. Equally, if not more importantly, we provided all clinicians with the opportunity to offer suggestions on how to improve the system, and since this survey, we have incorporated many of their suggestions into our environment. Our leadership team has taken a number of steps to address the most widespread concerns identified by this survey and validated through other avenues. We believe other institutions will find the methods used in this study helpful in assessing clinician satisfaction and in soliciting suggestions for improvement.

conclusIon Results of this survey suggest that MUSC physicians and nurses recognize and value having access to a single electronic patient record that is shared across the ambulatory care enterprise. However, they view the current system as less than ideal. Speed, performance and the user interface (e.g., need to simplify data entry) are of concern. Likewise, additional training and resources are needed to more effectively support the system and its users. Our leadership team has taken a number of steps to address the most widespread concerns identified by this survey and validated elsewhere. Steps to enhance system speed/performance and reliability are being taken, and multi-modal programs of improved training and user support are being implemented. We expect to observe the results of these initiatives and report outcomes in due course. Assessing user satisfaction with the EMR is important in providing leadership with additional insight into the issues and concerns. Conducting formal evaluation studies, however, are often not done because of lack of available expertise and resources. We believe that surveys such as this one can prove to be a useful resource not only to our healthcare organization’s leadership team, but to those in other healthcare institutions interested

in easily assessing EMR satisfaction across the enterprise.

reFerences Anderson, J., & Aydin, C. (Eds.). (2005). Evaluating the organizational impact of healthcare information systems, 2nd ed. Springer. Bates, D. (2005). Physicians and ambulatory electronic health records. Health Affairs, 24(5), 1180-1189. Brender, J. (2006). Handbook of evaluation methods for health informatics. Elsevier. Burkle, T., Ammenwerth, E., Prokosch, H., & Dudeck, J. (2001). Evaluation of clinical information systems. What can be evaluated and what cannot?. Journal of Evaluation in Clinical Practice, 7(4), 373-385. Chin, J., Diehl, V., & Norman, K. (1988). Development of an instrument measuring user satisfaction of the human-computer interface. Paper presented at the CHI, New Ork. DeLone, W., & McLean, E. (2003). The DeLone and McLean model of information systems success: A ten-year update. Journal of Management Information Systems, 19(4), 9-30. Despont-Gros, C., Mueller, H., & Lovis, C. (2005). Evaluating user interactions with clinical information systems: A model based on human-computer interactions models. Journal of Biomedical Informatics, 38, 244-255. Friedman, C., & Wyatt, J. (1997). Evaluation methods in medical informatics. New York: Springer. Gadd, C., & Penrod, L. (2001). Assessing physician attitudes regarding use of an outpatient EMR: A longitudinal, multi-practice study. Proceedings/ AMIA Annual Symposium, 194-198. Gamm, L., Barsukiewicz, C., Dansky, K., & Vasey, J. Pre- and post-control model research on 63

Assessing Physician and Nurse Satisfaction with an Ambulatory Care EMR

end-users’ satisfaction with an electronic medical record: Preliminary results. Paper presented at the AMIA. Hier, D., Rothschild, A., LeMaistre, A., & Keeler, J. (2005). Differing faculty and housestaff acceptance of an electronic health record one-year after implementation. International Journal of Medical Informatics, 74, 657-662. Hillestad, R., Bigelow, J., Bower, A., Girosi, F., Meili, R., Scoville, R., et al. (2005). Can electronic medical record systems transform healthcare? Potential health benefits, savings and costs. Health Affairs, 24(5), 1103-1117. Joos, D., Chen, Q., Jirjis, J., & Johnson, K. (2006). An electronic medical record in primary care: Impact on satisfaction, work efficiency and clinic processes. Paper presented at the AMIA, Washington, DC. Likourezos, A., Chalfin, D., Murphy, D., Sommer, B., Darcy, K., & Davidson, S. (2004). Physician and nurse satisfaction with an electronic medical record system. Journal of Emergency Medicine, 27(4), 419-424. Linder, J., Schnipper, J., Tsurikova, R., Melnikas, A., Volk, L., & Middleton, B. (2006). Barriers to electronic health record use during patient visits. Paper presented at the AMIA, Washington, DC. Miller, R., & Sim, I. (2004). Physicians’ use of electronic medical records: Barriers and solutions. Health Affairs, 23(2), 116-126.

satisfaction with two implementations under one roof. J Am Med Inform Assoc., 11(1), 43-49. Penrod, L., & Gadd, C. (2001). Attitudes of academic-based and community-based physicians regarding EMR use during outpatient encounters. Proceedings/AMIA Annual Symposium, 528-532. Pizziferri, L., Kittler, A., Volk, L., Honour, M., Gupta, S., Wang, S., et al. (2005). Primary care physician time utilization before and after implementation of an electronic medical record: A time-motion study. Journal of Biomedical Informatics, 38, 176-188. Scott, J., Rundall, T., Vogt, T., & Hsu, J. (2005). Kaiser Permanente’s experience of implementing an electronic medical record: A qualitative study. BMJ, 331, 1313-1316. Sittig, D., Kuperman, G., & Fiskio, J. (1999). Evaluating physician satisfaction regarding user interactions with an electronic medical record system. Paper presented at the AMIA. Slaughter, L.A., Harper, B.D., and Norman, K.L. (1994). Assessing the equivalence of the paper and online formats of the QUIS 5.5. In Proceedings of Mid Atlantic Human Factors Conference, (pp. 87-91)Washington, DC. Wager, K., Lee, F., & Glaser, J. (2005). Managing healthcare information systems: A practical approach for healthcare executives. San Francisco, CA: Jossey-Bass.

O’Connell, R., Cho, C., Shah, N., Brown, K., & Shiffman, R. (2004a). Take note: Differential EHR satisfaction with two implementations under one roof. J Am Med Inform Assoc., 11(1), 43-49.

Wager, K., Lee, F., White, A., Ward, D., & Ornstein, S. (2000). Impact of an electronic medical record system on community-based primary care practices. The Journal of the American Board of Family Practice, 13(5), 338-348.

O’Connell, R., Cho, C., Shah, N., Brown, K., & Shiffman, R. (2004b). Take note: Differential EHR

Windows, S. (2005). Version 14.0.2005. Chicago: SPSS Inc.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 1, edited by J. Tan, pp. 63-74, copyright 2008 by IGI Publishing (an imprint of IGI Global).

64

65

Chapter 5

Information Technology (IT) and the Healthcare Industry: A SWOT Analysis Marilyn M. Helms Dalton State College, USA Rita Moore Dalton State College, USA Mohammad Ahmadi University of Tennessee at Chattanooga, USA

aBstract The healthcare industry is under pressure to improve patient safety, operate more efficiently, reduce medical errors, and provide secure access to timely information while controlling costs, protecting patient privacy, and complying with legal guidelines. Analysts, practitioners, patients and others have concerns for the industry. Using the popular strategic analysis tool of strengths, weaknesses, opportunities, and threats analysis (SWOT), facing the healthcare industry and its adoption of information technologies (IT) are presented. Internal strengths supporting further industry investment in IT include improved patient safety, greater operational efficiency, and current investments in IT infrastructure. Internal weaknesses, however, include a lack of information system integration, user resistance to new technologies and processes, and slow adoption of IT. External opportunities including increased use of the Internet, a favorable national environment, and a growing call for industry standards are pressured by threats of legal compliance, loss of patient trust, and high cost of IT.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Information Technology (IT) and the Healthcare Industry

The healthcare industry faces many wellrecognized challenges: high cost of operations, inefficiency, inadequate safety, insufficient access to information, and poor financial performance. For years, many have called for a fundamental change in the way healthcare is delivered. And while there is yet no clear picture of what this change will be, many believe a paradigm shift in healthcare is imminent and that information technology (IT) is the catalyst. Increasingly, IT is seen as a way to promote the quality, safety, and efficiency of healthcare by bringing decision support to the point of care, providing vital links and closing open loop systems, and allowing routine quality measurement to become reality. IT can not only reduce operating costs, but IT can also ensure a reduction in the number of medical errors. IT in the healthcare industry provides new opportunities to boost patient confidence and reinforce patient trust in caregivers and healthcare facilities. With health insurers feeling pressure from all directions (new regulations, consumers, rising medical costs), IT is an even more important asset for carriers (Balas, 2000). When compared to other information intensive industries, healthcare organizations currently invest far less in IT. For many years, the healthcare industry has experienced only single digit growth in terms of IT investment (Gillette, 2004). As a result, current healthcare systems are relatively unsophisticated compared to those in industries such as banking or aviation. With the many issues and variables surrounding healthcare’s IT investment, a framework for better understanding of the current situation is needed before more improvements and enhancements can result. This article draws upon a comprehensive framework from the strategic planning literature to compile and summarize the major issues facing IT and the healthcare industry.

66

MetHodology By categorizing issues into strengths, weaknesses, opportunities, and threats, SWOT analysis is one of the top tools and techniques used in strategic planning (see Glaister & Falshaw, 1999). SWOT assists in the identification of environmental relationships as well as the development of suitable paths for countries, organizations, or other entities to follow (Proctor, 1992). Valentin (2001) suggests SWOT analysis is the traditional means for searching for insights into ways of crafting and maintaining a fit between a business and its environment. Other researchers (see Ansoff, 1965; Porter, 1991; and Mintzberg, Ahlstrand, & Lampel, 1998) agree SWOT provides the foundation to gather and organize information to realize the desired alignment of variables or issues. By listing favorable and unfavorable internal and external issues in the four quadrants of a SWOT analysis, planners can better understand how strengths can be leveraged, realize new opportunities, and understand how weaknesses can slow progress or magnify threats. In addition, it is possible to postulate ways to overcome threats and weaknesses (e.g., Hofer & Schendel, 1978; Schnaars, 1998; Thompson & Strickland, 1998; McDonald, 1999; and Kotler, 2000). SWOT has been used extensively to aid in understanding a variety of decisions and issues including: manufacturing location decisions (Helms, 1999); penetration strategy design for export promotions and joint ventures (Zhang & Kelvin, 1999); regional economic development (Roberts & Stimson, 1998); entrepreneurship (Helms, 2003); performance and behavior of micro-firms (Smith, 1999), and strategic planning (Khan & Al-Buarki, 1992). Hitt, Ireland, Camp, and Sexton (2001) suggest that identifying and exploiting opportunities is part of strategic planning. Thus, SWOT analysis is a useful way to profile the general environmental position of a new trend, technology, or a dynamic industry. By using SWOT analysis, it is possible to apply

Information Technology (IT) and the Healthcare Industry

strategic thinking toward the implementation of IT in healthcare. By examining the internal and external factors interacting both for and against IT in healthcare, healthcare providers and supply chain organizations can formulate a strategic IT plan for developing their information resources over the next several years. By uncovering and reviewing the issues, policy makers can enact changes to make the process of IT implementation easier while simultaneously working to change the culture to foster IT benefits for institutions and the patients and other stakeholders they serve. In SWOT analysis, strengths act as leverage points for new strategic initiatives, while weaknesses are limiting factors. Specifically applied to IT in healthcare, strengths should indicate areas where either IT or healthcare is particularly strong, (i.e., the technical skills of IT professionals or the quality of existing healthcare information systems). Weaknesses should display areas where either IT or healthcare requires improvement, and may range from personnel issues within IT to limited healthcare applications beyond routine transaction processing. The threats and opportunities identified during the external analysis should be both factual and attitudinal issues that must be addressed in any strategic plan being formulated, and should include both healthcare and IT issues (Martin, Brown, DeHayes, Hoffer, & Perkins, 2005). The following section presents the internal strengths and weaknesses currently confronting the implementation and proliferation of IT in healthcare.

Internal strengtHs Improved patient safety Patient safety, as expressed in the Hippocratic Oath (Classical Version)—“I will keep them from harm and injustice”—is an underlying principle of professional healthcare throughout the world. Improving patient safety is a primary objective at

all levels of the healthcare industry. The strategic initiative to increase the role of IT in healthcare can advance the cause of greater patient safety by enhancing the quality of that care. With comprehensive data available in a timely manner, healthcare providers can make better decisions about their patients’ care, thereby reducing errors due to incomplete or insufficient information at the point of decision (Goldberg, Kuhn, & Thomas, 2002). Lenz (2007) agrees IT has a huge potential to improve the quality of healthcare and that this aspect has not been fully explored by current IT solutions. Advanced process management technology is seen as a way to improve IT support for healthcare processes by improving the quality of those processes. The Joint Commission on Accreditation of Healthcare Organizations (JCAHO) established the Healthcare Information Technology Advisory Panel in 2005 to focus attention on the improvement of patient safety and clinical processes as new healthcare information systems are implemented. Members of the panel include researchers, physicians, nurses, chief information officers, educators and leaders of healthcare organizations, as well as representatives from the Office of the National Coordinator for Health Information Technology, the American Health Information Management Association, the Agency for Healthcare Research and Quality, the Veterans Health Administration, and the Healthcare Information and Management Systems Society. The panel was formed to recommend ways JCAHO’s accreditation process and the widespread use of technology can be used to help re-engineer patient care delivery and result in major improvements in safety, quality and efficiency. The panel was also charged with the task of examining such topics as the effect of electronic health records on performance benchmarking and public reporting capabilities. Based on the panel’s recommendations, JCAHO will evaluate its strategic plan and future direction relative to healthcare information technology (Anonymous, 2005b).

67

Information Technology (IT) and the Healthcare Industry

Two examples of existing information technologies and computer-based information systems contributing to improvements in patient safety through better quality of care and reduction of errors are smart cards (and/or compact discs) and computerized physician order entry (CPOE) systems. Smart cards containing a patient’s entire medical history can be designed to be accessible only by devices in a hospital, doctor’s office, or other medical facility. They not only eliminate the problems of lost and comprised hard copies of patient records, but also enable more secure electronic transfers of patient information to other healthcare providers and insurers (Anonymous, 1997). The technology for Java-based cards can securely support applications for multiple healthcare facilities and be combined with biometric measures for identification purposes (Sensmeier, 2004). With complete information available, physicians are able to make better decisions for the care of the patient, and order appropriate tests and treatments (Goldberg et al., 2002). Several studies indicate medication errors are the most likely type treatment error to occur because drug therapy is one of the most widely used interventions in healthcare (Kohn, 2001). Computerized physician order entry (CPOE) systems eliminate transcription errors and can warn of allergies and drug interactions. Such systems reduce errors by more accurately dispensing the correct dosage of the correct medication for the correct patient (Bates, Teich, Lee, Seger, Kuperman et al., 1999;Kuperman, Teich, Gandhi, & Bates, 2001; Mekhjian, Kumar, Kuehn, Bentley, Teater et al., 2002; Order Entry Rules, 2002; Scalise, 2002; Shane, 2002). Computerized Physician Order Entry (CPOE) systems reduce medication errors by 80%, and errors with serious potential patient harm by 55% (Bates et al., 1999). In pharmaceuticals, e-prescribing, or the electronic transmission of prescription information from the prescribing physician to a pharmacist can reduce medical errors. Since Americans receive

68

more than three billion prescriptions yearly and pharmacists have to call these physicians 150 million times a year because they cannot read or understand the prescriptions, e-prescribing could reduce injuries from medication errors (Brodkin, 2007). Technology-enabled improvements could also aid disease prevention and management. Other benefits could include lowering age-adjusted mortality by 18% and reducing annual employee sick days. Lieber (2007) stresses shared experiences regarding pandemic diseases can provide the best solutions and this is aided by IT solutions and global information exchange. When medical records are available electronically, patients too can have access to their personal health records. Five large U.S. employers have funded an institute where their current and retired employees and their families can have access to and maintain their lifelong personal health records (Five Large Companies, 2007). With access to longitudinal and comprehensive records, patient safety can continue to improve. A recent example of IT and improved patient safety is the North Mississippi Medical Center (NMMC) in Tupelo, Mississippi. Serving 24 rural counties, NMMC is the largest rural hospital in the country and the 2006 winner of the Malcolm Baldrige National Quality Award in the Healthcare Category. NMMC’s recognition was due largely to their success in utilizing IT. Patients’ electronic medical records can be accessed by nurses, by partner community hospitals, by physicians in their offices, and even by specialists and primary care providers in remote sites, reducing medical errors and duplication of effort. These enhancements have earned NMMC the distinction of one of the most wired facilities in the country. The organization has a shared radiography information system for all its hospitals and clinics which reduces report preparation time. For example, patients can have a radiology procedure, see their doctor, and obtain their results the same day (Baldrige Award Recipient, 2006; Anderson, 2007).

Information Technology (IT) and the Healthcare Industry

greater efficiency of operation Information technology, or the digital world of bits and bytes, delivers information faster, smarter, and cheaper (Conger & Chiavetta, 2006). In healthcare, IT has improved operational efficiency and increased productivity by reducing paperwork, automating routine processes, and eliminating waste and duplication. Lieber (2007) reports the use of electronic health records could save as much as $8 billion yearly in California alone through improvements in delivery efficiency. Picture archival and communication systems (PACS) not only save providers’ costs for file room, storage space and film supplies, but also decrease time spent reporting, filing and retrieving records. Web access enables physicians to view radiological images from their offices, homes, or other remote facilities. IT provides emergency rooms with tools for electronic prescriptions, order entry, provider documentation, and aftercare instructions for patients and their family. Updating electronic instructions is quick and easy. Purchasing departments are aided by the ability to buy products for specialty areas, such as anesthesia, infection control, substance-abuse programs, and home healthcare. Increased productivity and positive return on investment are seen in many areas of IT and are continuously improving (Parker, 2004b). Three other important technological advances for improved productivity are voice-technology systems, two-way communications systems, and radio-frequency identification (RFID). Voicetechnology systems can significantly reduce the time nurses and admissions personnel spend on pre-authorizations and pre-certifications required by third-party payer plans. Within six months of implementing a phone-based voice-technology system, Erlanger Hospital in Chattanooga, Tennessee, reported a greater than 50% drop in phone transaction times. Moreover, in May, 2006, after four years in operation, Erlanger reported a total payback of $920,201, decreased percentages in

days denied, and the reassignment of three fulltime equivalent employees to other departments within the hospital (Bowen & Bassler, 2006). Automated two-way communications systems for scheduling can greatly improve workflow by managing automatic appointment reminders, waiting lists, and cancellation notices. These systems call patients, and in a pleasant voice, remind them of a doctor’s appointment, ask them to confirm their intention to keep the appointment, and report the information to the provider’s office (Sternberg, 2005). RFID technology, particularly in the areas of human and material resources, offers healthcare facilities a way to measure and control their resources as well as the relevant workflow processes (Janz, Pitts, & Otondo, 2005). Some healthcare providers suggest Emergency Department Information Systems (EDIS) can improve operations efficiency in this extremely time-critical area by facilitating the flow of patients through the emergency department, eliminating redundant patient records, promoting information sharing, and providing quicker access to laboratory test results and radiology films (Parker, 2004a). Because electronic medical records allow tracking of patients’ conditions and medications, emergency room providers and hospitals have immediate access to detailed information; both patients and providers have a better sense of what occurred and when. Both groups also report increased satisfaction with the process. Interoperability of systems also makes patient information available across budgetary and functional units, thereby providing greater continuity of patient care (Cohen, 2005). Lopes (2007) agrees it is more efficient when an internal medicine physician can consult with a cardiologist electronically while viewing a patient’s medical work-up and history. Such systems create efficiencies, have a positive return on investment, and there are no misfiled lab results. IT tools and software are being developed to meet the growing needs of the healthcare community. As an example, VHA, Inc. recently

69

Information Technology (IT) and the Healthcare Industry

introduced an updated version of a Comparative Clinical Measurement tool to give member hospitals flexibility in both collecting and reporting clinical improvement data. VHA, Inc. has worked with the Joint Commission on Accreditation of Healthcare Organizations to include their measures into the new tool (VHA, 2007).

current Investment in It Is there a hospital in the United States that has not already made an investment in their IT infrastructure? Probably not. In the past ten years, advances in health information technologies have occurred at an unprecedented rate and healthcare organizations have responded by increasing their IT investments “threefold” (Burke & Menachemi, 2004). Today, albeit at varying levels of sophistication, all hospitals use IT to run their core administrative and clinical application systems, that is, patient accounting, insurance billing, human resources, staff and facilities scheduling, pharmacy, laboratory results reporting, and radiology (Cohen, 2005). Most healthcare organizations in the U.S. are spending between 2.1% and 10% of their capital operating budget on IT (Conn, 2007c). Within this spending on IT, healthcare providers cite electronic health record development as a top priority followed by development and implementation of clinical IT systems to improve patient care capability (Conn, 2007b). According to the U.S. Department of Health and Human Services (HHS), approximately 13% of the nation’s 4,000+ hospitals use electronic medical records and 14% to 28% of the 853,000 U.S. physicians are wired (Swartz, 2005). A recent study reported, on average, hospitals have acquired 10.6 clinical application systems, 13.5 administrative application systems, and 50.0 strategic application systems (Burke & Menachemi, 2004). Some healthcare organizations including the Cincinnati Children’s Hospital, Baylor Healthcare System in Dallas and The Heart Center of Indiana are going beyond their core systems to

70

develop, acquire, and integrate applications for decision support, benchmarking, facilities management, and workflow processes (Cohen, 2005; Kay & Clarke, 2005). If predictions are correct, by 2015 the electronic medical records market is expected to grow to more than $4 billion, up from $1 billion in 2005. The Kalorama Information Research firm, after studying the healthcare, diagnostics, pharmaceuticals, and medical devices markets, predicts the surge will be led by the increase in IT budgets of hospitals, physicians’ offices and other U.S. healthcare organizations (Study: U.S., 2007). Healthcare organizations, having already made investments in their computing and communications hardware and software, application software, and personnel, can leverage their existing IT investments as they expand their IT infrastructure to meet growing demands to achieve more efficient operations and more effective levels of healthcare. Whether or not the IT investments meet the ROI goals of financial departments, hospitals are going to implement IT, according to a survey by the Healthcare Information and Management Systems Society. Some 88% of hospitals have adopted electronic medical records and 24% already have them in place. Some 36% are implementing them and 28% have plans to. Only 12% lack IT plans (Greene, 2007).

Internal weaKnesses lack of system Integration Integrated systems offer seamless data and process integration over diverse information systems (Landry, Mahesh, & Hartman, 2005). Since a patient’s treatment involves receiving services from multiple budgetary units in a hospital, information system integration should exist between the computer-based applications within a single hospital. When healthcare organizations coordinate and integrate their internal data, they can

Information Technology (IT) and the Healthcare Industry

improve operations and decision making; however, most healthcare organizations are not yet at this level of system integration. Clinical, administrative, and financial systems are not linked, and as a result, many healthcare institutions are not yet maximizing their IT potential (Cohen, 2005). Moreover, system integration need not be confined to applications within a single facility. There are many types of healthcare providers and healthcare-related agencies in the complex healthcare network. Since a patient’s treatment usually involves receiving services from multiple providers and interacting with various other healthcare-related entities, information system integration should also exist between the computer-based applications of those separate agencies. Immediate benefits of information sharing between different agencies include the elimination of duplicate work in gathering and inputting the data, the immediate availability of the information, a lower probability of error, and greater convenience for the patient. Called a “vision of unsurpassed information technology integration,” The Heart Center of Indiana, a joint venture between St. Vincent Health, The Care Group, and CorVasc, reports improved quality of care at lower costs through its IT partnership (Kay & Clarke, 2005). System integration between agencies could also take increased efficiency to the industry or national level. A report issued by the Foundation of Research and Education of the American Health Information Management Association supports a fully integrated fraud management system which it believes could help address the growing problem of healthcare fraud (Swartz, 2006a).

user resistance User resistance, more commonly termed user acceptance in the information systems literature, is nothing new to IT. The original Technology Acceptance Model (TAM) put forth by Davis (1989) states a user’s level of system acceptance

is explained by two factors: the system’s perceived usefulness and its perceived ease of use. Perceived usefulness is defined as the degree to which a person believes that using a particular system would enhance job performance, while perceived ease of use is defined as the degree to which a person believes that using a particular system would be free of effort. Subsequent research across a variety of research settings confirms perceived usefulness as the strongest predictor of user acceptance (Adams, Nelson, & Todd, 1992; Taylor & Todd, 1995; Venkatesh & Davis, 1996; Mahmood, Hall, & Swanberg, 2001). Some believe that IT implementations in the healthcare environment, however, encounter more resistance than in any other environment (Adams, Berner, & Wyatt, 2004). The healthcare-related literature suggests physician resistance is a key weakness existing in the doctor’s office and at the hospital. Consistent with Davis’ (1989) principle of perceived usefulness, physician acceptance of new IT systems at the hospital is linked to the system’s impact on patient safety (Rhoads, 2004), while physician acceptance of new IT at their office largely depends on cost (Chin, 2005). Healthcare literature suggests nurse acceptance of new IT has steadily improved as applications demonstrate increased support of the practice of nursing and improvement of patient safety resulting from the reduction of human error (Sensmeier, 2005; Simpson, 2005). A study of 12 critical access hospitals found barriers to health information technology included funding, staff resistance to change, staff adaptation to IT and workflow changes. Other user resistance was noted by the time constraints on small staff, facility and building barriers, and lack of appropriate IT support. While all agree that IT will improve safety and reduce errors, barriers to implementation are numerous and must be addressed (Hartzema, Winterstein, Johns, de Leon, Bailey, McDonald, & Pannell, 2007).

71

Information Technology (IT) and the Healthcare Industry

slow It adoption

external opportunItIes

Traditionally, healthcare has been slow to adopt IT and has lagged significantly behind other industries in the use of IT (Ortiz & Clancy, 2003; Adams et al., 2004). A 2005 report from the National Academy of Engineering and the Institute of Medicine agrees healthcare’s failure to adopt new strategies and technologies has contributed to the list of problems now associated with the industry: thousands of preventable deaths a year, outdated procedures, billions of dollars wasted annually through inefficiency, and costs rising at roughly three times the rate of inflation. Lack of competition, resistance to change, and capital costs are among the major causes for healthcare’s slowness to adopt IT (Hough, Chen, & Lin, 2005). There are signs of progress, however, which offer promise of accelerated change. Many hospitals and physician groups are now digitizing their medical records and clinical data (Hough et al., 2005). As noted earlier, some hospitals like Cincinnati Children’s Hospital, Baylor Healthcare System in Dallas, and The Heart Center of Indiana have adopted IT at advanced levels (Cohen, 2005; Kay & Clarke, 2005). These hospitals are models for the industry, forging a path for other healthcare organizations to follow, and emerging as healthcare leaders in IT whose techniques can be benchmarked, emulated and implemented. As the healthcare technologies are developed to greater sophistication and functionality, it will be possible for other healthcare organizations to “leapfrog” over the slow, expensive evolutionary learning process experienced by the leaders (Conger & Chiavetta, 2006). The following section outlines external opportunities and threats facing IT and healthcare. Specific opportunities are the Internet, the national environment, and industry standards. Key threats include legal compliance, loss of patient trust, and the costs of IT systems, training, implementation, and support.

the Internet

72

Across the industry, healthcare facilities and providers are in various stages of incorporating the Internet into their operations to allow new ways to communicate with the general public, specific patients, patient groups, physicians, other providers, and employees. Notable Web-based services include public Web sites, various telemedicine applications for targeted patient audiences, physician portals, physician education sites, and facility intranets which serve an organization’s internal audiences. Generally, there is an increased focus throughout the healthcare industry to improve all Web-based applications (Sternberg, 2004). Through their public Web sites, hospitals and other healthcare agents provide medical information to the general public (Natesan, 2005). E-Health Web portals offer healthcare services and education to people with chronic conditions and to their caregivers (Moody, 2005). Through various telemedicine initiatives, the healthcare industry has reached significant numbers of people living in rural areas, providing access to expert advice and reducing their health risks (Harris, Donaldson, & Campbell, 2001). Webbased patient support systems educate patients and allow them better participation in their own care. Patients can research detailed information for their particular conditions, medications, and treatments to understand what is happening and to reduce their anxiety. Online surgery videos and graphics can be presented in user-friendly formats to assist patients in procuring information. The Internet has also had a major impact in the delivery of information and education to healthcare professionals (Kiser, 2001). Numerous organizations have Web sites for disseminating new medical information to physicians. Various Web-based physician education services have been established. Some hospitals offer physician portals allowing physicians to access patients’ medical

Information Technology (IT) and the Healthcare Industry

records, lab results, and radiological images and reports from their offices, homes, or other remote locations (Cohen, 2005). The Internet is also redefining communication channels between doctors and patients, as well as between healthcare providers and other healthcare-related agencies. DeShazo, Fessenden, and Schock (2005) suggest the top two emerging trends in healthcare are (1) online patient/physician communication and (2) secure connectivity and messaging among hospitals, labs, pharmacies, and physicians. Advances in home technology coupled with the aging of the baby-boom generation have created the demand for better communications with patients about their on-going care and monitoring. Improving the communication between the patient’s at-home technology and the provider’s technology is also a growth opportunity. Based on the adequacy of information transmitted to the healthcare provider, the physician saves appointment times and patients are freed from excessive office visits, thereby lowering transaction costs (Flower, 2005). The Internet and other advances in IT have enabled new models for electronic delivery of a variety of healthcare services. Kalyanpur, Latif, Saini, and Sarnikar (2007) describe the market forces and technological factors that have led to the development of Internet-based radiological services and agree the Internet has provided the platform for cost-effective and flexible radiological services. Wells (2007) agrees the practice of evidence-based medicine requires access to the Internet, mobile devices, and clinical decisionsupport tools to assist practitioners in improving preventable medical errors.

Favorable external environment There is growing support worldwide for the utilization of more IT in healthcare (Caro, 2005). Reports from Australia, Great Britain, India, Italy, and Norway, for example, document local, regional and national healthcare projects and initiatives

utilizing IT (Sharma, 2004; Grain, 2005; Marino & Tamburis, 2005; Bergmo & Johannessen, 2006; Fitch & Adams, 2006). In the U.S., more funds are being made available in the form of grants and demonstration projects by the Federal government to encourage greater adoption of IT in healthcare. In 2004, officials of the U.S. Department of Health and Human Services (HHS) disclosed a ten-year healthcare information infrastructure plan, the “Decade of Health Information Technology,” to transform the industry from a paper-based system to an electronic one. More than 100 hospitals, healthcare providers, and communities in 38 states were awarded $96 million over three years to develop and use IT for healthcare. Awards were focused on communities and small and rural hospitals. Five states, Colorado, Indiana, Rhode Island, Tennessee, and Utah, were awarded $25 million over five years to develop secure statewide networks for accessing patient medical information. The National Opinion Research Center at the University of Chicago was awarded $18.5 million to create a National Health Information Technology Resource Center to provide technical assistance, tools, and a best-practices repository as well as provide a focus for collaboration to grantees and other federal partners. In all, HHS awarded nearly $140 million in grants to promote the use of IT, develop state and regional networks, and encourage collaboration in advancing the adoption of electronic health records (Swartz, 2005). In 2005, the Agency for Healthcare Research and Quality, part of HHS, awarded over $22 million in grants to 16 institutions in 15 states to aid in implementing healthcare IT projects emphasizing patient safety and healthcare quality. The grants were designed to encourage the sharing of information among providers, labs, pharmacies, and patients, with the specific goal of decreasing medication errors and duplicate testing. Eleven of the 16 grants were awarded to small and rural communities (Anonymous, 2005c).

73

Information Technology (IT) and the Healthcare Industry

In his 2004 State of the Union address, President Bush called for the transformation of electronic health records within the next ten years in the United States and urged more healthcare organizations to consider implementing such health information technologies as electronic healthcare records (EHR), computerized ordering of prescriptions and medical tests, clinical decision support tools, digital radiology images, and secure exchange of authorized information, emphasizing that all of these technologies have been shown to improve patient care quality and reduce medical errors (Abrahamsen, 2005). In President Bush’s 2008 budget proposal, there is funding for a healthcare system and IT is the starting point for the system. Carolyn Clancy, Director of the Agency for Healthcare Research and Quality, agrees the data generated from the healthcare system could answer various medical inquiries and could draw on the data of EHRs of millions of individuals to advance the evidence base for clinical care. She further suggests the data could reveal why costs are increasing and what risks and benefits are associated with particular prescription drugs (Lubell, 2007).

Industry standards The development of industry standards for both data communications and data taxonomies may be the most profound of all the opportunities currently facing healthcare. As a crucial firststep in modernizing the U.S. healthcare system, all industry participants—providers, payers, and regulators—are being urged to adopt interoperable systems and common data standards for existing federal, state, and health networks along with standard practices to promote data sharing and protection of patient privacy (Swartz, 2006b). Standard data communications technology and standard data definitions are essential for such health information technologies as electronic health records and e-prescribing (Brailer, 2004).

74

A recent study of several disability compensation programs within the U.S. found each program uses its own terminology and disability definitions causing non-standard interpretation of terms, misinterpretation of data, and delay in the disability evaluation process. The study suggests defining and adopting a standard for disability evaluation could not only eliminate process inefficiencies in determining disabilities but could also facilitate innovative disability technology practices (Tulu, Hilton, & Horan, 2006). Some standardization of data has been introduced with the Health Insurance Portability and Accountability Act (HIPAA) legislation, but for the most part, standardization projects are voluntary and lack assurances the standards will be adopted by all parties. The National Quality Forum (NQF) endorsed a voluntary consensus standard for patient safety events and has been adopted by more than 260 healthcare providers, consumer groups, professional associations, federal agencies, and research and quality improvement organizations. The taxonomy is the first standardized, integrative classification system adopted by a group of medical agencies, organizations, providers, and U.S. states. The standard establishes a common taxonomy for healthcare errors and other patient safety problems. It can be used to classify data collected in different reporting systems, allowing the data about patient safety events to be combined and analyzed (Anonymous, 2005d). Standardization can result in greater levels of system integration, increased sharing of data between healthcare partners, greater information continuity throughout the healthcare industry, and more powerful data mining. Enterprise Resource Planning (ERP) systems can offer more online processing to all users and function, automate routine job processes, and redefine existing work processes (Landry et al., 2005). With integration of systems and standardization of data taxonomies, the healthcare industry will experience changes similar to those in other industries where “enterprise” systems have been adopted. Others agree a

Information Technology (IT) and the Healthcare Industry

seamless support of information flow for healthcare processes that are increasingly distributed requires the ability to integrate heterogeneous IT systems into a comprehensive system (Lenz, Beyer, & Kuhn, 2007). System standards resulting in a greater level of systems integration is a pressing need. Conn (2007a) reports the compromise reached by two rival standards groups for data communications standards can help to bridge the gap between physicians’ offices and hospitals in the electronic health record systems they use. The Continuity of Care Document standard combines the independent works by two standards development organizations on creating electronic summaries of care for discharged patients.

external tHreats legal compliance The Health Insurance Portability and Accountability Act (HIPAA), enacted by Congress in 1996, is the most significant Federal legislation affecting the U.S. healthcare industry since the Medicare and Medicaid legislation of 1965. Title I of HIPAA legislates improved portability and continuity of health insurance coverage for American workers. Title II addresses “administrative simplification” requiring the development of standards for the electronic exchange of personal health information (PHI). Administrative simplification requires rules to protect the privacy of personal health information, the establishment of security requirements to protect that information, and the development of standard national identifiers for providers, health insurance plans, and employers. Two significant sections of HIPAA are (1) the Privacy Rule and (2) the Security Rule. The Privacy Rule legislates in detail the collection, use, and disclosure of personal health information. To be in compliance with the Privacy Rule, covered entities must notify individuals of

uses of their PHI, keep a record of all disclosures of PHI, and document and disclose their privacy policies and procedures. Covered entities must have designated agents for receiving complaints and they must train all members of their workforce in proper procedures. The Security Rule complements the Privacy Rule and presents three types of security safeguards designated as administrative, physical, and technical. For each type, the Rule identifies various security standards and names (1) required implementation specifications which must be adopted and implemented as specified in the Act and (2) addressable implementation specifications which are more flexible and can be implemented by the covered entities as deemed appropriate. Covered entities face potentially severe penalties for failure to comply with the complex legalities of HIPAA and this has caused much concern throughout the industry. Physicians, medical centers, and other healthcare providers have experienced increased paperwork and cost to incorporate the requirements of this legislation into their current methods of operation. Future adoptions of new information technologies will be subject to its specifications as well (American College of Physicians, 2006).

loss of patient trust The Institute of Medicine (IOM) of the National Academy of Sciences released a report in 1999 that caused much attention to be focused on the U.S. healthcare industry. The report stated medical errors caused between 44,000 and 98,000 preventable deaths annually, and medication errors alone caused 7,000 preventable deaths (Kohn, Corrigan, & Donaldson, 1999). Within two weeks of the report’s release, Congress began hearings and the President ordered a government-wide feasibility study for implementing the report’s recommendations for (1) the establishment of a Center for Patient Safety, (2) expanded reporting of adverse events, and (3) development of safety programs

75

Information Technology (IT) and the Healthcare Industry

in healthcare organizations. According to a study by Healthgrades, a leading healthcare ratings organization, during the period 2000–2002 the estimated number of accidental deaths per year in U.S. hospitals had risen from the 98,000 reported by the IOM in 1999 to 195,000 (Shapiro, 2006). On July 29, 2005, President Bush signed into law the Patient Safety and Quality Improvement Act, establishing a federal reporting database. This was the first piece of patient safety legislation since the 1999 IOM report. Under this act, hospitals voluntarily report “adverse patient events” to be included in the database, and “patient safety organizations” under contract with the Federal government, analyze the events and recommend improvements. The reports submitted by the hospitals remain confidential and cannot be used in liability cases. The most recent Healthgrades report (April 7, 2007) covering the period 2003–2005, indicates that patient safety incidents have increased over the previous period to 1.16 million among the 40 million hospitalizations covered under the Medicare program. Healthcare must utilize all available means to maximize patient safety and retain patient trust. Healthcare is an information intensive industry and the delivery of high quality healthcare depends in part on accurate data, available at the point of decision. Handwritten reports, notes and orders, non-standard abbreviations, and poor legibility all contribute to substantial errors and injuries (Kohn et al., 1999). A doctor needs to know a patient’s medical history, ancillary providers need to be able to read the doctor’s orders and patients need to be able to understand what the doctor expects of them. IT solutions are available that address many of the data accuracy and availability problems in healthcare records (Poston, Reynolds, & Gillenson, 2007); however, the level of adoption for these technologies is not impressive. For example, a 2005 report predicts that by the end of 2007, only 59% of all medical groups will have implemented an Electronic Health Record (EHR) system (Gans, Kralewski, Hammons, & Dowd,

76

2005). And although Computerized Physician Order Entry (CPOE) systems reduce medication errors by 80% (Bates et al., 1999), a 2004 survey by Leapfrog found only 16% of hospitals, clinics, and medical practices expected to be utilizing CPOE by 2006.

cost One of the most immediate barriers to widespread adoption of technology is the high cost of implementation. A report by the Annals of Internal Medicine estimated that a National Health Information Network (NHIN) would cost $156 billion in capital investment over five years and $48 billion in annual operating costs. Approximately two-thirds of the capital costs would be needed to acquire the functionalities and one-third for interoperability. The present level of spending is only about one-fourth of the amount estimated for the model NHIN. While an NHIN would be expensive, $156 billion is equivalent to 2% of annual healthcare spending for 5 years (Kaushal et al., 2005). Industry reports from Datamonitor, Gartner, and Dorenfest & Associates predict increased spending on IT by healthcare providers at an annual rate of between 10% and 15% (Broder, 2004). A study conducted by Partners Healthcare System, Boston, concluded that a national healthcare information system would cost $276 billion, take 10 years to build, and require another $16.5 billion annually to operate. However, the study also concluded that such a system would save U.S. hospitals $77.8 billion annually because of more efficient communication (Anonymous, 2005a). According to a study by RAND Health (Health Information Technology, 2005), the U.S. healthcare system could save more than $81 billion annually, reduce adverse healthcare events, and improve quality of care if it were to widely adopt health information technology. Patients would benefit from better health and payers would benefit from lower costs; however, some hospitals fear loss of revenue due to reduced patient length of stay. A

Information Technology (IT) and the Healthcare Industry

Table 1. SWOT Analysis Strengths • Improved Patient Safety • Greater Efficiency of Operation • Current Investment in IT

Weaknesses • Lack of System Integration • User Resistance • Slow IT Adoption

Opportunities • The Internet • Favorable External Environment • Industry Standards

Threats • • •

recent study in Florida suggests that this fear may be unfounded. The results of the study suggest that there is a significant and positive relationship between increased levels of IT use and various measures of financial performance. The results indicated that IT adoption is consistently related to improved financial outcomes both overall and operationally (Menachemi, Burkhardt, Shewchuk, Burke, & Brooks, 2006). Lopes (2007) agrees that fully integrated electronic medical records systems can replace paper records and allow hospitals and physicians to share medical information electronically to improve response and lower costs from duplication. However, the adoption of such systems is slowed by the high cost of new technology, the complexity of the systems, training, and an unwillingness to adapt work processes to include new information technologies.

dIscussIon and conclusIon Table 1 summarizes the current SWOT analysis of IT implementation in the healthcare industry in the U.S. The healthcare industry faces multi-faceted challenges to improve patient safety and assure information security while containing costs and increasing productivity. The key area for addressing these concerns is more investment in IT to facilitate the flow of information and offer access to providers and partners along the healthcare supply chain, reduce medical errors, and increase

Legal Compliance Loss of Patient Trust Costs

efficiency. Implementation of IT networks to achieve the required level of information and data communications is complicated by the variety of systems already used by provider organizations as well as the lack of system integration within provider organizations. Various benchmarking studies are helping to educate healthcare providers about IT expenditures and offer comparison reports on expenditures. The availability of systems far exceeds the budget of most organizations to adopt them. However, the improved revenue cycles and cost-benefit offered by IT investments are becoming easier to quantify in faster turnaround and processing of patient-related transactions, shared data, reduced duplication of efforts, and increased provider and customer satisfaction. Information technology can help take the paper chart out of healthcare, and eliminate error, variance and waste in the care process. IT can help connect the appropriate persons, knowledge, and resources at the appropriate time and location to achieve the optimal health outcome, increase customer service and patient care with industry leading medication fill rates and timely deliveries, cut operation costs through advanced warehouse management, reduce internal labor costs, and improve enterprise efficiencies of healthcare organizations, all through tightly integrated applications. IT can ultimately transform the healthcare industry. Along with improved safety and greater patient trust, adopting IT in healthcare can only

77

Information Technology (IT) and the Healthcare Industry

improve current conditions and help the United States improve healthcare in general. Concerns remain about how smaller practices can afford the costs of new systems that require them to move their paper medical records to electronic media. These costs and start-up expenses mean an unequal playing field for small practices versus larger healthcare systems with more money to spend on IT integration.

areas For Future researcH IT applications in healthcare are reaching the growth phase of the lifecycle. The strengths, weaknesses, opportunities, and threats at this stage of the life cycle are clear, but few solutions have been proposed. Research is needed to forecast the SWOT issues as IT in healthcare moves from growth to maturity. Case studies in both large and small physician practices as well as in large and small healthcare systems are needed to better understand the IT implementation timeframe and costs. Studies that address ways to overcome human barriers to implementation are also needed. Using the supply chain model, the healthcare information system needs to be studied as to access and applicability for other providers including pharmacists, dietitians, insurance companies, home health service and equipment providers, and other vendors to healthcare. If the healthcare system it to be truly integrated, these additional players must be included. Protection of patient medical information, access to information, data security, and assurances of privacy should also be studied. International suppliers and other options for outsourcing should be investigated as a cost containment strategy. With the growing body of patient-related medical and healthcare data, other applications should be studied. Data is available to postulate ways to improve lifelong health and reduce the incidences of various diseases and ailments. Such

78

data mining should be studied by those in the management information systems area to determine cause and effect and recommend changes. As employers seek to contain healthcare costs of their employees, such data can aid in more active involvement in reducing health risks and making lifestyle changes (i.e., smoking cessation programs, dietary counseling, healthy cafeteria food, work-place gyms). Choosing the best approach to implement IT systems in healthcare settings is also an area for further study. These systems should meet healthcare goals in addition to functionality and integration criteria. Involving physicians and other clinicians in selecting IT systems can increase their support and lessen their resistance to technology. In fact, the more stakeholders are involved in IT selection and implementation planning, the greater their acceptance and rate of adoption will be. Studies in IT implementation outside the healthcare industry need to be reviewed and analyzed to determine where other industries have had success in implementation or have developed tools that could aid the healthcare arena. Human resource studies of executive ownership and accountability can help the healthcare industry better prepare physicians and other practice managers to overcome the user resistance to IT.

reFerences Abrahamsen, C. (2005). Washington taps into healthcare technology. Nursing Management, 36(3), 52. Adams, B., Berner, E. & Wyatt, J. (2004). Applying strategies to overcome user resistance in a group of clinical managers to a business software application: A case study. Journal of Organizational and End User Computing, 16(4), 55-65. Adams, D., Nelson, R. & Todd, P. (1992). Perceived usefulness, ease of use, and usage of information technology: A replication. MIS Quarterly, 16(2), 227-247.

Information Technology (IT) and the Healthcare Industry

American College of Physicians. (2006). Health Insurance Portability and Accountability Act Privacy Rule causes ongoing concerns among clinicians and researchers. Annals of Internal Medicine, 145(4), 313-316.

Bergmo, T. & Johannessen, L. (2006). The long road from potential to realized gains of information technology in healthcare – experiences from Norway. International Journal of Economic Development, 8(3), 682-716.

Anderson, K. (2007). The ASRT Salutes Your Excellence, North Mississippi Medical Center. ASRT Scanner, 39(5), 8-11.

Bowen, R., & Bassler, P. (2006). Can you hear me?. Healthcare Informatics. Retrieved from www.healthcare-informatics.com.

Anonymous. (1997). Just how smart are smart cards? Financial Executive, 13(5), 8-10.

Brailer, D. (2004). Perspective: Translating ideals for health information technology into practice. Health Affairs, W318-W320.

Anonymous. (2005a). Adoption of health information technology standards would save billions. (2005). Healthcare Strategic Management, 23(2), 4-6. Anonymous. (2005b). Healthcare information technology panel established. Association of Operating Room Nurses, AORN Journal, 82(6), 1018. Anonymous. (2005c). Millions awarded to healthcare IT projects. Quality Progress, 38(12), 12. Anonymous. (2005d). Patient safety event taxonomy standard endorsed. Association of Operating Room Nurses, AORN Journal, 82(6), 1040-1041. Ansoff, H. (1965). Corporate strategy. New York: McGraw-Hill. Balas E., Weingarten S., Garb C., Blumenthal D., Boren S., & Brown G. (2000). Improving preventive care by prompting physicians. Arch Intern Mcd, 160, 301-308. Baldrige Award Recipient Profile: Malcolm Baldrige National Quality Award 2006 Award Recipient, Healthcare: North Mississippi Medical Center. Retrieved from http://www.nist.gov/ public_affairs/releases/mississippi.html Bates, D., Teich, J., Lee, J., Seger, D., Kuperman, G., et al. (1999). The impact of computerized physician order entry on medication error prevention. Journal of the American Medical Informatics Association, 6(4) 313-321.

Broder, C. (2004). U.S. healthcare providers plan IT spending increase. iHealthBeat. Retrieved, from http://www.ihealthbeat.org/index.cfm Brodkin, J. (2007). A Growing Divide in HealthCare IT. Network World, 24(6), 17. Burke, D. & Menachemi, N. (2004). Opening the black box: Measuring hospital information technology capability. Healthcare Management Review, 29(3), 207-217. Caro, D. (2005). Forging e-health partnerships: Strategic perspectives from international executives. Healthcare Management Review, 30(2), 174. Chin, T. (2005). Are physicians at the infotech tipping point?. American Medical News, 48(10), 20-21. Cohen, S. (2005). Emerging benefits of integrated IT systems. Healthcare Executive, 20(5), 14-18. Conger, T. & Chiavetta, D. (2006). The pulse of industry. Industrial Engineer, 38(1), 30-35. Conn, J. (2007a). Agreement could put EHRs on fast track. Modern Healthcare, 37(9), 22-23. Conn, J. (2007b). Eyes on the HER. Modern Healthcare, 37(9), 37-38. Conn, J. (2007c). Still the matter of money. Modern Healthcare, 37(9), 36-37.

79

Information Technology (IT) and the Healthcare Industry

Davis, F. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319-340. DeShazo, C., Fessenden, R. & Schock, P. (2005). Realizing IT’s Potential. Physician Executive, 31(4), 26-29. Fitch, C. & Adams, C. (2006). Managing mobile provision for community healthcare support: Issues and challenges. Business Process Management Journal, 12(3), 299. Five Large Companies Working to Provide Lifelong Personal Health Records for Employees. (2007). Healthcare Financial Management, 61(2), 11. Flower, J. (2005). The 16 challenges. Hospitals & Health Networks, 4(4), 16-18. Gans, D., Kralewski, J., Hammons, T., & Dowd, B. (2005). Medical groups’ adoption of electronic health records and information systems. Health Affairs (Project Hope), 24(5), 1323-1333. Gillette, B. (2004). IT investment shows marked increase, but mostly on the clinical side. Managed Healthcare Executive, 14(9), 58. Glaister, K. & Falshaw, J. (1999). Strategic planning: Still going strong?. Long Range Planning, 32(1), 107-116. Goldberg, R., Kuhn, G., Louise, A., & Thomas, H. (2002). Coping with medical mistakes and errors in judgment. Annals of Emergency Medicine, 40(2), 287-292. Grain, H. (2005). Information systems in the new world: An emerging national approach. Australian Health Review, 29(3), 292-297. Greene, J. (2007). The costs & benefits of IT. Trustee, 60(2), 8-12. Harris, K., Donaldson, J. & Campbell, J. (2001). Introducing computer-based telemedicine in three rural Missouri counties. Journal of End User Computing, 13(4), 26-35. 80

Hartzema, A., Winterstein, A., Johns, T., de Leon, J., Bailey, W., McDonald, K., & Pannell, R. (2007). Planning for pharmacy health information technology in critical access hospitals. American Journal of Health-System Pharmacy, 64(3), 315-321. Health Information Technology: Can HIT lower costs and improve quality?. (2005). Rand Health. Retrieved from http://www.rand.org Health Grades, Inc. (2007). HealthGrades quality study: Fourth annual patient safety in American hospitals study. Retrieved from http://www. healthgrades.com/media/dms/pdf/PatientSafetyInAmericanHospitalsStudy2007.pdf. Helms, M. (1999). How to be successful in China: A SWOT analysis. Competitiveness Review, 9(2), 1-10. Helms, M. (2003). Japanese managers: Their candid views of entrepreneurship. Competitiveness Review, 13(1), 24-34. Hitt, M., Ireland, R., Camp, S., & Sexton, D. (2001). Guest editors’ introduction to the special issue: Strategic entrepreneurship: Entrepreneurial strategies for wealth creation. Strategic Management Journal, 22(6-7), 479-491. Hofer, C. & Schendel, D. (1978). Strategy formulation: Analytical concepts. St. Paul, MN: West Publishing Company. Hough, C., Chen, J. & Lin, B. (2005). Virtual health/electronic medical record: Current status and perspective. International Journal of Healthcare Technology & Management, 6(3), 257. Janz, B., Pitts, M. & Otondo, R. (2005). Information systems and healthcare II: Back to the future with RFID: Lessons learned – some old, some new. Communications of the Association for Information Systems, 15, 1.

Information Technology (IT) and the Healthcare Industry

Kalyanpur, A., Latif, F., Saini, S., & Sarnikar, S. (2007). Inter-Organizational e-Commerce in healthcare services: The case of global teleradiology. Journal of Electronic Commerce in Organizations, 5(2), 47-56. Kaushal, R., Blumenthal, D., Poon, E., Jha, A., Franz, C., Middleton, B., et al. (2005). The costs of a national health information network. Annals of Internal Medicine, 143(3), 165-173. Kay, D. & Clarke, J. (2005). Raise care quality, lower costs successfully with IT partnership. Journal of Nursing Administration, 12. Khan, G. & Al-Buarki, E. (1992). Strategic planning in bahrain. Management Decision, 30(5), 3-10. Kiser, K. (2001). Doctor, teach thyself. Online Learning, 23-28. Kohn, L. (2001). The Institute of Medicine report on medical error: Overview and implications for pharmacy. American Journal of Hospital Pharmacy, 58(22), 63-66. Kohn, L., Corrigan, J., & Donaldson, M. (Eds.). (1999). To err is human: Building a safer health system. Institute of Medicine, Washington: National Academy Press. Kotler, P. (2000). Marketing management. Upper Saddle River, NJ: Prentice-Hall.

Lenz, R., Beyer, M. & Kuhn, K. (2007). International Journal of Medical Informatics, 76(2-3), 201-207. Lieber, S.(2007). IT goes global. Modern Healthcare, 37(9), 30. Lopes, G. (2007). Paper still rules patient records. Washington Times (DC), March 5. Lubell, J. (2007). IT may play Research Role in ’08. Modern Healthcare, 37(5), 4. Mahmood, M., Hall, L. & Swanberg, D. (2001). Factors affecting information technology usage: A meta-analysis of the empirical literature. Journal of Organizational Computing and Electronic Commerce, 11(2), 107-130. Marino, A. & Tamburis, O. (2005). Technological innovation and organisational context information management in the high specialised hospital: The case of South Italy. International Journal of Healthcare Technology & Management, 6(3), 276. Martin, E., Brown, C., DeHayes, D., Hoffer, J., & Perkins, W. (2005). Managing Information Technology, 5th ed. Upper Saddle River, NJ: Prentice-Hall. McDonald, M. (1999). Marketing plans. Oxford, UK: Butterworth-Heinemann Press.

Kuperman, G., Teich, J., Gandhi, T., & Bates. D. (2001). Patient safety and computerized medication ordering at Brigham and Women’s Hospital. Journal of Quality Improvement, 27(10), 509-520.

Mekhjian, H., Kumar, R., Kuehn, L., Bentley, T., Teater, P., et al. (2002). Immediate benefits realized following implementation of physician order entry at an academic medical center. Journal of the American Medical Informatics Association, 9(5), 529-539.

Landry, B., Mahesh, S. & Hartman, S. (2005). The impact of the pervasive information age on healthcare organizations. Journal of Health and Human Services Administration, 27(3-4), 444-464.

Menachemi, N., Burkhardt, J., Shewchuk, R., Burke, D. & Brooks, R. (2006). Hospital information technology and positive financial performance: A different approach to finding an ROI. Journal of Healthcare Management, 51(1), 40-59.

Lenz, R. (2007). IT Support for healthcare processes – Premises, challenges, perspectives. Data & Knowledge Engineering, 61(1), 39-58.

Mintzberg H., Ahlstrand, B. & Lampel, J. (1998). Strategy safari: A guide through the wilds of strategic management. New York: Free Press.

81

Information Technology (IT) and the Healthcare Industry

Moody, L. (2005). E-Health Web portals: Delivering holistic healthcare and making home the point of care. Holistic Nursing Practice, 19(4), 156. Natesan, C. (2005). The impact of the Internet on healthcare services. Journal of American Academy of Business, 7(2), 247-252. Order entry rules: Healthcare enterprise achieves physician acceptance, reduced medication errors and improved patient outcomes through CIS and CPOE technology. (2002). Health Management Technology, 23(7), 34, 38. Ortiz, E. & Clancy, C. (2003). Use of information technology to improve the quality of healthcare in the United States. Health Services Research, 38(2), 10-13. Parker, P. (2004a). Move care to a higher level with emergency systems. Nursing Management, 35(9), 82-83. Parker, P. (2004b). Quantify technology’s benefits. Nursing Management, 35(2), 41. Porter, M. (1991). Towards a dynamic theory of strategy. Strategic Management Journal, 12, 95-117. Poston, R., Reynolds, R., & Gillenson, M. (2007). Technology solutions for data accuracy and availability problems in healthcare records. Information Systems Management, 24(1), 59-71. Proctor, R. (1992). Structured and creative approaches to strategy formulation. Management Research News, 15(1), 13-19. Rhoads, C. (2004). Rx for success. Computerworld, 38(34), 27-28. Roberts, B. & Stimson, R. (1998). Multi-sectoral qualitative analysis: A tool for assessing the competitiveness of regions and formulating strategies for economic development. Annals of Regional Science, 32(4), 469-494.

82

Scalise, D. (2002). CPOE: Is it worth it?. Hospitals & Health Networks, 76(1). Schnaars, S. (1998). Marketing strategy. New York: Free Press. Sensmeier, J. (2004). Transform workflow through selective implementation. Nursing Management, 35(12), 46-52. Sensmeier, J. (2005). Here and now: Healthcare’s top 10 IT breakthroughs. Journal of Nursing Administration, October, 8. Shane, R. (2002). Computerized physician order entry: Challenges and opportunities. American Journal of Health-System Pharmacy, 59(3), 286-288. Shapiro, S. (2006). HealthGrades patient safety study shows increase in U.S. hospital incidents. Medical News Today, Retrieved from http://www. medicalnewstoday.com/printerfriendlynews. php?newsid=40954. Sharma, D. (2004). Technologies for the people: A future in the making. Futures, 36(6-7), 733. Simpson, R. (2005). In direct proportion: Ratios, IT, and trust. Nursing Management, 36(2), 14. Smith, J. (1999). The behavior and performance of young micro firms: Evidence from business in Scotland. Small Business Economics, 13(3), 185-200. Sternberg, D. (2004). Trends worth watching. Marketing Health Services, 24(2), 44. Sternberg, D. (2005). Strategic innovation. Marketing Health Services, 24(2), 44. Study: U.S. electronic medical-records market count grow four-fold by 2015. (2006). Business Journal (Central New York), 21(7), 8. Swartz, N. (2005). U.S. hands out technology grants for healthcare. Information Management Journal, 39(1), 14.

Information Technology (IT) and the Healthcare Industry

Swartz, N. (2006a). E-records may end fraud. Information Management Journal, 40(1), 16. Swartz, N. (2006b). Tech leaders release e-health roadmap. Information Management Journal, 40(1), 18. Taylor, S. & Todd, P. (1995). Understanding information technology usage: A test of competing models. Information Systems Research, 6(2), 144-176. Thompson, A. & Strickland, A. (1998). Crafting and Implementing Strategy, 10th ed. New York: McGraw Hill. Tulu, B., Hilton, B. & Horan. T. (2006. Improving disability evaluation productivity: Linking innovative business models with information technology. International Journal of Healthcare Technology & Management, 7(1-2), 168.

Valentin, E. (2001). SWOT analysis from a resource-based view. Journal of Marketing Theory and Practice, 9(2), 54-69. Venkatesh, V. & Davis, F. (1996). A model of the antecedents of perceived ease of use: Development and test. Decision Sciences, 27(3), 451-481. VHA Inc. introduces new version of its comparative clinical measurement tool. (2007). Healthcare Purchasing News, 31(3), 8-14. Wells, L. (2007). Role of information technology in evidence-based medicine: Advantages and limitations. Internet Journal of Healthcare Administration, 4(2), 5-10. Zhang, J. & Kelvin, A. (1999). Strategic entry for products and technology into China. Journal of International Marketing and Marketing Research, 24(2), 14-29, 85-98.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 1, edited by J. Tan, pp. 75-92, copyright 2008 by IGI Publishing (an imprint of IGI Global).

83

84

Chapter 6

Using a Neural Network to Predict Participation in a Maternity Care Coordination Program George E. Heilman Winston-Salem State University, USA Monica Cain Winston-Salem State University, USA Russell S. Morton Winston-Salem State University, USA

aBstract Researchers increasingly use Artificial Neural Networks (ANNs) to predict outcomes across a broad range of applications. They frequently find the predictive power of ANNs to be as good as or better than conventional discrete choice models. This paper demonstrates the use of an ANN to model a consumer’s choice to participate in North Carolina’s Maternity Care Coordination (MCC) program, a state sponsored voluntary public health service initiative. Maternal and infant Medicaid claims data and birth certificate data were collected for 59,999 births in North Carolina during the years 2000-2002. Part of this sample was used to train and test an ANN that predicts voluntary enrollment in MCC. When tested against a hold-out production sample, the ANN model correctly predicted 99.69% of those choosing to participant and 100% of those choosing not to participant in the MCC program.

IntroductIon Information technology (IT) plays a pervasive role throughout the healthcare industry. In addition to providing the data storage and data process-

ing capabilities needed to support the business management, customer relations management, human resource management and office automations requirements of healthcare organizations, IT also is used increasingly to support decision making functions.

DOI: 10.4018/978-1-61692-002-9.ch006

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Using a Neural Network to Predict Participation in a Maternity Care Coordination

Most decision support methodologies rely on the mathematical modeling of historical data. Many of these systems, such as the widely accepted Acute Physiology and Chronic Health Evaluation (APACHE) system, are based on binary LOGIT regression estimations or other statistical analysis techniques. This type of modeling requires the specification of a priori functional relationships between dependent and independent variables based on assumptions such as correct model specification, error-free measurement of independent variables, and normally distributed, heteroscedastistic, independent, zero-mean residuals. It is more likely, however, that healthcare decisions will depend on a variety of factors involving complex, hidden interrelationships of both sociodemographic and health related characteristics. To address issues of non-linearity and complex relationships in study data, many modelers have turned to other methods of analysis that fall under the broader categorization of “artificial intelligence” (AI). AI, which attempts to give computers human-like reasoning capabilities, includes techniques such at expert systems, fuzzy systems, genetic algorithms, case based reasoning and a variety of classifier systems like the Artificial Neural Network (ANN) used in this study. Because of advantages like ease of optimization, prediction accuracy, easy knowledge dissemination, workload reduction and decision support, Artificial Neural Networks have been widely accepted and used for more than a decade in the healthcare arena (Lisboa & Taktak, 2006). When used in medical applications, ANNs are known to provide decision support assistance that can produce highly accurate results (Kaur & Wasan, 2006) and better predictive performance than other modeling alternatives (Alkan et al., 2005; Alpsan et al., 1995; Goss & Vozikis, 2002). Medical applications of ANN include diverse examples ranging from the analysis trauma data (Chesney et al., 2006; Eftekhar et al., 2005) to predicting the contact map structures of proteins (Chen et al., 2008). ANNs have proven useful in

medically related classifications including drug and non-drug chemical compounds (Pehlivanli et al., 2008), genes (Wang et al., 2008), heart sounds (Ari & Saha, 2008), liver abnormalities (Poonguzhali & Ravindran, 2007) and types of epileptic seizures (Najumnissa & ShenbagaDevi, 2008). ANNs also have been successfully applied to the diagnoses of cancer (Lisboa & Taktak, 2006), cardiac state (Samanta & Nataraj, 2008), diabetes (Kuar & Wasan, 2006), gastrointestinal hemorrhage (Das et al., 2003) and myocardial infarction (Baxt, 1991; Baxt et al., 2002). While medical diagnosis is probably the most common healthcare application for Artificial Neural Networks, ANNs also have been used successfully in other healthcare areas. Examples include assessing community-level vulnerability to methamphetamine manufacture (Dalmadge & Cain, 2008), evaluating if patient debt is likely to be repaid (Zurada & Lonial, 2005), evaluating the severity of risks on healthcare non-clinical business operations (Okoroh et al., 2007), identifying individuals at risk for high medical costs (Crawford et al., 2005), identifying sources of future high resource demand (Kudyba et al., 2006), and predicting nursing staff levels (Seomun et al., 2006). While ANNs are generally well accepted and frequently used in the healthcare industry, one sector that does not seem to have taken advantage of this technology is public healthcare services. Although some high-level examples of ANN use for public health issues exist, such as assessing community-level vulnerability to methamphetamine manufacture (Dalmadge & Cain, 2008) and forecasting demand for national immunization vaccines (Choy & Kuo, 2006), little has been developed at the public program level. While facing the expected pressures for professionalism and quality service, public healthcare programs also faces the additional burdens of budgetary restrictions and legislative oversight. As a result, public healthcare programs, like other sectors of healthcare, are intensifying their focus on the enhancement of operating efficiency through

85

Using a Neural Network to Predict Participation in a Maternity Care Coordination

effective resource allocation. One way to enhance efficiency is to more accurately identify resource demands. Since Artificial Neural Networks excel at identifying relationships in historical data for purposes of classification and prediction, it follows that using an ANN to predict participation in a public healthcare program should improve predictive capability, reduce inefficient resource allocation, and decrease variability in treatment processes (Kudyba et al., 2006). North Carolina’s Maternity Care Coordination (MCC) program represents one such public healthcare initiative. The Maternity Care Coordination program attempts to coordinate prenatal care for eligible Medicaid clients. MCC provides counseling, referrals, and resource assistance for women considered to be at high risk for poor birth outcomes. Participation in the program, however, is voluntary. The goal of this study is to develop an artificial neural network that can predict Medicaid women’s voluntary enrollment in the MCC program.

MaternIty care coordInatIon enHancIng prenatal care States have attempted to address the issue of poor neonatal outcomes by incorporating comprehensive, coordinated prenatal care programs into their Medicaid plans. These prenatal care programs are enhanced beyond the scope of the traditional medical model to include services such as health education, psychosocial risk assessment, enrollment in WIC (the Special Supplemental Food Program for Women, Infants, and Children), and other types of health promotion interventions (Gehshan et al., 2009). Unfortunately, attempts to evaluate the effectiveness of comprehensive prenatal care coordination have had mixed results (Buescher et al., 1991; Cain, 2006; Cain, 2007; Clarke et al., 1993; Herman & Berendes, 1996; Korenbrot & Patterson, 1995; Schulman et al., 1997; Nason & Alexander, 2002).

86

The North Carolina Maternity Care Coordination program (MCC) is an example of an early prenatal care coordination program, and is the focus of this study. MCC has the objective of reducing barriers to Medicaid clients’ use of health and social services. The program is geared toward helping eligible women receive nutritional care, psychosocial counseling, and other resource assistance. For example, women in MCC are encouraged to seek eligible services such as transportation, housing assistance, and job training. Counseling may include social and emotional support, stress reduction methods, and coaching in healthy behaviors. Referral for WIC enrollment is emphasized, and most women enrolled in MCC receive nutritional counseling through WIC (Buescher & Horton, 2002). Medicaid women who are perceived to be at very high risk for a poor birth outcome are most likely to be referred to the MCC program by their physician. Participation in MCC is, however, voluntary and thus at the discretion of the Medicaidenrolled woman. If she chooses to participate in MCC, she will have access to expanded prenatal care; if she chooses not to participate, she will receive the standard package of care available through the state’s Medicaid program. Various observable maternal medical and socioeconomic risks may influence a pregnant woman’s decision to participate in an enhanced prenatal care program and may even be the same risk factors that prompted the healthcare provider to recommend the program in the first place. In actuality, however, it may be the unobserved relationships among these factors that influence her final decision to participate in an enhanced prenatal care program. A variety of complex, hidden personal factors may not be reflected in the medical or intake records. These may include health or domestic issues that the woman is not willing to reveal or able to discern. For example, a woman at higher risk for a poor birth outcome may be facing stressors in the local environment such as violence, use of

Using a Neural Network to Predict Participation in a Maternity Care Coordination

alcohol or illicit drug activity in the community. These factors may influence her choice to participate, even while remaining unobserved by the healthcare provider. In this study we test whether neural network learning is able to model complex nonlinear patterns between a diverse set of predictor variables and the choice to participate in MCC.

study data This study uses data developed by the North Carolina State Center for Health Statistics (SCHS) called the Composite Linked Birth File. The Composite Linked Birth File is comprised of the linkage of a unique birth certificate record to any Medicaid-paid infant claims records, MCC and WIC enrollment records, and/or infant death certificate record. The data include a census of all births in the state of North Carolina in years 20002002. The total number of North Carolina resident live births was 120,247 in 2000, 118,112 in 2001, and 117,307 in 2002. Among these, Medicaidpaid births based on the infant’s Medicaid status numbered 49,188 in 2000, 51,720 in 2001, and 48,883 in 2002. Births to women enrolled in the Maternity Care Coordination program were 24,694 in 2000, 24,328 in 2001, and 19,637 in 2002. Thus, approximately 50 percent of North Carolina Medicaid women who delivered during the study period were enrolled in the MCC program. The sample includes only Medicaid births to white and African-American women aged 15-45 years. Three additional selection criteria were established to address issues of bias in the sample selection. First, women who enrolled in the MCC program after 32 weeks gestation were not included, thereby excluding late joiners. Second, only live singleton births were included. And third, births to women who received no prenatal care were excluded. In addition to the selection criteria described above, records with missing data for any study variables were excluded. The application of the selection criteria to the original

355,666 observations from the years 2000-2002 resulted in a sampling frame of 137,249 Medicaid births, including 57,635 births to women participating in the MCC program. We randomly selected 60,000 of these observations for use in the study’s data sample.

predIctor VarIaBles The state of North Carolina collects up to 300 pieces of maternal and infant information for each birth included in the Composite Linked Birth File. We selected nineteen variables that previously have been identified as important in predicting if a woman is at risk for a high risk pregnancy and, therefore, more likely to participate in the MCC program (Paneth, 1995; Schwethelm et al., 1989). These predictor variables fall into three general categories: 1) location, 2) overall health, and 3) pregnancy-related health. Specifically, the predictor variables include mother’s zip code (MOMZIP), county of residence (COUNTY), mother’s age (MOMAGE), mother’s race (MOMRACE), last year of mother’s schooling (MOMEDUC), marital status (MSTATUS), self-reported tobacco use (TOBACCO), selfreported alcohol use (ALCOHOL), the month prenatal care began (MMPNCBEG), presence of medical risk factor(s) (MEDRISK), and a previous death of a live newborn (PREVDETH). The input variables include several other dummy variables (0 or 1) indicating the presence of specific medical conditions such as anemia (ANEMIA), cardiac problems (CARDIAC), diabetes (DIABETES), chronic hypertension (HYPERCH), pregnancyrelated hypertension (HYPERPR), eclampsia (ECLAMPSIA), pre-term birth or small for gestation age in a previous pregnancy (SGA), and renal disease (RENAL). MCC is the outcome variable, where 1 indicates that the woman chooses to participate in the MCC program and 0 indicates the choice not to participate.

87

Using a Neural Network to Predict Participation in a Maternity Care Coordination

artIFIcIal neural networK Model Back propagation network learning is a commonly used algorithm that has been proven to be a reliable tool for general classification applications (Alspan et al., 1995; Hornik et al., 1989; Medsker & Liebowitz, 1994; White, 1990). For this study, the NeuroShell2 software package was used to define and test a three layer, feed forward, back propagation Artificial Neural Network model. Figure 1 shows a simple propagation network of the type used in this study. In the diagram, the circles are neurons (mathematical processing units) and the lines connecting the neurons are numeric weights. The models are typically composed of: 1) a layer of input neurons, wherein each input neuron represents one predictor variable, 2) one or more hidden layers of weighting neurons, and 3) an output layer containing a neuron for each dependent variable to be predicted. Our neural network model includes an input layer with 19 neurons, one hidden layer containing 100 neurons, and an output layer containing one neuron to predict MCC participation. During the training phase of network development, the ANN must be trained to predict actual output values based a correctly described set of Figure 1. Diagram of simple propagation artificial neural network

88

weighted connections in the input and hidden layers. In the beginning the weight values are randomly set, the input values of the first case in the training set are presented to the network’s input layer, and the calculated output of each neuron is fed forward through the network until an output value is derived. Next, the computed output value is compared to the actual value of the output variable in the training case. The error value (the squared difference between actual and calculate output) is then propagated backward through the network and small adjustments are made to the weights so that, in the future, the network will come closer to calculating the correct output value. This process is repeated for each case in the training set, and training set is processed over and over while the weights are continually tweaked. Periodically, a holdout (test) set of cases is run against the network to determine if the predictive power of the network is still improving. The network will never learn an exact predictive function, but it will slowly approach one. The purpose of the test set is to make sure that the network is not over trained in recognizing relationships within its training set at the expense of loosing its predictive power with new data. The original sample used in this study consisted of 60,000 observations. This number is a limitation of the NeuroShell2 software. After the sample was selected one observation was found to have a 4-digit zip code and was dropped, leaving a total of 59,999 observations. Table 1 presents the descriptive statistics for the variables used in the analysis. Seventy percent (70%) of the observations (42,000 patterns) were used to train the network and twenty percent (20%) of the observations (12,000 patterns) were used as a test set to evaluate model training. Ten percent (10%) of the observations (5,999 patterns) were used as a production set to validate the trained model. The neural network software randomly selected the observations for each of these three data sets.

Using a Neural Network to Predict Participation in a Maternity Care Coordination

Table 1. Descriptive statistics for variables Variable

Min. Value

Max. Value

Mean

Std.Dev.

Zip Code

22717

65109

County

0

100

Mother’s age

15

45

23.57821

5.350374

Mother’s Race 1= White; 2= Afr.Am.

1

2

1.361606

0.480470

Mother’s yrs. schooling

0

16

11.42652

5.421363

Marital Status 1=married; 2= single

1

2

1.600160

0.489869

TOBACCO 1=yes; 2 = no

1

2

1.793197

0.527171

ALCOHOL 1=yes; 2 = no

1

2

1.867

0.347672

MMPNCBEG

0

9

2.808147

1.577052

MEDRISK

0

1

0.123819

0.329377

PREVDETH

0

1

0.018384

0.134335

ANEMIA

0

1

0.027617

0.163875

CARDIAC

0

1

0.003533

0.059338

DIABETES

0

1

0.024734

0.155314

HYPERCH

0

1

0.008467

0.091626

HYPERPR

0

1

0.050718

0.219422

ECLAMPSI

0

1

0.004917

0.069948

SGA

0

1

0.012284

0.110149

RENAL

0

1

0.003217

0.056625

MCC

0

1

0.429857

0.495060

Dummy Variables 0=no; 1=yes

Training continued through 35 epochs of the training set. The model’s weights were evaluated against the test set every 200 events. After 62,000 events had been processed with no changes to the weight structure, training was stopped. Finally, the production set was processed through the trained model to produce the network’s predictions for each pattern in the production file.

results NeuroShell2 provides a number of statistical tools to assess the predictive ability of the network model. Table 2 presents a summary of their values.

Following the table is a narrative description of the statistics. The coefficient of multiple determination, R2, compares the accuracy of the model to a benchmark model consisting of the mean of all the samples. An R2 value of 1 indicates a perfect fit, near 1 indicates a very good fit, and near or below zero indicates a very poor fit. The square of the correlation coefficient, r2, provides a measure of the strength of the relationship between the actual and predicted outputs. An r2 closer to 1 indicates a strong linear relationship, while an r2 closer to 0 indicates no linear relationship.

89

Using a Neural Network to Predict Participation in a Maternity Care Coordination

Table 2. Production set assessment statistics Output R

2

5999 patterns 0.9917

correlation coefficient r

0.9960

r2

0.9920

Mean Squared Error

0.0020

Mean Absolute Error

0.0120

Min. Absolute Error

0

Max. Absolute Error

0.8460

Percent within 5%

41.357

Percent within 5% to 10%

0.1000

Percent within 10% to 20%

0.0170

Percent within 20% to 30%

0.1830

Percent over 30%

0.7830

Min Absolute Error is the minimum absolute value of (actual - predicted) for all patterns in the production set. Max Absolute Error is the maximum absolute value of (actual - predicted) for all patterns. Mean Absolute Error is the mean of the absolute value of (actual - predicted) for all patterns. Mean Squared Error is the mean of (actual – predicted)2 for all patterns. NeuroShell2 also lists the percent of network predictions that differ from the actual answers within a range of specified percentages. When an actual answer is 0, a percentage difference cannot be calculated. As a result, the total computed percentages may not total 100%. The 5999 patterns in the production set included observations for 3453 women who rejected participation in MCC and 2545 women who chose to participate in MCC. The model correctly predicted all 3453 non-participants (100%) and 2537 of 2545 participants (99.69%).

conclusIon States face increasing pressure to limit growth in their Medicaid budgets while still meeting the healthcare needs of vulnerable populations. Thus,

90

predicting the number of eligible recipients of public healthcare services becomes a significant challenge for program managers, especially when participation is voluntary. To meet this challenge, the use of predictive models has become much more important in an ever-widening variety of public health programs. Conventional prediction methods, which excel at identifying causal effects and linear relationships among variables, require a priori model specification and may not be able to detect underlying complex relationships among predictor variables. When prediction accuracy is more important than assessing causal effects, Artificial Neural Networks offer the user a powerful prediction tool without the need to understand all the subtle relationships that may exist within the target population. Additionally, significant advances in computer hardware and software technology have made Artificial Neural Networks an increasingly powerful, available, affordable, useful and user-friendly tool. Public healthcare resource allocation and planning can be greatly enhanced by the predictive power of ANNs, particularly in terms of voluntary program participation. This study develops an Artificial Neural Network to predict voluntary enrollment by Medicaid women in North Carolina’s Maternity Care Coordination (MCC) program for enhanced prenatal care. Observations from 59,999 randomly selected Medicaid births from the years 2000-2002 were used in the development and testing of the ANN. The model contained 19 predictor variables related to location, overall health, and pregnancy-related health to predict voluntary participation in the enhanced prenatal care program. The model correctly predicted the mother’s choice to participate in MCC over 99% of the time. Using the superior predictive power of Neural Networks, managers can make better funding, staffing, and other resource allocation decisions based on timely, correct predictions of program participation. Neural Network models can be updated as new information is obtained and re-

Using a Neural Network to Predict Participation in a Maternity Care Coordination

trained to recognize new relationships among predictor variables that may result from evolving changes in participant characteristics and/or program modifications. This research path may be extended in many directions. Since so little research has been published on the use of Artificial Neural Networks in the prediction of voluntary participation in public healthcare programs, one obvious extension is to expand the study of ANNs in the prediction of voluntary participation in other types of healthcare programs. Other extensions of the study could also be made to predict voluntary program participation in other areas outside of health services, such as education, child care, utility subsidies, and transportation.

Buescher, P. A., & Horton, S. J. (2002). Prenatal WIC participation in relation to low birth weight and Medicaid infant costs in North Carolina-An update. Center for Health Informatics and Statistics Studies, 122, 1–9.

reFerences

Cain, M. (2007). Self-selection and evaluation of health care programs. Journal of Academy of Business and Economics, 7(3), 82–89.

Alkan, A., Koklukaya, E., & Subasi, A. (2005). Automatic seizure detection in EEG using logistic regression and artificial neural network. Journal of Neuroscience Methods, 148(2), 167–176. doi:10.1016/j.jneumeth.2005.04.009 Alpsan, D., Towsey, M., Ozdamar, O., Tsoi, A. C., & Ghista, D. N. (1995). Efficacy of modified backpropogation and optimization methods on a realworld medical application. Neural Networks, 8(6), 945–962. doi:10.1016/0893-6080(95)00034-W Ari, S., & Saha, G. (2008). Classification of heart sounds using empirical mode decomposition based features. International Journal of Medical Engineering and Informatics, 1(1), 91–108. doi:10.1504/IJMEI.2008.019473 Baxt, W. G. (1991). Use of an artificial neural network for the diagnosis of myocardial infarction. Annals of Internal Medicine, 115(11), 843–848. Baxt, W. G., Shofer, F. S., Sites, F. D., & Hollander, J. E. (2002). A neural computational aid to the diagnosis of acute myocardial infarction. Annals of Emergency Medicine, 39(4), 366–373. doi:10.1067/mem.2002.122705

Buescher, P. A., Roth, M. S., & Williams, D. (1991). An evaluation of the impact of maternity care coordination on Medicaid birth outcomes in North Carolina. American Journal of Public Health, 81(12), 1625–1629. doi:10.2105/ AJPH.81.12.1625 Cain, M. (2006). A Two-step estimation applicaton in health care management evaluating the effects of choice of public health plan on health outcomes. Journal of Academy of Business and Economics, 6(2), 65–72.

Chen, P., Huang, D., Zhao, X., & Li, X. (2008). Predicting contact map using Radial Basis Function Neural Network and Conformational Energy Function. International Journal of Bioinformatics Research and Applications, 4(2), 123–136. doi:10.1504/IJBRA.2008.018340 Chesney, T., Penny, K., Oakley, P., Davis, S., Maffulli, N., & Templeton, J. (2006). Data Mining Medical Information: Should Artificial Neural Networks Be Used To Analyze Trauma Audit Data. International Journal of Healthcare Information Systems and Informatics, 1(2), 51–64. Chiu, R., & Kuo, C. (2006). A decision support model of demand forecast for national immunization vaccines. International Journal of Electronic Healthcare, 2(1), 92–104. doi:10.1504/ IJEH.2006.008699 Clarke, L. L., Miller, M. K., & Mahan, C. S. (1993). The Effectiveness of Florida’s Improved Pregnancy Outcome Program. Journal of Health Care for the Poor and Underserved, 4(2), 117–132.

91

Using a Neural Network to Predict Participation in a Maternity Care Coordination

Crawford, A. G., Fuhr, J. P., Clarke, J., & Hubbs, B. (2005). Comparative Effectiveness of Total Population versus Disease-Specific Neural Network Models in Prediciting Medical Costs. Disease Management, 8(5), 277–287. doi:10.1089/ dis.2005.8.277 Dalmadge, C. L., & Cain, M. (2008). Utilizing Neural Networks o Assess Community-Level Vulnerability to the Methamphetamine Epidemic. Issues in Information Systems, 9(2), 258–266. Das, A., Ben-Menachem, T., Cooper, G. S., Chak, A., Sivak, M. V. Jr, Gonet, J. A., & Wong, R. C. (2003). Predicition of outcome in acute lower-gastrointestinal hemorrahage based on and artificial neural network: Internal and external validation of a predictive model. Lancet, 362(9392), 1261– 1266. doi:10.1016/S0140-6736(03)14568-0 Eftekhar, B., Mohammad, K., Ardebili, H. E., Ghodsi, M., & Keabchi, E. (2005). Comparison of artificial neural network and logistic regression models for prediction of mortality in head trauma based on clinical data. BMC Medical Informatics and Decision Making, 5(3), 1–8. Gehshan, S., Belnap, D., & Snyder, A. (2009). Medicaid Outreach and Enrollment for Pregnant Women: What Is the State of the Art?Washington, DC: Urban Institute. Goss, E., & Vozikis, G. S. (2002). Improving Health Care Organizational Management Through Neural Network Learning. Health Care Management Science, 5(3), 221–227. doi:10.1023/A:1019760901191 Herman, A. A., & Berendes, H. W. (1996). The evaluation of the effectiveness of a communitybased enriched model of prenatal intervention project in the District of Columbia. Health Services Research, 33(5), 609–621.

92

Hornik, K., Stinchcombe, M., & White, H. (1989). Multilayer feedforward networks are universal approximations. Neural Networks, 2(5), 359–366. doi:10.1016/0893-6080(89)90020-8 Kaur, H., & Wasan, S. K. (2006). Empirical study on Applications in Data Mining Techniques in Healthcare. Journal of Computer Science, 2(2), 194–200. doi:10.3844/jcssp.2006.194.200 Korenbrot, C. C., Gill, A., & Patterson, E. (1995). Evaluation of California’s statewide implementation of enhanced perinatal services as Medicaid benefits. Public Health Reports (Washington, D.C.), 110(2), 125–133. Kudyba, S., Hamar, G. H., & Gandy, W. M. (2006). Utilising neural network applications to enhance efficiency in the healthcare industry: prediciting populations of future chronic illness. International Journal of Business Intelligence and Data Mining, 1(4), 371–383. doi:10.1504/ IJBIDM.2006.010780 Lisboa, P. J., & Taktak, A. F. G. (2006). The use of artificial neural networks in decision support in cancer: A systematic review. Neural Networks, 19, 408–415. doi:10.1016/j.neunet.2005.10.007 Medsker, L., & Liebowitz, J. (1994). Design and Development of Expert Systems and Neural Networks. New York: McMillan. Najumnissa, D. & ShenbagaDevi, S. (2008). Intelligent identification and classification of epileptic seizures using wavelet transform. International Journal of Biomedical Engineering and Technology, 1(3), 293–314. doi:10.1504/ IJBET.2008.016963 Nason, C. S., Alexander, G. R., Pass, M. A., & Bolland, J. M. (2001). An evaluation of a Medicaid managed maternity program: The impact of comprehensive care coordination on utilization and pregnancy outcome. Journal of Health and Human Resources Administration, 24(3/4), 493–521.

Using a Neural Network to Predict Participation in a Maternity Care Coordination

Okoroh, M. I., Ilozor, B. D., & Gombera, P. (2007). An artificial Neural Networks approach in managing healthcare. International Journal of Electronic Healthcare, 3(3), 279–292. doi:10.1504/ IJEH.2007.014548 Paneth, N. S. (1995). The Problem of Low Birth Weight. The Future of Children: Low Birth Weight, 5(1), 21–34. Pehlivanli, A. C., Ibrikci, T., & Ersoy, O. K. (2008). Consensual classification of drug and nondrug compounds. International Journal of Computational Biology and Drug Design, 1(3), 224–234. doi:10.1504/IJCBDD.2008.021416 Poonguzhali, S., & Ravindran, G. (2007). Evaluation of feature extraction methods for classification of liver abnormalities in ultrasound images. International Journal of Biomedical Engineering and Technology, 1(2), 134–143. doi:10.1504/ IJBET.2007.015856 Samanta, B., & Nataraj, C. (2008). Automated diagnosis of cardiac state in healthcare systems using computational intelligence. International Journal of Services Operations and Informatics, 3(2), 162–177. doi:10.1504/IJSOI.2008.019331 Schulman, E. F., Sheriff, D. J., & Momany, E. (1997). Primary Case Management and Birth Outcomes in the Iowa Medicaid Program. American Journal of Public Health, 87(1), 80–84. doi:10.2105/AJPH.87.1.80

Schwethelm, B., Margolis, L. H., Miller, C., & Smith, S. (1989). Risk Status and Pregnancy Outcome among Medicaid Recipients. American Journal of Preventive Medicine, 5(3), 157–163. Seomun, G., Chang, S. O., Lee, S. J., Kim, I. A., & Park, S. (2006). A Prediction Model for Patient Classification according to Nursing Need: Using Data Mining Techniques . In Park, H., Murray, P., & Delaney, C. (Eds.), Consumer-Centered Computer-Supported Care for Healthy People. Fairfax, VA: IOS Press. Subasi, A., & Ercelebi, E. (2005). Classification of EEG signals using neural network and logistic regression. Computer Methods and Programs in Biomedicine, 78(2), 87–99. doi:10.1016/j. cmpb.2004.10.009 Wang, H., Zheng, H., & Hu, J. (2008). Poisson approach to clustering analysis of regulatory sequence. International Journal of Computational Biology and Drug Design, 1(2), 141–157. doi:10.1504/IJCBDD.2008.020206 White, H. (1990). Connectionist nonparametric regression: multilayer feedforward networks can learn arbitrary mappings. Neural Networks, 5(6), 535–549. doi:10.1016/0893-6080(90)90004-5 Zurada, J., & Lonial, S. (2005). Comparison Of The Performance Of Several Data Mining Methods For Bad Debt Recovery In The Healthcare Industry. The Journal of Applied Business Research, 21(2), 37–53.

93

94

Chapter 7

Can IT Act as a Catalyst for Change in Hospitals? Some New Evidence Teemu Paavola LifeIT Plc, Finland

aBstract This chapter presents a succesful reorganization of a patient care process that was carried out in a middle sized Finnish hospital. The reorganization of the patient care process for joint replacement surgery succeeded in achieving a 50 per cent increase in operations. This study proposes that IT may have an indirect influence on the achievement of goals, such as productivity, as soon as the IT investment has been decided upon; in other words, IT benefits start accruing before the IT component is even in place. This is a new feature to add to the previous definitions, because this particular benefit cannot be logically derived from any of the features of the actual IT system. Paying enough attention to this phenomen at the planning stage can be vital to the success of new IT system investment.

IntroductIon In the last two decades, the assessment of the benefits of IT has given rise to an interactive dialogue between management sciences and information systems science in particular, but in health care the subject has received little attention. In the field of health care, IT investments are still seen as primarily an acquisition to replace earlier technology and expand current use, not as an investment project to be managed in line DOI: 10.4018/978-1-61692-002-9.ch007

with business goals. There are, however, many phenomena at play shaping the practices of the health care sector and identifying and allowing for these phenomena may be the key to successful IT system projects, indeed even more important than the technology itself. The literature on both IT management and process development is quite unanimous in its belief that both are necessary for achieving more efficient operation and a productivity increase. Therefore an IT system project is often a change project by nature, which can make it challenging particularly in the field of health care (Berg, 2001;

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Can IT Act as a Catalyst for Change in Hospitals?

Littlejohns, Wyatt, & Garvican, 2003), where resistance to change is virtually a characteristic of the profession (Weick & Sutcliffe, 2003). The Act on Specialized Medical Care concerning the maximum times to arrange treatment, which came into force in Finland in March 2005, has made many healthcare units look at the arrangement of the services they produce in a new light. Particular attention is fixed on the legal obligation concerning the waiting times between treatment decisions and treatment measures, which is to be no more than six months. The need to increase the number of operations has become a matter of current debate particularly in orthopaedics, where the length of queues has become unlawfully long at several hospitals in Finland. Improvements in controlling the queues have previously been achieved by the more efficient handling of referrals (Harno et al., 2000), but with orthopaedics this was felt to be ineffective (Harno et al., 2001). In special operative areas, making use of all the development potential available within the traditional treatment chains should be explored as a permanent remedy, after first-aid obtained in the form of outsourced services. This chapter illustrates a case where process management and process development tools where exploited to support new ways of work and improve productivity in healthcare. In Finland, Seinäjoki Central Hospital implemented a project to revise processes in order to reduce queues in surgery, particularly artificial-joint surgery. The project in question was originally classified as an IT project which also incorporated process development. Over the course of the project, however, the balance between the two components shifted, and there was no time to incorporate the new IT system before the changes were implemented. This chapter describes the results of the experiment for the benefit of, for example, other operation units that are taking a close look at their operations and of developers and management in health care as support for decision-making. Parts of the study have been published in Finnish language (Jokipii et al., 2006).

process ManageMent tHeorIes The term ‘process thinking’ refers to a number of management theories that have been used by industry in its quest for better operating processes over the last few decades. In many of these, the use of IT also has a significant role. Indeed, IT has become more important in a number of areas, including health care; yet process thinking has not always been employed. The populations of Europe and the Americas are ageing quickly. The healthcare system is struggling with the combination of rising demand and escalating costs in specialist medical care, while at the same time, there is strong support for reduced public-sector healthcare spending but firm rejection of any cuts in service levels. If the two targets are to become reality simultaneously, the methods enabling them to be achieved should be chosen on the basis of how deep the cuts should be. Cosmetic improvements would be fairly painless: for example, Total Quality Management (Crosby, 1979; Deming, 1991) would result in long-term improvements in operating processes as a more efficient use of resources would bring gradual savings. Some scholars have, however, likened some quality management theories to a rain dance (Schaffer & Thomson, 1992). In their view they look good, sound good and allow those involved to feel good, while at the same time they may have no influence on the rain itself. There are also other management theories in the field of process thinking. According to the time-based management approach, all development should focus on process lead-time (Stalk & Hout, 1990). In such an approach all other positive aspects, improved quality, cost savings and customer satisfaction will follow automatically. However, development measures do not need to mean squeezing more out of the stages intended to boost the value of the treatment process. In fact, industrial companies have been able to find larger savings in the way they use the time that brings no added value, which, after all, 95

Can IT Act as a Catalyst for Change in Hospitals?

accounts for more than 95% of the total (Stalk & Hout, 1990). In contrast to Total Quality Management, which emphasizes continuous development, Business Process Reengineering (BPR) proposes a radical revision of the business process. The aim is to start from scratch without the burden of old operating approaches (Oliver, 1993; Hammer & Champy, 1993). The reengineering starts with a definition of the desired end result. This will form the basis for the planning of the new process functions and sequences. The aim is to maximize value-adding functions and to get rid of all operations not adding to the value. Extensive use of the information technology is often used as the means for achieving the desired results. Effecting the operational changes required by BPR has been somewhat problematic. Resistance to change, which is inherent in human nature, and the fact that reengineering is often a zero-sum game, make the implementation of the change process more difficult (Buchanan, 1997). The theory, though somewhat worn-out, is still useful, as it underlines the importance of the information technology in performance improvement. Great potential for applying the theory and information technology can be found in sectors that for ages have relied on well-entrenched operating models, such as health care (Evans, Hwang, & Nagarajan, 1997). The importance of IT for productivity in different organizations has been discussed for decades. The discussion led by Straussmann (1990) and Brynjolfsson (1993) has particularly focused on explaining what is known as the IT Productivity Paradox. Although Brynjolfsson et al. (1993 and 1996) later declared that the problem had disappeared by 1991, not all researchers have agreed, and interest in explaining the phenomenon remains high — so much so that Texas-based reseacher Mark Anderson et al. (2003) proposed a new IT Productivity Paradox to replace the original. They observed a growth in the market worth of businesses after Y2K investments leading up to the turn of the millennium. These investments were made 96

for replacing systems already in use, yet they still resulted in significant increases in productivity. Anderson et al. proposed this as a new, inverse phenomenon to be explained, calling it the New Productivity Paradox. According to their theory, the weak IT impact on productivity especially immediately after the turn of the millennium (20002002) can be attributed to companies simply not investing enough in IT, in direct contradiction to the IT Productivity Paradox. Lee and Menon (2000) argue that hospitals that are characterized by hight technical efficiency are no more productive than hospitals characterized counterwise. In fact, in the hospitals studied IT capital seemed to have a negative correlation to productivity. They explain this by the fact that although hospital’s processes may have been efficient, resource allocation and budgeting between various categories of capital and labor had not been efficient. Devaraj and Kohli (2000) believe that the effect of IT on performance can been seen only after a time lag and cannot necessarily be observed in cross-sectional or snapshot data analyses. With data collected from eight hospitals, their study indicates support for the impact of technology contingent on BPR practiced by hospitals.

startIng poInts For cHange and experIMent Seinäjoki Central Hospital wanted to reorganize the operations for artificial-joint patients so that three operations could be performed in the same operating room in the course of a normal day’s work instead of two. The introduction of a new IT system was also planned as part of the project. The experimental period lasted from November 2004 to November 2005. The revision of the treatment process utilized process thinking and process development tools. Of these, the Theory of Constraints (Goldratt, 1990) was thought the best applicable for examining the process for treating artificial-joint patients. The point in this approach is to identify those stages

Can IT Act as a Catalyst for Change in Hospitals?

in the process that dictate the maximum current throughput. By allocating additional resources and development action to these bottlenecks, the throughput can be improved without needing to interfere in the other stages of the process. The main change for increasing the usage of the operating room capacity was transferring the anaesthetic stage from the operating room to separate induction facilities. Experiments on this had been reported earlier in medical journals (Hanss et al., 2005; Sandberg et al., 2005; Torkki et al., 2005). In the new arrangement, the anaesthetic stage was transferred outside the operating room. At the same time, one anaesthesia nurse was added to the operating team, working both in the operating room and in anaesthetic. Another anaesthetic nurse took the next patient in good time to the recovery room or to the operating room’s induction facilities to be anaesthetized. As soon as the operating room was cleaned after the previous operation, the next patient could be prepared for surgery. The next patient was brought to the operating room already anaesthetized and in the correct position for the operation. The duties of the orthopaedist that were not part of the operations or preparation for them were scheduled outside the operation days. Thus, the surgeon whose turn it was to operate was able to focus exclusively on the work in the operating room. At the beginning of the experimental period, the same orthopaedist operated for one week at a time, but this practice had to be changed so that the operation days were rotated among different practitioners. At the beginning of 2005, there were five orthopaedists working at Seinäjoki Central Hospital.

MaterIal, MetHods and results In Finland, every year about 6,800 artificial joint operations are carried out on the hip and some 7,200 on the knee, and there are more than 1,700 instances of further surgery. These operations are performed in almost 70 hospitals, but the mini-

mum number of 200 operations recommended by the Ministry of Social Affairs and Health is only exceeded in 25 units. Every year the Seinäjoki Central Hospital performs between 550 and 600 artificial-joint operations. In the study, quantitative material was collected from the operating days in the experimental period on which three artificial-joint operations were carried out (147 patients); because of the small number of orthopaedists, there were 2-3 of these days in a week. Comparative material consisted of the days on which two artificial-joint operations were carried out between January 1 and June 30, 2004 (54 patients). The time when patients were in the operating room and changeover times were recorded in the operation database. The time-monitoring material consisted of the times when the operating room was in use. The median time that patients were in the operating room and the median changeover time, when there is no patient in the operating room, were used for comparison purposes. Qualitative material was collected through interviews during the experimental period and by means of a work-satisfaction questionnaire carried out among doctors and nurses a year after the experiment started. The new operating model made it possible to carry out three operations during a normal working day (see Figure 1). The orthopaedists examined the patients during a pre-operative visit or on the day preceding the operation. The first patient of the morning was in the operating room in time, and the operation started on time at 8.30 a.m. The anaesthetization stages for the second and third patients, which were carried out staggered with the operation, took slightly longer than if carried out in the operating room. As it was possible to separate some of the steps previously carried out in the operating room and have them done outside, the hospital succeeded in increasing the throughput of the process by 50%, even though the usage capacity of the operating room remained almost the same.

97

Can IT Act as a Catalyst for Change in Hospitals?

Figure 1. The old arrangement and the new operating model for artificial joint surgery

Adding fourth nurse to the operating team (now 2 in anaesthetization and 2 in the operation) made it possible to shorten the changeover times considerably: the average time was reduced from 54 minutes to 13 minutes. This was because the team was able to take coffee and meal breaks in turn. One of the operation nurses was able to help the orthopaedist as necessary. In the three-operation model, anaesthetizing the second or third patient of the day in separate facilities reduced the time the patient was in the operating room by 20 minutes (149 minutes vs. 129 minutes). According to the questionnaire, fifty per cent or more of the doctors who took part in the experiment felt that the meaningfulness of their work and work motivation had increased and thought that the three-operation experiment should become a permanent fixture. The nursing staff felt that minimizing the idle waiting improves the atmosphere and increases work motivation to some extent. The doctors felt the new operating model improves the meaningfulness of the work and work motivation more than the nursing staff did.

dIscussIon The usage capacity of the operating room is generally considered to be the bottleneck in the operation process. This generalization leads easily to a

98

practice where outsourced services or increasing the number of a hospital’s own operating rooms are seen as the only options for increasing output. From our experiences the throughput of the process for artificial-joint operations can be increased while the usage capacity of the operating room remains the same or even decreases. Focusing the operations on one operating room proved to be effective. An increase in the throughput of the operation process was sought without increasing the workload of the staff. The hospital succeeded in doing this by firstly dealing with idle waiting. Targeting greater efficiency here and a simultaneous improvement in the throughput required development in several areas, e.g. adding one nurse to the operating team, a bigger work contribution from the hospital attendant in preparing patients, preparing the anaesthetic in a new way and changing the orthopaedist’s work schedule. The justification for adding one nurse was that in the revised staggered operation stage, there was also one patient more. It was not possible to anticipate all the effects of the change. In order to ensure that things went smoothly, specialist experienced doctors acted as anaesthetists and orthopaedists during the experimental period, but at the same time the arrangement narrowed the opportunities for training specializing doctors. Furthermore, not

Can IT Act as a Catalyst for Change in Hospitals?

enough preparation was made for the increase in the number of operations at all stages of the treatment process. At times, the growth in the throughput caused congestion on the ward and especially in further treatment at health centres. In financial terms, the transfer to the practice of three operations was worthwhile. The resources for arranging three operations were obtained principally by utilizing the fixed costs of the hospital more efficiently. In alternative cost accounting comparing the additional cost caused by a hospital’s own activities with the cost of an artificialjoint operation acquired from the private sector or another provider (minus the costs of the prosthesis, materials and cost of the treatment days) shows a difference of some USD 4,000 between the hospital’s own work and outsourcing with regard to the added third primary operation per day. Because of the limited number of orthopaedists, however, it was not possible in the experimental period to run ‘flat out’ five days a week. 15 complete operations a week would be enough to meet the need for artificial-joint surgery in the entire hospital district, and the revised treatment process would generate annual savings of between USD 700,000 and USD 800,000 for the Hospital District of South Ostrobothnia, even taking into account the additional recruitment required. Savings come from cutting back on services purchased from private hospitals, as the hospital itself can now perform a larger percentage of the joint replacement surgery required. Although planned as part of the project, the introduction of the new IT system did not take place during the present change project. The reason for this was that the software was not complete when the project was launched. Nevertheless, the IT system appears to have played a noteworthy role in the launching of the project because key players whose commitment was essential to the successful implementation of the change were motivated by the eventual benefits of the IT system to support the project. The IT system was intended to work as a tool for monitoring patients in rehabilitation

and to make it easier to organize follow-up visits in a timely fashion and based on actual needs. The new IT system would further reduce the workload at the hospital by making it possible to carry out post-operative patient monitoring at local health centres rather than at the hospital outpatient clinic.

conclusIon This chapter presents a succesful reorganization of the care process for an artificial joint patient. The project by a middle sized Finnish hospital offers an encouraging example of a way to exploit process management tools in health care. Seinäjoki Central Hospital succeeded in obtaining a 50% increase in flow-through in the process for treating artificial-joint patients with the transfer of the anaesthesia stage outside the operating room in the reorganization. For every two joint replacement operations previously conducted, there were now three operations performed in the same theatre and in a normal working day. In the longer term, the arrangement would mean that in Finland the entire country’s need for artificial-joint surgery, about 15,600 operations per annum, could be dealt with in 30 operating rooms. This could considerably streamline the publicly-financed health care system in Finland, as these operations are currently performed in almost 70 hospitals. The introduction of the new patient care process demonstrated that the operating theatre capacity was not causing a bottleneck, but that it was the orthopaedic surgeons brought in at the various intervals who formed the key resource. The reorganized care process for patients requiring joint replacement surgery should produce annual cost savings of USD 700,000 to USD 800,000 for the Seinäjoki Central Hospital. Following the experience gained in the project, the Seinäjoki Central Hospital has decided to adopt the project model on a permanent basis. A similar reorganization is also possible in other hospital districts. This

99

Can IT Act as a Catalyst for Change in Hospitals?

observation, however, should only be applied to orthopaedic joint replacement surgery. An interesting detail in this change project was that the role of the IT system itself was completely marginal. All kinds of justifications for investing in IT systems can be given, from economic calculations to managerial intuition (Paavola, 2007), but in the change project described in this chapter, IT seemed to act only as a catalyst (Paavola, 2008). Perhaps the project could not have been successfully implemented to the extent it was had not an IT component, which “always requires changes in the operational process”, been included. This observation of IT acting as a catalyst for performance enhancement and hence improved productivity deserves further study. It would be worth exploring to what extent indirect impact should be considered in making IT investments alongside direct effects, and how this phenomenon could be studied. Are there cases where IT investment was motivated as leverage for achieving a particular change in the operating environment, even though the same impact on productivity could have been achieved without IT investment?

reFerences Anderson, M., Banker, R., & Ravindran, S. (2003). The new productivity paradox. Communications of the ACM, 46(3), 91–94. doi:10.1145/636772.636776 Berg, M. (2001). Implementing information systems in health care organizations: myths and challenges. International Journal of Medical Informatics, 64(2/3), 143–156. doi:10.1016/ S1386-5056(01)00200-3 Brynjolfsson, E. (1993). The productivity paradox of information technology. Communications of the ACM, 36(12), 67–77. doi:10.1145/163298.163309

100

Brynjolfsson, E., & Hitt, L. (1996). Paradox lost? Firm level evidence on the returns to information systems spending. Management Science, 42(4), 541–558. doi:10.1287/mnsc.42.4.541 Buchanan, D. (1997). The limitations and opportunities of business process re-engineering in a politicized organizational climate. Human Relations, 50, 51–72. doi:10.1177/001872679705000103 Crosby, P. B. (1979). Quality is free. New York: McGraw-Hill. Deming, W. E. (1991). Out of crisis. Cambridge, MA: Cambribge University Press. Devaraj, S., & Kohli, R. (2000). Information technology payoff in the health-care industry: a longitudinal study. Journal of Management Information Systems, 16(4), 41–67. Evans, J., Hwang, Y., & Nagarajan, N. (1997). Cost reduction and process reengineering in hospitals. Journal of Cost Management, 11(3), 20–27. Goldratt, E. M. (1990). Theory of constrains. Croton-on-Hudson. NY: North River Press. Hammer, M., & Champy, J. (1993). Reengineering the corporation - A manifesto for business revolution. New York: Harper Business. Hanss, R., Buttgereit, B., Tonner, P. H., Bein, B., Schleppers, A., & Steinfath, M. (2005). Overlapping induction of anesthesia: An analysis of benefits and costs. Anesthesiology, 103(2), 391–400. doi:10.1097/00000542-200508000-00023 Harno, K., Arajärvi, E., Paavola, T., Carlson, C., & Arnala, I. (2001). Clinical effectiveness and cost analysis of patient referral by videoconferencing in orthopaedics. Journal of Telemedicine and Telecare, 7(4), 219–225. doi:10.1258/1357633011936435

Can IT Act as a Catalyst for Change in Hospitals?

Harno, K., Paavola, T., Carlson, C., & Viikinkoski, P. (2000). Patient referral by telemedicine: effectiveness and cost analysis of an intranet system. Journal of Telemedicine and Telecare, 6(6), 320–329. doi:10.1258/1357633001935996 Jokipii, P., Kalliovalkama, K., & Paavola, T. (2006). Tekonivelpotilaan hoitoprosessin uudistaminen. [Finnish Medical Journal]. Suomen Lääkärilehti, 61(14), 1626–1627. Lee, B., & Menon, N. (2000). Information technology value through different normative lenses. Journal of Management Information Systems, 16(4), 99–119. Littlejohns, P., Wyatt, J., & Garvican, L. (2003). Evaluating computerised health information systems: hard lessons still to be learnt. British Medical Journal, 326, 860–863. doi:10.1136/ bmj.326.7394.860 Oliver, J. (1993, August). Shocking to the core. Management Today, 18–22. Paavola, T. (2007). Justifying IT system investments. International Journal of Healthcare Technology and Management, 8(1/2), 97–106. doi:10.1504/IJHTM.2007.012104

Sandberg, W. S., Daily, B., Egan, M., Stahl, J. E., Goldman, J. M., Wiklund, R. A., & Rattner, D. (2005). Deliberate perioperative systems design improves operating room throughput. Anesthesiology, 103(2), 406–418. doi:10.1097/00000542200508000-00025 Schaffer, R., & Thomson, H. (1992). Successful change programs begin with results. Harvard Business Review, 22(1), 80–89. Stalk, G., & Hout, T. (1990). Competing against time - how time-based competition is reshaping global markets. New York: The Free Press. Strassmann, P. (1990). The business value of computers. New Canaan, CT: The Information Economic Press. Torkki, P. M., Marjamaa, R. A., Torkki, M. I., Kallio, P. E., & Kirvela, O. A. (2005). Use of anesthesia induction rooms can increase the number of urgent orthopedic cases completed within 7 hours. Anesthesiology, 103(2), 401–405. doi:10.1097/00000542-200508000-00024 Weick, K., & Sutcliffe, K. (2003). Hospitals as cultures of entrapment. California Management Review, 45(2), 73–84.

Paavola, T. (2008). Exploring IT system benefits in health care. Tampere University of Technology Publication 756. Tampere: Tampereen Yliopistopaino.

101

102

Chapter 8

Informatics Application Challenges for Managed Care Organizations: The Three Faces of Population Segmentation and a Proposed Classification System Stephan Kudyba New Jersey Institute of Technology, USA Theodore L. Perry Health Research Corporation, USA Jeffrey J. Rice Independent Scholar, USA

aBstract Organizations across industry sectors continue to develop data resources and utilize analytic techniques to enhance efficiencies in their operations. One example of this is evident as Managed Care Organizations (MCOs) enhance their care and disease management initiatives through the utilization of population segmentation techniques. This article proposes a classification system for population segmentation techniques for care and disease management and provides an evaluation process for each. The three proposed operational areas for Managed Care Organizations are: 1) Risk Status: early identification of high-risk patients, 2) Treatment Status: compliance with treatment protocols, and 3) Health Status: severity of illness or episodes of care groupings, all of which require particular analytic methodologies to leverage data resources. By applying this classification system an MCO can improve its ability to clarify internal goals for population segmentation, more accurately apply existing analytic methodologies, and produce more appropriate solutions.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Informatics Application Challenges for Managed Care Organizations

IntroductIon Population segmentation is the term broadly applied to technologies used to correctly identify and target the right patients for care and disease management program interventions. Many MCOs are using these technologies, including those containing predictive modeling techniques and quantitative applications, to enhance patient care and optimize available resources. With the rapidly growing use of population segmentation and predictive modeling today, it is essential to understand the relative strengths and weaknesses of different types of population segmentation techniques, including those employing predictive methods. The purpose of this article is to propose a classification system for the different types of population segmentation techniques and their usefulness in addressing the independent and interactive roles of risk status, treatment status, and health status in patient evaluation initiatives. Organizations across industry sectors have intensified their initiatives to increase operational efficiency through effective resource allocation, and the health care sector is no exception. Given the increased level of competition in today’s’ digital, information economy, organizations are faced with the task of increasing productivity by more efficiently allocating available resources in producing goods and services to meet the demands of their customers. One of the greatest issues facing the health care industry today is managing patients who suffer from chronic illnesses. Currently, approximately 100 million Americans have at least one chronic condition and this is expected to rise to over 150 million Americans in the next 20 to 30 years (Faughty, 1999; Institute of Medicine, 2001). Furthermore, chronic conditions are the leading cause of death, disability, and illness in the United States, accounting for over 75% of direct medical expenditures (Landro, 2002). The health care industry has been faced with a number of additional factors that have increased the complexity of managing available resources.

Some of these include an increase in the aging population, costs for defensive medicines, optimizing existing health care facility usage (e.g., staffing doctors and nurses along with designated bed utilization rates) and the introduction of new organizations such as HMOs and PPOs (SmithDaniels, Schweikhart, & Smith-Daniels,1988). One way Managed Care Organizations are attempting to improve their efficiencies in treating illnesses is through the development and management of robust data resources and the utilization of analytic techniques to identify patterns and trends in patient populations. With this information, efficiency can be enhanced by more accurately identifying the sources of resource demand of specific customer segments and initiating strategic health care management policies and better allocating available resources to meet those demands (Heskett, 1983; McLoughlin, Yan, & Van Deirdonck, 1995).

analytic Methods for Health care Management The utilization of analytic techniques in strategic management is increasing (Shook, 2000). More formal analytic techniques such as stochastic trees have been utilized to help increase operational efficiencies by enhancing the decision making process in medical treatment procedures (Hazen, 1992, 2000). Other analytic methodologies involving data mining techniques enable decision makers to identify patterns in clinical-, claimsand activity-based historical data, to better understand explanatory relationships in data and create models to more accurately predict future resource demand (Xiaohua, 2005). Artificial neural networks are computer algorithms that identify relationships in historical data that can be used for classification and prediction (Bishop, 1995; Swingler, 1996). Reducing the uncertainties in process resource requirements through enhanced predictive capabilities is seen to increase efficiency across industry sectors (Kudyba & Hoptroff, 2001). 103

Informatics Application Challenges for Managed Care Organizations

More effective disease management through data analysis can assist our current health care system to fill the ever-increasing gaps in health care service, management, and treatment. Disease management programs utilize both cost-based and health-based rationales for prevention and health promotion, emphasizing medical best practices and evidence-based medicine. Usually, outcomes are evaluated on the basis of clinical/therapeutic improvement or compliance, financial/cost reduction, and behavioral/emotional enrichment (Hunter & Fairfield, 1997; Leider & Krizan, 2001). Ultimately, the success of disease management programs often depends upon the ability to target an appropriate intervention to the appropriate patient, where population segmentation enhances MCOs’ ability to accomplish this. The following section of this article provides more detailed background on the application of population segmentation techniques to enhance health care efficiency and introduces three primary areas where they can be utilized by MCOs.A detailed analysis of the required analytic objective corresponding to three primary health care assessment areas (e.g. Risk Status, Health Status and Treatment Status) is then provided. The criteria for categorizing population segmentation methods according to the three primary assessment areas are then described.

approacHes to populatIon segMentatIon and predIctIVe ModelIng and a proposed classIFIcatIon systeM Research addressing the use of quantitative-based decision support systems to enhance efficiency in the health care sector is on the rise given the development of data resources and availability of sophisticated analytic methods (Walchzak, Brimhall, & Lefkowitz, 2006; Raghupathi, 2006). There is great interest today in the application of population segmentation and predictive model-

104

ing to enhance care and disease management programs by correctly focusing resources and interventions on the segment of a population who would benefit the most from interventions (Ash, Zhao, Ellis, & Sclein, 2001; Cousins, Shickle, & Bander, 2002; Ridinger & Rice, 2000). While most analytic approaches share a common goal, to enhance a care or disease management program’s ability to improve quality of care while lowering overall cost, they are optimized to address this goal in different ways. We propose a classification system to clarify the objective of the most commonly used population segmentation and predictive modeling approaches. The proposed classification system divides the methodologies based upon their useful functionality in addressing three primary operational areas for MCOs. The three proposed classifications are Risk Status, Treatment Status, and Health Status (see Figure 1). The benefits of this classification system are to enable MCOs to clarify their primary objective, thereby enabling them to select a methodology that best fits their need and ultimately achieve better disease management solutions. Disease management programs make use of population segmentation to improve the efficiency and effectiveness of their interventions. Population segmentation approaches often serve different objectives such as early identification of high-risk patients, identifying compliance with treatment protocols, or categorizing patients by severity of illness for payment adjustments or outcomes reporting. It is important to use the appropriate methodology to best meet the information requirements of a specific intervention. Depending upon the MCOs’ focus and intervention(s), some disease management programs might use one approach while other programs might use two or more approaches. For disease management purposes, predictive modeling is defined as utilizing currently available data to prospectively identify an individual’s risk of a specific outcome. Thus, one may look at a patient’s claims or survey data to determine if

Informatics Application Challenges for Managed Care Organizations

Figure 1. RISK STATUS

Future Risk of High Cost and Utilization (e.g., Predictive Modeling) Risk Status

Total Patient Patient Health

Total Health Health Treatment Status Status

Current Status of Physical and Mental Health (e.g., Current Health Assessment)

Current Type and Intensity of Health Care (e.g., Medical Best Practices)

HEALTH STATUS

TREATMENT STATUS A single patient may be looked at from each of the three perspectives.

the patient has an increased risk of future hospitalization, high cost of care, or some morbidity event (Kudyba, Hamer, & Gandy, 2005). As will be described next, each of the population segmentation applications (i.e., health status, treatment status, and risk status) has a predictive component, but not all are optimized for predicting future risk. The following section will describe the objectives required for each of the three areas of application (assessment) and clarify the appropriateness of corresponding population segmentation approaches.

tHree prIMary areas oF assessMent For Mcos (rIsK status, treatMent status and HealtH status) risk status Risk status refers to a patient’s likelihood for specific clinical outcomes (e.g., myocardial infarction), financial outcomes (e.g., significant increases in future health care costs), or utilization

outcomes (e.g., emergency department or hospital visits). It can also refer to the probability of future risk for developing certain medical conditions, such as obesity or cardiac disease (Perry, 2007). Risk status focuses on the probability of future events, making this classification of population segmentation optimally predictive in nature. Predictive modeling techniques for this area range from simple linear equations to complex neural network forecasting techniques (Grana, Preston, McDermott, & Hanchak, 1997; Kiernan, Kraemer, Winkleby, King, & Tylor, 2001). These may include seasonal adjustment methods and trend lines (Cote & Tucker, 2001), and heuristics including uncomplicated rules-based algorithms to multifaceted decision support techniques (Ferreira et al., 2001). Obviously, the mathematical or logical technique used to assess risk status is highly dependent upon the type of risk being evaluated. Regardless of the methodology, the objective of evaluating a patient’s risk status is the same, prediction of an unknown future event. A recent study conducted by Kiernan et al. (2001) compared two prediction techniques, logistic regression and signal detection, to assess

105

Informatics Application Challenges for Managed Care Organizations

individuals who are at risk for being overweight. This study was based on survey data from 1,635 White and Hispanic men and women. Body mass index (BMI) was used to define “overweight” for this population, and predictor variables used included gender, ethnicity, age, and educational level. Results from the study demonstrate that both methodologies had similar predictive accuracy and identified a similar set of risk predictor variables. Nevertheless, these methods did not classify the same individuals into population subgroups. Notably, a very high risk group (less educated, young Hispanic adults) was hidden by the logistic regression analysis but was revealed by the signal detection analysis. These results demonstrate the importance of selecting the best technique of population segmentation to identify a high-risk population. Logistic regression was utilized to predict the probability of asthma-related hospital admissions for asthma members in a large HMO (Grana et al., 1997). A predictive model was built from administrative data (e.g., medical, pharmacy, laboratory, and enrolment files) associated with over 54,000 asthma patients. Asthma-specific utilization, pharmacy data, and length of enrolment were found to be the best predictors of future asthma-related admissions. Results were evaluated by the predicted number of admissions compared with the actual number of admissions, broken down into 10 deciles. Analysis of the top three deciles resulted in a sensitivity and specificity score of 0.70 and 0.71, respectively. These results demonstrate how claims data alone can be used to accurately predict future hospitalizations for patients within a population. Risk status technologies are optimized for predicting future risk. In addition to applications in disease and care management programs, this category of population segmentation, which employs predictive modeling techniques, is also the type of segmentation technology that is currently emerging in underwriting, rating, and payment adjustment. Population segmentation that focuses on

106

Risk Status is best suited for correctly identifying and targeting the highest risk patients and thus is the most predictive of the three classifications. The next operational area which can be addressed by population segmentation methodologies involves Treatment Status for patients.

treatment status Treatment status focuses on the actual care that a patient is receiving, or in other words, the type and intensity of health care delivered. Treatment status encompasses an assessment of medical best practices or evidence-based medicine protocols. It is often evaluated as part of a physician profiling system. Obviously, not all patients are alike; there is variability in both mental and physiological responses to treatment regiments. Likewise, not all physicians are alike; there are differences in education and training as well as in personal attitudes. Nevertheless, evidence-based clinical practice guidelines can help to manage this inevitable variation. Within a disease management framework, treatment status becomes a very important population segmentation application, especially when attempting to manage and coordinate the health care delivery for thousands of chronically ill patients. Often, there is a gap between recommended standards of care and the treatment received by chronically ill patients (Muney, 2002). Only through careful evaluation of a patient’s treatment status can these treatment gaps be identified and addressed. Research on treatment status includes studies such as that conducted by O’Connor, SperlHillen, Pronk, and Murray (2001), to investigate clinical-based practice characteristics related to best practices within chronic disease care. The objective of the study was to identify those features shared by successful primary care clinics. Seven primary care practices managing patients with diabetes, hypertension, lipid disorders, or heart disease were examined. Data from each of these clinics were compiled and treated as individual

Informatics Application Challenges for Managed Care Organizations

case studies. Results from this study illustrate that focusing on treatment status of chronically ill patients can facilitate significant improvements in health outcomes within 1 to 2 years in adults with diabetes, hypertension, or lipid disorders. For example, a 20% reduction in risk for a major cardiovascular event (on a population basis) was reported. This study, as well as others (Heller & Arozullah, 2001; Solberg, Reger, & Pearson et al., 1997; Wagner, Austin, & Korff, 1996), illustrates the benefits of improving treatment status in chronically ill patients, which results in better clinical and financial outcomes. While treatment status has some predictive value and is an important component of care and disease management, it is not optimized for true predictive modeling. For example, we can use a population segmentation methodology to address treatment status to identify two patients that have elevated cholesterol and are not being appropriately treated with antilipidemic medication. Both patients are at increased risk for poor clinical outcomes, so there is some predictive component to this methodology. However, we cannot immediately infer if both patients are at co-equal risk. To add to the example, if one of the patients is 35 years of age with a family history of hyperlipidemia and cardiac disease and the other patient is 85 years of age without any current cardiac disease, then clearly the two patients are not at equal risk for cardiovascular complications in the future. Treatment Status is optimized to identify gaps in treatment, and is therefore not optimized to identify patients’ Risk Status or for predictive modeling. One final area of focus for MCOs entails analytic requirements for Health Status applications.

Health status Health status refers to the current standing of a patient’s clinical, physical, and mental health. Population segmentation techniques to address Health Status are utilized to describe a patient’s

current health relative to other patients. Health Status technologies generally serve one of two broad purposes: 1) classifying patients by health status for outcomes measurement and comparison purposes, or 2) categorizing patients into similar severity levels for assessing treatment intensity or for payment adjustment design. Other health status methods are utilized to group patients according to their likely treatment needs and resource consumption during a current episode of care (Baker, 2002). Currently, many health care organizations are using proprietary grouper software for patient/provider profiling, utilization/ clinical benchmarking, disease/case management activities, and quality improvement initiatives. There is a relationship between current health status and health outcomes. Goetzel, Anderson, Whitmer et al. (1998) conducted a study of the Health Enhancement Research Organization (HERO) database in order to estimate the impact of modifiable health status factors on health care expenditures. Overall, individuals who had poor health status had significantly higher expenditures than individuals who had a better health status in 7 of the 10 health status categories (depression, high stress, high blood glucose levels, high/low body weight, former tobacco users, current tobacco users, and high blood pressure). Moreover, individuals with poor health status in multiple categories had much higher medical expenditures than those without. For example, those with heart disease, psychosocial, and stroke profiles had 228%, 147%, and 85% higher expenditures, respectively. This study indicates that patients’ health status is associated with at least near-term increases in the likelihood of incurring future high medical costs (Anderson, Whitmer, & Goetzel, 2000; Leutzinger, Ozminkowski, Dunn et al., 2000). Despite the correlation between health status and future health outcomes, general population segmentation technologies that are optimized for Health Status are not optimized for predictive modeling. For example, two patients who are in critical condition may score similarly on a GHA.

107

Informatics Application Challenges for Managed Care Organizations

Both patients are likely to incur substantial costs in the near term and are obviously at short-term risk for poor clinical outcomes. However despite the same GHA score, we cannot immediately infer if both patients are at the same risk for longterm health outcomes or long-term expenditures. To continue the example, if one of the patients is 70 years of age with terminal cancer and the other patient is 25 years of age and was recently in a traumatic accident, then the outcomes may be dramatically different over a 12-month time frame. Health Status is optimized to correctly classify patients vis-à-vis their current health, and is therefore not optimized to identify patients’ Risk Status or for predictive modeling.

eValuatIon crIterIa oF populatIon segMentatIon MetHodologIes For Mco areas oF assesMent An important first step in evaluating population segmentation and predictive modeling techniques is to define the specific purpose and goal of the modeling initiative. To date, the evaluation process has often been complicated because of a lack of clarity in the specific goal for potential applications. In addition, there is often a lack of clarity within the MCO for what would be done with the information once obtained. For example, the care and disease management departments might envision early identification of at risk patients, the provider network department might envision provider profiling on quality metrics, and the finance department might plan to use the information to enhance underwriting or rating. While all of these are reasonable goals, they may not all be obtainable with a single technology, and it is highly unlikely that a single technology will perform optimally well for all purposes. The evaluation process should begin with a definition of the purpose and goal of the application, where multiple goals, should be prioritized.

108

This is important when balancing trade-offs in performance. Once this is achieved, the best class of population segmentation technology must be selected to meet this goal. For Risk Status applications, the goal is to apply and evaluate the predictive modeling accuracy. The focus is on accuracy metrics such as positive predictive value, true/false positive rates, true/false negative rates, sensitivity and specificity at various screening thresholds, and ROC or R-squared values. For each metric, it is important to ascertain how the metrics have been validated and if they will generalize to a particular MCO setting. For Treatment Status applications, the goal is to evaluate the comprehensiveness and accuracy of the assessments. The focus is on number of diseases covered, number of evidencebased standards covered, the quality of the algorithms used (i.e., are they disclosed or hidden in “black box” technology?), and the source of the standards (evidence-based, proprietary, etc.). The Treatment Status system’s accuracy must also be considered. This typically involves an assessment of the sensitivity and specificity of classification with respect to treatment status. Applications based upon both survey and claims data invariably produce some errors. Consequently, it is important that the assessments produce reasonable information that will be clinically acceptable, providing a sound basis for intervention decisions. For Health Status applications, the evaluation focuses on the system’s ability to correctly classify patients into comparable groups or cohorts. For example: (1) What is the accuracy of the system’s prediction in categorizing patients into a certain level of health or episode of care? (2) Can the system be generalized and can the results be used to compare to other groups or over time? and (3) Is the system well accepted by the provider community if it is to be included in payment schemes? If a single system purports to provide current health status and also predicted future health outcomes, then it is important to evaluate the accuracy metrics as described above for both the current health status prediction and the future health outcomes prediction.

Informatics Application Challenges for Managed Care Organizations

Existing population management approaches fail to fully utilize information derived from these three statuses. The difficulty in integrating multiple data sources and data types has resulted in the optimization of operational systems based on single sources. Obviously, positive outcomes have certainly been achieved using this type of single source methodology, which focuses interventions vis-à-vis highly specific, singular data sources or assessments (Huffman, 2005; Ropka, 2002). However, as described in this article, when faced with the penultimate operational challenge of population management (i.e., delivering limited resources to the right patients at the right time) it becomes increasingly important to use information derived from multiple sources in order to decrease the probability of incorrect allocation of these resources. The informatics approach described in this article is not without limitations. This approach requires extensive IT resources and IT system architectural changes. It is also very data intensive and requires the coordination and management of large amounts of data.

conclusIon In this article we have proposed a classification and evaluation method for population segmentation approaches used in care and disease management. Risk Status, Treatment Status, and Health Status assessments each provide useful information to forecast the best allocation of resources and interventions within a population of interest. For example, a disease management company might be interested in managing those patients who are at high risk for future hospitalizations. In this example the disease management company could use information from population segmentation approaches in order to focus resources on

the segment of the population with the highest probability of future hospitalizations. However, the analysis of each assessment application is optimized differently: Risk Status is optimized for the prediction of future events, Treatment Status is optimized for the identification of gaps in medical care, and Health Status is optimized for the classification of patients’ current state of physical and mental health. Clearly, in order to capitalize on the strengths of population segmentation methodologies, health plans need to understand the primary functionality of a given population segmentation methodology in conjunction with the required objective of the particular assessment application and, subsequently, select the most appropriate modeling approach that best fits their purposes. As MCOs increasingly rely upon care and disease management programs to fill gaps in health care service and delivery, correctly identifying and targeting patients for program interventions has risen in importance. Though risk status is clearly the application that is optimally predictive in nature, it cannot alone be used to evaluate, assess, and allocate interventions to a given population. Population segmentation methodologies addressing current treatment status and health status, as well as anticipated risk status, are the most effective and logical way to move forward with an overall intervention strategy. Meanwhile, it is important to differentiate between these three approaches, because they have been optimized for different goals. Clarity in purpose and focused evaluation against that purpose will provide MCOs with the best techniques. Consequently, MCO disease management programs that truly integrate/utilize all three approaches will best succeed in accomplishing the main goal of focusing the right resources on the right patients while increasing quality of care and decreasing overall medical expenditures.

109

Informatics Application Challenges for Managed Care Organizations

reFerences Anderson, D.R., Whitmer, R.W., & Goetzel, R.Z., et al. (2000). Health enhancement research organization (HERO) research committee. The relationship between modifiable health risks and group-level health care expenditures. American Journal of Health Promotion, 15, 45-52. Ash, A.S., Zhao, Y., Ellis, R.P., & Schlein, K.M. (2001). Finding future high-cost cases: Comparing prior cost versus diagnosis-based methods. Health Services Research, 36, 194-206. Baker, G.B. (2002). Integrating technology and disease management—the challenges. Healthplan Magazine, 43, 60-66. Bishop, C. (1995). Neural networks for pattern recognition. Oxford University Press. Cousins, M.S., Shickle, L.M., & Bander, J.A. (2002). An introduction to predictive modeling for disease management risk stratification. Disease Management, 5, 157-167. Cote, M.J., & Tucker, S.L. (2001). Four methodologies to improve health care demand forecasting. Healthcare Financial Management, 5, 54-58. Ferreira, F.L., Peres, Bota, D., Bross, A., Melot, C., & Vincent, J.L. (2001). Serial evaluation of the SOFA score to predict outcome in critically ill patients. Journal of American Medical Association, 286, 1754-1758. Foughty, T. (1999). Accomplishing health improvement through diabetes disease management. Drug Benefit Trends, 12, 43-45. Goetzel, R.Z., Anderson, D.R., Whitmer, R.W., et al. (1998). The relationship between modifiable health risks and health care expenditures: An analysis of the multi-employer HERO health risk and cost database. Journal of Occupational and Environmental Medicine, 40, 843-854.

110

Grana, J., Preston, S., McDermott, P.D., & Hanchak, N.A. (1997). The use of administrative data to risk-stratify asthmatic patients. American Journal of Medical Quality, 12, 113-119. Hazen, G.B. (1992). Stochastic trees: A new technique for temporal medical decision modeling. Medical Decision Making, 13, 227-236. Hazen, G. (2000). Preference factoring for stochastic trees. Management Science, 46(3), 389-403. Heller, C., & Arozullah, A. (2001). Implementing change: It’s as hard as it looks. Disease Management and Healthcare Outcomes, 9, 551-563. Heskett, L. (1983). Shouldice Hospital Ltd, Rev. 1/88. Boston, MA: Harvard Business School. Huffman, M. (2005). Implementing outcomebased homecare: A workbook of OBQI, care pathways and disease management. Jones and Bartlett. Hunter, D.J., & Fairfield, G. (1997). Managed care: Disease management. British Medical Journal, 315, 50-53. Institute of Medicine. (2001). Crossing the quality chasm: A new health system for the 21st century. Washington, DC: National Academy Press, Committee on Quality of Health Care in America. Kiernan, M., Kraemer, H.C., Winkleby, M.A., King, A.C., & Taylor, C.B. (2001). Do logistic regression and signal detection identify different subgroups at risk? Implications for the design of tailored interventions. Psychological Methods, 6, 35-48. Kudyba. S., Hamer, B., & Gandy, W. (2005, December). Enhancing efficiency in the health care industry. Communications of the ACM, 28(12), 107-110. Kudyba, S., & Hoptroff, R. (2001). Data mining and business intelligence: A guide to productivity. Hershey, PA: Idea Group.

Informatics Application Challenges for Managed Care Organizations

Landro, L. (2002, January 4). Technology programs make disease management easier. The Wall Street Journal, Health Journal.

Ridinger, M.H., & Rice, J.J. (2000). Predictive modeling points way to future risk status. Health Management Technology, 21, 10-12.

Leider, H.L., & Krizan, K. (2001). Disease management—a great concept, but can you implement it? Disease Management, 4, 111-119.

Ropka, M.E. (2002). Symptom status and functional status outcomes: Humanistic outcomes in obesity disease management. Obesity Research, 10, 42S-49S.

Leutzinger, J.A., Ozminkowski, R.J., Dunn, R.L., et al. (2000). Projecting future medical care costs using four scenarios of lifestyle risk rates. American Journal of Health Promotion, 15, 35-44. McLaughlin, C., Yang, S., &Van Dierdonck, R. (1995). Professional service organizations and focus. Management Science, 14(7), 1185-1193. Muney, A.M. (2002). Evidence-based medicine needs to be promoted more vigorously. Managed Care, 11, 10-12. O’Connor, P.J., Sperl-Hillen, J.M., Pronk, N.P., & Murray, T. (2001). Primary care clinic-based chronic disease care. Features of successful programs. Disease Management and HealthCare Outcomes, 9, 691-698. Perry, T. (2007), December). Identification and prediction of chronic conditions for health plan members using data mining techniques: In data mining methods and applications. Taylor Francis. Raghupathi, W. (2006). Designing clinical decision support systems in health care: A systemic approach. International Journal of Healthcare Information Systems and Informatics, 2(1).

Smith-Daniels, V., Schweikhart, S., & SmithDaniels, D. (1988). Capacity management in health care services: Review and future research directions. Decision Sciences, 19, 889-919. Solberg, L.I., Reger, L.A., & Pearson, T.L., et al. (1997). Using continuous quality improvement to improve diabetes care in populations: The IDEAL model: Improving care for diabetics through empowerment, active collaborations and leadership. Joint Commission Journal on Quality Improvement, 23, 581-592. Swingler, K. (1996). Applying neural networks: A practical guide. New York: Academic Press. Wagner, E.H., Austin, B.T., & Korff, M.V. (1996). Improving outcomes in chronic illness. Managed Care, 4, 12-25. Walczak, S., Brimhall, B., & Lefkowitz, J. (2006). Nonparametric decision support systems in medical diagnosis: Modeling pulmonary embolism. International Journal of Healthcare Information Systems and Informatics, (1), 2. Xiaohaua, H. (2005). Temporal rule induction for clinical outcome analysis. International Journal of Business Intelligence and Data Mining, 1(1), 122-136.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 2, edited by J. Tan, pp. 21-31, copyright 2008 by IGI Publishing (an imprint of IGI Global).

111

112

Chapter 9

Scrutinizing the Rule: Privacy Realization in HIPAA S. Al-Fedaghi Kuwait University, Kuwait

aBstract Privacy policies, laws, and guidelines have been cultivated based on overly verbose specifications. This article claims that privacy regulations lend themselves to a firmer language based on a model of flow of personal identifiable information. The model specifies a limited number of situations and acts on personal identifiable information. As an application of the model, the model is applied to portions of the Privacy Rule of Health Insurance Portability and Accountability Act (HIPAA).

IntroductIon The notion of privacy is becoming an important feature of modern society. In this context, deciding how to regulate the processing of personal identifiable information (PII) is a vital issue. Responding to the public’s awareness of the importance of protecting privacy of personal identifiable information, guidelines have evolved and converged around a set of basic privacy principles (e.g., OECD, 1980). How successful are these privacy principles? According to the OECD (1980) report, The choice of core principles and their appropriate level of detail presents difficulties… In particular, it is difficult to draw a clear dividing line between the level of basic principles or objectives and

lower level “machinery” questions which should be left to domestic implementation. (I. GENERAL BACKGROUND, 19, e) According to Bennett (2001), privacy principles enfold different interpretations, There are disputes for example: about … the distinction between collection, use and disclosure of information, and whether indeed these distinctions make sense and should not be subsumed under the overarching concept of “processing” … How these and other statutory issues are dealt with will, of course, have profound implications for the implementation of privacy protection standards within any one jurisdiction.(p. 12)

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Scrutinizing the Rule

Mechanisms to protect the privacy of personal identifiable information include legal measures, policies and privacy-enhancing technologies. Legislations such as the Health Insurance Portability and Accountability Act (HIPAA) and systems such as P3P are not sufficient to safeguard privacy because “they do not address how personal data is actually handled after it is collected” (He & Antón, 2003, p. 1). Also, according to FischerHübner and Ott (1998, p. 1) “privacy cannot be efficiently implemented solely by legislative means. Data protection commissioners are therefore demanding that legal privacy requirements should be technically enforced and should be design criteria for information systems.” Among types of personal identifiable information, health information is ranked as being the most sensitive, at the same level as financial information (GPC Alberta, 2003). In the health information area, “Survey research found that the public … was deeply worried about how their personal medical information was being accessed and used in other sectors, for secondary purposes such as insurance, employment, licensing, research, law enforcement, public health, and media activities” (Westin & Gelder, 2005, p. 4). The Health Insurance Portability and Accountability Act (HIPAA) of 1996 is the most significant health care legislation in U.S. history. The U.S. Department of Health and Human Services (HHS) issued the Privacy Rule (45 CFR Parts 160 and 164) to implement the requirement of HIPAA. According to U.S. Department of Health & Human Services (2003), “The Standards for Privacy of Individually Identifiable Health Information (“Privacy Rule”) establishes, for the first time, a set of national standards for the protection of certain health information” (HSS, 2003). While privacy laws have an important role, other approaches are valuable. The “Privacy by Design” approach that incorporates the Fair Information Practices standards into information

systems, tries to go beyond the HIPAA Privacy Rule (Westin & Gelder, 2005, p. 6). Nevertheless, it is also important to develop clearly defined privacy terms. Recent developments in the areas of privacy and information assurance have placed the elements of information privacy on firmer theoretical ground. This includes a model for personal information flow (Al-Fedaghi, 2006) that systematically categorizes subprocesses involved in personal information processing. This article illustrates a sample of the benefits of such an approach through inspecting portions of the Privacy Rule of HIPAA. Our purpose is not to take a position on any of the privacy principles or subprinciples, and neither is it to highlight conflicts and contradictions, but rather it is to provide precise specifications of privacy notions in order to apply them to different privacy standards, legislations, or codes of practice. Understanding the constitutive elements of privacy principles will help to determine any disparities and consistencies/inconsistencies when incorporating them in different laws and in their interaction with other requirements (e.g., enforcement mechanisms).

personal Identifiable Information This section focuses on personal identifiable information (PII) as our main object of study, and its flow model (Al-Fedaghi, 2005a, 2006). It is typically claimed that what makes the data “private” or “personal” is either specific legislation, for example, a company must not disclose information about their employees; or individual agreements, for example, a customer has agreed to an electronic retailer’s privacy policy. However, this line of thought blurs the difference between personal identifiable information and other “private” or “personal” information. Personal identifiable information has an “objective” definition in the sense that it is independent of such authorities as legislation or agreement.

113

Scrutinizing the Rule

Definition of Personal Identifiable Information In the information sphere, information is classified as personal identifiable information (PII) and nonidentifiable information (NII). Personal identifiable information is information about singly-identifiable persons, called its proprietors. PII is information that has referents that are natural persons. There are two types of PII: 1.

Atomic personal information where the information refers to a single proprietor, for example, John is 25 years old. “Referent,” here, implies an identifiable (natural) person. 2. Compound personal information where the information refers to more than one proprietor, for example, Mary donated her kidney to Alice. Any compound PII is privacy-reducible to a set of atomic PII (Al-Fedaghi, 2005a).

PII Flow Model The PII Flow Model divides functionality into stages that include informational entities and processes, as shown in Figure 1. Each stage may have several specific units that denote autonomous functions such as storage/retrieval and utilization of collected information. New PII is created by proprietors, nonproprietors (e.g., medical diagnostics by physicians), or is deduced by someone (e.g., data mining that generates new information from existing information). The created information is used either in some type of utilization (e.g., decision making), stored, or it is immediately disclosed. The processing stage involves acting (e.g., storing, anonymization, data mining, summarizing, translating) on PII for whatever purpose it is collected. The disclosure stage involves releasing PII to insiders or outsiders. The “disposal” or disappearance of PII can happen anywhere in the model, such as the transformation of PII to an

114

anonymous form in the processing stage. Double arrows in Figure 1 denote two acts on PII (will be defined) that for simplicity’s sake can be considered as one type of act. For example, the double arrow to Storage indicates storing or retrieving data. The double arrow between Collecting and Disclosing reflects the fact that every disclosure act of information is accompanied by a collecting of information by another agent. For example, if someone discloses his/her PII to an insurance company, then simultaneously the company is collecting this PII. The double arrow between Processing and Mining is necessary to indicate that mining, as a type of processing, produces two types of information: (1) processed/ mined information that is not new (e.g., implied information), and (2) processed/mined information that is new (e.g., classifying a customer as a risk, based on statistical analysis of all customers). Example: Consider the situation where we have one proprietor and two PII agents (e.g., companies, agencies, other individuals). Suppose that the roles of these three actors are defined as follows: Proprietor: Creates, stores, and discloses PII to agent 1. Agent 1: Collects, stores, utilizes, and discloses PII to agent 2. Agent 2: Collects and processes PII through a mining technique that creates new PII that is stored and utilized in some applications (e.g., decision making). The PII flow model for this simple environment can be drawn based on Figure 1, as shown in Figure 2. For each actor in this scenario we make a copy of the PII flow model. However, because the proprietor does not collect or process PII, these stages are not shown in his/her region. The creation and processing stages are not shown for agent 1 because it does not create or process PII. Similarly, the disclosure stage is not shown in the region of agent 2. Let t be a piece of PII of the proprietor. It originates in the Creating box in

Scrutinizing the Rule

Figure 1. PII flow model. Creation stage

Non-proprietor

Proprietor

Creating PII

Storage Collection stage

Utilize

Collecting PII 5

6

Storage

Processing stage

3

Utilize

Processing PII

Storage

Disclosure stage

Mining

Utilize

Disclosing PII

Figure 2. PII flow model for a proprietor and two agents. Proprietor’s Region

1st Agent’s Region

Creating

Storage

Utilize Storage Collecting

2nd Agent’s Region Store

Utilize Creating

Collecting

Mining Disclosing

Disclosing

Processing

115

Scrutinizing the Rule

the proprietor’s region. It is stored in Store, and moves through the Disclosing box to the Collecting box of the agent 1’s region. It is stored and utilized there, and moves through the Disclosing box to the Collection phase of agent 2. There, it moves to the Processing and to the Mining boxes where it generates new PII in the Creation phase for Storage and later Utilization. We can add details to any stage as the situation requires. For example, agent 2 may add Storage to keep a copy of the original PII or a Disclosure stage can be added if agent 2 discloses the resultant new PII to a third agent.

tHe proprIetor-otHers arcHItecture Using the PII flow model, we can built a system that involves a proprietor on one side and others (other persons, agencies, companies, etc.) who perform different types of activities in the PII transformations between different stages. We will refer to any of these as PII agents. PII agents may include anyone who participates in activities over PII. The proprietor is not accepted as agent with respect to his/her own PII. The EU Privacy Directive (1995) and HIPAA manage the type of system that organizes (1) The relationship between the proprietor and agents that utilize his/her PII, and (2) The relationship among the agents. It is a “binary” system that involves the proprietor on one side and all agents on the other side. As a result, we need two types of PII flow regions: one for proprietors and one for agents. The proprietor’s region includes activities on his/ her PII and the others’ region includes activities on the proprietor’s PII, as shown in Figure 3. We assume that there is no interest in how proprietors collect and process their own PII. We distinguish here among types of acts on PII (labeled A through Q) as shown in Figure 3 and described in Table 1.

116

Examples: Act A: A person discloses PII to a hospital (i.e., a colleting agent). Act B: A hospital discloses a student’s PII to an insurance company. Act C: PII is produced by a mining program (e.g., John is a high-risk customer) of a bank is disclosed to crediting company. Act D: Processed PII (changing the original data to another form through such operations as modification, translation, summarization, generalization) is released from a hospital to an insurance company. Act E: A hospital stores PII in its automated or manual system, accessing stored data, erasure of stored data. Act F: A hospital utilizes PII to communicate with its patients (e.g., e-mails). Act G: A hospital stores processed PII (e.g., mined PI) in its system, accessing stored data. Act H: An agent utilizes processed data for research. Act I: A collecting agent sends collected PII to a processing agent in the same enterprise. Act J: An agent mines PII to extract implied PII (e.g., the information Z is the grandfather of Y is taken from Z is the father of W and W is the father of Y). Act K: An agent Stores PII produced by a mining program. Act L: An agent makes decisions based on PII that is produced from a mining program. Act M: An agent produces new PII using a mining program (e.g., an analysis of customers produces the information that John is a high-risk person). Act N: A newspaper writer publishes PII about the proprietor. Act O: An agent collects PI from another agent. Act P: An agent creates PI (gossip) and processes it. Act Q: Informing a proprietor of results of medical tests.

Scrutinizing the Rule

Figure 3. Architecture of proprietor/agent PII flow. Proprietor’s Region

Agent’s Region N

Non-proprietor Utilize

L Utilize

Collecting

O

E

B

M

P I

Store

J

Store

Q

The advantage of using this type of categorization is that we can specify different types of codes, requirements, or restrictions for each type of act on PII. Example: Consider the act of creating PII by others. This type includes M (creating PII by a mining agent), and N (creating PII by nonproprietor). These two creating acts are different from the act of a proprietor creating PII (about him/ herself) and disclosing it (act A). For example, the EU Directive requirement that “the data subject must be in a position to learn of the existence of a processing operation” (EU Directive, 1995, Recitals 38), is applied to M and N, but not to A. Additionally, acts M and N are two different types of creating PII. Act M creates information by automatic means as in the case of Jane is a high risk person based on analysis of PII of all customers of a company. Act N involves nonautomatic means such as an informer claims that Jane is a terrorist based on his/her surveillance. These two types of creating PII may require different treatment in different contexts. For example, a law may forbid creating PII by automated means

Mining

Processing G

Collecting

K

F

A

Disclosing

Store

Creating

D

H

Utilize C

Disclosing

while it is difficult to apply this law to PII created by people about other people. Example: Consider the act of disclosing PII. It includes four types of disclosures. This type includes A (disclosing PII by a proprietor), B (disclosing PII by a collecting agent), C (disclosing PII by a creating agent), and D (disclosing PII by a processing agent). It is clear that the requirements and restrictions of disclosing PII will differ with respect to who is disclosing the information: a proprietor disclosing his/her PII or others disclosing their PII to one another. For example, The OECD’s restriction directed at such practices as “deceiving data subjects to make them supply information” (OECD, 1980, I. GENERAL BACKGROUND: The Problems (52)) makes sense only in act A. Furthermore, the difference extends to the type of agent. We apply different considerations to an agent who is creating new or implied PII using a mining technique compared to a simple collecting agent. The point is that we have a well-defined set of acts on PII that can be used in different applications such as writing codes and guidelines.

117

Scrutinizing the Rule

Table 1.Types of acts on personal identifiable information Description of the Act

Comments

A

Disclosing PII by a proprietor

Act A also represents collecting PII (by a collecting agent).

B

Disclosing PII by a collecting agent

B implies O (collecting PI by another collecting agent)

Disclosing PII by a creating agent

In Figure 3, B implies O, that is, the disclosed PII flows from a collecting agent to another collecting agent.

C D

Disclosing PII by a processing agent

In Figure 3, D implies B, that is, the disclosed PII flows from a processing agent to a collecting agent.

E

Storing PII by a collecting agent

E (double arrow in figure 3) includes retrieval of PII.

F

Utilizing PII by a collecting agent

“Utilize” indicates noninformational operations.

G

Storing PII by a processing agent

We can separate the storing and retrieval acts as two independent acts in the model.

H

Utilizing PII by a processing agent

Utilizing processed PII may be different from utilization of other types of PII.

I

Processing by an agent

PII flows from the collecting stage to a processing stage.

J

Mining PII by a mining agent

Mining is a type of processing. This type of mining produces implied PII but not new PII.

K

Storing PII by a creating agent

L

Utilizing PII by a creating agent

M

Creating PII by a mining agent

Automatic creation of PII. Mining is a type of processing.

N

Creating PII by a nonproprietor

Non-automatic creation of PII (e.g., gossip).

O

Collecting PII from nonproprietor

O occurs simultaneously with B: If an agent discloses PII then there is an agent that collects that PII.

P

Processing of created PI

Nonproprietor may create PII and process it immediately without storing or acting on it.

Q

Disclosure to proprietor

E.g., Informing a person of results of medical tests.

Each of the PII acts can be supplemented with various purposes (Utilizations) as shown. For example, purposes for Disclosing PII by a proprietor can be Internet purchase, government form, and security. While the number of acts is limited, the number of purposes (Utilizations) is unlimited. Also, purposes that are applied to a certain act may not be suitable for another act. For example, a nonproprietor who creates PII for an Internet purchase is an illegal act except with the appropriate authorization.

HIpaa The Health Insurance Portability and Accountability Act (HIPAA) of 1996 establishes uniform

118

national standards for handling health care information. The 169-page law is supplemented with about 1000 pages of related regulations and directions. The U.S. Department of Health and Human Services (HHS, 2003) issued the Privacy Rule (45 CFR Parts 160 and 164) to implement the requirement of HIPAA. The rule assures a right of patient access to/track of his/her health information and mandates notification of how it will be used and disclosed. It applies only to a subset of health providers. We will apply the PII FLOW MODEL architecture to some of the notions of the Privacy Rule in order to illustrate how our model can enhance the conceptual framework under which these notions can be interpreted.

Scrutinizing the Rule

definitions in the rule The Privacy Rule develops a set of definitions starting with health information, moving to Individually identifiable health information and lastly what is called Protected Health information. According to HIPAA (1996): Health information means any information, whether oral or recorded in any form or medium, that: 1.

2.

Is created or received by a health care provider, health plan, public health authority, employer, life insurer, school or university, or health care clearinghouse; and Relates to the past, present, or future physical or mental health or condition of an individual; the provision of health care to an individual; or the past, present, or future payment for the provision of health care to an individual (PART C--ADMINISTRATIVE SIMPLIFICATION, section 1171(4) of the Act).

The first question is, why are health care provider, health plan, public health authority, employer, life insurer, school or university, or health care clearinghouse mentioned in the definition of health information? If we want to restrict the handling of a particular type of information to certain players then such a restriction is not part of the definition of that information. Imagine that we define scientific information as information created or received by schools and universities. Building a foundation for the Privacy Rule requires giving, first, a definition of health information, and then later declaring that only information utilized by some users is considered in the context of HIPAA. This is a typical approach, for one reason, that if the set of users changed in the future then we do not have to change the definition of health information. The definition is also constructed such that if you measure your own temperature then the result

is not health information because it is not created or collected by the named agents. This approach confuses the ordinary meaning of health information with the meaning of health information in HIPAA. Common language is important when we think of the wide applications of HIPAA in the life of every citizen. The definition also complicates health information by bringing in a certain application area payment. Do we change the definition of health information if we have completely free health services? A better approach is to define health information, and then include or exclude such applications. So what is health information? In the PII flow model, it is simply defined in the context of personal identifiable information, as will be described next. According to the Health Insurance Portability and Accountability Act (HIPAA, 1996, PART C -ADMINISTRATIVE SIMPLIFICATION DEFINITIONS, SEC. 1171): Individually identifiable health information is information that is a subset of health information, including demographic information collected from an individual, and: 1.

Is created or received by a health care provider, health plan, employer, or health care clearinghouse; and 2. Relates to the past, present, or future physical or mental health or condition of an individual; the provision of health care to an individual; or the past, present, or future payment for the provision of health care to an individual; and i. That identifies the individual; or ii. With respect to which there is a reasonable basis to believe the information can be used to identify the individual. The definition involves, in addition to previous criticisms, ambiguity. What about John has been

119

Scrutinizing the Rule

infected by Mary’s disease, then is it John’s identifiable health information or Mary’s identifiable health information? Consider the information This animal has Parkinson’s disease, then is it Parkinson identifiable health information? It certainly identifies the known individual: Parkinson. In the PII flow model, John has been infected by Mary’s disease is compound PII that can be reduced to atomic PII. Also, the definition of personal identifiable information is based on referents–a well-known logical term—of type individual (i.e., a natural person). Thus, This animal has Parkinson’s disease is not PII because it does not refer to an individual. Based on the definition of personal identifiable information, we define health personal information (HPI) as a special kind of personal identifiable information related to the health of a referent of type person. There are many kinds of HPI: • • • •

Strict HPI: (physical and emotional descriptions): John’s blood type is A. Financial HPI: John’s kidney operation is paid for by insurance XYRE123. Location HPI: John’s kidney operation is performed in General Hospital. Compound HPI: John’s physician is Edward.

•

Operational HPI: John is injected with Penicillin.

Notice that there is no mention about whether the HPI is for payment or freely provided, or who “creates or receives” it. The definitions of PII and HPI can be complemented by restricting HPI to HPI handled by a health care provider, health plan, employer, or health care clearinghouse. For example, the so-called protected health (personal) information (PHI) is defined in HIPAA (Definitions, 160.103) as: •

Protected health information means individually identifiable health information: 1. Except … 2. Protected health information excludes individually identifiable health information in …

In our framework, it is merely a restricted subset of HPI. A hierarchy of definitions (Figure 4) simplifies the conceptual application of the Privacy Rule: This conceptual framework simplifies understanding of the application of the rule. For example,

Figure 4. Relationships among types of information

World Reference (uniquely identification) to a person

Personal Identifiable Information (PII) About health Health PII

R estri cti on s Protected Health PII

120

Scrutinizing the Rule

consider a researcher who obtains the information from student records maintained by a school. From Figure 4, it is clear that such an act is not protected health information because it is simply restricted to specified agents. Notice that in the PII flow model an agent may have two roles: as a PII system and as a HPI system. For example, a hospital may have a PII database system as an employer and a HPI database system as a HIPAA covered entity. Each system has its own PII flow model with its own privacy rules such that the other system is treated as another collecting agent.

use and disclosure Going beyond the basic definitions of health related information, we examine now a central notion in the Privacy Rule: use and/or disclosure. According to HHS (2003), The Privacy Rule standards address the use and disclosure of individuals’ health information—called “protected health information” by organizations subject to the Privacy Rule — called “covered entities,” as well as standards for individuals’ privacy rights to understand and control how their health information is used. (p. 1) Apparently, use and/or disclosure is a very important term in the Privacy Rule. In the SUMMARY OF THE HIPAA PRIVACY RULE, U.S. Department of Health & Human Services (HSS, 2003), “use or disclosure” occurs 32 times while “use(s) and disclosure(s)” occurs 35 times. The following are samples of how the rule utilizes use and/or disclosure. When “defining” Business Associate, the rule describes a business associate as “a person or organization … that performs certain functions or activities … or provides certain services to, a covered entity that involve the use or disclosure of individually identifiable health information” (HHS, 2003, p. 3). Also, HHS (2003, p. 6) states that, “The Privacy Rule does not require that

every risk of an incidental use or disclosure of protected health information be eliminated.” We can conclude from these samples that the Privacy Rule means by “use and/or disclosure” all types of operations or processing on PII. The rule’s “use and/or disclosure” refers in the PII flow model to collecting, storing, utilizing, processing, mining, creating, and disclosing HPI regardless of their types. Examining “use,” by itself, we find that HIPAA defines it as follows: Use means, with respect to individually identifiable health information, the sharing, employment, application, utilization, examination, or analysis of such information within an entity that maintains such information (Emphasis added). (Sec. 164.501) The terms, “sharing, employment, application, utilization, examination, or analysis” are overloaded terms. The PII flow model distinguishes between two types of using PII: 1.

2.

The informational acts on PII that are hard-wired in the PII flow model such as collecting, storing, processing, and so forth. In these acts information is what is acted on. In the PII flow model, Utilize denotes noninformational acts where what is acted on is not PII (e.g., a doctor treats a patient). It is the “use” to which health PII is put (treatment, payment).

HIPAA’s “use” does not distinguish types of acts. For example, sharing is an act on PI. It implies disclosure/collecting of PII among agents. If a doctor requested the medical file then this situation represents, the (hospital) system discloses the file to an internal agent (doctor). If the first doctor who receives the file shares the information with a new doctor then this is also disclosure (and collection) of PII between two internal agents.

121

Scrutinizing the Rule

Sharing in these cases involves disclosing and collecting information. However, “Employment” (see HIPAA definition of “use” mentioned previously) is not necessarily this type of act. Employment (partially) means utilizing information in the noninformational act such as treatment of patient. It is the “use” to which health PII is put. Treatment is represented in the PII flow model in the utilize box, where a doctor as a collecting agent utilizes the medical record disclosed to him/her by the system to treat (noninformational act) a patient. Thus, the rule mixes acts on information and the “use” to which health PII is put. Acts on information are limited as described in the PII flow model; however, the “use” to which information is put is infinite. Regulations for acts on PII ought to be very specific for acts on information, while regulations for the “use” to which information is put are usually open-ended. Other terms in the definition of “use” (application, utilization, examination, or analysis) can also be clarified in similar fashion. The distinction is important for defining several notions. For example, “consent” for informational acts is different from consent for noninformational acts. Consent to disclose information does not imply consent for the use to which health PII is put as described next.

consent According to Galpottage and Norris (2005, p. 6), after reviewing several definitions, “The term ‘patient consent’ implies that the patient is willing to share personal health information and, where appropriate, to submit to a course of medical treatment.” Notice how this definition assumes a single consent for two different types of acts: informational acts (sharing health PII) and noninformational acts (a course of medical treatment). Also, according to Galpottage and Norris (2005, p. 6) “In the context of informed

122

consent, deliberately omitting to tell the patient the consequences of their consent or the use to be made of their information would constitute an act of malfeasance.” But, “the consequences of their consent” and “the use to be made of their information” are two completely different actions. Consent to sharing health PII does not imply consent for a course of medical treatment. Informed consent in the sharing of health PII involves further acts on PII such as disclosure and processing (e.g., for research), while Informed consent in the context of a course of medical treatment involves medical complications, side effects, and so forth. In the Privacy Rule, consent is not required for purposes of the imprecisely defined operations of “treatment, payment and health operations.” The rule relates consent (acknowledgement) to use information in these operations not to the actual treatment, payment or health care operations. It states that: “a covered entity is permitted, but not required, to use and disclose protected health information, without an individual’s authorization, for … Treatment, Payment, and Health Care Operations.” Treatment is defined in the Privacy Rule at 45 CFR 164.501 as: “Treatment” generally means the provision, coordination, or management of health care and related services among health care providers or by a health care provider with a third party, consultation between health care providers regarding a patient, or the referral of a patient from one health care provider to another. Use and/or disclosure for treatment is an informational act (act on information). However, we observe that the information can be disclosed internally (covered entity’s own treatment) or disclosed externally to altogether different covered entity. The ideal situation for disclosing PII to another covered entity is shown in Figure 5. The consulted covered entity collects (act A) the disclosed PII and uses it as a consulted entity (F) to create medical opinion (N) that is disclosed

Scrutinizing the Rule

back to the original entity (C followed by Q). The PII is not acted on in any other way. There are several subsets of acts that may be specified for entities receiving the information. If the new entity is a specialist who actually treats the individual then the phrase without an individual’s authorization seems meaningless because the patient has already given consent for treatment (hence PII). The PII flow model presents a precise description of “minimally necessary” acts for disclosure for this type of treatment. This analysis can be used in rewriting the Privacy Rule or as a supplementary interpretation to the rule. In contrast, the rule gives a blank check to disclose protected health information, without an individual’s authorization, for external covered entities based on the ambiguous term treatment. The second use that a covered entity is permitted to disclose protected health information without an individual’s authorization is for payment. Payment is defined in the Privacy Rule at 45 CFR 164.501: “Payment” encompasses the various activities of health care providers to obtain payment or be reimbursed for their services and of a health plan to obtain premiums, to fulfill their coverage responsibilities and provide benefits under the plan, and to obtain or provide reimbursement for the provision of health care. Again we see here a blank check for any internal or external activities. There is no difference between disclosing PII to an employee of a covered entity and to another covered entity across the nation. Consider the case of misusing a social security number by an employee of the covered entity in contrast to misusing it while in the possession of consumer reporting agencies. Notice that other portions of the Rule may help in interpreting a certain part of it, however, the claim we make is the wording of each part can be made clearer using our approach.

Even if it is decided that an individual’s authorization is not required in either case, an acknowledgement of the two cases ought to be specified explicitly for several purposes including for the sake of individual’s awareness and the rule’s implementer’s alertness. Health Care Operations can be analyzed in a similar way to the discussion of “treatment” and “payment.”

disclosure The rule states examples of “disclosures that would require an individual’s authorization include disclosures to a life insurer for coverage purposes, disclosures to an employer of the results of a preemployment physical or lab test, or disclosures to a pharmaceutical firm for their own marketing purposes” (HHS, 2003, Authorized Uses and Disclosures section). However, in light of our categorizations of disclosures, “disclosure” may mean the disclosure of the original PII collected from the proprietor, the disclosure of collected PII from other collecting agent, the disclosure of processed PII, or ever the disclosure of PII created by mining program or input by another person (e.g., physician, health worker). Example: According to the rule, “A covered entity may use or disclose, without an individual’s authorization, the psychotherapy notes, for ...” (HHS, 2003, Authorized Uses and Disclosures section). Psychotherapy notes are defined: Psychotherapy notes” means notes recorded (in any medium) by a health care provider who is a mental health professional documenting or analyzing … that are separated from the rest of the of the individual’s medical record. Psychotherapy notes excludes medication prescription and monitoring, counseling session start and stop times, the modalities and frequencies of treatment furnished, results of clinical tests, and any summary of the following items: diagnosis, functional

123

Scrutinizing the Rule

status, the treatment plan, symptoms, prognosis, and progress to date. Figure 6 shows simplified versions of different regions in this situation. The psychotherapist collects PII from four sources shown in the figure. 1. 2. 3. 4.

The original medical record Created PII by the psychotherapist during the session New PII revealed by the patient during the session Created PII related to the session (e.g., start and stop times, e.g., John’s session is at 4 pm, John [the patient] arrived late for the session).

These are clear categorizations of information that can be supplemented by the type of descriptions given by the rule (e.g., summary of this type or that type). Suppose the patient made threats during the session to kill a person named Alice. How does the rule handle the issue of disclosure of this PII? In the PII flow model, (3) above can be divided into two types of PII: (a) atomic PII, and (b) compound PII. Because the threat involves Alice, then she has the right as a proprietor to know this information (Al-Fedaghi, 2005b). We conclude that the PII flow model provides a framework for deciding on disclosure issues such as the psychotherapist notes based on categorization of types of PII instead of the rule’s general description that lack preciseness. The PII flow model furnishes a foundation for categorizing PII and acts on PII. This involves distinguishing different types of PII, different types of regions of agents, and different types of acting on PII. This discretion of types of PII, of acts on PII, of regions of agents clearly enhances the important requirement of releasing “the minimum amount reasonably necessary to achieve the purpose of the disclosure” (HSS, 2003, Limiting Uses and Disclosures to the Minimum Necessary section).

124

conclusIon This article has introduced a systematic description of the “processing” of personal identifiable information. The description is based on a flow model that limits handling personal identifiable information to creating, collecting, processing and disclosing information. It also specifies a limited number of acts on personal information. This can provide a more methodological scheme for privacy policies, statues, and guidelines specifications. To substantiate this claim we applied the model to some portions of the Privacy Rule of the Health Insurance Portability and Accountability Act (HIPAA). Our analysis maps the notion of “use,” “disclosure,” and “consent” in the rule to its corresponding acts in the PII flow model. This analysis can be used in rewriting the rule or as a supplementary interpretation to the rule. Another aim of this article is to show that privacy regulations lend themselves to a firmer language based on a PII flow model. While this article uses the model to scrutinize the Privacy Rule of HIPAA, it is also applicable to the statutorybased privacy Directive of the European Union and to codes and guidelines (Al-Fedaghi, 2007).

reFerences Al-Fedaghi, S. (2005a). How to calculate the information privacy. In Proceedings of the Third Annual Conference on Privacy, Security and Trust: How to Calculate Information Privacy, St. Andrews, New Brunswick, Canada. Retrieved January 29, 2008, from http://www.lib.unb.ca/ Texts/PST/2005/pdf/fedaghi.pdf Al-Fedaghi, S. (2005b). Privacy as a base for confidentiality. In Proceedings of the Fourth Workshop on the Economics of Information Security, Harvard University, Cambridge, MA. Retrieved January 29, 2008, from http://infosecon. net/workshop/schedule.php

Scrutinizing the Rule

Al-Fedaghi, S. (2006, June 20-23). Aspects of personal information theory. In Proceedings of the Seventh Annual IEEE Information Assurance Workshop (IEEE-IAW): Aspects of Personal Information Theory. West Point, NY: United States Military Academy. Al-Fedaghi, S. (2007, July 12-14). When reinventing principles is necessary. In Proceedings of the Seventh International Computer Ethics Conference, University of San Diego, CA. Bennett, C. J. (2001, June 19). What government should know about privacy: A foundation paper. In Paper prepared for the Information Technology Executive Leadership Council’s Privacy Conference (Revised August 1, 2001). Retrieve January 29, 2008, from http://www.accessandprivacy.gov. on.ca/english/pub/wgskap.doc EU Directive. (1995, October 24). Directive 95/46/EC of the European Parliament and of the council. Retrieved January 29, 2008, from http://eur-lex.europa.eu/LexUriServ/LexUriServ. do?uri=CELEX:31995L0046:EN:HTML Fischer-Hübner, S, & Ott, A. (1998, October 5-8). From a formal privacy model to its implementation. In Proceedings of the 21st National Information Systems Security Conference, Arlington, VA. Retrieved January 29, 2008, from http://www. cs.kau.se/~simone/niss98.pdf Galpottage, A., & Norris, A. (2005). Patient consent principles and guidelines for e-consent: A New Zealand perspective. Health Informatics Journal, 11(1), 5-18. Retrieved January 29, 2008, from http://jhi.sagepub.com/cgi/reprint/11/1/5

GPC Alberta. (2003). Oipc stakeholder survey 2003—highlights report. Office of the Information and Privacy Commissioner of Alberta (Tech. Rep.). He, Q., & Antón, A. I. (2003, June 16-17). A framework for modeling privacy requirements in role engineering. In Proceedings of the International Workshop on Requirements Engineering for Software Quality (REFSQ 2003), Klagenfurt / Velden, Austria. Retrieved January 29, 2008, from http://crinfo.univ-paris1.fr/REFSQ/03/papers/P14-He.pdf HHS. (2003). Summary of the HIPAA privacy rule. U.S. Department of Health & Human Services. Retrieved January 29, 2008, from http://www. hhs.gov/ocr/privacysummary.pdf HIPAA. (1996, August 21). Health Insurance Portability and Accountability Act of 1996, Public Law 104-191, 104th Congress. Retrieved January 29, 2008, from http://aspe.hhs.gov/admnsimp/ pl104191.htm OECD. (1980). Guidelines on the protection of privacy and transborder f lows of personal data. Retrieved January 29, 2008, from http://www.oecd.org/document/18/0,2340, en_2649_34255_1815186_1_1_1_1,00.html Westin, A. F., & Gelder, V. (2005, August). Building privacy by design in health data systems. Center for Social and Legal Research: A report by The Program on Information Technology, Health Records and Privacy of the Center for Social and Legal Research. Retrieved January 29, 2008, from http://www.amia.org/inside/initiatives/healthdata/2006/ehrrept9-6-05_westin.pdf

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 2, edited by J. Tan, pp. 32-47, copyright 2008 by IGI Publishing (an imprint of IGI Global).

125

126

Chapter 10

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care? A Comparison among Norway, Denmark, and Sweden Agneta Ranerup Göteborg University, Sweden

aBstract The aim of this article is to evaluate the provision of Web support in choice reforms in health care in Norway, Denmark, and Sweden. Two main issues are investigated: (1) What institutional frameworks for choice in health care exist, and how is the exercise of choice supported by Web technology in these countries? (2) As a consequence of this, what roles of the individual are mediated by this technology? The present study provides a critical analysis of current technologies for providing information about health care. It is concluded that in Norway the individual is equipped to be a reasonably informed consumer, customer, and citizen. A similar situation exists in Denmark, but here the consumer role is even more prominent. In Sweden, there has been little technological support for these roles, but recently national actors have initiated a project aimed at creating a national portal for public health care. During the last 10 years, patients have become more informed than they previously were about various aspects of health care, often by using information technology (IT) (Josefsson, 2005; Tovey, 2006). IT, patients, and health care are

all core issues in several fields of research such as e-health and Consumer Health Informatics (Eysenbach, 2001; Mureo & Rice, 2006; Nelson & Ball, 2004; Tan, 2005). There are differences between these fields related to their history and

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

their focus of interest. However, it can be stated without controversy that a central rationale in both of these fields is to support the individual from the point of view of the medical rationality associated with the role of being a patient in need of care. This support might be provided in the form of supporting the provision of care and medical knowledge associated with the individual’s illnesses and treatments, including the larger administrative process in a hospital (Forducey, Kaur, Scheiderman-Miller, and Tan, 2005). The patient role and its associated medical rationality are seen in contrast to other roles, for example, that of citizen. A simple but yet telling example of the former role is a piece of research evaluating government health portals (Glenton, Paulsen, & Oxman, 2005). This article contains a broad comparative study of health portals in different countries. The focus is on assessing to what extent government health portals provide access to relevant, valid, and understandable information about the effects of various specified interventions (“treatments”). Against the background of the increasing importance of IT for patients, it is interesting to note an emerging phenomenon in health care: the introduction of choice reforms (Table 1). This development has its origins in the 1970s and the renewal of the public sector, often referred to as “New Public Management” (NPM), which dominated the reform agenda in many of

the OECD countries. Part of this renewal was an increase in the individual’s rights of choice in health care, mostly among public providers, but sometimes also among private providers (Le Grand & Bartlett, 1993). When discussing the introduction of choice reforms in health care, it is obvious that the institutional frameworks of choice (Table 1) varies among countries. This is also why comparative studies of institutional frameworks of choice in different countries are of interest (Silber, 2004; Vrangbæk, Østergren, Winblad-Spångberg, & Okkels, 2006). In a challenging way, a theoretical perspective outlined by Anttiroiko (2004) connects the issue of patients, IT, and institutional frameworks of choice reform in health care (Table 1). According to this perspective, IT and the institutional frameworks in conjunction mediate the relationship between the individual and public-sector services, for example, in the form of health care. The relationship between the individual and the public sector is, Anttiroiko argues, influenced by the coevolution of technology and institutional frameworks. In this way, the individual will be in a position to pursue different roles (Table 1) in his interactions with the public sector. This relationship is today mediated by technologies such as e-mail, workflow systems, and customer relationship management (CRM) systems, as well as the Web, which is used in the production and communication of

Table 1. Overview of fundamental theoretical concepts Fundamental concepts

Brief explanations

Choice reform

A type of public-sector reform emanating from the 1970s, which appeared in many OECD countries. Among other things, it intended to offer the individual a larger role either in making a direct choice among different providers of publicly financed services, or in exerting an indirect influence through the representative of a purchasing agency.

Institutional frameworks of choice

Laws or other forms of regulative agreements that delimit the right of choice of hospital for individuals in these countries.

Patient, citizen, consumer, customer

Different roles as well as accompanying rationalities and behavior that the individual may exercise toward health care.

127

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

health care services. In the relationship between the public sector and the individual, there exists also institutional frameworks of choice such as laws and rights (Table 1), in this case in particular, those that allow the individual certain rights to choose a preferred hospital. This perspective on health care may be primarily of interest to those countries in which health care is predominantly publicly financed (the United Kingdom, the Nordic countries, Spain, Italy, Australia, and Canada). However, it is of a more general interest to countries with predominately private health systems (the United States and others) (Palier, 2005) because it contributes to a richer understanding of how IT affects the individual’s potential to examine available services in today’s health care environment. For example, quality rankings that are offered through the Web are relevant for individuals that are trying to inform themselves about both public and private health care. With this as a background, the aim of this article is to evaluate the role of IT in general, and the Web in particular, in choice reforms in health care in three Nordic countries (Norway, Denmark, and Sweden) from the point of view of the individual seeker of care. Two main issues to be considered are the following: (1) What institutional frameworks for choice in health care exist, and how is the exercise of choice supported by Web technology in these countries? and (2) As a consequence of this, what roles of the individual are mediated by this technology? In summary, this research contributes to e-health and Consumer Health Informatics by taking the issue of institutional frameworks seriously when discussing the roles of the individual as mediated by technology. The article continues as follows: the first section briefly elaborates on the positioning of the study in relation to previous research and is followed by an overview of theoretical concepts for discussing the potential roles of the individual toward health care. Then the research method is described. After that, the empirical study is pre-

128

sented, with a focus on the framework for choice in health care in the three countries studied and a thorough description of the technical facilities in each case. This is followed by a critical analysis of these observations in the light of the potential roles of the individual that might be mediated by Web technologies in the three countries. Finally, some conclusions and issues for further research end the article.

InstItutIonal FraMeworKs For cHoIce wHIcH aFFect tHe role oF tHe IndIVIdual In HealtH care One basis for this study is the viewpoint that the national institutional frameworks that regulate choice are of importance when discussing the rights of choice in health care (Anttiroiko, 2004; Vrangbæk et al., 2006). However, questions about institutional frameworks and individual rights are seldom addressed in e-health and Consumer Health Informatics (Eysenbach, 2001; Mureo & Rice, 2006; Nelson & Ball, 2004). One reason for this might be that, at face value, incorporating institutional issues into the discussion would diminish the validity of studies for health care and patients in different countries. Even in research in which the focus is on the individual’s information needs in consultation situations (Coulter, Entwistle, & Gilbert, 1999; Tovey, 2006), this aspect is missing. Much research in these fields is based on experiences from the United States, where health care is generally not based on universal rights of large groups in society to receive subsidized health care (Palier, 2005; Tan, 2005). This might explain why it is less common to take the institutional framework into consideration in this context; however, there are a few exceptions. One example is found in a book about Consumer Health Informatics which discusses collaborative healthware with a special focus on a group of patients receiving Medicaid support (Goldsmith

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

& Safran, 2004). In a discussion of evidencebased patient choice, Elwyn and Edwards (2001) include aspects of the institutional framework in the medical choice situation, using experiences from the United Kingdom. In contrast, Peckham (2002) discusses the harmfulness of omitting the institutional framework in research about patients, especially when dealing with issues of choice in health care. The focus of the present study is on the institutional and technological components of choice reform in health care. In line with the theoretical framework proposed by Anttiroiko (2004), studying the institutional as well as the technological components of choice reform will be taken as the basis for discussing the individual roles that are mediated in each country.

concepts For dIscussIng tHe role oF tHe IndIVIdual In HealtH care It is interesting to note that the concepts of patient, citizen, and consumer describe roles related to the individual’s relationship with health care that are central in current health care discussions (Baggot, 2005; Harrison, Dowswell, & Milewa, 2002). When discussing the general public in e-health and Consumer Health Informatics, the most prevalent words used are “patient” and “consumer,” whereas the concept of “citizen” is much less common (Oh, Rizo, Enkin, & Jadad, 2005). Some authors argue that these concepts are marked by lack of clarity in terminology and usage (Baggot, 2005; Elwyn & Edwards, 2001). The intention here is not to provide an extensive discussion of the definitions of these concepts. However, an overview of how roles and rationalities are defined in research is a necessary basis for the rest of this study. The first role to discuss in the individual’s relationship with health care is that of citizen. In her study of choice in health care, Østergren (2004) describes individualistic and collectivis-

tic views of being a citizen. The individualistic perspective is the moral standpoint, meaning that the relationship with the collective becomes important and individual rights are emphasized. The collective tradition puts the individual’s interests aside to value collective ideas. In this case, it is the citizen’s obligation to the collective that is emphasized (Østergren, 2004). In a similar vein, Willems writes about doctors as gatekeepers guaranteeing fairness in health care. Thus, he argues, this arrangement “[…] constructs, in other words, patients as citizens” (Willems, 2001, p. 27). A further dimension stems from Baggot (2005), who describes citizens’ specificity in their potential to exercise “voice” rather than “choice.” Potential “voice” activities are, for example, taking part in societal discourse on health care or even pre-election debates about health care issues. This line of thinking can be contrasted with other ways of having an impact on health care by choosing certain services and providers rather than others (“choice”), and being in this way an active part of the larger development and renewal of health care (Clarke, 2006). Thus, the concept of citizen implies both individual capacities, such as knowledge about rights, and more collectivistic attributes such as having a sense of fairness and being equipped to take part in discourse about the conditions of health care. The role of consumer in health care can be defined by incorporating aspects of the ideal of the “calculating consumer” emanating from economics (Elwyn & Edwards, 2001). In line with this thinking, to behave as a consumer in the choice of a doctor or hospital, the individual must be able to obtain an overview of alternatives and to compare and rank these alternatives (Greener, 2003). No less important, when discussing the issue of offering individuals choice in health care reform, Østergren (2004) argues that the capacity to exit is the essential attribute of consumers. In other words, they can vote with their feet (Clarke, 2006). In contrast to these definitions and roles, when Howgill (1998) speaks about health care

129

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

consumers, he is talking about the private market for care, in which the individual depends on private economic resources to pay for care received. Thus, the consumer role implies capacities to consider different alternatives in a more elaborate way and to make an informed choice among them. Alternatively, one may consider the related role of health care customers. According to Greener (2003), the preferred as well as de facto attainable role of individuals in their relationship toward health care is to be a health customer. This means being, not a full-fledged calculating consumer, but an individual who is able to judge services received based on preferences and expectations. This judgment might be in form of a performance measurement regime in which actual patients would be asked for their opinions on the level of service they have received, as well as the progress they have made as a result of treatment. Last but not least, there is of course also the role of being a patient, or in other words, pursuing activities related to medical rationality (Glenton et al., 2005), as presented in the introduction.

researcH context and MetHod The context of this study is the institutional, but even more so, the technological component of choice reforms in health care in three of the Nordic countries: Norway, Denmark, and Sweden. The institutional framework of all these countries offers the option for a patient to choose a hospital other than the one that is closest to his or her place of residence (Vrangbæk et al., 2006). These three countries are marked by significant similarities from an economic and cultural point of view. In addition, it is no news that the access to Internet in these countries is high. Thus, it is argued here that they are an interesting test bed for examining the “state of the art” of technologies that support the choice of hospitals.

130

Because of the exploratory nature of the issues investigated here, the chosen research approach is the case-study method (Yin, 1999). This method allows for the study of a single phenomenon within its real-life context, while at the same time tolerating the condition that the boundaries between a phenomenon and its context are unclear. Data collection should ideally cover several sources. In addition, an operational framework should be in place, even though the current intention is exploratory. Having responded to requests for a consciously planned research strategy rather than an informal procedure, this type of study can generate reliable results. All these aspects have inspired the research approach of this study. The institutional frameworks for choice in the three countries are investigated by using other researchers’ writings on the issue of choice in health care. Furthermore, official documents describing the institutional frameworks in the three countries have been consulted. The intention has been to give an adequate, although not detailed, picture of the various frameworks. However, the main focus of attention is the analysis of technologies in each country from the point of view of choice reform in health care. Only public national portals are included in this study, because they provide a balanced view of how significant societal agencies present and support choice reforms in health care in these countries. More specifically, the portals are owned and implemented by national public agencies and associations in the three countries, for example, the Ministry of Health in Norway, the Association of County Councils, the Department of the Interior, and the Department of Health in Denmark, and the Association of County Councils in Sweden. The focus of attention is the content and functionality of the three countries’ technologies. This means that in the Norwegian case, the Web site www.sykehusvalg.no, introduced in 2003 as a prominent component of choice reform, is included as a whole. In Denmark, a national portal, www.sundhed.dk, for health care was introduced

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

in 2003. In this portal, all texts explaining choice reforms and other associated facilities are included in this study. This also means that links to other national public agencies’ facilities related to choice reform in health care are included in the analysis: a Web site describing waiting times in health care for different treatments (www.venteinfo. dk), a Web site describing available public and private hospitals included in the choice reform (www.sygehusvalg.dk), and an extended facility for the evaluation of quality of care (www.sundhedskvalitet.dk). In Sweden, there is no national public portal explicitly designed for supporting choice in health care. However, there are facilities offering an overview of waiting times for different treatments (www.vantetider.se). For the sake of completeness, a brief view is also given of the Web site Sjukvårdsrådgivningen (Health advice online), www.sjukvardsradgivningen.se, which is owned by the Association of County Councils in Sweden. This author provides a straightforward, nontheoretical walkthrough of the various facilities that are available through the portals in the three countries. An account of this investigation is also offered in the results section. This is used as the basis for analyzing the roles of the individual in health care, which are mediated by their contribution in view of the different role concepts presented in the literature (see “Concepts for discussing the role of the individual in health care”). More specifically, the features of the different role concepts are compared to the results of this technological walkthrough to see whether they “match.” For example, with regard to the role of a customer in health-care facilities, seeing evaluations of services from the ordinary service users’ point of view is fundamental for individual seekers of care (Greener, 2003). Furthermore, it is argued here that the mediated roles must be discussed with the help of an approximate, informal qualitative scale rather than an absolute and quantitative measure. The reason is the exploratory and qualitative nature of study in this field and the associated

application of the concepts for comparisons (Yin, 1999) of the mediated roles.

results overview of regulations Norway. In the early 1990s, free choice of hospitals was established in one region as a pilot project. In Norway, a new act expanding choice to the whole country was passed in 1999 and became effective in January 2001 (Sosial- og helsedepartementet, 1998). It allowed choice among all public hospitals. This reform was a part of a patients’ rights act which also included rights to assessment and second opinions, treatment within national and not only regional capacity limits, as well as the right of involvement and the right to information. One provision was that the patients themselves should pay little or no travel costs. In September 2004, hospital choice was extended to private hospitals and particular hospital units within multihospital trusts, as well as to child and psychiatric care. A waiting-time guarantee, procompetition legislation, and expanded capacity were also put into place (Vrangbæk et al., 2006). This meant that once a time limit has been exceeded in a region, the national office for social insurance must help the patient find either a private hospital or a hospital abroad (Sosial- og helsedirektoratet, 2004; Vrangbæk et al., 2006). This reform has been characterized as late, but, at that time, it was more radical than anything in Denmark or Sweden (Vrangbæk & Østergren, 2004). Denmark. In the early 1990s, free choice of hospitals became an issue in public discourse. The county councils initially opposed the idea, but entered a voluntary agreement before a formal parliamentary decision was made. When a formal decision was made in 1992, the issue of free choice was formulated as “extended choice” (Vrangbæk & Østergren, 2004). This meant that hospitals were allowed to refuse access to extended-choice

131

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

patients in times of heavy workload. The patients themselves had to pay their travel costs to receive nonacute treatment in other counties. Moreover, choice was limited to the same level of specialization and did not include private hospitals. This can be characterized as a somewhat restricted model (Vrangbæk & Østergren, 2004). However, since May 2002, patients have been free to choose any public Danish hospital and a few hospitals owned by patient associations for elective treatments, in case their GP finds that this is relevant. If a patient has waited more than 2 months in his or her home county, this right also includes private hospitals and hospitals abroad (Pedersen, Christiansen, & Bech, 2005; Vrangbæk et al., 2006). Sweden. In contrast to Norway and Denmark, free choice of hospitals in Sweden is not mandated in formal law, but was adopted as a recommendation by the Federation of County Councils (Landstingsförbundet) in 1989. This meant that since the beginning of 1991, patients, with some exceptions, have had the right to seek care throughout the entire country at primarycare centers and certain hospitals and private clinics. In 2000, these recommendations were clarified and simplified (Landstingsförbundet, 2000a), but they were still perceived as controversial by some county councils. In 2003, this new recommendation was accepted by all county councils (Vrangbæk et al., 2006). Through this recommendation, patients are given the right to choose a hospital or a specialist anywhere in the country, except in the case of highly specialized care. In some counties (8 out of 21), the patient must obtain a referral from his or her GP to seek hospital care outside his or her own county. In the other county councils, this is not necessary (Vrangbæk et al., 2006). Summary. The above discussion has revealed that all three countries have introduced institutional frameworks that give the individual the right of choice of hospital in the health care system. To choose a different hospital, patients need a

132

referral from their general practitioner (GP) or another doctor. Some authors argue that patients lack information about rights, and because of this, they are unable to benefit from them (Vrangbæk et al., 2006; Winblad-Spångblad, 2003). In addition, the GP plays a significant role as a source of knowledge and also in the role of issuing referrals (Vrangbæck et al., 2006). This would imply that the individual patient is in a difficult situation in regard to his or her potential to make use of these rights of choice.

overview of technologies for supporting choice in Health care Norway. In Norway, there has been some experimentation with Web-based league tables containing quality indicators and free telephone lines in 1998-2001 (Vrangbæk et al., 2006), or in other words, the period immediately preceding the introduction of choice reform in health care. However, when choice in health care was introduced de facto in 2001, it was soon supported by a Web site called Free Hospital Choice (Fritt sykehusvalg), www.sykehussvalg.no, which was publicly launched in May 2003. A text on this Web site, accessed on November 24, 2006* (all information marked by * comes from the same source), explains its aim: [It] supports the government’s goal of facilitating patients’ right to choose where to receive treatment. The service offers patients, next of kin, and clinical personnel up-to-date quality information concerning patient’s rights, waiting times, and quality information about the different hospitals, as well as other relevant information. Moreover, according to this text, this purpose of this Web site is to empower Norwegian citizens and to enable them to make better-informed decisions about which hospital to choose for different types of treatment. The stated background to the Web site is the Patients’ Rights Act. This law gives

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

a patient the right to choose in which hospital he or she is to be treated. The Web site contains several sections with different features. The first section is called Waiting times (Ventetider).* A comprehensive list of 17 kinds of treatments and illnesses (pediatric treatments, heart ailments, psychiatric disorders, etc.) is found on the front page of this section. By clicking on one of these areas, the user can access a list of treatments. By selecting one of these treatment choices, he or she obtains a list of waiting times in number of weeks for medical examination, walk-in clinic treatment, and inpatient treatment for all hospitals in the country that have adequate competence in each particular case. One can also go to an alphabetical list of treatments or click on a picture of the human body. Here one can select a type of treatment and in this way obtain the relevant waiting time. The database containing the waiting times for treatments is searchable, but the search is limited to one or a few of the five hospital regions in Norway. In the list of waiting times for treatments, a user can also find information about quality indicators and type of institution (public or private) that are relevant for each hospital. A second section is called Attention to quality (Pekepinn på kvalitet).* Here a text explains the concept of quality indicators, saying that the general public can use the indicators to get information about the quality of different services, predominantly at the hospital level. It also states that the significant factor is the experiences of patients, whereas the indicators do not necessarily say much about the result of the medical treatment as such. There are links leading to descriptions of various quality indicators, such as: (1) the number of planned operations that subsequently are postponed; (2) indicators expressing different views of quality of care from the perspective of both walk-in clinic patients and inpatients. For each one of these, the figures are compared with mean value statistics for all hospitals in Norway: standard of premises, communication, organization, information, ease of getting around in the

hospital, and overall experience of care received; and (3) the number of patients residing in corridors as opposed to ordinary patient rooms at the hospital. A third section is called Rights (Rettigheter).* This contains a general and easily understandable document about the right to choose a hospital for treatment. There are documents stating the types of hospitals that are eligible for choice and an overview of how health care will help the individual defray the costs of going to another hospital. There are also links to the official documents describing the framework for regulating choice. A fourth section is called, For health care personnel (For helsepersonal).* These facilities can also be accessed by patients and are similar to the ones provided to patients. However, in this facility, the waiting-time data and quality indicators contain a longitudinal dimension. With regard to the quality indicators, through the facilities for health personnel, it is possible to compare a larger number of indicators for each hospital than by using the facilities designed for citizens. Denmark. In Denmark, a portal called Health Care (Sundhed), www.sundhed.dk, was launched in 2003. On its home page, accessed November 24, 2006** (all information marked by ** comes from the same source), is the statement: “Sundhed.dk is your main entrance to health care in Denmark. Here you get information that is useful for patients and health personnel.” There are also facilities that enable the individual patient to follow his or her recorded treatments and diagnoses through the facilities provided. The portal contains several sections: Health, Treatments, Medicine, What about the law?, Facts and figures, Organization of health care, and News. There are also links to all the regions in Denmark and their hospitals. Furthermore, there is a special section for patients containing links to the complaints center, facilities for self-service, donation of organs, and patient records. A button labeled Treatments provides access to a special database with waiting times in weeks for treatment, Waiting times (Ventetider).**

133

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

This facility is created and administered by the National Board of Health. A text says: “Here you can find waiting time for elective treatments and operations in public and private hospitals. This information is only approximate, which is why you should seek further advice through your GP, patient advisor, or hospital personnel.” This facility offers the option to select one of 21 kinds of treatment, as well as associated treatment subtypes. Then the relevant hospitals for the selected treatment appear, as well as the waiting time. It is possible to include private hospitals in this search. There is also an option to restrict the search to various regions in Denmark. Moreover, by clicking on the Treatments, button, one can also access the new facility that was introduced in November 2006: Quality in Health Care (Sundhedskvalitet).** A press release introducing this facility says that, “This Web site is the first version of how a complete picture of quality and service can be made available through the Internet offering information about individual hospitals.”** This intention is further described on its home page as a means to support the choice of hospital. It is also stated that by looking at statistics about different aspects of hospital operation, it is possible to obtain information about the quality of both public and private hospitals. The user is advised to make a choice of quality indicators from those that are available: (1) standard of premises: number of beds per room, plus toilets; 2) sanitary conditions: hand washing, postoperative infections, kitchen facilities, and sanitary services in general; (3) rights: contact person, waiting-time guarantee, and extended choice of hospital; (4) patient safety: mistakes in medication, injuries during surgery; and (5) patient satisfaction: general, sense of inclusion in care, safety when leaving the hospital. There are also quality indicators for the different types of treatments or kinds of illness. For example, for cataract operations, there are quality indicators for: (1) activities: number of patients, average number of days for inpatient

134

treatments; (2) expected waiting time: for medical examination, hospital inpatient treatment, and walk-in clinic treatment; (3) complications. In the case of heart failure, quality indicators include: (1) types of medical examination performed: electrocardiogram (EKG), X-ray; (2) treatments: physical exercise, patient education, beta-blocker use; (3) mortality. In contrast, for some types of treatments, for example infertility treatments, there are no indicators related to the medical examinations or results of treatment, but only the expected waiting time for treatment. There are 21 kinds of treatment to choose from, followed by associated treatments for each primary treatment chosen. Search can also be limited on a geographical basis (by region). When using these facilities, the quality indicators for the hospitals implementing the selected treatments are shown. The Treatments section in the Health care (Sundhed) portal also contains a subsection called Hospital statistics. Here it is possible to search statistics for both public and private hospitals to determine the total number of patients and the most prevalent types of treatment for each hospital. There are also statistics for the duration of treatments in general (for inpatients) and for the most prevalent types of treatment in each hospital. In the Treatments section, there is a final subsection related to choice in health care: Quality of care. Here one can find information and databases related to different quality studies in association with different types of treatments. There is also a database containing patient views of care, derived from a patient survey conducted in 2004. It contains patient views about, for example, their relationship with hospital personnel, their sense of inclusion in care, information provided when leaving the hospital, hospital premises, and waiting times. Another section accessible from the home page, What about the law?, is also relevant to the issue of choice in health care. This section has several subsections, of which one is called Free choice of hospitals. Here the frameworks for choice are summarized, with links to documents

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

containing the actual frameworks for choice. Equally important, on the Health care portal, there is also a link to a Web site called Free Hospital Choice (Sygehusvalg.dk)** implemented by the Association of County Councils. The home page of this Web site states that: Patients that do not receive care by a public hospital after two months of waiting now have the option to seek care at private clinics in Denmark as well as hospitals abroad. […] To use the right of free choice of hospital, there is a precondition that the chosen hospital has an agreement with the county councils and hospitals about the relevant treatment.** At this level, there is a link called Patients. By clicking on this link, one finds general information about patient rights as well as the Web site itself. There are also links to waiting times (the database described above) and a leaflet about the free choice of hospitals. There is an option to search among all hospitals included in the free choice of hospitals after the 2-month waiting period. In this database, there are options to search in an alphabetical list containing all private hospitals and clinics, to search within selected geographical boundaries, and to search among 18 areas of treatment (breast, heart, orthopedic, and others). For each one of these alternatives, there is a selection of relevant treatments. Clicking on any one of these brings up a list with all the private hospitals that are in a position to offer the selected type of treatment. Sweden. In Sweden, a Web site called Waiting Times in Care (Väntetider i vården), www.vantetider.se, was introduced in April 2000, containing a database of waiting times for various types of treatments. It was created by the Association of County Councils (Landstingsförbundet). At that time, the database contained waiting times for approximately 30 different treatments and included about 30 specialist hospitals (Landstingsförbundet, 2000b). The database is part of efforts by the county councils and regions to

strengthen the position of patients by increasing the transparency of health care and making the various waiting times public. In 2005, waiting times for both specialized care and primary care were included. This Web site could be accessed directly as well as through the Web sites of the county councils in Sweden. The home page of Waiting Times in Care, accessed on November, 26, 2006*** (all information marked by *** comes from the same source), indicates that the site contains waiting times for both specialized and primary care. There are also links to a news section as well as to reports. Both of these are primarily directed toward specialists because their main focus is on the finer technicalities of waiting times and how they are reported. The database offers the possibility of searching for a certain hospital, treatment, kind of illness, or region. 26 kinds of treatment are available for choice, six types of specialized medical examination, and approximately 40 treatments. For each treatment, the relevant hospitals, waiting time in weeks, information about free capacity, and contact information of hospitals with free capacity are presented. The information about free or excess capacity is defined as the potential of each clinic to receive patients coming from county councils other than its own. The search can be limited to data about first visit, medical examination, and treatment. Concerning the issue of choice in health care, a document*** available on the Web site expands on the relationship between the waiting-time guarantee and the free choice of hospitals and explains that the ways of exercising choice are decided by the individual county council. It is also pointed out that a prerequisite for exercising the right of choice is that a doctor in your home county council has issued a referral. This treatment can then be carried out in another county council’s jurisdiction. Sometimes, when the treatment is costly, some county councils may request special permission before the right of choice can be requested. The rights to information about options are described

135

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

as follows: “Normally, it is the responsibility of the patient to find an alternative provider of care.” This formulation can be contrasted with the recommendation about choice in health care that since 2003 has been adopted by all county councils in Sweden: “It is an important task of the county councils to inform their inhabitants on a continuous basis about the options for choice in health care. Such information must be directed towards the population as a whole as well as to those that are seeking care” (Landstingsförbundet, 2000a, p. 2). Furthermore, in 2001, the National Board of Health and Welfare in Sweden published a report on quality indicators in health care written as a government project. In this report, the use of quality indicators from internal and external viewpoints was very briefly discussed as a basis for choice of health care provider. The Web site, Waiting Times in Care, was characterized as an important basis for the choice of health care provider as pursued by patients and doctors (Socialstyrelsen, 2001). A new report including a broad spectrum of quality indicators for different treatments was published in 2006 (SKL & Socialstyrelsen 2006). However, this report did not form part of an online service. The indicators included showed data that compare the different county councils, not the individual hospitals. Furthermore, the indicators were explicitly introduced as not to be used as a basis for the choice of hospital (SKL & Socialstyrelsen 2006, p. 12). In the autumn of 2007, a new report was produced in which the individual hospital’s values for some of the indicators were published as a test case (SKL & Socialstyrelsen, 2007). In Sweden, there is also a portal, Health advice online (Sjukvårdsrådgivningen), www.sjukvardsradgivningen.se, owned by the county councils in Sweden. The focus of this Web site is on illnesses, injuries, anatomy, health, drugs, and treatments. There is also a section on rights, focusing mostly on disabilities, means of assistance, and dental care. There is information about the waiting-time

136

guarantee, but no information about the rights of choice in health care. In 2006, the Association of County Councils in Sweden initiated a project to build a national portal for health care: Health care on the Web (Vården på webben). The portal Health advice online and its developer provided the basic infrastructure for this work, but since 2007, all 21 county councils have joined in. A prominent goal behind the new national portal is that an individual should have access to information about all kinds of providers of care, including those that are not part of the publicly financed health care system. A further goal is that the national portal should enhance the individual’s ability to compare conditions and regulations, waiting times, and quality of care (MD, National project, personal communication, March 29, 2007). This ambition has been backed up by government policies (Socialdepartementet et. al., 2007). The launch of the national portal is planned for 2009. The design of individual facilities was discussed intensively in the autumn of 2007, and requirements documentation is planned to be available for the purchasing process to start in December 2007 (Designer, National project, personal communication, September 18, 2007).

dIscussIon technology in choice reforms in norway, denmark, and sweden There is a national Web site aimed at supporting choice of hospitals in Norway, whereas in Denmark there is a national portal for health care as a whole that contains many facilities to support choice. Sweden, by contrast, does not have facilities at the national level for supporting the free choice of hospitals other than those dedicated to the waiting-time guarantee and some very brief information about rights of choice in connection with this. These three cases illustrate the different ways that technology is implemented by important

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

national authorities and associations in health care. More specifically, such an implementation might be in the form of a standalone facility for supporting choice of hospital, as in Norway, or of a facility that is part of a larger portal incorporating facilities from different public agencies, as in Denmark. Thus, in the Danish case, there is a common interface to health care as a whole. This is good from the point of view of the individual for whom the concept of a one-stop portal is convenient, offering a single window to all available services (Wimmer, 2002). From a technological viewpoint, the facilities for providing these kinds of information in the three countries are quite simple. However, when it comes to the facilities showing waiting times, Norway provides several types of interfaces: graphical (a picture of the human body) as well as textual (lists with sublevels for kinds of illnesses and treatments, or an alphabetical list of treatments). The two other countries provide textual interfaces only. The Norwegian technology in this respect gives the individual the option of using various interfaces according to his or her preferences and levels of knowledge. For finding relevant hospitals for treatments, it is possible to restrict the search in several ways, by geographical area as well as by private or public provider. Also in Norway, the facilities that show waiting times provide links to quality information for the chosen hospitals, thus making it easier for the individual to combine these two aspects of choice. Last but not least, the Danish facilities for quality indicators provide various options for restricting search (illnesses, treatments, geography, or type of quality indicator). This is good if, as in Denmark, a broader spectrum of quality indicators is offered and care is taken to avoid information overload when the results are shown. Moreover, preferences regarding indicators might vary among individuals and situations. These are the most obvious examples of the fact that technological facilities can provide an extra performance-enhancing capability beyond the most obvious and simple aspect of the Web:

making information available at a distance and around the clock. Against this background, it is possible to conclude that the Web technology features implemented in these three countries are fairly simple: adding capabilities to restrict search in various ways, connecting various types of information which support choice, and providing a choice of preferred interfaces. However, there also exists the potential to provide more advanced, composite interactive decision support to the individual seeker of care. In this case, simple and straightforward Web technology might be used for supporting choice by arranging a logical chain of operations that the individual can follow. One example is from the field of educational and career guidance, for which computer-based decision support has been provided through the Web for some years (Tait, 1999). Moreover, in the Swedish labor market, the National Labor Market Board has, since 2004, provided computerized decision support through the Web through which an individual states his or her preferences regarding choice of education and vocation. This part of the decision chain is followed by further facilities to delimit search, examine available options, and rank these options (Norén & Ranerup, 2005). Recently, a similar model has been introduced to provide computerized decision support to the individual citizen in the new premium pension system in Sweden (Ranerup, 2006).

Mediated roles of the Individual with respect to Health care Norway. In Norway, the facilities provided in www.sykehusvalg.no have the potential to support the individual in becoming a reasonably informed citizen with regard to the rights of choice (Østergren, 2004). An individual user can find more general information about rights and benefits as well as links to the relevant official documents regulating choice. The sense of fairness (Willems, 2001) in the distribution of and access to care is to

137

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

some extent supported by the waiting-time information for each care provider. More specifically, fairness is supported by transparency regarding such strategic information, which is fundamental from a democratic point of view: in this way, the differences in waiting times for various hospitals and types of illness become visible. As in the provision of all kinds of public services, there is the potential for citizens to exercise voice as opposed to choice (Baggot, 2005), in this particular context by taking part in the societal discourse on health care. This is supported by information about rights, waiting times, available options in the form of public and private hospitals, and, last but not least, quality indicators. A limitation here is the focus on choice in health care in the technological facilities provided, which means that the individual is not offered an overview of health care in general through a one-stop portal (Wimmer, 2002). Such an approach would make the individual more knowledgeable about health care in general, but would also imply that individuals are using the available facilities on a regular basis. In Norway, the individual is also reasonably well equipped through the Web to behave as a consumer (Greener, 2003; Harris, 2003), by being offered an overview of available hospitals for requested treatments and information about quality of care. There are various types of quality indicators in health care (Rygh & Mörland, 2006; Socialstyrelsen, 2001): indicators for structural issues (capacity for care, education of personnel), for process issues (to what extent clinical praxis is in accordance with the optimal process for medical examination and treatment), for the result of treatments, for example mortality, and for the patient’s view and experiences (Socialstyrelsen, 2001). In the Norwegian quality indicators, there is a focus on structural issues in general, which appears to some extent also in the form of patients’ views on structural conditions. However, indicators for comparing the treatment received with the optimal course of medical examination

138

and treatment for a specific disease as well as with the results of treatment are absent. This means that the individual in Norway is equipped to compare different hospitals, although not in a deep and elaborate way. It is argued here that the quality indicators provided in Norway are quite limited in the sense that the individual is equipped to function as a customer rather than as a fullyfledged calculating consumer who is capable of comparing and ranking a considerable number of available options. According to Greener: “[H] ealth customers would judge the level of service they believe they have received, and assess health services according to their own expectations of the services as well as to national standards” (Greener, 2003, p. 86). The most important thing here is, according to Greener, that it is the patients’ opinion of the level of service they have received that is the focus of attention. Moreover, the role focusing on medical or patient rationality is supported by the facilities provided, because these facilities offer information that is of relevance to those seeking care. A general overview of available hospitals and their competence in various of treatments and illnesses is part of this information. However, in the Norwegian case, as already noted, there is the limitation that the focus of the information on facilities is solely on the choice of hospital. Denmark. In Denmark, individuals are reasonably well equipped to function as citizens through the available technological facilities which provide an overview of rights of choice and fairness in a manner equivalent to Norway. However, in their role as consumers, the Danes are a little better off than the Norwegians. This is mainly because of the recently implemented facilities offering quality indicators for health care, Quality of Care (Sundhedskvalitet), which provide a basis for making comparisons. In fact, the introduction and continued development of quality indicators was explicitly a part of supporting choice reform in health care (Socialstyrelsen, 2001). In terms of the categories of quality indicators classified as

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

structural-, process- and results-oriented (Rygh & Mörland, 2006), the Danish facilities contain instances of all of these. Interestingly, this includes the result of treatments both at a general level (complications, mortality) and connected with specific treatments. As for the issue of supporting individuals as customers, there are also indicators expressing the more personal views of patients (Socialstyrelsen, 2001) with regard to care received. These are in the form of five indicators covering different aspects of patients’ safety and satisfaction from the patients’ point of view. In fact, in Norway and Denmark, the individual patient is the target of these attempts that, as a further step in these processes, are made available through the Web and therefore more accessible as well as searchable. However, this is not to say that all patients have the indicators they need to make a choice or that they have a sufficient understanding of available indicators to use them in their decisions (Rygh & Mörland, 2006). Last but not least, the Danes are also better off with regard to facilities offered to support the individual as a patient. Not only are facilities that are relevant for choice offered (available hospitals and treatments), but also many types of general and specific information about health care, covering quality of care as well as health in general. Recently (autumn 2006), the connections between the portal www.sundhed.dk and the individual patient have been made stronger by the introduction of technological facilities which enable the patient to review his or her own treatments. In this manner, the facilities provided can be seen as a means of connecting the individual to many aspects of health care, including those that have to do with institutional and medical issues. Sweden. The Web facilities supporting choice in Sweden must be characterized as modest. The brief information about rights and the waitingtime information to some extent support the individual in the role of citizen. As for support for the role of consumer, there is not much to be found. In the updated portal, Health advice

online (Sjukvårdsrådgivningen), the individual is, to some extent, equipped to act as a patient, because many forms of support in line with the patient or medical rationality are offered. However, these are not connected to individual hospitals or primary-care providers, but are kept on a general level. In Sweden, the regulatory framework says that it is an important task for the county councils to keep the general public, as well as those that are seeking health care, continuously informed about their rights of choice (Landstingsförbundet, 2000a), but not much is said about how or in what forms this should be done. However, in the information about choice that does exist at a national and regional level (see above) the individual is defined as being responsible for finding information about the hospitals that he or she wants to use as a part of the right of choice. This exposes a mixed attitude toward the individual in regard to his or her rights to information in choice reform in general and information about available options for choice in particular. Of course, there is a significant difference between, on the one hand, Norway and Denmark and, on the other hand, Sweden. In Norway and Denmark there are laws guaranteeing the right of choice, whereas in Sweden there is an agreement among the county councils. In Sweden, regional authorities have a comparatively large influence when it comes to de facto regulation of citizens’ rights. In addition, as argued by Karlsson (2003), in Sweden, the individual generally has no rights in his or her role as a citizen that can be requested from an authority or institution in health care or elsewhere, in the way that such a request could be made, for example, in the United Kingdom (UK). It is seen as a paradox that less generous welfare states, for example the UK, offer citizens more individual rights than does Sweden (Karlsson, 2003). Furthermore, the quality-indicator projects in Sweden are marked by the absence of indicators supporting choice, because there are no indica-

139

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

tors for individual hospitals. However, attempts are being made to introduce such indicators as of the autumn of 2007 (SKL & Socialstyrelsen, 2007). The indicators cover various aspects such as waiting times, treatments used, results of treatments, and patients’ views. The rationale behind the work with indicators in Sweden can be expressed in the following way: “Citizens and patients have the right to know what health care accomplishes and to compare their own county council with others. An equally important aspect is that comparing results enhances learning and the further development of health care” (Socialstyrelsen & SKL, 2006, p. 10). Thus, it appears that a discursive rationale about enhancing transparency for citizens through quality indicators is considered important in Sweden, although implemented rather half-heartedly. The reason for this judgment is that the actual reports describing the quality indicators are not more specifically targeted to be understood by citizens. During the autumn of 2006, there was a change of government in Sweden. The intention to increase choice in health care is supported in policy writings of the new coalition (Allians för Sverige, 2006). It is argued that hospitals should be responsible for informing the individual about where he or she can obtain care with minimal waiting time. In this way, current political events might change the technological aspects of choice reform in health care in Sweden. It is also interesting to note that recently (January 2007), all the county councils aligned themselves with a new project which aims to launch a national health care portal in 2009. The purchasing process is planned to start in 2008, followed by a complex implementation phase involving the 21 county councils. This initiative can be interpreted as a late attempt to combine the resources of the comparatively independent county councils, with, to date, an unclear prognosis when it comes to the chances of a successful result in a reasonable time.

140

conclusIon This study contributes to research in e-health and Consumer Health Informatics by investigating the roles of the individual that are mediated by technology in choice reforms in health care, thus emphasizing the joint role of institutional frameworks of choice and technology. From a general point of view, the technological support for this objective is considerable in Norway and Denmark, but much less developed in Sweden. However, it is fair to say that the Web technology features that are implemented are fairly simple, offering capacities for delimiting search, connecting various types of information, and choosing the preferred type of interface. However, in the facilities offered by the three countries, there is no more advanced computerized decision support to enhance the choice of hospital. In spite of this, Web technology offers information about hospitals, rights, and quality of care right at hand to the individual, thus bypassing, for example, doctors, who have often had the role of information gatekeepers (Winblad-Spångberg, 2003). As for the roles of the individual as mediated by technology, in Norway, the technology supports the individual in becoming a reasonably informed citizen regarding the rights of choice, as well as a consumer with some capacity to make comparisons and choose a hospital. In Denmark, the individual is equipped to behave as a reasonably informed citizen, but also to a significant extent as a consumer because of having access to more fully developed information about services (hospitals, treatments, quality indicators) and facilities to compare these. In Sweden, there is considerably less potential for the individual to become an informed citizen or consumer by means of available technology. A difference between the three countries is that the rights of choice in health care are implemented by means of laws in Norway and Denmark, but in Sweden, rights of choice are a recommendation that has been accepted by all the County Councils.

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

The focus of this study has been on technological support for choice and not on other potential sources of support, for example patient advisors or paper documents that are not a part of a technological infrastructure such as the Web. The approach chosen has been considered to be relevant against the background of the increasing importance of IT for patients and health care alike. It is based on a theoretical framework that sees the relationship between the individual and the state as more than that previously mediated by institutional arrangements and technology (Anttiroiko, 2004) and thus contributes to the discourse in this field. The research presented here also contributes to the contemporary praxis of health care. At a general level, this study has shown how different types of technology supporting choice in health care can be designed in today’s health care environment. This is of interest to countries that are expanding choice reform in health care and wish to use IT as a part of this effort. A recent example in this respect is the UK, with its intention to offer technological facilities to support choice (Department of Health, 2003), which have been more recently (2005-2007) rolled out. As for further research into technology to support choice in health care, the finer aspects of providing quality indicators to patients through the Web are of interest. The problem is one of providing understandable and relevant information about important aspects of care and how this might be communicated to the individual (Rygh & Mörland, 2006). It has been argued that the use of quality indicators is relevant from many viewpoints. However, the fact that there are methodological problems with, for example, reporting the result of health care makes the use of these indicators for comparing hospitals problematic when they are used as a basis for choice (Schen, 2005). The question is, then, whether this is a surmountable problem, or whether new types of quality indicators for this particular area of use will have to be developed and provided through

the Web. Another interesting issue for research is what indicators citizens actually prefer as a basis for choice. Yet another issue for further research is the potential to introduce computerized decision support that combines different stages and aspects of a decision process for hospital choice, in line with similar attempts in the fields of educational and career guidance and premium pension issues, as discussed earlier. These forms of support are based on a model for relevant decision-making in the area of concern. In the area of educational and career guidance, there are certain theories that support this (Law, 1999). In regard to the choice of premium pension funds, the decision support provided is designed on the basis of financial theory and psychology (SOU, 2005). An interesting issue, then, is what theoretical frameworks might form a part of similar design attempts in health care. Last but not least, the newly initiated project of implementing a national health care portal is both relevant for the issues discussed here and a challenging issue for further research. This is even truer because one of the major objectives of the portal is to provide information about all kinds of health care providers as well as to support comparisons of rights, quality, and waiting times. However, implementing an infrastructure for a portal at a national level is one thing, but ensuring that the 21 county councils in Sweden actively implement the requested regional structure with information and interactive services is quite another. As summarized by a high-ranking civil servant, “The local leadership of this process is the most important thing” (MD, The Association of County Councils Purchasing Agency, personal communication, December 10, 2007).

acKnowledgeMent Thanks are due to the Bank of Sweden Tercentenary Foundation for funding this research.

141

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

reFerences Allians för Sverige. (2006, August 23). Election program [Valmanifest]. Retrieved January 29, 2008, from www.maktskifte06.se/fileadmin/ Upload/pdf/Valmanifest_2006.pdf Anttiroiko, A. (2004). Introduction to democratic e-governance. In M. Mälkiä, A.-V. Anttiroiko, & R. Savolainen (Eds.), E-transformation in governance: New directions in government and politics (pp. 22-49). Hershey, PA: Idea Group. Baggot, R. (2005). A funny thing happened on the way to the forum? Reforming patient and public involvement in the NHS in England. Public Administration, 83(3), 533-551. Clarke, J. (2006). Consumers, clients or citizens? Politics, policy and practice in reform in social care. European Societies, 8(3), 423-442. Coulter, A., Entwistle, V., & Gilbert, D. (1999). Sharing decisions with patients: Is the information good enough? BMJ, 318, 318-322. Degoulet, P., Fieschi, M., Jaulent, M-C., & Ménard, J. (2004). Patient empowerment, cybermedicine, and citizen education. In R. Nelson & M. Ball (Eds.), Consumer informatics. Applications and strategies in cyber health care (pp. 120-128). Berlin: Springer. Department of Health. (2003). Choice of hospitals: Guidance for PCTs, NHS Trusts and SHAs on offering patient choice of where they are treated. London: Department of Health. Elwyn, G., & Edwards, A. (2001). Evidence-based patient choice? In A. Edwards & G. Elwyn, G. (Eds.), Evidence-based patient choice. Inevitable or impossible? (pp. 3-18). Oxford University Press. Eysenbach, G. (2001). What is e-health? Journal of Medical Internet Research, 3(2), Article e20.

142

Forducey, P., Kaur, K., Scheiderman-Miller, C., & Tan, J. (2005). Surveying the e-health landscape: Cases and applications. In J. Tan (Ed.), E-health care information systems. An introduction for students and professionals (pp. 203-224). San Fransisco, CA: Jossey Bass. Glenton, C., Paulsen, E., & Oxman, A. (2005). Portals to wonderland: Health portals lead to confusing information about the effects of health care. Medical Informatics and Decision Making, 5(7). Goldsmith, D., & Safran, C. (2004). Collaborative healthware. In R. Nelson & M. Ball (Eds.), Consumer informatics. Applications and strategies in cyber health care (pp. 9-19). Berlin: Springer. Greener, I. (2003). Patient choice in the NHS: The view from economic sociology. Social Theory & Health, 1, 72-89. Harris, K. (2003). How do patients choose physicians? Enrolees in employment-related health plans. Health Services Research, 38(2), 711-732. Harrison, S., Dowswell, G., & Milewa, T. (2002). Guest editoral: Public and user “involvement” in the UK National Health Service. Health and Social Care in the Community, 10(2), 63-66. Howgill, M. (1998). Health care consumerism, the information explosion, and branding: Why ’tis better to be the cowboy than the cow. Managed Care Quarterly, 6(4), 33-43. Josefsson, U. (2005). Coping with illness online: The case of patients’ online communities. The Information Society, 21(2), 143-153. Karlsson, L. (2003). Conflict or harmony? Individual rights in Swedish and British health care [Konflikt eller harmoni? Individuella rättigheter och ansvarsutkrävande i svensk och brittisk sjukvård]. Doctoral dissertation, Göteborg: CEFOS, Göteborgs universitet.

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

Landstingsförbundet. (2000a). Recommendation about choice in health care. [Rekommendation om valmöjligheter inom hälso- och sjukvården]. Dnr Lf924/00. Stockholm: Landstingsförbundet. Landstingsförbundet. (2000b). Waiting times in care on the Internet [Väntetider i vården på Internet. Pressmeddelande] 2000-04-10. Stockholm: Landstingsförbundet. Law, B. (1999). Career-learning space: newDOTS thinking for careers education. British Journal for Guidance & Counselling, 27(1), 35-55. Le Grand, J., & Bartlett, W. (1993). Introduction. In J. Le Grand & Bartlett W. (Eds.), Quasi-markets and social policy (pp. 1-12). London: Macmillan. Mureo, M., & Rice, R. (2006). The Internet and health care. Theory, research, and practice. NJ: Lawrence Erlbaum. Nelson, R., & Ball, M. (2004). Consumer informatics. Applications and strategies in cyber health care. Berlin: Springer. Norén, L., & Ranerup, A. (2005). The Internet Web portal as an enrolment device, In B. Czarniawska & T. Hernes (Eds.), Actor-network theory and organizing (pp. 188-207). Malmö: Liber and Copenhagen Business School Press. Oh, H., Rizo, C., Enkin, M., & Jadad, A. (2005). What is e-health (3): A systematic review of published definitions. Journal of Medical Internet Research, 7(1), article e1. Østergren, K. (2004). The institutional construction of consumerism. A study of implementing quality indicators (Working paper 15). Bergen: Stein Rokkan Centre for Social Studies. Palier, B. (2005). How to understand the differences between various health care systems [Att förstå skillnadena prestation mellan olika hälso- och sjukvårdssystem]. Stockholm: Sveriges kommuner och Landsting.

Peckham, S. (2002). Book review. Health and Social Care in the Community, 10(6), 520-522. Pedersen, K., Christiansen, T., & Bech, M. (2005). The Danish health care system: Evolution–not revolution–in a decentralized system. Health Economics, 14, 41-57. Ranerup, A. (2006). Electronic markets connecting citizens to pension reform. Electronic Markets, 16(4), 282-291. Rygh & Mörland. (2006). Hunting for good quality indicators [Jakten på de gode kvalitetsindikatorerne]. Tidsskrift for Den Norsk Lægeforening, 21, 2822-2825. Silber, D. (2004). The case for e-health. In T. Iakovidis, P. Wilson, & J. Healy (Eds.), E-health (pp. 3-27). The Netherlands: IOS Press. SKL & Socialstyrelsen. (2006). Open comparisons of quality and efficiency in health care. A comparison between county councils in 2007 [Öppna jämförelser av hälso- och sjukvårdens kvalitet och efektivitet. Jämförelser melan landsting 2007]. Stockholm: SKL, Socialstyrelsen. SKL & Socialstyrelsen. (2007). Open comparisons of quality and efficiency in health care. A comparison between county councils in 2007 [Öppna jämförelser av hälso- och sjukvårdens kvalitet och efektivitet. Jämförelser melan landsting 2007]. Stockholm: SKL, Socialstyrelsen. Socialdepartementet, SKL, Socialstyrelsen. (2007). National IT strategy in health care. Where do we stand? Report 2007 [Nationell IT-strategi för vård och omsorg. Var står vi idag? Lägesrapport 2007]. Stockholm: Socialdepartementet, SKL, Socialstyrelsen. Socialstyrelsen. (2001). A comprehensive view of quality indicators in health care. A proposal by the National Swedish Board of Health and Welfare [Övergripande kvalitetsindikatorer inom hälso- och sjukvården] Stockholm: Socialstyrelsens förslag.

143

In What Ways Does Web Technology Support the Individual in Choice Reforms in Health Care?

Sosial og helsedepartementet. (1998). Law about patient rights [Lov om patientretttigheter]. Ot prpnr 12 (1998-99). Oslo: Sosial- og helsedepartementet. Sosial-og helsedirektoratet. (2004). Law on patient rights. Instructions [Lov om pasientrettigheter. Rundskriv]. IS12/2004. Oslo: Sosial- og helsedirektoratet. SOU. (2005). Difficult waters? Premium pension savings on course [Svårnavigerat? Premiepensionssparande på rätt kurs?], SOU 2005:8, Swedish Government Official Reports. Stockholm: Fritzes. Tait, A. (1999). Face-to-face and at a distance: The mediation of guidance and counselling through new technologies. British Journal of Guidance & Counselling, 27(1), 113-122. Tan, J. (2005). E-health care information systems. An introduction for students and professionals. San Fransisco, CA: Jossey Bass. Tovey, D. (2006). The informed patient. Singapore Medical Journal, 47(9), 741-745.

Vrangbæk, K., Østergren, K., Winblad-Spångberg, U., & Okkels, B. (2006). Patients’ reactions to free choice of hospitals in Norway, Denmark and Sweden. Bergen: Norwegian School of Economics and Business Administration. Willems, D. (2001). Balancing rationalities: Gatekeeping in health care. Journal of Medical Ethics, 27, 25-29. Wimmer, M. (2002). Integrated service modelling for online onestop government. Electronic Markets, 12(3),149-156. Winblad-Spångberg, U. (2003). From decision to practice: The role of the physician in implementing patient choice in Swedish health care [Från beslut till verklighet: Läkarnas roll vid implementeringen av valfrihetsreformer i hälso- och sjukvården]. Uppsala: Institutionen för folkhälsooch vårdvetenskap, Uppsala Universitet. Yin, R.K. (1999). Enhancing the quality of case studies in health service research. Health Service Research, 34(5), Part II, 1209-1224.

Vrangbæk, K., & Østergren, K. (2004). The introduction of choice in Scandinavian hospital systems. Arguments and policy processes in the Danish and the Norwegian case (Working paper 5). Bergen: Stein Rokkan Centre for Social Studies.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 2, edited by J. Tan, pp. 32-47, copyright 2008 by IGI Publishing (an imprint of IGI Global).

144

145

Chapter 11

Characteristics of Good Clinical Educators from Medical Students’ Perspectives: A Qualitative Inquiry Using a Web-Based Survey System Gary Sutkin University of Pittsburgh School of Medicine, USA Hansel Burley Texas Tech University, USA Ke Zhang Wayne State University, USA Neetu Arora Texas Tech University, USA

aBstract Medical educators have a unique role in teaching students how to save lives and give comfort during illness. This article reports a qualitative inquiry into medical students’ perspectives on the key qualities which differentiate excellent and poor clinical teachers, using a Web-based questionnaire with a purposeful sample of third- and fourth-year medical students. Thirty-seven medical students responded with 465 characteristics and supportive anecdotes. All participants’ responses were analyzed through reviewing, coding, member checking, recoding and content analysis, which yielded 12 codes. Responses from 5 randomly chosen participants were recoded by two authors with an inter-rater reliability coefficient of 0.72, implying agreement. Finally, 3 larger categories emerged from the data: Content Competence, Teaching Mechanics, and Teaching Dynamics. We incorporate these codes into a diagrammatic model of a good clinical teacher, discuss the relationships and interactions between the codes and categories, and suggest further areas of research.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

IntroductIon Medical teachers must be creative and effective teachers in addition to being sound clinicians and successful researchers (Bowen, 2006). Curriculum reform increases the need for skilled teachers in our medical schools. Although most faculty do not undergo formal teaching training during their medical training, many United States medical schools have created faculty development workshops to help faculty, among other things, become better teachers (Searle, Hatem, Perkowski, & Wilkerson, 2006). But what makes a good medical school teacher? Some educators have attempted to answer this question by surveying their own medical students: some through the coding and categorization of answers from surveys (Boendermaker, Conradi, Schuling, Meyboom-de-Jong, & Swierstra, & Metz, 2003; Cote & Leclere, 2000; Ker, 2003; Pinsky, Monson, & Irby, 1998) and others from the analysis of answers to Likert-type scales (Morrison, Hitchcock, Harthill, Boker, & Masunaga, 2005; Cox & Swanson, 2002; Elzubeir & Rizk, 2002). In our medical school, student assessments of clinical faculty who teach third and fourth-year medical students are heterogeneous, performed on a departmental basis, inconsistently administered, and not standardized or validated. Multiple sources are recommended to improve the content of educational assessment (Epstein, 2007); students are one source of assessment of faculty teaching. To identify quality teaching in clinical settings, it is critical to understand what contributes to good clinical teaching from students’ perspectives. The advancement of Internet technologies also provides new opportunities for dynamic, instant Web-based data collection efforts. In addition to the fast speed and higher accuracy rate of data collection, a Web-based questionnaire also makes it possible to reach and engage prospective participants despite the physical distances and diverse geographical locations. More importantly, a well designed Web-based survey system may provide

146

dynamic, automatic features such as e-mail invitations for participation, participation tracking and records, e-mail reminders and other necessary follow-ups about the survey research. Thirdly, a Web-based questionnaire may easily incorporate interactive components into the survey such as a pop-up reminder, as necessary, to encourage participants to provide more verbiage in their responses to open-ended questions. Additionally, a Web-based survey makes it easy to collect data fast and in a predetermined analyzable format. Finally, the Web-based approach is usually less expensive than the other options (Schleyer, 2000).

reVIew oF tHe lIterature Medical educators enjoy high status among professional educators, with good reason. Complex medical innovations, falling mortality rates, and the growth of the field of medical education have made the need for good medical teaching all the more critical. What makes an effective medical educator, in particular, and an effective teacher, in general? Interestingly, with a metaphorical flourish, Oser and Baeriswyl (2001) compared a teacher to an expert in an emergency room, someone who reacts constantly to the immediate events, despite having a plan. They describe a lesson period as a chain of operations guided by rules, with the teacher employing innumerable helping, piloting, and controlling activities to meet lesson goals. In the last 3 decades, in research on public school teaching, the emphasis has been on linking such teaching behaviors to student performance (Borich, 1996). The real answer to the above question is that teaching is a very complex activity that is influenced by myriad factors, both personal and environmental. In an empirical study, Good and Brophy (2000) identified effective elementary and secondary teacher behaviors that were associated with increased student performance. These factors included teacher efficacy, student opportunity to

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

learn, classroom management and organization, curriculum pacing, active teaching, teaching to mastery, and a supportive learning environment. The Handbook of Research on Teaching (4th ed), atomized the topic, deeply surveying the literature by subject field, including writing, reading, mathematics, science, health education, physical education, visual arts, history, social studies and more (Richardson, 2001), and emphasized observable teaching behaviors. The act of teaching itself seems to be as complex as the content in any field being taught. However, educational research in postsecondary institutions is less focused on teacher behavior and more focused on faculty development, even though faculty development can be a fuzzy concept. Menges and Ausin (2001) surveyed the literature on postsecondary teaching and suggested that discipline and institutional contexts shape instructional practices. In particular, they concluded that instructional practices diverge based on differences between the paradigmatic hardness of the field, with hard fields (e.g., chemistry and physics) having rather fixed practices and softer fields (e.g., sociology and English) being more receptive to instructional innovation. In the postsecondary literature, instructional development programs are often buried in faculty development efforts, with “faculty development” often used synonymously with effective teaching development. There also appears to be considerable criticism of faculty development efforts. Murry (2003) suggests that these programs lack cohesiveness and should be focused on teaching. He called for formal programs, strong support from leadership and ties of teaching effectiveness to the reward structure. Gottlieb, Rogers, and Rainey (2002) suggested that self-assessment of instruction is important to teaching effectiveness, assessing skill, educational needs, progress, and strengths and weaknesses of performance. Finally, some research on postsecondary teaching identifies instructor personality as an important precursor to student performance and

satisfaction. For example, Prichard and Sawyer (1994) identified the instructor’s personality as an important factor in how postsecondary teachers teach, along with philosophy, teaching methods, outside influences and past role models. The central theme is that personality and method must be the right fit for instructor, subject matter, and students. In another study, Lempp and Seale (2004) interviewed medical students on the quality of their teaching in British medical schools. While students reported positive role models, they also found a hidden curriculum, one focused on the importance of hierarchy in the teaching of medicine. Students reported many incidents of humiliation. They concluded that clinical practice and research have dominated clinical educator’s time. Teaching in a clinical setting is considered to be a lesser activity, unrecognized in incentive plans, and in fact, few physicians had received any training in teaching, learning theory, or learning assessment.

MetHods Characteristics of effective clinical teachers have not been studied extensively in the literature. Qualitative inquiry methods are well-suited for relatively underexplored areas of research and for theory-building. The primary goal in this study was to identify key qualities that differentiate good and poor clinical teachers using a qualitative research design. Thirty seven medical students from a large Southwestern medical school voluntarily responded to an open ended, Web-based questionnaire regarding characteristics of good and poor clinical teachers. Internet as a qualitative research tool is gaining popularity (Mann & Stewart, 2000) and has many advantages such as low cost (Schaefer & Dillman, 1998), convenience, and context of noncoercive and antihierarchical dialogue (Boshier, 1990), which lends itself to collaborative research. The Internet has become an important mode of com-

147

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

munication and safe expression. Its reach within the academic community is even higher. When research participants are difficult to reach and the data of interest is sensitive in nature, as was the case in this research, the Web offers the unique advantage of collecting and generating data in a centralized and noninvasive manner. It provides an opportunity for researchers to capture larger audiences, thereby enhancing cost effectiveness. Participants of this study were asked in an electronic questionnaire to list three characteristics of a good clinical teacher and a 3-5 sentence description of an instance when a faculty member demonstrated that quality in his or her teaching activities. Additionally, they were asked to describe a time when a faculty member was a nonexample of that characteristic. The questions were pilot tested by three recent graduates to enhance the clarity, relevance, user-friendliness, and applicability of the questionnaire. Their recommendations were incorporated into the final version (see appendix for an abbreviated Web form). Upon receiving approval from the Institutional Review Board, an invitation for participation was sent to a pool of potential participants. Because we were interested in understanding the perceptions of medical students regarding clinical teaching, the sample selected for this research was purposeful. In general, qualitative studies rely on the power of carefully selecting information-rich participants for whom the topic is meaningful and relevant (Sandelowski, 1995). The sample of participants for this research consisted of (a) medical students in the last month of their thirdor fourth-year, and (b) graduates in their first year of residency. Their voluntary participation was solicited through personal contact, e-mails, and flyers distributed at shelf examinations. The flyers and e-mails contained information about informed consent and listed a Web-link to the questionnaire. In order to ensure anonymity of responses, no identifying information was collected from participants. We invited dialog as participants’ schedules permitted.

148

Thirty seven students participated in this research (21 males; 16 females). The majority of them were White-American (70%) and in the age-group range of 25-29 years (59%). Nearly half of the participants (49%) were in their third year of medical school. Their preferred medical specialties varied widely. Due to their textual nature, Web-based questionnaire responses lend themselves to contentanalysis, which was our choice of methodology. According to Weber (1990), content analysis is “a systematic research method for analyzing textual information in a standardized way that allows evaluators to make inferences about that information.” In terms of the sampling units, we received 465 total response posts, in 12-point Font, which totaled approximately 30 pages of single-spaced text data. Three authors (GS, HB, NA) independently reviewed and coded the textual data to find patterns among the responses. These codes represented overarching themes of characteristics of good clinical teachers as identified by the students. Use of research team members to interpret and double-check the coding schemes is a way to enhance reliability in qualitative research. A commonly used means of establishing reliability is the use of multiple coders and the closely related technique of peer examiners (LeCompte & Preissle, 1993), which reduces potential bias in the analysis and reporting phase by using multiple perspectives to validate results (Kvale, 1996). The authors of this study met weekly for 2 months, exploring the patterns they observed in the data. One author (KZ) attended these meetings as an external auditor in order to maximize reliability in the multiple rater coding process. The preliminary labels for the codes were agreed upon, and they were routinely evaluated for agreement. Discrepancies were resolved by discussion. This process of refining led to establishing the coding schemes. Codes were retained if at least two of the three authors agreed. Codes were expressed in the positive sense. For example, quotes about

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

an attending not having time to spend with the student and about an attending taking extra time with the student were both classified under the code “Available.” During the final stages, the authors defined the codes and classified them into larger categories. The data analyses resulted in 12 codes which were classified into three final categories. The resultant codes and categories were then used by two authors (GS, NA) to re-code responses from five randomly chosen students. This recoding yielded an adequate inter-rater reliability coefficient of 0.72 for qualitative research.

1.

FIndIngs

2.

The findings presented below, describe the codes and the categories that emerged from content analysis. The analysis yielded 12 codes, which we classified into three larger categories: (1) Competence, (2) Teaching Mechanics, and (3) Teaching Dynamics. See Tables 1, 2, and 3 for full descriptions, definitions, and positive and negative examples. In our examples, we have replaced faculty names with “Dr. X” and gender pronouns with alternating use of “he” and “she.”

coMpetence We defined Competence as the set of knowledge, skills, and abilities that physicians must be able to perform and demonstrate. Many students described effective clinical teachers as excellent physicians worthy of modeling. These comments could be separated into two main codes: knowledgeable and relationships with patients. The students were impressed not only by their teachers who were intelligent, well-read, and technically skilled, but also by those adept at interacting with their patients.

We defined knowledgeable according to the ACGME core competency definition: demonstrates knowledge of the biomedical sciences and its application to patient care (ACGME, 2007). Examples of quotes include: Dr. X is a man of extreme knowledge; he has many years of experience on top of many years of education. I think knowledge is essential to a teacher because how can you be a great teacher without having information to teach? We defined forming positive relationships with patients according to the ACGME core competency definition: provides patient care that is compassionate, appropriate and effective (ACGME, 2007). Students frequently used descriptors like “respect,” “caring,” “approachable,” and “positive.” Examples of quotes include: Dr. X gives each patient’s concerns the time and attention that he or she desires. As a result, Dr. X’s approach to interacting with everyone is something that students can feel comfortable emulating.

More examples and full definitions of these codes can be found in Table 1.

teacHIng MecHanIcs We defined Teaching Mechanics as the instructional toolset that helps students analyze and synthesize biomedical information. The good teachers used these techniques in their every-day clinical teaching. These tools ranged from good questioning techniques to unique learning activities. We categorized these instructional practices into five codes: guided practice, presentation techniques,

149

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Table 1. Competence–the knowledge, skills, and abilities essential to being a physician Characteristics of Good Medical School Teachers (Codes, Definitions and Examples) Codes

Knowledgeable

Forms positive relationships with patients

Definitions

Quotes: Positive Examples

Quotes: Nonexamples

Demonstrates knowledge of the biomedical sciences and its application to patient care

♦ “Dr. X can quote you all the landmark journal articles to support all sides of a given clinical situation and acknowledges that there are multiple views/ways of doing things” ♦ “a database of information…she not only explains the details, but the process behind the details

♦ “Some teachers cannot stand to be wrong or admit they do not know something” ♦ “Occasionally you have questions that do not have an answer. Bad teachers will try to sidetrack and confuse you.”

Provides patient care that is compassionate, appropriate and effective

♦ “I admire the way that Dr. X treats all patients … with respect. He initiates these interactions by being approachable and making people feel comfortable.” ♦ “taught the patient about his disease”

♦ “talking about patients … with unkind remarks” ♦ “Many would try to make up for lack of … patient interaction (on rounds) with the ‘what would you guys like me to talk about today’ speech”

Dr. X’s method of diagramming concepts is not unique but it does help make the vast amount of information manageable. By organizing thoughts in a concept diagram, a foundation can be formed on which details can be added.

patient-related teaching, constructivist teaching, and enrichment. 1.

2.

The technique of guided practice allows the teacher to take a student step by step through the learning process. It can be accomplished verbally with questions and prompts, or visually, such as the supervised teaching of a procedure. One student noted: Dr. X would often put the instrument in your hand and walk you through a procedure. Superior presentation techniques were also often mentioned. They sometimes took the form of an organized lecture or a novel way of expounding on clinical topics. Two examples are: Dr. X had very clear learning objectives and made sure that we all participated and completed learning goals. She knew what your limitations were and had reasonable expectations of students at our level.

150

3.

Many students appreciated patient-related teaching. Two examples are: When he visits a patient on rounds, each patient becomes an opportunity to learn from him. He knows his patients very well and can tell each one’s story in remarkable detail with little to no paperwork to prompt him. He explains patients’ conditions and disease processes in ways that everyone in the room can understand. She makes us present in front of the patient and discusses the case openly in front of the patient so as to include them in their treatment plan and disease process.

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

4.

Comments were also made about faculty skilled in constructivist teaching, or interactive teaching that helps students construct the meaning for themselves from a given context. Constructivist teaching takes advantage of what a student already knows, and then fits new information to construct a deeper level of understanding (Caine & Caine, 1991). For example: Dr. X would bring picture atlas to rounds. He would bring a list of board questions and even made an amazing handout that seemed to fit perfect for boards.

5.

Enrichment is used by teachers to allow students to come to a much deeper understanding of a topic. Students often commented on feeling “challenged” and “intellectually stimulated.” Two examples are:

nonthreatening, available, patient, enthusiastic, and respectful. 1.

Dr. X is non-intimidating and doesn’t make you feel like you’re constantly being graded. Dr. X never talked down to us or belittled us. 2.

3.

We defined Teaching Dynamics as the personality characteristics that improve the conditions of learning for the students. These characteristics generally represented the more “human” side of the doctor and included attitudes, emotions, values, and reflective ability. Effective teachers were often described as “caring” about their students. We divided the comments into five codes:

Patient: Dr. X gives the students and residents time to present the H&P without interrupting. Listens to everything we say. Dr. X allowed me to close the fascia and was very patient. When Dr. X noticed how nervous I was, Dr. X used humor to make me feel better and reassured me that every physician has been there.

4.

Enthusiastic: Dr. X has an energetic teaching style... (and) definitely keeps our attention.

More examples and full definitions of these codes can be found in Table 2.

teacHIng dynaMIcs

Available: I felt free to ask Dr. X any question that I wanted, regardless of how simple.

Dr X had the perfect combination of asking questions to pull knowledge and make you think, but balanced it with a quick didactic giving her thoughts. Dr. X would have you read all the time over your patients and expected that you would know EVERYTHING about them. He always pushed you to know the next step.

Nonthreatening:

Dr. X loved to teach and would tell you that daily. 5.

Respectful: Dr X loves to hear students’ ideas about management and treatment. We talk it out amongst (us) so we are almost all equals.

More examples and full definitions of these codes can be found in Table 3.

151

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Table 2. Teaching Mechanics—instructional toolset that helps students analyze and synthesize biomedical information Characteristics of Good Medical School Teachers (Codes, Definitions and Examples) Codes

Guided practice

Presentation techniques

Patient-related teaching

Constructivist teaching

Enrichment

152

Definitions

Quotes: Positive Examples

Quotes: Nonexamples

Questions, prompts, and models in order to clarify what learners are to do.

♦ “Dr. X spends many hours making sure that students understand her lectures. She asks to see if every student knows how to perform a good physical examination on a newborn. Having [been] checked by Dr. X, students feel confident in their physical [examination] skill.”

♦ “Faculty members that lose patience in an OR setting and grab the needle driver out of your hand” ♦ “Dr. X. got mad at me for not driving the (laparoscopic) camera right but wouldn’t show me how to do it”

Delivers educational content with superior organization, scope, sequencing, and pace.

♦ “Presents knowledge in an understandable easily learnable format” ♦ I’ve never heard REI/amenorrhea explained in a more memorable/simple fashion than he did”

♦ “Unorganized … it was very confusing” ♦ “when they don’t know how to convey that knowledge into something interesting to listen to”

Uses the patient as a teaching tool, integrating knowledge and skill while maintaining patient dignity.

♦ “Includes patient in conversations” ♦ “Taught clinical skills by demonstrating them and then having us briefly practice the skills on patients”

♦ “how the material related to patient care and how it related to other topics remained a mystery” ♦ “Fast rounds in the hall, no teaching.”

Teaches interactively to help learners “construct” meaning for themselves.

♦ “Dr. X demonstrates an ability to be creative in both his lectures as well as his clinical teaching points. He gave a lecture just the other day in which he had us recreate the pelvic floor with play-do. This will forever be in my mind because it was interactive and imaginative. Not just another regurgitation of material. This type of anatomy lecture was made much more memorable by imagination and the interactive nature of the lecture.”

♦ “Things don’t seem to stick as well when there is very little interaction” ♦ “One-way conversation is particularly poor in small group settings such as rounds”

Provides in-depth content or challenging examples in order to deepen the students’ understanding

♦ “and I gain so much more understanding of the topic because Dr. X explains the “why” and “how” of pathophysiology” ♦ “Later in the week Dr. X rereviewed the discussion with us briefly. That helped solidify the points he made“ ♦ “gave us extra assignments”

♦ “what you need to be learning is the core medical knowledge, not random ‘resident level’ facts” ♦ “Many staff would simply follow bullets from power points without making any clinical correlations”

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Table 3. Teaching Dynamics–personality characteristics that improve learning conditions for students Characteristics of Good Medical School Teachers (Codes, Definitions and Examples) Codes

Definitions

Quotes: Positive Examples

Quotes: Non-examples

Uses nonmenacing language and actions

♦ “I never heard Dr. X raise her voice … she instructed without humiliating” ♦ “Dr. X didn’t criticize me when I didn’t know the answer”

♦ “One attending is so freakin’ intense and intimidating that you stress out the entire day, night” ♦ “You are afraid to ask questions because he will make you feel like it is a stupid question”

Available

Spends extra time teaching the students

♦ “Says ‘not now, we’ll discuss it later’ and in fact, does discuss later” ♦ “We had a number of conversations and she really listened – this from someone who was working late into the night to take care of her patients” ♦ “Took the time to … demonstrate how he did that surgery”

♦ “Many physicians are pressured by the bottom line, seeing patients and billing. Teaching should take a priority …” ♦ “There were a couple of instances in which the faculty made me feel as though I was an impediment to his/her work and that I was ignorant and unwanted.” ♦ “Some attendings would not initiate interactions with students. Any interaction is limited to residents”

Patient

Appears calm, composed, steadily persevering

♦ “Remains calm/patient in all settings” ♦ “Ensured everyone understood” ♦ “Calm and unwavering”

♦ “Not staying calm” ♦ “when we’re in their clinic, they emphasize that we need to go fast and if you’re not done, they’ll walk in on you”

Passionate about being a doctor and eager to teach

♦ “She took sincere joy in showing us proper exam technique” ♦ “Because he enjoys his work, he places his specialty in a positive spotlight & encourages students to follow in his footsteps” ♦ “She loves the field”

♦ “reviewed topics in a very detached fashion” ♦ “These faculty members were merely teaching as a fulfillment of a requirement of their contract”

Approaches others with a sense of deference for their worth and esteem

♦ I admire the way that Dr. X treats all patients, students, and staff with respect. ♦ Dr. X at all times maintains respect for everyone’s personal predilections and teaches with an abject neutrality

Nonthreatening

Enthusiastic

Respectful

“pIMpIng” Although we did not put “pimping” as a separate code or category, we noted the frequency of passionate comments made it. Pimping was noted as both malignant and positive.

♦ “Another attending … was even more rude, inflammatory and derogatory to me” ♦ “Other teachers were obviously not interested in … showing students any respect”

Two examples of positive “pimping” included: I think being pimped is an ok way to teach, as long as it is not the kind of pimping where you are SO freaked out about their next question and getting it wrong that... you can no longer concentrate. Dr. X was not like that at all. Dr. X’s questions were

153

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

relevant to each patient and... it created such a better learning environment. One day in the OR, Dr. X asked basic anatomy questions concerning branches of the abdominal aorta. I did not recall at that time from anatomy 2 1/2 years prior all the answers Dr. X was looking for. That experience pushed me to learn more. Pimping could also be threatening and noneducational: Dr. X made us feel like we were being separated and stoned. Some attendings like to ask very specific and pointed questions with the intent of being able to ask you questions that you are unable to answer. This method of learning has the advantage of being easily remembered (because it is always easier to remember uncomfortable situations that made you feel unintelligent); however, they do not incorporate this knowledge into a usable framework. Instead you just fill your mind with factoids that stand alone. We assigned each example of “pimping” to relevant codes. For example, “if you didn’t know the answer, Dr. X would make you feel stupid” was coded as a negative example of nonthreatening, and “Dr. X expects you to know everything about your patients” was coded as a positive example of enrichment.

dIscussIon Through coding and categorizing the descriptive responses obtained using a Web-based questionnaire, we were able to construct a paradigm of the excellent clinical teacher from the students’ perspective. The excellent clinical educator is both knowledgeable and good with patients, utilizes a toolset of effective teaching techniques, and possesses personality characteristics that draw the

154

student into the teaching moment. These qualities are not exhaustive, but are certainly illustrative. Even though the data easily separated into these 12 codes and three categories, interaction and overlap among them was evident, and thus we created a pictorial representation of their interaction (Figure 1). We call this troika of domains the Student Inspired Model for Effective Clinical Teaching (SIMECT). According to our model, each category is affected by its associated teaching characteristics (i.e., codes), which can have either a positive or negative effect on its effectiveness. The personality attributes were especially easy to view in multiple intersecting continuums (for example, ranging from threatening to nonthreatening, disrespectful to respectful, and the like). We incorporated the negative comments into our model and drew the arrows to show how these continuums affected clinical teaching. We believe this model to be predictive in nature, so that the right mix of educator competence, teaching dynamics, and use of teaching mechanics can increase or decrease and speed or slow student learning of critical knowledge, skills, and dispositions. The codes were stable when analyzed by teachers and were not altered by changes in context. A clinical educator who would take time to teach during a stressful surgery, for example, was also described as a nonthreatening questioner, enthusiastic about teaching, and patient with students and those under his or her care. The domains are connected in such a way that improving practice in one area may help behavior in another area. This model can be a very useful tool in developing teacher education programs. For instance, a program designed to improve the enrichment (teaching mechanics) of a clinical lesson would also improve the students’ perception of the teacher’s enthusiasm (teaching dynamics) and knowledge of the content (competence). Although all three domains influenced effective clinical teaching, we learned through our coding process that Teaching Dynamics was the

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Figure 1. Student inspired model for excellent clinical teching (SIMECT)

student Inspired Model for excellent clinical teaching (sIMect)

Knowledgeable

Competence

Interaction with patients

NonEnthusiastic

Available

Threatening Respectful Impatient

Teaching Dynamics Patient

Outcome: Excellent Clinical Teaching

Disrespectful NonThreatening Busy

Enthusiastic

Guided Practice

Enrichment

Presentation Techniques

Teaching Mechanics

Constructivist Teaching Patient Related Teaching

most frequently mentioned. In fact, it seemed to be the centerpiece feature of students’ observations of the effective clinical educator, in that it served as a catalyst for the overall teaching-learning process. Teaching dynamics often made evident the instructor’s competence, while influencing (positively or negatively) any particular strategy used to deliver content to the learner. For example, a knowledgeable instructor who taught

nonthreateningly appeared to have presentation techniques more effective than one who berated while presenting. This finding was completely unexpected, and it dominated respondent discussions and stories. Without dissent, students concluded that the positive teaching personalities mediated both the display of competence and the use of teaching techniques.

155

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

The Teaching Dynamics finding is reflective of personality research found in personality development psychology, particularly the personality orientation known as the Big Five Model. The Big Five personality factors are agreeableness, conscientiousness, extraversion, emotional stability, and openness (Goldberg, 1993). These dimensions of personality were derived from factor analyses of language people used to describe themselves, and they are useful because they can represent diverse systems of personality description (Oliver & Srivastava, 1999). According to Oliver and Srivastava (1999), these traits are stable and are predictive of behaviors, including workplace behaviors. Our findings suggest that “expressions” of these personality traits (e.g., nonthreatening, available, and patient) are also predictive of teaching effectiveness that can be targeted for personal development and change. For example, a clinical educator can consciously use language that is less threatening or can learn more effective ways of providing feedback on student performance. The use of pimping as an instructional technique is one example of how an expression of personality type can affect teacher-student relationships. We were not surprised to receive so many comments surrounding the art of pimping. “Pimping” is a loosely defined term that connotes a person in power publicly probing a junior colleague’s knowledge base. It usually manifests as the questioning of a medical student in front of their peers. Wear, Kokinova, Keck-McNulty and Aultman (2005) recently surveyed fourth-year medical students and discovered that although pimping can be intended for humiliation or can be inappropriate for the student’s level, it is often seen by students as a positive pedagogical tool. Our students also noticed this dual nature of pimping, and many appreciated the stimulus for additional learning. Therefore, a highly motivated and extraverted clinical educator could pimp enthusiastically and respectfully, increasing students’ interest in learning a content or a skill. Conversely, a negative teaching dynamic,

156

like using pimping to humiliate, in the hands of a cold and distant clinical educator could shut down learning for students. Although our qualitative data were taken from one medical school, we suspect that these ideal clinical teaching qualities have some universality in the relationship among competence, teaching dynamics, and teaching methods, with teaching dynamics being the feature of teacher that most affects adaptations in the face of student, contextual, and environmental differences. Elzubeir and Rizk (2001) asked medical students what they look for in a faculty role model and then coded and categorized the results into three categories, similar to ours. Their students especially appreciated faculty who demonstrated respect, honesty, politeness, and enthusiasm. Paukert and Richards (2000) reviewed written descriptions of faculty who had “significantly and positively influenced their clinical education.” They coded and categorized the responses into five categories: person, physician, teacher, supervisor, and unspecified (global). While the first three were similar to our categories, “supervisor” included attendings that provided their students with opportunities to participate in patient care and practice skills. Irby, Ramsey, Gillmore, and Schaad (1991) used observations of faculty teaching to create a 44-item, 7-point scale questionnaire, which they administered to students to identify the characteristics of effective ambulatory clinical teachers. They found that good teachers were described as being actively involved with learners, promoting learner autonomy, and demonstrating patient care skills. The dynamic relationships between the three major components of good clinical teaching also have practical implications on the design and implementation of quality e-learning in health education. Teaching dynamics, which were so important to our students, can be incorporated and translated into e-learning or blended learning teaching activities.

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

The Web-based questionnaire, designed and developed particularly for this study, worked very well as a data collection instrument with a few customized features. We were able to track and record participant’s login attempts, identify incomplete submissions and provide reminders for participants to complete the survey in one or multiple attempts. The Web-based system also provided friendly prompts as necessary to encourage elaborations on open-ended questions. These technology-enhanced features have certainly helped increase response rate and obtain complete and richer data, in a more timeefficient, user-friendly, and cost-efficient fashion. We highly recommend future developments of similar Web-based assessment tools in health education research. As with most studies of this type, our study limitations include a small sample that may not be representative of the population under study. We anticipate that our backgrounds and biases might have influenced our coding and subsequent grouping into characteristics. Another group of coders might classify the data in a different way. We attempted to mitigate our biases through collaboration between medical (GS) and general (HB, KZ, NA) educators. The Web-based survey greatly expedited the pilot testing of our survey. The qualities submitted by our pilot-testers became our first coding categories, helping us to develop an initial framework for categorization. Adjustments to the Web survey after pilot-testing were finished immediately. From then on, respondents could access the survey from the Internet at their convenience, and the data was easily retrievable by the researchers. Because of the physical and academic distances from each other and because this project was funded by a time-sensitive grant, the researchers initially planned to analyze the collected data using a qualitative software program, which we believed would help to bridge those differences among us. The idea was to keep the process as seamless and paperless as possible, with data

entered by the respondents, rarely needing to be printed and never needing to be retyped. We especially wanted to avoid the delay and disconnect associated with large amounts of raw data in the hands of a transcriptionist. Additionally, we hoped that the speed and reliability of the Web-survey itself would be replicated with the data analysis performed by the qualitative software package. After reviewing several qualitative software packages, we chose Qualrus, a relatively new product that promised artificial intelligence (AI) algorithms that would help speed data analysis (Brent, Slusarz, & Thompson, 2002). This software application has graphical coding tools that can guide and simplify the coding process. Our choice of this particular product was based on the earlier experience of one of the researchers with the product. We also liked the idea of the AI features. Having the software learn from user codes and make recommendations of possible codes could be very helpful. However, ours was a simple project, made even simpler by the Web survey; so we did not fully use the software. Our project had one timepoint for respondents to answer the survey. Additionally, our data was already highly structured, and we were interested in speed. The data was easily organized by simply cutting and pasting with a word processor. The software’s process mirrors classical qualitative analysis procedures, including coding, grouping the codes, establishing relationships among the groups of codes, and managing reliability issues. We decided to meet face-to-face and used this exact logic to resolve coding, recoding, grouping, and reliability issues. Though the meetings were time-consuming, no software package could replace the dynamism and efficient problemsolving of those meetings. Like Seidel (1991), we desired to be closer to our data and did not want our software “bridge” to become an impediment. Qualrus software is better suited for complex longitudinal projects that amass large amounts of data from multiple sources using various me-

157

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

dia. If we had massive amounts of data with no ostensible organization, it is easy to see how our implicit inductive and iterative processes could have been made more explicit, and hence more manageable, using qualitative software. Future areas of research might include content analysis of in-person interviews with medical students and faculty as well as direct observation of faculty teaching medical students. Although these methodologies have been previously used (Boendermaker, Schuling, Meyboom-de-Jong, & Swierstra, & Metz, 2000; Irby, 1994; Irby et al., 1991), we would like to conduct them in our medical school and compare the findings to the SIMECT model. Our findings can have an important impact on medical school assessment of clinical teaching. Our medical school faculty are not regularly evaluated on the 12 teaching characteristics identified as most important by our medical students. Future research might aim at testing some of these characteristics in order to create new assessment toolsets. Finally, our study does not address the effects of these shared characteristics on student learning. Further research on instructor intentions and attitudes may help illuminate the effect of these teacher characteristics on student learning. Eventually, interventions based on these domains could be developed to improve clinical instruction. In conclusion, we present a paradigm of effective clinical teaching as viewed by our medical students: a good clinical teacher possesses a comfortable competence, specific instructional strategies, and a dynamic set of positive personality traits.

reFerences ACGME Outcome Project. (2007). Accreditation Council for Graduate Medical Education. Retrieved January 29, 2008, from http://www. acgme.org/outcome/comp/compFull.asp.

158

Boendermaker, P.M., Conradi, M.H., Schuling, J., Meyboom-de-Jong, B., Swierstra, R.P., & Metz, J.C.M. (2003). Advances in Health Science Education Theory Practice, 8, 111-116. Boendermaker, P.M., Schuling, J., Meyboom-deJong, B., Zwierstra, R.P., & Metz, J.C.M. (2000). What are the characteristics of the competent general practitioner trainer? Family Practice 17, 547-553. Borich, G. (1996). Effective teaching methods (3rd ed.). Englewood Cliffs, NJ: Merrill/Prentice Hall. Boshier, R. (1990). Socio-psychological factors in electronic networking. International Journal of Lifelong Education, 9(1), 49-64. Bowen, J.L. (2006). Medical education: Educational strategies to promote clinical diagnostic reasoning. New England Journal of Medicine, 355, 2217-2225. Brent, E., Slusarz, P., & Thompson, A. (2002). Qualrus: The intelligent qualitative analysis program (manual). Columbia, MO: IdeaWorks. Caine, R.N., & Caine, G. (1991). Making connections: Teaching and the human brain. Alexandria, VA: Association for Supervision and Curriculum Development. Côté, L., & Leclère, H. (2000). How clinical teachers perceive the doctor-patient relationship and themselves as role models. Academic Medicine, 75, 1117-1124. Cox, S.S., & Swanson, M.S. (2002). Identification of teaching excellence in operating room and clinic settings. American Journal of Surgery, 183, 251-255. Elzubeir, M.A., & Rizk, D.E.E. (2001). Identifying characteristics that students, interns and residents look for in their role models. Medical Education, 35, 272-277. Epstein, R.M. (2007). Assessment in medical education. New England Journal of Medicine, 356, 387-396.

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Goldberg, L. (1993). The structure of phenotypic personality traits. American Psychologists, 48, 26-34. Good, T., & Brophy, J. (2000). Looking in classrooms (8th ed.). New York: Longman. Gottlieb, R., Rogers, J., & Rainey, K. (2002). Using curriculum evaluation as a basis for clinical faculty development. College Student Journal, 36(2), 279+. Retrieved January 29, 2008, from Questia database, http://www.questia.com/ PM.qst?a=o&d=5000799148 Irby, D.M. (1994). What clinical teachers in medicine need to know. Academic Medicine, 69, 333-342. Irby, D.M., Ramsey, P.G., Gillmore, G.M., & Schaad, D. (1991). Characteristics of effective clinical teachers of ambulatory care medicine. Academic Medicine, 66, 54-55. Ker, J.S. (2003). Attributes of trainers for postgraduate training in general surgery–a national consensus. Surgeon, 1, 215-20. Kvale, S. (1996). Interviews: An introduction to qualitative research interviewing. Thousand Oaks, CA: Sage. LeCompte, M.D., & Preissle, J. (1993). Ethnography and qualitative design in educational research (2nd ed.). San Diego, CA: Academic Press. Lempp, H., & Seale, C. (2004). The hidden curriculum in undergraduate medical education: Qualitative study of medical students’ perceptions of teaching. British Medical Journal, 320, 770-773. Mann, C., & Steward, F. (2000). Internet communication and qualitative research: A handbook for researching online. Thousand Oaks, CA: Sage. Menges, R., & Austin, A. (2001). Teaching in higher education. In V. Richardson(Ed.), Handbook of research on teaching (4th ed). Washington, DC: American Educational Research Association.

Morrison, E.H., Hitchcock, M.A., Harthill, M., Boker, J.R., & Masunaga, H. (2005). The online clinical teaching perception inventory: A “snapshot” of medical teachers. Family Medicine, 37, 48-53. Murray, J. P. (1999). Faculty development in a national sample of community colleges. Community College Review, 27(3), 47. Retrieved January 29, 2008, from Questia database: http://www.questia. com/PM.qst?a=o&d=5001886124 Oliver, J., & Srivastava, S. (1999). The Big Five taxonomy: History, measurement, and theoretical perspectives. In L. Pervin & O.P. John (Eds.), Handbook of personality: Theory and research (2nd ed.). New York: The Guilford Press. Oser, F., & Baeriswyl, F. (2001). Choreographies of teaching: Bridging instruction to learning. In V. Richardson (Ed.), Handbook of research on teaching (4th ed). Washington, DC: American Educational Research Association. Paukert, J.L., & Richards, B.F. (2000). How medical students and residents describe the roles and characteristics of their influential clinical teachers. Academic Medicine, 75, 843-845. Pinsky, L.E., Monson, D., & Irby, D.M. (1998). How excellent teachers are made: Reflecting on success to improve teaching. Advances in Health Science Education, 3, 207-215. Prichard, K. W., & Sawyer, R.M. (Eds.). (1994). Handbook of college teaching: Theory and applications. Westport, CT: Greenwood Press. Retrieved January 29, 2008, from Questia database: http://www.questia.com/PM.qst?a=o&d=9413558 Richardson, V. (2001). (Ed.). Handbook of research on teaching (4th ed). Washington, DC: American Educational Research Association. Sandelowski, M. (1995). Focus on qualitative methods: Sample size in qualitative research. Research in Nursing and Health, 18, 179-183.

159

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

Schafer, D.R., & Dillman, D.A. (1998). Development of a standard e-mail methodology: Results of an experiment. Public Opinion Quarterly, 62, 378-397. Searle, N.S., Hatem, C.J., Perkowski, L., & Wilkerson, L. (2006). Why invest in an educational fellowship program? Academic Medicine, 81, 936-40.

Seidel, J. (1991). Method and madness in the application of computer technology to qualitative data analysis. In N. Fielding & R.M. Lee (Eds.). Using computers in qualitative research (pp. 107-116). London: Sage. Wear, D., Kokinova, M., Keck-McNulty, C., & Aultman, J. (2005). Pimping: Perspectives of 4th year medical students. Teaching and Learning in Medicine, 17, 184-191. Weber, R.P. (1990). Basic content analysis (2nd ed.). Newbury Park, CA.

160

Characteristics of Good Clinical Educators from Medical Students’ Perspectives

appendIx: clInIcal teacHIng weB-Based QuestIonnaIre (aBBreVIated) clinical teaching General Instructions Please read the following instructions before completing the questionnaire. The purpose of this survey is to identify characteristics you believe are associated with effective clinical teaching. Participation is voluntary. Your responses will be kept anonymous. Filling out this survey implies consent to participate in this project. Use examples from the third and fourth years of medical school.

Clinical Teaching Instructions: The following open-ended questions include several subquestions. Please answer each of the subquestions to the best of your knowledge. In your opinion, what makes a good clinical teacher? Please list the 3 characteristics or qualities of good clinical teaching in the appropriate text boxes below. Please list only ONE characteristic in each of 3 cells. Below each characteristic, please also: A) nominate a faculty member from the third or fourth year of medical school who best exemplifies a positive example of that characteristic, B) write a short story that illustrates how that faculty member demonstrated that quality in his/her teaching activities, and C) write a short story that describes a situation in which any faculty member was a nonexample of that characteristic. Please do NOT list the name of the faculty member in part C. For these short stories, please write a minimum of 3-5 sentences, but feel free to write as much as the box will allow. Please report examples from the third and fourth year of medical school.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 2, edited by J. Tan, pp. 69-86, copyright 2008 by IGI Publishing (an imprint of IGI Global).

161

162

Chapter 12

Open Source Software: A Key Component of E-Health in Developing Nations David Parry Auckland University of Technology, New Zealand Emma Parry National Women’s Health, Auckland District Health Board, New Zealand Phurb Dorji Jigme Dorji Wanchuck National Referral Hospital, Bhutan Peter Stone University of Auckland, New Zealand

aBstract The global burden of disease falls most heavily on people in developing countries. Few resources for healthcare, geographical and infrastructure issues, lack of trained staff, language and cultural diversity and political instability all affect the ability of health providers to support effective and efficient healthcare. Health information systems are a key aspect of improving healthcare, but existing systems are often expensive and unsuitable. Open source software appears to be a promising avenue for quickly and cheaply introducing health information systems that are appropriate for developing nations. This paper describes some aspects of open-source e-health software that are particularly relevant to developing nations, issues and problems that may arise and suggests some future areas for research and action. Suggestions for critical success factors are included. Much of the discussion will be related to a case study of a training and E-health project, currently running in the Himalayan kingdom of Bhutan.

organIZatIon oF tHIs paper This paper is organised around a number of sections. The introduction outlines the rationale of the paper and deals with some aspects of Open DOI: 10.4018/978-1-61692-002-9.ch012

source software that make it attractive for software development in the health domain for low income countries. The methodology section then introduces the framework of assessment that is being used. The majority of this paper describes a case study of a project run by the authors in Bhutan in

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Open Source Software

the obstetric domain. Critical success factors for such a project are then analysed and some conclusions are drawn. The discussion covers some of the issues that have arisen from this experience, and articulates some lessons learned. This.

Figure 1. Research domains

IntroductIon This project deals with the intersection of a number of domains, as shown in Figure 1

e-Health E-health has become a popular term for the transformation of healthcare that has occurred through the use of electronic communications, in a conscious imitation of “ebusiness”. E-health encompasses more than the traditional electronic health record. It involves the use of information and communications technologies in the widest sense, including telemedicine, web-based health and mobile devices for healthcare. A definition has been proposed, after comprehensive analysis, in (Pagliari et al., 2005) “e-health is an emerging field of medical informatics, referring to the organization and delivery of health services and information using the Internet and related technologies. In a broader sense, the term characterizes not only a technical development, but also a new way of working, an attitude, and a commitment for networked, global thinking, to improve health care locally, regionally, and worldwide by using information and communication technology” This definition is actually adapted from a previous one in an editorial(Eysenbach, 2001). The globalised and networked aspects are particularly important in our case study – the emphasis is on communication and collaboration rather than distance

Health Information systems Health information systems (HIS) often have three main objectives, to improve patient care, improve management and form part of a quality improvement programme. However, these objectives – as described by (Littlejohns, Wyatt, & Garvican, 2003) are not always achieved. As part of a HIS implementation there are often major changes to workflow and practice, large expenditures on hardware including computing and communications, and system integration, as well as software development, training and implementation. (Littlejohns et al., 2003) Points out that failures occur in HIS development – often due to a lack of understanding of the complexity of the project. Interestingly OSS appears to answer some of these issues by providing more stable – if less feature-rich – software and providing a generally larger pool of developers and users than for proprietary software.

open source software Open source software (OSS) has gained very wide acceptance particularly in the web server community. Projects such as Apache (Mockus, Fielding, & Herbsleb, 2000) have involved large scale participation, and dominant market share.. In the healthcare domain, Sourceforge.

163

Open Source Software

net lists 58 applications for download. Many of the applications are extremely specialized, but the other hand, some like WIRM (Jakobovits, Rosse, & Brinkley, 2002) are effectively complete development environments. This paper will argue that successful development and use of OSS in healthcare requires a number of critical success factors, and that these reflect both the nature of OSS projects in the wider world and particular aspects relevant to healthcare. OSS can be seen as part of a wider movement that has been characterized as innovation from the user community (Hippel, 2001). This emphasizes the point that that OSS is not just “free” but also is able to be modified by the community that uses it. There is a very large and continuing project (OpenEHR) Kalra, 2005 #3034, that is attempting a to produce reliable, semantically correct representations of health information. This project is not really focussed on developing specific implementations, but rather a common means of representation.

developing nations and the case of Bhutan Health information systems are important for developing nations as well as industrialized ones. A large review of the use of information technology in primary care in developing countries (Tomasi, 2004) identified five main areas of application – data processing in the health care system, decision support, electronic data transmission, electronic patient records and telemedicine. Many developing countries have low levels of trained clinical staff, and this can increase the load on secondary and tertiary providers. In order to audit their performance, and increase efficiency, electronic records and workflow systems can reduce the workload on the staff available. Both of these aspects are particularly important for developing nations for a number of reasons. •

164

Developing nations have extremely limited health budgets, but the burden of dis-

•

•

•

ease on individual households can be very large. For example up to 100% of household income being spent on end-stage care for AIDS patients in some nations in Africa(Russell, 2004). Developing nations often have a diverse mixture of groups within them, and it cannot be assumed that all citizens have a common language. Even when a common language exists, it may be spoken by a relatively small percentage of the world’s population, and commercial development of software using that language may not be feasible. Infrastructure and resource constraints in particular for network connectivity may reduce the utility of high-performance systems routinely used in the west. For example PACS systems involving transmission of large images via network connections may not be practical, but memory-stick based approaches may be feasible (Parry, Sood, & Parry, 2006). Open source approaches allow the development of expertise in multiple sites away from large commercial organizations. Therefore they can encourage the upskilling of software developers in smaller centres. This expertise can be applied to the localization of standard packages and the development of a solid base for software support. In this aspect both the development and the use of open source software can be beneficial in the education sector. Developing nations often have large and increasing numbers of young, educated people available for project development. OSS tools are attractive for teaching information systems development because of cost, wide availability of documentation and localized versions being available. For example, linux is available in x language versions including dzonga – the national language of Bhutan.

Open Source Software

•

•

•

•

•

Commercial software suppliers may be reluctant to sell advanced software packages in developing nations because of the difficulty of arranging support and the perceived threat of piracy Developing nation’s health systems often have a complex collection of groups working within them including governments, commercial organizations, local and international charities and international official organizations. The requirement for reporting and data analysis may well be more complex than in industrialised nations. Infrastructure developed to support callcentre development or tourism, including internet and telecommunications technology, is easily adapted to allow links between nations. Because OSS tools are supported via the web, this approach avoid the reliance on expensive and out-of-date paper manuals and development kits. OSS’s licensing structure allows cross-national projects to be completed much more easily. Mobile devices have particular promise for e-health in developing nations(Iluyemi, Fitch, & Parry, 2007). Mobile OSS development is a particularly active area of research(Raento, Oulasvirta, Petit, & Toivonen, 2005). An exhaustive survey of African nations e-health status(Kirigia, Seddoh, Gatwiri, Muthuri, & Seddoh, 2005) has shown that many nations currently have very low preparedness, given the relative paucity of internet and telephone connections, however recent developments in increasing bandwidth capacity see for example http:// www.fibreforafrica.net/ may be improving this state of affairs.

The case study in this paper deals with the intersection of a number of research domains, (Figure 1), which means that the choice of methodology for analysis may be challenging. There have been

some recent papers on the use of OSS in health information systems, focused mainly on developed country applications,(Kantor, Wilson, & Midgley, 2003; McDonald et al., 2003). Interestingly these papers point out that the use of OSS in healthcare is not new and that although perhaps small-scale this work has been consistently ongoing. However, these papers do emphasize the potential gains to be had by the use of OSS in healthcare both in terms of health providers and also developers. Because of the widely varying state of communications infrastructure in developing countries, western models of development which emphasize in-hospital systems linked by fixed line high capacity networks may not be appropriate. In the context of less developed countries, there have been a number of telemedicine projects, often concerned with communication from centres of excellence in western nations eg (Swinfen, Swinfen, Youngberry, & Wootton, 2005), or within developing countries (Deodhar, 2002), but a shared approach to development is vital (Wooton, 2001).

MetHodology In order to analyses a case study, some sort of framework of analysis should be adopted. The development project was actually quite complex with elements of telemedicine, knowledge management and information processing included in the overall design (see Figure 2). A survey of Telemedicine projects in India (Pal, Mbarika, Cobb-Payton, Datta, & McCoy, 2005) identified six critical success factors for Telemedicine success, these were used as practical and simple measures that could be applied in this complex if small-scale project. The success factors identified were:: 1. 2. 3.

Set clear program objectives Garner Government Support Adapt User-Friendly Interfaces

165

Open Source Software

Figure 2. Overall system plan

4. 5. 6.

Determine Accessibility Via Telecommunications and Internet Access Implement Standards and protocols Measure Cost-effectiveness and User Satisfaction

The case study will deal with these areas, although the project is wider than a simple telemedicine project as it includes database development and integration with the audit system, along with web-based protocols.

case report e-HealtH support For oBstetrIc serVIces In BHutan Bhutan is a small Buddhist Kingdom located in the Himalayas, with a population of under 700,000. Land transport is extremely slow because of the geography - for example it takes 3 days to travel across the country, a distance of around 300km. There is only one airport and no facility for helicopter transport. Seventy percent of the population live in rural areas with 30% more than 1 hours

166

walk from the nearest road head. Bhutan has had major successes in increasing life expectancy and improving health care but avoidable neonatal and maternal mortality and morbidity remains an issue. Current figures for Bhutan suggest an infant mortality rate of 67/1000 and a maternal mortality rate of 4.2/1000 – compare with New Zealand’s rate of 4/1000 and 0.07/1000 respectively (World Health Organization, 2006a). Large numbers of preventable neonatal deaths continue to occur in the less developed counties. However, recent work has suggested (Darmstadt et al., 2005) that evidence based interventions in antenatal and intrapartum care could reduce these rates by between 37% and 67%. These interventions are not complex and are relatively inexpensive. The overarching imperative is to ensure appropriate care for pregnant women that involves patient education and cooperation with antenatal and intrapartum services. Although there have been many studies on the use of e-health in obstetrics and perinatology and in less developed nations (Deodhar, 2002)there remains relatively little work on the evaluation of these systems, especially in terms of outcome and integration

Open Source Software

of these systems into existing structures, and the changes that occur because of their introduction, although the 1:45 cost benefit ratio quoted in this study is impressive. The World Health Organisation has been running a “Making Pregnancy Safer” Initiative (World Health Organization) in order to reduce the level of neonatal and maternal mortality. Previous work in Bhutan had developed a protocol book for emergency obstetric care (EMOC). Other countries using EMOC have recorded improvement in outcomes, for example Bangladesh (Islam MT, 2006) and Peru (Kayango et al., 2006).One of the major lessons learned in these trials was that a record of outcomes via a perinatal database, and the wide dissemination of protocols, for example those that identify potentially high-risk patients, are vital for success. Surprisingly perhaps, the identification of appropriate procedures for dealing with high-risk patients has been shown to be effective in reducing the demand for interventions (Islam MT, 2006).

the Bhutanese Health system Healthcare is free in Bhutan, and is delivered via a tertiary structure. The primary healthcare unit is the “Basic Health Unit” (BHU), of which there are around 170 around the country. These units do not always have medically qualified staff, but they run outreach and clinic services and usually a number of beds are available. Delivery services are sometimes available run by nurses. District health units (around 30) will have at least one generalist medical officer, some of these units have the capacity to perform caesarean sections and ultrasound scanning. There are 3 Referral hospitals in the country which have at least one obstetrician and theatre services. The Jigme Dorji Wanchuck National Referral Hospital (JDWNRH) in the captial has four obstetricians and is the tertiary referral unit for the whole country. Because there is no general practitioner service, patients have the opportunity to refer themselves directly

to hospital-based consultants even in cases where primary care would be more appropriate. This and the paucity of qualified obstetricians, results in a large workload and the Obstetricians are busy and often difficult to contact for advice. A current project is running to introduce the emergency obstetric care (EMOC) system to Bhutan, and the protocols are being integrated with these to provide seamless care.

project description The aim of the project was to collaboratively develop a number of treatment and/or diagnosis protocols to allow the clinical staff to apply appropriate evidence-based care for the major problems that would be dealt with by a perinatal service. The role of the Perinatal service is described in (Mascarenhas, Eliot, & MacKenzie, 1992), essentially it provides care for mothers and baby between conception and birth, and aims to reduce the risks to mother and baby in this process, by appropriate intervention and monitoring In addition to the staff applying the protocols in practice, the aim is to raise awareness of the issues that affect perinatal outcomes amongst others, for example referring clinicians. The development of a collegial editing and review process involving Clinical staff in Bhutan and New Zealand was also seen as a vital part of the project. The project also included the development of a perinatal web-based database to allow for more effective management of the service on a day-to-day basis and also allow for analysis of clinical performance. OSS occurs in a number of places in the system. The perinatal database is written in PHP with a mySQL database engine. Web page development was done using open source tools, as was the xml protocol development. However Microsoft products were used for the operating systems and web server software in Bhutan, along with the Linux/Apache setup for the webserver in New Zealand.. In addition, standard MSOffice products were used to develop the protocols.

167

Open Source Software

reVIew oF success Factors

adapt user-Friendly Interfaces

clear objectives

The user interface adopted was a standard webbrowser, whatever the source of the data – even locally stored protocols would display in a browser. The native protocols were stored as XML documents, which were then displayed in a humanreadable format via a web browser. XML was chosen for ease of updating – in that the editing process could alter content without a great deal of formatting issues and with the awareness that other display methods such as voice responses or mobile devices may be used in the future. As an open standard, XML is very well suited to this approach. The XML design is intended to be expandable and able to represent both diagnostic and therapeutic protocols. A fragment of the XML representation is shown in Table 1. The initial outline was based on the PubMED schema, but simplified to remove excessive bibliographic elements. The XML documents identify the responses to particular diagnoses or symptoms which would be expected to be encountered commonly. The aim is to allow clinical staff - who may be at a remote site – to identify what emergency care is needed, whether the patient needs to be referred or transferred and the degree of urgency of that referral. Also the protocol can identify what additional tests or procedures need to be performed.

The objectives of the project were identified in initial discussions and codified in the agreement signed between the stakeholders. The objectives included the development of a perinatal medicine service, continuing support for this service and standardisation of treatment based on the best possible evidence. An additional objective was sustainability of the service, OSS supported this by allowing low or no cost technical documentation and development tools to be made available to local staff.

government support The Royal Government of Bhutan (RGOB) is the sole supplier of healthcare in the kingdom. The RGOB runs a series of five-year plans which identify objectives and priorities as well as sources of funding. Plans developed by overseas providers are examined and extensive negotiation takes place to ensure that the country receives appropriate and sustainable help that is consistent with the RGOB objectives. This process began in the case of this project, two years before the initiation, when representatives of the funders – The Magee Family – met with other stakeholders including government representatives, clinical staff from New Zealand and Bhutan associated with the project, and UNICEF. This resulted in a project agreement that was signed off in a formal ceremony. The project composed a number of other elements including funding for hardware and training of clinical workers in the perinatal medicine area. Continuing involvement of the stakeholders has been a great asset to the project. RGOB department of IT has been running a longterm project to support OSS and is getting closer to the development of a policy on its use (Bhutan Department of Information Technology, 2007)

168

determine accessibility Although land transport is difficult, Bhutan is in the process of increasing the availability of internet access. Apart from dial-up connections there are microwave links and recently the International Telecommunications Union (ITU) e-post initiative(International Telecommunications Union, 2006), has recently been launched in Bhutan using very small aperture (VSAT) satellite ground stations for rural access to electronic communications, and this may be useful for rollout to

Open Source Software

Table 1. XML Protocol Fragment

Cord prolapse

The cord that normally presents itself is within intact membranes. When the mem-

branes rupture, the cord prolapses. This is an emergency as cord compression and/ or occlusion can cause fetal asphyxia.

Rupture of Membrane Prolapse



Palpable cord on vaginal exam

Observed cord protruding onto vulva





remote areas. OSS tools are often very efficient in terms of file size, and machine footprint (use of processor time and memory) so they can be used in a wider range of scenarios than might be possible for the latest proprietary operating systems or applications.

Implement standards The protocols themselves were developed in a standard format (Table 2), and as seen above, implemented using XML. In addition to this an attempt was made to standardize the production of the protocols, so that candidate protocols from other sources would go through an editorial process and be routinely revised. This process has been followed by paper-based systems previously, but electronic approaches allow instantaneous updating of the live protocol without fear of version control issues and also allow a trail to be kept of previous versions that can be linked to any events linked historically to the implementation. Recent work (Yellowlees, Marks, Hogarth, & Turner, 2008) has reiterated the importance of

standards in health IT and the importance of standards in the open source environment.

Table 2. Major Elements of the Protocol Document Element

Comment

Name

Name of protocol

Definition Keywords

Used for searching

Diagnosis

For diagnostic protocols

Diagnostic Step Procedure

For procedural protocols

Procedure Step Audience

Intended user, includes country and location

Evidence

A small selection of the supporting evidence

Author

Multiple Authors Possible

Last Update Review Date

169

Open Source Software

Measure cost-effectiveness and user satisfaction This aspect is perhaps the most difficult part of the project. As part of the process protocols will be regularly reviewed by stakeholders. In addition, a Perinatal database is being implemented in order to record outcomes, and assess performance against that expected in the protocol – in particular areas where the protocols are not being followed, and whether the protocols or behavior or both should be modified. It is hoped that improvements in the mortality and morbidity figures will also be noticeable. Finally a rise in awareness of the general maternity service and increased access to it by women – only half of whom currently have an attended birth – will accompany improved outcomes among those who have contact with the Perinatal service. Current work is focused on examining the effect of improved ultrasound training and outreach clinics.

lessons learned Integration of protocols from diverse sources was one of the major challenges facing the team. Protocols were sourced from National Women’s Hospital Auckland, the World Health Organisation and EmOC protocols in Bhutan. Collaborative review of protocols was extremely important, as buy-in from clinical staff is vital. However the process of maintaining a common electronic repository was technically difficult as each of the reviewers tended to work asynchronously using paper copies. The final approach used was to produce paper prototypes and distribute them, collect back annotated versions and then combine them in a final word document. This was then converted to XML. Development of the Perinatal database was restricted by the very small numbers of users available to test and comment on the system, and a wide user community,

170

which may only be available online, may well increase the quality and speed of development..

dIscussIon and Future worK There are some general issues that affect eHealth initiatives, and the use of OSS in the developing world, in particular connectivity, computing resources and skills.

connectivity Less developed nations have generally much lower availability of fixed telephone lines. In addition, geographic, economic and governmental issues often conspire to make conventional dial-up access less common than in western countries. However, wireless and satellite solutions such as VSATS including the International Telecommunications Union (ITU) e-post initiative (International Telecommunications Union, 2006) are overcoming these issues. It is important to recognize that not every nation’s infrastructure is developing in the same way, and many nations may leapfrog to wireless solutions without the use of landline-based solutions. However high bandwidth solutions may not be appropriate for developing countries. One of the most successful e-health projects has been the Swinfen Project – currently expanding in Iraq (Swinfen et al., 2005). This project uses e-mail in a store-and forward model, between clinicians in various countries. The prospects of advanced telepresence approaches being effective in routine care seem slight because of issues concerning quality of service, bandwidth and reliability of connection. Even though the “trauma pod” and other projects financed by the US Department of Defence are beginning to show results (Romano, Lam, Moses, Gilbert, & Marchessault, 2006), costs are likely to render this approach problematic in other contexts.

Open Source Software

computing Devices such as the Simputer (The Simputer Trust, 2000) and the sub $100 laptop (OLPC, 2006), promise much cheaper access to computing power. It should be emphasized that for e-health applications, the computing device can be fairly simple, indeed mobile devices may become the preferred means of access. Along with cost, the ability to survive rough treatment, extremes of temperature and humidity and long battery life – or even the use of clockwork power in the case of the sub $100 laptop – are more important in developing countries than in Organization for economic cooperation and development (OECD) member countries. Parts supply and transport cost can make the repair of computers extremely expensive. However organizations such as Global Assistance for Medical Equipment (GAME) (http://www.global-medical-equipment.org/ whatwedo.html) have established links between professional organizations in the developed and less-developed world. These approaches move beyond the shipping of obsolescent equipment, to an integrated and well thought and sustainable out collaboration between donors and recipients.

skills and Information At present consumer e-health is of limited usefulness in the developing world. Low levels of literacy and information literacy cause difficulties. However the fact that the vast majority of web resources are written in English, and are US-centric in terms of organization of healthcare, availability of drugs and medical devices and naming make even materials designed for health consumers in the OECD countries less useful for those in other nations. However, these issues are much less important when the provision of e-health services for medical professionals is considered. Adapting general principles to specific cases is a key skill of medical professionals. Indeed the traffic is not all one-way, less developed nation professionals

often have skills that are no longer available in more developed nations. Collaboration in training of medical professionals, where trainees from different nations are exchanged, can improve the training in both systems. This can be supported by the use of e-health tools such as websites, e-mail and instant messaging. Other skills required include the support of the e-health infrastructure in terms of technical support for computing devices and connectivity. Fortunately the requirement of tourists from western countries for internet connectivity wherever they are, along with the burgeoning industries of call centres and ‘off-shoring’ of software development are providing a strong push for training in these areas. OSS use in education and training allows nations with limited resources to devote more funding to the human side of education, as well as allowing projects that involve software localization to advance quickly. Open-source clinical protocols may become important repositories of clinical knowledge allowing rapid development and input from experience, especially is based on standard electronic forms. Another important aspect of skill transfer and collaboration is the use of early warning networks for disease surveillance such as the Global Outbreak Alert and Response Network (GORAN) that played a very large part in the early detection of SARS (Heymann & Rodier, 2004). Such networks link health workers throughout the world and the transfer of information is by no means one-way. There remains a dearth of well-controlled studies of e-health initiatives in developing nations, but the need for effective collaboration remains paramount (Wooton, 2001). However there are a number of pointers to success; 1.

The e-health system must be compatible with existing organizational and cultural structures. Some “western” assumptions do not apply in less developed nations and vice versa. For example routine ultrasound

171

Open Source Software

2.

3.

4.

examination in early pregnancy has not been shown to be effective in reducing mortality in a Cochrane review(Neilson, 1998). However an environment where mortality due to unsuspected problems is much greater, and the availability of on-demand scans is lower, may give different results. Collaboration and training between the professionals involved is vital. This applies to both clinical and technical staff. This may in fact be the area of greatest benefit. Ingenuity is more important than technology. Store-and-forward email may be of greater utility than telepresence. Open-source technology is particularly suited to this area of work. Lower costs, availability of technical skills, greater range of customized languages and often lower technology requirements make Open Source approaches and especially web-based Opensource tools particularly attractive.

Future work in this area will include greater use of multicentre collaboration, both within existing networks such as GORAN and GAME and outside them. Lower bandwidth costs, and easier access to high bandwidths will enable richer media to be used, such as telesonography via store and forward (Parry et al., 2006).Common health problems are starting to afflict North and South – ageing populations, the rapid spread of new infectious diseases and chronic conditions. Common approaches to these issues – including the use of low-cost assistive technology, and offshoring of medical procedures such as radiology (Larson & Janower, 2005) may be controversial, but at least the discussion has started. There are enormous potential benefits in the development of e-health in collaboration with developing nations, and the benefit to the people of world may be immense. An additional benefit of the open source approach may be an increased ability for IT specialists in developed nations to assist people around the world. As virtually every nation now

172

has a web presence, the technical barriers to such collaboration are much lower than they were even ten years ago. It is hoped that further work will refine the system sufficiently to allow the software to be placed in a repository such as Sourceforge. net. Furthermore, it is hoped that such an approach will encourage increasing collaboration and development in this area.

acKnowledgMent This project work would not have been possible without the generosity of the Magee family and the hard work of the representatives of UNICEF and the Royal Government of Bhutan. Staff of Jigme Dorji Wanchuck National Referral Hospital, Thimphu assisted in a very wide range of roles and continue to work on this project

reFerences Bhutan Department of Information Technology. (2007). Bhutan’s Journey Towards Open Source. Paper presented at the DebConf7, Edinburgh, Scotland. https://penta.debconf.org/~joerg/ events/25.en.html Darmstadt, G., Bhutta, Z., Cousens, S., Adam, T., Walker, N., & Berni, L. d. (2005). Evidence-based, cost-effective interventions: how many newborn babies can we save? Lancet, 365(9463), 977–988. doi:10.1016/S0140-6736(05)71088-6 Deodhar, J. (2002). Telemedicine by email-experience in neonatal care at a primary care facility in rural India. Journal of Telemedicine and Telecare, 8(Suppl 2), 20–21. doi:10.1258/135763302320301867 Eysenbach, G. (2001). What is e-health? Journal of Medical Internet Research, 3(2), e20. doi:10.2196/ jmir.3.2.e20

Open Source Software

Heymann, D. L., & Rodier, G. (2004). Global Surveillance, National Surveillance, and SARS. Emerging Infectious Diseases Journal, 10(2). Retrieved 1st December 2006, from http://origin. cdc.gov/ncidod/EID/vol10no2/03-1038.htm

Kirigia, J., Seddoh, A., Gatwiri, D., Muthuri, L., & Seddoh, J. (2005). E-health: Determinants, opportunities, challenges and the way forward for countries in the WHO African Region. BMC Public Health, 5(1), 137. doi:10.1186/1471-2458-5-137

Hippel, E. v. (2001). Innovation by User Communities: Learning from Open-Source Software. MIT Sloan Management Review, 42(4), 82.

Larson, P. A., & Janower, M. L. (2005). The Nighthawk: Bird of Paradise or Albatross? Journal of the American College of Radiology, 2(12), 967–970. doi:10.1016/j.jacr.2005.08.002

Iluyemi, A., Fitch, C. J., & Parry, D. T. (2007, April 18-20th 2007). Mobile and Wireless Technologies for e-Health Provision in Africa: Opportunities for GSM-Based Patient Centric Services. Paper presented at the Med-e-Tel Luxembourg International Telecommunications Union. (2006). Bhutan to be Testbed for ITU’s E-Post Venture with Universal Postal Union. Retrieved 20th April 2006, from http://www.itu.int/newsarchive/ press_releases/2002/10.html Islam, MT, H. Y., Waxman R, Bhuiyan AB. (2006). Implementation of emergency obstetric care training in Bangladesh: lessons learned. Reproductive Health Matters, 14(27), 61–72. doi:10.1016/ S0968-8080(06)27229-X Jakobovits, R. M., Rosse, C., & Brinkley, J. F. (2002). WIRM An Open Source Toolkit for Building Biomedical Web Applications. Journal of the American Medical Informatics Association, 9(6), 557–570. doi:10.1197/jamia.M1138 Kantor, G. S., Wilson, W. D., & Midgley, A. (2003). Open-source software and the primary care EMR. Journal of the American Medical Informatics Association, 10(6), 616. doi:10.1197/jamia.M1403 Kayango, M., Esquiche, E., Luna, M. R., Frias, G., Vega-Centeno, L., & Bailey, P. (2006). Strengthening emergency obstetric care in Ayacucho, Peru. International Journal of Gynaecology and Obstetrics: the Official Organ of the International Federation of Gynaecology and Obstetrics, 92, 299–307. doi:10.1016/j.ijgo.2005.12.005

Littlejohns, P., Wyatt, J. C., & Garvican, L. (2003). Evaluating computerised health information systems: hard lessons still to be learnt. British Medical Journal, 326(7394), 860–863. doi:10.1136/ bmj.326.7394.860 Mascarenhas, L., Eliot, B. W., & MacKenzie, I. Z. (1992). A comparison of perinatal outcome, antenatal and intrapartum care between England and Wales, and France. British Journal of Obstetrics and Gynaecology, 99(12), 955–958. McDonald, C. J., Schadow, G., Barnes, M., Dexter, P., Overhage, J. M., & Mamlin, B. (2003). Open Source software in medical informatics--why, how and what. International Journal of Medical Informatics, 69(2-3), 175–184. doi:10.1016/ S1386-5056(02)00104-1 Mockus, A., Fielding, R. T., & Herbsleb, J. (2000). A case study of open source software development: the Apache server. Paper presented at the Software Engineering, 2000. Proceedings of the 2000 International Conference on. Neilson, J. P. (1998). Ultrasound for fetal assessment in early pregnancy. Cochrane Database of Systematic Reviews, 4. OLPC. (2006). One laptop per Child. Retrieved 1st December 2006, 2006, from http://laptop.org/ Pagliari, C., Sloan, D., Gregor, P., Sullivan, F., Detmer, D., & Kahan, P. J. (2005). What Is eHealth (4): A Scoping Exercise to Map the Field. Journal of Medical Internet Research, 7(1), e9. doi:10.2196/jmir.7.1.e9

173

Open Source Software

Pal, A., Mbarika, V. W. A., Cobb-Payton, F., Datta, P., & McCoy, S. (2005). Telemedicine diffusion in a developing Country:The case of India (march 2004). Information Technology in Biomedicine. IEEE Transactions on, 9(1), 59–65. Parry, E. C., Sood, R., & Parry, D. T. (2006). Investigation of optimization techniques to prepare ultrasound images for electronic transfer [Abstract]. Ultrasound in Obstetrics and Gynecology, 28(4), 487-488(482). Raento, M., Oulasvirta, A., Petit, R., & Toivonen, H. (2005). ContextPhone: A Prototyping Platform for Context-Aware Mobile Applications, 4, 51-59. Romano, J. A., Lam, D. M., Moses, G. R., Gilbert, G. R., & Marchessault, R. (2006). The Future of Military Medicine Has Not Arrived Yet, but We Can See It from Here. Telemedicine and e-Health, 12(4), 417-425. Russell, S. (2004). The economic burden of illness for households in developing countries: a review of studies focusing on malaria, tuberculosis, and human immunodeficiency virus/acquired immunodeficiency syndrome. Am J Trop Med Hyg, 71(2_suppl), 147-155. Swinfen, P., Swinfen, R., Youngberry, K., & Wootton, R. (2005). Low-cost telemedicine in Iraq: an analysis of referrals in the first 15 months. Telemedicine and Telecare, 11(Suppl 2), 113.

174

The Simputer Trust. (2000). Simputer(TM) - Welcome. Retrieved December 1, 2006, from http:// www.simputer.org/ Tomasi, E. F., Luiz Augusto and Maia, Maria de Fatima Santos. (2004). Health information technology in primary health care in developing countries: a literature review. Bulletin of the World Health Organization, 82(11), 867–874. Wooton, R. (2001). Telemedicine and developing countries–successful implementation will require a shared approach. Telemedicine and Telecare, 7(Suppl 1), 1–6. doi:10.1258/1357633011936589 World Health Organization. (2006a). Estimates of child and adult mortality and life expectancy at birth by country. Retrieved January 17, 2007, from http://www.who.int/healthinfo/statistics/ mortlifeexpectancy/en/index.html World Health Organization. (2006b, 6th June 2006). Introduction to the ‘Making Pregnancy Safer’ initiative. Retrieved October 1, 2007, from http://w3.whosea.org/pregnancy/introf.htm Yellowlees, P. M., Marks, S. L., Hogarth, M., & Turner, S. (2008). Standards-Based, Open-Source Electronic Health Record Systems: A Desirable Future for the U.S. Health Industry. Telemedicine and e-Health, 14(3), 284-288.

175

Chapter 13

An Empirical Investigation into the Adoption of Open Source Software in Hospitals Gilberto Munoz-Cornejo University of Maryland Baltimore County, USA Carolyn B. Seaman University of Maryland Baltimore County, USA A. Güneş Koru University of Maryland Baltimore County, USA

aBstract Open source software (OSS) has gained considerable attention recently in healthcare. Yet, how and why OSS is being adopted within hospitals in particular remains a poorly understood issue. This research attempts to further this understanding. A mixed-method research approach was used to explore the extent of OSS adoption in hospitals as well as the factors facilitating and inhibiting adoption. The findings suggest a very limited adoption of OSS in hospitals. Hospitals tend to adopt general-purpose instead of domain-specific OSS. We found that software vendors are the critical factor facilitating the adoption of OSS in hospitals. Conversely, lack of in-house development as well as a perceived lack of security, quality, and accountability of OSS products were factors inhibiting adoption. An empirical model is presented to illustrate the factors facilitating and inhibiting the adoption of OSS in hospitals.

IntroductIon The open source software (OSS) phenomenon has become an important area of interest in information systems research due in part to the

large and fast-growing number of OSS users and software products in a large variety of domains. OSS is already being adopted and used as a software platform in a number of fields other than healthcare (Dedrick & West, 2003; 2004; Norris,

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

2004; Waring & Maddocks, 2005), and it has the potential to be equally promising for the hospital industry (Fitzgerald & Kenny, 2004). Studying OSS adoption in any domain can help reveal patterns and phenomena that are applicable to adoption in general, in addition to revealing insights into the domain being studied. In particular, the adoption and use of OSS in a hospital context remains a poorly understood phenomenon; only a handful of researchers have addressed the factors inhibiting or facilitating such adoption. Such an understanding is important in helping hospitals make better decisions about whether and how adoption of OSS could benefit them. The first step in developing a better understanding is to explore the current state of OSS adoption, and the factors inhibiting and influencing it in hospitals. Such an exploration is the goal of this study. Once this current state is well described, it will be possible to seek answers to higher-level questions about the pros and cons, the costs and benefits, the advantages and disadvantages of OSS adoption in this domain, which is the second goal. Therefore, the present study is of considerable interest for both practitioners and researchers. It will provide hospitals and healthcare organizations that are considering the adoption of OSS technologies with an understanding of how technological, environmental and organizational factors affect the adoption process. This way hospital IT practitioners, or others attempting to introduce OSS technology into hospitals, can prepare against the expected barriers and can utilize the facilitators for successful adoption. This research also provides scholars with an empirical model for better understanding facilitating and inhibiting factors, as well as providing the foundations for further research that may validate and expand on the empirical model in other healthcare organizations and other domains. The main objective of this investigation was to explore and analyze the extent of OSS adoption in hospitals, along with the factors influencing or inhibiting this adoption process. Hospital IT

176

managers were chosen to represent the hospitals’ perspective on this topic. The following three questions guided this investigation: 1. 2. 3.

What are the types and names of OSS products that hospitals choose to adopt? What is the extent of OSS adoption for these products in hospitals? What are the factors facilitating and inhibiting the adoption of OSS in hospitals?

To research these questions, a survey and interviews were used to acquire both breadth and depth of understanding. The purpose of the survey was to answer the first two questions—to explore and characterize the types of OSS products adopted in hospitals and to discover the extent to which these products have been adopted. The interviews were used to answer question three to attain a deeper understanding of the factors that are facilitating and inhibiting the adoption of OSS in hospitals. In the following sections of this article, we first present the related work in this area. Then, we introduce the methodology for our survey and interview studies. After that, we present our data analysis and results. Then, we introduce our empirical model of the adoption of OSS in hospitals. Finally, we present our conclusions and the implications of our work.

lIterature reVIew open source software adoption in Healthcare Over the past few years, a small number of researchers have focused on the study of the potential advantages and risks of adopting and implementing OSS in the healthcare domain. Prior research encouraged the adoption and use of OSS in healthcare organizations because of OSS’s potential to both enhance healthcare delivery and lower software acquisition costs (Carnall, 2000;

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Kantor, Wilson, & Midgley, 2003; McDonald et al., 2003; Valdes, Kibbe, Tolleson, Kunik, & Petersen, 2004). OSS could potentially be more reliable and secure than proprietary software because its source code can be inspected and reviewed (Carnall, 2000). Past research introduced and extended the idea of OSS as a software development model that could definitively improve clinical and research software in the field of medical informatics (Yackel, 2001). A paper by Kantor, Wilson, and Midgley (2003) also presents the potential benefits that OSS could provide in the area of primary care. Kantor et al., also proposed that the adoption of OSS would reduce the excessive costs, the frequent turnover of vendors, and the lack of common data standards that are afflicting electronic medical records (EMR) systems in primary care. More recently, McDonald, Schadow, Barnes, et al. (2003) also investigated the potential role that the OSS model of software development may have in the medical informatics area. They also described a number of OSS products that have been used in the medical informatics domain over the years, including: OpenEMed, a patient record system; OSCAR, a family practice office management and medical record system; as well as the internationally well-known VistA system, a computer-base patient records system (CBPR) developed in MUMPS (Massachusetts General hospital Utility Multi-Programming System) by the U.S. Department of Veteran’s Affairs (Brown, Lincoln, Groen, & Kolodner, 2003; Longman, 2007). A more recent study by Valdes et al. (2004) also pointed out that OSS could be an effective solution for the problems that distress the healthcare industry such as high costs, business failures and barriers of standardization (Valdes et al., 2004). Other papers by Erickson, Langer, and Nagy (2005), Scarsbrook (2007) and Nagy (2007) supported the growth and adoption of OSS in radiology because OSS may significantly lower the entry cost for standards-compliant practices

in the healthcare industry. They also proposed that OSS might allow rapid scientific advancement due to the sharing of information and software (Erickson et al., 2005; Scarsbrook, 2007). Other authors such as DeLano (2005) presented some reasons for the potential success of OSS predicting that the pharmaceutical research and development process may benefit from the OSS development model.

open source software adoption in Hospitals A case study of OSS adoption was conducted at the Beaumont hospital in Ireland, where the IT department, under limited financial resources, made the decision to adopt OSS. Several OSS products were adopted and implemented successfully. The authors reported that there were important initial start-up and future operational costs when OSS products were preferred in the hospital (Fitzgerald & Kenny, 2004). Another study by Glynn, Fitzgerald and Exton (2005) investigated the commercial adoption of OSS using an innovation adoption theory framework based on Tornatzky and Fleischer’s (1990) model. They derived a framework that was then used to investigate the adoption process of OSS in the case of the Beaumont hospital (Fitzgerald & Kenny, 2004). The OSS products and processes were also seen as promising in terms of enabling rapid evolution and proliferation of applications in the medical domain through their use of open standards and higher degrees of interoperability (Raghupathi & Gao, 2007). The authors argued that the development processes in the Eclipse project (http:// eclipse.org) could improve scalability, prevent vendor lock-ins, and reduce costs in the medical information systems including electronic health record and clinical decision support systems. There are some recent studies focusing only on the managerial and technical barriers to the adoption of OSS (Holck, Larsen, & Pedersen,

177

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

2005). Past research on OSS and healthcare also proposed that OSS would reduce the number of bugs and failures in medical systems, as well as reduce their overall cost (Yackel, 2001). A study by Hogarth and Turner (2005) focused on creating a catalogue of existing OSS clinical projects and on determining metrics for their viability. The authors mentioned that many of the factors that are required to make a “successful and vibrant” OSS community within the mainstream software applications systems (e.g., Linux, Apache, etc.) may not necessarily be applicable to the clinical software applications systems. Another study by Kantor, Wilson and Midgley (2003) presented a set of potential advantages that the adoption of OSS may provide with regards to lowering the resistance of hospitals to the adoption of electronic medical records (EMR). These included: 1) the potential of OSS to reduce EMR ownership and software development costs, 2) the removal of vendor lock-in, and 3) the adherence of OSS to standards for the compatibility and data interchange among systems. In another study by Valdes, Kibbe, Tolleson, et al. (2004) dealing with the barriers to the proliferation of electronic health records/electronic medical records (EHR/EMR), the authors concluded that OSS is a viable solution to the barriers of high cost, business failure and standardization that the healthcare industry is facing when adopting EHR/EMR. The authors mentioned that, for example, interconnectivity problems are more easily solved when using OSS, since no technical information can be hidden. They also added that OSS can help alleviate the high costs associated with the adoption and implementation of EHR/EMR (Valdes et al., 2004). Although this article presents a good case for the adoption of OSS solving the barriers that EHR/EMR is facing, the authors do not support their case with empirical data. In summary, even though we have witnessed a widespread, significant OSS research and industry adoption of OSS, there are still few studies on OSS adoption and use, especially in the hospital

178

industry. Only a handful of researchers have addressed the factors inhibiting or facilitating OSS adoption in hospitals (Carnall, 2000; Kantor et al., 2003; Valdes et al., 2004; Glynn, Fitzgerald & Exton, 2005). Each of the aforementioned studies in this section found that top management support, limited financial resources, past experiences using OSS-like systems, and the flexibility to modify, combine, and tailor OSS are the most important facilitating factors for the adoption of OSS within a hospital scenario. The factors inhibiting adoption range from the fear of IT personnel becoming de-skilled by not using mainstream commercial applications, the lack of OSS-literate IT personnel, the lack of other successful OSS examples in the industry, to the lack of reliable procurement models for the adoption of OSS. Finally, many of the papers and studies reported are cases from European countries, with healthcare systems that are very different from that in the U.S. Table 1 presents only a summary of the facilitators and inhibitors shown to influence the adoption of OSS as found in the literature.

MetHodology A mixed methods design was used in this research to explore the extent of OSS adoption in hospitals as well as to investigate the influencing and inhibiting factors. The exploratory approach of this study is warranted by the fact that, as of yet, the adoption and use of OSS in U.S. hospitals has not been accompanied by any theoretical grounding or by empirical analysis that explains how or why OSS products are being adopted and used. That is, thus far, there are few existing conceptual frameworks to guide a research effort in this area. Similarly, there are no theoretical guidelines that have been empirically evaluated to support a rigorous understanding of the complex factors that inhibit the adoption and successful implementation of OSS technologies in hospitals. For these reasons, a mixed methods approach using a

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Table 1. Main facilitators/inhibitors of OSS adoption Author(s)

Major Factor Findings Facilitators

Inhibitors

Fitzgerald and

• Limited financial resources

• Lack of support from vendors

Kenny (2004)

• Top management support

• Perception that OSS would threaten local proprietary

Adoption of Open Source Software in Hospitals

• Software functionality Gynn, Fitzgerald and Exton (2005)

software companies

• User’s past experience

• Fear by users to become de-skilled

• Perception that the benefits of OSS outweigh its

• Perception of work under-valued if using OSS products

disadvantages

• Having to change operating model to OSS

• OSS-literate IT personnel

• Fear by users to be de-skilled

• Top management support

• Lack of OSS champion example

• Personal support for OSS ideology

• Lack of tolerance to technical problems with OSS

• Network externalities

• Favorable arrangements with proprietary vendors

• The OSS champion example Holck, Larsen

• Limited financial resources

and Pedersen

• Pressure to upgrade IT systems

(2005)

• Top management support • User’s past experience • Government support

• Lack of reliable procurement models 





Legal (licenses) Technical (functionality, security, usability) Corporate and business

policy

(vendor,

customer support, and software alliances)

Tomas Yakel

• Access to real-world systems

• Lack of a mature OSS beyond prototype phase

(2001)

• Reduction of bugs in medical systems

• High level of technical expertise required for OSS

• Reduction of software ownership and develop-

• Proprietary mindset of the medical community

ment cost

• Technology complexity of the medical domain • Lack of OSS-IT personnel support, specifically for medical software applications

MacDonald et al. (2003)

• Public policy encouraging that all software developed by the government must be released under an OSS license

Adoption of Open Source Software in Healthcare

• Information mechanisms to disseminate to the community about OSS developments and benefits

• Medical software currently in use is proprietary software • Leadership and top management in healthcare is risk adverse • Elimination of in-house personnel due to outsourcing • Technology complexity of the medical domain

Hogarth and Turner (2005)

• Reduction of software ownership and development cost

(2003)

medical software applications

• Disappearance of vendor lock-in

• Technology complexity in the medical domain

• OSS adherence to standards for compatibility

• Success of mainstream applications might not translate

and data interchange Kantor et al.

• Lack of OSS-IT personnel support, specifically for

to clinical software

• OSS can reduce EMR ownership and development cost • Disappearance of the vendor lock-in • OSS adherence to standards for compatibility and data interchange

179

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

grounded theory perspective was selected over a confirmatory or causal research design approach. Grounded theory is a systematic, qualitative research procedure used to develop an inductively grounded theory that explains a process, an action, or interaction about a phenomenon (Glaser & Strauss, 1967; Glaser, 1978; 1999; Creswell, 1994; 2005; Strauss & Corbin, 1998; Charmaz, 2006).

study desIgn The data collection methods used in this research are a survey and interviews, allowing both breadth and depth of information concerning the adoption of OSS in hospitals. We focused on Baltimore, Washington and Northern Virginia (BWNV) area hospitals instead of a nationwide area. This allowed us to spend more time cultivating each contact from the target population through initial phone calls, and to obtain richer data in the form of personal face-to-face and telephone exchanges. First, a survey was used to gather data from a wide variety of hospitals dispersed across a geographic area. This was done in order to explore and characterize the extent and the types of OSS products adopted by hospitals. Following the survey, semi-structured interviews were conducted in-person and by telephone with IT managers in order to attain deeper understanding of the factors that facilitate or inhibit OSS adoption in hospitals. Interviews are the quintessential qualitative method for data collection and one of the most widely used techniques for acquiring qualitative data in order to collect impressions and opinions about the particular research issue (Tashakkori & Teddlie, 1998; Patton, 2002). The target population for the study consists of hospital executives, directors and managers that are involved in IT within BWNV area hospitals. Although we selected the BWNV area largely because of our own location, it is an appropriate choice because it is one of the most diverse areas in the U.S. socioeconomically, politically, and cul-

180

turally. The survey sample was selected from the Healthcare Information and Management Systems Society (HIMSS) from their electronic mailing list database of chief information officers (CIO), chief technology officers (CTO), vice presidents (VP) (of information technology (IT), information systems (IS) and management information systems (MIS)), and directors and managers of other IT departments within hospitals. HIMSS was selected because it is a leading non-profit organization dedicated to improving healthcare through the application of information technology (HIMSS, 2006). This research takes a key informant approach that allowed the responses of the IT managers to represent those of the hospital being surveyed. The use of managers as key informants has been successfully applied in many IT studies that involve organizations (Huff & Munro, 1985; Gatignon & Robertson, 1989; Chau & Tam, 1997; Eyler et al., 1999; Goode, 2005).

surVey adMInIstratIon Prior to sending the survey invitation e-mail out, an attempt was made to contact each of the IT managers in the target population by telephone in an effort to encourage participation and receive a verbal commitment from them to complete the survey. After the initial telephone contact, an e-mail invitation letter was sent to the potential respondents. The survey link was appended to the bottom of the e-mail cover letter and upon clicking the survey link, the participant was directed to the online survey (Appendix A). Descriptive statistics, such as frequency distributions, percentages, standard deviations, confidence intervals, Chi-square and Fisher’s tests were computed in order to analyze the survey results. Moreover, to ensure better reporting and complete description of our Web-based survey results, we applied a checklist of recommendations from the Checklist for Reporting Results

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

of Internet e-Surveys (CHERRIES) (Eysenbach, 2004) (Appendix B).

InterVIews The interview population consisted of the subset of survey respondents who responded positively to a survey item that specifically asked if they were willing to share their thoughts and experiences in an interview. A total of 11 survey respondents initially agreed to be interviewed. All such respondents were sent an e-mail letter introducing the objectives of the interview and asking to schedule a meeting. By the end of this process, only five IT hospital managers ultimately agreed to be interviewed. The other six managers, for reasons unknown, chose not to respond to the many invitations by e-mail and telephone to participate and were unreachable to be interviewed. Each interview lasted 30-60 minutes and was conducted between January and May 2007. The interviews focused on the organizational, technological, and environmental factors that facilitated or inhibited the adoption of OSS at their hospitals (Appendix C). Before conducting each interview, the participant was briefed on the nature and purpose of the study. All the participants were asked for their authorization to be recorded during the interview and were asked to sign an informed consent. The interviews were coded and analyzed employing grounded theory consistent with the systematic procedures recommended by Strauss and Corbin (1998), namely open coding, axial coding and selective coding. Coding is the process that dissects, differentiates, combines, and discovers concepts and relevant features from the data (Seaman, 1999). We developed concepts and categories emerging from the data using the line-by-line analysis as described by Straus and Corbin (1998) and Glaser and Straus (1967). The concepts and categories were generated by our analysis of the data and validated applying the

constant comparative method. Each interview was treated as an individual case. NVivo® was used to assist the qualitative analysis process, to manage data, to store the interview transcripts, and to help in coding text (Bazeley & Richards, 2000).

results survey results This research finds that 23% (n=7) of the hospitals within the survey sample have adopted OSS. Conversely, 76% (n=23) of the hospitals indicated that they have not adopted any type of OSS. All of the hospital adopters of OSS reported having general-purpose products. Among them, only 57% (n=4) reported having adopted domain-specific products. Table 2 presents descriptive statistics profiling the hospitals in our survey sample. Key findings from this research indicate that hospitals are adopters of both general-purpose and domain-specific products, but they have adopted general-purpose products to a greater extent than domain-specific products. General-purpose OSS adoption in hospitals clusters mainly in databases, desktop software, programming languages, and operating systems, as well as Web development tools and server products. Well-known OSS products such as MySQL, Linux, Apache, Firefox, PHP and Perl were the leading software products that hospitals selected to adopt. The scale used in the survey to indicate extent of adoption was adapted following Fichman and Kemerer (1997) and it ranges from unawareness (no knowledge of OSS), to awareness, interest (actively learning), evaluation/trial (acquisition and initiation of an evaluation or trial version), commitment (use for one or more deployment projects), limited deployment (regular, but still limited, deployment and use), and general deployment stages (a stable and regular part of the IT infrastructure). The survey results show that the vast majority of general-purpose products are positioned from

181

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Table 2. Descriptive statistics of surveyed hospitals Frequency

Percent

(n =30)

%

Healthcare system hospital

12

40.0

Hospital as a part of a multi-system network

11

36.7

Stand-alone hospital

5

16.7

Ambulatory care facility

1

3.3

Other

1

3.3

<50 beds

1

3.3

101-200 Beds

2

6.7

201-300 Beds

6

20.0

301-400 Beds

4

13.3

401-500 Beds

2

6.7

>501

11

36.7

Not classified by beds

4

13.3

< $5M

1

3.3

$5M-$25M

3

10.0

$26M-$50M

3

10.0

$51M-$200M

2

6.7

$201M- $350M

8

26.7

$351M-$500M

2

6.7

> $501M

11

36.7

<2%

5

16.7

2.1-3.0%

15

50.0

3.1-4.0%

1

3.3

4.1-5.0%

5

16.7

5.1-6.0%

1

3.3

>8%

3

10.0

In-house

27

90.0

Outsourced

3

10.0

≤10

5

16.7

10-30

10

33.3

31-60

7

23.3

≥91

8

26.7

Hospital type

Number of beds in the hospital

Hospital’s annual gross revenue

Annual IT operating budget

Type of IT personnel

Number of in-house IT staff employed full time

continued on the following page

182

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Table 2. continued Years of experience of in-house IT staff ≤2 years

1

3.3

2-5 years

6

20.0

5-10 years

15

50.0

≥10 years

8

26.7

the evaluation/trial stages to the limited deployment stages. Well-known OSS products, for example, MySQL, Linux, Apache, and Perl, are in the limited deployment stages, whereas OSS desktop software applications, such as Firefox and Mozilla, are in the evaluation/trial stages. The extent of adoption of domain-specific products is lower than that of general-purpose products. The predominant adoption stages for all the domain-specific OSS products are awareness to interest. Domain-specific adoption occurs mainly in the telemedicine, electronic medical records, radiology, laboratory and pharmacy information systems products. Furthermore, the results of the survey provide information about relevant contextual and structural characteristics of the hospitals that tend to adopt OSS. These characteristics may have a determinant effect on the adoption of OSS. First, the majority of the adopting hospitals are very large hospitals, with 500 beds or more. Second, these hospitals tend to have high annual revenue, more than $500 million. Third, hospital adopters of OSS have a propensity to have a large number of IT support staff. Finally, hospitals that have adopted OSS also tend to have IT budgets that are less than 3% of the hospital’s total budget.

adoption of OSS in hospitals. Further, hospitals rely heavily on software vendors for all of their IT solutions. The results also show that hospital software vendors enlarge their product lines and the services they provide to hospitals to include general-purpose and domain-specific OSS products. In addition, IT managers have a positive satisfaction level, in general, with the software vendor services and products, and, overall, have a good relationship with them. Table 3 presents a concise summary of the results of the interviews. The majority of the hospital IT managers reported that lack of in-house development, and a perceived lack of security, quality, and accountability of OSS products were the most significant factors that inhibit the adoption of OSS in hospitals. IT managers also identified the lack of medical informaticians, patient-privacy protection and privacy legislation as major inhibitors to adopt OSS, particularly domain-specific products. Based upon our findings (from both the survey and interviews), the following section presents an empirical model describing the factors facilitating and inhibiting the adoption of OSS in hospitals and the relationships between them.

Interview results

adoptIon oF oss In HospItals: an eMpIrIcal Model

This study also identifies, through the interview data, key categories that facilitate and inhibit the adoption of OSS in the hospitals within the sample. The interview data reveal that hospital software vendors are the most critical factor influencing the

We have used Strauss and Corbin’s (1998) paradigm to develop an empirical model describing the adoption of OSS in hospitals, based on our data. This empirical model helps us to develop and propose connections between the factors that

183

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Table 3. Emerging code categories and subcategories of the adoption of OSS in hospitals Core Categories

Subcategories

1. Hospital IT human resources

• In-house software development • IT personnel • Medical informaticians

2. Hospital regulatory landscape

• Patient-privacy protection and privacy legislation • Lack of liability/accountability provided by OSS

3. Hospital software vendors

• Software vendor providers of OSS • Satisfaction level with software vendors • New software business models

4. Hospital organizational factors

• Hospital organizational culture • Hospital organizational structure

5. Hospital technological factors

• Perceived lack of quality • Perceived lack of security

6. International development of OSS

• Labor cost and qualified programmers • Type of healthcare systems

emerged from our findings. Figure 1 presents the empirical model that lays out the analysis of the factors that emerged from our results and the relationships between them. The empirical model identifies temporal and inferential, rather than causal, relationships between the factors relevant to the adoption of OSS in hospitals. For example, the mix of causal conditions in a particular hospital at a particular point in time (as defined by the level of in-house development, the number of IT personnel, etc.) sets the stage and shapes what happens when an event occurs related to the core category (e.g., when a software vendor offers an open source solution to the hospital). This core category then directly influences the strategic actions (i.e., adoption or non-adoption of OSS) that lead to the consequences. The contextual factors and intervening conditions moderate and mediate the strategic actions that are employed to bring about certain consequences (Strauss & Corbin, 1998; Creswell, 2005). So, in terms of the symbology in Figure 1, an arrow from one construct to another cannot be interpreted to mean that the first

184

construct in any way causes the second, but that the mix of factors and actions described by the first construct influence the mix of factors and actions described by the second construct in any particular instance. The constructs of the model are described in more detail below.

causal conditions Causal conditions, as the term is being used in our empirical model, are factors that are identified as influencing the core category. There is evidence from our findings that all of these causal conditions have an influence on whether or not a hospital is open to an offer of OSS by a software vendor. The subject of technical personnel in hospitals came up often in our interview data (see “Interview Results” and Table 3). Hospital IT managers report that the lack of in-house development is the rule rather than the exception; hospitals do not develop their own software systems, and thus they depend on software vendors for all their IT operations and software needs. Managers also mentioned that much of their IT staff personnel

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

are exclusively devoted to the on-site support of IT systems provided by vendors. The degree to which a hospital lacks in-house development activity, and IT personnel with technical development skills influences how dependent they are on their software vendors, and thus influences how they would react to the offerings of their vendors. Such a dependence would make a hospital more likely to accept a technology solution from a vendor that included OSS. A related causal condition is the lack of personnel who possess an amalgamation of medicine and information systems expertise and thus who would be able to develop and maintain software systems tailored to hospitals and healthcare organizations. The perceived lack of general quality and perceived lack of security of OSS products are persistent themes that emerged from our data analysis (these are described under the core category “Hospital technological factors” in Table 3). As one manager commented “OSS is not going to have the same level of quality and not nearly the same level of documentation and rigor you can get from a corporate environment.” Another IT manager opined that “the majority of the OSS are probably of inferior quality because they are just gifts that any research lab puts together and hands out from a couple graduate students.” Managers also perceive OSS as a high-risk product when it concerns security. As one manager commented, “It is not the fact that the OSS won’t be able to provide the functionality that we need in the hospital. The major concern is going to be how secure OSS is.” Managers perceive OSS to be highly vulnerable to attacks from hackers or other parties, which may inhibit them from adopting OSS, even from a vendor. These quality and security factors will color a hospital’s openness to a vendor’s offer of an open source solution. The lack of accountability of OSS providers is also a concern for the hospital IT managers we interviewed. Having a vendor that can be held liable or accountable if there is inadequate or insufficient quality or security of the software

product strongly influences the decision to adopt products from software vendors. As one IT manager expressed: “the factor that caused us not to adopt OSS is the support and accountability that comes with writing a check to a commercial software vendor.” The negative perceptions of quality, security and lack of liability reinforce the hospitals’ dependence on software vendors. Finally, our findings report that IT managers have a positive satisfaction level, in general, towards the products, support and services that software vendors provide in their hospitals, as noted in the interview results. This further reinforces the hospitals’ dependence on software vendors. In summary, these causal conditions all shape and impact the core category, that is, they influence what happens when and if a vendor offers a hospital a solution that includes OSS.

core category The mix of causal conditions in a particular hospital setting sets the stage for the “core category,” that is, the hospital software vendors. While our survey did not address the issue of software vendors, there was unanimous consensus amongst all the hospital IT managers interviewed that hospital software vendors play a pivotal role in the adoption process of OSS in hospitals, as discussed in the interview results. IT managers identify hospital software vendors who supply OSS products and services as the key facilitators for the adoption of both general-purpose and domain-specific OSS products. In terms of the empirical model presented in Figure 1, the actions of the software vendors is the trigger, or the gateway, that creates the situation where a hospital must decide to adopt or not adopt OSS. Such a decision does not even arise except through the actions of a software vendor, according to the findings of this study. As one manager commented, “hospitals are so dependent on vendors of hospital IT products that we are not in the position to kind of ‘buck the rules’ and go it alone for the adoption of OSS.”

185

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Figure 1. Empirical model for the adoption of OSS in hospitals Contextual factors · · · ·

Causal conditions · Lack of in-house software development · Lack of IT personnel · Lack of medical informaticians · Lack of general quality · Lack of security · Lack of liability of OSS · Satisfaction level with software vendors

Hospital type Hospital size Hospital IT budget Hospital organizational culture · Hospital organizational structure

Consequences Core category · Hospital software vendors

Strategic actions · Adoption of OSS

· Reducing software development and implementation costs · Avoiding vendor lock-in · Promoting common data standards · Increasing software quality · Increasing security

Intervening conditions · Patient-privacy protection and privacy legislation · Type of health care system · International development of OSS

However, sometimes this decision is not even explicit. As one IT manager adopter of OSS expressed, “we don’t have a conscious decision to adopt OSS because our hospital outsources a lot of our technical knowledge to vendors, so the adoption of OSS is coming throughout the vendor’s decisions for the most part.” The hospitals’ decision to adopt OSS from software vendors is linked to their belief that the OSS offered this way has “a professional level of quality control” that is greater than the OSS available from other sources, such as the Internet. As one IT manager who adopted vendor-supported OSS stated, “I am very happy using OSS because, for me, the best of two worlds is when vendors support an OSS solution. I am willing to pay for OSS, because I feel I have professional quality and control over the software.”

contextual Factors Contextual factors are the “specific set of conditions (patterns of conditions) that intersect dimensionally at this time and place to create a set of circumstances or problems to which persons respond through actions/interactions” (Strauss 186

& Corbin, 1998, p. 132). Our data, especially the survey data presented in “Survey Results”, reveal that several contextual factors are expected to moderate the adoption of OSS in hospitals. The combined qualitative and quantitative results of this study provide evidence that the following contextual factors may facilitate or inhibit the adoption of OSS in hospitals: 1) hospital type, 2) hospital size, 3) hospital IT budget, 4) hospital organizational culture, and finally 5) hospital organizational structure. These factors are different from the causal conditions listed earlier, in that they are more general, static factors that apply to the hospital as a whole and do not specifically form the hospital’s attitude towards OSS, or towards the software vendor. Depending on the hospital type (such as a standalone hospital versus a multi-hospital network, or a university hospital versus a private hospital, and so on), the importance of IT adoption within the hospital may differ. Different types of hospitals seem to have different requirements to adopt software. For example, a university hospital may allow experimentation with new software products while a private hospital in a multi-hospital network may not allow any type of experimentation. Such

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

factors may have an effect on the adoption of OSS by hospitals. Hospital size is likely to be related to organizational characteristics such as slack in resources or a large professional workforce that can also have a positive effect on the adoption of OSS in hospitals. Hospital IT budget is another contextual factor that emerged in our study as a meaningful factor since hospitals with smaller relative IT budgets (with 3% or less of the total hospital budget) have a propensity to adopt OSS. Other contextual factors within the hospital such as organizational culture and organizational structure can also have an effect on the adoption of OSS. As one manager commented, “the organizational design of the hospitals has a major influence on the adoption of software within the hospital, I don’t want to use the word power structure, but it is almost the political landscape of the organization that influences the way we adopt any technology.” Our findings support the effect that all the aforementioned factors have on the strategic actions (i.e., adoption or non-adoption) as depicted in Figure 1 with regards to OSS adoption within hospitals.

Intervening conditions Intervening conditions are those conditions that “mitigate or otherwise impact causal conditions” (Strauss & Corbin, 1998, p. 131). The intervening conditions identified in this study included: 1) patient-privacy protection and privacy legislation, 2) type of healthcare system, and 3) international development of OSS. These intervening conditions are factors, external to the immediate hospital setting that may inhibit the adoption of OSS in hospitals. IT managers we interviewed (see Table 3) report that factors such as patientprivacy protection and privacy legislation may act as deterrents for the adoption of OSS in general, especially with regards to the domain-specific OSS products. For example, hospital IT managers were reluctant to adopt domain-specific OSS products

because they perceived OSS as posing a threat to patients’ privacy and confidentiality as well as to HIPAA compliance mandates. Consequently, we conclude that the aforementioned three intervening conditions also mediate the adoption of OSS in hospitals.

strategic actions Strategic actions are “purposeful or deliberated acts that are taken to resolve a specific problem” (Strauss & Corbin, 1998, p. 133). The interaction outcome of the core category (hospital software vendors) with the contextual factors and the intervening conditions may result in a decision by hospitals to make full use of a technology—in this case OSS—as a plausible or implausible alternative to proprietary (closed source) or commercial software products.

consequences Consequences are the outcomes of the interaction of the core category with the contextual factors, intervening conditions and the strategic actions. The outcomes of this empirical model are closely aligned with the potential benefits of OSS claimed in the literature reviewed in literature section of this article. However, we can only speculate about the actual consequences, as that part of the model is beyond the scope and objectives of this research. However, investigating the consequences of OSS adoption in hospitals is a vital area for future research.

IMplIcatIons Implications for the literature OSS has created a stir of interest in many disciplines ranging from computer science to sociology, and a growing body of literature has emerged to explain many aspects of OSS. However, no work

187

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

has investigated the adoption of OSS in hospitals. The research presented here addresses this gap. A number of respondents from the interviews noted that the lack of IT personnel and the lack of medical informaticians are inhibiting factors for adoption of OSS by hospitals. This is consistent with previous authors (Yackel 2001; MacDonald et al., 2003; Fitzgerald & Kenny, 2004; Hogarth & Turner, 2005; Waring & Maddocks, 2005) who have noted the importance of IT personnel with high levels of technical expertise required in order to deal with OSS applications and the technological complexity in the medical domain that needs personnel that understand both medicine and information systems. In contrast to other studies claiming that the reduction of ownership and development cost is one of the main advantages of adopting OSS in healthcare (Yackel, 2001; Kantor et al., 2003; Fitzgerald & Kenny, 2004; Glynn et al., 2005; Hogarth & Turner, 2005; Holck, Larsen, & Pedersen, 2005), the findings from this research indicated that cost factors are not a core, important category for hospital IT managers when deciding to adopt OSS. The IT managers in our study were found to be more concerned about the quality, security and liability issues surrounding OSS than about the potential cost-benefit factors associated with the adoption and use of OSS. This finding also compares with a prior study by Goode (2005), which also noted that managers see software with high cost as an indicator of quality. Prior research (Fitzgerald & Kenny, 2004; Glynn et al. 2005; Holck et al. 2005; Waring & Maddocks, 2005) has noted the importance of top management support for the successful adoption of technology within organizations. Our findings, by contrast, show that not only top management support is important to the adoption of OSS by hospitals, but clinical personnel within hospitals (e.g., physicians, nurses, etc.) also exert a significant influence on the decision to adopt not only OSS but any technology. Many IT managers recognized the political influence of these groups

188

as a critical factor in how OSS would be used in the future, even before getting to the technology portion of the adoption of OSS by hospitals. This study also shows that the hospital industry is a very conservative industry when it concerns adopting new technologies. Managers repeatedly indicated the “conservative aspects and risk adverse” behavior of the hospital industry to adopt not only OSS but also any new technology. This finding is consistent with MacDonald et Al. (2003) and Glynn (2005) who also pointed out hospitals’ risk averse behavior when adopting IT. Finally, our core finding about the central role of software vendors in the adoption decision in hospitals has some relationship to prior literature. Some existing studies have indicated that avoiding vendor lock-in is perceived to be an advantage of adopting open source (Carr, 2003; 2004; Fink, 2003; Kantor et al., 2003; Fitzgerald, 2004; Goldman & Gabriel, 2005; Goode, 2005). In contrast, in our study, the role of vendors emerged quite differently. The role of vendors as OSS adopters, who then transfer their adoption decisions on to their client hospitals, has not previously been described in the literature. This finding describes vendors as innovating the way they develop, distribute, support and maintain software systems within hospitals. Prior studies have not shown software vendors to be such key enablers of OSS in the hospital industry.

Implications for Future research This research is unique within the field of OSS and healthcare. That is, there is no study that has been published to date presenting an empirical model for the adoption of OSS in hospitals. This model, grounded in empirical data collected from surveys and interviews, identifies the factors and relationships facilitating and inhibiting the adoption of OSS in hospitals. This model provides the basis for future testing of the interactions among the key concepts proposed in this study. Furthermore, there are numerous significant issues for

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

researchers, including ourselves. Our findings, while not highly generalizable due to the limitations of the study, provide sufficient grounding for future confirmatory studies. In particular, a number of very interesting propositions or hypotheses are suggested by our empirical model, and by the survey and interview data. Future research aimed at validating these hypotheses would be a significant contribution to the field. Examples of such propositions include: •

•

• •

Proposition: Adoption of OSS is more likely to be found in hospitals that have in-house technical staff with experience in software development, OSS, and/or with medical informatics. Proposition: Hospitals with an existing relationship with a software vendor who offers OSS solutions are more likely to adopt OSS. The likelihood increases with the degree of dependence on the vendor and the degree of satisfaction with the vendor. Proposition: The adoption of OSS in a hospital is more likely when there is a centralized IT strategy within the hospital. Proposition: The likelihood of a hospital’s adoption of OSS is negatively correlated with the IT manager’s perception of the general quality and security of OSS products.

To further validate propositions such as those above, as well as the whole empirical framework derived from this study, the following future research is planned: 1.

2.

Validation of the model by collecting data from a large sample of hospitals, either in the U.S. and/or internationally, would allow for further conclusions about the causal relationships and interactions suggested by our empirical model. A case study in a hospital setting to analyze the consequences of the adoption and implementation of OSS.

3.

Further empirical investigation into the relationship between hospital software vendors and adoption of OSS.

Implications for practice The present research provides a better understanding to hospital IT managers and practitioners about the extent of OSS adoption in hospitals in conjunction with the factors facilitating or inhibiting this adoption process. Hospitals and healthcare organizations that are considering the adoption and implementation of OSS technologies need to understand how technological, environmental and organizational factors affect the adoption process. This way IT hospital practitioners can prepare against the expected barriers and can utilize the facilitators for successful OSS technology adoption. The first implication for practitioners is that, contrary to theoretical and anecdotal expectations about the cost-benefit advantages of OSS versus proprietary or commercial-based software, the findings from this research indicated that financial factors are not deemed to be a core concern for IT managers when deciding to adopt OSS. The IT managers in our study were found to be more concerned about the quality, security and liability issues surrounding OSS. This implies that, when building a business case, or justification, for the adoption of OSS, the analysis must take into account issues related to quality, security, and accountability with at least as much prominence as cost-benefit issues. The second implication for hospital IT practitioners would be to involve all the stakeholders within the hospital in the adoption decisionmaking; for this particular point, our finding indicated that physicians, nurses, and other clinical personnel are key stakeholders to address in the adoption process of not only OSS but any type of technology introduced to a hospital. Thus the receptivity to the idea and philosophy of OSS must be assessed with these stakeholders, and

189

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

any ideas and concerns that might surface during the assessment must be documented and taken into account. The third recommendation for hospitals that are considering OSS is that they can start adopting OSS with a small pilot project in order to test and experiment with the quality issues of interest, as well as the costs and benefits, of OSS to the hospital. In addition, it is very important to collect data and metrics from the pilot project and communicate the results to all the stakeholders, including vendors, within the hospital. It is important to mention that OSS is not “free,” and never will be without a cost. Another implication for practitioners who want to promote the use of OSS within the hospital and healthcare industry is for them to liaise with hospital software vendors and the OSS community. Coordinating with hospital IT vendors is important because, as our findings reported, any tendency towards adoption of OSS in hospitals is occurring because healthcare IT vendors are embracing, providing, and maintaining OSS products. Under this business model, hospital software vendors are not only offering the software to hospitals but also offering services for installation, customization, and maintenance of OSS applications, either domain-specific or generalpurpose. Furthermore, there are good examples of software partnerships amongst IT businesses, open source communities, and researchers such as Eclipse and even Linux (Capek, 2005; Goldman & Gabriel, 2005; Zeller & Krinke, 2005) that can be replicated in the hospital and healthcare industry. Moreover, the hospital industry is probably the most influential and powerful industry operating today in the healthcare area. If this industry sees the benefits from OSS, then partnerships between IT businesses, OSS communities, and universities could result in research, development and promotion of OSS hospital products and policies that further the evolution of the OSS movement, as well as provide substantial benefits to the hospital

190

industry. Therefore, such partnerships could be a potentially transforming development in promoting and adopting OSS in hospitals.

lIMItatIons oF tHe study Notwithstanding the important contributions of the current study, it has its own shortcomings. For example, our findings may not apply to the full spectrum of U.S. hospitals. This research is exploratory in nature, so that the design, data collection methods, and analysis were broad by design, and not intended for confirmation. This research also examined the adoption or non-adoption of OSS in a limited geographical area and over a particular time period, which makes any attempts to generalize the results across hospitals in the U.S. difficult without further empirical analysis and investigation. Another limitation was the modest sample size of response in the survey (n=30) and interviews (n=5). Through the evolution of this study, it became clear that IT managers in the hospital industry in the BWNV area were less than enthusiastic about discussing and sharing information about open source adoption within their hospitals. Many attempts to influence a higher rate of response and interview participation were made, including initial contacts, follow-up contacts, reminders, and even financial incentives. While the small sample size affects the ability to generalize results, it does not affect what was the intent of the study, to explore and identify relevant issues and factors for further study. However, it is important to mention that these are important limitations for any future similar study because of the unwillingness of the managers and executives to share their views on issues concerning IT adoption. Finally, another limitation of this research is that the data appears not to represent all types of OSS products. While it was not the intent of the

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

study, it is clear from the responses (in particular the types of OSS that survey respondents report adopting) that our respondents were referring primarily to large, enterprise-level OSS applications (e.g., database servers, Web servers, operating systems, etc.). This limits our ability to extend our findings to the entire spectrum of OSS products available to hospitals. It also limits our ability to compare the results of this study to prior research, which mostly addresses the adoption of smaller, stand-alone, download-and-install types of OSS applications.

reFerences

conclusIon

Carnall, D. (2000). Medical software’s free future. BMJ, 321(7267), 976.

This research identifies the factors that could lead to more effective adoption of OSS by hospitals. In addition, this research sheds light and broadens the understanding of OSS adoption within hospitals by offering to IT practitioners information on the extent of that adoption currently. Finally, the insight gained from this research serves as a guide and foundation for future work to investigate more determinants of OSS adoption in hospitals and healthcare organizations. It is also the researcher’s hope that this study will be the seminal stone to pave the way for future studies on OSS adoption and implementation in organizations, public and private, national and international.

Carr, N. (2003). IT doesn’t matter. Harvard Business Review, 81(5), 41-49.

acKnowledgMent The authors would like to thank Khaled El Emam, Medha Umarji, Roy Rada, Katherine Stewart Dongsong Zhang and Stephen Russell, as well as the anonymous IJHISI reviewers for their editorial insights. Additionally, we deeply appreciate the time, interest, and participation of hospital IT managers in the Baltimore-Washington area surveyed and interviewed in this study.

Bazeley, P., & Richards, L. (2000). The NVivo qualitative project book. London; Thousand Oaks, CA: Sage Publications. Brown, S., Lincoln, M., Groen, P., & Kolodner, R. (2003). VistA—U.S. Department of Veterans Affairs national-scale HIS. International Journal of Medical Informatics, 69(2-3), 135-156. Capek, P., Frank, S., Gerdt, S., & Shields. D. (2005). A history of IBM’s open source involvement and strategy. IBM Systems Journal, 44(2), 249-248.

Carr, N. (2004). Does IT matter?: Information technology and the corrosion of competitive advantage. Boston, MA: Harvard Business School Press. Charmaz, K. (2006). Constructing grounded theory. London; Thousand Oaks, CA: Sage Publications. Chau, P., & Tam, K. (1997). Factors affecting the adoption of open systems: An exploratory study. MIS Quarterly: Management Information Systems, 21(1), 1-20. Creswell, J. (1994). Research design: Qualitative & quantitative approaches. Thousand Oaks, CA: Sage Publications. Creswell, J. (2005). Educational research: Planning, conducting, and evaluating quantitative and qualitative research (2nd ed.). Upper Saddle River, NJ: Merrill. Dedrick, J., & West, J. (2003). Why firms adopt open source platforms: A grounded theory of innovation and standards adoption. Paper presented at the Proceedings of the Workshop on Standard Making: A Critical Research Frontier for Information Systems. Seattle, WA. 191

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Dedrick, J., & West, J. (2004). An exploratory study into open source platform adoption. Paper presented at the Proceedings of the 37th Hawaii International Conference on System Sciences. DeLano, W. (2005). The case for open source software in drug discovery. Drug Discovery Today, 10(3). Erickson, B., Langer, S., & Nagy, P. (2005). The role of open source software in innovation and standardization in radiology. Journal of the American College of Radiology, 2(11), 927-931. Eyler, A., Mayer, J., Rafii, R., Housemann, R., Brownson, R., & King, A. (1999). Key informant surveys as a tool to implement and evaluate physical activity interventions in the community. Health Educ. Res., 14(2), 289-298. Eysenbach, G. (2004). Improving the quality of Web surveys: The checklist for reporting results of Internet e-surveys (CHERRIES). J Med Internet Res, 6(3), e34. Fichman, R., & Kemerer, C. (1997). Object technology and reuse: Lessons from early adopters. Computer, 30(10), 47-59. Fink, M. (2003). The business and economics of Linux and open source. Upper Saddle River, NJ: Prentice Hall PTR. Fitzgerald. (2004). A critical look at open source. Computer, 37(7), 92-94. Fitzgerald, & Kenny. (2004). Developing an information systems infrastructure with open source software. Software, IEEE, 21(1), 50-55. Gatignon, H., & Robertson, T. (1989). Technology diffusion: An empirical test of competitive effects. Journal of Marketing, 53(1), 35-49. Glaser, B. (1978). Theoretical sensitivity: Advances in the methodology of grounded theory. Mill Valley, CA: Sociology Press.

192

Glaser, B. (1999). The future of grounded theory. Qualitative Health Research, 9(6), 836-845. Glaser, B., & Strauss, A. (1967). The discovery of grounded theory; strategies for qualitative research. Chicago, IL: Aldine Publishing Company. Glynn, E., Fitzgerald, B., & Exton, C. (2005). Commercial adoption of open source software: An empirical study. Paper presented at the Proceedings of the 2005 International Symposium on Empirical Software Engineering (ISESE’05). Noosa Heads, QLD, Australia. Goldman, R., & Gabriel, R. (2005). Innovation happens elsewhere: Open source as business strategy. Amsterdam; Boston, MA: Morgan Kaufmann. Goode, S. (2005). Something for nothing: Management rejection of open source software in Australia’s top firms. Information and Management, 42(5), 669-681. HIMSS. (2006). HIMSS bylaws. Retrieved June 1, 2006, from http://www.himss.org/content/files/ himss_bylaws.pdf. Hogarth, M., & Turner, S. (2005). A study of clinical related open source software projects. Paper presented at the American Medical Informatics Association Conference (AMIA). Washington, D.C. Holck, J., Larsen, M., & Pedersen, M. (2005). Managerial and technical barriers to the adoption of open source software. Berlin-Heidelberg: Springer. Huff, S., & Munro, M. (1985). Information technology assessment and adoption: A field study. MIS Quarterly, (December), 327-338. Kantor, G., Wilson, W., & Midgley, A. (2003). Open source software and the primary care EMR. J Am Med Inform Assoc, 10(6), 616.

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

Longman, P. (2007). Best care anywhere: How America’s most socialized healthcare system became its finest: What the VA miracle means for you and your country. Sausalito, CA: PoliPointPress.

Strauss, A., & Corbin, J. (1998). Basics of qualitative research: Techniques and procedures for developing grounded theory (2nd ed.). Thousand Oaks, CA: Sage Publications.

McDonald, C., Schadow, G., Barnes, M., Dexter, P., Overhage, J., Mamlin, B., et al. (2003). Open source software in medical informatics—why, how and what. International Journal of Medical Informatics, 69(2-3), 175-184.

Tashakkori, A., & Teddlie, C. (1998). Mixed methodology: Combining qualitative and quantitative approaches. Thousand Oaks, CA: Sage Publications.

Nagy, P. (2007). Open source in imaging informatics. Journal of Digital Imaging, 20(0), 1-10. Norris, J. (2004). Mission-critical development with open source software: Lessons learned. Software, IEEE, 21(1), 42-49. Patton, M. (2002). Qualitative research and evaluation methods (3rd ed.). Thousand Oaks, CA: Sage Publications. Raghupathi, W., & Gao, W. (2007). An eclipsebased development approach to health information technology. International Journal of Electronic Healthcare, 3(4), 433-452. Seaman, C. B. (1999). Qualiative methods in empircal studies of software engineering. Software Engineering, IEEE Transactions on, 25(4), 557-572. Scarsbrook, A. (2007). Open source software for radiologists: A primer. Clinical Radiology, 62(2), 120-130.

Valdes, I., Kibbe, D., Tolleson, G., Kunik, M., & Petersen, L. (2004). Barriers to proliferation of electronic medical records. Informatics in Primary Care, 12(1), 3-9. Van Latum, F., Van Solingen, R., Oivo, M., Hoisl, B., Rombach, D., & Ruhe, G. (1998). Adopting GQM-based measurement in an industrial environment. Software, IEEE, 15(1), 78-86. Waring, T., & Maddocks, P. (2005). Open source software implementation in the UK public sector: Evidence from the field and implications for the future. International Journal of Information Management, 25(5), 411-428. Yackel, T. (2001). How the open source development model can improve medical software. Medinfo, 10, 68-72. Zeller, A., & Krinke, J. (2005). Essential open source toolset: Programming with Eclipse, JUnit, CVS, Bugzilla, Ant, Tcl/Tk and more. Chichester, England: John Wiley & Sons.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 3, edited by J. Tan, pp. 16-37, copyright 2008 by IGI Publishing (an imprint of IGI Global).

193

An Empirical Investigation into the Adoption of Open Source Software in Hospitals

appendIx a. surVey InstruMent http://userpages.umbc.edu/~gimunoz1/Appendix%20A-%20Survey%20Instrument.pdf

appendIx B. cHerrIes http://userpages.umbc.edu/~gimunoz1/Appendix%20B-%20CHERRIES.pdf

appendIx c. InterVIew guIde protocol http://userpages.umbc.edu/~gimunoz1/Appendix%20C-%20Interview%20Guide.pdf

194

195

Chapter 14

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching Masoud Mohammadian University of Canberra, Australia Ric Jentzsch Compucat Research Pty Limited, Australia

aBstract Radio frequency identification (RFID) is a promising technology for improving services and reduction of cost in health care. Accurate almost real time data acquisition and analysis of patient data and the ability to update such a data is a way to improve patient’s care and reduce cost in health care systems. This article employs wireless radio frequency identification technology to acquire patient data and integrates wireless technology for fast data acquisition and transmission, while maintaining the security and privacy issues. An intelligent agent framework is proposed to assist in managing patients’ health care data in a hospital environment. A data classification method based on fuzzy logic is proposed and developed to improve the data security and privacy of data collected and propagated.

IntroductIon Research into the use of developing and evolving technologies needs to be expanded in order that society as a whole can benefit. Radio Frequency Identifiers (RFID) have been around for many years. Their use and projected use has only begun to be researched in hospitals [Fuhrer, P. and Gui-

nard, D. 2007]. This chapter research considers the use of RFIDs and its potential in hospitals and similar environments. Furthermore RFIDs are used to collect data at its source while developing profiles for patients and their care. There are four areas where using RFIDs and their data collection can have significant positive effects in hospitals. These four areas are:

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

• • •

•

Care tracking: this is getting the right care to the right patient at the right time. Quality of care: improving the services given to the right patient at the right time in a timely manner. Cost of care: finding ways to be effective in the use of available resources such that the cost per patient per incident does not adversely increase to the cost of the resources. Service of care: better, more timely information for a more informed decision making process, to provide more knowledgeable individual tailored care.

RFID tags and readers are most commonly are associated with tracking goods in manufacturing and warehousing, but hospitals are starting to apply RFID to new purposes [Kowalke, M. 2006]. RFID technology does not require contact or line of sight for communication, like bar codes. RFID data can be read through the human body, through clothing, read wirelessly, and through non-metallic materials. Both research and practical application of the use of RFIDs in hospitals continues to be of importance. For hospitals this has meant the potential of managing inventories in a more efficient manner. Inventories in hospitals take on a variety of differences than to manufacturing. The nature of the inventory and assets in a hospital can include various types of equipment (that is often very expensive, comes in many sizes, and uses), drugs (that come in a variety of sizes, shapes, color, and governing regulations), beds, chairs, patients (the primary reason hospitals exist), and staff. The percentage of worldwide radio frequency identification (RFID) projects concerning peopletagging has increased from eight percent to 11 percent since 2005 [Tindal, S. 2008]. However, the healthcare sector has yet to quantify or provide evidence of the benefit to people-tagging. Human chipping is not new but does bring up a lot of ethical questions [Angeles, R. 2007].

196

RFIDs are used in hospitals for tracking highvalue assets and setting up automated maintenance routines to improve operational efficiencies. However the use of RFIDs in tracking beds and tracking mobile equipment is in its infancy. RFIDs is used to monitor equipment for example how long a bed was used at a particular location to determine a sterilization schedule as well as bed location tracking. However RFID technology is already being deployed across the pharmaceutical industry to combat drug counterfeiting, drugs shelf life tracking [Kowalke, M. 2006]. Managing expensive, often difficult to replace, and legal drugs can only be improved using RFIDs. The management of patients and their condition is paramount in a hospital. RFIDs can assist in asset and personnel tracking, patient care, and billing where unnecessary expenses will be cut, the average length of stay of a patient is reduced, where more patient lives will be saved due to timely efficient services, and where patient records are actively continuously updated to provide better patient care. [Kowalke, M. 2006]. An RFID chip stores the wearer’s data that can be accessed by a hand-held reader. This makes patient identification more reliable, provides updated patient condition nearly instantly, and improves the cost of health care. The health sector is already taking up peopletagging where it allows nurses to radio their location if they are being assaulted, reduce mother baby mismatches and baby theft, help severe diabetics with getting correct treatment, and monitoring disoriented elderly patients without the need for a dedicated member of staff [Tindal, S. 2008]. The need is not to keep track of staff but be able to locate the staff with the particular skills that are needed at the right time and place. Staff wearing badges with RFIDs embedded can be found to help provide that needed and timely care that a patient may need. However privacy concerns have been aired over patient tracking using RFIDs.

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

A patient upon arrival to a hospital can be is issued with an RFID tag which can contain information concerning the patient such as their surname, first name, reason admitted to hospital, date of admission, their doctor’s name, a patient number and a section for monitoring. Monitoring could include heart rate, blood pressure, and some other vital signs. Monitoring would be settable to the need of the patient. For example once an hour might be sufficient for most patients, but for others every 15 minutes might be sufficient. This illustration shows how a particular section of a hospital might be configured according to the needs of that section. Each patient has a patienttag. Each patient’s bed has a bed-tag. Spaced out within rooms, hall ways, and hospital staff stations are receivers. Every 15 minutes the receiver interrogates the bed and the patient tags. The patient’s vital signs are sent to the patient database where the patient’s condition is recorded. The patient care profile is then updated with this information. If anything is out of range or an exception is identified the nearest nurse station to the patient is then contacted. There is a need for more research into applications and innovative architectures for secure access, retrieval and update of data in healthcare systems [Finkenzeller. K. (1999), Glover. B and Bhatt H. (2006), Hedgepeth W. O. (2007). Lahiri, S. (2005), Schuster E. W., Allen S. J. and Brock D. L. (2007). Shepard S. (2005), Angeles, R. (2007), Pramatari, K.C., Doukidis, G.I. and Kourouthanassis, P. (2005), Qiu R, Sangwan R. (2005), Mickey, K. (2004), Whiting, R. (2004), Weinstein, R. (2005)]. Although many organizations are developing and testing the possible use of RFIDs the real value of RFID is achieved in conjunction with the use of intelligent software agents for processing and monitoring data obtained via RFIDs. Thus the issue becomes the integration of these two great technologies for the benefit of assisting health care services. This article considers a framework using RFID and Intelligent Software Agents for managing

patients’ health care data in a hospital environment. A fuzzy data classification system is also developed to improve the application of regulatory data requirements for security and privacy of data exchange. The chapter is divided into four main sections. Next section considers issues relate to data collection and profiling. Section three is based on the patient to doctor profiling and intelligent software agents. The fourth section covers RFID background and provides a good description of RFIDs and their components. This section discusses several practical cases of RFID technology in and around hospitals. It will also list three possible applicable cases assisting in managing patients’ medical data. The final section discusses the important issue of maintaining patients’ data security and integrity and relates that to RFIDs.

data collectIon Large amount of health care data such as patients, doctors, nurses, institution itself, drugs and prescriptions, diagnosis, and many other areas is collected and stored in hospitals. It is not feasible or effective to use RFID to collect and retrieve such large amount of data. This chapter concentrates on a subset with the understanding that all areas could, directly or indirectly, benefit from the use of RFID and intelligent software agents in a health care environment. The RFID [Bhuptani, M., & Moradpour, S. (2005)] provides the passive vehicle to obtain the data via its monitoring capabilities. The intelligent software agent provides the active vehicle in the interpretation profiling of the data and reporting capacity. By investigating and analyzing collect patient data the patient’s condition can be monitored and abnormal situations can be reported on time. Using this information an evolving profile of each patient can be constructed and analysed. Analyzing the data can assist in deciding what kind of care a patient requires, the effects of ongoing care, and how to best care for this patient using

197

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

available resources (doctors, nurses, beds, etc…) for the patient. The intelligent software agent builds a profile of each patient as they are admitted to the healthcare institution by analyzing the recorded and stored data about each patient. The same way a profile for each doctor is developed based on stored data about each doctor. Therefore patients and doctors profile can be correlated to obtain the specialization and availability of the doctors to suit the patients.

patient profiling Profiling is combined with personalization, and user modeling [Wooldridge, M. and Jennings, N. (1995)]. The use of profile in hospitals and healthcare so far has been limited. Tracking of information about consumers’ interests by monitoring their movements online is considered profiling or user modeling in e-commerce systems. By analyzing the content, URL’s, and other information about a user’s browsing path/click-stream a profile of a user behavior is constructed. However patient profiling differ from user profiling in ecommerce systems. The patient profiling is useful in a variety of situations such as providing a personalized service based on the patient and not on symptoms or illness to a particular patient as well as assisting in identifying the medical facilities in trying to prevent the need for the patient to return to the hospital any sooner than necessary. Patient profiling also assist in matching a doctor’s specialization to the right patient. A patient profile can also assist in providing information about the patient on continuous bases for the doctors so that a tailored and appropriate care can be provided to the patient.

and patient. A static profile is kept in pre-fixed data fields where the period between data field updates is long such as months or years. A dynamic profile is constantly updated as per evaluation of the situation in which the situation occurs. The updates may be performed manually or automated. The automated user profile building is especially important in real time decision-making systems. Real time systems are dynamic. The profiling of patient doctor model is based on the patient / doctor information. These are: •

•

•

A value denoting the degree of association can be created form the above evaluation of the doctor to patient’s profile. The intelligent agent based on the denoting degrees and appropriate, available doctors can be identified and be allocated to the patient. In the patient / doctor profiling the agent will make distinctions in attribute values of the profiles and match the profiles with highest value. It should be noted that the agent creates the patient and doctor profiles based on data obtained from the doctors and patient namely:

patIent to doctor proFIlIng

•

A patient or doctor profile is a collection of information that can be used in a decision analysis situation between the doctor, domain environment,

•

198

The categories and subcategories of doctor specialization and categorization. These categories will assist in information processing and patient / doctor matching. Part of the patients profile based on their symptoms (past history problems, dietary restrictions, etc.) can assist in prediction of the patients needs specifically. The patients profile can be matched with the available doctor profiles to provide doctors with information about the arrival of patients as well as presentation of the patients profile to a suitable, available doctor.

Explicit profiling occurs based on the data entered by hospital staff about a patient. Implicit profiling can fill that gap for the missing data by acquiring knowledge about the patient from its past visit or other relevant

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Figure 1. Agent profiling model using RFID

patient data

Patient database

Staff database

RFID patient data Patient profile

staff data

Staff profile Patient Profile Agent Engine

Staff Profile Agent Engine Rule Base

Rule Base

Profile Agent Engine Rule Base

Matching Profile Agent

databases if any and then combining all these data to fill the missing data. Using legacy data for complementing and updating the user profile seems to be a better choice than implicit profiling. This approach capitalizes on user’s personal history (previous data from previous visit to doctor or hospital). The proposed agent architecture allows user profiling and matching in such a time intensive important application. The architecture of the agent profiling systems using RFID is given in Figure 1. Profile matching done is based on a vector of weighted attributes. To get this vector, a rule based systems can be used to match the patient’s attributes (stored in patient’s profile) against doctor’s attributes (stored in doctor’s profile). If there is a partial or full match between them then the doctor will be informed (based on their availability from the hospital doctor database).

Profile matching [Doan, A-H. Lu, y. Lee, T. Han, J (2003)] performed is based on a vector of weighted attributes using an intelligent agent system. To get this vector, the intelligent agent uses a rule-based system to match the patient’s attributes (stored in patient’s profile) against doctor’s attributes (stored in doctor’s profile). If there is a partial or full match between them then the doctor will be informed (based on their availability from the hospital doctor database). Such a rules based system is built based on the knowledge of domain experts. This expert system is extensible as new domain knowledge can be added to its knowledge base as rules. Large amount of research in the area of profiling in e-commerce, schema matching, information extraction and retrieval has been shown promising results [(Do, H. Rahin, E. (2002), Doan, A-H. Lu, y. Lee, T. Han, J (2003)]. However profiling in healthcare is new and innovative.

199

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Staff and patient / doctor profiling and profile matching could be the missing link in providing more tailored healthcare professionals and facility to patients in a hospital environment. Profile matching may consist of: • • •

determining the matching algorithms required for matching patient / doctor profile, determining the availability of staff and facilities required for a given patient. understanding of government policies related patient healthcare,

Another issue related to patient / doctor profiling is defining the level of matching of the patient and doctor profile. There is not always possible to provide the services of the doctor/s and facilities identified as exact match for a patient healthcare because the matching doctor may be unavailable or unreachable. Some guidelines include issues such as the critical nature of the patient illness, its level of sensitivity and regulatory rules. As such the rules that govern patient / doctor profile matching can be expressed in human linguistic terms which can be vague and difficult to represent formally. Fuzzy Logic (Zadeh, L. A., 1965) has been found to be useful in its ability to handle vagueness. As such the profiling patient / doctor matching is based on fuzzy logic. The profiling matching system consists of a fuzzy rule based system uses and inference engine to a weighted value between a patient profile and doctor/s profile. The matching between a patient profile and doctor/s is then divided into the following classes: “total match”, “medium match”, “low match” and “no match”. Based on these class categories a weighted match of patient / doctor profile can be identified. The doctors then can be categorized and ranked based on the matching profile value. The doctors can be classified into classes based on their matching profile as well as their availability such as “highly available”, “more and less available” and “not available”. Of course the data

200

about availability of the doctors are obtained and updated and the profiling agent continuously checks such information from the staff database. The matched doctors then can be ranked as: “high”, “medium” and “low”. A fuzzy rule the profile matching system then may look like: IF patient_doctor_profile is total match and doctor_availability is highly available Then doctor_ranking = high

The integration of RFID capabilities and intelligent agent techniques provides promising development in the areas of performance improvements in RFID data collection, inference and knowledge acquisition and profiling operations. Due to the important role of intelligent agents in this system, it is recognized that there is a need for a framework to coordinate intelligent agents so that they can perform their task efficiently. Intelligent agent coordination [Wooldridge, M. and Jennings, N. (1995). Odell, J. and Bigus, J. P. and Bigus, J. (1998). Shaalana, K. El-Badryb,M. and Rafeac, A. (2004)] has shown to be promising. The Agent Language Mediated Activity Model (ALMA) agent architecture currently under research is based on the mediated activity framework. We believe that such a framework is able to provide RFID with the necessary framework to profile a range of internal and external medical/patient profiling communication activities performed by wireless multi-agents.

rFId descrIptIon RFID or Radio Frequency Identification is a progressive technology that has been said to be easy to use and well suited for collaboration with intelligent software agents. Basically an RFID can: • • • •

be read-only; volatile read/write; or write once / read many times RFID are:

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

• •

non-contact; and non-line-of-sight operations.

Being non-contact and non-line-of-sight will make RFIDs able to function under a variety of environmental conditions and while still providing a high level of data integrity [Finkenzeller. K. (1999), Glover. B and Bhatt H. (2006), Hedgepeth W. O. (2007). Lahiri, S. (2005), Schuster E. W., Allen S. J. and Brock D. L. (2007). Shepard S. (2005)]. Next section will discuss the environment that RFIDs operate in and their relationship to other available wireless technologies such as the IEEE 802.11b, IEEE 802.11g, IEEE 802.11n etc… in order to fulfill their requirements effectively and efficiently. RFID or Radio Frequency Identification is a progressive technology that has been said to be easy to use and well suited for collaboration with intelligent software agents. Basically an RFID can be read-only, volatile read/write; or write once / read many times. RFID are non-contact; and non-line-of-sight operations. Being non-contact and non-line-of-sight will make RFIDs able to function under a variety of environmental conditions and while still providing a high level of data integrity [Finkenzeller. K. (1999), Glover. B and Bhatt H. (2006), Hedgepeth W. O. (2007). Lahiri, S. (2005), Schuster E. W., Allen S. J. and Brock D.

L. (2007). Shepard S. (2005)]. A basic RFID system consists of four components namely, the RFID tag (sometimes referred to as the transponder), a coiled antenna, a radio frequency transceiver and some type of reader for the data collection.

transponders The reader emits radio waves in ranges of anywhere from 2.54 centimeters to 33 meters. Depending upon the reader’s power output and the radio frequency used and if a booster is added that distance can be increased. When RFID tags (transponders) pass through a specifically created electromagnetic zone, they detect the reader’s activation signal. Transponders can be on-line or off-line and electronically programmed with unique information for a specific application or purpose. A reader decodes the data encoded on the tag’s integrated circuit and passes the data to a server for data storage or further processing. There are four major frequency ranges that RFID systems operate at. As a rule of thumb, low-frequency systems are distinguished by short reading ranges, slow read speeds, and lower cost. Higher-frequency RFID systems are used where longer read ranges and fast reading speeds are required, such as for vehicle tracking, automated toll collection, asset management, and tracking of mobile equipment.

Table 3. Frequency ranges for RFID systems Frequency

Range

Applications

Low-frequency

3 feet

Pet and ranch animal identification; car key locks

3 feet

library book identification; clothing identification; smart cards

25 feet

Supply chain tracking: Box, pallet, container, trailer tracking

100 feet

Highway toll collection; vehicle fleet identification

125 - 148 KHz High-frequency 13.56 MHz Ultra-high freq 915 MHz Microwave: 2.45GHz

201

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

coiled antenna The coiled antenna is used to emit radio signals to activate the tag and read or write data to it. Antennas are the conduits between the tag and the transceiver that controls the system’s data acquisition and communication. RFID antennas are available in many shapes and sizes. They can be built into a doorframe, book binding, DVD case, mounted on a tollbooth, embedded into a manufactured item such as a shaver or software case (just about anything) so that the receiver tags the data from things passing through its zone [Finkenzeller. K. (1999), Glover. B and Bhatt H. (2006), Hedgepeth W. O. (2007). Lahiri, S. (2005), Schuster E. W., Allen S. J. and Brock D. L. (2007). Shepard S. (2005).]. Often the antenna is packaged with the transceiver and decoder to become a reader. The decoder device can be configured either as a handheld or a fixed-mounted device.

types of rFId transponders RFID tags can be categorized as active, semiactive, or passive. Each has and is being used in a variety of inventory management and data collection applications today. The condition of the application, place and use determines the required tag type. Active RFID tags are powered by an internal battery and are typically read / write. Tag data can be rewritten and / or modified as the need dictates [Finkenzeller. K. (1999), Glover. B and Bhatt H. (2006)]. The semi-active tag comes with a battery. The battery is used to power the tags circuitry and not to communicate with the reader [Shepard S. (2005)]. Passive RFID tags operate without a separate external power source and obtain operating power generated from the reader. Passive tags, since they have no power source embedded in themselves, are consequently much lighter than active tags, less expensive, and offer a virtually unlimited operational lifetime. However, the trade off is that they have shorter

202

read ranges, than active tags, and require a higherpowered reader.

Hospital environment In a hospital environment, in order to manage patient medical data we need both types; fixed and handheld transceivers. Also, transceivers can be assembled in ceilings, walls, or doorframes to collect and disseminate data. Hospitals have become large complex environments. In a hospital nurses and physicians can retrieve the patient’s medical data stored in transponders (RFID tags) before they stand beside a patient’s bed or as they are entering a ward. Given the descriptions of the two types and their potential use in hospital patient data management we suggest that: •

•

It would be most useful to embed a passive RFID transponder into a patient’s hospital wrist band; It would be most useful to embed a passive RFID transponder into a patient’s medical file;

Doctors should have PDAs equipped with RFID or some type of personal area network device. Either would enable them to retrieve some patient’s information whenever they are near the patient, instead of waiting until the medical data is pushed to them through the hospital server. After examining both ranges for Active and Passive RFID tags, we can suggest the following: •

Low frequency range tags are suitable for the patients’ band wrist RFID tags. Since we expect that the patients’ bed will not be too far from a RFID reader. The reader might be fixed over the patient’s bed, in the bed itself, or over the door-frame. The doctor using his/her PDA would be aiming to read the patient’s data directly and within a relatively short distance.

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

•

High frequency range tags are suitable for the physician’s tag implanted in their PDAs. As physicians move from one location to another in the hospital, data on their patients could be continuously being updated.

transceivers The transceivers / interrogators can differ quite considerably in complexity, depending upon the type of tags being supported and the application. T\ he overall function of the application is to provide the means of communicating with the tags and facilitating data transfer. Functions performed by the reader may include quite sophisticated signal conditioning, parity error checking and correction. Once the signal from a transponder has been correctly received and decoded, algorithms may be applied to decide whether the signal is a repeat transmission, and may then instruct the transponder to cease transmitting or temporarily cease asking for data from the transponder. This is known as the “Command Response Protocol” and is used to circumvent the problem of reading multiple tags over a short time frame. Using interrogators in this way is sometimes referred to as “Hands Down Polling”. An alternative, more secure, but slower tag polling technique is called “Hands Up Polling.” This involves the transceiver looking for tags with specific identities, and interrogating them in turn. Hospital patient data management deals with sensitive and critical information (patient’s medical data). Hands Down polling techniques in conjunction with multiple transceivers that are multiplexed with each other, form a wireless network. The reason behind this choice is that, we need high speed for transferring medical data from medical equipment to or from the RFID wristband tag to the nearest RFID reader then through a wireless network or a network of RFID transceivers or LANs to the hospital server. From

there it is a short distance to be transmitted to the doctor’s PDA, a laptop, or desktop through a WLAN or wired LAN. The “Hand Down Polling” techniques as previously described, provides the ability to detect all detectable RFID tags at once (i.e. in parallel). Preventing any unwanted delay in transmitting medical data corresponding to each RF tagged patient. Transponder programmers are the means, by which data is delivered to write once, read many (WORM) and read/write tags. Programming can be carried out off-line or on-line. For some systems re-programming may be carried out on-line, particularly if it is being used as an interactive portable data file within a production environment, for example. Data may need to be recorded during each process. Removing the transponder at the end of each process to read the previous process data, and to program the new data, would naturally increase process time and would detract substantially from the intended flexibility of the application. By combining the functions of a transceiver and a programmer, data may be appended or altered in the transponder as required, without compromising the production line. It can be concluded from this section that RFID systems differ in type, shape, and range; depending on the type of application, the RFID components shall be chosen. Low frequency range tags are suitable for the patients’ band wrist RFID tags. Since we expect that the patients’ bed not to be too far from the RFID reader, which might be fixed on the room ceiling or door-frame. High frequency range tags are suitable for the physician’s PDA tag. As physicians move from on location to another in the hospital, long read ranges are required. On the other hand, transceivers which deal with sensitive and critical information (patient’s medical data) need the Hands Down polling techniques. These multiple transceivers should be multiplexed with each other forming a wireless network.

203

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

applications of the rFId technology in a Hospital

ii.

The following section describes steps involved in the process of using RFID in hospital environment for patient information management: 1.

A biomedical device equipped with an embedded RFID transceiver and programmer will detect and measure the biological state of a patient. This medical data can be an ECG, EEG, BP, sugar level, temperature or any other biomedical reading. After the acquisition of the required medical data, the biomedical device will write this data to the RFID transceiver’s EEPROM using the built in RFID programmer. Then the RFID transceiver with its antenna will be used to transmit the stored medical data in the EEPROM to the EEPROM in the patient’s transponder (tag) which is around his/her wrist. The data received will be updated periodically once new fresh readings are available by the biomedical device. Hence, the newly sent data by the RFID transceiver will be updated (and may be accumulated as needed) to the old data in the tag. The purpose of the data stored in the patient’s tag is to make it easy for the doctor to obtain medical information regarding the patient directly via the doctor’s PDA, tablet PC or laptop. 2. Similarly, the biomedical device will also transfer the measured medical data wirelessly to the nearest WLAN access point. Since high data rate transfer rate is crucial in transferring medical data, IEEE 802.11b or g is recommended for the transmission purpose. 3. Then the wirelessly sent data will be routed to the hospitals main server; to be then sent (pushed) to: i. Other doctors available throughout the hospital so they can be notified of any newly received medical data.

204

4.

5.

6.

7.

To an on-line patient monitoring unit or a nurse’s workstation within the hospital. iii. Or the acquired patients’ medical data can be fed into an expert (intelligent agent) software system running on the hospital server and to be then compared with other previously stored abnormal patterns of medical data, and to raise an alarm if any abnormality is discovered. Another option could be using the in-builtembedded RFID transceiver in the biomedical device to send the acquired medical data wirelessly to the nearest RFID transceiver in the room. Then the data will travel simultaneously in a network of RFID transceivers until reaching the hospital server. If a specific surgeon or physician is needed in a specific hospital department, the medical staff in the monitoring unit (e.g. nurses) can query the hospital server for the nearest available doctor to the patient’s location. In our framework an intelligent agent can perform this task. The hospital server traces all doctors’ locations in the hospital through detecting the presences of their wireless mobile device; e.g. PDA, tablet PC or laptop in the WLAN range. Physicians may also use RFID transceivers built-in the doctor’s wireless mobile device. Once the required physician is located, an alert message will be sent to his\her PDA, tablet PC or laptop indicating the location to be reached immediately including a brief description of the patient’s case. The doctor enters into the patient’s room or ward according to the alert he/she has received. The doctor wants to check the medical status of a certain patient and interrogates the patient’s RFID wrist tag with his RFID transceiver equipped in his\her PDA, tablet PC or laptop, etc.

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

classIFIcatIon oF data For encryptIon Data security, availability, privacy and integrity [McGraw, G. (2006)] are of paramount important in healthcare and hospital environment. Data security and privacy policies in healthcare are governed by hospital, medical requirements and government regulations. These requirements demand not only for data security but also for data accessibility and integrity. Implementing data security using data encryption solutions remain at the forefront for data security. Data encryption algorithms are implemented to protect the actual data. However data is stored and transferred through several devices and simply protection of data by data encryption fails to secure the resource on which it is stored or transferred. On the other hand the issue of keys and overall process of data encryption process remains complex. The best data encryption solutions are those that balance information protection with on-demand access to

that encrypted data for security professionals or other designated individuals within an organization. Since managing the keys for encrypted data can become difficult many large organizations choose to encrypt their regulated data. Once data is transmitted wirelessly, security becomes a more crucial issue. Unlike wired transmission, wirelessly transmitted data can be easily sniffed out leaving the transmitted data vulnerable to many types of attacks. For example, wireless data could be easily eavesdropped on using any mobile device equipped with a wireless card. In worst cases wirelessly transmitted data could be intercepted and then possibly tampered with, or in best cases, the patient’s security and privacy would be compromised. Hence emerges the need for data to be initially encrypted from the source. Two main layers of encryption can be used when using RFID’s in hospital environment, they are, Physical (hardware) layer encryption and Application (software) layer encryption. This means encrypting all collected medical data at the source or hardware level before transmitting it. Thus,

Figure 2. Sample patient-doctor relational database with and associate fuzzy set

205

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

we insure that the patient’s medical data would not be compromised once exposed to the outer world on its way to its destination. So even if a person with a malicious intent and also possessing a wireless mobile device steps into the coverage range of the hospitals’ WLAN, this intruder will gain actually nothing since all medical data is encrypted, making all intercepted data worthless. In application (software) layer encryption all collected medical data at the destination or application level is encrypted once it is received. Application level encryption runs on the doctor’s wireless mobile device (e.g. PDA, tablet PC or laptop) and on the hospital server. Once the medical data is received, it will be protected by a secret pass-phrase (encryption\decryption key) created by the doctor who possesses this device. This type of encryption would prevent any person from accessing patient’s medical data if the doctor’s wireless mobile device gets lost, or even if a hacker hacks into the hospital server via the Internet, intranet or some other mean. Sensitive and mission critical data are stored in databases, in server applications, and middleware. However many solutions to data encryption at this level are expensive, disruptive, and demanding intensive resource. Using a data classification process organizations can identify and encrypt only the relevant data thereby saving time and processing power. Without data classification organizations using encryption process would simply encrypt everything and consequently impacting users more than necessary (Cline, J., 2007; Butterfield, R., 2007). Understanding the value of the data is significant information for an organization in determining and deploying the proper data classification, security and risk assessment. Data classification is essential and can assist organizations with their data security, privacy and accessibility needs. Such a classification process needs to be able to determine the value, sensitivity, privacy, government regulations and corporate strategic objectives. However data classification is a ma-

206

jor difficulty for many organizations as it is an expensive and time consuming task. Classification of and data process may consist of: •

• •

• •

•

•

Determining the business and corporate objectives and the level of protection required for data in that organization (e.g. security measures required, intellectual property protection, strategic use of data, privacy policies), Understanding of government policies for data protection and accessibility, Determining the corporate vales of the data and its sensitivity (private business critical data, internal usage of data, public release of data), Determining who needs access to the data (e.g. user security level), Determining the processes that manipulate the data (internal processes, external applications, mix internal-external applications), Determine the life time of the data and issues related to storage, backups and removal of data (tape, sever, outsourcing), and Determining the level of protection required for the audience that may view the organization’s data.

Another issue related to data classification is defining the level of classification for data. There is no exact and firm rule on the level of classifications. Some guidelines include issues such as the data type, its level of sensitivity and corporate objective and regulatory rules. Such data classification will be different for different organizations based on policies of each organization and government regulatory polices. As such policies are expressed in human understandable language and are vague and difficult to represent formally. The excessive gap between precision of classic logic and imprecision and vagueness in definition of polices creates difficulty in representing this policies in formal logic. Fuzzy

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Logic [Zadeh, L. A. (1965).] has been found to be useful in its ability to handle vagueness. In this article a data classification method based on fuzzy logic [Zadeh, L. A. (1965)] is presented to determine data classification levels for data in an organization. The level of sensitivity and corporate objective and regulatory rules are determined using this classification method. Classification levels could divide data into classes such as “top secret”, “secret”, “confidential”, “mission critical”, “not critical”, “private but not top secret”, and “public”. Based on these class categories the business processes and individuals that access and use the data and the level of encryption can be identified. The users can be categorized to determine access to any of these data classes. The users can be classified into classes based on need-to-know such as “very high”, “high”, “medium” and “low” users. The need for encryption level of the data can also be determined to be high, medium, zero (not necessary). To classify data with minimal resources impact and without needing to re-design databases one option is to add extra information to each data item by adding meta-data information to the attributes of each entity in relational-data bases and domains in classes in object-oriented databases. These meta-data information could be the value or degree of security, privacy or other related policies for that data item. This can be demonstrated using a simple relational database

as described below. For example consider the following entities of a relational database system: Patient(PatientID, Name, Address, TelNo, InsuranceID) Insurance(InsuranceID, Type, InsuranceProviderID) InsuranceProvider(InsuranceProviderID, Name, Address, TelNo, FaxNo) Doctor(DoctorID, Name, OfficeNo, TelNo, PagerNo) PatientDoctor(PatientDoctorID, PatientID, DoctorID, VisitDate, Notes)

Adding meta-data values can then be used for adaptation and implementation of classification of data in databases for an organization. The meta-data values can be can be obtained from the knowledge workers of the organization based on organization policies, procedure and business rules as well as government requirements for data privacy and security. For example Table 3 shows the metadata value related to security of attributes of table Patient based on organization’s security policy and government security and privacy policy. The values are in the range of 0 to 70, where zero indicates the meta-data for a data item that is public and 70 indicates the meta-data for a data item that is top secret (note that other meta data values are also possible and for this application we have chosen between values zero to seventy). Now assume that the following domain metadata values for these linguistic variable, TP = top secret, SE = “secret”, CO =“confidential”,

Table 2. Formulas for calculation triangular fuzzy memberships

m A ( x ) = 0 , x < a1

x - a1 mA ( x ) = , a1 ≤ x ≤ a2 a2 - a1 a -x mA ( x ) = 3 , a2 ≤ x ≤ a3 a3 - a2 m A ( x ) = 0 , x > a3

Formulas for calculation trapezoidal fuzzy memberships

m A ( x ) = 0 , x < a1 mA ( x ) =

x - a1 , a1 ≤ x ≤ a2 a2 - a1

m A ( x ) = 1 , a2 ≤ x ≤ a3 mA ( x ) =

a4 - x , a3 ≤ x ≤ a4 a4 - a3

m A ( x ) = 0 , x > a4

207

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Table 3. Metadata values for table customer Patient Table

Meta-data Value base on organization

Meta-data Value base on government

policy

regulatory policy

PatientID

68

39

Name

64

70

Address

30

60

TelNo

44

68

MC = “mission critical”, NC = “not critical”, PR = “private but not top secret”, PU = “Public”. Assume that the linguistic terms describing the meta-data for the attributes of entities in the above database has are: TP = [58,..,70], SE = [48,..,60], CO =[37,..,50], MC = [28,..,40], NC = [16,..,30], PR = [8,..,20], PU = [0,..,10]. Based on the metadata value for each attribute the membership of that attribute to each data classification can be calculated. In the Figure 3 triangular and trapezoidal fuzzy set was used to represent the data security classifications (e.g. Data security classification levels: TP = “top secret”, SE = “secret”, CO =“confidential”, MC = “mission critical”, NC = “not critical”, PR = “private but not top secret”, PU = “Public”). The membership value of PatientID based on its meta-data can be calculated for all these classification using the formulas in Table 2. Where x is metadata value for the attribute PatientID and α1, α2 and α3 are the lower middle and upper bound values of the fuzzy set data security classification. The degree of membership value

of the attribute PatientID to fuzzy set security classification based on meta-data from Table 3 can then be calculated as shwon in Figure 3. Now that the data can be classified and categorized into fuzzy sets (with membership value), a process for determining precise actions to be applied must be developed. This task involves writing a rule set that provides an action for any data classification that could possibly exist. The formation of the rule set is comparable to that of an expert system, except that the rules incorporate linguistic variables with which human are comfortable. We write fuzzy rules as antecedent-consequent pairs of If-Then statements. For example: IF Organizational_Security_Classification is TopSecret and Government_Security_Classification is Confidential Then Level of Encryption is High

The overall fuzzy output is derived by applying the “max” operation to the qualified fuzzy outputs each of which is equal to the minimum of the firing strength and the output membership

Figure 3. Fuzzy membership of metadata value of PatientID based on

μ(PatientID)

TP

SE

CO

MC

NC

PR

PU

0.8

0

0

0

0

0

0

(a) Organization policy

μ(PatientID)

TP

SE

CO

MC

NC

PR

PU

0

0

0.3

0.16

0

0

0

(b) government regulatory policy

208

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Figure 4. Center of gravity inference method A 1

A 2

B 1

X

Y

A 34

w 1

A

Z B 2

w 2 X x

y

Y

Z

intersec tion

Max

or min

z (Centroid of area)

function for each rule. Various schemes have been proposed to choose the final crisp output based on the overall fuzzy output. In this article a type of inference method called centre of gravity and illustrated in Figure 4. Securing medial data seems to be uncomplicated, yet the main danger of compromising such data comes from the people managing it, e.g. doctors, nurses and other medical staff. For that, it is noted that even though the transmitted medical data is classified and encrypted, doctors have to run application level encryption on their wireless mobile devices in order to protect this important data if the devices gets lost, left behind, robbed, etc. Nevertheless, there is a compromise. Increasing security through using multiple layers, and increasing length of encryption keys decreases the encryption\decryption speed and causes unwanted time delays, whether we were using application or hardware level of encryption. As a result, this could delay medical data sent to doctors or on-line monitoring units.

Future trends RFID in medical environment is an innovative and applicable idea. Linking RFID’s and wireless technologies will provide the required information to achieve timely services to patients as fast as possible. It also will pave the way for future paperless hospitals. RFID technology has many potential important applications in hospitals. With the progress the RFID technology is currently gaining, it seems to become a standard as other wireless technologies, and eventually manufacturers building them in electronic devices; biomedical devices with reduced cost.

conclusIon Managing patients’ data wirelessly can prevent errors, enforce standards, make staff more efficient, simplify record keeping and improve patient care. This research in the wireless medical environment

209

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

introduces new ideas in conjunction to what is already available in the RFID technology and wireless networks. Linking both technologies with each other to achieve the research main goal, delivering patients medical data as fast and secure as possible, to pave the way for future paperless hospitals. With the reduction in cost of radio frequency identification (RFID) technology, it is expected the increased use of RFID technology in healthcare in monitoring patients and assisting in health care administration. A fuzzy logic based for profile matching and data classification system is also developed to improve the application of regulatory data requirements for security and privacy of data exchange. Finally it should be noted that the world of RFIDs could be frustrating. Often potential users have to wade through product information to select the right RFID technology for their needs. The terminology, concepts, and uses seem to be vendor driver rather than potential user driver. The RFID vendor websites often describe how and where to use RFIDs using arcane and unfamiliar language.

acKnowledgMent The author would like to acknowledge the initial research work performed on this project at University of Canberra by MIT students.

reFerences Angeles, R. (2007), An empirical study of the anticipated consumer response to RFID product item tagging, Industrial Management & Data Systems, Vol. 107 No. 4, pp. 461-583. Bhuptani, M., & Moradpour, S. (2005). RFID Field Guide - Deploying Radio Frequency Identification Systems. NJ: Prentice Hall.

210

Bigus, J. P. and Bigus, J. (1998). Constructing Intelligent software agents with Java – A Programmers Guide to Smarter Applications, Wiley, ISBN: 0-471-19135-3. Butterfield, R. (2007). “Data classification: A prerequisite to ILM”, http://www.snwonline.com/ implement/data_class_05-30-05.asp Cline, J. (2007). “Growing pressure for data classification”, http://www.computerworld.com/ action/article.do?articleId=9014071&command =viewArticleBasic, Last accessed on 22/11/2007 Do, H. Rahin, E. (2002), “Coma: A System for Flexible Combination of Schema Matching Approaches”, Proceedings of 28th Conference on Very Large Databases, Morgan Kaufmann, pp 610-621USA. Doan, A-H. Lu, y. Lee, T. Han, J (2003), “ProfileBased Object Matching for Information Integration”, IEEE Intelligent Systems Magazine, pp 54-59. USA. Finkenzeller. K. (1999), RFID Handbook, John Wiley and Sons Ltd, USA. Fuhrer, P. and Guinard, D. (2007). Building a Smart Hospital using RFID technologies: Use Cases and Implementation. rfidehealth.pdf Glover. B and Bhatt H. (2006), RFID Essentials, O’Reilly Media, Inc.USA, (ISBN: 10-0596009445) Gruber, T.R. (1993) A translation approach to portable ontology specifications. Knowledge Acquisition, 5:199-220. Hedgepeth W. O. (2007). RFID metrics: decision making tools for today’s supply chains.Boca Raton, FL : CRC Press, (ISBN: 9780849379796) Kowalke, M. (2006) TMCnet. Wireless Mobility Blog. (2006). RFID vs. WiFi for Hospital Inventory Tracking Systems. http://blog.tmcnet.com/ wireless-mobility/rfid-vs-wifi-for-hospital-inventory-tracking-systems.asp, October 23, 2006.

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

Lahiri, S. (2005), RFID Sourcebook, IBM Press, USA, (ISBN: 10-0131851373 ) Last accessed on 22/11/2007 Lewan, T. (2007). ZDNet Australia. Human RFID chipping invading the US?. 23 July 2007. McGraw, G. (2006), “Software Security”, Addison-Wesley, USA. Messmer, E. (2006). Network World. Hospital finds a bloody good RFID application. http://www.computerworld.com.au/index.php/ id;2117167769;fp;4;fpid;18. Mickey, K. (2004), RFID grew in 2002. Traffic World 5: 20-21. Odell, J. (Ed.), (2000). Agent Technology, OMG Document 00-09-01, OMG Agents interest Group, September 2000 Pramatari, K.C., Doukidis, G.I. and Kourouthanassis, P. (2005), “Towards ‘smarter’ supply and demand-chain collaboration practices enabled by RFID technology”, in Vervest, P., Van Heck, E., Preiss, K. and Pau, L.F. (Eds), Smart Business Networks, Springer, Germany, pp. 197-210. Qiu R, Sangwan R. (2005), Toward collaborative supply chains using RFID. CIO Wisdom II, more best practices. New Jersey: Prentice-Hall; pp. 127–44.

Schuster E. W., Allen S. J. and Brock D. L. (2007) Global RFID : the value of the EPCglobal network for supply chain management. Berlin ; New York : Springer, (ISBN: 9783540356547). Shaalana, K. El-Badryb,M. and Rafeac, A. (2004) , “A multiagent approach for diagnostic expert systems via the internet”, Expert Systems with Applications, Volume No. 27 Issue No. 1, Elsevier Publishing, USA pp1-10. Shepard S. (2005). RFID : radio frequency identification, New York : McGraw-Hill, (ISBN: 0071442995). Tindal, S. (2008). ZDNet Australia.. One in ten RFID projects tag humans. http://www.zdnet.com.au/ news/hardware/soa/RFID-people-tagging-benefits-health-sector-/0,130061702,339285273,00. htm. 21 January 2008. Weinstein, R. (2005). RFID: A Technical Overview and Its Application to the Enterprise. IT Professional Magazine, 7(3), 27-33.Whiting, R. (2004), MIT = RFID + Rx. Information Week 988: 16. Wooldridge, M. and Jennings, N. (1995). Intelligent software agents: Theory and Practice. The Knowledge Engineering Review. vol. 10, no 2, pp 115-152, USA. Zadeh, L. A. (1965). “Fuzzy sets”, Information and control, Vol. 8. pp 338-352.

211

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

appendIx Fuzzy logic data classification Knowledgebase: IF Organizational_Classification is Top Secret and Government_Classification is Top Secret Then Level of Encryption isl High IF Organizational_Classification is Top Secret and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Top Secret and Government_Classification is Confidential Then Level of Encryption is High IF Organizational_Classification is Top Secret and Government_Classification is Mission Critical Then Level of Encryption is High IF Organizational_Classification is Top Secret and Government_Classification is Not Critical Then Level of Encryption is High IF Organizational_Classification is Top Secret and Government_Classification is Private but not Top Secret Then Level of Encryption is High IF Organizational_Classification is Top Secret and Government_Classification is Public Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Top Secret Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Confidential Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Mission Critical Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Not Critical Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Private but not Top Secret Then Level of Encryption is High IF Organizational_Classification is Secret and Government_Classification is Public Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Top Secret Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Confidential Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Mission Critical Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Not Critical Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Private but not Top Secret Then Level of Encryption is High IF Organizational_Classification is Mission Critical and Government_Classification is Public Then Level of Encryption is High IF Organizational_Classification is Not Critical and Government_Classification is Top Secret Then Level of Encryption is High IF Organizational_Classification is Not Critical and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Not Critical and Government_Classification is Confidential Then Level of Encryption is High IF Organizational_Classification is Not Critical and Government_Classification is Mission Critical Then Level of Encryption is High IF Organizational_Classification is Not Critical and Government_Classification is Not Critical Then Level of Encryption is Medium IF Organizational_Classification is Not Critical and Government_Classification is Private but not Top Secret Then Level of Encryption is Medium

212

Intelligent Agent Framework for Secure Patient-Doctor Profiling and Profile Matching

IF Organizational_Classification is Not Critical and Government_Classification is Public Then Level of Encryption is Zero IF Organizational_Classification is Private but not top secret and Government_Classification is Top Secret Then Level of Encryption is High IF Organizational_Classification is Private but not top secret and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Private but not top secret and Government_Classification is Confidential Then Level of Encryption is Medium IF Organizational_Classification is Private but not top secret and Government_Classification is Mission Critical Then Level of Encryption is Medium IF Organizational_Classification is Private but not top secret and Government_Classification is Not Critical Then Level of Encryption is Medium IF Organizational_Classification is Private but not top secret and Government_Classification is Private but not Top Secret Then Level of Encryption is Medium IF Organizational_Classification is Private but not top secret and Government_Classification is Public Then Level of Encryption is Medium IF Organizational_Classification is Public and Government_Classification is Top Secret Then Level of Encryption is High IF Organizational_Classification is Public and Government_Classification is Secret Then Level of Encryption is High IF Organizational_Classification is Public and Government_Classification is Confidential Then Level of Encryption is Medium IF Organizational_Classification is Public and Government_Classification is Mission Critical Then Level of Encryption is Medium IF Organizational_Classification is Public and Government_Classification is Not Critical Then Level of Encryption is Zero IF Organizational_Classification is Public and Government_Classification is Private but not Top Secret Then Level of Encryption is Medium IF Organizational_Classification is Public and Government_Classification is Public Then Level of Encryption is Zero

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 3, edited by J. Tan, pp. 38-57, copyright 2008 by IGI Publishing (an imprint of IGI Global).

213

214

Chapter 15

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care Jongtae Yu Mississippi State University, USA Chengqi Guo James Madison University, USA Mincheol Kim Jeju National University, South Korea

aBstract In the advent of pervasive computing technologies, the ubiquitous healthcare information system, or U-health system, has emerged as an innovative avenue for many healthcare management issues. Drawing upon practices in healthcare industry and conceptual developments in information systems research, this paper aims to explain the latent relationships amongst user-oriented factors that lead to individual’s adoption of the new technology. Specifically, this study focuses on the introduction of chronic disease U-health system. Using the Ordinary Line Square (OLS) regression analysis, we are able to discover the insights concerning which constructs affect service subscriber’s behavioral intention of use. Based on the data collected from over 440 respondents, empirical evidences are presented to support that factors such as medical conditions, perceived need, consumer behavior, and effort expectancy significantly influence the formation of usage intention.

IntroductIon Ubiquitous computing can be defined as the “contemplation of today’s computers commuDOI: 10.4018/978-1-61692-002-9.ch015

nicating each other through wireless network in actual activities of everyday life” (Weiser, 1993a; Weiser, 1993b). Some articulate such concept as the embedded computers in walls, refrigerator, tables, and objects in the surrounding environment (Rhodes & Mase, 2006). In other words,

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

computer is expected to become an ubiquitous resource, much similar to the light with a switch and water with a tap. Two functionalities, computing tasks and telecommunications, are required to realize the features of ubiquitous computing such as localized information, localized control, and resource management (Weiser, 1993a; Rhodes & Mase, 2006). The evolving mobile technology has expanded the applicability of ubiquitous computing to areas including virtual reality, head mounted display (HMD), wearable computing, and smart office room (Weiser, 1993b). An agile, responsive, and location-aware service delivery system is highly desired by the healthcare management business. Correspondingly, ubiquitous computing allows patients to receive prompt medical care anywhere (home, office, outdoor, or hospital) and any time (24 hours / 7 days), thus improving the service quality and decreasing the risk of medical treatment failures. For example, a doctor can check the status of a patient in a real time manner using the sensor which is installed in patient’s home or attached to the patient’s body. In case of emergency, the sensor can detect changes of the patient’s health condition in the early stage and automatically contact the designated hospital to initiate treatment procedure. Moreover, medical expenses have been continuously surging in the past and present years due to aging population and increasing demand for chronic disease management. Facing the soaring healthcare cost, many countries attempt to encourage patients to leave hospitals as early as possible and introduce medical services that are portable and effective for treatments. These efforts require treatment that delivers remote medical service to patient’s own residence (Kang & Lee, 2007). Hence, a medical service utilizing ubiquitous healthcare technology offers an innovative solution that allows patients to access medical service whenever and wherever. In the sense or market demand, taking South Korea as example, the U-Health market is expected to be worth $150

million (in USD) in 2010 with 7 million subscribers in their 30s~40s (Jee et. al., 2005). In summary, we define the U-Health system as the use of ubiquitous computing technologies to support expeditious and personalized communications, activities, and transactions between a medical service provider and its various stakeholders. In the literature review section, Figure 3 describes the layout of U-health system and its associated stakeholders. Despite the strong potential of the technology, however, the studies of ubiquitous healthcare services are not widely conducted. Existing literature has focused mostly on pure technical concerns or system development process; whereas managerial issues and behavior perspectives of U-health system application are largely overlooked. In this paper, the authors discuss the issue of applying ubiquitous computing technology to healthcare management from end user’s perspective. The main goal of this paper is to identify and investigate the factors and their inter-relationships that affect end user’s intention of adopting U-Health system. For example, one critical factor is effort expectancy, which refers to the degree to which the user believes it is easy to use the technology. Theoretically, the effort expectancy is positively correlated with behavioral intention of use, namely, the easier the user finds to use the product, the more likely s/he will adopt it. A more thorough articulation of these factors can be found in the Research Model and Hypotheses section. Realizing that the scope of healthcare discussion can be extremely broad, we focus our attention on chronic disease treatment in this research. A main question to be addressed is what factors influence end user’s perception about the U-health system and how these factors are related with each other. Therefore, an investigation of people’s subjective perception is warranted. The authors select South Korea as the place for data collection because: first, people in South Korea have a high level of concern about health issues, especially chronic diseases. Largely due to the

215

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

rapidly changing socio-economic structure, more than 16% of total population is reported to suffer from chronic diseases. According to a report by the Korean National Insurance Corporation (2007), South Korea is among one of the top countries that are mostly concerned with chronic diseases; second, South Korea has one of the highest mobile technology penetration rates in the world, thus providing a suitable environment for studying ubiquitous healthcare technology. In South Korea, more than 85% of Koreans use at least one cellular phone and subscribe wireless value added service (Shim, 2007). Regarding the organization of this study, the authors first examine the current knowledge base in both medical research and supportive technology aspects. Such literature review process enables us to consolidate theoretical and practical findings so that a research model is developed to cater for our core research issue – understanding the perceptions of chronic disease patient towards U-Health system. Next, we present the data analysis results that empirically validate the proposed hypotheses. At last, we conclude our findings and contributions.

lIterature reVIew It is widely recognized that the IT innovation has historically and continually benefited healthcare business in numerous aspects. For instance, collaborative computer system allows healthcare workers to reengineer and streamline business processes within medical service life cycle by leveraging software and network applications. In the case of U-Health system implementation, we need to consider both soft and hard factors. Soft factors are subtle factors containing people’s experiences, feelings and perceptions; whereas hard factors refer to the instantiations of computing technology (e.g. a HIPPA compliant patient accounts management system). Following such idea, the purpose of reviewing previous literature is two fold: to establish the scope of this study; to

216

build preliminary ground work for utilizing earlier findings in the area of healthcare management to guide our research activities.

prior research in Healthcare Management Information systems The evolution of HMIS can be summarized as a transferring process from isolation to integration (Briggs, Nunamaker, & Sprague, 2005). Initially installed in stand alone PCs, HMIS took shape among segments of healthcare management process, such as drug inventory and account administration (Tan, 1995). Due to the soaring operational cost of healthcare and increasing need for cross-boundary collaboration (Bandyopadhyay & Schkade, 2000), a true HMIS was introduced and adopted by organizations including the Health Maintenance Organization (HMO). Information and communication technology provides healthcare industry with capabilities to perform tasks that are deemed impossible in the past. For instance, a medical center uses broadband video conferencing technology to enable a group of specialists to examine the patient, who is located in a far distanced hospital. Previous studies also indicate that new interaction technologies utilizing human movements may provide more flexible, naturalistic interfaces and support the ubiquitous or pervasive computing paradigm (Abawajy, 2009). Practical implications – In pervasive computing environments the challenge is to create intuitive and user-friendly interfaces. Application domains that may utilize human body movements as input are surveyed here and the paper addresses issues such as culture, privacy, security and ethics raised by movement of a user’s body-based interaction styles. Since technology adoption has been considered as one of the main driving forces behind the growth of healthcare expenditures (Barros, Pinto, & Machado, 1999), numerous studies have been done to investigate the factors influencing HMIS adoption on the organization’s side (Berta, et al.,

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

2005; Wong & Legnini, 2000). These studies make contributions in terms of designing and validating cost-effective procedures that facilitate system adoption process and identifying environmental factors that influence the success of systems adoption. On the other hand, the authors found that extensive attention has been put on physician’s intention of using invested Information Technology (IT) products. However, further research is needed to investigate the role of patient, who is the ultimate service subscriber, in the process of healthcare system implementation (Hu, Chau, Sheng, & Tam, 1999). Another viable lens for explaining subscriber side information system usage is the Technology Adoption Model (TAM), which was formalized by F.D. Davis in 1989. Davis argued that the user’s intention of use is essentially determined by two constructs: Perceived Usefulness (PU) and Perceived Ease of Use (PEOU), as seen in Figure 1. Parsimonious and generalizable notwithstanding, TAM becomes less robust in this research context because it fails to address some unique dimensions of U-Health technology adoption. Therefore, we need modifications (e.g. broken down PU) that customize TAM constructs so as to accommodate the issue of chronic disease healthcare management. For instance, PU can be articulated as the formative relationship between Health Condition and Health Concern, as shown in Figure 2. Patient’s awareness of his/her health condition leads to health concern, which essentially form the evaluation of PU of suggested medical solutions. In this way, an example has

been set up concerning how PU can be adjusted to apply technology acceptance theory under different research topics. A comprehensive discussion of research model construction is presented in the Research Model and Hypotheses section.

the u-Health system concept and targeted problem domain Global companies including IBM, Intel and Google are investing in U-Health market while GE and Philips are leaders in the world’s medical equipment market. As Table 1 indicates, the U-Health sector in medical industry has become a promising business. Leveraging computing devices that are available yet invisible in the physical environment, U-Health system is a builtin application combining technologies, methods, and procedures that aim to monitor, maintain, and improve individual’s health condition (Park, 2003). One salient feature of U-Health system is the anytime anywhere accessibility, which allows the real time information of patient to be collected and then transmitted to medical organization for a cure or diagnosis (Yoo, 2006). With its surging

Figure 2. Modifying PU (Davis, 1989)

Figure 1. The TAM model (Davis, 1989)

217

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Table 1. The movement of overseas companies for U-Health business (Kang & Lee, 2007) Company

U-Health Business Activities

IBM

Providing insurance companies and medical service suppliers with various U-Health solutions such as remote monitoring and individual health measurement.

Microsoft

Investing on innovative health care technology including software and patents from Global Care Solutions (GCS) located in Bangkok.

Philips

Sold its semiconductor business in 2006 and focused on healthcare and lifestyle market. Introduced its TV-based custom health care service (Motiva) for old patients who are not familiar with computer network.

Intel

Created Digital Health department to enter digital health market of hospital computerization and home care service in 2005. Announced healthcare solution in joint with LG (Korea) to branch out into home health care service market 2007.

Google

Cooperated with Cleveland Clinic (US) for a program by which patients keep their own medical information in Google account in 2008.

penetration rate in global range, personal mobile device such as cellular phone offers an important platform on which the U-Health system concept can be realized. The challenge, however, lies in the seamless integration between mobile technology and the existing data network infrastructure. The current global healthcare industry is confronted with numerous issues. For instance, modern e-health systems incorporate different healthcare providers in one system. Further, medical information is distributed and shared on one common electronic platform, which contains large quantity of cross-context communications among stakeholders (Deng, Cock & Preneel, 2009). Since U-Health valuates medical treatments and information portability, we need to look deeper in a context oriented perspective. This study hence puts a specific focus on the implementation of U-Health system in South Korea for chronic disease patients, whose number is increasing steadily and raises health concerns of the society. Treating chronic disease such as diabetes, tuberculosis, and anemia requires long term treating efforts and cares from both patient and medical center. In addition, effective communication between patients and medical service provider is critical in stabilizing patient’s condition (e.g. diabetes). Hence, U-Health system offers a

218

promising solution that enhances information dissemination throughout the healthcare service delivery cycle. A visual description of a typical U-Health system implementation is presented in Figure 3. Using wired and wireless network, UHealth system has the following advantages (Korean National Insurance Corporation, 2007): • • • • •

Delivering a time-efficient medical service. Automating communications among stakeholders. Real time detection of patient’s conditions. Easy portability of patient’s historical medical records and prescriptions. High level of accessibility and flexibility mitigating the effects of time and place constraints.

In Figure 3, a sensor, which can be a patient’s cellular phone or wearable computing device, measures the health status of patient and transmits the information to U-Health service provider, who maintains a large collection of patients’ records in database server. The patient’s status is then forwarded to medical personnel in hospital for diagnose and possible treatment, which can be transmitted back to the patient over the same network.

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Figure 3.The U-Health System

researcH Model and HypotHeses Many of the contemporary health problems, especially chronic diseases, are mostly related with life style changes such as smoking or unhealthy dietary pattern (Riska, 1982; McKeown, 1971). Medical research model that is based on individualisticmechanistic notion of disease argues that an individual’s willingness to improve his/her health status results in the changes of his/her behavior and thus promotes health status positively (Cohen & Cohen, 1978). Such willingness can be generated and motivated by the perceived personal health condition, namely, the medical concern. Individual who has a negative evaluation of his/her health

condition is proactive to adopt healthcare service or activity that is believed to improve one’s health condition (Lahiri & Xing, 2004). Further, health condition is considered a critical starting point in the healthcare utilization process (Windmeijer & Santos Silva, 1997). Therefore, it is assumed that negatively perceived health condition leads to more medical service usage or activity to improve the health status. When people perceive health problems, they are likely to raise their health concerns about subsequent medical consequences. Such concern, in turn, can lead to certain activities including health information seeking, adherence to treatment, and interpretation of symptoms (Uskul & Hynie, 2007). Earlier studies have argued that the

219

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Figure 4.Mediating Effect of Medical Concern

Internet has become the new alternative source of healthcare information (Madden, 2003; Eysenbach & Köhler, 2003; Morahan-Martin, 2004). Other research has suggested that patients diagnosed with chronic diseases appear to be active in seeking the medical information. Therefore, a high level of concern about health problem and consequence leads to more active medical service usage. Hence, it is important to specify and validate the mediating effect of medical concern, as shown in Figure 4, between medical condition and action taking. Following the previous discussion, the authors contend that individuals who are actively involved with preventive or restorative healthcare measures are more likely to use the U-health service. Such argument can be further decomposed into the following hypotheses: H1: High level of medical concern leads to active medical activity. H2: Perceived medical condition leads to active medical activity. H3: Perceived medical concern mediates the relationship between medical condition and medical activity. H4: Active medical activity leads to behavioral intention of use. Specifying the relation between consumer needs and purchase intention has been a major research topic across various fields (Bhaskaran & Hardley, 2002; Odom, Kumar, & Saunders, 2002; Burke & Payton, 2006; Kervenoael, Soopra-

220

manien, Elms, & Hallsworth, 2006; Chae, Black, & Heitmeyer, 2006). The capability of identifying and satisfying the consumer’s needs basically determines the success of promoting a certain product or service (Lahiri & Xing, 2004). Several studies (e.g. Junginger et al., 2006; Zhang, Fang, & Olivia, 2007) have provided strong evidence indicating the effects of consumer’s perceived need on the purchase intention. On the other hand, such need is positively correlated with perceived quality and perceived usefulness of the service or product. Moreover, users accept new technology when they expect certain performance improvement in completing tasks. If people believe the technology will be useful in increasing their job performance, they are more likely to adopt the technology (Davis, 1989; Venkatesh et al., 2003). Hence, we hypothesize that: H5: High level of perceived need for U-health system leads to behavioral intention of use. It has been established in previous IS adoption literature that the behavioral intention of use is affected, although not exclusively, by Effort Expectancy, which is defined as the degree of ease associated with the use of the system (Venkatesh et al., 2003). The concept of effort expectancy is rooted in earlier constructs such as perceived ease of use in Technology Adoption Model (TAM) (Davis, 1989) and perceived level of difficulty in Theory of Reasoned Action (TRA) (Ajzen & Fishbein, 1975). In our research model, we

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

articulate effort expectancy in two dimensions: experience and consumer behavior. Several studies have identified how significantly experience (e.g. learning) is related to job performance (Guido, 2007; Ngwenyama, Guergachi, & McLaren, 2007). These results strongly support the argument that experienced workers are more efficient and more productive in performing tasks than less experienced ones. Hence, it is inferred that experienced user of ubiquitous or mobile technology are more likely to use the medical service offered through U-health system, namely, the less time a user needs to learn how to use U-Health system indicates a lower level of adoption barrier. Consumer behavior theory classifies the types of consumers according to how early they purchase the products. Innovators, or early adopters, are willing to take perceived risks to adopt newly introduced technology. So it is expected that individual who follows the behavior of innovator would adopt the U-health service earlier than the follower or late adopter. Some researchers have identified certain correlation between the early adoption behavior and business success on the organizational level. For instance, Hendricks and his colleagues (2007) point out that early adopter firms report higher profitability improvements of ERP systems than late ERP adopter firms. Moreover, Smith (2006) suggests that large companies are inclined to be the early adopter of mobile commerce and thus reap the first mover benefits. To sum up, we hypothesized that: H6: High level of mobile technology experience leads to behavioral intention of use. H7: Low level of expected time to learn how to use U-Health system leads to behavioral intention of use. H8: Early adopter behavior leads to behavioral intention of use. Finally, the recommendation construct is considered in our model because, for managers, it is critical to estimate the future purchase or

repurchase of the services or products (Gupta & Stewart, 1996; Morgan & Rego, 2006). Hence, the hypothesis of the inter-relationship between behavioral intention of U-Health system use and customer recommendation is stated as the following: H9: High level of behavioral intention of use leads to recommendation. Based on the above hypotheses, we propose our research model in Figure 5. In addition, we summarize the previous hypotheses in Table 2.

researcH MetHodology To test the hypotheses, the authors employ the Ordinary Line Square (OLS) regression analysis to pinpoint predictor variables’ contribution in explaining the variances of dependent variable – behavioral intention of use, which is also a mediator variable that describes how recommendation effects will occur. In academia, it has always been a fundamental dilemma for social researchers when attempting to maximize the three dimensions of research methodology: generalizability, precision, and realism. In other words, one cannot increase one of these three features without reducing the other one or two (McGrath, 1982). Therefore, it is up to researchers to justify their selection of research process that has the best goodness of fit with the research context. In this case, the authors determine to focus on the generalizability dimension mainly because the aim of this research is to obtain a finding that can be generalized to wider population, namely, chronic disease patients. Although a certain level of the in-depth knowledge is lost, it would essentially lead to another research project, such as a qualitative case study, that can make up for the deficiency. As mentioned earlier, the research design of this study aims to account for usage intention formulation of U-Health system that provides

221

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Figure 5. The Research Model

Table 2.The Nine Hypotheses Independent Variable

Dependent Variable

Description

H1

Medical Concerns

Medical Activities

High level of medical concern leads to active medical activity.

H2

Medical Conditions

Medical Activities

Perceived medical condition leads to active medical activity.

H3

Medical Conditions

Medical Concerns

Perceived medical concern mediates the relationship between medical condition and medical activity.

H4

Medical Activities

Behavioral Intention

Active medical activity leads to behavioral intention of use.

H5

Perceived Need

Behavioral Intention

High level of perceived need for U-health system leads to behavioral intention of use.

H6

Technology Experience

Behavioral Intention

High level of mobile technology experience leads to behavioral intention of use.

H7

Expected Learning Time

Behavioral Intention

Low level of expected time to learn how to use U-Health system leads to behavioral intention of use.

H8

Early Adoption Behavior

Behavioral Intention

Early adopter behavior leads to behavioral intention of use.

H9

Behavioral Intention

Recommendation

High level of behavioral intention of use leads to recommendation.

services to chronic patients. However, it is generally difficult to directly assess the quality of prescription service, especially in the case of chronic patient. The expertise of healthcare

222

professional, the capability of healthcare system, and communication with customer play respective roles in determining the service quality. The authors contend that it is not our top priority

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Table 3. Test of the influence of medical concern and condition on medical activities Model

Un-standardized Coefficients B

Std. Error

(Constant)

1.214

.204

medical concern

.241

.037

medical condition

.225

.023

Standardized Coefficients

t

Sig.

5.949

.000

.270

6.541

.000

.399

9.670

.000

Beta

F=73.698, p=0.000, R2=0.249

to validate strengths of U-Health system in the aspect of service quality improvement, but to scrutinize the factors influencing user’s perception of adopting the system. To administrate the data collection process, a broad range of survey was distributed in South Korea, where there is a high level of concern about chronic disease. As a rule of thumb in social research, the satisfactory sample size is 200 effective responses at least. In this research the authors were able to obtain effective surveys from 447 respondents and thus achieving a good sample size in order to warrant the significance of statistical findings. In terms of respondent’s characteristics, most of feedbacks were collected from people who are aware of or understand chronic disease and have experience in using wireless computing technology.

data analysis and results In this section, the results are presented according to the following order: Healthcare Related Factors, Perceived Need, Effort Expectancy, and Recommendation.

Healthcare Related Factors As shown in Table 3, the medical concerns (H1) and medical conditions (H2) positively affect medical activities. In other words, individuals with a high level of medical concern, or poor medical condition, are more likely to be involved with medical activities such as taking medicine

and regular exercising to maintain or to improve their health status. To test the mediate effect, in which medical condition leads to medical activity through medical concern, the authors employ the four-step approach proposed by Baron & Kenny (1986). The first step is to examine the direct effect of independent variable (medical condition) on dependent variable (medical activity). The second step is to test the relation between independent variable and mediate variable (medical concern). The third step is to conduct regression analysis focusing on the inter-relationship between mediate variable and dependent variable. The final step is to conduct regression analysis of the overall effects of independent variable and mediate variable on dependent variable. The mediate effect is proved to be significant when the effect of the coefficient in the first step is not significantly different from zero and the coefficient in the third step is significantly greater than zero. If both conditions are satisfied, a fully mediated model is said to be found (Frazier & Tix, 2004). In this study, Table 4 indicates that the effect of medical condition on the medical activity through the medical concern (H3) denotes partially instead of fully mediated relationship. The major reason is because the coefficient of medical condition is significantly different from zero while the coefficient of the medical concern is also significantly greater than zero. Hence, it is argued that medical concern partially mediates the formative relationship between perceived health condition and medical activity.

223

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

The results in Table 5 show that the effect of medical activities on the behavioral intention (H4) is not significant in this case. However, the scatter plot, as shown in Figure 6, provides an alternative explanation that the relationship between the two variables is not linear. In the plot, behavioral intention (Y-axis) appears to be significant in the left and right areas across the line of medical activities (X-axis). In the middle area of medical activities, however, the correlation is scattered and loose. Such U-shaped regression curve is subject to receiving two transformations in statistical analysis: logarithmic form and quadratic form. In this study, quadratic model is applied and the correlation between medical activities and behavioral intention is significant at LOS.05, as shown in Table 6 and Figure 7. It can be inferred from previous discussion that individuals in two extreme groups are more prone to use U-health system. Those in the right

extreme (high medical activity group) have higher concerns and poorer health condition than other groups. Such concerns and conditions would lead to the usage of innovative technology that might be helpful to improve their health status. On the other hand, the individuals in the left extreme (low medical activity group) do not give much attention to their health status due to several restrictions. For example, because of high work pressure, many employees have a difficult time arranging healthcare consulting service on a regular basis, which serves as a complementary reason for increasing cases of chronic diseases. Therefore, these people may realize the potential advantage of U-health system and thus believe that U-health system is able to eliminate the restrictions of accessing medical services. Individuals who are in the middle group may be satisfied with the current treatments for their health status management and find few benefits in seek-

Table 4. Test of the influence of medical condition on medical concerns Step

Model

Step 1

Un-standardized Coefficients

T

Sig.

.024

9.779

.000

.051

.030

1.700

.090

.270

.040

6.680

.000

.225

.023

9.670

.000a

.241

.037

6.541

.000b

B

Std. Error

Medical condition → medical activity

.237

Step 2

Medical condition → medical concern

Step 3

Medical concern → medical activity

Step 4

Medical condition & concern → medical activity

a-medical condition / b-medical concern

Table 5. Test of the influence of medical activities on behavioral intention Un-standardized Coefficients

Standardized Coefficients

B

Std. Error

Beta

(Constant)

156.926

14.136

medical activities

-2.869

4.237

Model

F=.458, p=.499

224

-.032

t

Sig.

11.101

.000

-.677

.499

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Figure 6. The scatter plot between medical activities and behavioral intention

Table 6. Test of the influence of medical activities on behavioral intention (Quadratic model) Parameter Estimates

Equation Quadratic

Constant

b1

b2

236.413

-55.900

7.737

F=3.889, p=0.021, R =0.017 2

Figure 7. The scatter plot of quadratic estimation between medical activities and behavioral intention

225

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

ing other solutions due to the economic lock-in effect.

Referring back to the literature review, we have found a plethora of theoretical arguments validating and explaining the relationship between customer’s perceived need and purchase attention. Such argument is also verified in this research context. According to the statistical results shown in Table 7, the perceived need (H5) for U-health system positively affects behavioral intention. It strongly supports the fact that individuals having a high demand for U-health system are more likely to use the service than those who have lower need.

to consider that the less learning time is needed to use the system, the more likely the behavioral intention of use can be formed. Whereas for H8, since we use a reverse scale, that is, 1 represents strong and 5 means weak, for measuring early adoption behavior, there is negative correlation between those two factors. In congruence with the theory, individuals who are early adopters are more likely to accept U-health service. However, experience of using mobile technology (H6) is not significant at the 0.05 level. It can be attributed to the high mobile technology penetration rate in South Korea. Since people are already skillful in using mobile device, there is lack of concern in technology experience that leads to behavioral intention of U-Health system use.

Effort Expectancy

Recommendations

In Table 8, the analysis of effort expectancy indicates that early adoption behavior (H8) and expected learning time (H7) for considering adoption are negatively connected with behavioral intention at the 0.05 level of significance. It is relatively easy

According to the data analysis, which is shown in Table 9, higher behavioral intention leads to higher recommendation, which encourages the continuance of system use and reuse through customer recommendation after initial adoption behavior.

Perceived Need

Table 7. Test of the influence of perceived need on behavioral intention Unstandardized Coefficients

Model (Constant) Perceived Need

B

Std. Error

51.293

16.126

20.586

3.229

Standardized Coefficients

t

Sig.

3.181

.002

6.375

.000

Beta .323

F=40.640, p=0.000, R =0.104 2

Table 8. Test of the influence of adoption behavior, experience, and expected time on behavioral intention Un-standardized Coefficients

Model

Standardized Coefficients

t

Sig.

14.335

.000

-1.976

.049

.100

1.693

.092

-.352

-5.637

.000

B

Std. Error

(Constant)

311.662

21.742

Adoption behavior

-13.440

6.803

-.126

2.143

1.266

-23.756

4.215

Experience Expected time F=18.103, p=0.000, R =0.180 2

226

Beta

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

The results for all hypotheses testing are summarized in Table 10. An individual would not take action until s/he perceives his/her medical condition, which are supported in H1, H2 and H3. The impact of medical activities to behavioral intention of using medical product (e.g. U-health system) is non-linear. Two groups of individuals: high medical activity group, those who have higher concerns and poorer health condition, and low medical activity group, those who do not give much attention to their health status due to some restrictions (e.g. tight working schedule), are estimated to have high desire of adopting convenient health care services enabled by IT. The perceived need of individual leads to positive formation of behavioral intention of use. Whereas people will be reluctant in adopting the product if it takes too much efforts and the learning curve is too long, which leads to another inference that

the early adopters of new technology is estimated to have a lower barrier to embrace the new system. Once the behavioral intention is formed, the user is likely to recommend the product to others based on the previous positive experience obtained.

dIscussIon and conclusIon The authors argue that consumer’s intention of using U-Health system can be better understood if it is viewed through a multi-faceted framework. In this paper, we draw upon literature from different paradigms including specific medical research and technology acceptance and use. The paper’s basic premise is that medical factors, perceived economic needs, and effort expectancy exert formative impacts on consumer’s behavioral intention. Such premise is then broken into a series of hypotheses

Table 9. Test of the influence of behavioral intention on recommendation Un-standardized Coefficients

Model (Constant) Behavioral Intention

B

Std. Error

.511

.035

.001

.000

Standardized Coefficients

t

Sig.

14.498

.000

6.980

.000

Beta .314

F=48.718, p=0.05, R =0.099. 2

Table 10. Summary of Hypotheses Testing Independent Variable

Dependent Variable

Hypothesized Effect

Support

H1

Medical Concerns

Medical Activities

Positive

Yes1

H2

Medical Conditions

Medical Activities

Positive

Yes

H3

Medical Conditions

Medical Concerns

Positive

Yes (Partially)

H4

Medical Activities

Behavioral Intention

Positive

No (Non-linear)

H5

Perceived Need

Behavioral Intention

Positive

Yes

H6

Technology Experience

Behavioral Intention

Positive

No

H7

Expected Learning Time

Behavioral Intention

Negative

Yes

H8

Early Adoption Behavior

Behavioral Intention

Negative2

Yes

H9

Behavioral Intention

Recommendation

Positive

Yes

227

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

in Table 9 extending the discussion to a further detailed level. By validating these assumptions, we are able to obtain more insights in terms of how key adoption factors interact with each other. The statistical findings contend that high level of health concern and perceived health condition positively lead to relevant medical activities including regular sport exercises and consulting with doctors. Although the data analysis doesn’t support direct formative relationship between medical activities and actual U-Health system adoption (e.g. people who visit doctors might not intend to use U-Health system), a further investigation on the data indicates that two groups of potential users people who have high level of health concerns and poor health condition and people who are unable to manage their health issues due to limitations such as severe work pressure – are more likely to use the system. Moreover, as many other studies have argued, the perceived need or perceived value of the system is found positively related with behavioral intention of use. In congruence with seminal theories including TAM, the study shows that individuals who are early adopters are more likely to accept U-health service since the expected learning cycle is short, or, the system is easy to use. The expected learning time is measured using reversed scale hence it has a negative relationship with the behavioral intention of use. Some of the limitations in this study lie in the measurements development and some insignificant statistical findings, which hinder us from better validating our proposal. Future efforts are required to shed the light on designing and validating measures that would fill the voids in this study. Also, since no single research can maximize the three features of research process: generalizability, precision, and realism (McGrath, 1982), a different perspective can be obtained through a qualitative case study of U-health system or other innovative health care services.

228

reFerences Abawajy, J. H. (2009). Human-computer interaction in ubiquitous computing environments. International Journal of Pervasive Computing and Communications, 5(1), 61–77. doi:10.1108/17427370910950311 Ajzen, I., & Fishbein, M. (1973). Attitudinal and normative variables as predictors of specific behavior. Journal of Personality and Social Psychology, 27(1), 41–57. doi:10.1037/h0034440 Bandyopadhyay, K., & Schkade, L. L. (2000). Disaster recovery planning by HMOs: theoretical insights. Health Care Management Review, 25(2), 74–85. Barros, P. P., Pinto, C. G., & Machado, A. (1999). A signaling theory of excessive technological adoption. Health Care Management Science, 2(2), 117–125. doi:10.1023/A:1019075425763 Berta, W. Teare, G. F., Gilbart, E., Soberman G., L. Lemieux-Charles, L., Davis, D., & Rappol, S. (2005). The contingencies of organizational learning in long-term care: factors that affect innovation adoption. Health Care Management Review, 30(4), 282–292. Bhaskaran, S., & Hardley, F. (2002). Buyer beliefs, attitudes and behaviour: foods with therapeutic claims. Journal of Consumer Marketing, 19(7), 591–606. doi:10.1108/07363760210451410 Briggs, R. O., Nunamaker, J. F. Jr, & Sprague, R. H. Jr. (2005). Special issue: crossing boundaries in information systems research. Journal of Management Information Systems, 22(4), 7–11. doi:10.2753/MIS0742-1222220401 Burke, R., & Payton, P. E. (2006). Measuring and managing shoppability: Ten principles to convert shoppers into buyers. European Retail Digest, 50, 66–78.

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

Chae, M. H., Black, C., & Heitmeyer, J. (2006). Pre-purchase and post-purchase satisfaction and fashion involvement of female tennis wear consumers. International Journal of Consumer Studies, 30(1), 25–33. doi:10.1111/j.14706431.2005.00434.x Cohen, C. I., & Cohen, E. J. (1978). Health education: panacea, pernicious or pointless? The New England Journal of Medicine, 28, 718–720. Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. Management Information Systems Quarterly, 13(3), 319–340. doi:10.2307/249008 Deng, M., Cock, D. D., & Preneel, B. (2009). Towards a cross-context identity management framework in e-health. Online Information Review, 33(3), 422–442. doi:10.1108/14684520910969880 Eysenbach, G., & Köhler, C. (2003). What is the prevalence of health-related searches on the World Wide Web? Qualitative and quantitative analysis of search engine queries on the Internet. In Proceedings of the AMIA Symposium (pp. 225-229). Frazier, P. A., & Tix, A. P. (2004). Testing moderator and mediator effects in counseling psychology. Journal of Counseling Psychology, 51(1), 115–134. doi:10.1037/0022-0167.51.1.115 Guido, F. (2007). The organizational learning curve. European Journal of Operational Research, 177(3), 1375–1384. doi:10.1016/j. ejor.2005.04.009 Gupta, K., & Stewart, D. W. (1996). Customer satisfaction and customer behavior: the differential role of brand and category expectations. Marketing Letters, 7(3), 249–263. doi:10.1007/BF00435741 Hendricks, K. B., Singhal, V., & Stratman, J. K. (2007). The impact of enterprise systems on corporate performance: a study of ERP, SCM, and CRM system implementations. Journal of Operations Management, 25(1), 65–82. doi:10.1016/j. jom.2006.02.002

Hu, P. J., Chau, P. Y. K., Sheng, O. R. L., & Tam, K. Y. (1999). Examining the technology acceptance model using physician acceptance of telemedicine technology. Journal of Management Information Systems, 16(20), 91–112. Jee, K. Y. (2005). Healthcare in ubiquitous era. Jinhan M&B. Junginger, M., de Visser, E., Hjort-Gregersen, K., Koornneef, J., Raven, R., Faaij, A., & Turkenburg, W. (2006). Technological learning in bioenergy systems. Energy Policy, 34(18), 4024–4041. doi:10.1016/j.enpol.2005.09.012 Kang, S., & Lee, S. (2007). Economic effect and growth strategy of U-Health. SERI (Samsung Economic research Institute). Issue Paper. Kervenoael, R. D., Soopramanien, D., Elms, J., & Hallsworth, A. (2006). Exploring value through integrated service solutions: the case of e-grocery shopping. Managing Service Quality, 16(2), 185–202. doi:10.1108/09604520610650646 Korean National Insurance Corporation (2007), Daily Report, Jan/29. Lahiri, K., & Xing, G. (2004). An econometric analysis of veterans’ healthcare utilization using two-part models. Empirical Economics, 29, 431–449. doi:10.1007/s00181-003-0178-0 Lim, H., Widdows, R., & Jungkun, J. (2006). M-loyalty: winning strategies for mobile carriers. Journal of Consumer Marketing, 23(4/5), 208–218. doi:10.1108/07363760610674338 Madden, M. (2003). America’s online pursuits: the changing picture of who’s online and what they do. Washington DC: Pew Internet and American Life Project. Retrieved from www.pewinternet. org/reports/toc.asp?Report=106 McGrath, J. E. (1982). Judgment Calls in Research. Beverly Hills, CA: Sage Publications, Inc.

229

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

McKeown, T. (1971). A Historical Appraisal of the Medical Task. In McLachlan, G., & McKeown, T. (Eds.), Medical History and Medical Care (pp. 29–50). London: Oxford University Press. Morahan-Martin, J. M. (2004). How internet users find, evaluate, and use online health information: a cross-cultural review. Cyberpsychology & Behavior, 7(5), 497–510. Morgan, N. A., & Rego, L. L. (2006). The value of different customer satisfaction and loyalty metrics in predicting business performance. Marketing Science, 25(5), 426–439. doi:10.1287/ mksc.1050.0180 Ngwenyama, O., Guergachi, A., & McLaren, T. (2007). Using the learning curve to maximize IT productivity: a decision analysis model for timing software upgrades. International Journal of Production Economics, 105(2), 524–535. doi:10.1016/j.ijpe.2006.02.013 Odom, M., Kumar, A., & Saunders, L. (2002). Web assurance seals: how and why they influence consumers’decisions. Journal of Information Systems, 16(2), 231–250. doi:10.2308/jis.2002.16.2.231 Park, K. (2003). Background to Ubiquitous Healthcare. No.272 Conference between Industry, Academic, Research Centers, Korea. Rhodes, B., & Mase, K. (2006). Wearables in 2005. IEEE Pervasive Computing / IEEE Computer Society [and] IEEE Communications Society, 5(1), 92–95. doi:10.1109/MPRV.2006.21 Riska, E. (1982). Health education and its ideological content. Acta Sociologica, 25, 41–46. Smith, A. D. (2006). Exploring m-commerce in terms of viability, growth and challenges. International Journal of Mobile Communications, 4(6), 4. Tan, J. K. H. (1995). Health management information systems: theories, methods, and applications. Gaithersburg, MD: Aspen Publishers.

230

Uskul, A. K., & Hynie, M. (2007). Self-construal and concerns elicited by imagined and real health problems. Journal of Applied Social Psychology, 37(9), 2156–2189. doi:10.1111/j.15591816.2007.00283.x Venkatesh, V., Morris, M. G., Davis, G., & Davis, F. D. (2003). User acceptance of information technology - toward to unified view. Management Information Systems Quarterly, 27(3), 425–478. Weiser, M. (1993a, October). Hot topics: ubiquitous computing. IEEE Computer. Weiser, M. (1993b). Some computer science issues in ubiquitous computing. Communications of the ACM, 36(7), 75–84. doi:10.1145/159544.159617 Windmeijer, F. A. G., & Silva, J. S. (1997). Endogeneity in count data models: an application to demand for healthcare. Journal of Applied Econometrics, 12, 281–294. doi:10.1002/ (SICI)1099-1255(199705)12:3<281::AIDJAE436>3.0.CO;2-1 Wong, H. J., & Legnini, M. W. (2000). The diffusion of decision support systems in healthcare: are we there yet? Journal of Healthcare Management, 45(4), 240–254. Wong, H. K. (2007). Consumerism and quality standards for subvented social services: a case study of Hong Kong. International Journal of Public Administration, 30(4), 441–458. doi:10.1080/01900690601158287 Yoo, S. (2006). Ubiquitous Health, Industrial Health. Korea & World Affairs, 5, 13–18. Zhang, J., & Fang, X., L. S., & Olivia, R. (2007). Online consumer search depth: theories and new findings. Journal of Management Information Systems, 23(3), 71–95. doi:10.2753/MIS07421222230304

Towards a Conceptual Framework of Adopting Ubiquitous Technology in Chronic Health Care

endnotes 1

2

The default level of significance for all hypotheses testing is.05. Revere scales are used for the measurement.

231

232

Chapter 16

E-Patients Empower Healthcare: Discovery of Adverse Events in Online Communities1 Roy Rada University of Maryland Baltimore County, USA

aBstract E-patients can empower themselves and improve healthcare. In online communities, patients may discuss adverse events that are inadequately addressed in the literature. The author as a patient joined various online patient discussion groups and identified several such adverse events. For each such adverse event, the patient findings, the medical literature, and the implications are noted. Extracts from the literature that were provided to the patients were welcomed by the patients. Possible approaches to financially supporting such activities are sketched.

IntroductIon The President of the United States in an address to the nation said (Obama, 2008): “In addition to connecting our libraries and schools to the internet, we must also ensure that our hospitals are connected to each other through the internet. That is why the economic recovery plan I’m proposing will help modernize our health care system – and that won’t just save jobs, it will save lives. We will make sure that every doctor’s office and hospital in this country is using cutting DOI: 10.4018/978-1-61692-002-9.ch016

edge technology and electronic medical records so that we can cut red tape, prevent medical mistakes, and help save billions of dollars each year.” The emphasis on electronic medical records is a natural one for a provider-centric nation. However, another benefit from greater digitization and connectivity of the world in health care matters is the emergence of patient power or the e-patient. A Dutch study showed that patients may use blogs to advance some of the principles of Web 2.0, as the author noted (Adams, 2008): “enabling patients to be more active in documenting and managing information related to their health experiences”. What is the role of the patient in the e-health world?

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

E-Patients Empower Healthcare

In the already classic, oft-cited article “Will Disruptive Innovations Cure Health Care”, the authors propose two key steps to improving the health care system (Christensen et al., 2000): •

•

create and then embrace a system where the clinician’s skill is matched to the difficulty of the medical problem and invest less money in high-end, complex technologies and more in technologies that simplify complex problems.

The empowerment of patients online fits nicely into this desiderata. Patients have varying expertises and common experiences across the globe. They can meet with one another and share stories and insights both about their medical conditions and their doctors. The technology is simple and already well developed for other commercial purposes. What evidence exists to support this contention that patients by moving online and sharing information about health can help the health care system? A E-Patient Network with support from the Robert Woods Johnson Foundation has developed a wikipedia-based white paper titled “e-Patients: How they can help us heal healthcare” for which the conclusion is this (Ferguson, 2009): “The creation of optimal health care may depend on our ability to embrace our first generation of e-patients, providing them with the autonomy, authority, and empowerment they desire and deserve and inviting them to join us in a combined effort to improve healthcare for everyone. It will be only by joining forces with these new colleagues that we can hope to solve the pervasive problems that plague the healthcare system: quality, cost, access, and consumer satisfaction.” Many different aspects of this e-patient revolution are being explored. For instance, the support that might be provided for patients seeking health care information online has been addressed with

the conclusion that patients seek information differently and sometimes inefficiently and ineffectively but online support tools could improve that search behavior (Keselman et al., 2008). Customer Relationship Management systems are extensively used in healthcare systems (Calhoun et al., 2006). Data mining of web information is an alternative way to learn what consumers think. This paper explores the means by, and extent to, which participants in online patient-patient discussion groups provide useful information about medical adverse events.An adverse event occurs when some intervention by a healthcare provider produces an unwanted reaction. For instance, radiation treatment for oral cancer can cause obstructive sleep apnea. The literature on adverse events addresses their causes, how to reduce them, and the impact they have on patients, staff, and health care organizations (Misson, 2001). Typically, health care professionals investigate adverse events through the medical record (Duff et al., 2005). Many online patient groups are established by volunteers on free sites, such as groups.yahoo.com (Rada, 2006). However, some healthcare entities maintain patient online discussion groups. For instance, Joslin Diabetes Center runs an online, diabetes discussion group for the public, and experts from the Center provide feedback online. Kaiser Permanente maintains numerous discussion groups moderated by Kaiser’s professionals, but access is restricted to enrollees in the Kaiser Plan. Healthcare professional in online moderator roles might address adverse events, among other things. Listening to patients is a key to reducing adverse events (Cleary, 2003): “by relying on the observations and insights of patients such as Mr. Q., the physicians and staff will be able to close the gap between Mr. Q.’s experience and what they can achieve.”

233

E-Patients Empower Healthcare

The book “Partnering with Patients to Reduce Medical Errors” states (Spath, 2004): “The health care team can only be as strong as its weakest link, and unfortunately, the weakest link is often the recipient of care – the patient.” Patients in online groups hold a unique and valuable position because of their sheer numbers and an intense focus on their shared illness. Patient groups may have contact with larger numbers of disease-specific patients than many physicians and have the luxury of spending many hours discussing similarities and differences. After hundreds of hours of conversation, patterns can begin to emerge. These patterns might lead to new insights about adverse events. Norman Scherzer started a patient discussion group to explore issues of how cancer patients, such as his wife, might be best treated (Ferguson, 2002). Through sharing information the patients and their significantothers made discoveries about side effects of the treatment for the cancer that were published in a scholarly journal. This paper explores the discovery of adverse events through online patient groups.The hypothesis is that patients share information online, that when combined with information from the literature, can help identify important gaps in the medical literature. More generally, the argument is that these online groups can be an important resource for both patients and healthcare providers.

MetHod The author was trained as a medical doctor and became a cancer patient. As doctors become ill and see the world from the patient’s side, they often have useful insights to share about the relationship between patients and healthcare providers (Rosenbaum, 1988). Since an online discussion group is self-documenting by nature, the opportunity exists for a participant in a group to review the discussion and to engage in a kind of retrospective ethnographic analysis. Studying

234

online groups via ethnography is in many ways easier than studying face-to-face groups (Paccagnella, 1997). The term netnography has been coined by Kozinets to apply to such ethnography. According to (Kozinets, 2002): “As a method, netnography is faster, simpler, and less expensive than traditional ethnography and more naturalistic and unobtrusive than focus groups or interviews.” In the context of this research, an online group uses a software system that provides a searchable archive of previous messages. Members of the group create messages and post them to the system, and the system in turn distributes these messages to the group. The system may interface to a group member via an email client or a web site. The online groups noted in this paper include patients, their significant others, and sometimes others who want to help. This population will be typically represented with the umbrella term ‘patients’ with its meaning apparent in the context. This author joined two cancer patients groups as a patient in 2003. Both groups had formally welcomed participants to use personally de-identified information in the messages for research purposes. The groups had a total membership of several hundred. The author read the patient messages, identified messages of interest, studied relevant clinical, journal articles, and where appropriate shared extracts from the literature with the group. The author identified several cases where the information needs of the patients led to the discovery of adverse events and gaps in the medical literature. For each case, the patient findings, the medical literature, and the implications are noted. Four years after the treatment of his cancer, the author developed a long-term adverse effect of the cancer treatment. This lead the author to join a long-term care discussion group (also with hundreds of members) and to discover yet again the marvels of what patients can share that medical professionals rarely have the time or ability to share -- namely, the personal experiences of chronic disease suffers who are coping with their life and the medical establishment.

E-Patients Empower Healthcare

results The four cases follow: 1.

2.

Patients discussed ways to cope with their fatigue. In the online discussion groups, over a dozen patients reported signs and symptoms of Obstructive Sleep Apnea (OSA) in relation to this fatigue. The literature reveals incomplete information about OSA in headand-neck cancer patients. Two articles provide interestingly different perspectives on OSA as a complication of the treatment. In one article the incidence of OSA was 92% in treated patients (Friedman et al., 2001), while in the other article 8% of treated patients developed OSA (Rombaux et al., 2000). The literature at that time provided no mention of a radiated-only patient developing OSA, but one of the patients developed OSA after only radiation. The outcomes of these observations were two-fold. On one hand, extracts of the literature were shared with the online group, and the patients expressed gratitude for that literature information. Secondly, the observation of a gap in the literature became the basis of published, medical literature by this author (Rada, 2005a, 2007). Hyperbaric oxygen treatment for osteoradionecrosis of the mandible is routine in the United States. A patient in a discussion group presented his concerns about hyperbaric for osteoradionecrosis and said: “Every dentist that I have seen in San Antonio has recommended hyperbaric oxygen, but does anyone know if hyperbaric oxygen is worth the $50,000 cost?” The patient went to Mexico and was told hyperbaric oxygen was unnecessary. European studies have shown that hyperbaric oxygen is not appropriate (Annane et al., 2004), but the American literature defends it (Mendenhall, 2004). Differences in the standard of care in one country versus another and the standard of

3.

care versus the ability to pay for the care create a kind of adverse event for the patient. Again, when extracts from the literature were shared with the patients, they replied with messages including a ‘thank you’. A patient reported severe allergic response to a drug (amifostine) that was first being used for cancer patients as a radioprotectant. The literature at the time suggested that severe reactions to amifostine treatment were rare: “Amifostine administration was well tolerated, with a low incidence of side effects” (Antonadou et al., 2002). However, the patient group included two patients with severe allergic reaction. A year later the results of a clinical trial were published which confirmed what the patients feared (Rades et al., 2004): “Administration of amifostine during radiotherapy is associated with a high rate of serious adverse effects.” When a new drug use appears, detecting uncommon adverse events may be supported by having patient groups monitoring and discussing their reactions to their treatments. Some of the patients in turn took this group information to their doctors, and helped their doctors appreciate the problem.

Four years after treatment that patient began developing spasms in the muscles. Two years later this was diagnosed as radiation-induced neuropathy. The doctors said “progressive and irreversible”. Furthermore, no treatment existed and the disease progression was variable. Doctors are hardly able to say more than that. However, patients who are experiencing this problem are eager to say more. Unfortunately, in meeting with one’s doctor, the patient gets no access to other patients with this condition and generally speaking no other such patients are locally accessible -- the condition is relatively rare in that it affects about 1% of cancer survivors. The author joined a long-term care survivors group and discovered that radiation-induced

235

E-Patients Empower Healthcare

neuropathy was experienced by multiple people in that group. Their fate had been invariably an unpleasant one. The author’s doctor suggested an experimental treatment which had been tried with two patients elsewhere. The author found a few patients in the online group who had tried a similar treatment unsuccessfully. However, the opportunity to understand what is happening proved itself a powerful elixir -- “knowing is half the battle”. When patients are confused about an adverse event, the possibility exists that the health profession itself is also confused. The patients often do not share their confusion in a compelling way with the healthcare professional. However, in the patient group, the patients may be comfortable to elaborate. From this study, topics in which patients felt particularly unable to get adequate explanations from their healthcare professionals where topics which the healthcare industry had not adequately addressed and which would require the coordinated attention of healthcare professionals from different disciplines. Patients appreciated receiving extracts of the medical literature that pertained to their questions. While the information might have improved health outcomes, it also could lead to other positive outcomes. For instance, patients could contribute to community awareness initiatives. Or a healthcare professional might author a scholarly paper about a particular adverse event. The majority of the discussion in the online groups was not about the preceding adverse events. Much of the discussion was about emotional topics,such as a patient reporting the good news that the latest checkup with the oncologist revealed no progression of the disease and other patients congratulating the patient on the good news. Patients often complained about diffuse ill effects of the cancer treatment. However, the importance of these groups to the well-being of the participants seems obvious, and, at times, discoveries are possible that would seem unlikely to practically be obtained any other way.

236

categorIes and procedures Each of the preceding adverse events concerned more than one clinical specialty. The adverse events might be categorized as follows: •

•

•

A diffuse symptom: Obstructive sleep apnea secondary to treatment for head-andneck cancer may tend to be overlooked by otolaryngologists because the symptoms are diffuse and obstructive sleep apnea is often addressed by sleep specialists rather than otolaryngologists. Standard of care: Hyperbaric oxygen as part of the national standard of care for osteoradionecrosis of the mandible is not supported by clinical trials internationally, but the practicing otolaryngologist is not expected to dispute the national standard of care. Uncommon reaction to new drug: When the drug amifostine was initiated for a new purpose, researchers needed further experience to uncover adverse events.

The foremost causes of adverse events as reported by the US Institute of Medicine (Kohn et al., 2000) are technical errors, diagnostic errors, failure to prevent injury, and medication errors. That classification is, however, not necessarily the optimal one for understanding what can be gleaned from patient online groups. If a provider has decided to support an online discussion group and to provide moderators, then it might guide moderators relative to the findings of this study. To find evidence of adverse events that are inadequately appreciated in the literature, a healthcare professional might: 1. 2.

Join an online discussion group for patients with a particular chronic disease. Identify a finding that is highlighted by a patient as a problem. Findings may include symptoms, signs, laboratory or test results,

E-Patients Empower Healthcare

3.

4.

5.

observations, or specific events (such as hospitalization or receiving a bill). A finding is a problem when patient says so. Review the medical literature to determine whether a medical intervention experienced by the patient might have a causal relation to the problematic finding. Relevant PubMed ‘Medical Subject Headings’ are identified, a query is posed to PubMed, and full-text copies of journal articles are retrieved through membership in a medical library subscription program. Temporal relationship, strength of association, biological plausibility, and other relationships contribute to a judgment of causality (Darden & Rada, 1988). Determine whether the literature provides conflicting or unclear guidance. Sometimes the published literature suggests conflicting algorithms for diagnosis or treatment, and more research is needed to harmonize the literature. Extract information from the literature and return that information to the group. The extract should be clear to the intended audience, embedded within a personal context, and made as a reply to recently posted message that has not already received a similar response.

If an extract from the literature is simply posted without context or explanation, then the impact, as measured by patient response, is less. Information systems can support this work be parsing patient messages and semi-automatically linking to relevant citations from PubMed (Rada, 2005b).

dIscussIon If one takes the preceding categorization of adverse events from online groups and tries to generalize further, one might note that the problems occur where the otolaryngologist’s responsibility is blurred because someone else is also responsible.

In general, adverse events may be least well understood where 1) the responsibility for the adverse event falls among several medical specialties and 2) the medical specialists inadequately communicate with one another. The data from the online groups leads to qualitative results. In online groups, most participants are typically lurkers (Preece et al., 2004). To obtain accurate incidence data, clinical trials might be needed. The online patient information supported the identification of a problem. An article in the Archives of Internal Medicine supports this position by saying (DeMonaco, 2009): Postmarketing surveillance and the determination of the real-world safety profile of prescription drugs is arguably flawed. Recent identification of significant adverse effects associated with newly approved prescription drugs support the sometimes-held view that a new system needs to be introduced. The present voluntary system has not provided a sufficient early warning system, and some have called for active systems that probe for potential adverse effects of approved prescription drugs. Patient-oriented Web sites may provide an opportunity to identify potential adverse effects early in a drug’s postmarket history. The interest in collecting adverse events from online patient groups is clearly growing. If a healthcare provider wanted to support its employees in online patient discussion groups with the intent of also helping identify adverse events, then a proposal to the provider’s Institutional Review Board would be in order. The patients joining the group would be provided a consent form that detailed the conditions, the patient alternatives, and other components of a proper consent form. Given that patients had to register to join the group, their successful registration would only occur after they noted online that they consented. While patients might be asked to sign a consent form, they are not invited to the online groups to get a diagnosis or a treatment. Rather the groups support patient-patient interaction, and the patients are responsible for the content of the message

237

E-Patients Empower Healthcare

that they share. If a knowledgeable person brings extracts from the literature to the discussion, those extracts cite the original source and are informational only. Responsibility for taking action based on the information rests with the patient. A healthcare provider that wants to sponsor an online discussion group has a range of options, including the option to provide a healthcare professional as a moderator or to have no moderator. One problem with providing a moderator is the cost in human effort. Healthcare professionals have many demands on their time and often do not see participation in an online patient discussion group as a cost-effective use of their time. For the typical healthcare provider in the United States, efforts invested in an online discussion group cannot be billed to a health insurance company on behalf of the patients’ in the group. At least one healthcare provider continues to support online patient discussion groups because the provider has found that appreciative patients donate money to the provider that offsets that provider’s cost of maintaining the discussion group. A financial cost-benefit analysis that considered a wide-range of factors, such as healthcare professional labor costs, patient loyalty, and patient health outcomes, would be appropriate before an entity decided how much, if anything, to invest in online patient discussion groups. The world is eager for methods to reduce the costs of healthcare without decreasing the quality of healthcare. The advent of the digital, Internet age is creating new opportunities for people to cooperate. Disruptive innovations are needed (Christensen et al., 2000) that lower cost. Encouraging patients to help one another can save costs. Might it be possible to create some kind of a financial reward system to encourage patients to actively participate? The sharing of information online is not considered a reimbursable event by most American health insurance entities. A way could be found to identify online participants who were recognized by their peers as having expertise and having in-

238

surance companies reimburse such contributors when their contributions are deemed adequately significant. One could worry that such an approach would lead to rampant charlatanism. However, peer-peer evaluations in online systems have proven remarkably robust (Rada & Hu, 2002). Given the dearth of options for reducing costs but improving quality, ways to involve patients themselves more effectively into the health care solution should be explored.

conclusIon Empowering patients is vital to improving healthcare. One source of information that has been largely overlooked by the healthcare industry comes from online patient discussion groups. Online patient groups may provide an opportunity for healthcare providers to both build customer relationships and explore adverse effects. The author participated as a patient, though he is also a doctor, in several online patient groups. Patients discussed various types of adverse events, but several types were particularly intriguing for the gaps between what the patients needed to know and what the literature offered. These adverse events have been categorized as involving a diffuse symptom, a standard of care, and uncommon reaction to a drug. The cases are multidisciplinary in nature. The gaps in the literature create an opportunity for someone to 1) produce a synthesis of the literature that highlights the gap and 2) publish that synthesis in a scholarly medical journal. As measured by their responses, patients appreciated receiving information from the literature about their adverse events. A systematic approach to identifying such adverse events and providing relevant literature to patients is sketched based on the experiences of the author. Software can support the retrieving of relevant literature, but posing the response in the context of the patient’s concerns requires human judgment.

E-Patients Empower Healthcare

People seeking health-related support are one of the most common users of the Internet. They can contribute to the health care of one another by sharing their experiences. Such contributions to health care are essentially cost-free to the society but leverage the power of the patient.

reFerences Adams, S. (2008). Blog-based applications and health information: Two case studies that illustrate important questions for consumer health informatics (chi) research. International Journal of Medical Informatics, 1–8. Retrieved from http://dx.doi. org/10.1016/j.ijmedinf.2008.06.009. Annane, D., Depondt, J., Aubert, P., Villart, M., Gehanno, P., & Gajdos, P. (2004). Hyperbaric oxygen therapy for radionecrosis of the jaw: A randomized, placebo-controlled, double-blind trial from the orn96 study group. Journal of Clinical Oncology, 22(24), 4893–4900. doi:10.1200/ JCO.2004.09.006 Antonadou, D., Pepelassi, M., Synodinou, M., Puglisi, M., & Throuvalas, N. (2002). Prophylactic use of amifostine to prevent radiochemotherapyinduced mucositis and xerostomia in head-andneck cancer. International Journal of Radiation Oncology*Biology*Physics, 52(3), 739-747. Calhoun, J., Banaszak-Hol, J., & Hearld, L. (2006). Current marketing practices in the nursing home sector. Journal of Healthcare Management, 51(3), 185–200. Christensen, C., Bohmer, R., & Kenagy, J. (2000). Will disruptive innovations cure health care. Harvard Business Review, (September-October): 102–112. Cleary, P. D. (2003). A hospitalization from hell: A patient’s perspective on quality. Annals of Internal Medicine, 138(1), 33–39.

Darden, L., & Rada, R. (1988). Hypothesis formation via interrelations. In Prieditis, A. (Ed.), Analogica (pp. 109–128). London: Pitman. DeMonaco, H. (2009). Patient- and physicianoriented web sites and drug surveillance: Bisphosphonates and severe bone, joint, and muscle pain. Archives of Internal Medicine, 169(12), 1164–1166. doi:10.1001/archinternmed.2009.133 Duff, F. L., Daniel, S., Kamendje, B., Le Beux, P., & Duvauferrier, R. (2005). Monitoring incident report in the healthcare process to improve quality in hospitals. International Journal of Medical Informatics, 74(2-4), 111–117. doi:10.1016/j. ijmedinf.2004.06.007 Ferguson, T. (2002, September 2002). Key concepts in online health: E-patients as medical researchers. The Ferguson Report Retrieved August 2005, from http://www.fergusonreport.com/ articles/fr00903.htm Ferguson, T. (2009). E-patients: How they can help us heal healthcare. Retrieved June 27, 2009, 2009, from http://www.acor.org/epatientswiki/ Friedman, M., Landsberg, R., Pryor, S., Syed, Z., Ibrahim, H., & Caldarelli, D. (2001). The occurrence of sleep-disordered breathing among patients with head and neck cancer. The Laryngoscope, 111, 1917–1919. doi:10.1097/00005537200111000-00008 Keselman, A., Browne, A., & Kaufman, D. (2008). Consumer health information seeking as hypothesis testing. Journal of the American Medical Informatics Association, 15(4), 484–495. doi:10.1197/jamia.M2449 Kohn, L., Corrigan, J., & Donaldson, M. (Eds.). (2000). To err is human: Building a safer health system. Washington: National Academy Press. Kozinets, R. (2002). The field behind the screen: Using netnography for marketing research in online communications. JMR, Journal of Marketing Research, 39(1), 61–72. doi:10.1509/ jmkr.39.1.61.18935 239

E-Patients Empower Healthcare

Mendenhall, W. M. (2004). Mandibular osteoradionecrosis. Journal of Clinical Oncology, 22(24), 4867–4868. doi:10.1200/JCO.2004.09.959 Misson, J. C. (2001). A review of clinical risk management. Journal of Quality in Clinical Practice, 21, 131–134. doi:10.1046/j.14401762.2001.00421.x Obama, B. (2008). Remarks of president-elect barack obama: Radio address on the economy. Retrieved June 27, 2009, from whitehouse.gov Paccagnella, L. (1997). Getting the seats of your pants dirty: Strategies for ethnographic research on virtual communities. Journal of ComputerMediated Communication, 3(1). Retrieved from http://ascusc.org/jcmc/vol3/issue1/paccagnella. html. Preece, J., Nonnecke, B., & Andrews, D. (2004). The top five reasons for lurking: Improving community experiences for everyone. Computers in Human Behavior, 20(2), 201–223. doi:10.1016/j. chb.2003.10.015 Rada, R. (2005a). Obstructive sleep apnea and head and neck neoplasms. Otolaryngology - Head and Neck Surgery, 132(5), 794–799. doi:10.1016/j. otohns.2004.12.002 Rada, R. (2005b). Software for patient-patient discussions. Montreal Conference on E-Technologies 2005, Montreal, Canada (pp. 205-209). Rada, R. (2006). Membership and online groups. e-Society 2006: IADIS International Conference, Dublin, Ireland (pp 290-293).

240

Rada, R. (2007). Sleep and quality of life in head and neck neoplasm. In Verster, J. C., PandiPerumal, S. R., & Streiner, D. L. (Eds.), Sleep and quality of life in head and neck neoplasm. Totowa, New Jersey: Humana Press. Rada, R., & Hu, K. (2002). Patterns in student–student commenting. IEEE Transactions on Education, 45(3), 262–267. doi:10.1109/ TE.2002.1024619 Rades, D., Fehlauer, F., Bajrovic, A., Mahlmann, B., Richter, E., & Alberti, W. (2004). Serious adverse effects of amifostine during radiotherapy in head and neck cancer patients. Radiotherapy and Oncology, 70, 261–264. doi:10.1016/j.radonc.2003.10.005 Rombaux, P., Hamoir, M., Plouin-Gaudon, I., Liistro, G., Aubert, G., & Rodenstein, D. (2000). Obstructive sleep apnea syndrome after reconstructive laryngectomy for glottic carcinoma. European Archives of Oto-Rhino-Laryngology, 257(9), 502–506. doi:10.1007/s004050000267 Rosenbaum, E. E. (1988). A taste of my own medicine, when the doctor is the patient. New York: Random House. Spath, P. (2004). Preface. In P. Spath (Ed.), Partnering with patients to reduce medical errors (pp. xix-xxiii). Chicago: American Hospital Association Press. 1 This paper borrows heavily from Roy Rada (2008) “Ethnographic discovery of adverse events in patient online discussions: customer relationship management” International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 3, pp 77-85

241

Chapter 17

Towards Process-of-Care Aware Emergency Department Information Systems: A Clustering Approach to Activity Views Elicitation Andrzej S. Ceglowski Monash University, Australia Leonid Churilov The University of Melbourne, Australia

aBstract The critical role of emergency departments (EDs) as the first point of contact for ill and injured patients has presented significant challenges for the elicitation of detailed process models. Patient complexity has limited the ability of ED information systems (EDIS) in prediction of patient treatment and patient movement. This article formulates a novel approach to building EDIS Activity Views that paves the way for EDIS that can predict patient workflow. The resulting Activity View pertains to “what is being done,” rather than “what experts think is being done.” The approach is based on analysis of data that is routinely recorded during patient treatment. The practical significance of the proposed approach is clinically acceptable, verifiable, and statistically valid process-oriented clusters of ED activities that can be used for targeted process elicitation, thus informing the design of EDIS. Its theoretical significance is in providing the new “middle ground” between existing “soft” and “computational” process elicitation methods.

IntroductIon Information system (IS) design principles call for requirements definition as an intermediate stage in

the design and development of IS (Mertins, Bernus, & Schmidt, 1998). The requirements definition is a document that outlines all the needs that users require of the prospective system. The requirements

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Towards Process-of-Care Aware Emergency Department Information Systems

definition is designed to allow for the translation of the physical needs of a process into an automated environment. Programmers should be able to work from this document without going back to the users for clarification. The requirements definition can thus “be compared to a schematic of a plan or a diagram of how a technical device works” (Langer, 2008). The requirements definition commonly takes the organisation from a functional view of activities (who, in which department, does what) to a process-oriented view of operations (what happens, when, and where). This shift from functional data structuring to process event recording is best described in process models. Process models are formalised representations of the activities enacted by a human or a machine that are considered important to the achievement of the objective of the process (Dumas, van der Aalst, & ter Hofstede, 2005). The process models provide a structured framework for IS specification and design, one example is ARIS, a widely used reference architecture and methodology (Scheer, 1999), and allow configuration of the IS to support or control the flow of work in the operational process (Rozinat & van der Aalst, 2008). Comprehensive process models combine different views (Poulymenopoulou, Malamateniou, & Vassilacopoulos, 2003; Seltsikas, 2001) that describe: 1.

what activities are being performed within a process and their interactions (an activity view);

2. 3.

4.

what data are relevant as inputs to, or outputs from these activities (data and output views); who performs each activity and where it is performed (an organisational or resource view); and when and how the process activities are being performed (a control view).

Information systems designed for hospital emergency departments (ED), also known as emergency rooms (ER) or accident and emergency departments (A&E), are commonly called emergency department information systems (EDIS). EDIS as they exist today mainly address two aspects of ED operations: (i) providing for tracking of patients and (ii) making patient information available to clinicians and administrators (Figure 1). The first function, having patient information available online, promotes efficiency of operations through enhanced data entry capabilities, transferability between hospital departments and locations, and potential for bedside update of patient records. Electronic health records have wide reaching implications for EDIS design that are being addressed through various electronic health record initiatives (e.g., OpenEHR in Australia; GEHR in Europe; and HL7 in the USA). The second function of existing EDIS relates to patient flow management. Most systems provide a list of patients awaiting treatment along with presenting problem, urgency, and disposition. This

Figure 1. Current EDIS deal with patient management retrospectively. There is an absence of information about the pathways future treatment might take and likelihood of patient admission. P atient m anag em ent (trackin g) P a tien ts w aiting f or:  R egistratio n  T riage  A R oom  A D octor or N urse  A dm ission  A p ro cedure

242

Labora to ry, Im agin g and P harm ac y:  O rd er s tatus R esources:  R oo m A vailability  E qu ipm ent l ocation

C lin ical r ecords s uppo rt P a tient sum m aries:  D em ographics  P revious history  U rgency  P re senting P roblem  S ta tus

eH ealth R ecord s:  D iagnosis  P ro c edures  T re atm e nt  D isposition  F ollo w -up

Towards Process-of-Care Aware Emergency Department Information Systems

allows clinicians to select patients for treatment and to see when test results are available. Many systems can provide tracking while patients are in the ED. Patient tracking is passive with the system reporting the location of patients in an effort to assist with efficient use of resources. Unfortunately, existing EDIS do not have predictive functions. This limits their capacity to provide prediction of workflow (where patients will go next) and optimisation of throughput (by minimising queues within the system) and resource use (by making sure that resources are aware of the impending need). In this article, we argue that existing EDIS are unable to provide such benefits because every workflow is instantiated with every patient presentation (or “instance”). By this, we mean that the EDIS sees every patient presentation as a unique instance that will follow an undefined sequence of activities and use unknown resources—there are no “template” processes programmed into the EDIS that might be followed, even if the treatment is highly reproducible. In making EDIS “process aware,” the system is able to apply process templates to patients and so is capable of predicting the pathway that a patient might follow through the ED, what resources might be required, and what the overall state of the ED might be at any time. Process awareness does not imply only that the system designers considered processes when they specified the system requirements, but rather that the systems allow for explicit definition of the process logic, execution, and monitoring. If EDIS are going to progress beyond the limitations of existing systems, then work flows need to be defined so that patient instances can be linked to specific treatments. Once this is achieved, it will be possible to build higher level understanding of ED operations. This understanding will allow decision support functions such as prediction of resource need, optimisation of bed use, and warning of impending blockage. The principles of IS design described in the opening paragraphs clearly indicated that the

ability to logically represent the groups of activities that occur in the course of patient treatment within ED through a clear and comprehensive activity view constitutes a necessary (although not sufficient) condition for moving EDIS to a process aware basis. The objective of this article is to: (i) describe an activity view that supports the requirements definition of process aware systems and (ii) provide a method for the development of this activity view. The search for clusters of activities within existing ED patient data records represents a new middle ground between “soft” (or ethnographic) methods for the elicitation of process models (Bustard, Oakes, & He, 1999) and “hard” methods such as workflow (or process) mining (Business Process Management Center, 2003) that arrive directly at process models from data analysis. The article’s contribution to practice lies in the potential to design process aware EDIS since the activity view developed in this article fits with standard IS design methodologies. The formulation in this article of an activity view for EDIS design is new because of the lack of existing of comprehensive process models for patient treatment in EDs (Djorhan & Churilov, 2003).

Background: context and Motivation for process aware edIs EDIS focus on patient management and data retrieval and storage (Amouh, Gemo, Macq, et al., 2005). Patient management aspects of EDIS have functions that prioritise patients according to their urgency, allocate beds and rooms, and register occurrence of key events such as nurse and doctor assessments and patient departure. EDIS data management revolves around the transport and storage of electronic patient records and the ordering and reporting of pathology tests and imaging investigations. Supplementary hardware such as wireless devices can track patients and enhance the gathering of data for both patient and data management (Amouh et al., 2005). The

243

Towards Process-of-Care Aware Emergency Department Information Systems

systems do not forecast patient flow and resource use, nor do they facilitate exchange of information with outside organisations. EDIS functionality supports resource allocation and patient selection for treatment based on priority (Aronsky, Jones, Lanaghan et al., 2007, gives an example of this), but it fails to coordinate patient movement and patient treatment (for instance, there is no tracking of patients sent for X-rays, estimation of the time they will take at the X-ray location, nor booking of space at the next resource, such as the plaster cast room, that they might require once their X-rays have been assessed.). An extreme view of an “ideal” process aware EDIS would be one that is able to incorporate every step in patient care in which human handoffs are automated, each step in patient treatment automatically logged and tracked, and timing and sequencing of steps analysed for performance evaluation. Human interactions with networked electronic devices such as personal computers, CT scanners, lab systems, telephones, IV pumps, and wireless patient tracking tags would be linked to the EDIS for automation of process control. Order entering by physicians, bar-coding of medication by nurses, patient registration by clerks, and surgery scheduling for surgeons should be linked and coordinated automatically. Overall, the EDIS should be able to sequence, monitor, track, alert,

and reroute any step in each of the patient care processes (Rucker, 2003). This extreme view, while attractive to proponents of workflow systems (systems that automatically direct work throughout a process), is unlikely to be achieved in the real world. The ED environment is simply too complex, too many decisions need to be made by clinicians based on patient observations that are impossible to include in any IS, and patient variation is too vast to be included in a single configurable workflow. However, this extreme view can guide thinking towards EDIS that support real processes and provide avenues for decision support while remaining sufficiently flexible to sustain the unique treatment needs of individual patients. This is impossible without understanding the interaction of activities involved in the processes, as depicted in the explicit process aware EDIS/activity view relationship presented in Figure 2. Thus, the research problem addressed in this article is to how to elicit an activities view for an EDIS. The following sections will discuss (i) the methods available for elicitation of activity views and (ii) the work that has been done to group ED treatment processes.

Figure 2. Process aware EDIS are supported by diverse views: organisational, data, control, activity and output views. Existing EDIS tend to be derived solely from listings organisational (resource) and data views and so are only able to encompass a part of ED operations. P rocess A w are E D IS

S upport

diverse View(s)

244

F acilitate

E ntity R elationship

Include

E D tr eatm ent O perations

D epend o n F igure 1 R esources and d ata

e m

Towards Process-of-Care Aware Emergency Department Information Systems

The Elicitation of Activity Views and Views of ED Activities Activity View Elicitation The most common way in which activities are elicited is through interviews with experts and people who perform the work (Earl, 1994; Kotonya & Sommerville, 1998; Weerakkody & Currie, 2003). This is frequently termed an ethnographic approach (Schuler & Namioka, 1993). As a technique for building the Activity View, it is prone to subjective views of the work that may be distorted according to social dynamics unrelated to the work (Rennecker, 2004) and may encounter situations where the interviewees are unable to provide generalized pictures of the work (Gospodarevskaya, Churilov, & Wallace, 2005). EDs appear to fall into the latter class. Process mining (van der Aalst et al., 2003) is a purely computational technique that is able to extract feasible “as is” process models directly from workflow data without resorting to interviews. It has its roots in a data mining idea that associations between variables in a relation can be counted and granted some level of confidence (Agrawal, Imielinski, & Swami, 1993). Combination of this concept with inference algorithms (Angluin & Smith, 1983) provided a way in which timeseries data could be mined (Cook & Wolf, 1998) to retrospectively build a picture of sequences of events in software (Agrawal, Gunopulos, & Leymann, 1998). This idea has been extended so that the branches, loops, and joins common in most processes may be inferred from event logs (de Medeiros, van der Aalst, & Weijters, 2003). Process mining requires access to a log that records the sequence of defined tasks in workflow for a large number of cases. Such data is readily accessible for most work that takes place within or on computer systems, as evidenced by business activity management software tools for the

analysis of logged data such as ARIS-Process Performance Manager (IDS-Scheer, 2004). Process mining provides an excellent solution to process elucidation in situations where workflow is fully automated (“mature” information system applications) but little process elicitation help is available for less mature “case handling” applications where process is undefined or much of the work is performed by an expert who initiates activities based on the particulars of each job (van der Aalst et al., 2003). Detailed event logs are seldom available for activities that take place outside computer systems. Fortunately, “computer external” activities are often logged in databases for billing and other purposes. Such information may be captured in batches after the activities have been completed but lack information about sequence or timing of events. The existence of these nonsequential “activity logs” provides an avenue for the identification of patterns of activity. Activity logs of this sort are commonly associated with ill-defined processes where experts make complex decisions while performing the work (such as hospital EDs). The binary “event” logs do not have the detail necessary for process mining and the mere volume of data is likely to make the logs inscrutable to traditional data analysis techniques (past efforts in this regard are discussed in the next section), but they do provide an avenue for the identification of patterns of activity if non-traditional methods are used. The use of these logs in clustering of activities falls between the “soft” ethnographic and “hard” process mining that were introduced above (depicted in Figure 3). The clustering “compromise” approach will be described later in this article, but it is first necessary to describe existing views of ED activities in order show to current understanding of ED processes.

Existing Views of ED Activities There are well over a thousand diagnoses listed in the Victorian Emergency Data Set guidelines

245

Towards Process-of-Care Aware Emergency Department Information Systems

Figure 3. An extension of Figure 2, showing that elicitation is achieved through social and technical elicitation extremes. The middle ground is explored in this paper. P rocess A w are E D IS

S upport

diverse View(s)

E ntity R elationship

D epend on

F igure 2

R esources and d ata

Include

D erived fr om

W orkflow M ining

R equire

S equential Logs

elicitation Methods

C lustering

R equire

Lists of activities

E thnographic

R equire

D escriptions

(2007), and many other governments have a similar proliferation of diagnoses. It is not feasible to incorporate separate activity views for every one of these diagnoses in a requirements definition because this will probably lead to a situation where overhead costs for the control systems surpass the benefit of efficient coordination of every variant (Becker, Kugeler, & Rosemann, 2003). Simplification is necessary to reduce the large number of diagnoses and permutations of treatment to a number suitable for cost effective implementation in an IS. Some grouping of the activities involved in patient treatment is needed in order to reduce the complexity of the requirements definition. Without this grouping, it is difficult to conceive systems that can indicate what the next step in patient care may be, where it should take place, and who should be responsible for it. Groupings and simplifications have been attempted many times in the past. Analysts have used output data such as “the time treatment takes,” and input data such as “the complexity of presentation” to group patients (for example,

246

E D tr eatm ent O perations

F acilitate

T he technique described in this paper

Hoffenberg, Hill, & Houry, 2001; Walley, 2003). These approaches address the activities related to patient movement (whether patients are likely to be admitted to hospital or not, or whether they may be treated in a chair, rather than a bed, for example) but not the detailed activities involved in patient treatment. Patient movements have been analysed in simulation studies (Brailsford, Churilov, & Liew, 2003; Mahapatra, Koelling, Patvivatsiri et al., 2003; Sinreich & Marmor, 2004), but the models, in keeping with the philosophy of discrete event simulation, tend to imitate the physical movement of patients through the ED, rather than the treatment provided to patients. The studies result in recommendations on how to reduce wait or facilitate throughput but they fail to provide insight into the management of patients in accord with their treatment requirements. Clinicians have tried to get a simplified picture of ED operations by grouping patient “cases,” characteristically according to combinations of age, urgency of complaint, diagnosis, time in

Towards Process-of-Care Aware Emergency Department Information Systems

ED, and outcome of visit linked to cost (Bond, Baggoley, Erwich-Nijhout et al., 1998; Cameron, Baraff, & Sekhon, 1990; Jelinek, 1995). These clinical classification schemes have failed to group patients by similarity of treatment because they have used the use of cost as an objective function. Low and high cost attributes trade off, so dissimilar patients end up in similar groups. Clinical guidelines (Clinical Pathways) exist that detail every aspect of patient treatment and act as checklists for common chronic ailments (for example, the Action plan for anaphylaxis available at http://www.medicalobserver.com. au/clinicalguidelines. Porter, Cai, Gribbons et al. (2004) provide an example of decision support for a single treatment). Clinical Pathways provide a notion of a sequence of prescribed activities involved in patient treatment that would be ideal for a process aware EDIS. Unfortunately, they only cover a small number of narrow treatments so they do not provide the variety of ED treatments that would need to be included for predictive decision support. A different approach is needed to unravel the complexity of patient treatment in EDs. Such an approach is described in the following sections.

MetHod descrIptIon: FroM clInIcal procedures to an actIVIty VIew For a process aware edIs Mertins et al. (1998) suggest that it is possible to build information systems from a subset of the organisational, data, control, activity, and output views (defined as aspects of process models in the Introduction). This approach is apparent in the design of most existing EDIS, where the data and organisational views are well represented (for exampl, the linking of Figure 1 to the resources and data component in Figure 2), but the activity view is less well represented or not present at all (Rucker, 2003). The resulting EDIS’ cannot be

process aware because treatment activities have not been functionally related in the activity view, treatment objectives are not available for the processes, and supporting applications (bedside monitoring, for instance) cannot be integrated into the process. Elucidation of the activity view is necessary to allow coordination of the resources, communication, and technology associated with treatment. Since the activity view is, in its simplest form, a collection of objects such as activities, objectives, and software applications, clinical procedures that are manually recorded (as described at the end of the previous section) may be used to provide a library of activities. Even though there is a wide range of patients and presentations, much of the patient treatment related work in EDs is based on application of a short list of medical procedures such as patient observation, administration of drugs, and laboratory and imaging investigations. Just 62 procedures are used for reporting ED treatment in Victoria, Australia (Metropolitan Health and Aged Care Services Division of the Victorian State Government, 2007). Thirty six procedures account for 99% of all procedures in Victorian hospitals. Almost 17% are classed as “Other,” which includes observation of patients by medical staff; 6% are “No procedures”; some 10% are drug administrationl; and over 9% Xray imaging. Other significant procedures are venipuncture, intravenous catheter access in preparation for infusion of fluid or drugs, and echocardiogram diagnostics (figures derived from Victorian Emergency Medical database for 2002). An activity view that simply lists all these procedures would result in processes models based on a vast number of possible permutations of causal linkages between all procedures. Individual procedures could have associated objectives but it would be difficult to understand how the objectives were related to each other. Some form of simplification is necessary to advance the activity view beyond this primitive.

247

Towards Process-of-Care Aware Emergency Department Information Systems

The grouping of procedures into treatment specific clusters is a way of providing this simplification—procedures are logically linked as activities within treatments. While the grouping of activities in this way does not provide a true process view (the activities are not sequenced), it significantly reduces the number of permutations in order to provide a simplified view of ED operations. It is this simplified view that has eluded designers of ED information systems up to now. Each group of activities, or treatment cluster, may be scrutinised to determine the most likely (or most desirable) sequence of activities. The grouping of procedures for this activity view may be achieved through nonparametric clustering. One nonparametric technique is self organising mapping (SOM) (Kohonen, 1995). The SOM is a grouping technique that is algorithm driven and relies on data rather than domainspecific expertise. The objectives of the technique are to minimise diversity within groups and to maximise differences between groups. The technique generally employs large datasets, works well with many input variables, and produces arbitrarily complex models unlimited by human comprehension (Kennedy, Lee, Van Roy et al., 1998). Self organising maps provide a visual understanding of patterns in data through a two dimensional representation of all variables. Records that have similar characteristics are adjacent in the map, and dissimilar records are situated at a distance determined by degree of dissimilarity. Viscovery SOMine, the software tool used in this analysis, employs a variant of Kohonen’s BatchSOM (Kohonen, 1995), enhanced with a scaling technique for speeding up the learning process (Eudaptics Software Gmbh, 1999). The concept of providing a simplified activity view of treatment through the grouping of procedures is illustrated in the next section.

248

ValIdatIon tHrougH exaMple: clusterIng clInIcal procedures to BuIld an actIVIty VIew Victorian Emergency Medical Data (VEMD) made up of de-identified records of all ED presentations across 31 anonymous hospital campuses was obtained and five similar-sized campuses (by number of records) were selected for analysis and comparison. Each record contained demographic particulars and details of the visit such as “apparent severity of complaint,” “key time points,” and “disposition,” plus all medical procedures performed, but cost data was not available. All cases where patients underwent more than one procedure were included for analysis. This was generally around 60% of all patient presentations. The 13 least-used procedures were omitted from the analyses, as was the “NONE” procedure. These exclusions totalled less than 1% of procedures in cases where patients had more than one procedure. Random samples of approximately 10,000 cases having Departure Status “Discharged home” were extracted. Where there were less than 10,000 cases with this departure status at a campus, all cases were included for analysis. The data was saved into a tab delimited format. Each record had 50 fields—a record identifier and 49 procedures coded as “0” for “not performed” to 1,2,3...16 for repeated applications. The procedure data was sparse, but the number of records gave some assurance that patterns of recurring groups of procedures could be found. Self organising mapping was applied to the data using the same clustering settings. These settings included: • The initial map height that gave the initial number of vertical nodes in the two dimensional representation of the data (the number of horizontal nodes was automatically set from the map ratio); • A scaling factor that specified the increase in horizontal number of nodes as the clus-

Towards Process-of-Care Aware Emergency Department Information Systems

tering algorithm progressed through each growth step (the vertical number of nodes was adjusted according to the desired map ratio); • The influence radius of the neighbourhood interaction of the intermediate maps (the reach of the Gaussian neighbourhood function). A high intermediate tension “averages” the data distribution, while a low tension allows adaptation to display finer features. • The number of iterations and the Wegstein factor (a convergence parameter in the batch SOM algorithm comparable to the momentum factor as it is commonly used in supervised neural network algorithms). Between 13 and 27 clusters were identified across the five data sets. The clusters were validated through a range of internal measures of cluster quality. These measures were analogous to traditional indicators of cluster quality such as the Rand Statistic, Jaccard Coefficient, and the Folkes and Mallows Index; however, they operated on two dimensions, rather than the single figure of the traditional measures. Maps that displayed the frequency of records across two dimensional cluster space, the quantisation error, the proximity of nodes to neighbours, and curvature of the map through n-dimensional space were scrutinised to ensure that the spread of clusters was even, the clustering error within reasonable limits, and the cluster shapes regular. This clustering of procedures into prototype “workflows” was verified through discussions with a specialist in emergency medicine and found to be clinically sensible. It was concluded that the clusters reflected “as is” core ED treatments. Similar sets of procedures were apparent in clusters across campuses. Typical prototype workflows that were common across multiple campuses are (figures in brackets indicate likelihood of procedure):

• • • • • •

12 Lead ECG (1.0), with Peripheral IV catheter and Venipuncture (0.6); 12 Lead ECG and ECG monitoring (1.0), with Venipuncture (0.9); Peripheral IV catheter and IV drug infusion (1.0), with Venipuncture (0.7); Suture, Steri-strip, glue (1.0) with Dressing (between 0.6 and 0.9); Plaster of Paris (1.0) and X-ray (0.9) with Drug administration (around 0.5); Splint (1.0), X-Ray (0.8) with Drug administration (between 0.2 and 0.6);

These clusters each comprise between 2% and 9% of patients and add up to between 20% and 30% of patients. In addition to these widespread clusters, there are typically two to three large clusters at each campus that provide for some 25% of presentations. These large clusters often include drug administration, venipuncture and full ward test (urine) or 12 Lead ECG, but the proportions differ between campuses. It is important to note that these clusters represent “core treatment” activities within the EDs. These treatments employ resources most frequently The truncated example in Table 1 indicates that the patient was likely to be experiencing breathing difficulties (possibly an asthma attack) and might have been classed as urgent or nonurgent (note that this is in direct contrast to previous studies, which have segmented patients by urgency—a patient movement, rather than patient treatment orientation). Nebulised medication was provided to all patients. This treatment was supported with X-rays, drugs, venipuncture, electrocardiogram, peripheral intravenous catheter, and full ward tests. Two campuses also indicated that intravenous injections were commonly recorded in this group. The numbers in Table 1 may be considered to represent the likelihood between 0 and 1 of patients in that cluster undergoing the procedure. In the example given in Table 2, it can be seen that there is much agreement across campuses regarding the reporting of “head injury observa-

249

Towards Process-of-Care Aware Emergency Department Information Systems

Table 1. Nebulised medication procedures across three hospital campuses, A, B, and C

Procedure Description

Campus

A

B

C

% Patients

2.9%

2.4%

3.6%

Nebulised Medication

NEB

1.00

1.00

1.00

X-ray

XRAY

0.43

0.26

0.47

Oral/ sublingual/ topical/ rectal drug administration

DRUG

0.81

0.89

0.80

Venipuncture

VB

0.28

0.21

0.29

12 Lead ECG

ECG

0.16

0.14

0.15

Peripheral intravenous catheter (IV access)

IV

0.28

0.22

0.24

0.09

Full ward test - urine

FWT

0.06

Intravenous infusion

IVS

0.13

0.06 0.12

Table 2. Head injury observation across five campuses Campus

% Patients

HIO

XRAY

IV

IVI

VB

DRS

SUT

ECG

FWT

INF

DRUG

CT

IVS

RBG

SPR

A

2.3%

1.0

0.4

0.5

0.0

0.4

0.1

0.1

0.3

0.2

0.2

0.5

0.1

0.3

0.2

0.0

B

0.9%

1.0

0.4

0.4

0.1

0.3

0.2

0.2

0.1

0.3

0.1

0.5

0.1

0.2

0.1

0.0

C

4.2%

1.0

0.5

0.4

0.0

0.5

0.1

0.0

0.4

0.1

0.2

0.6

0.2

0.2

0.0

0.0

D

3.4%

1.0

0.3

0.1

0.2

0.2

0.1

0.1

0.1

0.1

0.0

0.0

0.0

0.0

0.0

0.0

E

0.6%

1.0

0.4

0.3

0.0

0.4

0.2

0.1

0.3

0.3

0.1

0.4

0.2

0.1

0.5

0.5

Key: HIO = Head Injury Observation; XRAY = X-ray examination; IV = Peripheral intravenous catheter; IVI =IV drug infusion; VB = Venipuncture; DRS = Dressing; SUT = Suture, steristrip, glue; ECG = 12 lead Electrocardiogram; FWT = Full Ward Test of urine; INF = Infusion of fluid (not blood); DRUG = Oral, sublingual, topical, rectal drug administration; CT = Computerised tomography scan; IVS = Intravenous drip; RBG = Test of Random Blood Glucose; SPR = Spirometry

tion” (HIO), despite the varying percentage of patients (second column of the table). The campus in the last row displays anomalous random blood glucose (RBG) and Spirometry (SPR) that may indicate different patient and treatment profiles. Similar studies were carried out for admitted as well as discharged patients and core treatments identified that overlapped the above and expanded to reflect the treatment accorded to patients who were later admitted to hospital. The studies were reinforced with text mining exercises that compared patient reason for presentation with their treatment and through simulation of treatment activities. It was also found that the nonexclusive

250

nature of the core treatments (i.e., one procedure might be used in multiple core treatments) that arose for the clustering technique made the technique far superior (in defining treatments that were clinically realistic) to other methods that were attempted such as association mining and CART. This section set out to describe the realisation of the idea that a simplified activities view could be developed if procedures involved in patient treatment were clustered into process-oriented groups. Distinct clusters of procedures were identified in the treatment of patients at a number of EDs. The grouping was validated and verified and considered a reasonable “as is” simplification

Towards Process-of-Care Aware Emergency Department Information Systems

of ED activities into core treatments (Ceglowski, Churilov, & Wassertheil, 2007, describes simulation of ED operations based on treatment clusters). The implications of these findings are discussed in the next section.

dIscussIon: How clustered actIVItIes support systeMs desIgn In providing core treatments, the previous section has expanded the simple activity view of individual procedures to one with grouped procedures in a handful of frequently applied core treatments. The simplification reduced the number of objects from 62 procedures to around 20 core treatments. This reduction in complexity facilitates the association of activities with other aspects of the activities view such as process goals because each core treatment represents a process-oriented class of treatment. The primary benefit of identifying prototype workflows and linking them to patient attributes has been the potential for predictive decision support. Certain core treatments take longer than others take and have a higher rate of hospital admission associated with them. When many patients requiring such long duration treatments reside in the ED concurrently, then throughput slows and the ED becomes prone to blockage—regardless of its capacity. By linking patient presentation to core treatments and the time they take, it is possible to update prediction of potential blockage as each new patient presents at the ED. An extension of this high level prediction is the potential for prediction of future workflow requirements. Having patient presentation linked to their most likely future treatment allows their future workflow to be mapped and planned for, so handoffs can be identified, responsibilities indicated, and resources specified. In this way the organisational, data, and output views may be linked to core treatments.

The core treatments add knowledge about common treatment activities that can be further explored using various techniques, such as ethnographic studies or additional data investigations. The control view can be structured once sequencing and mapping of cause-effect linkages has been done (possibly through expert clinical input) because the roles and responsibilities may be identified and linked to each activity within the treatment. This moves EDIS towards process aware systems that can not only coordinate patient treatment and movement, but predict them, too. The clustering technique may further be extended through use of techniques such as association mining to provide more detailed segmentations of the groups and better understanding of the business rules, especially if the data is linked to attributes such as urgency, age, and whether the patient was admitted or not. The insights provided by such investigations give rise to another, more refined, activity view where patient attributes may be causally linked to treatment activities, making prediction of patient routings feasible.

conclusIon This article indicated that existing EDIS suffer from inadequate understanding of patient treatment processes that manifest as EDIS that are unable to predict future patient treatment and patient movement needs. It was noted that development of an activities view would assist movement towards process aware, predictive EDIS. A suggestion was made that clinical procedures recorded for other purposes could be used to populate the activity view, but simplification was necessary in order to reduce the number of possible process models that could be deduced. A method for simplification was described that resulted in procedures being grouped in clinically acceptable core treatments that covered 99% of ED treatment operations. The utility of these core treatments in providing an activity view was discussed.

251

Towards Process-of-Care Aware Emergency Department Information Systems

This article has described a route to take existing EDIS (which deal primarily with the management of historic data such as patient records and the queue of patients awaiting treatment) to event-based systems that can “look forward” to predict what pathways patients might take through the ED. EDIS that are capable of managing workflows (predicting the work to be performed at each successive stage of treatment) facilitate:

Agrawal, R., Imielinski, T., & Swami, A. (1993). Mining association rules between sets of items in large databases. In ACM SIGMOD Conference on Management of Data, Washington, DC.

•

Angluin, D., & Smith, C.H. (1983). Inductive inference: Theory and methods. Computing Surveys, 15(3), 237-269.

•

•

Logistics management, by “booking” resources for when patients will need them; Predictive reporting, by linking presentation problem and severity with likely length of stay, probably treatment, and whether a patient might be admitted to a ward; and Management decision support, through the ability to assimilate the status of patients in the ED with the new presentations awaiting treatment.

Clustering represents an additional tool in the elicitation of process models for EDIS design. It is one that is not limited in breadth in the way that ethnographic studies often have to be—all history (as it exists) can be included in the analysis. Nor does clustering have the requirement of workflow mining that detailed logs exist—simple activity records will suffice to provide a working activities view to supplement existing organisational, data, and input views for EDIS design. Transforming such a supplementation into a truly integrated process aware EDIS design presents a challenging direction for future research.

reFerences Agrawal, R., Gunopulos, D., & Leymann, F. (1998). Mining process models from workflow logs. In Sixth International Conference on Extending Database Technology, Valencia, Spain.

252

Amouh, T., Gemo, M., Macq, B., et al. (2005). Versatile clinical information system design for emergency departments. IEEE Transactions on Information Technology in Biomedicine, 9(2), 174-183.

Aronsky, D., Jones, I., Lanaghan, K., et al. (2007). Supporting patient care in the emergency department with a computerized whiteboard system. Journal of the American Medical Informatics Association. Becker, J., Kugeler, M., & Rosemann, M. (2003). Process management: A guide for the design of business processes. Berlin, Heidelberg, New York: Springer. Bond, M., Baggoley, C., Erwich-Nijhout, M., et al. (1998). Urgency, disposition and age groups: A casemix model for emergency departments. Emergency Medicine Australasia, 10(2), 103-110. Brailsford, S., Churilov, L., & Liew, S.K. (2003). Treating ailing emergency departments with simulation: An integrated perspective. In International Conference on Health Sciences Simulation. Orlando, FL: The Society for Modeling and Simulation International. Business Process Management (BPM) Center. (2003). Process mining. Retrieved May 24, 2008, from http://tmitwww.tm.tue.nl/research/ processmining/ Bustard, D., Oakes, W.R., & He, Z. (1999). Models to promote effective system change. In IEEE International Conference on Software Maintenance (ICSM ‘99).

Towards Process-of-Care Aware Emergency Department Information Systems

Cameron, J., Baraff, L., & Sekhon, R. (1990). Case-mix classification for emergency departments. Medical Care, 28, 146-158. Ceglowski, A., Churilov, L., & Wassertheil, J. (2007). Combining data mining and discrete event simulation for a value-added view of a hospital emergency department. Journal of the Operational Research Society, 58(2), 246-254. Cook, J.E., & Wolf, A.L. (1998). Discovering models of software processes from event-based data. ACM Transactions on Software Engineering and Methodology, 7(3), 215-249. de Medeiros, A.K.A., van der Aalst, W.M.P., & Weijters, A.J.M.M. (2003). Workflow mining: Current status and future directions. In R. Meersman, Z. Tari, & D.C. Schmidt (Eds.), On the move to meaningful Internet systems 2003: CoopIS, DOA, and ODBASE (pp. 389-406). Berlin, Heidelberg: Springer-Verlag. Djorhan, V., & Churilov, L. (2003). Business interactions in an acute emergency department in Australia: A clinical process modeling perspective. In the 6th Pacific Asia Conference on Information Systems (PACIS 2002). Tokyo, Japan: The Japan Society for Management Information (JASMIN).

Hoffenberg, S., Hill, M.B., & Houry, D. (2001). Does sharing process differences reduce patient length of stay in the emergency department? Annals of Emergency Medicine, 38(5), 533-540. IDS-Scheer. (2004). ARIS process performance manager. Retrieved May 24, 2008, from http:// www.ids-scheer.com Jelinek, G.A. (1995). Casemix classification of patients attending hospital emergency departments in Perth, Western Australia: Development and evaluation of an urgency-based casemix system. Perth, Australia: University of Western Australia. Kennedy, R., Lee, Y., Van Roy, B., et al. (1998). Solving data mining problems through pattern recognition. Prentice Hall. Kohonen, T. (1995). Self-organizing maps. Berlin, Germany: Springer. Kotonya, G., & Sommerville, I. (1998). Requirements engineering: Processes and techniques. New York: John Wiley. Langer, A.M. (2008). Analysis and design of information systems. London: Springer (pp. 418; p. 249 illus.).

Dumas, M., van der Aalst, W., & ter Hofstede, A. (2005). Process-aware information systems: Bridging people and software through process technology. Hoboken, NJ: Wiley-Interscience.

Mahapatra, S., Koelling, C.P., Patvivatsiri, L., et al. (2003). Pairing emergency severity index 5-level triage data with computer aided system design to improve emergency department access and throughput. In The 2003 Winter Simulation Conference. New Orleans, LA: INFORMS.

Earl, M.J. (1994). The new and the old of business process redesign. The Journal of Strategic Information Systems, 3(1), 5-22.

Mertins, K., Bernus, P., & Schmidt, G. (1998). Handbook on architectures of information systems. Berlin, Germany: Springer.

Eudaptics Software Gmbh. (1999). Viscovery SOMine. Wien: Eudaptics Software Gmbh.

Metropolitan Health and Aged Care Services Division of the Victorian State Government. (2007). Victorian emergency minimum dataset (VEMD). Retrieved May 24, 2008, from http:// www.health.vic.gov.au/hdss/vemd/2004-05/ manual/index.htm

Gospodarevskaya, E., Churilov, L., & Wallace, L. (2005). Modeling the patient care process of an acute care ward in a public hospital: A methodological perspective. Hawaii: HICSS.

253

Towards Process-of-Care Aware Emergency Department Information Systems

Porter, S.C., Cai, Z., Gribbons, W., et al. (2004 ). The asthma kiosk: A patient-centered technology for collaborative decision support in the emergency department. Journal of the American Medical Informatics Association, 11, 458-467. Poulymenopoulou, M., Malamateniou, F., & Vassilacopoulos, G. (2003). Specifying workflow process requirements for an emergency medical service. Journal of Medical Systems, 27(4), 325-335. Rennecker, J. (2004). Updating ethnography to investigate contemporary organisational forms. In M.E. Whitman & A.B. Woszczynski (Eds.), The handbook of information systems research (pp. xi, 349). Hershey, PA: IGI Global. Rozinat, A., & van der Aalst, W.M.P. (2008). Conformance checking of processes based on monitoring real behavior. Information Systems Frontiers, 33(1), 64-95. Rucker, D. (2003). Finally, a tool to re-engineer health care: The workflow engine. San Francisco, CA: Montgomery Research, Inc. Scheer, A.W. (1999). Business process frameworks. Berlin, Germany: Springer.

Seltsikas, P. (2001). Organizing the information management process in process-based organizations. In 34th Hawaii Int. Conf. Syst. Sci., Hawaii. Sinreich, D., & Marmor, Y. (2004). A STAFFING emergency. Industrial Engineering, 36(6). van der Aalst, W.M.P., ter Hofstede, A.H.M., & Weske, M. (2003). Business process management: A survey. In Business Process Management (BPM2003) (pp. 1-12). Berlin, Germany: Springer-Verlag. van der Aalst, W.M.P., van Dongena, B.F., Herbst, J., et al. (2003). Workflow mining: A survey of issues and approaches. Data & Knowledge Engineering, 47(2), 237-267. Walley, P. (2003). Designing the accident and emergency system: Lessons from manufacturing. Emerg Med J, 20(2), 126-130. Weerakkody, V., & Currie, W. (2003). Integrating business process reengineering with information systems development: Issues and implications. In W.M.P. van der Aalst, A. H.M. ter Hofstede, & M. Weske (Eds.), Business process management (BPM2003) (pp. 302-320). Berlin, Germany: Springer-Verlag.

Schuler, D., & Namioka, A. (1993). Participatory design: Principles and practices. Hillsdale, NJ: L. Erlbaum Associates.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 4, edited by J. Tan, pp. 1-16, copyright 2008 by IGI Publishing (an imprint of IGI Global).

254

255

Chapter 18

Applying Dynamic Causal Mining in Health Service Management Yi Wang Nottingham Trent University, UK

aBstract This article describes an application that illustrates the role of data mining technology in identifying hidden causal knoledge from health and medical data repositories. Across the health care and medical enterprises, a wide variety of data is being generated at a rapid rate. Current information technologies tends to focus on a more statical side of causal knowledge and do not address the dynamic causal knowledge. This article shows that the dynamic causal relation data can be captured for treatment, payment, operations purposes and administrative directed insights. Accessing this currently unrealized knowledge potential would enable the delivery of actionable knowledge to medical practitioners, healthcare system managers, policy planners and even patients to make a significant difference in overall healthcare.

lIterature reVIew Medical and Health Management Patient record management systems is desired in clinical settings (Abidi, 2001; Heathfield & Louw, 1999; Jackson, 2000). The major reasons include physicians’ significant information needs (Dawes & Sampson, 2003) and clinical information overload. Hersh (1996) classified textual health information into two main categories: patient-

specific clinical information and knowledge-based information. Knowledge management capabilities have been incorporated in many clinical systems since the 1980s in order to provide a better understanding and management basis. In the HELP system, decision logic was stored to allow it to respond to new data entered (Kuperman, Gardner, & Pryor, 1991). The SAPHIRE system performs automatic indexing of radiology reports by utilizing the UMLS Metathesaurus (Hersh, Mailhot,

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Applying Dynamic Causal Mining in Health Service Management

Arnott-Smith, & Lowe, 2002). The clinical data repository at Columbia-Presbyterian Medical Center (Friedman, Hripcsak, Johnson, Cimino, & Clayton, 1990) is a database used for decision support (Hripcsak, 1993). Another clinical data repository is the University of Virginia Health System (Schubart & Einbinder, 2000). Case based reasoning also has been proposed in Montani and Bellazzi (2002). Janetzki, Allen, and Cimino (2004) use a natural language processing approach to link electronic health records to online information resources.

applyIng data MInIng In MedIcal and HealtH ManageMent Data mining has been used to extract diagnostic rules from breast cancer data (Kovalerchuk, Vityaev, & Ruiz, 2001). Data mining has also been applied to clinical databases to identify new medical knowledge (Hripcsak, Austin, Alderson, & Friedman, 2002; Prather, Lobach, Goodwin, Hales, Hage, & Hammond, 1997). Dreiseitl, Ohno-Machado, Kittler, Vinterbo, Billhardt, and Binder (2001) compare five classification algorithms for the diagnosis of pigmented skin lesions. This is similar to the measurement of algorithmic performances in other areas applications (Yang & Liu, 1999). For example, Acir and Guzelis (2004) apply support vector machines in automatic spike signal detection in ElectroEncephaloGrams (EEG). Kandaswamy, Kumar, Ramanathan, Jayaraman, and Malmurugan (2004) use artificial neural network to classify lung sound signals into different categories. Chrisman et al. (2003) incorporated biological knowledge and expression data using a Bayesian framework. Imoto et al. (2003) incorporated protein-protein interaction, protein-DNA interactions, and transcriptional factor binding site

256

information. Text mining is used to study of protein interaction network inference (Iossifov et al., 2004). System Dynamics has a number of strengths that make it especially useful in health care settings (Roberts & Hirsch, 1976). Some of the System Dynamics work in health care has been done by Pugh-Roberts Associates (Hirsch & Miller, 1974). The other area in which early work took place was community mental health (Levin & Roberts, 1976). The Soundview-Throgs Neck community examined the forces contributing to a rapid rise in heroin addiction (Levin, Kirsch, & Roberts, 1972). A generic model of ambulatory care was developed to analyse economic performance of organization (Hirsch & Bergan, 1973). Another model developed for the State of Minnesota examined the factors affecting the success of new organizations, called Human Service Boards (Hirsch, Bergan, & Frohman, 1974). Two comprehensive System Dynamics models were developed to aid manpower policy formulation in dentistry (Hirsch & Killingsworth, 1975) and nursing (Bergan & Hirsch, 1976). One model reflected the effects of such factors as the “technology gap” between the hospital and nearby community hospitals (Hirsch, Forsyth, Bergan, & Goodman, 1976). A model helped a medical school examine problems in its relationships with affiliated hospitals and restructure those relationships accordingly (Stearns, Bergan, Roberts, & Cavazos, 1978). Models created for project changes in areas’ populations, health care consequences of those changes, and shifts in utilization patterns as a result of changes in resources available, insurance coverage, and various health policies (Hirsch & Henderson, 1977). Several modelling efforts, in fact, were designed to stop at the point where a set of causal diagrams had been completed (Stearns, Bergan, Roberts, & Quigley, 1976).

Applying Dynamic Causal Mining in Health Service Management

dynamic causal Mining The Dynamic Causal Mining (DCM) algorithm was discovered in 2005 (Pham, Wang, & Dimov, 2005) using only counting algorithm to integrate with Game Theory. It was extended in 2006 (Pham, Wang, & Dimov, 2006) with delay and feedback analysis, and was further improved for the analysis in Game Theory with Formal Concept analysis (Wang, 2007). DCM enables the generation of dynamic causal rules from data sets by integrating the concepts of Systems Thinking (Senge, Kleiner, Roberts, Ross, & Smith, 1994) and System Dynamics (Forrester, 1961) with Association mining (Agrewal et al., 1996). The algorithm can process data sets with both categorical and numerical attributes. Compared with other Association Mining algorithms, DCM rule sets are smaller and more dynamically focused. The pruning is carried out based on polarities. This reduces the size of the pruned data set and still maintains the accuracy of the generated rule sets. The rules extracted can be joined to create dynamic policy, which can be simulated through software for future decision making. The rest of this section gives a brief review of Association Mining and System Dynamics.

association Mining and process Mining Association mining was discovered by Agrewal, Mannila, Srikant, Toivonen, and Inkeri (1996). It was further improved in various ways, such as in speed (Agrewal et al., 1996; Cheung, Han, Ng, & Fu, 1996) and with parallelism (Zaki, Parthasarathy, Ogihara, & Li, 1997) to find interesting associations and/or correlation relationships among large sets of data items. It shows attributes value conditions that occur frequently together in a given dataset. It generates the candidate itemsets by joining the large itemsets of the previous pass and deleting those subsets which are small in the previous pass without considering the transactions

in the database. By only considering large itemsets of the previous pass, the number of candidate large itemsets is significantly reduced. The idea of process mining was investigated in contexts of software processes and workflow (Agrawal, Gunopulos, & Leymann, 1998). Cook and Wolf (1998) propose methods for process discovery in case of software sequential processes. Herbst and Karagiannis used a hidden Markov model in the context of workflow management and other sequential processes (Herbst, 2000; Herbst & Karagiannis, 1999, 2000). Maruster, van der Aalst, Weijters, van den Bosch, and Daelemans (2001) suggest a technique for discovering the underlying process from hospital data assuming that the workflow log does not contain any noisy data. A heuristic method that can handle noise is presented in Weijters and van der Aalst (2001).

system thinking and system dynamics System thinking is based on the belief that the component parts of a system will act differently when isolated from its environment or other parts of the system. Systems thinking is about the interrelated actions which provide a conceptual framework or a body of knowledge that makes the pattern clearer Senge et al. (1994). It is a combination of many theories such as soft systems approach and system theory (Coyle, 1996). Systems thinking seeks to explore things as wholes, through patterns of interrelated actions. System Dynamics can be defined as “a qualitative and quantitative approach to describe model and design structures for managed systems in order to understand how delays, feedback, and interrelationships among attributes influence the behaviour of the systems over time” (Coyle, 1996), or “the purpose of the model is to solve a problem, not simply to model a system. The model must simplify the system to a point where the model replicates a specific problem”(Sterman, 1994).

257

Applying Dynamic Causal Mining in Health Service Management

System dynamics is a tool to visualize and understand such patterns of dynamic complexity, which is built up from a set of system archetypes based on principles in System thinking (Sterman, 2000). System dynamics visualizes complex systems through causal loop diagrams. A causal loop diagram consists of a few basic shapes, that together describe the action modelled. System dynamics addresses two types of behaviour, sympathetic and antipathetic (Pham et al., 2005). Sympathetic behaviour indicates an initial quantity of a target attributes starts to grow, and the rate of growth increases. Antipathetic behaviour indicates an initial quantity of a target attributes starts either above or below a goal level and over time moves toward the goal.

revision of the models. The stages are depicted in Figure 1. Stage 1: Problem definition. In this phase, the problem is identified and the key variables are given. Also, the time horizon is defined so that the cause and effects can be identified. Stage 2: Data preparation. Data are collected from various sources and a homogeneous data source is created to eliminate the representation and encoding differences. Stage 3: Data mining. This stage involves transforming data into rules. This thesis suggests using DCM as a data mining tool. The details of DCM are explained in the following sections and next article. Stage 4: Policy formulation. Policies are groups of the rules extracted by mining techniques. Policies improve the understanding of the system. The interactions of different policies must also be considered since the impact of combined policies is usually not the sum of their impacts alone. These interactions may reinforce each other or have an opposite ef-

BasIc concepts oF dynaMIc causal MInIng This section provides a general framework for DCM. It is an iterative and continual process of mining rules, formulating policies, testing, and

Figure 1. DCM process

Model Simulation

Policy Formulation

Data mining

Data preparation

Problem definition

258

Applying Dynamic Causal Mining in Health Service Management

fect. The policy can be used for behaviour simulation to predict the future outcome. Stage 5: Model Simulation. This stage tests the accuracy of the policies. The policies will predict results for new cases so the managers can alter the policy to improve future behaviour of the system. It is necessary to capture the appropriate data and generate a prediction in real time, so that a decision can be made directly, quickly, and accurately.

dataset To find dynamic causality among a set of attributes means to identify correlation and interdependencies between them. The DCM algorithm is a way of describing the state of a target system as it evolves in time. It discovers dynamic causality in a data set by matching the dynamic behaviour between separated attributes.

time stamp Definition 1: A dynamic time stamp is created from two time stamps. Consider two time stamps ti and ti+1. The dynamic time stamp ∆ti is equal to the difference between two consecutive time stamps. ∆ti = ti+1 – ti

(1)

Time stamps are used for identifying the range of variables. The size of each time stamp is selected by the specific need and may vary in different situations. For instance, a time stamp for an increase in production may be in the order of months, while for a change in a cell may be in the order of milliseconds. The time stamp also can help to determine how detailed the variables need to be. The attribute may increase or decrease dramatically if the time stamp is in the order of seconds, however it may be assumed to be constant if the time stamp is in the order of years. All the time stamps should be of uniform length.

In order to carry out DCM, the time stamps are summarised or partitioned into equal-sized time stamps. A time stamp is useful for describing and prescribing changes to the systems and objects.

data Definition 2. A dynamic attribute is the change or the difference between two attribute values with consecutive time stamps. The two types of value do not have the same nature. Let D denote a data set which contains a set of n records with attributes {A1, A2, A3,… Am}, where each attribute is of a unique type (for example; cost of medicine, treatment volume, inventory volume, etc). Each attribute is associated with a time stamp ti, where i ={1,2,3,…n}. Let Dnew be a new database constructed from D such that dynamic attribute ∆Am,∆t in Dnew is given by: i

∆Am,∆t = Am,t i

i+1

- Am,t

i

(2)

where m identifies the attribute of interest. The classical Association Mining algorithms can be applied only to data in the original form (attribute form), for example, in the market basket problem (Agrawal et al., 1993) the focus is on the items of each purchase. On the other hand, DCM is interested in the dynamic changes between data. To apply DCM, the records are arranged in a temporal sequence (t = 1, 2 ,…, n). Definition 2 is only for numerical attributes and the causality between categorical attributes in D can be identified by examining the differences of corresponding changes in attribute values. An example of such a database is shown in Table 1. A1 and A2 represent attributes, such as from a tax database like income and tax. Definition 3. In the case of categorical attributes, the dynamic causal attributes can be identified by joining the polarities of corresponding changes in attribute values. Let Dnew be a new data set

259

Applying Dynamic Causal Mining in Health Service Management

constructed from D such that attribute ∆Am,∆t in i Dnew is given by: ∆Am,∆t = join(Am,t , Am,t ) i

i+1

i

Table 1. Original database D Time

(3)

where join is a function combining Am,t and Am,t . i+1 i For example, ( t=1, Am,t = Red) and (t=2, Am,t = i+1 i Blue) then ( ∆t =1, ∆Am,∆t = RedBlue). i

The attribute is gathered first and the dynamic attribute is derived from the attribute. The dynamic attribute identifies the significant relationship between the dynamics of the attribute.

MeasureMents Since the input of the DCM algorithm can be quite large, it is important to prune away the redundant attributes. Definition 4. A polarity indicates the direction of a change of an attribute. There are three types of polarity (+, -, 0); where + indicates an increase, - indicates a decrease, and 0 indicates neutrality, that is, no change at all. Definition 5. A polarity combination is a joint set of two or more polarities. The simultaneous presence of combinations (+,+) and (-,-) indicates sympathetic changes and will produce a sympathetic rule. The simultaneous presence of combinations (+,-) and (-,+) indicates antipathetic changes and produces antipathetic rules. There are four different combinations of polarity (+,-) (antipathetic negative), (+,+) (sympathetic positive), (-,+) (antipathetic positive), and (-,-) (sympathetic positive) to indicate the degree of causality. This differs from the classical causal loops relation which has only + and -, due to a simultaneous increase of an attribute set not automatically leading to a simultaneous decrease of the same set.

260

A1

A2

1

9

2

2

17

3

3

10

12

4

4

16

5

7

24

Definition 6. A support is the ratio of records of a certain polarity combination over the total number of records in the dynamic attributes. Three supports are applied in DCM; sympathetic support, antipathetic support, and single support. For data set Dnew and any two attributes ∆A1,∆t and ∆A2,∆t i i the three kinds of supports are defined as follows: Sympathetic Support (∆A1,∆t , ∆A2,∆t ) = freq (+, + ) i i n (4a) or

freq ( -, -) n

(4b)

Antipathetic Support (∆A1,∆t , ∆A2,∆t ) = freq (+, -) i i n (5.a) or

freq ( -, +) n

(5.b)

Single Attribute Support (∆Am,∆t ) = freq ( ) i n (6.a)

Table 2. Derived database Dnew ∆t

∆A1

∆A2

∆t1

+8

+1

∆t2

-7

+9

∆t3

-6

+4

∆t4

+3

+8

Applying Dynamic Causal Mining in Health Service Management

Dynamic attribute value

Figure 2. Graph of two dynamic attributes

Time stamps

or

freq ( ) n

(6.b)

or

freq (0) n

(7.c)

Where freq (+,+) is a function counting the number of times where an increase in ∆A1,∆t is i associated with a simultaneous increase in ∆A2,∆t , i freq (-,-) is a function counting the number of times where an decrease in ∆A1,∆t is associated i with a simultaneous decrease in ∆A2,∆t , freq i (-,+) is a function counting the number of times where an decrease in ∆A1,∆t is associated with a i simultaneous increase in ∆A2,∆t , and freq (+,-) is a i function counting the number of times where an decrease in ∆A1,∆t is associated with a simultanei ous decrease in ∆A2,∆t . i All supports relate to the frequencies of the occurring patterns. For a given user specified support, the problem of DCM is to find all rules where the support is greater than the user defined

support. The support is the frequency of occurrences of attribute sets that support a rule. Figure 2 shows the results of the comparison between two dynamic attributes. The X-axis represents the time stamp and the Y-axis represents the value of the dynamic attributes. Figure 3 shows the graphical representation of two dynamic attributes, where the polarity combination is indicated. Definition 7. A support level is the value threshold for each dynamic attribute. Every record in a dynamic attribute must have an absolute value larger or equal to the support level in order to be considered as candidate for the dynamic rule. For a given support level, a positive and negative value of the support level can be then drawn as shown in Figure 3. The support is the occurrence of the polarity combination above the value of the support level. Definition 8. A frequent dynamic set is a pair of dynamic attributes which contain a polarity combination with frequency occurrence above a user-defined support threshold.

261

Applying Dynamic Causal Mining in Health Service Management

Figure 3. Illustration of polarity combination Value

sympathetic *

Support level

antipathetic * * *

0 - Support level

* sympathetic

Table 3. Derived database Dnew with arrows indicating support counting direction ∆t

∆A1

∆A2

∆t1

+8

+1

∆t2

-7

+9

∆t3

-6

+4

∆t4

+3

+8

Theorem 1. If a pair of dynamic attributes (∆A1, ∆A2) is infrequent, then either one individiual dynamic attribute is infrequent or both dynamic attributes are infrequent. Proof: If a dynamic attribute set is frequent, then this indicates that both the dynami attributes are above the user-defined threshhold. The above theorem is a consequence of Definition 7 for the frequent dynamic set. This observation forms the basis of the pruning strategy in the search procedure for frequent dynamic sets, which has been leveraged in many Association Mining algorithms (Zaki, 2000), that only the single dynamic attribute found to be frequent needs to be extended as candidate for the rule. Theorem 2. Confidence measure is not useful in DCM analysis. Proof: The confidence measure is not used here because the total numbers of records in

262

*

time stamp *

* antipathetic

Table 4. Counting result

Supports(ΔA1, ΔA2)

(+,+)

(-,-)

(+,-,)

(-,+,)

2/4

0

0

2/4

an attribute is equal to the total number of time stamps. Thus it makes the confidence equal to the support. Instead, multiple supports are introduced to further reduce the running time and size of the relevant rules.

rule representatIon A dynamic causal rule consists of variables connected by arrows denoting the causal influences among the attributes. Figure 4 shows an example of the notation. Two attributes, A1 and A2, are linked by a causal arrow. Each causal link is assigned a polarity and the link indicates the direction of the change. In System Dynamics, a symbol x →+y can be interpreted as δy/δx > 0 and x →-y can be interpreted as δy/δx < 0. This analogy is applied in DCM and the dynamic causal rules produced by DCM can form causal diagrams, which will be used to simulate future behaviour.

Applying Dynamic Causal Mining in Health Service Management

Figure 4. Notation of a dynamic causal rule

Link polarity

+,-

Variable A1 Attribute

Variable A2 Attribute

Causal link

Definition 9. A dynamic causal rule is derived from a frequent dynamic attribute set. A dynamic causal rule can be either strong or weak. A weak rule is a set of attributes with polarity that partially fulfils equation (4), (5), or (6). A strong rule is a set of attributes with polarity that completely fulfils equation (4), (5), or (6). There are two types of strong rule, sympathetic and antipathetic. Figure 2 shows that variables A1 and A2 are causally dependent. If any two variables A1 and A2 are truly causally related, then a change of A1 causes a change of A2. This article focuses on the discovery of the causality with no time delay, which means that attributes are in same time period or interval. This implies that attributes should occur within the same time stamp. Theorem 3. The support for a strong rule is less than or equal to the half of the total time stamps. Proof: Each dynamic attribute can have only one type of polarity at one time stamp. If the occurrence of one polarity is huge, the occurrence of the other two polarities will diminish. Following the definition of strong rules, the occurrence of the polarities + and – have both to be highly frequent. Since support indicates the times each polarity pair occurred over the total time stamp, the support cannot be more than ½ of the total time stamps. A sympathetic rule causes an increase or decrease in the output of a target system. It reinforces a change with more change in the same

direction. An antipathetic rule represents an adjustment to achieve a certain goal or objective. It indicates a system attempting to change from its current state to a goal state. This implies that if the current state is above the goal state, then the system forces it down. If the current state is below the goal state, the system pushes it up. An antipathetic rule provides useful stability but resists external changes.

dynamic policy Definition 10. A dynamic policy consists of one or more dynamic causal rules. In a dynamic policy with several dynamic causal rules, each rule should share at least one dynamic attribute with other rules. Definition 11. A dynamic policy is single if it consists only of one rule and if that rule does not share any common attribute with other rules in a dynamic policy. Given a single dynamic policy where an attribute A1 is dynamically causally related to another attribute A2, then such a rule can be represented as the following function. ∆A2 = k∆A1

(8)

where kt =

A2,t A1,t

(9)

263

Applying Dynamic Causal Mining in Health Service Management

Given A1new indicating new values added in A1, a ∆A1new can be calculated based on A1new, and based on equation 8 a ∆A2new can be calculated. If given a value A2,t=0 a new set of A2 can be derived, where A2,t=2 = A2,t=1 + Anew , A2,t=3 = A2,t=2 + Anew 2,t 1 2,t 2 , and so forth. Definition 12. A dynamic policy is serial if each rule shares only one attribute with one other rule in a dynamic policy. A serial dynamic policy is open if each attribute in the policy has only one causal link, in other words, if there is a start and an ending attribute. A serial dynamic policy is closed if there is no start or ending attribute.

T = ( 1, 2, 3, ….t) and Ct = A2,t* A3,t *A4,t

IllustratIVe exaMple •

•

Given an open serial dynamic policy where an attribute A1 is dynamically causally related to another attribute A2, and A2 is dynamically causally related to another attribute A3, then such a rule can be represented as the following function. A2,T = k1,T A1,T and A3,T = k2,T A2,T= k2,T k1,T A1,T (10) •

where T = ( 1, 2, 3, ….t), k1,t =

A2,t A1,t

and k2,t =

A3,t A2,t (11)

Definition 15. A dynamic policy is complex if the policy consists of a mixture of open and closed serial dynamic policies. For a complex or closed dynamic policy there are attributes which are connected with more than one attribute. For instance, A1 is connected with A2, A3, and A4. A2, A3 and A4 are not connected. Then A1 can be represented with: A2,T* A3,T *A4,T = CT where

264

Stage 1: To establish a dynamic data set. The dynamic set is calculated based on the dataset illustrated in Table 5 and Equation 1. Table 6 shows the dynamic set after the subtraction. A1 ,A2 … An could represent treatment volume, doctor capacity, and so forth, which all varies through time. Stage 2: To prune the dynamic data set based on the specified support. Pruning is carried out to remove columns (attributes) where the level of support is below the minimum set. In this example, the support is set to 2, which means columns with two or more 0s are removed (value 0 indicates no dynamics in the attribute at the corresponding time stamp and more occasionally the set value of 0 indicates the attribute is not dynamic). Table 7 illustrates the “pruned” dynamic data set. Stage 3: To create dynamic rules, the dynamic supports are counted and calculated according to Equations (4), (5), and (6). Table 8 shows the dynamic supports for the pairs of attributes in Table 7. In this example, the user-specified support is set to 0.3, which

Table 5. Input dataset ΔA1

ΔA2

ΔA3

ΔA4

ΔA5

ΔA6

ΔA7

+8

+1

0

+6

0

0

0

-7

+9

0

-1

0

0

-7

-6

+4

0

-5

0

0

-4

+3

+8

0

-8

0

-97

+8

-1

-6

0

-3

0

97

-5

0

0

0

-6

0

0

-1

+5

+3

0

0

0

0

+5

+9

-8

0

-4

0

0

-7

+1

-5

0

+7

-15

0

+6

Applying Dynamic Causal Mining in Health Service Management

means any attribute pair with dynamic support with value larger than or equal to 0.3 is a dynamic rule. Table 7 shows how the DCM rules are generated. Note that (F1&F7) is the only strong sympathetic rule because when one of the attributes increases its value, the other will automatically increase its value and vice versa. Table 8 shows the attribute pair A1 and A7 and dynamic attributes ∆A1 and ∆A7.The strokes indicate redundant attributes and redundant dynamic attributes. Given a new attribute A1,new as shown in Table 9, from it ∆A1,new and ∆A2,new can be calculated as shown in Figure 6. Figure 4.3 shows the plots Table 6. Dynamic dataset Time

A1

A2

A3

A4

A5

A6

A7

1

9

2

10

22

20

100

13

2

17

3

10

28

20

100

13

3

10

12

10

27

20

100

6

4

4

16

10

22

20

100

2

5

7

24

10

14

20

3

10

6

6

18

10

11

20

100

5

7

6

18

10

5

20

100

4

8

11

21

10

5

20

100

9

9

20

13

10

1

20

100

2

10

21

8

10

8

5

100

8

Table 7. Pruned dataset

of A1,new and A2,new and it is clear that the two plots are causally related.

MInIng algorItHM problem Formulation Let D denote a database which contains a set of n records with attributes {A1, A2, A3,… An.}, where each attribute is of a unique type (sale price, production quantity, inventory volume, etc). Each attribute is linked to a time stamp t. To apply DCM, the records are arranged in a temporal sequence (t = 1, 2,…, n). The causality between attributes in D can be identified by examining the polarities of corresponding changes in attribute values. Let Dnew be a new data set constructed from D. A generalized dynamic association rule is an implication of the form A1 →p A2, where A1 ⊂ D, A1 ⊂ D, A1∩ A2= f and p is the polarity. The implementation of the DCM algorithm must support the following operations:

Table 8. Dynamic support (+,+)

(-,-)

(-,+)

(+,-)

F1&F2

0.3

0.1

0.2

0.2

F1&F4

0.2

0.3

0

0.2

F1&F7

0.3

0.3

0

0.1

Δ A1

Δ A2

Δ A4

Δ A7

F2&F4

0.1

0.2

0.1

0.3

+8

+1

+6

0

F2&F7

0.2

0.2

0.1

0.2

-7

+9

-1

-7

F4&F7

0.1

0.5

0.1

0

-6

+4

-5

-4

+3

+8

-8

+8

-1

-6

-3

-5

0

0

-6

-1

+5

+3

0

+5

(+,+)

(F1&F2), (F1&F7)

+9

-9

-4

-7

(-,-)

(F1&F4), (F1&F7), (F4&F7)

+6

(+,-)

(F2&F4)

+1

-5

+7

Table 9. The rules generated

265

Applying Dynamic Causal Mining in Health Service Management

Figure 5. Plot for A1,new and A2,new dynamic causal behaviour for new values

attribute value

30 20 10

A1,new

0 -10

1

2

3

4

5

6

7

A2,new

-20 -30 timestamps

1. 2.

To add new attributes. To maintain a counter for each polarity with respect to every dynamic value set. While making a pass, one dynamic set is read at a time and the polarity count of candidates supported by the dynamic sets is incremented. The counting process must be very fast as it is the bottleneck of the whole process.

The process time would be n2-n, where n is the number of attributes. This becomes a huge problem if n becomes too large. It is obvious that the task becomes much simpler if the size of n could be reduced before the search. Table 14 shows the pruned percentage of the total database based on the support. The support level is set to 0.

algorithm description

experIMent

DCM makes two passes over the data as shown in Figure 6 and Figure 7. In the first pass, the support of individual attributes is counted and the frequent attributes are determined. The dynamic values are used for generating new potentially frequent sets and the actual support of these sets is counted during the pass over the data. In subsequent passes, the algorithm initializes with dynamic value sets based on dynamic values found to be frequent in the previous pass. After the second of the passes, the causal rules are determined and they become the candidates for the dynamic policy. In the DCM process, the main goal is to find the strong dynamic causal rule in order to form a policy. It also represents a filtering process that prunes away static attributes, which reduces the size of the data set for further mining.

data preparation

266

The overall aim is to identify hidden dynamic changes. It was taken from a local hospital The original data was given as shown in Table 11. The only data of interest are the data with changes, for example sale amounts of a medicine, the time stamp, and so forth. The rest of the static data, such as the weight and the cost of the product, can be removed. This data consists of real life data. This dataset contains 65,536 attributes of metal manufacturing, with eight records in each attribute. After cleaning the data, the dynamic attributes are found as shown in Figure 16. The dynamic attribute is calculated by finding the difference between sales amounts in one month and sales amounts in the previous month.

Applying Dynamic Causal Mining in Health Service Management

Figure 6. The steps of DCM Part 1: – Preprocessing: Removal of the “least” causal data from database Part 2: – Mining: Formation of a rule set that covers all training examples with minimum number of rules Part 3: – Checking: Check if an attribute pair is self contradicting (sympathetic and antipathetic at the same time) Input: The original database, the values of the pruning threshold for the neutral, sympathetic and antipathetic supports. Output: Dynamic sets Step 1: Check the nature of the attributes in the original database (numerical or categorical). Initialize a new database with dynamic attributes based on the attributes and time stamps from original database. Step 2: Initialize a counter for each of the three polarities. Step3: Prune away all the dynamic attributes with supports above the input thresholds.

Figure 7. The checking step of DCM Input: The mined database, the values of the pruning threshold for the supports of the polarity combinations. Output: Dynamic sets Step 1. Check weather a rule is self-contradictory (a rule is both sympathetic and antipathetic). Step 2. If step 1 returns true then Retrieve the attribute pair form the preprocessed database Step Initialize a counter that includes polarity combination Step 4. For the pair of attributes Count the occurrence of polarity combination with two records each time. Prune away the pairs if the counted support is below the input threshold.

In the next step, the neutral attributes are pruned. The idea of pruning is to remove redundant dynamic attributes; thus fewer sets of attributes are required when generating rules. The first pruning is based on the single attribute support. In this case, the single attribute support is defined to be 0.5, which means that if an attribute with polarity +, -, or 0 occurs in more than half of total

time stamps, it will be pruned. In this case, 429 attributes remain for the rule generation. In this experiment, dynamic sets are compared based on a simultaneous time stamp. Then the support of sympathetic and antipathetic rules for each dynamic set is calculated. The support is used as the threshold to eliminate unsatisfactory dynamic sets and to obtain the rules from the

267

Applying Dynamic Causal Mining in Health Service Management

Table 10. Pruned results Data set

Single Support 0.05

0.10

0.15

0.20

0.25

0.30

0.35

Adult

5%

20%

27%

74%

100%

100%

100%

Bank

11%

20%

60%

94%

100%

100%

100%

Cystine

5%

33%

70%

100%

100%

100%

100%

Market basket

6%

10%

50%

86%

100%

100%

100%

Mclosom

1%

13%

38%

72%

90%

100%

100%

ASW

1%

8%

40%

68%

70%

97%

100%

Weka-base

6%

25%

52%

86%

100%

100%

100%

satisfactory sets. In Table 12 the left most colum indicates the codes for each attributes. The top row indicates the times stamps and the rest of table is dynamic attributes. The algorithm was run based on the procedures described in previous sections. Figure 8 shows the plot of sympathetic and antipathetic support. The x-axis represents the support and the yaxis represents the number of rules. This database shows that there are more sympathetic rules than antipathetic rules. The figure shows that increasTable 11. Original data sets

268

ing support will lead to exponential growth of the rules. As the support reaches 0.05 or 5, as it indicates on the figure, the number of rules is 630. Most of these rules are redundant and have no meaning due to the low support. Figure 9 shows the rule plot with support equal to average value, where the +support = the average of all positive records and –support = the average of all negative records. The number of rules has decreased by applying the support level.

Applying Dynamic Causal Mining in Health Service Management

Table 12. Dynamic attributes

Figure 8. Rule plot with support level = 0 support level = 0 700 number of rules rules

600 500 400

Sympathetic rules Antipathetic rules

300 200 100 0 0.11

>0.10

>0.09

>0.08

>0.07

>0.06

>0.05

support

Table 13 shows the extracted strong rules with support level equal to average value and support larger than 0.08. C15276179, A04004004 …… are the code for each attributes. There are only dynamic pairs so there is no need to do the simulation

dIscussIon Apriori provides some form of causal information, that is, suggesting a possible direction of causation between two attributes, but there is no basis to conclude that the arrow indicates direct

269

Applying Dynamic Causal Mining in Health Service Management

Figure 9. Rule plot with support level = frequent support level = average vale

number of rules

350 300 250 200

Sympathetic rules Antipathetic rules

150 100 50 0 >0.11

>0.10

>0.09

>0.08

>0.07

>0.06

>0.05

support

Table 13. Result generated by the algorithm Strong rules Sympathetic {C15276179, F030008} {J08008008, F060010} {A04004004, A05005005} {A05005006, C10251104} {A04004004, F100020}

0,093 0,089 0,086 0,084 0,082

Antipathetic {A05005008, C15276179} {C10251104, F070010} {A05005008, F030008}

0.092 0.083 0.082

or even indirect causation. The DCM algorithm, on the other hand, shows causality between attributes. Thus, where association rule generation techniques find surface associations, causal inference algorithms identify the structure underlying such associations. Each type of relationship generated by the DCM algorithm provides additional information. The DCM algorithm finds four kinds of relationships, each of which deepens the user’s understanding of their target system by constructing the possible models. For example, A1 →+ A2 provides more information than A1→ A2 because the latter indicates that A1 coexists with A2. The condition of the rule is not stated (whether sym270

Support

pathetic or antipathetic). A genuine causality such as A1 →+ A2 provides useful information because it indicates that the relationship from A1 to A2 is strictly sympathetic causal. The rules extracted by DCM can be simulated by using software to model the future behaviour. The rules extracted by association algorithm cannot be simulated.

conclusIon and suMMary This article provides insight into how to use dynamic causal mining approaches to medical and health management research can be addressed

Applying Dynamic Causal Mining in Health Service Management

through management strategies specifically designed to respond. The algorithm shows opportunities for more extensive examination of dynamics between attributes as well as produces a framework for such analysis. A mining framework will provide medical and health management researchers, as well as researchers interested in other complex social phenomena, with insights into determining the best research design given a specific context. This article addresses important and challenging issues of causal analysis within e-government and implementation of causality extraction and construction algorithms for medical and health management data. This article also studies the computational costs of causal, and implements a multipruning approach for building models that incurs minimal computational cost while maintaining accuracy and unsupervised causality extraction algorithms to reduce the reliance on medical and health management data.

reFerences Abidi, S.S.R. (2001). Knowledge management in healthcare: Towards knowledgedriven decisionsupport services. International Journal of Medical Informatics, 63, 5-18. Acir, N., & Guzelis, C. (2004). Automatic spike detection in EEG by a two-stageprocedure based on support vector machines. Computers in Biology and Medicine,34(7), 561-575. Agrawal, R., Gunopulos, D., & Leymann, F. (1998). Mining process models from workflow logs. In Sixth International Conference on Extended Database Technology (pp. 469-483). Agrawal, R., Mannila, H., Srikant, R., Toivonen, H., & Inkeri, A. (1996). Fast discovery of association rules, advances in knowledge discovery and data mining. The Association for the Advancement of Artificial Intelligence and the MIT Press (pp. 307-328).

Bergan, T., & Hirsch, G.B. (1976). Analysis and Planning for improved distribution of Nursing personnel and services. A national model of supply, demand, and distribution. Cambridge, MA: Pugh-Roberts Associates, Inc. Cheung, D., Han, J., Ng, V.T., & Fu, A.W. (1996). Fast distributed algorithm for mining association rules. In International Conference on Parallel and Distributed Information Systems, Tokyo, Japan (pp. 31-42). Cook, J.E., & Wolf, A.L. (1998, November). Eventbased detection of concurrency. In Proceedings of the Sixth International Symposium on the Foundations of Software Engineering (FSE-6), Orlando, FL (pp. 35-45). Coyle, R.G. (1996). System dynamics modelling: A practical approach. London: Chapman and Hall. Dawes, M., & Sampson, U. (2003). Knowledge management in clinical practice: Asystematic review of information seeking behavior in physicians. International Journal of Medical Informatics, 71, 9-15. Dreiseitl, S., Ohno-Machado, L., Kittler, H., Vinterbo, S., Billhardt, H., & Binder, M. (2001). A comparison of machine learning methods for the diagnosis of pigmented skin lesions. Journal of Biomedical Informatics, 34, 28-36. Flood, R. (1999). Rethinking the fifth discipline. Learning within the unknowable. London: Ruthledge. Forrest, S., Hofmeyr, S.A., Somayaji, A., & Longstaff, T.A. (1996). A sense of self for unix processes. In 1996 IEEE Symposium on Security and Privacy (pp. 120-128). IEEE Computer Society. Building the virtual state. Information technology and institutional change. Washington, DC: Brookings Institution Press. Forrester, J.W. (1961, July-August). Industrial Dynamics: A Major Breakthrough for Decision Makers. Harvard Business Review, 37-66.

271

Applying Dynamic Causal Mining in Health Service Management

Friedman, C., Hripcsak, G., Johnson, S.B., Cimino, J.J., & Clayton, P.D. (1990). Ageneralized relational schema for an integrated clinical patient database. In Proceedings of the 14th Annual Symposium on Computer Applications in Medical Care (pp. 335-339). Heathfield, H., & Louw, G. (1999). New challenges for clinical informatics: Knowledge management tools. Health Informatics Journal, 5(2), 67-73. Herbst, J. (2000). A machine learning approach to workflow management. In 11th European Conference on Machine Learning. Berlin, Germany: Springer (LNCS 1810, pp. 183-194). Herbst, J., & Karagiannis, D. (1999, August), An inductive approach to the acquisition and adaptation of workflow models. In M. Ibrahim & B. Drabble (Eds.), Proceedings of the IJCAI’99 Workshop on Intelligent Workflow and Process Management: The New Frontier for AI in Business (pp. 52-57). Stockholm, Sweden. Herbst, J., & Karagiannis, D. (2000). Integrating machine learning and workflow management to support acquisition and adaptation of workflow models. International Journal of Intelligent Systems in Accounting, Finance and Management, 9, 67-92. Hersh, W., Mailhot, M., Arnott-Smith, C., & Lowe, H. (2002). Selective automated indexing of findings and diagnoses in radiology reports. Journal of Biomedical Informatics, 34, 262-273. Hripcsak, G. (1993). Monitoring the monitor: Automated statistical tracking of a clinical event monitor. Computers and Biomedical Research, 26(5), 449-466. Hripcsak, G., Austin, J.H., Alderson, P.O., & Friedman, C. (2002). Use of natural language processing to translate clinical information from a database of 889,921 chest radiographic reports. Radiology, 224(1), 157-163.

272

Hirsch, G.B., & Bergan, T.A. (1973). Simulating ambulatory care systems: The relationship between structure and behaviour. In Proceedings of the Summer Computer Simulation Conference, Montreal, Quebec, Canada (pp. 877-884). Hirsch, G.B., Bergan, T.A., & Frohman, A.L. (1974). Organizational change in the provision of human services. In Proceedings of the Summer Computer Simulation Conference, Houston, Texas. Hirsch, G.B., Forsyth, B.R., Bergan, T.A., & Goodman, M.R. (1976). Projecting a teaching hospital’s future utilization: A dynamic simulation approach. Journal of Medical Education, 51(11), 937-939. Hirsch, G.B., & Henderson, J.D. (1977). A simulation model for “what if?” questions in plan development. American Journal of Health Planning, 2(1), 40-46. Hirsch, G.B., & Killingsworth, W.R. (1975). A new framework for projecting dental manpower requirements. Inquiry, 12(2), 126-142. Hirsch, G.B., & Miller, S. (1974). Evaluating HMO policies with a computer simulation model. Medical Care, 12(8), 668-681. Jackson, J.R. (2000). The urgent call for knowledge management in medicine. The Physician Executive, 26(1), 28-31. Janetzki, V., Allen, M., & Cimino, J.J. (2004). Using natural language processing to link from medical text to online information resources. In Proceedings of Medinfo (pp. 1665). Kandaswamy, A., Kumar, C.S., Ramanathan, R.P., Jayaraman, R., & Malmurugan, N. (2004). Neural classification of lung sounds using wavelet coefficients. Computers in Biology and Medicine, 34, 523-537.

Applying Dynamic Causal Mining in Health Service Management

Kovalerchuk, B., Vityaev, E., & Ruiz, J.F. (2001). Consistent and complete data and expert mining in medicine. In K.J. Cios (Ed.), Medical data mining and knowledge discovery. New York: Physica-Verlag.

Prather, J.C., Lobach, D.F., Goodwin, L.K., Hales, J.W., Hage, M.L., & Hammond, W.E. (1997). Medical data mining: Knowledge discovery in a clinical data warehouse. In Proceedings of the AMIA Annual Symposium Fall 1997 (pp. 101-105).

Kuperman, G.J., Gardner, R.M., & Pryor, T.A. (1991). The HELP system. New York: SpringerVerlag.

Richardson, G.P. (1996). System dynamics. In S.I. Gass & C.M. Harris (Eds.), Encyclopedia of operations research and management science (pp. 656-660). Boston, MA: Kluwer Academic Publishers.

Levin, G., Hirsch, G.B., & Roberts, E. (1972). Narcotics and the community: A system simulation. American Journal of Public Health, 62, 8-6. Levin, G., & Roberts, E.B. (1976). The dynamics of human service delivery. Cambridge, MA: Ballinger Publishing Company. Maruster, L., van der Aalst, W.M.P., Weijters, T., van den Bosch, A., & Daelemans, W. (2001, October 25-26). Automated discovery of workflow models from hospital data. In B. Kröse et al. (Eds.), Proceedings 13th Belgium- Netherlands Conference on Artificial Intelligence (BNAIC’01), Amsterdam, The Netherlands (pp. 183-190). Montani, S., & Bellazzi, R. (2002). Supporting decisions in medical applications: The knowledge management perspective. International Journal of Medical Informatics, 68, 79-90. Pham, D.T., Wang, Y., & Dimov, S. (2005). Intelligfent manufacturing strategy selection. In Proc 1st International Virtual Conference on Intelligent Production Machines and Systems (pp. 312-318). Oxford: Elsevier. Pham, D.T., Wang, Y., & Dimov, S. (2006). Incorporating delay and feedback in intelligent manufacturing strategy selection. In Proc 2nd International Virtual Conference on Intelligent Production Machines and Systems (pp. 246-252). Oxford: Elsevier.

Richardson, G.P. (1999). Citation for winner of the 1999 Jay W. Forrester Award: Jac Vennix. System Dynamics Review, 15 4, 375-377. Richardson, G.P., & Andersen, D.F. (1995). Teamwork in group model building. System Dynamics Review, 11(2), 113-137. Roberts, E.B., & Hirsch, G.B. (1976, Winter). Strategic modelling for health care managers. Health Care Management Review, 69-77. Schubart, J.R., & Einbinder, J.S. (2000). Evaluation of a data warehouse in an academic health sciences center. International Journal of Medical Informatics, 60, 319-333. Senge, P., Kleiner, A., Roberts, C., Ross, R., & Smith, B. (1994). The fifth discipline fieldbook. New York: Doubleday Stearns, N.S., Bergan, T.A., Roberts, E.B., & Cavazos, L.F. (1978). A systems intervention for improving medical school-hospital interrelationships. Journal of Medical Education, 53, 464-472. Stearns, N.S, Bergan, T.A., Roberts E.B., & Quigley, L. (1976, Fall). Systems intervention: New help for hospitals. Health Care Management Review, 1, 9-18. Sterman, J.D. (1994). Learning in and about complex systems. System Dynamics Review, 10(2-3), 291-330.

273

Applying Dynamic Causal Mining in Health Service Management

Sterman, J. D. (2000). Business dynamics: Systems thinking and modelling for a complex world. Boston, MA: Irwin McGraw-Hill. Yang, Y., & Liu, X. (1999). A re-examination of text categorization methods. In Proceedings of the 22nd Annual International ACM Conference on Research andDevelopment in Information Retrieval (SIGIR’99) (pp. 42-49). Wang, Y. (2007). Integration of data mining with game theory. Journal of International Federation for Information Processing, 207/2006, 275-280. Boston, MA: Springer

Weijters, T., & van der Aalst, W.M.P. (2001, October 25-26). Process mining: Discovering workflow models from event-based data. In B. Kröse et al. (Eds.), Proceedings 13th Belgium-Netherlands Conference on Artificial Intelligence (BNAIC’01), Amsterdam, The Netherlands (pp. 283-290). Zaki, M.J., Parthasarathy, S., Ogihara, M., & Li, W. (1997). New parallel algorithms for fast discovery of association rules. Data Mining and Knowledge Discovery, 1(4), 343-37.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 4, edited by J. Tan, pp. 17-38, copyright 2008 by IGI Publishing (an imprint of IGI Global).

274

275

Chapter 19

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems: An Introduction and Literature Survey Christos Vasilakis University College London, UK Dorota Lecnzarowicz University of Westminster, UK Chooi Lee Kingston Hospital, UK

aBstract The unified modelling language (UML) comprises a set of tools for documenting the analysis of a system. Although UML is generally used to describe and evaluate the functioning of complex systems, the extent of its application to the health care domain is unknown. The purpose of this article is to survey the literature on the application of UML tools to the analysis and modelling of health care systems. We first introduce four of the most common UML diagrammatic tools, namely use case, activity, state, and class diagrams. We use a simplified surgical care service as an example to illustrate the concepts and notation of each diagrammatic tool. We then present the results of the literature survey on the application of UML tools in health care. The survey revealed that although UML tools have been employed in modelling different aspects of health care systems, there is little systematic evidence of the benefits.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

IntroductIon Health care systems are known to be complex and, as a result, difficult to analyse and re-engineer (Berwick, 2005). Health system engineers often rely on computer modelling and simulation to assist with the analysis of existing systems and the pretesting of suggested changes. To this extend, a variety of software engineering techniques and tools have been employed (Jun, 2007). Examples include data flow diagram (Pohjonen et al., 1994), state transition diagram (Mehta, Haluck, Frecker, & Snyder, 2002), entity relationship diagram (Kalli et al., 1992), integrated definition or IDEF (Hoffman, 1997), and more recently, Unified Modelling Language, commonly known as UML (Object Management Group, 2005). UML provides a comprehensive set of tools that can be used for documenting the analysis of a system and for developing model requirements. UML diagrams are graphical depictions that demonstrate the flow of events within the system (Object Management Group, 2005). Depending on the perspective chosen for the study (e.g., actor oriented, activity oriented), different tools are available to the analyst. Due to its versatility and the ability to analyse systems from different perspectives, UML is said to be effective in describing and evaluating the functioning of complex systems such as health care (Kumarapeli, De Lusignan, Ellis, & Jones, 2007). However, there seems to be very little systematic evidence on its benefits. The focus of the article is to review the literature on the application of UML tools to the analysis and modelling of health care systems. To this end, we first briefly introduce four of the most common UML diagrammatic tools, namely use case, activity, state, and class diagrams. We use a simplified surgical care service as an example to illustrate the notation and concepts of each diagrammatic tool. Next, we present the results of the literature survey on the application of UML in health. The survey revealed that studies of the benefits of UML to health evaluation are

276

an exception and most studies have used UML without an evaluative component. We conclude with a brief discussion of the results.

uMl dIagraMMatIc tools UML 2.0 has 13 types of diagrams, which can be categorised hierarchically as follows (Object Management Group, 2005): •

•

•

Structure diagrams used to represent the elements of the system being modelled. They include class, component, composite structure, deployment, object, and package diagrams. Behaviour diagrams that allow the representation of what happens in the modelled system in the activity, state, and use case diagrams. Interaction diagrams, a subset of behaviour diagrams, that allow the representation of the control and data flow among the elements of the system being modelled. These are communication, interaction overview, sequence, and timing diagrams.

We briefly introduce here the four UML diagrammatic tools that appear in the surveyed literature, namely, use cases and use case diagram, activity, state, and class diagram. A full description of the concepts and syntax of UML diagrams is beyond the scope of this article. A plethora of user guides and technical notes are available on the subject, with the monograph by Ambler (2004) a particularly useful introduction. We illustrate the basic concepts and notation of each diagrammatic tool by presenting simple models of a simplified care process of surgical consultation with a patient in an outpatient clinic. In general, physicians refer patients for surgical consultation if they believe the underlying health problem is amenable to surgical intervention. Following the referral, the outpatient clinic books

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Figure 1. UML use case diagram of a simplified surgical care service (surgical consultation in outpatient clinic) Surgical Consultation «extends»

Order tests «extends»

Book appointment Take samples

Nurse

Register on surgical waiting list «extends» Educate on operation

Decide on treatment

Patient requiring surgical consultation «uses»

Surgeon «uses»

Report on samples

Evaluate symptoms and test results

«uses»

Analyse samples

Patient

Lab technician

the patient an appointment with the surgeon and also arranges for samples to be taken if further diagnostic tests are required. At the consultation, the surgeon assesses the need for an operation by evaluating symptoms and test results. Following a decision to operate, the patient’s name is registered on a prioritised surgical wait list so that appropriate time can be booked at the operating theatre of a hospital. The patient may also be educated about the operation by a specialist nurse. If an operation is not deemed suitable, then the patient may be further referred for medical treatment.

use cases and uMl use case diagrams In software and system engineering, a use case is a technique for capturing the functional require-

ments of a system (Object Management Group, 2005). Each use case provides one or more scenarios that convey how a specific part of the system interacts with the users (called actors) to achieve a business goal or function. There is no standard format for detailing use cases but some tabular layout is commonly used. The UML use case diagram, on the other hand, allows the graphical representation of a set of use cases. The UML standard sets out a specific graphical notation (Object Management Group, 2005). Use cases and UML use case diagrams not only provide clarity in terms of actors and sequence of steps involved in the event but also serve as a useful tool to present details of the actor’s progression in the system. Figure 1 shows a use case diagram depicting the process of surgical consultation with a patient 277

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

in outpatient clinic. Actors are represented by stick figures and use cases by ovals. Associations between actors and use cases are depicted by edges. The “uses” arrow points to a use case that is always invoked, while the “extends” to a use case that is conditionally invoked. In the example, the surgeon always evaluates the patient’s symptoms and the test results. The outpatient clinic nurse may order additional tests when booking the appointment, which in turn, may require taking samples (e.g., blood) from the patient. The actor “Patient requiring surgical consultation” is a type of the generic actor “Patient”, an example of the common construct of specialisation/generalisation.

uMl activity diagram The purpose of the activity diagram is to depict the procedural flow of actions that are part of a larger activity (Object Management Group, 2005). In projects in which use cases are generated, activity diagrams can model a specific use case at a more detailed level. Activity diagrams can be also used independent of use cases for modelling a function, such as admission to hospital or discharge procedure. They can also be used to model system functions, such as computerised physician order systems, and complete patient pathways, such as from admission to hospital to discharge. Activity diagrams also allow the depiction of parallel activities that often occur in health systems. In the UML activity diagram, which is based on the semantics of Petri nets, each activity is represented by a rounded rectangle. An arrow represents the transition from one activity to another. The starting point is represented by a filled-in circle and the endpoint by a bull’s-eye. Activities enclosed within parallel bars happen at the same time. Diamond shaped objects denote a decision mandated by conditions stated in the brackets above the arrows. Figure 2 shows a UML activity diagram that models the exemplar surgical care process. Fol-

278

lowing the referral to the outpatient clinic, booking the appointment and ordering of the diagnostic tests happen in parallel (for the purposes of this example). Depending on the outcome of the consultation, the patient’s name is registered on the surgical waiting list if operation is deemed necessary and the patient may receive education about the operation, otherwise the patient is referred for medical treatment.

uMl state diagram The UML state diagram is essentially a Harel (1987) Statechart with standardized notation that can describe any system that is (or can be conceptualised as) reactive, from computer programs to business processes. In this context, a reactive system—as opposed to a transformational system—is a system that constantly responds to internal and external stimuli by changing states or by performing some action. Like state machines, a UML state diagram includes state-transition diagrams that represent the operations of a system through discrete states and transitions from one state to another. In addition, state diagrams include notions of state hierarchy, parallelism, and event broadcasting (Sobolev, Harel, Vasilakis, & Levy, 2008). In UML state diagrams, rectangles represent states and arrows represent transitions. An arrow may have a transition label that controls the transition. The label includes the events that trigger the transition, and the condition that needs to be true for the transition to occur in square brackets. The actions associated with the transition also appear on the labels following the forward slash. Drawing states inside other states represents hierarchy. Dashed rectangles symbolise parallel states. Figure 3 shows a UML state diagram of patient states in the example of surgical care service. Following the referral to clinic, the initial substates called “pending” of parallel states “appointment” and “diagnostic tests” are activated. When the event “make booking” is fired and if there are

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Figure 2. UML activity diagram of the care process of a simplified surgical care service (surgical consultation in outpatient clinic)

Referral to clinic

Booking of appointment with surgeon

Ordering of diagnostic tests

Consultation with surgeon

[Operation is deemed necessary ]

[Otherwise]

Referral for medical treatment

Registration on surgical waiting list

Education about operation

available slots in the clinic, the patient is considered to have the appointment booked. Once the treatment has been decided and depending on the outcome, the patient state transitions to “surgical waiting list” and “education,” or to “waiting for medical treatment.”

uMl class diagram A class diagram describes the static structure of a system by showing the system’s classes, their attri-

butes and methods, and the relationships between the classes (Object Management Group, 2005). A class, indicated by a rectangle, can be thought of as a blueprint for defining similar objects. Each object is an instance of a class and encapsulates both state, in terms of attributes, and behaviour, in terms of methods. Attributes (or properties) are shown in an optional compartment below the class name. Each attribute is shown with at least its name, and optionally with its type, initial value, and other information. The class methods (or operations) appear in a second optional compartment.

279

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Figure 3. UML state diagram of patient states in a simplified surgical care service (surgical consultation in outpatient clinic) Patient at surgical consultation in outpatient clinic

Waiting for referral to clinic referral to clinic

Appointment

Diagnostic tests

Pending

Pending

take samples

make booking [ slots available]

Samples taken

Booked

patient presents at clinic

At consultation treatment decided [operation deemed necessary ]

Surgical waiting list

E ducation

Pending

Pending

register

educate

On waiting list

Educated

book operation

Each method is shown with at least its name, and additionally with its parameters and return type. The association between two classes is indicated by a line. The number of objects participating in the association, known as multiplicity, is given by an optional notation at each end of the line (“0..1” if none or only one object participates in the association, “1” exactly one, “0..*” zero or more, “1..*” at least one).

280

[otherwise]

Waiting for medical treatment

referral for medical treatment

In the example shown in Figure 4, a patient may have none, one, or more referrals (on different dates). Each referral has a priority, is made to the named surgeon, and the appointment slot is updated once it has been scheduled. Patient names are placed on the surgical list of a surgeon following the consultation. A surgeon may be associated with no patients or many. A class diagram may convey a lot more information that is omitted here in the interests of brevity.

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Figure 4. UML class diagram of a simplified surgical care service (surgical consultation in outpatient clinic)  placed 1

patient -name -dob -gender +add patient () +remove patient ()

 is made for 1

1 has 0..*

0..*

0..*

surgical waiting list -patient name -type of operation -priority -surgeon name -placement date -removal date +add patient name () +remove patient name () +audit waiting list()

referral -patient name -date -priority -surgeon -appointment date +add referral () +remove referral () +update appointment date ()

diagnostic test -test -patient name -surgeon name -result -date ordered -date of result +add order () +update result ()

0..*

0..*  orders

0..*

1

 operates on 1

surgeon -name -specialty +add surgeon () +remove surgeon ()

application of uMl tools to Health care For the literature survey, we searched the medical literature for articles that demonstrate the application of UML tools to the broad area of health care. We included all articles that demonstrated the application of any of use case and UML use case diagram, UML activity diagram, UML state diagram, and UML class diagram. Eighteen papers were identified to be relevant to this review, found via published (as listed in Pubmed) and grey literature search. The literature survey identified use cases and use case diagrams (11 papers), and activity diagrams (13) as the most common UML tools of the four included in this survey used in health system analysis. Some studies have reported on the use of class diagrams (3) but none on the use of UML state diagrams. The search yielded very little evidence of a systematic evaluation

consults w ith 1

of the benefits of using UML in the analysis of health care systems. The retrieved literature can broadly be classified into three categories according to the application domain: modelling health care processes, evaluating and modelling clinical guidelines, and evaluating and generating requirements of information systems in health care. We now briefly review each collected article according to the these categories.

Modelling Health care processes Recent work by Jun (2007) and Jun, Ward, and Clarkson (2005) aimed at providing the effective application of various modelling methods to health care with the end goal of enabling professionals and managers to understand care processes more clearly, manage risk, and as a result improve patient safety. Among the modelling methods evaluated was the swim lane activity diagram, which is a

281

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

variation of UML activity diagram where activities are grouped according to the actors involved or in a single thread. Three case studies were used to illustrate the methods: patient discharge, diabetic care process in GP practise, and prostate cancer diagnosis process in the hospital. The findings of the case studies were evaluated against benchmark goals of enhancing understanding and validated via key user review of findings. Cruz, Gramaxo Silva, Soares, Oliveira, Serrano, and Paulo Cunha (2002) reported on the experience of the using UML use case diagrams as a tool to optimise hospital processes. The diagrams were developed at two levels of abstraction: a global use case diagram to represent the main processes of the care services examined, and detailed use case diagrams to study parts of the system seen as critical bottlenecks. Activity diagrams were to be developed in subsequent project phases to model dynamic concepts of each detailed use case. Although Cruz et al. (2002) concluded that UML helped in communication, discussion, and validation of the different steps of the project, there was no evaluation of these benefits. Goossen et al. (2004) looked at the feasibility of mapping and modelling of nursing care process information to some international standards. They represented the nursing care process as a dynamic sequence of phases, each containing information specific to the activities of the phase, and used UML to represent this domain knowledge in models. A UML activity diagram was developed as a model of a generic nursing care process. After creating a structural model of the information collected at each stage of the nursing process, various working groups mapped that information to other standards as a means of validation. An activity diagram of a generic nursing process was also developed as a problem solving approach to patient care and a UML class diagram enhanced this view and demonstrated further the care process from a nursing perspective. The authors concluded that their study produced a good model of the nursing care process but that improvements could still

282

be made. However, despite the self critique of the study, the authors did not evaluate the UML activity and class diagram benefits and drawbacks as a means to study health care processes. Spyrou, Bamidis, Pappas, and Maglaveras (2005) proposed an extension of UML to model processes in the health care domain using workflow modelling techniques. The work presented in the article extends the UML activity diagram to support workflow characteristics as well as standardised clinical documents that are handled by the processes. The extended notation was then used to model the flow of patients in a regional health system. No evaluation of the benefits was reported. Lyalin and Williams (2005) aimed at improving the way cancer registration and other processes are described through enhancements in the notation of UML activity diagrams. The article illustrates a UML activity diagram used to describe the process of cancer registration and which was enhanced by allowing the depiction of timeline, duration for individual activities, responsibilities for individual activities, and descriptive text. The authors claim that this provides for clarification of the process of cancer registration and can broaden its understanding among different specialists. Lyalin and Williams (2005) included an extensive description of benefits and weaknesses of activity diagrams and a comprehensive activity diagram of death clearance process at cancer registry. The authors conclude that the enhancements add value to the tool and cite the positive response they received after using the enhanced UML activity diagrams in a cancer registry best practices development workshop. Similarly, Saboor, Ammenwerth, Wurz, and Chimiak-Opoka (2005) aimed at improving UML activity diagram by developing and testing a process modelling method which included details of clinical processes necessary for a systematic and even semiautomatic quality assessment. The premise of the authors’ work was that UML is only a descriptive tool that does not allow for

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

comprehensive quality assessment. Subsequently, Saboor et al. (2005) proposed a new modelling method based on the UML activity diagram with extra notations to allow for evaluation. The method was validated by modelling various versions of the process of ordering a radiological examination. It was suggested that further evaluation of improvements on the UML activity diagram were needed.

technology. However the authors provided no systematic evaluation of the benefits of employing UML tools in health systems, but rather focused on the utility of UML activity diagram in general.

evaluating and Modelling clinical guidelines

Maij, Toussaint, Kalshoven, Poerschke, and Zwetsloot-Schonk (2002) looked at the problem of alignment between information and communication technology (ICT) infrastructure and business processes in health care organisations. The paper investigated whether the combination of Dynamic Essential Modelling of Organisations, that is a business process modelling methodology, with UML can solve the problem. It used the example of a screening case study on the management of preoperative centres and focused on developing an efficient information system. Maij et al. (2002) used UML use cases to derive the functionality of the information systems. It also provided a use case diagram and description for a transaction at the preoperative centre. The authors concluded that the combination of the two techniques is useful in aligning business processes and functional features of ICT infrastructure and should help the end-user to develop a better understanding with regards to the relationship between the two areas. Although the paper did include a brief discussion of UML, it did not provide a systematic assessment of its utility in health care. Lee, Bakken, and John (2006) briefly reported on the use of UML tools (use case, activity, and sequence diagrams) to store and present the functional requirements of a handheld-based decision support system for morbid obesity screening and management. The authors stated that UML is useful in depicting processes related to management of clinical based guidelines, facilitating discussion and agreement in developing data model, and in aiding the design of Web-based prototype.

Sutton, Taylor, and Earle (2006) developed a computerised system to allow hypertensive patients to be monitored and assessed without visiting their family doctors. The Web-based system, created using PROforma, made recommendations for continued monitoring and for changes in medication. PROforma is a language that allows clinical guideline to be expressed in a computer-interpretable manner. The study concluded that PROforma proved adequate as a language for the implementation of the clinical reasoning but lacked notational convenience. Hence, UML activity diagrams were employed instead to create the models that were used during the knowledge acquisition and analysis phases of the project. Sutton et al. (2006) also reported on the application of UML activity diagram to represent the clinical guidelines in the management of hypertensive patients. The authors praised the notational convenient of UML but did not systematically evaluate its benefits. In similar fashion, Hederman, Smutek, Wade, and Knape (2002) compared a technique for representing and sharing clinical guidelines (GLIF) with UML activity diagram. The authors concluded that there are clear potential benefits in using a mainstream modelling language such as UML as opposed to a specific clinical guideline representation technique such as GLIF. The potential benefits include availability of modelling tools, the ability to transfer between modelling tools, and to automate via business workflow

evaluating and generating requirements of Information systems

283

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Lunn, Sixsmith, Lindsay, and Vaarama (2003) reported using UML activity diagram to model various processes of monitoring care and use case diagrams to generate requirements for the development of an information system intended to support planning in the provision of elderly care services. The study is a good example of the application of UML tools in health but provides little guidance in terms of actual benefits and weaknesses. Weber et al. (2001) aimed at developing a tool to support clinical trial centres in developing trial specific modules for the computer-based documentation system of paediatric oncology. The research carried out an object-oriented business process analysis for a clinical trial conducted at a German hospital. The results comprised a comprehensive business process model consisting of UML diagrams and use case specifications, which included use case diagrams (“manage trial,” “plan trial,” “conduct trial,” “document course of therapy,” and documentation view of the latter) and an example of use case specification. Weber et al. (2001) concluded by recommending the use of object-oriented analysis in the context of therapeutic trials but did not carry out the an evaluation. LeBozec, Jaulent, Zapletal, and Degoulet (1998) described a UML approach to the designing of a case-based medical imaging retrieval system for pathologists. The authors created UML use case and class diagrams to illustrate the steps of the case-based reasoning systems methodology used to develop sound knowledge systems in pathology. The diagrams were used to visualize the relevant objects and to evaluate the model before implementation, and included use case diagram of the image retrieval system and use case with corresponding scenarios chart. The authors concluded that, although further evaluation is required, UML seems to be a promising formalism improving the communication between the developers and users.

284

Aggarwal (2002) highlighted the benefits of the UML in specifying, visualising, constructing, documenting, and communicating the model of a health care information system. It illustrated the usage of use cases and use case diagram, activity and class diagram by employing simplified examples of a nurse submitting a blood-count order, a physician order system, and of an emergency room. Ganguly and Ray (2000) discussed the development of a methodology for the design of interoperable telemedicine systems based on UML. Their research focused on the feasibility of the development of agent-based interoperable telemedicine systems and used the example teleelectrocardiography in the case study. Among the tools suggested for system design, Ganguly and Ray (2000) used a UML use case diagram to describe a distributed electrocardiogram system. Finally, Hoo, Wong, Laxer, Knowlton, and Wan (2000) had as an objective to develop software that facilitates more efficient and effective utilisation of medical images and associated data in biomedical research. The area of focus was assisting clinicians in presurgical evaluation of patients with medically refractory epilepsy as an example. The authors drafted use cases to summarize operational scenarios of clinicians using the system and used UML class diagrams to describe object-oriented concepts of the system.

dIscussIon In this article, we introduced four common UML diagrammatic tools (use case, activity, state, and class diagrams) and used original models of a simplified example of surgical service to illustrate their usage. We also presented the findings of a literature survey on the application of these tools to the modelling of health care systems. The survey identified a number of articles in which

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

UML tools were used but very limited systematic evaluation of their benefits. One notable exception is the recent work by Jun (2007) where a variety of modelling methods, including a variation of UML activity diagrams, was evaluated for their utility in modelling health systems. Jun (2007) identified that only a limited number of modelling methods have been considered or evaluated for purpose of use in health care settings. Following systematic evaluation by health care professionals, Jun (2007) concluded that there is no single method preferred by all users or applicable to all areas but there is a strong case for using a variety of modelling techniques in enhancing the understanding of care process among practitioners. This work is, to our knowledge, the only comprehensive evaluation of different process mapping tools in health care and provided a clear insight to the benefits those methods can offer to the system, practitioners, and the patient. Similar evaluation may also be needed for all UML tools before firm recommendations can be made as to their applicability. Despite the obvious analytical applicability, also pointed out by Jun (2007), state diagrams have not been used in the analysis of health care systems. The closest case is in an analysis of biological systems by Roux-Rouquie, Caritey, Gaubert, and Rosenthal (2004), where the utility of UML state diagrams to describe and specify biological systems and processes was examined. Roux-Rouquie et al. (2004) mapped biological concepts to UML ones and presented state diagrams with states and substates of the active and inactive enzymes, and with concurrent substates at molecular and phenotype levels. Outside the UML notation, the original Statecharts notation that forms the basis of the UML state diagram notation has recently been used in the modelling and simulation of a cardiac surgical service (Sobolev et al., 2008; Vasilakis, Sobolev, Kuramoto, & Levy, 2007).

conclusIon It is apparent from this literature survey that UML has a role in the analysis of health care systems. There are clear benefits, especially in terms of clarity of communication and repeatability, if a standardised and rigorous notation is employed broadly. However, the application of UML to the modelling of health care systems is probably not as prevalent as in other application domains, at least as it is documented in the medical literature. Therefore, it is essential to conduct a thorough evaluation of the use and potential benefits of UML in a health care context if more wide spread application is to be recommended.

acKnowledgMent This work was partially supported by an award from the Strategic Promotion of Ageing Research Capacity (SPARC) initiative. The authors acknowledge the support they received from the Harrow School of Computer Science, University of Westminster, as well as the generous advice they received from Prof. Peter Millard, Prof. Peter Lansley, and Dr. Elia El-Darzi.

reFerences Aggarwal, V. (2002). The application of the unified modeling language in object-oriented analysis of healthcare information systems. Journal of Medical Systems, 26, 383-397. Ambler, S. (2004). The object primer: Agile modeling-driven development with UML 2.0. Cambridge: Cambridge University Press. Berwick, D.M. (2005). The John Eisenberg lecture: Health services research as a citizen in improvement. Health Services Research, 40, 317-336.

285

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Cruz, I.A., Gramaxo Silva, A., Soares, P., Oliveira, I., Serrano, J.A. V., & Paulo Cunha, J. (2002). Modelling the hospital into the future with UML. International Journal of Healthcare Technology and Management, 4, 193-204. Ganguly, P., & Ray, P. (2000). A methodology for the development of software agent based interoperable telemedicine systems: A teleelectrocardiography perspective. Telemedicine Journal, 6, 283-294. Goossen, W., Ozbolt, J., Coenen, A., Park, A., Mead, C., Ehnfors, M., et al. (2004). Development of a provisional domain model for the nursing process for use within the health level 7 reference information model. Journal of the American Medical Informatics Association: JAMIA, 11, 186-194. Harel, D. (1987). Statecharts: A visual formalism for complex systems. Science of Computer Programming, 8, 231-274. Hederman, L., Smutek, D., Wade, V., & Knape, T. (2002). Representing clinical guidelines in UMl: A comparative study. Studies in Health Technology and Informatics, 90, 471-477. Hoffman, K.J. (1997). Demystifying mental health information needs through integrated definition (IDEF) activity and data modeling. Journal of the American Medical Association, Symposium.l, 111-115. Hoo, K.S., Wong, S.T.C., Laxer, K.D., Knowlton, R.C., & Wan, C. (2000). Design patterns in medical imaging information systems. Proc. SPIE Medical Imaging, 3980, 256-263. Jun, G.C.T. (2007). Design for patient safety: A systematic evaluation of preocess modelling appraoches for healthcare system safety improvement, PhD Thesis. Cambridge: University of Cambridge, Department of Engineering.

286

Jun, G.C.T., Ward, J.R., & Clarkson, P.J. (2005). Mapping the healthcare process in order to design for patient safety. In International Conference on Engineering Design (ICED 05) (pp. 353-364). Kalli, S., Ambroso, C., Gregory, R., Heikela, A., Ilomaki, A., Leaning, M., et al. (1992). Inform: Conceptual modelling of intensive care information systems. International Journal of Clinical Monitoring and Computing, 9, 85-94. Kumarapeli, P., De Lusignan, S., Ellis, T., & Jones, B. (2007). Using Unified Modelling Language (UML) as a process-modelling technique for clinical-research process improvement. Medical Informatics and the Internet in Medicine, 32, 51-64. LeBozec, C., Jaulent, M.C., Zapletal, E., & Degoulet, P. (1998). Unified modeling language and design of a case-based retrieval system in medical imaging. In AMIA Annual Symposium Proceedings (pp. 887-891). Lee, N.J., Bakken, S., & John, R. (2006). Functional requirements specification and data modeling for a PDA-based decision support system for the screening and management of obesity. In AMIA Annual Symposium Proceedings (pp. 1002). Lunn, K., Sixsmith, A., Lindsay, A., & Vaarama, M. (2003). Traceability in requirements through process modelling, applied to social care applications. Information and Software Technology, 45, 1045-1052. Lyalin, D., & Williams, W. (2005). Modeling cancer registration processes with an enhanced activity diagram. Methods of Information in Medicine, 44, 11-13. Maij, E., Toussaint, P.J., Kalshoven, M., Poerschke, M., & Zwetsloot-Schonk, J.H. (2002). Use cases and DEMO: Aligning functional features of ICTinfrastructure to business processes. International Journal of Medical Informatics, 65, 179-191.

Application of Unified Modelling Language (UML) to the Modelling of Health Care Systems

Mehta, N.Y., Haluck, R.S., Frecker, M.I., & Snyder, A.J. (2002). Sequence and task analysis of instrument use in common laparoscopic procedures. Surgical Endoscopy, 16, 280-285. Object Management Group. (2005). Unified Modeling Language: Superstructure. Version 2.0. OMG Inc. Pohjonen, H., Bondestam, S., Karp, P., Kinnunen, J., Korhola, O., Sakari, P., et al. (1994). Quality assurance of the bedside chest radiography process. Acta Radiologica, 35, 235-243. Roux-Rouquie, M., Caritey, N., Gaubert, L., & Rosenthal, S. (2004). Using the Unified Modelling Language (UML) to guide the systemic description of biological processes and systems. Biosystems, 75, 3-14. Saboor, S., Ammenwerth, E., Wurz, M., & Chimiak-Opoka, J. (2005). MedFlow - improving modelling and assessment of clinical processes. Studies in Health Technology and Informatics, 116, 521-526. Sobolev, B., Harel, D., Vasilakis, C., & Levy, A.R. (2008). Using the statecharts paradigm for simulation of patient flow in surgical care. Health Care Management Science, 11, 79-86.

Spyrou, S., Bamidis, P., Pappas, K., & Maglaveras, N. (2005). Extending UML activity diagrams for workflow modelling with clinical documents in regional health information systems. In Proceedings of MIE2005 (pp. 1160-1165). Sutton, D.R., Taylor, P., & Earle, K. (2006). Evaluation of PROforma as a language for implementing medical guidelines in a practical context. BMC Medical Informatics and Decision Making, 6, 20. Vasilakis, C., Sobolev, B., Kuramoto, L., & Levy, A. R. (2007). A simulation study of scheduling clinic appointments in surgical care: Individualsurgeon vs. pooled lists. Journal of the Operational Research Society, 58, 202-211. Weber, R., Knaup, P., Knietig, T., Hauz, R., Merzweiler, A., Mludek, V., et al. (2001). Objectoriented business process analysis of the cooperative soft tissue sarcoma trial of the German Society for Paediatric Oncology and Haematology (GPOH). In Proceedings of the 10th World Congress on Medical Informatics (MEDINFO 2001) (pp. 58-62).

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 4, edited by J. Tan, pp. 39-52, copyright 2008 by IGI Publishing (an imprint of IGI Global).

287

288

Chapter 20

TreeWorks:

Advances in Scalable Decision Trees Paul Harper Cardiff University, UK Evandro Leite Jr. University of Southampton, UK

aBstract Decision trees are hierarchical, sequential classification structures that recursively partition the set of observations (data) and are used to represent rules underlying the observations. This article describes the development of TreeWorks, a tool that enhances existing decision tree theory and overcomes some of the common limitations such as scalability and the ability to handle large databases. We present a heuristic that allows TreeWorks to cope with observation sets that contain several distinct values of categorical data, as well as the ability to handle very large datasets by overcoming issues with computer main memory. Furthermore, our tool incorporates a number of useful features such as the ability to move data across terminal nodes, allowing for the construction of trees combining statistical accuracy with expert opinion. Finally, we discuss ways that decision trees can be combined with Operational Research health care models, for more effective and efficient planning and management of health care processes.

IntroductIon Since the second half of the 20th Century, an upsurge of electronic data has been taking place worldwide. Studies such as Frawley, PiatestskyShapiro, and Matheus (1991) show that the amount of data is doubling each year. Contributing factors for the data explosion include the widespread use of computer systems for nearly any commercial, financial, governmental, or research activity;

easier access to large storage capacity media; and advances in data collection tools. The Internet and all of its related services like the World Wide Web, e-mail, and online databases as a global information system have flooded humanity with a tremendous amount of data and information. With the continuous growth in size and complexity of information, there is an urgent need for a new generation of computational theories and tools to assist us in extracting useful information

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

TreeWorks

(knowledge) from the rapidly growing volumes of digital data. These theories and tools are the subject of the fields of data mining and knowledge discovery in databases (KDD). The need to understand data is extremely important for business, government, and research. Examples are numerous and varied and include applications in optimising market shares by increasing competitive advantage using knowledge extracted from sales transactions (Rygielski, Wang, & Yen, 2002), maximising customer retention (Edelstein, 1998), predicting the size of television audiences (Brachman, Khabaza, Kloesgen, Piatetsky-Shapiro, & Simoudis, 1996), and improving ovarian cancer detection (Li et al., 2004). Decision trees are one such data mining technique that learn from data and generate models containing explicit rule-like relationships among the variables. Decision tree algorithms begin with the entire training set of data, split into two or more subsets until the split size reaches an appropriate level. The entire modelling process can be visualised in a tree structure. This structure maps observations between dependent and independent variables. An arc between two nodes in the tree represents a partition of the parent node into child nodes. All observations follow a path from the root (initial node) and are assigned to a leaf (terminal node) based on splitting criteria (values of the independent variables). The two best-known and most widely used decision tree algorithms are Classification and Regression Trees (CART) and C4.5 (a successor of ID3). CART was developed by the statisticians Breiman, Friedman, Olshen, and Stone (1984), and C4.5 was developed by Quinlan, a computer scientist in the field of machine learning (Quinlan, 1993). The original methods to grow decision trees are not ideal for handling some of the features that are present in modern-day data sets such as categorical variables with many distinct values and the ability to handle extremely large datasets. In order to help overcome such issues, Catlett (1991) proposed sampling at each node of the classifica-

tion tree, but considers in his studies only datasets that could fit in main memory which were rather limited in size. Methods for partitioning the dataset such that each subset fits in main memory are considered by Chan and Stolfo (1993). Even though this method makes classification of large datasets possible, their studies show that the quality of the resulting decision tree is not as good as if the classifier had used all the available data. Most existing methods for automatic construction of classification trees utilise greedy heuristics, choosing locally optimal splits to divide the data at each level. Unfortunately these locally optimal values cannot be obtained quickly when a large dataset is analysed. In this article, we present TreeWorks, a CART tool that enhances existing decision tree theory and overcomes some of these common decision tree limitations. Whilst the primary focus of our research, as presented here, is on scalable decision trees, the resulting TreeWorks tool is highly practical and user-friendly and has already found considerable application by the UK National Health Service (NHS). The NHS handles millions of patient records each year and TreeWorks has assisted the NHS Information Centre with the redesign of Healthcare Resource Groups (HRGs). HRGs, which are similar to DRGs as used in the U.S., are standard groupings of clinically similar treatments which use common levels of healthcare resource, and are fundamental for standardising healthcare commissioning across the country as part of the UK Government’s policy of Payment by Results (PbR) (Department of Health, 2008). In this article, we also highlight ways in which decision trees can support Operational Researchers building health care models. A particular feature of health care processes is the inherent variation and uncertainty in treating individuals. Homogeneity leads to increased certainty in individual patient predictions (resource consumption, outcomes, pathways, etc.), which in turn results in the potential for more effective and efficient planning and management of health care processes.

289

TreeWorks

Thus decision trees can play a useful role here and the benefits of a combined data mining and modelling approach are illustrated. In the next section, a brief overview of decision trees (both for classification and regression) is provided together with methods for minimising impurity of trees. This is followed by a section on our approach for handling large datasets. A heuristic for handling categorical variables is then presented and evaluated, and some other features of TreeWorks are discussed. Finally we discuss the benefits of a combined decision treemodelling approach.

classification and regression trees Based on the type of the dependent variable, decision trees can be divided into classification trees (categorical dependent variable) and regression trees (continuous dependent variable). TreeWorks is a CART tool that is able to handle both classification and regression problems. CART works on the principle of binary splitting of data based on improving gain (or improving the purity of the tree into more homogeneous nodes). A number of purity measures are available both for classification and regression problems. These include information gain, gini index of diversity, gain ratio criterion, MaxDif, and Generalized Gini. The interested reader may consult Berzal et al. (2003) for a comprehensive review. Classification trees are suitable when the dependent variable is categorical, for example in predicting patient survival or outcome in the healthcare domain. The nodes are split on values of the independent variables by minimising purity measures for categorical data. Let s(i) be the size of node i, and suppose we have a measure for the impurity I(i) of node i, then the gain in purity made by splitting node i into nodes i0 and i1 is: Gain(i, i0,i1) = I(i) - (I(i0)p(i0)+I(i1)p(i1)

290

where p(i0) = s(i0)/s(i) and p(i1) = s(i1)/s(i) are the proportion of records assigned to node i0 and i1 respectively. The tree is grown recursively: at each step we choose which node to split and the variable to split on in order to maximise the gain, that is, minimise the impurity. We continue to do this until all nodes are pure enough, according to some stopping rule. Two examples of ways to evaluate the homogeneity of categorical data are the Gini index of diversity and the information entropy. Gini is based on squared probabilities of membership for each target category in the node. It reaches its minimum (zero) when all cases in the node fall into a single target category. Suppose that a variable y takes on values in 1, 2, . . . , m, and let f(i, j) be the frequency of value j in node i. That is, f(i, j) is the proportion of records assigned to node i for which y = j. m

I G (i ) = 1 - ∑ f (i, j ) 2 j =1

Information entropy is based on the concept of entropy as used in information theory, and takes the form: m

I E (i ) = -∑ f (i, j ) log f (i, j ) j =1

Whilst classification trees classify objects into discrete classes, regression trees are used when the class is continuous. A function y(x1, x2, . . . , xn) of n continuous or discrete attributes can be implemented. Values of the independent variables are split by a measure for continuous variable such as least square deviance (LS). Let y( j) be the value of y for record j and let be the mean value of y over node i, then:

IV

∑ (i ) =

j∈i

( y ( j ) - y (i)) 2 s (i ) - 1

TreeWorks

Other methods used to build regression trees include the least absolute deviation (LAD) which minimises the differences between the values and the median of the dependent variable. Studies carried out to compare LS and LAD regression trees, such as Torgo (1999), reveal that both models have different preference biases that can be considered useful depending on the application. LAD trees tend to produce predictions that, on average, are more accurate. However, these trees do commit large errors more often than LS trees.

Figure 1. Row referencing framework for handling data

Handling large datasets There exist a number of algorithms to construct decision trees. Most algorithms in the machine learning and statistics community are main memory algorithms, even though today’s databases are in general much larger than main memory. There have been several approaches to dealing with large databases, and the interested reader is referred to Gehrke, Ramakrishnan, and Ganti (2000) for a review of approaches. One particular approach is to group into similar subgroups each ordered attribute and run the algorithm on the grouped data. However all grouping methods for classification that take the class label into account assume that the database fits into main memory (Fayyad & Irani, 1993; Maass, 1994; Quinlan, 1993). We follow the same principle as described in Gehrke et al. (2000) that “the scalable versions of the algorithms produce exactly the same decision tree as if sufficient main memory were available to run the original algorithm on the complete database in main memory.” Therefore, below we describe our approach that does not sample the data for gaining speed but instead is a heuristic that uses all the available data for the split. The TreeWorks software utilizes a variation on the RainForest framework (Gehrke et al., 2000) that enables large blocks of data to be analysed. The basic concept revolves around proposing a model that does not implement the data storage and access methods on a low level, as these functions

are executed by a DBMS (Database Management System). In order to evaluate purity functions, the algorithm will examine notably smaller groups of data called the Attribute-Value-Class-set, or AVC-set, and Attribute-Value-Square-set, or AVS-set. The AVC-set and the AVS-set, which are respectively used for explaining categorical and numerical data, are extracted using SQL queries and are subsequently used to measure the value of the purity functions. The size of the AVC and AVS sets depends only upon the number of distinct values of the independent variable being tested, thus most of the complex and time-consuming process for accessing data is undertaken internally by the database management system. Unlike the RainForest framework, TreeWorks uses row references for the node’s data, linking to data rows available in all the sets rather than replicating a dataset for the node. The implemented variation of RainForest AVC (which is illustrated in Figure 1) is: SELECT ALL AllData.Xi, AllData.Y,

291

TreeWorks

COUNT(*) FROM AllData, WHERE AllData.Id = IndexesNode∂.Id and Fn GROUP BY AllData.Xi, AllData.Y Where: AllData is the table that represents all data available for training the model, IndexesNode∂ represents a node’s data table containing only one column that links to rows of AllData which should be the records that belongs to IndexesNode∂. In Figure 1 this can be any of root node, child1 or child2, and so forth. Y is the dependent variable. X1, X2, X3, and so forth, are the independent variables. Fn corresponds to the table where the node training data is stored. In our version, it is the join between the main database table that contains the training cases and the one column table which contains the symbolic links to the training case rows. Although this variation requires the joining of two tables, an aspect that takes some extra time to create the AVC set, it saves considerable time during the split step where only one column will need to be populated with data. TreeWorks utilises the PosgreSQL (Momjian, 2001) database management system. There were many reasons for this choice, such as PosgreSQL’s excellent performance and scalability, and crossplatform functionality with versions for both Unix and Windows computers enabling it to be used in a dedicated database server or on a user’s workstation.

292

a proposed Heuristic for classification problems Categorical splits have frequently presented a problem to classification trees due to the fact that categories cannot be ordered and the number of possible partitions to be tested explodes at the rate of 2m-1− 1, where m is the number of values belonging to a given independent variable, which is NP-hard. In our studies, using a PC Intel Core2 Duo @2.13GHz with 2GB of RAM, the time in seconds, s, to evaluate just one split with a categorical variable containing m distinct values and using the Gini purity function, was found to be s=0.00185e0.569m. This relationship is shown in Figure 2. With just 28 distinct values, this would take over 4 hours just to explore one possible split. With many categorical variables to consider at each level in the tree, clearly this could be very time-consuming. With 30 distinct values it would take approximately 13 hours for the full-search and with 35 more than 5,000 days. Yet in our review of commercially available CART packages, all appeared to use the full-search algorithm, with some well-know tools complaining of memoryspace limitations and others simply crashing. We propose a heuristic that is straightforward to implement and in tests has shown to be sufficiently accurate. The heuristic is based on the notion that the inclusion of values will either increase or reduce the purity of a split, thus we can recursively select a value of the independent variable to be analysed as a split candidate. We then include all of the rows with this value in the left node and compute the node purity, then remove these row from the left and try adding them to the right node and recompute the purity, finally choosing the side that has the better gain based on the chosen split criteria. This procedure is divided into two parts. Part (1) is a set of steps executed on every occasion and is where the majority of the improvement occurs, and part (2) contains optional steps that can be executed in order to try and achieve further improvement.

TreeWorks

Figure 2. Relationship between number of distinct categorical values and CPU time

The algorithm for a parent node p to be split into a left node (left), and a right node (right), by using the independent variable i, is described bellow:

part (2) i.

Execute all the steps of part one.

part (1)

ii.

For every element in left test if the gain will improve if it is moved to right.

Sort the values of i contained in p by its lexicographical order. ii. Assign the data cases that contain first value of i to left. iii. Assign the data cases that contain the second value of i to right. iv. From the third to last values, try putting the values in left or right and assign the value to the side that minimises the impurity of the split. v. Three last combinations need to be tested: a. Remove the first value from left and put in right. b. Remove the second value from right and send to left. c. Put back the first value to left. vi. Use the smallest impurity combination from the steps iv) or v). End of part one i.

For every element in right test if the gain will improve if it is moved to left. iv. Use that combination of values and impurity value for the split. End of part two iii.

evaluation of the Heuristic Once again using a PC Intel Core2 Duo @2.13GHz with 2GB of RAM, the time in seconds, s, to evaluate just one split with a categorical variable containing m distinct values using the heuristic was found to be s=0.000000195m2.8+0.18. Even with m = 4,000 the split takes less than 40 minutes. One clear advantage of the heuristic search algorithm is that it permits the construction of all trees including some categorical variables that would otherwise have not been possible using the full-search with 2m-1-1 combinations. In our tests

293

TreeWorks

on various size datasets, TreeWorks has never run out of memory and has always been able to produce trees in reasonable time even for datasets with categorical variables containing many thousands of distinct values. In TreeWorks, the user defines the threshold value of m, above which the heuristic will be implemented. Categorical variables with distinct values below this threshold will be split using the full-search algorithm. We have compared the accuracy of the fullsearch and heuristic partial-search algorithms. The heuristic performed extremely well on all tests with only a few tests where there were small differences in gain between the two approaches. Three such experiments are detailed below. For experiments 1 and 2, we use a large datasets supplied by the NHS Information Centre as part of the redesign of HRGs. The dataset contains over 100,000 records for hospital inpatient stays in the UK and more than 50 variables. The dependent variables of interest include patient length of stay

and costs. Independent variables include such fields as patient source (emergency or elective), NHS Trust (or hospital), current version of the HRG, age, sex and procedure code (operation type) where applicable. Due to sensitivity of the data, we are unable here to provide explicit names of variables in our results. For experiment 3, we use datasets that are freely available from UCI Machine Learning Repository, School of Information and Computer Science, University of California (Asuncion & Newman, 2007)

experiment 1 We created 10 data samples, with each sample containing 20% of the overall dataset with data randomly sampled without reposition. We tested all 10 datasets using both the full-search and our developed heuristic, and recorded the gain for one split using 6 different categorical independent variables. Each chosen categorical

Table 1(a). Gain values obtained from the full-search. Variable

2

3

4

5

6

7

8

9

10

A

35.02

Set 1

37.28

34.67

35.14

33.88

37.96

35.07

35.68

37.42

37.40

B

8.353

6.372

5.683

7.096

5.385

7.545

8.243

8.603

7.281

7.771

C

7.945

6.431

5.559

6.636

4.909

7.959

8.141

7.969

8.101

7.078

D

86.27

84.98

84.98

87.18

84.20

85.46

86.74

85.23

86.79

84.88

E

35.12

37.84

34.71

35.38

34.00

38.45

35.14

36.18

37.50

37.68

F

0.920

1.140

1.529

1.553

0.813

1.082

0.913

2.395

1.673

1.159

Table 1(b). Gain values obtained from the heuristic search. †denotes a difference in gain compared to Table 1(a). Variable

294

Set 1

2

3

4

5

6

7

8

9

10

A

35.02

37.28

34.67

35.14

33.88

37.96

35.07

35.68

37.42

37.40

B

8.353

6.372

5.683

7.096

5.385

†7.478

8.243

8.603

7.281

†7.613

C

7.945

6.431

5.559

6.636

4.909

7.959

8.141

7.969

8.101

7.078

D

86.27

84.98

84.98

87.18

84.20

85.46

86.74

85.23

86.79

84.88

E

35.12

37.84

34.71

35.38

34.00

38.45

35.14

36.18

37.50

37.68

F

0.920

1.140

1.529

†1.531

0.813

1.082

0.913

2.395

1.673

1.159

TreeWorks

variable contained different numbers of distinct values but in this first experiment were chosen to be of reasonable size (less than 30) such that the full-search was possible. In Table 1, sections (a) and (b) show results from the full-search and heuristic (implemented with both part (1) and (2) of the algorithm) respectively. There were only three occasions where the gain values differed, marked by † in (b) of Table 1. Overall the heuristic achieved the optimal gain in 57 of the 60 tests (95% of the time) and with an overall gain accuracy, measured using the absolute difference in gain values, of 99.8%. Chi-squared tests showed that there was no evidence to reject the hypothesis that the gains were the same for the full and partial searches.

experiment 2 The second experiment evaluates the performance of the heuristic, with a focus on both accuracy and CPU time, for a varying number of data cases and distinct values for both categorical dependent and

independent variables. presents results from the comparison and shows CPU time and gain. We have used three different NHS datasets covering intensive care, HRG data, and inpatient admission data. Fifteen tests and 30 comparisons across the three different datasets have been performed. For some tests, such as with those with 3,754 distinct values of the independent variable, it was not possible to evaluate the gain for the full-search since the CPU time would have been in the order of days, months, or even years, and in such cases we can only estimate what the CPU might have been based on the exponential relationship as described earlier. In summarising the performance of the heuristic, there are clear benefits in CPU time with the partial-search at most taking 1.2 hours when handling in excess of 115,000 cases and 3,754 distinct values of the independent variable (test N). With 4 and 37 distinct values of the dependent and independent variables respectively for 115,000 cases, TreeWorks takes less than one second whereas a full search is estimated to take

Table 2. Comparison of CPU time between full-search and heuristic for experiment 2. (e) indicates an estimated CPU time and (†) denotes a difference in gain between the full-search and heuristic. Number of distinct values

Full-search

Test

Dependent Variable

Independent Variable

Cases

A B C D E F G H I J K L M N O

2 2 2 2 3 3 3 3 7 37 37 4 37 37 4

23 19 4 7 20 23 26 28 20 4 4 37 563 3754 3754

1332 768 564 602 2582 2582 2582 2582 2582 80815 115451 115451 115451 115451 115451

Heuristic

CPU Time

Gain

CPU Time

Gain

253.906 11.828 0.016 0.016 50.384 422.528 3984 11729 90.7 0.156 0.294 10808496 (e) 2.3E+166 (e) 8.7E+1038 (e) 3.2E+1038 (e)

6.266 2.715 0.706 0.524 6.238 5.996 7.026 9.815 9.019 19.108 19.12 N/A N/A N/A N/A

0.016 0.016 0.016 0.015 0.017 0.018 0.019 0.320 0.017 0.158 0.201 0.212 42.723 4305.9 3758.8

6.266 2.666 (†) 0.706 0.524 6.238 5.996 7.026 9.815 9.019 19.108 19.12 22.565 20.533 20.536 30.319

295

TreeWorks

125 days (Test L). For those tests that that could be compared, in only one (test B) was the heuristic not optimal with a discrepancy of less than 2%. The overall accuracy of the heuristic was greater than 99.95%. It is also worthy of note that some of the more significant and valuable split gains in the region of 20% to 30% were achieved in tests L, M, N, and O, which would have never been possible with the full-search.

experiment 3 The third set of experiments makes use of available extensive datasets from UCI Machine Learning Repository, University of California (Asuncion & Newman, 2007). We repeat the same testing structure as experiment 2, focusing on both accuracy and CPU time. Results are given in section

a of Table 3. Tests 1 – 4 use the Adult Data Set (http://archive.ics.uci.edu/ml/datasets/Adult) and tests 5 – 9 the Census-Income (KDD) Data Set (http://archive.ics.uci.edu/ml/datasets/CensusIncome+%28KDD%29). In all tests, TreeWorks, is used to predict whether an individual’s income exceeds $50,000 per annum. Varying number of independent variables and cases are included across the range of tests we have performed and are shown in section b of Table 3. The KDD dataset contains 299,285 instances (records) and for tests 5 through 8, we sample 5% of records to allow a quicker comparison between the heuristic and the full search but varied the selected independent variable and hence associated number of distinct values (from 9 to 42). However in test 9, we use all 299,285 instances. A full search gain was possible in 8 of the 9 tests. For each of these 8 tests, the heuristic

Table 3a. Comparison of CPU time between full-search and heuristic for experiment 3. Number of distinct values

Full-search

Test

Dependent Variable

Independent Variable

Cases

1 2 3 4 5 6 7 8 9

2 2 2 2 2 2 2 2 2

9 16 15 42 17 24 15 28 24

48842 48842 48842 48842 14964 14964 14964 14964 299285

Heuristic

CPU Time

Gain

CPU Time

Gain

0.092 1.429 0.778 Not possible 3.631 648.845 1.3 8877.493 677.945

2.101 10.704 9.479 N/A 11.92 6.295 12.821 4.632 6.438

0.075 0.077 0.079 0.093 0.557 0.522 0.553 0.595 1.055

2.101 10.704 9.479 0.764 11.92 6.295 12.821 4.632 6.438

Table 3b. Summary of independent variables used in experiment 3.

296

Test

Dataset

Independent Variable

1 2 3 4 5 6 7 8 9

Adult Adult Adult Adult Census-income Census-income Census-income Census-income Census-income

workclass education occupation native_country education-detailed major_industry_code major_occupation_code household_and_family_stats major_industry_code

TreeWorks

achieved the same gain as the full search. In test 9, which contained the full 299,285 instances of the census-income data, TreeWorks computed the same gain as the full search in only 1 second compared to over 11 minutes for the full search. We conclude that given the accuracy and CPU time across a range of different datasets and experiments, both health care and nonhealth care related, that TreeWorks is sufficiently accurate and responsive for real-time business applications handling realistic large sized databases.

additional treework Features In addition to the ability to handle large databases and many categorical values, TreeWorks incorporates a number of other features, some of which are mentioned here. Many of these features were based on requests from users of TreeWorks, such as those from staff at the NHS Information Centre, which further enhance the tool and permit it to be both powerful and highly user-friendly and interactive. Features include: •

• • • • •

•

Provision of tree model validation (training and testing datasets) with user-chosen percentage for training set. Possibility of moving data across terminal nodes. Description of the node paths. Visualisation of the data at each node with standard database functionality. Support of missing values in the data. Allowing the user to interact with the model, choosing the split variable, split values and growing and pruning sub-trees. User-friendly wizards guiding the user through the data handling processes.

The interactive nature of the tool was particularly important in the redesign of HRGs, since it was a preferred NHS approach to build classification and regression trees interactively with teams of expert consultants and medical

personnel. The resulting tool allows for new HRG groups to be agreed using a combination of the CART statistical algorithm combined with expert opinion. For example, it was often the case that a purely statistical tree with optimal gain (reduction in variance) did not make complete clinical sense, and subsequently could not be considered in its entirety as an HRG. Consequently, it was necessary to move some surgical procedures and patient clusters from one node to another. TreeWorks allows for this to be conducted in real-time and reports on the consequence on the reduction in variance before the data is moved. An illustrative screen-shot is shown in Figure 3. The move data form allows the user to create a series of SQLtype commands to select the data to move. For example, in Figure 3, data for emergency patients aged 16 and under are to be moved from node 1 to 2. The user can preview the selected data rows and reduction in gain prior to confirming the move. Other well-known software does not permit for this level of interaction. Model validation is also important, and TreeWorks allows the user to select the percentage split between the training and testing datasets. The two samples are randomly selected using pseudorandom numbers based on Knuth’s subtractive random number generator algorithm (Knuth 1981). Figure 4 illustrates TreeWork’s model validation. A summary of each node (with terminal nodes highlighted) provides statistics such as number of cases, average value and standard deviation. The final columns show whether the node has passed or failed a t-test which is used to compare the training and testing models for each node.

decision trees for Health care Modelling Operational Research methods are widely used for health care modeling and have found considerable application (Brandeau, Sainfort, & Pierskalla, 2004; Ozcan, 2005). Health care modelling, however, is beset with many challenges (Harper

297

TreeWorks

Figure 3. Moving data between nodes

Figure 4. Terminal node model validation

& Pitt, 2004). A particular feature is the inherent variation and uncertainty in treating individuals. For example, length of stay in hospital or the infectious period for a given disease typically varies from patient to patient. From both a clinical and operational perspective, it is desirable to be able to understand and capture this variability (Harper, 298

2005). Homogeneity leads to increased certainty in individual patient predictions (resource consumption, outcomes, pathways, etc.), which in turn results in the potential for more effective and efficient planning and management of health care processes.

TreeWorks

In designing and building health care models, there are a number of approaches when considering how to capture patient variability: •

•

•

•

Ignore variability: build deterministic models. Essentially here we have one patient group (all the available data) and are using average values. Resample all individuals. In this model we recreate every observed individual to exactly recreate real-life. This is time-consuming and still lacks the ability to provide insight for future predictions, such as case-mix or demand for services. Build a stochastic model with one “generic” patient group. In this model we define one distribution for each parameter in the model. We use all the available date to define each distribution, thus we are sampling individuals from the entire possible range of (observed) values. Create patient groups. Each patient group will have their own set of parameters, distributions, care-pathways, and so forth.

The benefit of the last approach is that we are able to construct clinically and/or statistically meaningful patient groups that we can then use as patient groups in developed models in order to capture variability. As we create more groups, naturally we capture more of the variability and increase information content. However, what typically happens is that we reach a point when creating more patient groups does not lead to a further significant capture of variability or increased understanding. This is similar to Pareto’s principle (80-20 rule). If we pursue the patient grouping approach, then we need to know how many groups to create and group definitions (e.g., for hospital length of stay we might create groups using indictors such as age, sex, elective or emergency, speciality, etc.). We suggest therefore that a combined data mining and modelling approach is beneficial. First

we construct decision trees. The chosen dependent variable would be of relevance to the nature of the patient-based model, such as dwelling time in a particular state of the patient pathway or probability of transition from one state to another. All patients will be assigned to one of the terminal nodes in the decision tree. Each terminal node becomes a unique patient group in the model. Here we take the model to represent the individual patient pathways, such as movements through a health service provider or transitions through a natural history of a disease. Each individual patient that enters the model will belong to a patient group. Dwelling times and other parameters in the model will be taken from the group of which the patient is a member. To capture any dynamic effects, we may decide to create multiple decision trees for different parts of the model, and re-assign patients to groups as appropriate A combined data mining and modelling approach has been adopted for various studies by the authors. These include hospital capacities (Harper, 2002), intensive care (Costa, Ridely, Shahani, Harper, de Senna, & Nielsen, 2003), diabetic retinopathy (Harper, Sayyad, de Senna, Shahani, Yajnik, & Shelgika, 2003) and screening for chlamydia (Evenden, Harper, Brailsford, & Harindra, 2005). To illustrate the concept, for modelling hospital resource capacities, Harper (2002), developed a discrete event simulation to capture individual patient pathways through hospital and monitor corresponding resource needs. The challenge was to adequately handle the variability such as length of stay, operating times and workforce needs. Decision trees were constructed to define patient groups and fit distributions for various parameters in the model. We were able to work with hospital clinicians and managers to create groups that were both statistically and clinically meaningful. For example, in one hospital it was possible to categorize all in-patients into 15 patient groups that were then fed into the simulation model. Hospital managers could then change any of the parameter values

299

TreeWorks

for any of the 15 groups for scenario analysis, such as a reduction in length of stay or change in workforce needs by skill-mix of staff for that patient group. Future research will examine a framework for formally interfacing TreeWorks with a developed simulation shell to facilitate the combined data mining and operational research modelling approach.

conclusIon This article presents TreeWorks, a classification and regression tree tool that overcomes some of the common limitations of decision tree software. We suggest a heuristic for allowing decision trees to overcome issues of scalability and handle observation sets that contains several distinct values of categorical data. Experimentation has demonstrated that the heuristic nearly always achieves the optimal gain and for practical purposes is sufficiently accurate. The distinct advantage is the CPU time and the ability to produce models in reasonable times even for situations with thousands of distinct values of categorical data. Indeed some of the best gains (reductions in variance) were found in splits which would not be possible to consider using the standard full-search approach. All the commercial software that we have reviewed implements the full-search and so are severely limited. With the continuous growth in size and complexity of information, there is an urgent need for a new generation of tools to assist in extracting knowledge from the rapidly growing volumes of digital data. We believe that TreeWorks represent a new chapter in enhancing the scalability of decision trees, and the tool has already found considerable application in handling millions of healthcare records by the UK National Health Service. Further features of our tool increase the ease of use and functionality, such as the ability to combine the CART algorithm with expert opinion by moving data across terminal nodes.

300

There are a number of future TreeWork enhancements proposed. These include the ability to produce hybrid trees that automatically select the best splitting function (for example from Gini index of diversity or information entropy) at each level to create the overall best tree. Also we wish to explore the incorporation of “fuzzy” splitting rules. Rule-based trees, with “crisp” splitting criteria such as CART, can occasionally seem overly harsh. In reality, individual membership could well be in different terminal nodes and each new case may be assigned node membership probabilities. Finally, future research will also examine a framework for interfacing TreeWorks with a developed simulation modelling shell to facilitate a combined decision tree and operational research modelling approach.

reFerences Asuncion, A., & Newman, D.J. (2007). UCI machine learning repository. Irvine, CA: University of California, School of Information and Computer Science. Retrieved May 26, 2008, from http:// www.ics.uci.edu/~mlearn/MLRepository.html Berzal, F., Juan-Carlos, C., Cuenca, F., & MartínBautista, M.J. (2008). On the quest for easy-tounderstand splitting rules. Data and Knowledge Engineering, 44(1), 31-48. Brachman, R.J., Khabaza, T., Kloesgen, W., Piatetsky-Shapiro, G., & Simoudis, E. (1996). Mining business databases. Communications of the Association for Computing Machinery, 39(11), 42-48. Brandeau, M.L., Sainfort, F., & Pierskalla, W.P. (2004). Operations research and health care. Boston, MA: Kluwer. Breiman, L., Friedman, J., Olshen, R., & Stone, C. (1984). Classification and regression trees. Belmont, CA: Wadsworth International Group.

TreeWorks

Catlett, J. (1991). Megainduction: Machine learning on very large databases. Unpublished doctoral dissertation, Basser Department of Computer Science, University of Sydney. Chan, P.K., & Stolfo, S.J. (1993). Meta-learning for multistrategy and parallel learning. In Proceedings of the Second International Workshop on Multistrategy Learning. Costa, A.X., Ridely, S.A., Shahani, A.K., Harper, P.R., de Senna, V., & Nielsen, M.S. (2003). Mathematical modelling and simulation for planning critical care capacities. Anaesthesia, 58, 320-327. Department of Health. (2008). Payment by results (PbR). Retrieved May 26, 2008, from www.dh.gov. uk/en/Policyandguidance/Organisationpolicy/ Financeandplanning/NHSFinancialReforms/ index.htm. Edelstein, H. (1998). Data mining: Myth and reality. GIGA Information Group Telepresentation. Evenden, D., Harper, P.R., Brailsford, S.C., & Harindra, V. (2005). Improving the cost-effectiveness of chlamydia screening with targeted screening strategies. Journal of the Operational Research Society, 57(12), 1400-1412. Fayyad, U.M., & Irani, K. (1993). Multi-interval discretization of continuous-valued attributes for classification learning. In Proceedings of the International Joint Conferences on Artificial Intelligence 1993, San Francisco, CA. Frawley, Piatestsky-Shapiro, & Matheus (1991). Knowledge discovery in databases: An overview. Menlo Park, CA: AAAI Press. Gehrke, J., Ramakrishnan, R., & Ganti, V. (2000). RainForest - a framework for fast decision tree construction of large datasets. Data Mining and Knowledge Discovery, 4(2-3), 127-162.

Harper, P.R. (2002). A framework for operational modelling of hospital resources. Health Care Management Science, 5(3), 165-173. Harper, P.R. (2005). A review and comparison of classification algorithms for medical decision making. Health Policy, 71, 315-331. Harper, P.R., & Pitt, M. (2004). On the challenges of health care modelling and a proposed project life-cycle for successful implementation. Journal of the Operational Research Society, 55, 657-661. Harper, P.R., Sayyad, M.G., de Senna, V., Shahani, A.K., Yajnik, C.S., & Shelgika, K.M. (2003). A systems modelling approach for the prevention and treatment of diabetic retinopathy. European Journal of Operational Research, 150, 81-91. Li, Tang, Wu, Gong, Gruidl, Zou, et al. (2004). Data mining techniques for cancer detection using serum proteomic profiling. Artificial Intelligence in Medicine, 32(2), 71-83. Maass, W. (1994). Efficient agnostic pac-learning with simple hypothesis. Momjian, B. (2001). PostgreSQL: Introduction and concepts. Addison-Wesley Longman Publishing Co., Inc. Ozcan, Y.A. (2005). Quantitative methods in health care management. San Francisco, CA: Jossey-Bass (Wiley). Quinlan, J.R. (1993). C4.5: Programs for machine learning. San Mateu: Morgan Kaufmann series in machine learning. Rygielski, C., Wang, J.-C., & Yen, D.C. (2002). Data mining techniques for customer relationship management. Technology in Society, 24, 483-502.

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 4, edited by J. Tan, pp. 53-68, copyright 2008 by IGI Publishing (an imprint of IGI Global).

301

302

Chapter 21

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor Hussein Atoui Université de Lyon and INSERM, France David Télisson Université de Lyon and INSERM, France Jocelyne Fayn Université de Lyon and INSERM, France Paul Rubel Université de Lyon and INSERM, France

aBstract Recent years have witnessed a growing interest in developing personalized and nonhospital based care systems to improve the management of cardiac care. The EPI-MEDICS project has designed an intelligent, portable Personal ECG Monitor (PEM) embedding an advanced decision making system. We present two of the ambient intelligence models embedded in the PEM: the neural-network based ischemia detection module and the Bayesian-network risk stratification module. Ischemia detection was expanded to take into account the patient ECG, clinical data, and medical history. The neural-network ECG interpretation module and the Bayesian-network risk factors module collaborate through a fuzzylogic-based layer. We also present two telemedicine solutions that we have designed and in which the PEM is integrated. The first telemedical architecture was created to allow the collection of medical data and their transmission between healthcare providers to get an expert opinion. The second one is intended for improving healthcare in old people’s homes.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

IntroductIon Recent years have witnessed a growing interest in developing personalized and nonhospital based care systems to improve the management of cardiac care (Axisa, Schmitt, Gehin, Delhomme, McAdams, & Dittmar, 2005; Campbell et al., 2005; Kerkenbush & Lasome, 2003). The reason behind such interest is due to the fact that cardiovascular diseases now represent the leading cause of mortality in Europe and reducing the time before hospitalization is crucial to reducing cardiac morbidity and mortality (McMurray & Rankin, 1994; Task Force Report, 1998). Event recorders and transtelephonic ECG recorders are thus increasingly used to improve decision making in the prehospital phase. However, such systems are usually unable to capture transient ECG events such as infrequent arrhythmias or ischemic episodes. In addition, all these systems require setting up new information technology infrastructures and medical services and need skilled personnel to interpret the ECG and make decisions for the patient care.

This approach has thus proved to be very impractical for patients with infrequent symptoms such as arrhythmias and ischemia that represent 85% of the cardiac diseased patients, and would be very expensive if adopted for every citizen at risk. The European project EPI-MEDICS has designed a solution based on the interpretation of ECG derived cardiological syndromes and developed a friendly and easy-to-use, cost-effective intelligent personal ECG monitor (PEM) (Rubel et al, 2004; Rubel et al., 2005). The device (Figure 1) is capable of recording a simplified 4-electrode, professional quality 3-lead ECG, to derive the missing 5 leads (V1, V3… V6) of the standard 12-lead ECG (Atoui, Fayn, & Rubel, 2004), to store the derived 12-lead ECG according to the SCP-ECG standard (EN 1064, 2007), to analyze and interpret the recorded ECG, to detect arrhythmias and ischemia or acute myocardial infarction, and to send an alarm message to the appropriate health care providers. To develop the PEM software platform, we were confronted to a variety of problems related to the system intelligence and functioning: from recording and storing the ECG according to the

Figure 1. The personal ECG monitor (PEM) device allows for early detection of arrhythmia and ischemia in the pre-hospital phase and thus for better and more adapted treatment

303

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

SCP-ECG standard, to analyzing the recorded ECG and combining the result of interpretation with the patient’s risk factors in a unique risk score, to finally triggering and managing an appropriate alarm message according to the calculated risk score. The objective of this article is to present an overview of the current intelligence aspects embedded in the PEM prototypes in terms of services and scenarios and to describe some of the latest ambient intelligence and pervasive solutions that have been designed to: •

•

allow the PEM user to record a good quality ECG without any professional assistance using a reduced electrode set in a simplified positioning and to derive the missing leads to generate an artificial standard 12-lead ECG as accurate and as similar as possible to the original 12-lead ECG. adapt the decision making process within the system to the specificity of the user by using advanced neural network-based decisionmaking methods, taking into account the serial ECG measurements and the patient risk factors and clinical data.

Subsequently, we present a prospect of the deployment scheme of these solutions in terms of modules and their integration within the PEM telemedical platform and tele-expertise applications.

eMBedded IntellIgence In an e-cardIology perVasIVe MultI-actor enVIronMent The EPI-MEDICS project developed an intelligent pervasive PEM for the early detection of patients and citizens at cardiac risk. Decision making embedded in the PEM is performed at 4 different levels: detection, diagnosis, generation of alarms, and intelligent management of the alarm messages

304

for handling any communication problem with the contacted health care providers. Arrhythmia detection and ischemia diagnosis are based on the comparison of the last recorded ECG with a reference ECG. Arrhythmia detection is rule based. For the diagnosis of ischemia, a multi-expert decision making module based on artificial neural network committees has been designed (Rubel et al., 2004). Different levels of alarms are forwarded to the relevant health care providers depending on the severity of the cardiac event (Rubel et al., 2005). Major alarms (acute ischemia/infarction, severe arrhythmia) are automatically transmitted to the nearest emergency call center. In case of a medium alarm level (suspicion of ischemia and/ or atypical arrhythmia), all information is sent to and temporarily stored on an alarm server that automatically sends an SMS to the attending health professional (cardiologist or general practitioner) stored in the patient’s contact list of the PEM Card (Figure 2). In case of a minor alarm (small ECG changes), the PEM displays a short message inviting the user to report about the message at the occasion of one of his next visits to his cardiologist or attending physician.

a Multiclassifier Multimodule decision Making The use of a neural network-based classifier to analyze the recorded ECG and to trigger an appropriate alarm according to risk thresholds (major, medium, minor, or no alarm thresholds) allows an optimal discrimination between the different alarm levels by simply moving upward or downward these thresholds. However, there is a major interest in reducing the rate of false-positives while maintaining the highest possible sensitivity and specificity in order to reduce the rate of false alarms that could generate expensive and unnecessary costs. One possible way to enhance the diagnostic accuracy of the PEM device consists of including a patient’s

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

Figure 2. The EPI-MEDICS medium alarm level tele-expertise architecture (from Rubel et al., 2005)

risk stratification, based on the patient medical history stored in an electronic health record in the PEM smartcard, in the decision-making process. Consequently, merging the ECG interpretation with the user (patient/citizen) cluster of risk factors can be of a great benefit since it would help to better quantify the patient’s risk for developing cardiac complications and thus to recognize whether the clinical situation is changing and turning into a cardiac disease that could quickly turn a healthy citizen into a patient at high risk of major cardiac events or of even sudden death because of life-threatening arrhythmias, cardiac ischemia or myocardial infarction.

risk Factors Quantification using Bayesian networks For risk factors quantification, to merge quantitative and qualitative data into one decision process, we have designed a solution based on Bayesian networks to calculate the probability of the event

(i.e., infarction) knowing the symptoms or signs (cholesterol rate, age, and so forth). The choice of using Bayesian networks to calculate a risk score upon the patient’s risk factors is justified by the capacity of Bayesian networks both to model the uncertainty inherent in medical reasoning and to make decisions based on incomplete data (Lagor, Aronsky, Fiszman, & Haug, 2001). We configured and assessed the accuracy of the Bayesian network over the INDANA database which is a very large collection of individualpatient data with at least a six year follow-up for cardiovascular events and deaths: myocardial infarction, stroke, or cardiovascular death (Gueyffier et al., 1995). The Bayesian network uses the patient data (age, gender, diabetes, and so forth) to predict the risk of cardiovascular event (Figure 3). The assessment over the INDANA database of the Bayesian network in comparison to logistic regression and discriminant analysis has shown a clear advantage of more than 5% in terms of Area Under the Roc Curve (AUC) when compared with

305

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

these two conventional statistical methods (Atoui, Fayn, Gueyffier, & Rubel, 2006).

•

collaboration scheme

•

The collaboration between the ECG interpretation module and the risk factors module has to be smooth to ensure a more accurate outcome. For this reason, a fuzzy-logic-based layer to control the dialogue between both modules can be helpful in a flexible decision-support system as the PEM. The choice of fuzzy-logic is due to the fact that such a technology allows an efficient collaboration between classifiers and thus enhances the global decision-making process. The overall decision scenario operates as follows (Figure 4): •

A patient, suffering from fatigue or any other common symptoms known to precede a possible cardiac event, uses the PEM device to acquire an ECG.

•

Once successfully recorded and stored in the PEM smartcard, the ECG is interpreted by the neural-network module. If the neural-network module output passes the high risk threshold, the ECG is considered as abnormal, and the system will trigger a major alarm sequence. If the output falls within the medium or minor risk interval, the ECG is considered as suspect, and the ECG classifier output is combined with a second output issued by the Bayesian module to produce a global risk score that will in turn trigger the appropriate alarm level: major, medium, minor.

The determination of the different risk thresholds and of the fuzzy logic rules are for the moment being based on human expert considerations about the desired sensitivities and specificities that can be deduced from the different Receiver Operating Characteristic (ROC) curves. Indeed,

Figure 3. Schematic representation of the bayesian network used to estimate the risk of cardiovascular outcome (from Atoui et al., 2006)

306

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

Figure 4. Schematic representation of the overall decision making platform that will be embedded in the final PEM devices for detecting ischemia after adding the bayesian and the fuzzy logic modules in the decision-making process

the development of an optimized fuzzy-logic interaction layer and the validation of the global decision making process would require a very large database that contains not only serial 12lead and PEM 3-leads ECGs but also exhaustive electronic health records of the patients of the database. Such a PEM-specific database is currently under construction through clinical trials conducted both by health professionals and by the patients in emergency centers, coronary care units, cardiology clinics, offices of cardiologists and GPs, home care, and ambulatory care situations.

exaMples oF peM IntegratIon In teleMedIcal applIcatIons tele-expertise architecture adapted to e-cardiology In medical applications, there are numerous situations in addition to the selfcare scenarios

developed during the EPI-MEDICS project, where healthcare providers might request an expert opinion about a patient’s health record in almost real time, no matter the degree of intelligence that has been embedded in the ECG recording devices. One typical example is a GP who wants to have an expert advice from a cardiologist for the ECG of his patient. Another example is a patient who is admitted in an emergency department of a general hospital. The interpretation of the ECG tracing may be not clear for the nonexpert medical staff. A solution is to send the ECG to a cardiologist at the nearest cardiology hospital for an expert opinion. To allow these transactions between the requester (GP, emergency doctor, paramedics, etc.) and the recipient (emergency call center, competence center, expert), a client-server type architecture has been designed (Figure 5) (Télisson, Fayn, & Rubel, 2004). This architecture is based on TCP/IP Internet technology and on the XML metalanguage for data representation, storage, and communication. It is composed of

307

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

Figure 5. The tele-expertise architecture. A web server provides intelligent management of messages exchanges and a set of distributed client applications for sending and receiving XML tele-expertise requests via HTTPS (from Télisson et al., 2004)

a main application installed on a Web server that provides an intelligent management of the messages exchanges and a set of distributed client applications for sending and receiving the tele-expertise requests via HTTPS (HTTP with a Secure Socket Layer). The requests are represented in the XML meta-\language. A message contains information about the communicating parties, the Electronic Health Record (EHR), and any attached file like SCP-ECGs. The architecture is designed to support the pervasive paradigm. In order to guarantee the reception of the message in due time, the expert should be able to receive anywhere different kinds of notifications by means of different types of devices that depend on the available technologies (Pager, SMS, PDA, and so

308

forth). The telerequester may use a standard digital ECG recorder, a PEM, or any other personal ECG monitor with Bluetooth and GPRS transmission capabilities. This infrastructure has been experimented within the Lyon area in a setting that includes an emergency department, regional and general hospitals, the Cardiology Hospital of Lyon, patients homes, and the informatics department of the Lyons hospitals which hosts the servers. To facilitate the recording of digital ECGs and their automatic transmission, in several situations we have used the EPI-MEDICS Personal ECG Monitor (PEM). Preliminary evaluation results have shown the potential of such an architecture to improve patient care by allowing to complete

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

the global medical patient record and to guarantee the traceability of the exchanges.

shared care telemedical solution dedicated to elderly patient nursing services: the tele-nurse project In modern societies, the number of elderly populations is steadily growing and the mean age at which the retirees are entering elderly nursing homes is also steadily increasing. The paramedical staff is thus increasingly confronted with medical requests that usually result, because of the absence of medical in-house support, in sending the patient to the hospital for security reasons. It is thus felt that the development of a medical trolley equipped with several medical instruments and video-telephony allowing remote decision making can reduce the number of hospitalizations. The goal of the Tele-Nurse project (Télisson, Fayn, Placide, Rubel, & Comet, 2006) is to set up a remote decision support system connecting the elderly patients’ rooms or apartments in the nursing homes to a PC based, easy to use and to deploy clinical workstation enabling the remote GP to provide quicker and more effective actions or treatments in case of disease episodes leading to emergency situations such as a heart attack. The medical trolley allows data transmission and information exchange, via an intelligent server, between the doctor and the nurse. The doctor can thus establish a diagnosis within minutes after he has been called and remotely decide the actions to be taken, which can instantly be carried out by the nurse. To meet these targets, the system must allow real time information sharing as well as bilateral medical data update, such as, for example, the recording and the upload of an ECG by the nurse whilst the remote GP is keying in the clinical signs. The medical trolley includes useful equipment allowing the medical staff to collect information. It consists of specific, high quality audio-visual equipment functioning like an enhanced vid-

eoconferencing system with the possibility to remotely control the position and the zooming of the camera, and of several medical acquisition devices: tensiometer, a standard portable SCP-ECG compliant 12-lead interpreting electrocardiograph or a PEM device from the EPI-MEDICS project, oxymeter, glucometer, thermometer, digital stethoscope, and a weight scale. Most of these devices automatically transmit their measurements wireless via Bluetooth or an industry standard radio link to the laptop PC that is embedded in the trolley which sends the collected data to the server. Two platforms were deployed in two French elderly homes, one equipped with a PEM device and the other with a portable 12-Lead ECG recorder. The main evaluation result was a high level of acceptance of the Tele-Nurse system concept.

conclusIon In this article, we described some of the latest ambient intelligence and pervasive solutions that have been designed and are being embedded in the PEM device, and, more specifically, the ANN-based ECG interpretation and the Bayesian-network risk factors modules and their integration into the overall PEM telemedical platform. At this stage of development, and after conducting several clinical trials by both patients and health professionals, the PEM and the associated software tools were judged extremely easy-to-use and user-friendly and have allowed the detection of several arrhythmia-based cases (Fayn, Restier, Li, Chevalier, & Rubel, 2006; Rubel et al., 2005). The capability of the PEM to detect acute infarction in self-care situations remains, however, to be demonstrated because no such event occurred during the clinical trials. Ongoing works are intended to create and deploy a palette of predictors for a variety of profiles representative of the PEM potential user (male ≥ 45 years, female ≥ 55 years, and so forth). The next

309

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

Figure 6. The global architecture of the Tele-Nurse project. The medical trolley is equipped with a high performance visioconferencing system (Visadom), a laptop and several medical devices (PEM, oxymeter, etc.). The remote PC based GP workstation is staffed with two screens, one for the display of the Electronic Health Record, the other for the video conferencing (from Télisson et al., 2006)

step is to validate the global decision platform after deploying the newly developed decision support tools (the Bayesian network and the fuzzy alarm generator), and work on ways to make the system self-improving on both the embedded intelligence level (adapted to the patient medical history) and the communication level, in order to reduce the number of false positives and/or false negatives.

reFerences Atoui, H., Fayn, J., Gueyffier, F., & Rubel, P. (2006). Cardiovascular risk stratificationin decision support systems: a probabilistic approach. Application to p-health. In Proceedings of the Computers in Cardiology 2006 Conference (Vol. 33, pp. 281-284). Atoui, H., Fayn, J., & Rubel, P. (2004). A neural network approach for patient-specific 12-lead ECG synthesis in patient monitoring environments. In Proceedings of the IEEE Computers in Cardiology 2004 Conference (Vol. 31, pp. 161-164).

310

Axisa, F., Schmitt, P., Gehin, C., Delhomme, G., McAdams, E., & Dittmar, A. (2005). Flexible technologies and smart clothing for citizen medicine, home healthcare, and disease prevention. IEEE Transactions on Information Technology in Biomedicine, 9(3), 325-336. Campbell, P., Patterson, J., Cromer, D., Wall, K., Adams, G., Albano, A., et al. (2005). Prehospital triage of acute myocardial infarction: Wireless transmission of electrocardiograms to the on-call cardiologist via a handheld computer. Journal of Electrocardiology, 38(4), 300-309. EN 1064:2005+A1:2007. (2007) Medical informatics - standard communication protocol - computer-assisted electrocardiography (SCPECG). Brussels. Retrieved May 26, 2008, from http://www.cen.eu/ Fayn, J., Restier, L., Li, B., Chevalier, P., & Rubel, P. (2006, September 2-6). Assessment of the benefit of pervasive self-care ECG recording in arrhythmogenic patients. World Congress of Cardiology 2006, Barcelona, Spain. European Heart Journal, 27(Suppl. 1), 140.

Ambient Intelligence and Pervasive Architecture Designed within the EPI-MEDICS Personal ECG Monitor

Gueyffier, F., Boutitie, F., Boissel, J., Pocock, S., Coope, J., Cutler, J., et al. (1995). INDANA: A meta-analysis on individual patient data in hypertension. Protocol and preliminary results. Therapie, 50(4), 353-362.

Rubel, P., Fayn, J., Simon-Chautemps, L., Atoui, H., Ohlsson, M., Télisson, D., et al. (2004). New paradigms in telemedicine: Ambient intelligence, wearable, pervasive and personalized. Studies in Health Technology and Informatics, 108, 123-132.

Kerkenbush, N.L., & Lasome, C.E. (2003). The emerging role of electronic diaries in the management of diabetes mellitus. AACN Clinical Issues: Advanced Practice in Acute and Critical Care, 14, 371-378.

Task Force Report. (1998). Recommendations of a task force of the European society of cardiology and the European resuscitation council on the prehospital management of acute heart attacks. European Heart Journal, 19, 1140-1164.

Lagor, C., Aronsky, D., Fiszman, M., & Haug, P. (2001). Automatic identification of patients eligible for a pneumonia guideline: Comparing the diagnostic accuracy of two decision support models. In Proceedings of the 10th World Congress on Medical Informatics, MEDINFO (Vol. 10, pp. 493-497).

Télisson, D., Fayn, J., Placide, J., Rubel, P., & Comet, B. (2006). A new shared care telemedical solution dedicated to elderly patient nursing services for remote GP decision support. In Proceedings of the Computers in Cardiology 2006 Conference (Vol. 33, pp. 229-232).

McMurray, J., & Rankin, R. (1994). CardiologyI: Treatment of myocardial infarction, unstable angina, and angina pectoris. British Medical Journal, 309(6965), 1343-1350. Rubel, P., Fayn, J., Nollo, G., Assanelli, D., Li, B., Restier, L., et al. (2005). Toward personal ehealth in cardiology. Results from the EPI-MEDICS telemedicine project. Journal of Electrocardiology, 38(4), 100-106.

Télisson, D., Fayn, J., & Rubel, P. (2004). Design of a tele-expertise architecture adapted to pervasive multi-actor environments. Application to e-cardiology. In Proceedings of the IEEE Computers in Cardiology 2004 Conference (Vol. 31, pp. 749-752).

This work was previously published in International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 4, edited by J. Tan, pp. 53-68, copyright 2008 by IGI Publishing (an imprint of IGI Global).

311

312

Compilation of References

Abawajy, J. H. (2009). Human-computer interaction in ubiquitous computing environments. International Journal of Pervasive Computing and Communications, 5(1), 61–77. doi:10.1108/17427370910950311

Agrawal, R., Gunopulos, D., & Leymann, F. (1998). Mining process models from workflow logs. In Sixth International Conference on Extended Database Technology (pp. 469-483).

Abidi, S. S. R. (2001). Knowledge management in healthcare: Towards knowledgedriven decision-support services. International Journal of Medical Informatics, 63, 5–18.

Agrawal, R., Imielinski, T., & Swami, A. (1993). Mining association rules between sets of items in large databases. In ACM SIGMOD Conference on Management of Data, Washington, DC.

ACGME Outcome Project. (2007). Accreditation Council for Graduate Medical Education. Retrieved January 29, 2008, from http://www.acgme.org/outcome/comp/ compFull.asp. Acir, N., & Guzelis, C. (2004). Automatic spike detection in EEG by a two-stageprocedure based on support vector machines. Computers in Biology and Medicine, 34(7), 561–575. Adams, S. (2008). Blog-based applications and health information: Two case studies that illustrate important questions for consumer health informatics (chi) research. International Journal of Medical Informatics, 1–8. Retrieved from http://dx.doi.org/10.1016/j.ijmedinf.2008.06.009. Aggarwal, V. (2002). The application of the unified modeling language in object-oriented analysis of healthcare information systems. Journal of Medical Systems, 26, 383–397. Agrawal, R., Mannila, H., Srikant, R., Toivonen, H., & Inkeri, A. (1996). Fast discovery of association rules, advances in knowledge discovery and data mining (pp. 307–328). The Association for the Advancement of Artificial Intelligence and the MIT Press.

Ajzen, I., & Fishbein, M. (1973). Attitudinal and normative variables as predictors of specific behavior. Journal of Personality and Social Psychology, 27(1), 41–57. doi:10.1037/h0034440 Alberta, G. P. C. (2003). Oipc stakeholder survey 2003— highlights report. Office of the Information and Privacy Commissioner of Alberta. Tech. Rep. Al-Fedaghi, S. (2005a). How to calculate the information privacy. In Proceedings of the Third Annual Conference on Privacy, Security and Trust: How to Calculate Information Privacy, St. Andrews, New Brunswick, Canada. Retrieved January 29, 2008, from http://www.lib.unb.ca/ Texts/PST/2005/pdf/fedaghi.pdf Al-Fedaghi, S. (2005b). Privacy as a base for confidentiality. In Proceedings of the Fourth Workshop on the Economics of Information Security, Harvard University, Cambridge, MA. Retrieved January 29, 2008, from http:// infosecon.net/workshop/schedule.php Al-Fedaghi, S. (2006, June 20-23). Aspects of personal information theory. In Proceedings of the Seventh Annual IEEE Information Assurance Workshop (IEEE-IAW): Aspects of Personal Information Theory. West Point, NY: United States Military Academy.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Compilation of References

Al-Fedaghi, S. (2007, July 12-14). When reinventing principles is necessary. In Proceedings of the Seventh International Computer Ethics Conference, University of San Diego, CA. Alkan, A., Koklukaya, E., & Subasi, A. (2005). Automatic seizure detection in EEG using logistic regression and artificial neural network. Journal of Neuroscience Methods, 148(2), 167–176. doi:10.1016/j.jneumeth.2005.04.009 Allians för Sverige. (2006, August 23). Election program [Valmanifest]. Retrieved January 29, 2008, from www.maktskifte06.se/fileadmin/Upload/pdf/Valmanifest_2006.pdf Alpsan, D., Towsey, M., Ozdamar, O., Tsoi, A. C., & Ghista, D. N. (1995). Efficacy of modified backpropogation and optimization methods on a real-world medical application. Neural Networks, 8(6), 945–962. doi:10.1016/0893-6080(95)00034-W Al-Qirim, N. (2007). Championing telemedicine adoption and utilization healthcare organizations in New Zealand. International Journal of Medical Informatics, 76, 42–54. doi:10.1016/j.ijmedinf.2006.02.001 Ambler, S. (2004). The object primer: Agile modelingdriven development with UML 2.0. Cambridge: Cambridge University Press. Amenssisa, J., & Dabi, S. (2003). District-based telemedicine project in Ethiopia. Addis Ababa, Ethiopia: Ministry of Health. American Library Association. (1989). Presidential Committee on Information Literacy - Final Report. Retrieved from http://www.ala.org/ala/acrl/acrlpubs/whitepapers/ presidential.htm. Amouh, T., Gemo, M., & Macq, B. (2005). Versatile clinical information system design for emergency departments. IEEE Transactions on Information Technology in Biomedicine, 9(2), 174–183. Anderson, D. R., Whitmer, R. W., & Goetzel, R. Z. (2000). Health enhancement research organization (HERO) research committee. The relationship between modifiable health risks and group-level health care expenditures. American Journal of Health Promotion, 15, 45–52.

Anderson, J., & Aydin, C. (Eds.). (2005). Evaluating the organizational impact of healthcare information systems (2nd ed.). Springer. Anderson, M., Banker, R., & Ravindran, S. (2003). The new productivity paradox. Communications of the ACM, 46(3), 91–94. doi:10.1145/636772.636776 Angeles, R. (2007). An empirical study of the anticipated consumer response to RFID product item tagging. Industrial Management & Data Systems, 107(4), 461–583. Angluin, D., & Smith, C. H. (1983). Inductive inference: Theory and methods. Computing Surveys, 15(3), 237–269. Annane, D., Depondt, J., Aubert, P., Villart, M., Gehanno, P., & Gajdos, P. (2004). Hyperbaric oxygen therapy for radionecrosis of the jaw: A randomized, placebo-controlled, double-blind trial from the orn96 study group. Journal of Clinical Oncology, 22(24), 4893–4900. doi:10.1200/ JCO.2004.09.006 Antman, E., Lau, J., Kupelnick, B., Mosteller, F., & Chalmers, T. (1992). A comparison of results of meta-analyses of randomized control trials and recommendations of clinical experts. Journal of the American Medical Association, 268, 240–248. Antonadou, D., Pepelassi, M., Synodinou, M., Puglisi, M., & Throuvalas, N. (2002). Prophylactic use of amifostine to prevent radiochemotherapy-induced mucositis and xerostomia in head-and-neck cancer. International Journal of Radiation Oncology*Biology*Physics, 52(3), 739-747. Anttiroiko, A. (2004). Introduction to democratic e-governance. In Mälkiä, M., Anttiroiko, A.-V., & Savolainen, R. (Eds.), E-transformation in governance: New directions in government and politics (pp. 22–49). Hershey, PA: Idea Group. Apache Jakarta Project. (2007). Retrieved from http:// jakarta.apache.org Ari, S., & Saha, G. (2008). Classification of heart sounds using empirical mode decomposition based features. International Journal of Medical Engineering and Informatics, 1(1), 91–108. doi:10.1504/IJMEI.2008.019473

313

Compilation of References

Aronsky, D., Jones, I., Lanaghan, K., et al. (2007). Supporting patient care in the emergency department with a computerized whiteboard system. Journal of the American Medical Informatics Association. Aronson, A. (2001). Effective mapping of biomedical text to the UMLS Metathesaurus: The MetaMap program. Proceedings of the 2001 AMIA Symposium, pp. 17-21. Ash, A. S., Zhao, Y., Ellis, R. P., & Schlein, K. M. (2001). Finding future high-cost cases: Comparing prior cost versus diagnosis-based methods. Health Services Research, 36, 194–206. Asuncion, A., & Newman, D. J. (2007). UCI machine learning repository. Irvine, CA: University of California, School of Information and Computer Science. Retrieved May 26, 2008, from http://www.ics.uci.edu/~mlearn/ MLRepository.html ATA. (2009). Americal Telemedicine Association. Retrieved from http://www.atmeda.org Atoui, H., Fayn, J., & Rubel, P. (2004). A neural network approach for patient-specific 12-lead ECG synthesis in patient monitoring environments. In Proceedings of the IEEE Computers in Cardiology 2004 Conference (Vol. 31, pp. 161-164).

Bandyopadhyay, K., & Schkade, L. L. (2000). Disaster recovery planning by HMOs: theoretical insights. Health Care Management Review, 25(2), 74–85. Barros, P. P., Pinto, C. G., & Machado, A. (1999). A signaling theory of excessive technological adoption. Health Care Management Science, 2(2), 117–125. doi:10.1023/A:1019075425763 Bates, D. (2005). Physicians and ambulatory electronic health records. Health Affairs, 24(5), 1180–1189. Baxt, W. G. (1991). Use of an artificial neural network for the diagnosis of myocardial infarction. Annals of Internal Medicine, 115(11), 843–848. Baxt, W. G., Shofer, F. S., Sites, F. D., & Hollander, J. E. (2002). A neural computational aid to the diagnosis of acute myocardial infarction. Annals of Emergency Medicine, 39(4), 366–373. doi:10.1067/mem.2002.122705 Bazeley, P., & Richards, L. (2000). The NVivo qualitative project book. London; Thousand Oaks, CA: Sage Publications. Becker, J., Kugeler, M., & Rosemann, M. (2003). Process management: A guide for the design of business processes. Berlin, Heidelberg, New York: Springer.

Atoui, H., Fayn, J., Gueyffier, F., & Rubel, P. (2006). Cardiovascular risk stratificationin decision support systems: a probabilistic approach. Application to p-health. In Proceedings of the Computers in Cardiology 2006 Conference (Vol. 33, pp. 281-284).

Bennett, C. J. (2001, June 19). What government should know about privacy: A foundation paper. In Paper prepared for the Information Technology Executive Leadership Council’s Privacy Conference (Revised August 1, 2001). Retrieve January 29, 2008, from http://www. accessandprivacy.gov.on.ca/english/pub/wgskap.doc

Axisa, F., Schmitt, P., Gehin, C., Delhomme, G., McAdams, E., & Dittmar, A. (2005). Flexible technologies and smart clothing for citizen medicine, home healthcare, and disease prevention. IEEE Transactions on Information Technology in Biomedicine, 9(3), 325–336.

Berg, M. (2001). Implementing information systems in health care organizations: myths and challenges. International Journal of Medical Informatics, 64(2/3), 143–156. doi:10.1016/S1386-5056(01)00200-3

Baggot, R. (2005). A funny thing happened on the way to the forum? Reforming patient and public involvement in the NHS in England. Public Administration, 83(3), 533–551. Baker, G. B. (2002). Integrating technology and disease management—the challenges. Healthplan Magazine, 43, 60–66.

314

Bergan, T., & Hirsch, G. B. (1976). Analysis and Planning for improved distribution of Nursing personnel and services. A national model of supply, demand, and distribution. Cambridge, MA: Pugh-Roberts Associates, Inc. Berghout, B. M., Eminovic, N., de Keizer, N. F., & Birnie, E. (2007). Evaluation of general practitioner’s time investment during a store-and-forward teledermatology consultation. International Journal of Medical Informatics, 76, 384–391. doi:10.1016/j.ijmedinf.2007.04.004

Compilation of References

Berta, W. Teare, G. F., Gilbart, E., Soberman G., L. Lemieux-Charles, L., Davis, D., & Rappol, S. (2005). The contingencies of organizational learning in long-term care: factors that affect innovation adoption. Health Care Management Review, 30(4), 282–292. Berwick, D. M. (2005). The John Eisenberg lecture: Health services research as a citizen in improvement. Health Services Research, 40, 317–336. Berzal, F., Juan-Carlos, C., Cuenca, F., & Martín-Bautista, M. J. (2008). On the quest for easy-to-understand splitting rules. Data & Knowledge Engineering, 44(1), 31–48. Bhaskaran, S., & Hardley, F. (2002). Buyer beliefs, attitudes and behaviour: foods with therapeutic claims. Journal of Consumer Marketing, 19(7), 591–606. doi:10.1108/07363760210451410 Bhuptani, M., & Moradpour, S. (2005). RFID Field Guide - Deploying Radio Frequency Identification Systems. NJ: Prentice Hall. Bhutan Department of Information Technology. (2007). Bhutan’s Journey Towards Open Source. Paper presented at the DebConf7, Edinburgh, Scotland. https://penta. debconf.org/~joerg/events/25.en.html Biermann, E., Dietrich, W., Rihl, J., & Standl, E. (2002). Are there time and cost savings by using telemanagement for patients on intensified insulin therapy: A randomised, controlled trial. Computer Methods and Programs in Biomedicine, 69, 137–146. doi:10.1016/ S0169-2607(02)00037-8 Bigus, J. P., & Bigus, J. (1998). Constructing Intelligent software agents with Java – A Programmers Guide to Smarter Applications, Wiley, ISBN: 0-471-19135-3. Bishop, C. (1995). Neural networks for pattern recognition. Oxford University Press. Bobrie, G., Postel-Vinay, N., Delonca, J., & Corvol, P. (2007). Self-Measurement and Self-Titration in Hypertension. American Journal of Hypertension, 20, 1314–1320. doi:10.1016/j.amjhyper.2007.08.011 Boendermaker, P. M., Conradi, M. H., Schuling, J., Meyboom-de-Jong, B., Swierstra, R. P., & Metz, J. C. M. (2003)... Advances in Health Sciences Education : Theory and Practice, 8, 111–116.

Boendermaker, P. M., Schuling, J., Meyboom-de-Jong, B., Zwierstra, R. P., & Metz, J. C. M. (2000). What are the characteristics of the competent general practitioner trainer? Family Practice, 17, 547–553. Bond, M., Baggoley, C., & Erwich-Nijhout, M. (1998). Urgency, disposition and age groups: A casemix model for emergency departments. Emergency Medicine Australasia, 10(2), 103–110. Borich, G. (1996). Effective teaching methods (3rd ed.). Englewood Cliffs, NJ: Merrill/Prentice Hall. Boshier, R. (1990). Socio-psychological factors in electronic networking. International Journal of Lifelong Education, 9(1), 49–64. Bowen, J. L. (2006). Medical education: Educational strategies to promote clinical diagnostic reasoning. The New England Journal of Medicine, 355, 2217–2225. Brachman, R. J., Khabaza, T., Kloesgen, W., PiatetskyShapiro, G., & Simoudis, E. (1996). Mining business databases. Communications of the ACM, 39(11), 42–48. Brailsford, S., Churilov, L., & Liew, S. K. (2003). Treating ailing emergency departments with simulation: An integrated perspective. In International Conference on Health Sciences Simulation. Orlando, FL: The Society for Modeling and Simulation International. Brandeau, M. L., Sainfort, F., & Pierskalla, W. P. (2004). Operations research and health care. Boston, MA: Kluwer. Breiman, L., Friedman, J., Olshen, R., & Stone, C. (1984). Classification and regression trees. Belmont, CA: Wadsworth International Group. Brender, J. (2006). Handbook of evaluation methods for health informatics. Elsevier. Brent, E., Slusarz, P., & Thompson, A. (2002). Qualrus: The intelligent qualitative analysis program (manual). Columbia, MO: IdeaWorks. Briggs, R. O., Nunamaker, J. F. Jr, & Sprague, R. H. Jr. (2005). Special issue: crossing boundaries in information systems research. Journal of Management Information Systems, 22(4), 7–11. doi:10.2753/MIS0742-1222220401

315

Compilation of References

Brown, S., Lincoln, M., Groen, P., & Kolodner, R. (2003). VistA—U.S. Department of Veterans Affairs nationalscale HIS. International Journal of Medical Informatics, 69(2-3), 135–156.

Cain, M. (2006). A Two-step estimation applicaton in health care management evaluating the effects of choice of public health plan on health outcomes. Journal of Academy of Business and Economics, 6(2), 65–72.

Brynjolfsson, E. (1993). The productivity paradox of information technology. Communications of the ACM, 36(12), 67–77. doi:10.1145/163298.163309

Cain, M. (2007). Self-selection and evaluation of health care programs. Journal of Academy of Business and Economics, 7(3), 82–89.

Brynjolfsson, E., & Hitt, L. (1996). Paradox lost? Firm level evidence on the returns to information systems spending. Management Science, 42(4), 541–558. doi:10.1287/ mnsc.42.4.541

Caine, R. N., & Caine, G. (1991). Making connections: Teaching and the human brain. Alexandria, VA: Association for Supervision and Curriculum Development.

Buchanan, D. (1997). The limitations and opportunities of business process re-engineering in a politicized organizational climate. Human Relations, 50, 51–72. doi:10.1177/001872679705000103 Buescher, P. A., & Horton, S. J. (2002). Prenatal WIC participation in relation to low birth weight and Medicaid infant costs in North Carolina-An update. Center for Health Informatics and Statistics Studies, 122, 1–9. Buescher, P. A., Roth, M. S., & Williams, D. (1991). An evaluation of the impact of maternity care coordination on Medicaid birth outcomes in North Carolina. American Journal of Public Health, 81(12), 1625–1629. doi:10.2105/AJPH.81.12.1625 Burke, R., & Payton, P. E. (2006). Measuring and managing shoppability: Ten principles to convert shoppers into buyers. European Retail Digest, 50, 66–78. Burkle, T., Ammenwerth, E., Prokosch, H., & Dudeck, J. (2001). Evaluation of clinical information systems. What can be evaluated and what cannot? Journal of Evaluation in Clinical Practice, 7(4), 373–385. Business Process Management (BPM) Center. (2003). Process mining. Retrieved May 24, 2008, from http:// tmitwww.tm.tue.nl/research/processmining/ Bustard, D., Oakes, W. R., & He, Z. (1999). Models to promote effective system change. In IEEE International Conference on Software Maintenance (ICSM ‘99). Butterfield, R. (2007). “Data classification: A prerequisite to ILM”, http://www.snwonline.com/implement/ data_class_05-30-05.asp

316

Calhoun, J., Banaszak-Hol, J., & Hearld, L. (2006). Current marketing practices in the nursing home sector. Journal of Healthcare Management, 51(3), 185–200. Cameron, J., Baraff, L., & Sekhon, R. (1990). Case-mix classification for emergency departments. Medical Care, 28, 146–158. Campbell, P., Patterson, J., Cromer, D., Wall, K., Adams, G., & Albano, A. (2005). Prehospital triage of acute myocardial infarction: Wireless transmission of electrocardiograms to the on-call cardiologist via a handheld computer. Journal of Electrocardiology, 38(4), 300–309. Capek, P., Frank, S., Gerdt, S., & Shields, D. (2005). A history of IBM’s open source involvement and strategy. IBM Systems Journal, 44(2), 249–248. Carlson, C. (2003). Information overload, retrieval strategies and Internet user empowerment. In: L., Haddon (Ed.), The Good, the Bad and the Irrelevant (COST 269) 1(1) (pp. 169-173). Helsinki, Finland. Carnall, D. (2000). Medical software’s free future. BMJ (Clinical Research Ed.), 321(7267), 976. Carr, N. (2003). IT doesn’t matter. Harvard Business Review, 81(5), 41–49. Carr, N. (2004). Does IT matter?: Information technology and the corrosion of competitive advantage. Boston, MA: Harvard Business School Press. Catlett, J. (1991). Megainduction: Machine learning on very large databases. Unpublished doctoral dissertation, Basser Department of Computer Science, University of Sydney.

Compilation of References

CEBM. (2007). Why the sudden interest in EBM? Retrieved from http://www.cebm.utoronto.ca/intro/interest. htm. Ceglowski, A., Churilov, L., & Wassertheil, J. (2007). Combining data mining and discrete event simulation for a value-added view of a hospital emergency department. The Journal of the Operational Research Society, 58(2), 246–254. Chae, Y. M., Lee, J. H., Ho, S. H., Kim, H. J., Jun, K. H., & Won, J. U. (2001). Patient satisfaction with telemedicine in home health services for the elderly. International Journal of Medical Informatics, 61, 167–173. doi:10.1016/ S1386-5056(01)00139-3 Chae, M. H., Black, C., & Heitmeyer, J. (2006). Prepurchase and post-purchase satisfaction and fashion involvement of female tennis wear consumers. International Journal of Consumer Studies, 30(1), 25–33. doi:10.1111/j.1470-6431.2005.00434.x Chan, P. K., & Stolfo, S. J. (1993). Meta-learning for multistrategy and parallel learning. In Proceedings of the Second International Workshop on Multistrategy Learning. Chang, I. C., Hwang, H. G., Hung, M. C., Lin, M. H., & Yen, D. C. (2007). Factors affecting the adoption of electronic signature: Executives’ perspective of hospital information department. Decision Support Systems, 44(1), 350–359. doi:10.1016/j.dss.2007.04.006 Charmaz, K. (2006). Constructing grounded theory. London; Thousand Oaks, CA: Sage Publications. Chau, P., & Tam, K. (1997). Factors affecting the adoption of open systems: An exploratory study. MIS Quarterly: Management Information Systems, 21(1), 1–20. Chen, I. J., Yang, K. F., & Tang, F. I. (2008). Applying the technology acceptance model to explore public health nurses’ intentions towards web-based learning: A cross-sectional questionnaire survey. International Journal of Nursing Studies, 45, 869–878. doi:10.1016/j. ijnurstu.2006.11.011

Chen, P., Huang, D., Zhao, X., & Li, X. (2008). Predicting contact map using Radial Basis Function Neural Network and Conformational Energy Function. International Journal of Bioinformatics Research and Applications, 4(2), 123–136. doi:10.1504/IJBRA.2008.018340 Chesney, T., Penny, K., Oakley, P., Davis, S., Maffulli, N., & Templeton, J. (2006). Data Mining Medical Information: Should Artificial Neural Networks Be Used To Analyze Trauma Audit Data. International Journal of Healthcare Information Systems and Informatics, 1(2), 51–64. Cheung, D., Han, J., Ng, V. T., & Fu, A. W. (1996). Fast distributed algorithm for mining association rules. In International Conference on Parallel and Distributed Information Systems, Tokyo, Japan (pp. 31-42). Chin, J., Diehl, V., & Norman, K. (1988). Development of an instrument measuring user satisfaction of the humancomputer interface. Paper presented at the CHI, New Ork. Chiu, R., & Kuo, C. (2006). A decision support model of demand forecast for national immunization vaccines. International Journal of Electronic Healthcare, 2(1), 92–104. doi:10.1504/IJEH.2006.008699 Christensen, C., Bohmer, R., & Kenagy, J. (2000). Will disruptive innovations cure health care. Harvard Business Review, (September-October): 102–112. Clarke, J. (2006). Consumers, clients or citizens? Politics, policy and practice in reform in social care. European Societies, 8(3), 423–442. Clarke, L. L., Miller, M. K., & Mahan, C. S. (1993). The Effectiveness of Florida’s Improved Pregnancy Outcome Program. Journal of Health Care for the Poor and Underserved, 4(2), 117–132. Cleary, P. D. (2003). A hospitalization from hell: A patient’s perspective on quality. Annals of Internal Medicine, 138(1), 33–39. Cline, J. (2007). “Growing pressure for data classification”, http://www.computerworld.com/action/article.do ?articleId=9014071&command=viewArticleBasic, Last accessed on 22/11/2007 Cohen, C. I., & Cohen, E. J. (1978). Health education: panacea, pernicious or pointless? The New England Journal of Medicine, 28, 718–720.

317

Compilation of References

Computing in the Physician’s Practice. (2000). Retrieved from http://www.harrisinteractive.com/harris_poll/index. asp?PID=58.

Creswell, J. (1994). Research design: Qualitative & quantitative approaches. Thousand Oaks, CA: Sage Publications.

Cook, J. E., & Wolf, A. L. (1998). Discovering models of software processes from event-based data. ACM Transactions on Software Engineering and Methodology, 7(3), 215–249.

Creswell, J. (2005). Educational research: Planning, conducting, and evaluating quantitative and qualitative research (2nd ed.). Upper Saddle River, NJ: Merrill.

Cook, J. E., & Wolf, A. L. (1998, November). Eventbased detection of concurrency. In Proceedings of the Sixth International Symposium on the Foundations of Software Engineering (FSE-6), Orlando, FL (pp. 35-45). Costa, A. X., Ridely, S. A., Shahani, A. K., Harper, P. R., de Senna, V., & Nielsen, M. S. (2003). Mathematical modelling and simulation for planning critical care capacities. Anaesthesia, 58, 320–327. Cote, M. J., & Tucker, S. L. (2001). Four methodologies to improve health care demand forecasting. Healthcare Financial Management, 5, 54–58. Côté, L., & Leclère, H. (2000). How clinical teachers perceive the doctor-patient relationship and themselves as role models. Academic Medicine, 75, 1117–1124. Coulter, A., Entwistle, V., & Gilbert, D. (1999). Sharing decisions with patients: Is the information good enough? BMJ (Clinical Research Ed.), 318, 318–322. Cousins, M. S., Shickle, L. M., & Bander, J. A. (2002). An introduction to predictive modeling for disease management risk stratification. Disease Management, 5, 157–167. Cox, S. S., & Swanson, M. S. (2002). Identification of teaching excellence in operating room and clinic settings. American Journal of Surgery, 183, 251–255. Coyle, R. G. (1996). System dynamics modelling: A practical approach. London: Chapman and Hall. Crawford, A. G., Fuhr, J. P., Clarke, J., & Hubbs, B. (2005). Comparative Effectiveness of Total Population versus Disease-Specific Neural Network Models in Prediciting Medical Costs. Disease Management, 8(5), 277–287. doi:10.1089/dis.2005.8.277 Crawford, W. (2006). Why aren’t ebooks more successful? Retrieved from http://www.econtentmag.com/Articles/ ArticleReader.aspx?ArticleID=18144.

318

Crosby, P. B. (1979). Quality is free. New York: McGrawHill. Cruz, I. A., Gramaxo Silva, A., Soares, P., Oliveira, I., Serrano, J. A. V., & Paulo Cunha, J. (2002). Modelling the hospital into the future with UML. International Journal of Healthcare Technology and Management, 4, 193–204. Dalmadge, C. L., & Cain, M. (2008). Utilizing Neural Networks o Assess Community-Level Vulnerability to the Methamphetamine Epidemic. Issues in Information Systems, 9(2), 258–266. Darden, L., & Rada, R. (1988). Hypothesis formation via interrelations. In Prieditis, A. (Ed.), Analogica (pp. 109–128). London: Pitman. Darmstadt, G., Bhutta, Z., Cousens, S., Adam, T., Walker, N., & Berni, L. d. (2005). Evidence-based, cost-effective interventions: how many newborn babies can we save? Lancet, 365(9463), 977–988. doi:10.1016/S01406736(05)71088-6 Das, A., Ben-Menachem, T., Cooper, G. S., Chak, A., Sivak, M. V. Jr, Gonet, J. A., & Wong, R. C. (2003). Predicition of outcome in acute lower-gastrointestinal hemorrahage based on and artificial neural network: Internal and external validation of a predictive model. Lancet, 362(9392), 1261–1266. doi:10.1016/S01406736(03)14568-0 Davis, D., Thomson, M., Oxman, A., & Haynes, R. (1997). Changing physician performance: A systematic review of the effect of continuing medical education strategies. Journal of the American Medical Association, 274, 700–705. Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. Management Information Systems Quarterly, 13(3), 319–340. doi:10.2307/249008

Compilation of References

Dawes, M., & Sampson, U. (2003). Knowledge management in clinical practice: Asystematic review of information seeking behavior in physicians. International Journal of Medical Informatics, 71, 9–15.

Deodhar, J. (2002). Telemedicine by email--experience in neonatal care at a primary care facility in rural India. Journal of Telemedicine and Telecare, 8(Suppl 2), 20–21. doi:10.1258/135763302320301867

de Medeiros, A. K. A., van der Aalst, W. M. P., & Weijters, A. J. M. M. (2003). Workflow mining: Current status and future directions. In Meersman, R., Tari, Z., & Schmidt, D. C. (Eds.), On the move to meaningful Internet systems 2003: CoopIS, DOA, and ODBASE (pp. 389–406). Berlin, Heidelberg: Springer-Verlag.

Department of Health. (2003). Choice of hospitals: Guidance for PCTs, NHS Trusts and SHAs on offering patient choice of where they are treated. London: Department of Health.

Dedrick, J., & West, J. (2003). Why firms adopt open source platforms: A grounded theory of innovation and standards adoption. Paper presented at the Proceedings of the Workshop on Standard Making: A Critical Research Frontier for Information Systems. Seattle, WA. Dedrick, J., & West, J. (2004). An exploratory study into open source platform adoption. Paper presented at the Proceedings of the 37th Hawaii International Conference on System Sciences. Degoulet, P., Fieschi, M., Jaulent, M.-C., & Ménard, J. (2004). Patient empowerment, cybermedicine, and citizen education. In Nelson, R., & Ball, M. (Eds.), Consumer informatics. Applications and strategies in cyber health care (pp. 120–128). Berlin: Springer. DeLano, W. (2005). The case for open source software in drug discovery. Drug Discovery Today, 10(3). DeLone, W., & McLean, E. (2003). The DeLone and McLean model of information systems success: A ten-year update. Journal of Management Information Systems, 19(4), 9–30. Deming, W. E. (1991). Out of crisis. Cambridge, MA: Cambribge University Press. DeMonaco, H. (2009). Patient- and physician-oriented web sites and drug surveillance: Bisphosphonates and severe bone, joint, and muscle pain. Archives of Internal Medicine, 169(12), 1164–1166. doi:10.1001/archinternmed.2009.133 Deng, M., Cock, D. D., & Preneel, B. (2009). Towards a cross-context identity management framework in e-health. Online Information Review, 33(3), 422–442. doi:10.1108/14684520910969880

Department of Health. (2008). Payment by results (PbR). Retrieved May 26, 2008, from www.dh.gov.uk/en/Policyandguidance/Organisationpolicy/Financeandplanning/ NHSFinancialReforms/index.htm. Despont-Gros, C., Mueller, H., & Lovis, C. (2005). Evaluating user interactions with clinical information systems: A model based on human-computer interactions models. Journal of Biomedical Informatics, 38, 244–255. Devaraj, S., & Kohli, R. (2000). Information technology payoff in the health-care industry: a longitudinal study. Journal of Management Information Systems, 16(4), 41–67. Directive, E. U. (1995, October 24). Directive 95/46/EC of the European Parliament and of the council. Retrieved January 29, 2008, from http://eur-lex.europa.eu/LexUriServ/ LexUriServ.do?uri=CELEX:31995L0046:EN:HTML Djorhan, V., & Churilov, L. (2003). Business interactions in an acute emergency department in Australia: A clinical process modeling perspective. In the 6th Pacific Asia Conference on Information Systems (PACIS 2002). Tokyo, Japan: The Japan Society for Management Information (JASMIN). Do, H. Rahin, E. (2002), “Coma: A System for Flexible Combination of Schema Matching Approaches”, Proceedings of 28th Conference on Very Large Databases, Morgan Kaufmann, pp 610-621USA. Doan, A.-H. Lu, y. Lee, T. Han, J (2003), “Profile-Based Object Matching for Information Integration”, IEEE Intelligent Systems Magazine, pp 54-59. USA. Documentation, C. (2007). Retrieved from www.cisco. com

319

Compilation of References

Dreiseitl, S., Ohno-Machado, L., Kittler, H., Vinterbo, S., Billhardt, H., & Binder, M. (2001). A comparison of machine learning methods for the diagnosis of pigmented skin lesions. Journal of Biomedical Informatics, 34, 28–36.

Erickson, B., Langer, S., & Nagy, P. (2005). The role of open source software in innovation and standardization in radiology. Journal of the American College of Radiology, 2(11), 927–931.

Duff, F. L., Daniel, S., Kamendje, B., Le Beux, P., & Duvauferrier, R. (2005). Monitoring incident report in the healthcare process to improve quality in hospitals. International Journal of Medical Informatics, 74(2-4), 111–117. doi:10.1016/j.ijmedinf.2004.06.007

Ethiopian Telecommunications Corporation. (2007). www.telecom.net.et

Dumas, M., van der Aalst, W., & ter Hofstede, A. (2005). Process-aware information systems: Bridging people and software through process technology. Hoboken, NJ: Wiley-Interscience. Earl, M. J. (1994). The new and the old of business process redesign. The Journal of Strategic Information Systems, 3(1), 5–22. Edelstein, H. (1998). Data mining: Myth and reality. GIGA Information Group Telepresentation. Eftekhar, B., Mohammad, K., Ardebili, H. E., Ghodsi, M., & Keabchi, E. (2005). Comparison of artificial neural network and logistic regression models for prediction of mortality in head trauma based on clinical data. BMC Medical Informatics and Decision Making, 5(3), 1–8. Elmasri, R. (2000). Fundamentals of database systems (3rd ed.). Addison Wesley. Elwyn, G., & Edwards, A. (2001). Evidence-based patient choice? In A. Edwards & G. Elwyn, G. (Eds.), Evidencebased patient choice. Inevitable or impossible? (pp. 3-18). Oxford University Press. Elzubeir, M. A., & Rizk, D. E. E. (2001). Identifying characteristics that students, interns and residents look for in their role models. Medical Education, 35, 272–277. EN 1064:2005+A1:2007. (2007) Medical informatics - standard communication protocol - computer-assisted electrocardiography (SCP-ECG). Brussels. Retrieved May 26, 2008, from http://www.cen.eu/ Epstein, R. M. (2007). Assessment in medical education. The New England Journal of Medicine, 356, 387–396.

320

Eudaptics Software Gmbh. (1999). Viscovery SOMine. Wien: Eudaptics Software Gmbh. Evans, J., Hwang, Y., & Nagarajan, N. (1997). Cost reduction and process reengineering in hospitals. Journal of Cost Management, 11(3), 20–27. Evenden, D., Harper, P. R., Brailsford, S. C., & Harindra, V. (2005). Improving the cost-effectiveness of chlamydia screening with targeted screening strategies. The Journal of the Operational Research Society, 57(12), 1400–1412. Eyler, A., Mayer, J., Rafii, R., Housemann, R., Brownson, R., & King, A. (1999). Key informant surveys as a tool to implement and evaluate physical activity interventions in the community. Health Education Research, 14(2), 289–298. Eysenbach, G. (2004). Improving the quality of Web surveys: The checklist for reporting results of Internet e-surveys (CHERRIES). Journal of Medical Internet Research, 6(3), e34. Eysenbach, G. (2001). What is e-health? Journal of Medical Internet Research, 3(2), e20. doi:10.2196/jmir.3.2.e20 Eysenbach, G., & Köhler, C. (2003). What is the prevalence of health-related searches on the World Wide Web? Qualitative and quantitative analysis of search engine queries on the Internet. In Proceedings of the AMIA Symposium (pp. 225-229). Fayn, J., Restier, L., Li, B., Chevalier, P., & Rubel, P. (2006, September 2-6). Assessment of the benefit of pervasive self-care ECG recording in arrhythmogenic patients. World Congress of Cardiology 2006, Barcelona, Spain. European Heart Journal, 27(Suppl. 1), 140. Fayyad, U. M., & Irani, K. (1993). Multi-interval discretization of continuous-valued attributes for classification learning. In Proceedings of the International Joint Conferences on Artificial Intelligence 1993, San Francisco, CA.

Compilation of References

Ferguson, T. (2002, September 2002). Key concepts in online health: E-patients as medical researchers. The Ferguson Report Retrieved August 2005, from http:// www.fergusonreport.com/articles/fr00903.htm Ferguson, T. (2009). E-patients: How they can help us heal healthcare. Retrieved June 27, 2009, 2009, from http://www.acor.org/epatientswiki/

Forrester, J. W. (1961, July-August). Industrial Dynamics: A Major Breakthrough for Decision Makers. Harvard Business Review, ▪▪▪, 37–66. Foughty, T. (1999). Accomplishing health improvement through diabetes disease management. Drug Benefit Trends, 12, 43–45. FountainJ.E. (2001).

Ferreira, F.L., Peres, Bota, D., Bross, A., Melot, C., & Vincent, J.L. (2001). Serial evaluation of the SOFA score to predict outcome in critically ill patients. Journal of the American Medical Association, 286, 1754–1758.

Frawley, Piatestsky-Shapiro, & Matheus (1991). Knowledge discovery in databases: An overview. Menlo Park, CA: AAAI Press.

Fichman, R., & Kemerer, C. (1997). Object technology and reuse: Lessons from early adopters. Computer, 30(10), 47–59.

Frazier, P. A., & Tix, A. P. (2004). Testing moderator and mediator effects in counseling psychology. Journal of Counseling Psychology, 51(1), 115–134. doi:10.1037/0022-0167.51.1.115

Fink, M. (2003). The business and economics of Linux and open source. Upper Saddle River, NJ: Prentice Hall PTR. Finkenzeller. K. (1999), RFID Handbook, John Wiley and Sons Ltd, USA. Fischer-Hübner, S., & Ott, A. (1998, October 5-8). From a formal privacy model to its implementation. In Proceedings of the 21st National Information Systems Security Conference, Arlington, VA. Retrieved January 29, 2008, from http://www.cs.kau.se/~simone/niss98.pdf Fitzgerald, & Kenny. (2004). Developing an information systems infrastructure with open source software. Software, IEEE, 21(1), 50-55. Fitzgerald. (2004). A critical look at open source. Computer, 37(7), 92-94. Flood, R. (1999). Rethinking the fifth discipline. Learning within the unknowable. London: Ruthledge. Forducey, P., Kaur, K., Scheiderman-Miller, C., & Tan, J. (2005). Surveying the e-health landscape: Cases and applications. In Tan, J. (Ed.), E-health care information systems. An introduction for students and professionals (pp. 203–224). San Fransisco, CA: Jossey Bass. Forrest, S., Hofmeyr, S. A., Somayaji, A., & Longstaff, T. A. (1996). A sense of self for unix processes. In 1996 IEEE Symposium on Security and Privacy (pp. 120-128). IEEE Computer Society. Building the virtual state. Information technology and institutional change. Washington, DC: Brookings Institution Press.

Fridman, S. (2000). Doctors lag when it comes to computer use - Industry trend or event. Retrieved from http://www.findarticles.com/p/articles/mi_m0HDN/ is_2000_March_28/ai_60907019. Friedman, C., & Wyatt, J. (1997). Evaluation methods in medical informatics. New York: Springer. Friedman, M., Landsberg, R., Pryor, S., Syed, Z., Ibrahim, H., & Caldarelli, D. (2001). The occurrence of sleep-disordered breathing among patients with head and neck cancer. The Laryngoscope, 111, 1917–1919. doi:10.1097/00005537-200111000-00008 Friedman, C., Hripcsak, G., Johnson, S. B., Cimino, J. J., & Clayton, P. D. (1990). Ageneralized relational schema for an integrated clinical patient database. In Proceedings of the 14th Annual Symposium on Computer Applications in Medical Care (pp. 335-339). Fuhrer, P. and Guinard, D. (2007). Building a Smart Hospital using RFID technologies: Use Cases and Implementation. rfidehealth.pdf Gadd, C., & Penrod, L. (2001). Assessing physician attitudes regarding use of an outpatient EMR: A longitudinal, multi-practice study. Proceedings/AMIA Annual Symposium, 194-198.

321

Compilation of References

Gagnona, M. P., Godinb, G., Gagnéb, C., Fortina, J. P., Lamothec, L., Reinharza, D., & Cloutierd, A. (2003). An adaptation of the theory of interpersonal behaviour to the study of telemedicine adoption by physicians. International Journal of Medical Informatics, 71, 103–115. doi:10.1016/S1386-5056(03)00094-7 Galpottage, A., & Norris, A. (2005). Patient consent principles and guidelines for e-consent: A New Zealand perspective. Health Informatics Journal, 11(1), 5-18. Retrieved January 29, 2008, from http://jhi.sagepub.com/ cgi/reprint/11/1/5 Gamm, L., Barsukiewicz, C., Dansky, K., & Vasey, J. Pre- and post-control model research on end-users’ satisfaction with an electronic medical record: Preliminary results. Paper presented at the AMIA. Ganguly, P., & Ray, P. (2000). A methodology for the development of software agent based interoperable telemedicine systems: A tele-electrocardiography perspective. Telemedicine Journal, 6, 283–294. Gatignon, H., & Robertson, T. (1989). Technology diffusion: An empirical test of competitive effects. Journal of Marketing, 53(1), 35–49. Gehrke, J., Ramakrishnan, R., & Ganti, V. (2000). RainForest - a framework for fast decision tree construction of large datasets. Data Mining and Knowledge Discovery, 4(2-3), 127–162. Gehshan, S., Belnap, D., & Snyder, A. (2009). Medicaid Outreach and Enrollment for Pregnant Women: What Is the State of the Art?Washington, DC: Urban Institute. Giles, J. (2005). Internet encyclopedias go head to head. Retrieved from http://www.nature.com/news/2005/051212/ full/438900a.html. Glaser, B. (1978). Theoretical sensitivity: Advances in the methodology of grounded theory. Mill Valley, CA: Sociology Press. Glaser, B. (1999). The future of grounded theory. Qualitative Health Research, 9(6), 836–845. Glaser, B., & Strauss, A. (1967). The discovery of grounded theory; strategies for qualitative research. Chicago, IL: Aldine Publishing Company.

322

Glenton, C., Paulsen, E., & Oxman, A. (2005). Portals to wonderland: Health portals lead to confusing information about the effects of health care. Medical Informatics and Decision Making, 5(7). Glover, B., & Bhatt, H. (2006). RFID Essentials, O’Reilly Media, Inc (pp. 10–0596009445). USA: ISBN. Glynn, E., Fitzgerald, B., & Exton, C. (2005). Commercial adoption of open source software: An empirical study. Paper presented at the Proceedings of the 2005 International Symposium on Empirical Software Engineering (ISESE’05). Noosa Heads, QLD, Australia. Goetzel, R. Z., Anderson, D. R., & Whitmer, R. W. (1998). The relationship between modifiable health risks and health care expenditures: An analysis of the multiemployer HERO health risk and cost database. Journal of Occupational and Environmental Medicine, 40, 843–854. Goldberg, L. (1993). The structure of phenotypic personality traits. The American Psychologist, 48, 26–34. Goldman, R., & Gabriel, R. (2005). Innovation happens elsewhere: Open source as business strategy. Amsterdam; Boston, MA: Morgan Kaufmann. Goldratt, E. M. (1990). Theory of constrains. Croton-onHudson. NY: North River Press. Goldsmith, D., & Safran, C. (2004). Collaborative healthware. In Nelson, R., & Ball, M. (Eds.), Consumer informatics. Applications and strategies in cyber health care (pp. 9–19). Berlin: Springer. Good, T., & Brophy, J. (2000). Looking in classrooms (8th ed.). New York: Longman. Goode, S. (2005). Something for nothing: Management rejection of open source software in Australia’s top firms. Information & Management, 42(5), 669–681. Google. (2007). Search returned http://www.google.com/ search?hl=en&q=GI+Motility. Goossen, W., Ozbolt, J., Coenen, A., Park, A., Mead, C., & Ehnfors, M. (2004). Development of a provisional domain model for the nursing process for use within the health level 7 reference information model. Journal of the American Medical Informatics Association: JAMIA, 11, 186–194.

Compilation of References

Gore, A. (1992). Infrastructure for the global village. Scientific American, 265, 150–153. Gospodarevskaya, E., Churilov, L., & Wallace, L. (2005). Modeling the patient care process of an acute care ward in a public hospital: A methodological perspective. Hawaii: HICSS. Goss, E., & Vozikis, G. S. (2002). Improving Health Care Organizational Management Through Neural Network Learning. Health Care Management Science, 5(3), 221–227. doi:10.1023/A:1019760901191 Gottlieb, R., Rogers, J., & Rainey, K. (2002). Using curriculum evaluation as a basis for clinical faculty development. College Student Journal, 36(2), 279+. Retrieved January 29, 2008, from Questia database, http://www. questia.com/PM.qst?a=o&d=5000799148 Grana, J., Preston, S., McDermott, P. D., & Hanchak, N. A. (1997). The use of administrative data to risk-stratify asthmatic patients. American Journal of Medical Quality, 12, 113–119. Greener, I. (2003). Patient choice in the NHS: The view from economic sociology. Social Theory & Health, 1, 72–89. Gruber, T. R. (1993). A translation approach to portable ontology specifications. Knowledge Acquisition, 5, 199–220. Gueyffier, F., Boutitie, F., Boissel, J., Pocock, S., Coope, J., & Cutler, J. (1995). INDANA: A meta-analysis on individual patient data in hypertension. Protocol and preliminary results. Therapie, 50(4), 353–362. Guido, F. (2007). The organizational learning curve. European Journal of Operational Research, 177(3), 1375–1384. doi:10.1016/j.ejor.2005.04.009 Gupta, K., & Stewart, D. W. (1996). Customer satisfaction and customer behavior: the differential role of brand and category expectations. Marketing Letters, 7(3), 249–263. doi:10.1007/BF00435741 Guyatt, G., Cairns, J., Churchill, D., Cook, D., Haynes, B., & Hirsch, J. (1992). Evidence-Based Medicine Working Group: Evidence-based medicine. A new approach to teaching the practice of medicine. Journal of the American Medical Association, 268, 2420–2425.

Haile-Mariam, A. (2002). Renaissance: Strategies for ICT Development in Ethiopia. M.Sc Thesis, School of Engineering Postgraduate Engineering Program. Hall, M. (2007). Core servlet and Java server pages. Sun Microsystems Press, Retrieved from http://www. coreservlets.com. Hammer, M., & Champy, J. (1993). Reengineering the corporation - A manifesto for business revolution. New York: Harper Business. Hanss, R., Buttgereit, B., Tonner, P. H., Bein, B., Schleppers, A., & Steinfath, M. (2005). Overlapping induction of anesthesia: An analysis of benefits and costs. Anesthesiology, 103(2), 391–400. doi:10.1097/00000542200508000-00023 Harel, D. (1987). Statecharts: A visual formalism for complex systems. Science of Computer Programming, 8, 231–274. Harno, K., Arajärvi, E., Paavola, T., Carlson, C., & Arnala, I. (2001). Clinical effectiveness and cost analysis of patient referral by videoconferencing in orthopaedics. Journal of Telemedicine and Telecare, 7(4), 219–225. doi:10.1258/1357633011936435 Harno, K., Paavola, T., Carlson, C., & Viikinkoski, P. (2000). Patient referral by telemedicine: effectiveness and cost analysis of an intranet system. Journal of Telemedicine and Telecare, 6(6), 320–329. doi:10.1258/1357633001935996 Harper, P. R. (2002). A framework for operational modelling of hospital resources. Health Care Management Science, 5(3), 165–173. Harper, P. R. (2005). A review and comparison of classification algorithms for medical decision making. Health Policy (Amsterdam), 71, 315–331. Harper, P. R., & Pitt, M. (2004). On the challenges of health care modelling and a proposed project life-cycle for successful implementation. The Journal of the Operational Research Society, 55, 657–661.

323

Compilation of References

Harper, P. R., Sayyad, M. G., de Senna, V., Shahani, A. K., Yajnik, C. S., & Shelgika, K. M. (2003). A systems modelling approach for the prevention and treatment of diabetic retinopathy. European Journal of Operational Research, 150, 81–91.

Hendricks, K. B., Singhal, V., & Stratman, J. K. (2007). The impact of enterprise systems on corporate performance: a study of ERP, SCM, and CRM system implementations. Journal of Operations Management, 25(1), 65–82. doi:10.1016/j.jom.2006.02.002

Harris, K. (2003). How do patients choose physicians? Enrolees in employment-related health plans. Health Services Research, 38(2), 711–732.

Herbst, J., & Karagiannis, D. (2000). Integrating machine learning and workflow management to support acquisition and adaptation of workflow models. International Journal of Intelligent Systems in Accounting Finance & Management, 9, 67–92.

Harrison, S., Dowswell, G., & Milewa, T. (2002). Guest editoral: Public and user “involvement” in the UK National Health Service. Health & Social Care in the Community, 10(2), 63–66. Hawking, S. (2001). The universe in a nutshell. New York, NY: Bantam Books. Haynes, R. (1993). Where’s the meat in clinical journals? [editorial]. ACP Journal Club, 119, A22–A23. Hazen, G. (2000). Preference factoring for stochastic trees. Management Science, 46(3), 389–403. Hazen, G. B. (1992). Stochastic trees: A new technique for temporal medical decision modeling. Medical Decision Making, 13, 227–236. He, Q., & Antón, A. I. (2003, June 16-17). A framework for modeling privacy requirements in role engineering. In Proceedings of the International Workshop on Requirements Engineering for Software Quality (REFSQ 2003), Klagenfurt / Velden, Austria. Retrieved January 29, 2008, from http://crinfo.univ-paris1.fr/REFSQ/03/ papers/P14-He.pdf Heathfield, H., & Louw, G. (1999). New challenges for clinical informatics: Knowledge management tools. Health Informatics Journal, 5(2), 67–73. Hederman, L., Smutek, D., Wade, V., & Knape, T. (2002). Representing clinical guidelines in UMl: A comparative study. Studies in Health Technology and Informatics, 90, 471–477. Hedgepeth, W. O. (2007). RFID metrics: decision making tools for today’s supply chains.Boca Raton, FL: CRC Press, (ISBN: 9780849379796) Heller, C., & Arozullah, A. (2001). Implementing change: It’s as hard as it looks. Disease Management & Health Outcomes, 9, 551–563.

324

Herbst, J. (2000). A machine learning approach to workflow management. In 11th European Conference on Machine Learning. Berlin, Germany: Springer (LNCS 1810, pp. 183-194). Herbst, J., & Karagiannis, D. (1999, August), An inductive approach to the acquisition and adaptation of workflow models. In M. Ibrahim & B. Drabble (Eds.), Proceedings of the IJCAI’99 Workshop on Intelligent Workflow and Process Management: The New Frontier for AI in Business (pp. 52-57). Stockholm, Sweden. Herman, A. A., & Berendes, H. W. (1996). The evaluation of the effectiveness of a community-based enriched model of prenatal intervention project in the District of Columbia. Health Services Research, 33(5), 609–621. Hersh, W., Mailhot, M., Arnott-Smith, C., & Lowe, H. (2002). Selective automated indexing of findings and diagnoses in radiology reports. Journal of Biomedical Informatics, 34, 262–273. Heskett, L. (1983). Shouldice Hospital Ltd, Rev. 1/88. Boston, MA: Harvard Business School. Heymann, D. L., & Rodier, G. (2004). Global Surveillance, National Surveillance, and SARS. Emerging Infectious Diseases Journal, 10(2). Retrieved 1st December 2006, from http://origin.cdc.gov/ncidod/EID/ vol10no2/03-1038.htm HHS. (2003). Summary of the HIPAA privacy rule. U.S. Department of Health & Human Services. Retrieved January 29, 2008, from http://www.hhs.gov/ocr/privacysummary.pdf

Compilation of References

Hier, D., Rothschild, A., LeMaistre, A., & Keeler, J. (2005). Differing faculty and housestaff acceptance of an electronic health record one-year after implementation. International Journal of Medical Informatics, 74, 657–662. Hillestad, R., Bigelow, J., Bower, A., Girosi, F., Meili, R., & Scoville, R. (2005). Can electronic medical record systems transform healthcare? Potential health benefits, savings and costs. Health Affairs, 24(5), 1103–1117. HIMSS. (2006). HIMSS bylaws. Retrieved June 1, 2006, from http://www.himss.org/content/files/himss_bylaws. pdf. HIPAA. (1996, August 21). Health Insurance Portability and Accountability Act of 1996, Public Law 104-191, 104th Congress. Retrieved January 29, 2008, from http:// aspe.hhs.gov/admnsimp/pl104191.htm

Hoffenberg, S., Hill, M. B., & Houry, D. (2001). Does sharing process differences reduce patient length of stay in the emergency department? Annals of Emergency Medicine, 38(5), 533–540. Hoffman, K. J. (1997). Demystifying mental health information needs through integrated definition (IDEF) activity and data modeling. Journal of the American Medical Association, Symposium.l, 111-115. Hogarth, M., & Turner, S. (2005). A study of clinical related open source software projects. Paper presented at the American Medical Informatics Association Conference (AMIA). Washington, D.C. Holck, J., Larsen, M., & Pedersen, M. (2005). Managerial and technical barriers to the adoption of open source software. Berlin, Heidelberg: Springer.

Hippel, E. v. (2001). Innovation by User Communities: Learning from Open-Source Software. MIT Sloan Management Review, 42(4), 82.

Hoo, K. S., Wong, S. T. C., Laxer, K. D., Knowlton, R. C., & Wan, C. (2000). Design patterns in medical imaging information systems. Proc. SPIE Medical Imaging, 3980, 256-263.

Hirsch, G. B., Forsyth, B. R., Bergan, T. A., & Goodman, M. R. (1976). Projecting a teaching hospital’s future utilization: A dynamic simulation approach. Journal of Medical Education, 51(11), 937–939.

Hornik, K., Stinchcombe, M., & White, H. (1989). Multilayer feedforward networks are universal approximations. Neural Networks, 2(5), 359–366. doi:10.1016/08936080(89)90020-8

Hirsch, G. B., & Henderson, J. D. (1977). A simulation model for “what if?” questions in plan development. American Journal of Health Planning, 2(1), 40–46.

Horsch, A., & Balbach, T. (1999). Telemedicine information systems. IEEE Transactions on Information Technology in Biomedicine, 3, 166–175.

Hirsch, G. B., & Killingsworth, W. R. (1975). A new framework for projecting dental manpower requirements. Inquiry, 12(2), 126–142.

Howgill, M. (1998). Health care consumerism, the information explosion, and branding: Why ’tis better to be the cowboy than the cow. Managed Care Quarterly, 6(4), 33–43.

Hirsch, G. B., & Miller, S. (1974). Evaluating HMO policies with a computer simulation model. Medical Care, 12(8), 668–681. Hirsch, G. B., & Bergan, T. A. (1973). Simulating ambulatory care systems: The relationship between structure and behaviour. In Proceedings of the Summer Computer Simulation Conference, Montreal, Quebec, Canada (pp. 877-884). Hirsch, G. B., Bergan, T. A., & Frohman, A. L. (1974). Organizational change in the provision of human services. In Proceedings of the Summer Computer Simulation Conference, Houston, Texas.

Hripcsak, G. (1993). Monitoring the monitor: Automated statistical tracking of a clinical event monitor. Computers and Biomedical Research, an International Journal, 26(5), 449–466. Hripcsak, G., Austin, J. H., Alderson, P. O., & Friedman, C. (2002). Use of natural language processing to translate clinical information from a database of 889,921 chest radiographic reports. Radiology, 224(1), 157–163.

325

Compilation of References

Hu, P. J., Chau, P. Y. K., Sheng, O. R. L., & Tam, K. Y. (1999). Examining the technology acceptance model using physician acceptance of telemedicine technology. Journal of Management Information Systems, 16(20), 91–112.

Islam, MT, H. Y., Waxman R, Bhuiyan AB. (2006). Implementation of emergency obstetric care training in Bangladesh: lessons learned. Reproductive Health Matters, 14(27), 61–72. doi:10.1016/S0968-8080(06)27229-X

Huff, S., & Munro, M. (1985). Information technology assessment and adoption: A field study. Management Information Systems Quarterly, (December): 327–338.

Jackson, J. R. (2000). The urgent call for knowledge management in medicine. Physician Executive, 26(1), 28–31.

Huffman, M. (2005). Implementing outcome-based homecare: A workbook of OBQI, care pathways and disease management. Jones and Bartlett. Hunter, D. J., & Fairfield, G. (1997). Managed care: Disease management. British Medical Journal, 315, 50–53. IDS-Scheer. (2004). ARIS process performance manager. Retrieved May 24, 2008, from http://www.ids-scheer.com Iluyemi, A., Fitch, C. J., & Parry, D. T. (2007, April 18-20th 2007). Mobile and Wireless Technologies for e-Health Provision in Africa: Opportunities for GSMBased Patient Centric Services. Paper presented at the Med-e-Tel Luxembourg International Telecommunications Union. (2006). Bhutan to be Testbed for ITU’s E-Post Venture with Universal Postal Union. Retrieved 20th April 2006, from http://www.itu.int/newsarchive/ press_releases/2002/10.html IMDB. (2007). How/where do you get your information? How accurate/reliable is it? Retrieved from http://imdb. com/help/show_leaf?infosource. Institute of Medicine. (2001). Crossing the quality chasm: A new health system for the 21st century. Washington, DC: National Academy Press, Committee on Quality of Health Care in America. Internet, A. (2007a). Three-month traffic statistics for wikipedia.org. Retrieved from http://www.alexa.com/ data/details/main?q=&url=wikipedia.org. Internet, A. (2007b). Top Sites United States. Retrieved from http://www.alexa.com/site/ds/top_sites?cc=US&ts_ mode=country&lang=none. Irby, D. M. (1994). What clinical teachers in medicine need to know. Academic Medicine, 69, 333–342. Irby, D. M., Ramsey, P. G., Gillmore, G. M., & Schaad, D. (1991). Characteristics of effective clinical teachers of ambulatory care medicine. Academic Medicine, 66, 54–55.

326

Jakobovits, R. M., Rosse, C., & Brinkley, J. F. (2002). WIRM An Open Source Toolkit for Building Biomedical Web Applications. Journal of the American Medical Informatics Association, 9(6), 557–570. doi:10.1197/ jamia.M1138 Janetzki, V., Allen, M., & Cimino, J. J. (2004). Using natural language processing to link from medical text to online information resources. In Proceedings of Medinfo (pp. 1665). Java Server Pages Documentation. (2007). Retrieved from http://java.sun.com/products/jsp. Jee, K. Y. (2005). Healthcare in ubiquitous era. Jinhan M&B. Jelinek, G. A. (1995). Casemix classification of patients attending hospital emergency departments in Perth, Western Australia: Development and evaluation of an urgency-based casemix system. Perth, Australia: University of Western Australia. Jokipii, P., Kalliovalkama, K., & Paavola, T. (2006). Tekonivelpotilaan hoitoprosessin uudistaminen. [Finnish Medical Journal]. Suomen Lääkärilehti, 61(14), 1626–1627. Joos, D., Chen, Q., Jirjis, J., & Johnson, K. (2006). An electronic medical record in primary care: Impact on satisfaction, work efficiency and clinic processes. Paper presented at the AMIA, Washington, DC. Josefsson, U. (2005). Coping with illness online: The case of patients’ online communities. The Information Society, 21(2), 143–153. Jun, G. C. T. (2007). Design for patient safety: A systematic evaluation of preocess modelling appraoches for healthcare system safety improvement, PhD Thesis. Cambridge: University of Cambridge, Department of Engineering.

Compilation of References

Jun, G. C. T., Ward, J. R., & Clarkson, P. J. (2005). Mapping the healthcare process in order to design for patient safety. In International Conference on Engineering Design (ICED 05) (pp. 353-364). Junginger, M., de Visser, E., Hjort-Gregersen, K., Koornneef, J., Raven, R., Faaij, A., & Turkenburg, W. (2006). Technological learning in bioenergy systems. Energy Policy, 34(18), 4024–4041. doi:10.1016/j.enpol.2005.09.012 Kahen, G., & Sayers, B. M. (1997). Health-care technology transfer: Expert and information systems for developing countries. Methods of Information in Medicine, 36, 69–78. Kalli, S., Ambroso, C., Gregory, R., Heikela, A., Ilomaki, A., & Leaning, M. (1992). Inform: Conceptual modelling of intensive care information systems. International Journal of Clinical Monitoring and Computing, 9, 85–94. Kandaswamy, A., Kumar, C. S., Ramanathan, R. P., Jayaraman, R., & Malmurugan, N. (2004). Neural classification of lung sounds using wavelet coefficients. Computers in Biology and Medicine, 34, 523–537. Kang, S., & Lee, S. (2007). Economic effect and growth strategy of U-Health. SERI (Samsung Economic research Institute). Issue Paper. Kantor, G., Wilson, W., & Midgley, A. (2003). Open source software and the primary care EMR. Journal of the American Medical Informatics Association, 10(6), 616. Kantor, G. S., Wilson, W. D., & Midgley, A. (2003). Open-source software and the primary care EMR. Journal of the American Medical Informatics Association, 10(6), 616. doi:10.1197/jamia.M1403 Karahanna, E., Straub, D. W., & Chervany, N. L. (1999). Information technology adoption across time: a crosssectional comparison of pre-adoption and post-adoption beliefs. Management Information Systems Quarterly, 23, 183–213. doi:10.2307/249751 Karlsson, L. (2003). Conflict or harmony? Individual rights in Swedish and British health care [Konflikt eller harmoni? Individuella rättigheter och ansvarsutkrävande i svensk och brittisk sjukvård]. Doctoral dissertation, Göteborg: CEFOS, Göteborgs universitet.

Kasper, D., Fauci, A., Longo, D., Braunwald, E., Hauser, S., & Jameson, J. (2005). Harrison’s principles of internal medicine (16th ed.). New York, NY: McGraw-Hill. Kaufman, D. R., Patel, V. L., Hilliman, C., Morin, P. C., Pevzner, J., & Weinstock, R. S. (2003). Usability in the real world: assessing medical information technologies in patients’ homes. Journal of Biomedical Informatics, 36, 45–60. doi:10.1016/S1532-0464(03)00056-X Kaur, H., & Wasan, S. K. (2006). Empirical study on Applications in Data Mining Techniques in Healthcare. Journal of Computer Science, 2(2), 194–200. doi:10.3844/ jcssp.2006.194.200 Kayango, M., Esquiche, E., Luna, M. R., Frias, G., VegaCenteno, L., & Bailey, P. (2006). Strengthening emergency obstetric care in Ayacucho, Peru. International Journal of Gynaecology and Obstetrics: the Official Organ of the International Federation of Gynaecology and Obstetrics, 92, 299–307. doi:10.1016/j.ijgo.2005.12.005 Kennedy, R., Lee, Y., & Van Roy, B. (1998). Solving data mining problems through pattern recognition. Prentice Hall. Ker, J. S. (2003). Attributes of trainers for postgraduate training in general surgery–a national consensus. The Surgeon, 1, 215–220. Kerkenbush, N. L., & Lasome, C. E. (2003). The emerging role of electronic diaries in the management of diabetes mellitus. AACN Clinical Issues: Advanced Practice in Acute and Critical Care, 14, 371–378. Kervenoael, R. D., Soopramanien, D., Elms, J., & Hallsworth, A. (2006). Exploring value through integrated service solutions: the case of e-grocery shopping. Managing Service Quality, 16(2), 185–202. doi:10.1108/09604520610650646 Keselman, A., Browne, A., & Kaufman, D. (2008). Consumer health information seeking as hypothesis testing. Journal of the American Medical Informatics Association, 15(4), 484–495. doi:10.1197/jamia.M2449 Kiernan, M., Kraemer, H. C., Winkleby, M. A., King, A. C., & Taylor, C. B. (2001). Do logistic regression and signal detection identify different subgroups at risk? Implications for the design of tailored interventions. Psychological Methods, 6, 35–48.

327

Compilation of References

Kirigia, J., Seddoh, A., Gatwiri, D., Muthuri, L., & Seddoh, J. (2005). E-health: Determinants, opportunities, challenges and the way forward for countries in the WHO African region. BMC Public Health, 5, 137.

Kudyba, S., & Hoptroff, R. (2001). Data mining and business intelligence: A guide to productivity. Hershey, PA: Idea Group.

Kirigia, J., Seddoh, A., Gatwiri, D., Muthuri, L., & Seddoh, J. (2005). E-health: Determinants, opportunities, challenges and the way forward for countries in the WHO African Region. BMC Public Health, 5(1), 137. doi:10.1186/1471-2458-5-137

Kudyba, S., Hamar, G. H., & Gandy, W. M. (2006). Utilising neural network applications to enhance efficiency in the healthcare industry: prediciting populations of future chronic illness. International Journal of Business Intelligence and Data Mining, 1(4), 371–383. doi:10.1504/ IJBIDM.2006.010780

Kohn, L., Corrigan, J., & Donaldson, M. (Eds.). (2000). To err is human: Building a safer health system. Washington: National Academy Press.

Kumar, A. (2005). Ontology-driven access to biomedical information (ODABI). Undergraduate Thesis, University of Arizona.

Kohonen, T. (1995). Self-organizing maps. Berlin, Germany: Springer.

Kumarapeli, P., De Lusignan, S., Ellis, T., & Jones, B. (2007). Using Unified Modelling Language (UML) as a process-modelling technique for clinical-research process improvement. Medical Informatics and the Internet in Medicine, 32, 51–64.

Korean National Insurance Corporation (2007), Daily Report, Jan/29. Korenbrot, C. C., Gill, A., & Patterson, E. (1995). Evaluation of California’s statewide implementation of enhanced perinatal services as Medicaid benefits. Public Health Reports (Washington, D.C.), 110(2), 125–133. Kotonya, G., & Sommerville, I. (1998). Requirements engineering: Processes and techniques. New York: John Wiley. Kovalerchuk, B., Vityaev, E., & Ruiz, J. F. (2001). Consistent and complete data and expert mining in medicine. In Cios, K. J. (Ed.), Medical data mining and knowledge discovery. New York: Physica-Verlag. Kowalke, M. (2006) TMCnet. Wireless Mobility Blog. (2006). RFID vs. WiFi for Hospital Inventory Tracking Systems. http://blog.tmcnet.com/wireless-mobility/rfidvs-wifi-for-hospital-inventory-tracking-systems.asp, October 23, 2006. Kozinets, R. (2002). The field behind the screen: Using netnography for marketing research in online communications. JMR, Journal of Marketing Research, 39(1), 61–72. doi:10.1509/jmkr.39.1.61.18935 Kudyba, S., Hamer, B., & Gandy, W. (2005, December). Enhancing efficiency in the health care industry. Communications of the ACM, 28(12), 107–110.

328

Kuperman, G. J., Gardner, R. M., & Pryor, T. A. (1991). The HELP system. New York: Springer-Verlag. Kvale, S. (1996). Interviews: An introduction to qualitative research interviewing. Thousand Oaks, CA: Sage. Kwak, N. K., McCarthy, K. J., & Parker, G. E. (1997). A human resource planning model for hospital/medical technologists: An analytic hierarchy process approach. Journal of Medical Systems, 21(3), 173–187. doi:10.1023/A:1022812322966 Lagor, C., Aronsky, D., Fiszman, M., & Haug, P. (2001). Automatic identification of patients eligible for a pneumonia guideline: Comparing the diagnostic accuracy of two decision support models. In Proceedings of the 10th World Congress on Medical Informatics. Medinfo, 10, 493–497. Lahiri, K., & Xing, G. (2004). An econometric analysis of veterans’ healthcare utilization using two-part models. Empirical Economics, 29, 431–449. doi:10.1007/ s00181-003-0178-0 Lahiri, S. (2005), RFID Sourcebook, IBM Press, USA, (ISBN: 10-0131851373) Last accessed on 22/11/2007 Landro, L. (2002, January 4). Technology programs make disease management easier. The Wall Street Journal, Health Journal.

Compilation of References

Landstingsförbundet. (2000a). Recommendation about choice in health care. [Rekommendation om valmöjligheter inom hälso- och sjukvården]. Dnr Lf924/00. Stockholm: Landstingsförbundet.

Leutzinger, J. A., Ozminkowski, R. J., & Dunn, R. L. (2000). Projecting future medical care costs using four scenarios of lifestyle risk rates. American Journal of Health Promotion, 15, 35–44.

Landstingsförbundet. (2000b). Waiting times in care on the Internet [Väntetider i vården på Internet. Pressmeddelande] 2000-04-10. Stockholm: Landstingsförbundet.

Levin, G., Hirsch, G. B., & Roberts, E. (1972). Narcotics and the community: A system simulation. American Journal of Public Health, 62, 8–6.

Langer, A. M. (2008). Analysis and design of information systems. London: Springer (pp. 418; p. 249 illus.).

Levin, G., & Roberts, E. B. (1976). The dynamics of human service delivery. Cambridge, MA: Ballinger Publishing Company.

Larson, P. A., & Janower, M. L. (2005). The Nighthawk: Bird of Paradise or Albatross? Journal of the American College of Radiology, 2(12), 967–970. doi:10.1016/j. jacr.2005.08.002 Law, B. (1999). Career-learning space: newDOTS thinking for careers education. British Journal of Guidance & Counselling, 27(1), 35–55.

Lewan, T. (2007). ZDNet Australia. Human RFID chipping invading the US?. 23 July 2007. Li, T., Wu, G., & Gruidl, Z. (2004). Data mining techniques for cancer detection using serum proteomic profiling. Artificial Intelligence in Medicine, 32(2), 71–83.

Le Grand, J., & Bartlett, W. (1993). Introduction. In Le Grand, J., & Bartlett, W. (Eds.), Quasi-markets and social policy (pp. 1–12). London: Macmillan.

Likourezos, A., Chalfin, D., Murphy, D., Sommer, B., Darcy, K., & Davidson, S. (2004). Physician and nurse satisfaction with an electronic medical record system. The Journal of Emergency Medicine, 27(4), 419–424.

LeBozec, C., Jaulent, M. C., Zapletal, E., & Degoulet, P. (1998). Unified modeling language and design of a case-based retrieval system in medical imaging. In AMIA Annual Symposium Proceedings (pp. 887-891).

Lim, H., Widdows, R., & Jungkun, J. (2006). Mloyalty: winning strategies for mobile carriers. Journal of Consumer Marketing, 23(4/5), 208–218. doi:10.1108/07363760610674338

LeCompte, M. D., & Preissle, J. (1993). Ethnography and qualitative design in educational research (2nd ed.). San Diego, CA: Academic Press.

Lindberg, D. (2003). NIH: Moving research from the bench to the bedside. Presentation to the Subcommittee on Health.

Lee, B., & Menon, N. (2000). Information technology value through different normative lenses. Journal of Management Information Systems, 16(4), 99–119.

Linder, J., Schnipper, J., Tsurikova, R., Melnikas, A., Volk, L., & Middleton, B. (2006). Barriers to electronic health record use during patient visits. Paper presented at the AMIA, Washington, DC.

Lee, N. J., Bakken, S., & John, R. (2006). Functional requirements specification and data modeling for a PDA-based decision support system for the screening and management of obesity. In AMIA Annual Symposium Proceedings (pp. 1002). Leider, H. L., & Krizan, K. (2001). Disease management— a great concept, but can you implement it? Disease Management, 4, 111–119. Lempp, H., & Seale, C. (2004). The hidden curriculum in undergraduate medical education: Qualitative study of medical students’ perceptions of teaching. British Medical Journal, 320, 770–773.

Lisboa, P. J., & Taktak, A. F. G. (2006). The use of artificial neural networks in decision support in cancer: A systematic review. Neural Networks, 19, 408–415. doi:10.1016/j.neunet.2005.10.007 Littlejohns, P., Wyatt, J., & Garvican, L. (2003). Evaluating computerised health information systems: hard lessons still to be learnt. British Medical Journal, 326, 860–863. doi:10.1136/bmj.326.7394.860

329

Compilation of References

Littlejohns, P., Wyatt, J. C., & Garvican, L. (2003). Evaluating computerised health information systems: hard lessons still to be learnt. British Medical Journal, 326(7394), 860–863. doi:10.1136/bmj.326.7394.860

Martin, J. L., Norris, B. J., Murphy, E., & Crowe, J. A. (2008). Medical device development: The challenge for ergonomics. Applied Ergonomics, 39, 271–283. doi:10.1016/j.apergo.2007.10.002

Longman, P. (2007). Best care anywhere: How America’s most socialized healthcare system became its finest: What the VA miracle means for you and your country. Sausalito, CA: PoliPointPress.

Maruster, L., van der Aalst, W. M. P., Weijters, T., van den Bosch, A., & Daelemans, W. (2001, October 25-26). Automated discovery of workflow models from hospital data. In B. Kröse et al. (Eds.), Proceedings 13th Belgium- Netherlands Conference on Artificial Intelligence (BNAIC’01), Amsterdam, The Netherlands (pp. 183-190).

Lunn, K., Sixsmith, A., Lindsay, A., & Vaarama, M. (2003). Traceability in requirements through process modelling, applied to social care applications. Information and Software Technology, 45, 1045–1052. Lyalin, D., & Williams, W. (2005). Modeling cancer registration processes with an enhanced activity diagram. Methods of Information in Medicine, 44, 11–13. Lyman, P., & Varian, H. (2003). How much information? Retrieved from http://www2.sims.berkeley.edu/research/ projects/how-much-info-2003/index.htm. Maass, W. (1994). Efficient agnostic pac-learning with simple hypothesis. Madden, M. (2003). America’s online pursuits: the changing picture of who’s online and what they do. Washington DC: Pew Internet and American Life Project. Retrieved from www.pewinternet.org/reports/toc.asp?Report=106 Magrabi, F., Lovell, N. H., & Celle, B. G. (1999). A webbased approach for electrocardiogram monitoring in the home. International Journal of Medical Informatics, 54, 1145–1153. doi:10.1016/S1386-5056(98)00177-4 Mahapatra, S., Koelling, C. P., Patvivatsiri, L., et al. (2003). Pairing emergency severity index 5-level triage data with computer aided system design to improve emergency department access and throughput. In The 2003 Winter Simulation Conference. New Orleans, LA: INFORMS. Maij, E., Toussaint, P. J., Kalshoven, M., Poerschke, M., & Zwetsloot-Schonk, J. H. (2002). Use cases and DEMO: Aligning functional features of ICT-infrastructure to business processes. International Journal of Medical Informatics, 65, 179–191. Mann, C., & Steward, F. (2000). Internet communication and qualitative research: A handbook for researching online. Thousand Oaks, CA: Sage.

330

Mascarenhas, L., Eliot, B. W., & MacKenzie, I. Z. (1992). A comparison of perinatal outcome, antenatal and intrapartum care between England and Wales, and France. British Journal of Obstetrics and Gynaecology, 99(12), 955–958. McDonald, C., Schadow, G., Barnes, M., Dexter, P., Overhage, J., & Mamlin, B. (2003). Open source software in medical informatics—why, how and what. International Journal of Medical Informatics, 69(2-3), 175–184. McDonald, C. J., Schadow, G., Barnes, M., Dexter, P., Overhage, J. M., & Mamlin, B. (2003). Open Source software in medical informatics--why, how and what. International Journal of Medical Informatics, 69(2-3), 175–184. doi:10.1016/S1386-5056(02)00104-1 McGrath, J. E. (1982). Judgment Calls in Research. Beverly Hills, CA: Sage Publications, Inc. McGraw, G. (2006). Software Security. USA: AddisonWesley. McKeown, T. (1971). A Historical Appraisal of the Medical Task. In McLachlan, G., & McKeown, T. (Eds.), Medical History and Medical Care (pp. 29–50). London: Oxford University Press. McLaughlin, C., Yang, S., & Van Dierdonck, R. (1995). Professional service organizations and focus. Management Science, 14(7), 1185–1193. McMurray, J., & Rankin, R. (1994). Cardiology-I: Treatment of myocardial infarction, unstable angina, and angina pectoris. British Medical Journal, 309(6965), 1343–1350. Mechaca, L. (2007). Goodbye, Axim. Message posted to http://direct2dell.com/one2one/archive/2007/04/11/11397.aspx.

Compilation of References

Medsker, L., & Liebowitz, J. (1994). Design and Development of Expert Systems and Neural Networks. New York: McMillan. Mehta, N. Y., Haluck, R. S., Frecker, M. I., & Snyder, A. J. (2002). Sequence and task analysis of instrument use in common laparoscopic procedures. Surgical Endoscopy, 16, 280–285. Mendenhall, W. M. (2004). Mandibular osteoradionecrosis. Journal of Clinical Oncology, 22(24), 4867–4868. doi:10.1200/JCO.2004.09.959 Menges, R., & Austin, A. (2001). Teaching in higher education. In Richardson, V. (Ed.), Handbook of research on teaching (4th ed.). Washington, DC: American Educational Research Association. Mertins, K., Bernus, P., & Schmidt, G. (1998). Handbook on architectures of information systems. Berlin, Germany: Springer.

Montani, S., & Bellazzi, R. (2002). Supporting decisions in medical applications: The knowledge management perspective. International Journal of Medical Informatics, 68, 79–90. Morahan-Martin, J. M. (2004). How internet users find, evaluate, and use online health information: a cross-cultural review. Cyberpsychology & Behavior, 7(5), 497–510. Morgan, N. A., & Rego, L. L. (2006). The value of different customer satisfaction and loyalty metrics in predicting business performance. Marketing Science, 25(5), 426–439. doi:10.1287/mksc.1050.0180 Morrison, E. H., Hitchcock, M. A., Harthill, M., Boker, J. R., & Masunaga, H. (2005). The online clinical teaching perception inventory: A “snapshot” of medical teachers. Family Medicine, 37, 48–53. Muney, A. M. (2002). Evidence-based medicine needs to be promoted more vigorously. Managed Care (Langhorne, Pa.), 11, 10–12.

Messmer, E. (2006). Network World. Hospital finds a bloody good RFID application. http://www.computerworld.com.au/index.php/id;2117167769;fp;4;fpid;18.

Mureo, M., & Rice, R. (2006). The Internet and health care. Theory, research, and practice. NJ: Lawrence Erlbaum.

Metropolitan Health and Aged Care Services Division of the Victorian State Government. (2007). Victorian emergency minimum dataset (VEMD). Retrieved May 24, 2008, from http://www.health.vic.gov.au/hdss/ vemd/2004-05/manual/index.htm

Murray, J. P. (1999). Faculty development in a national sample of community colleges. Community College Review, 27(3), 47. Retrieved January 29, 2008, from Questia database: http://www.questia.com/ PM.qst?a=o&d=5001886124

Mickey, K. (2004). RFID grew in 2002. Traffic World, 5, 20–21.

MySQL. (2007). http://www.mysql.com.

Miller, R., & Sim, I. (2004). Physicians’ use of electronic medical records: Barriers and solutions. Health Affairs, 23(2), 116–126. Misson, J. C. (2001). A review of clinical risk management. Journal of Quality in Clinical Practice, 21, 131–134. doi:10.1046/j.1440-1762.2001.00421.x Mockus, A., Fielding, R. T., & Herbsleb, J. (2000). A case study of open source software development: the Apache server. Paper presented at the Software Engineering, 2000. Proceedings of the 2000 International Conference on. Momjian, B. (2001). PostgreSQL: Introduction and concepts. Addison-Wesley Longman Publishing Co., Inc.

MySQL Connector/J. (2007). Retrieved from http://www. mysql.com/products/connector-j/. Nagy, P. (2007). Open source in imaging informatics. Journal of Digital Imaging, 20(0), 1–10. Najumnissa, D. & ShenbagaDevi, S. (2008). Intelligent identification and classification of epileptic seizures using wavelet transform. International Journal of Biomedical Engineering and Technology, 1(3), 293–314. doi:10.1504/ IJBET.2008.016963 Nason, C. S., Alexander, G. R., Pass, M. A., & Bolland, J. M. (2001). An evaluation of a Medicaid managed maternity program: The impact of comprehensive care coordination on utilization and pregnancy outcome. Journal of Health and Human Resources Administration, 24(3/4), 493–521.

331

Compilation of References

Neilson, J. P. (1998). Ultrasound for fetal assessment in early pregnancy. Cochrane Database of Systematic Reviews, 4.

Odell, J. (Ed.). (2000). Agent Technology, OMG Document 00-09-01, OMG Agents interest. Group, (September): 2000.

Nelson, R., & Ball, M. (2004). Consumer informatics. Applications and strategies in cyber health care. Berlin: Springer.

Odom, M., Kumar, A., & Saunders, L. (2002). Web assurance seals: how and why they influence consumers’ decisions. Journal of Information Systems, 16(2), 231–250. doi:10.2308/jis.2002.16.2.231

Ngwenyama, O., Guergachi, A., & McLaren, T. (2007). Using the learning curve to maximize IT productivity: a decision analysis model for timing software upgrades. International Journal of Production Economics, 105(2), 524–535. doi:10.1016/j.ijpe.2006.02.013 Norén, L., & Ranerup, A. (2005). The Internet Web portal as an enrolment device. In Czarniawska, B., & Hernes, T. (Eds.), Actor-network theory and organizing (pp. 188–207). Malmö: Liber and Copenhagen Business School Press. Norris, J. (2004). Mission-critical development with open source software: Lessons learned. Software, IEEE, 21(1), 42–49. Nydick, R. L., & Hill, R. P. (1992). Using the analytic hierarchy process to structure the supplier selection procedure. Journal of Purchasing and Materials Management, 25(2), 31–36. O’Connell, R., Cho, C., Shah, N., Brown, K., & Shiffman, R. (2004a). Take note: Differential EHR satisfaction with two implementations under one roof. Journal of the American Medical Informatics Association, 11(1), 43–49. O’Connell, R., Cho, C., Shah, N., Brown, K., & Shiffman, R. (2004b). Take note: Differential EHR satisfaction with two implementations under one roof. Journal of the American Medical Informatics Association, 11(1), 43–49. O’Connor, P. J., Sperl-Hillen, J. M., Pronk, N. P., & Murray, T. (2001). Primary care clinic-based chronic disease care. Features of successful programs. Disease Management and HealthCare Outcomes, 9, 691–698.

OECD. (1980). Guidelines on the protection of privacy and transborder flows of personal data. Retrieved January 29, 2008, from http://www.oecd.org/document/18/0,234 0,en_2649_34255_1815186_1_1_1_1,00.html Oh, H., Rizo, C., Enkin, M., & Jadad, A. (2005). What is e-health (3): A systematic review of published definitions. Journal of Medical Internet Research, 7(1), article e1. Okoroh, M. I., Ilozor, B. D., & Gombera, P. (2007). An artificial Neural Networks approach in managing healthcare. International Journal of Electronic Healthcare, 3(3), 279–292. doi:10.1504/IJEH.2007.014548 Oliver, J., & Srivastava, S. (1999). The Big Five taxonomy: History, measurement, and theoretical perspectives. In Pervin, L., & John, O. P. (Eds.), Handbook of personality: Theory and research (2nd ed.). New York: The Guilford Press. Oliver, J. (1993, August). Shocking to the core. Management Today, 18–22. OLPC. (2006). One laptop per Child. Retrieved 1st December 2006, 2006, from http://laptop.org/ Oser, F., & Baeriswyl, F. (2001). Choreographies of teaching: Bridging instruction to learning. In Richardson, V. (Ed.), Handbook of research on teaching (4th ed.). Washington, DC: American Educational Research Association. Østergren, K. (2004). The institutional construction of consumerism. A study of implementing quality indicators (Working paper 15). Bergen: Stein Rokkan Centre for Social Studies.

Obama, B. (2008). Remarks of president-elect barack obama: Radio address on the economy. Retrieved June 27, 2009, from whitehouse.gov

Ozcan, Y. A. (2005). Quantitative methods in health care management. San Francisco, CA: Jossey-Bass (Wiley).

Object Management Group. (2005). Unified Modeling Language: Superstructure. Version 2.0. OMG Inc.

Paavola, T. (2007). Justifying IT system investments. International Journal of Healthcare Technology and Management, 8(1/2), 97–106. doi:10.1504/IJHTM.2007.012104

332

Compilation of References

Paavola, T. (2008). Exploring IT system benefits in health care. Tampere University of Technology Publication 756. Tampere: Tampereen Yliopistopaino. Paccagnella, L. (1997). Getting the seats of your pants dirty: Strategies for ethnographic research on virtual communities. Journal of Computer-Mediated Communication, 3(1). Retrieved from http://ascusc.org/jcmc/vol3/issue1/ paccagnella.html. Pagliari, C., Sloan, D., Gregor, P., Sullivan, F., Detmer, D., & Kahan, P. J. (2005). What Is eHealth (4): A Scoping Exercise to Map the Field. Journal of Medical Internet Research, 7(1), e9. doi:10.2196/jmir.7.1.e9 Pal, A., Mbarika, V. W. A., Cobb-Payton, F., Datta, P., & McCoy, S. (2005). Telemedicine diffusion in a developing Country:The case of India (march 2004). Information Technology in Biomedicine. IEEE Transactions on, 9(1), 59–65. Palier, B. (2005). How to understand the differences between various health care systems [Att förstå skillnadena prestation mellan olika hälso- och sjukvårdssystem]. Stockholm: Sveriges kommuner och Landsting. Paneth, N. S. (1995). The Problem of Low Birth Weight. The Future of Children: Low Birth Weight, 5(1), 21–34. Park, K. (2003). Background to Ubiquitous Healthcare. No.272 Conference between Industry, Academic, Research Centers, Korea. Parry, E. C., Sood, R., & Parry, D. T. (2006). Investigation of optimization techniques to prepare ultrasound images for electronic transfer [Abstract]. Ultrasound in Obstetrics and Gynecology, 28(4), 487-488(482). Patton, M. (2002). Qualitative research and evaluation methods (3rd ed.). Thousand Oaks, CA: Sage Publications. Paukert, J. L., & Richards, B. F. (2000). How medical students and residents describe the roles and characteristics of their influential clinical teachers. Academic Medicine, 75, 843–845. Peckham, S. (2002). Book review. Health & Social Care in the Community, 10(6), 520–522.

Pehlivanli, A. C., Ibrikci, T., & Ersoy, O. K. (2008). Consensual classification of drug and nondrug compounds. International Journal of Computational Biology and Drug Design, 1(3), 224–234. doi:10.1504/ IJCBDD.2008.021416 Penrod, L., & Gadd, C. (2001). Attitudes of academicbased and community-based physicians regarding EMR use during outpatient encounters. Proceedings/AMIA Annual Symposium, 528-532. Perednia, D., & Allen, A. (1995). E-medicine technology and clinical applications. Journal of the American Medical Association, 273(6), 483–488. Perry, T. (2007), December). Identification and prediction of chronic conditions for health plan members using data mining techniques: In data mining methods and applications. Taylor Francis. Pham, D. T., Wang, Y., & Dimov, S. (2005). Intelligfent manufacturing strategy selection. In Proc 1st International Virtual Conference on Intelligent Production Machines and Systems (pp. 312-318). Oxford: Elsevier. Pham, D. T., Wang, Y., & Dimov, S. (2006). Incorporating delay and feedback in intelligent manufacturing strategy selection. In Proc 2nd International Virtual Conference on Intelligent Production Machines and Systems (pp. 246-252). Oxford: Elsevier. Pinsky, L. E., Monson, D., & Irby, D. M. (1998). How excellent teachers are made: Reflecting on success to improve teaching. Advances in Health Science Education, 3, 207–215. Pizziferri, L., Kittler, A., Volk, L., Honour, M., Gupta, S., & Wang, S. (2005). Primary care physician time utilization before and after implementation of an electronic medical record: A time-motion study. Journal of Biomedical Informatics, 38, 176–188. Pohjonen, H., Bondestam, S., Karp, P., Kinnunen, J., Korhola, O., & Sakari, P. (1994). Quality assurance of the bedside chest radiography process. Acta Radiologica, 35, 235–243.

Pedersen, K., Christiansen, T., & Bech, M. (2005). The Danish health care system: Evolution–not revolution–in a decentralized system. Health Economics, 14, 41–57.

333

Compilation of References

Poonguzhali, S., & Ravindran, G. (2007). Evaluation of feature extraction methods for classification of liver abnormalities in ultrasound images. International Journal of Biomedical Engineering and Technology, 1(2), 134–143. doi:10.1504/IJBET.2007.015856 Porter, S. C., Cai, Z., & Gribbons, W. (2004). The asthma kiosk: A patient-centered technology for collaborative decision support in the emergency department. Journal of the American Medical Informatics Association, 11, 458–467. Poulymenopoulou, M., Malamateniou, F., & Vassilacopoulos, G. (2003). Specifying workflow process requirements for an emergency medical service. Journal of Medical Systems, 27(4), 325–335. Pramatari, K. C., Doukidis, G. I., & Kourouthanassis, P. (2005). Towards ‘smarter’ supply and demand-chain collaboration practices enabled by RFID technology. In Vervest, P., Van Heck, E., Preiss, K., & Pau, L. F. (Eds.), Smart Business Networks (pp. 197–210). Germany: Springer. Prather, J. C., Lobach, D. F., Goodwin, L. K., Hales, J. W., Hage, M. L., & Hammond, W. E. (1997). Medical data mining: Knowledge discovery in a clinical data warehouse. In Proceedings of the AMIA Annual Symposium Fall 1997 (pp. 101-105).

Rada, R. (2005a). Obstructive sleep apnea and head and neck neoplasms. Otolaryngology - Head and Neck Surgery, 132(5), 794–799. doi:10.1016/j.otohns.2004.12.002 Rada, R., & Hu, K. (2002). Patterns in student–student commenting. IEEE Transactions on Education, 45(3), 262–267. doi:10.1109/TE.2002.1024619 Rada, R. (2007). Sleep and quality of life in head and neck neoplasm. In Verster, J. C., Pandi-Perumal, S. R., & Streiner, D. L. (Eds.), Sleep and quality of life in head and neck neoplasm. Totowa, New Jersey: Humana Press. Rada, R. (2005b). Software for patient-patient discussions. Montreal Conference on E-Technologies 2005, Montreal, Canada (pp. 205-209). Rada, R. (2006). Membership and online groups. e-Society 2006: IADIS International Conference, Dublin, Ireland (pp 290-293). Rades, D., Fehlauer, F., Bajrovic, A., Mahlmann, B., Richter, E., & Alberti, W. (2004). Serious adverse effects of amifostine during radiotherapy in head and neck cancer patients. Radiotherapy and Oncology, 70, 261–264. doi:10.1016/j.radonc.2003.10.005 Raento, M., Oulasvirta, A., Petit, R., & Toivonen, H. (2005). ContextPhone: A Prototyping Platform for Context-Aware Mobile Applications, 4, 51-59.

Preece, J., Nonnecke, B., & Andrews, D. (2004). The top five reasons for lurking: Improving community experiences for everyone. Computers in Human Behavior, 20(2), 201–223. doi:10.1016/j.chb.2003.10.015

Raghupathi, W. (2006). Designing clinical decision support systems in health care: A systemic approach. International Journal of Healthcare Information Systems and Informatics, 2(1).

Prichard, K. W., & Sawyer, R. M. (Eds.). (1994). Handbook of college teaching: Theory and applications. Westport, CT: Greenwood Press. Retrieved January 29, 2008, from Questia database: http://www.questia.com/ PM.qst?a=o&d=9413558

Raghupathi, W., & Gao, W. (2007). An eclipse-based development approach to health information technology. International Journal of Electronic Healthcare, 3(4), 433–452.

Qiu, R., & Sangwan, R. (2005). Toward collaborative supply chains using RFID. CIO Wisdom II, more best practices (pp. 127–144). New Jersey: Prentice-Hall. Quinlan, J.R. (1993). C4.5: Programs for machine learning. San Mateu: Morgan Kaufmann series in machine learning.

334

Ranerup, A. (2006). Electronic markets connecting citizens to pension reform. Electronic Markets, 16(4), 282–291. Rennecker, J. (2004). Updating ethnography to investigate contemporary organisational forms. In M.E. Whitman & A.B. Woszczynski (Eds.), The handbook of information systems research (pp. xi, 349). Hershey, PA: IGI Global.

Compilation of References

Rhodes, B., & Mase, K. (2006). Wearables in 2005. IEEE Pervasive Computing / IEEE Computer Society [and] IEEE Communications Society, 5(1), 92–95. doi:10.1109/ MPRV.2006.21 Rialle, V., Lamy, J. B., Noury, N., & Bajolle, L. (2003). Telemonitoring of patients at home: a software agent approach. Computer Methods and Programs in Biomedicine, 72, 257–268. doi:10.1016/S0169-2607(02)00161-X Richardson, G. P. (1999). Citation for winner of the 1999 Jay W. Forrester Award: Jac Vennix. System Dynamics Review, 15(4), 375–377. Richardson, G. P., & Andersen, D. F. (1995). Teamwork in group model building. System Dynamics Review, 11(2), 113–137.

Rosenbaum, E. E. (1988). A taste of my own medicine, when the doctor is the patient. New York: Random House. Roux-Rouquie, M., Caritey, N., Gaubert, L., & Rosenthal, S. (2004). Using the Unified Modelling Language (UML) to guide the systemic description of biological processes and systems. Bio Systems, 75, 3–14. Rozinat, A., & van der Aalst, W. M. P. (2008). Conformance checking of processes based on monitoring real behavior. Information Systems Frontiers, 33(1), 64–95. Rubel, P., Fayn, J., Nollo, G., Assanelli, D., Li, B., & Restier, L. (2005). Toward personal ehealth in cardiology. Results from the EPI-MEDICS telemedicine project. Journal of Electrocardiology, 38(4), 100–106.

Richardson, G. P. (1996). System dynamics. In Gass, S. I., & Harris, C. M. (Eds.), Encyclopedia of operations research and management science (pp. 656–660). Boston, MA: Kluwer Academic Publishers.

Rubel, P., Fayn, J., Simon-Chautemps, L., Atoui, H., Ohlsson, M., & Télisson, D. (2004). New paradigms in telemedicine: Ambient intelligence, wearable, pervasive and personalized. Studies in Health Technology and Informatics, 108, 123–132.

Richardson, V. (2001). (Ed.). Handbook of research on teaching (4th ed). Washington, DC: American Educational Research Association.

Rucker, D. (2003). Finally, a tool to re-engineer health care: The workflow engine. San Francisco, CA: Montgomery Research, Inc.

Ridinger, M. H., & Rice, J. J. (2000). Predictive modeling points way to future risk status. Health Management Technology, 21, 10–12.

Russell, S. (2004). The economic burden of illness for households in developing countries: a review of studies focusing on malaria, tuberculosis, and human immunodeficiency virus/acquired immunodeficiency syndrome. Am J Trop Med Hyg, 71(2_suppl), 147-155.

Riska, E. (1982). Health education and its ideological content. Acta Sociologica, 25, 41–46. Roberts, E. B., & Hirsch, G. B. (1976, Winter). Strategic modelling for health care managers. Health Care Management Review, ▪▪▪, 69–77. Romano, J. A., Lam, D. M., Moses, G. R., Gilbert, G. R., & Marchessault, R. (2006). The Future of Military Medicine Has Not Arrived Yet, but We Can See It from Here. Telemedicine and e-Health, 12(4), 417-425. Rombaux, P., Hamoir, M., Plouin-Gaudon, I., Liistro, G., Aubert, G., & Rodenstein, D. (2000). Obstructive sleep apnea syndrome after reconstructive laryngectomy for glottic carcinoma. European Archives of Oto-Rhino-Laryngology, 257(9), 502–506. doi:10.1007/s004050000267 Ropka, M. E. (2002). Symptom status and functional status outcomes: Humanistic outcomes in obesity disease management. Obesity Research, 10, 42S–49S.

Rygh & Mörland. (2006). Hunting for good quality indicators [Jakten på de gode kvalitetsindikatorerne]. Tidsskrift for Den Norske Laegeforening, 21, 2822–2825. Rygielski, C., Wang, J.-C., & Yen, D. C. (2002). Data mining techniques for customer relationship management. Technology in Society, 24, 483–502. Saaty, T. L. (1977). A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology, 15, 234–281. doi:10.1016/0022-2496(77)90033-5 Saaty, T. L. (1996). The analytic hierarchy process. Pittsburgh: RWS Publications. Saboor, S., Ammenwerth, E., Wurz, M., & ChimiakOpoka, J. (2005). MedFlow - improving modelling and assessment of clinical processes. Studies in Health Technology and Informatics, 116, 521–526.

335

Compilation of References

Sackett, D. (1997). Using evidence-based medicine to help physicians keep up-to-date. The Journal for the Serials Community, 9, 178–181. Sackett, D., & Straus, S. (1998). Finding and applying evidence during clinical rounds: The evidence cart. Journal of the American Medical Association, 280, 1336–1338. Samanta, B., & Nataraj, C. (2008). Automated diagnosis of cardiac state in healthcare systems using computational intelligence. International Journal of Services Operations and Informatics, 3(2), 162–177. doi:10.1504/ IJSOI.2008.019331 Sandberg, W. S., Daily, B., Egan, M., Stahl, J. E., Goldman, J. M., Wiklund, R. A., & Rattner, D. (2005). Deliberate perioperative systems design improves operating room throughput. Anesthesiology, 103(2), 406–418. doi:10.1097/00000542-200508000-00025 Sandelowski, M. (1995). Focus on qualitative methods: Sample size in qualitative research. Research in Nursing & Health, 18, 179–183. Sanger, L. (2001). Let’s make a wiki. Message posted to http://web.archive.org/web/20030414014355/http:// www.nupedia.com/pipermail/nupedia-l/2001-January/000676.html. Sarnikar, S. & Gupta, A. (2007). A context-specific mediating schema approach for information exchange between heterogeneous hospital systems. Forthcoming in International Journal of Healthcare Technology and Management. Sarnikar, S., Zhao, J., & Gupta, A. (2005). Medical information filtering using content-based and rule-based profiles. Proceedings of the AIS Americas Conference on Information Systems (AMCIS 2005), Omaha, NE. Scarsbrook, A. (2007). Open source software for radiologists: A primer. Clinical Radiology, 62(2), 120–130. Schafer, D. R., & Dillman, D. A. (1998). Development of a standard e-mail methodology: Results of an experiment. Public Opinion Quarterly, 62, 378–397. Schaffer, R., & Thomson, H. (1992). Successful change programs begin with results. Harvard Business Review, 22(1), 80–89.

336

Scheer, A. W. (1999). Business process frameworks. Berlin, Germany: Springer. Schubart, J. R., & Einbinder, J. S. (2000). Evaluation of a data warehouse in an academic health sciences center. International Journal of Medical Informatics, 60, 319–333. Schuler, D., & Namioka, A. (1993). Participatory design: Principles and practices. Hillsdale, NJ: L. Erlbaum Associates. Schulman, E. F., Sheriff, D. J., & Momany, E. (1997). Primary Case Management and Birth Outcomes in the Iowa Medicaid Program. American Journal of Public Health, 87(1), 80–84. doi:10.2105/AJPH.87.1.80 Schuster, E. W., Allen, S. J., & Brock, D. L. (2007) Global RFID: the value of the EPCglobal network for supply chain management. Berlin; New York: Springer, (ISBN: 9783540356547). Schwartz, B. (2004). The paradox of choice. New York, NY: HarperCollins Publishers. Schwethelm, B., Margolis, L. H., Miller, C., & Smith, S. (1989). Risk Status and Pregnancy Outcome among Medicaid Recipients. American Journal of Preventive Medicine, 5(3), 157–163. Scott, J., Rundall, T., Vogt, T., & Hsu, J. (2005). Kaiser Permanente’s experience of implementing an electronic medical record: A qualitative study. BMJ (Clinical Research Ed.), 331, 1313–1316. Seaman, C. B. (1999). Qualiative methods in empircal studies of software engineering. Software Engineering. IEEE Transactions on, 25(4), 557–572. Searle, N. S., Hatem, C. J., Perkowski, L., & Wilkerson, L. (2006). Why invest in an educational fellowship program? Academic Medicine, 81, 936–940. Seidel, J. (1991). Method and madness in the application of computer technology to qualitative data analysis. In Fielding, N., & Lee, R. M. (Eds.), Using computers in qualitative research (pp. 107–116). London: Sage. Seltsikas, P. (2001). Organizing the information management process in process-based organizations. In 34th Hawaii Int. Conf. Syst. Sci., Hawaii.

Compilation of References

Senge, P. (1994). The fifth discipline: The art and practice of the learning organization. New York, NY: Doubleday/ Currency. Senge, P., Kleiner, A., Roberts, C., Ross, R., & Smith, B. (1994). The fifth discipline fieldbook. New York: Doubleday. Seomun, G., Chang, S. O., Lee, S. J., Kim, I. A., & Park, S. (2006). A Prediction Model for Patient Classification according to Nursing Need: Using Data Mining Techniques. In Park, H., Murray, P., & Delaney, C. (Eds.), ConsumerCentered Computer-Supported Care for Healthy People. Fairfax, VA: IOS Press. Shaalana, K. El-Badryb,M. and Rafeac, A. (2004), “A multiagent approach for diagnostic expert systems via the internet”, Expert Systems with Applications, Volume No. 27 Issue No. 1, Elsevier Publishing, USA pp1-10. Sharma, R. (2005). Automatic integration of text documents in the medical domain. Undergraduate Thesis, University of Arizona. Shepard, S. (2005). RFID: radio frequency identification, New York: McGraw-Hill, (ISBN: 0071442995). Silber, D. (2004). The case for e-health. In Iakovidis, T., Wilson, P., & Healy, J. (Eds.), E-health (pp. 3–27). The Netherlands: IOS Press. Simon, J. S., Rundall, T. G., & Shortell, S. M. (2007). Adoption of Order Entry with Decision Support for Chronic Care by Physician Organizations. Journal of the American Medical Informatics Association, 14, 432–439. doi:10.1197/jamia.M2271 Sinreich, D., & Marmor, Y. (2004). A STAFFING emergency. Industrial Engineering (American Institute of Industrial Engineers), 36(6). Sittig, D., Kuperman, G., & Fiskio, J. (1999). Evaluating physician satisfaction regarding user interactions with an electronic medical record system. Paper presented at the AMIA. SKL & Socialstyrelsen. (2006). Open comparisons of quality and efficiency in health care. A comparison between county councils in 2007 [Öppna jämförelser av hälso- och sjukvårdens kvalitet och efektivitet. Jämförelser melan landsting 2007]. Stockholm: SKL, Socialstyrelsen.

SKL & Socialstyrelsen. (2007). Open comparisons of quality and efficiency in health care. A comparison between county councils in 2007 [Öppna jämförelser av hälso- och sjukvårdens kvalitet och efektivitet. Jämförelser melan landsting 2007]. Stockholm: SKL, Socialstyrelsen. Slaughter, L. A., Harper, B. D., & Norman, K. L. (1994). Assessing the equivalence of the paper and online formats of the QUIS 5.5. In Proceedings of Mid Atlantic Human Factors Conference, (pp. 87-91)Washington, DC. Slawson, D., Hauck, F., Strayer, S., & Rollins, L. (2007). What is information master? Retrieved from http://www. healthsystem.virginia.edu/internet/familymed/docs/ info_mastery.cfm#Information. Smith, A. D. (2006). Exploring m-commerce in terms of viability, growth and challenges. International Journal of Mobile Communications, 4(6), 4. Smith-Daniels, V., Schweikhart, S., & Smith-Daniels, D. (1988). Capacity management in health care services: Review and future research directions. Decision Sciences, 19, 889–919. Sobolev, B., Harel, D., Vasilakis, C., & Levy, A. R. (2008). Using the statecharts paradigm for simulation of patient flow in surgical care. Health Care Management Science, 11, 79–86. Socialdepartementet, S. K. L. Socialstyrelsen. (2007). National IT strategy in health care. Where do we stand? Report 2007 [Nationell IT-strategi för vård och omsorg. Var står vi idag? Lägesrapport 2007]. Stockholm: Socialdepartementet, SKL, Socialstyrelsen. Socialstyrelsen. (2001). A comprehensive view of quality indicators in health care. A proposal by the National Swedish Board of Health and Welfare [Övergripande kvalitetsindikatorer inom hälso- och sjukvården] Stockholm: Socialstyrelsens förslag. Solberg, L. I., Reger, L. A., & Pearson, T. L. (1997). Using continuous quality improvement to improve diabetes care in populations: The IDEAL model: Improving care for diabetics through empowerment, active collaborations and leadership. The Joint Commission Journal on Quality Improvement, 23, 581–592.

337

Compilation of References

Sosial og helsedepartementet. (1998). Law about patient rights [Lov om patientretttigheter]. Ot prpnr 12 (1998-99). Oslo: Sosial- og helsedepartementet. Sosial-og helsedirektoratet. (2004). Law on patient rights. Instructions [Lov om pasientrettigheter. Rundskriv]. IS12/2004. Oslo: Sosial- og helsedirektoratet. SOU. (2005). Difficult waters? Premium pension savings on course [Svårnavigerat? Premiepensionssparande på rätt kurs?] [Swedish Government Official Reports. Stockholm: Fritzes.]. SOU, 2005, 8. Spath, P. (2004). Preface. In P. Spath (Ed.), Partnering with patients to reduce medical errors (pp. xix-xxiii). Chicago: American Hospital Association Press. 1 This paper borrows heavily from Roy Rada (2008) “Ethnographic discovery of adverse events in patient online discussions: customer relationship management” International Journal of Healthcare Information Systems and Informatics, Vol. 3, Issue 3, pp 77-85 Spyrou, S., Bamidis, P., Pappas, K., & Maglaveras, N. (2005). Extending UML activity diagrams for workflow modelling with clinical documents in regional health information systems. In. Proceedings, MIE2005, 1160–1165. Stalk, G., & Hout, T. (1990). Competing against time how time-based competition is reshaping global markets. New York: The Free Press. Stearns, N. S., Bergan, T. A., Roberts, E. B., & Cavazos, L. F. (1978). A systems intervention for improving medical school-hospital interrelationships. Journal of Medical Education, 53, 464–472. Stearns, N. S., Bergan, T. A., Roberts, E. B., & Quigley, L. (1976, Fall). Systems intervention: New help for hospitals. Health Care Management Review, 1, 9–18. Sterman, J. D. (1994). Learning in and about complex systems. System Dynamics Review, 10(2-3), 291–330. Sterman, J. D. (2000). Business dynamics: Systems thinking and modelling for a complex world. Boston, MA: Irwin McGraw-Hill. Strassmann, P. (1990). The business value of computers. New Canaan, CT: The Information Economic Press.

Strauss, A., & Corbin, J. (1998). Basics of qualitative research: Techniques and procedures for developing grounded theory (2nd ed.). Thousand Oaks, CA: Sage Publications. Subasi, A., & Ercelebi, E. (2005). Classification of EEG signals using neural network and logistic regression. Computer Methods and Programs in Biomedicine, 78(2), 87–99. doi:10.1016/j.cmpb.2004.10.009 Sutton, D. R., Taylor, P., & Earle, K. (2006). Evaluation of PROforma as a language for implementing medical guidelines in a practical context. BMC Medical Informatics and Decision Making, 6, 20. Swinfen, P., Swinfen, R., Youngberry, K., & Wootton, R. (2005). Low-cost telemedicine in Iraq: an analysis of referrals in the first 15 months. Telemedicine and Telecare, 11(Suppl 2), 113. Swingler, K. (1996). Applying neural networks: A practical guide. New York: Academic Press. Tait, A. (1999). Face-to-face and at a distance: The mediation of guidance and counselling through new technologies. British Journal of Guidance & Counselling, 27(1), 113–122. Tam, M. C. Y., & Tummala, V. M. R. (2001). An Application of the AHP in vendor selection of a telecommunications system. Omega, 29(2), 171–182. doi:10.1016/ S0305-0483(00)00039-6 Tan, J. (2001). Health management information systems: Methods and practical applications (2nd ed.). Gaithersburg, MD: Aspen Publishers, Inc. Tan, J. (2005). E-health care information systems. An introduction for students and professionals. San Fransisco, CA: Jossey Bass. Tan, J. K. H. (1995). Health management information systems: theories, methods, and applications. Gaithersburg, MD: Aspen Publishers. Tan, J., Kifle, M., Mbarika, V., & Okoli, C. (2005). EMedicine in developed and developing countries. In Tan, J. (Ed.), E-healthcare information systems. Jossey-Bass (A Wiley Imprint). Tanenbaum, A. (2004). Computer networks (4th ed.). Prentice Hall, Inc.

338

Compilation of References

Tashakkori, A., & Teddlie, C. (1998). Mixed methodology: Combining qualitative and quantitative approaches. Thousand Oaks, CA: Sage Publications. Task Force Report. (1998). Recommendations of a task force of the European society of cardiology and the European resuscitation council on the prehospital management of acute heart attacks. European Heart Journal, 19, 1140-1164. Télisson, D., Fayn, J., & Rubel, P. (2004). Design of a tele-expertise architecture adapted to pervasive multi-actor environments. Application to e-cardiology. In Proceedings of the IEEE Computers in Cardiology 2004 Conference (Vol. 31, pp. 749-752). Télisson, D., Fayn, J., Placide, J., Rubel, P., & Comet, B. (2006). A new shared care telemedical solution dedicated to elderly patient nursing services for remote GP decision support. In Proceedings of the Computers in Cardiology 2006 Conference (Vol. 33, pp. 229-232). The Simputer Trust. (2000). Simputer(TM) - Welcome. Retrieved December 1, 2006, from http://www.simputer.org/ Tindal, S. (2008). ZDNet Australia. One in ten RFID projects tag humans. http://www.zdnet.com.au/news/ hardware/soa/RFID-people-tagging-benefits-health-sector-/0,130061702,339285273,00.htm. 21 January 2008. Tiruneh, M. (2006). ETC, broadband network infrastructure for e-health. ICT-H-2006 Workshop, Addis Ababa, Ethiopia. Tomasi, E. F., Luiz Augusto and Maia, Maria de Fatima Santos. (2004). Health information technology in primary health care in developing countries: a literature review. Bulletin of the World Health Organization, 82(11), 867–874. Topacan, U., Basoglu, A. N., & Daim, T. U. “Exploring the Success Factors of Health Information Service Adoption. In Proceedings of Portland International Conference for Management of Engineering and Technology’08 (pp. 2453-2461). Torkki, P. M., Marjamaa, R. A., Torkki, M. I., Kallio, P. E., & Kirvela, O. A. (2005). Use of anesthesia induction rooms can increase the number of urgent orthopedic cases completed within 7 hours. Anesthesiology, 103(2), 401–405. doi:10.1097/00000542-200508000-00024

Tovey, D. (2006). The informed patient. Singapore Medical Journal, 47(9), 741–745. Tufte, E. (2001). The visual display of quantitative information. Cheshire, CT: Graphics Press. Tung, F. C., Chang, S. C., & Chou, C. M. (2008). An extension of trust and TAM model with IDT in the adoption of the electronic logistics information system in HIS in the medical industry. International Journal of Medical Informatics, 77, 324–335. doi:10.1016/j.ijmedinf.2007.06.006 Turri, J. J. (1988). Program eases decision making. Health Progress (Saint Louis, Mo.), 69, 40–44. Uskul, A. K., & Hynie, M. (2007). Self-construal and concerns elicited by imagined and real health problems. Journal of Applied Social Psychology, 37(9), 2156–2189. doi:10.1111/j.1559-1816.2007.00283.x Vaidya, O. S., & Kumar, S. (2006). Analytic hierarchy process: An overview of applications. European Journal of Operational Research, 169, 1–29. doi:10.1016/j. ejor.2004.04.028 Valdes, I., Kibbe, D., Tolleson, G., Kunik, M., & Petersen, L. (2004). Barriers to proliferation of electronic medical records. Informatics in Primary Care, 12(1), 3–9. van den Brink, J. L., Moorman, P. W., de Boer, M. F., Pruyn, J. F. A., Verwoerd, C. D. A., & van Bemmel, J. H. (2005). Involving the patient: A prospective study on use, appreciation and effectiveness of an information system in head and neck cancer care. International Journal of Medical Informatics, 74, 839–849. doi:10.1016/j. ijmedinf.2005.03.021 van der Aalst, W. M. P., ter Hofstede, A. H. M., & Weske, M. (2003). Business process management: A survey. In Business Process Management (BPM2003) (pp. 1–12). Berlin, Germany: Springer-Verlag. van der Aalst, W. M. P., van Dongena, B. F., & Herbst, J. (2003). Workflow mining: A survey of issues and approaches. Data & Knowledge Engineering, 47(2), 237–267. Van Latum, F., Van Solingen, R., Oivo, M., Hoisl, B., Rombach, D., & Ruhe, G. (1998). Adopting GQM-based measurement in an industrial environment. Software, IEEE, 15(1), 78–86.

339

Compilation of References

Vanhecke, T., Barnes, M., Zimmerman, J., & Shoichet, S. (2006). PubMed vs. HighWire Press: A head-to-head comparison of two medical literature search engines. [Epub ahead of print] Computers in Biology and Medicine. Vasilakis, C., Sobolev, B., Kuramoto, L., & Levy, A. R. (2007). A simulation study of scheduling clinic appointments in surgical care: Individual-surgeon vs. pooled lists. The Journal of the Operational Research Society, 58, 202–211. Venkatesh, V., & David, F. D. (2000). A theoretical extension of the technology acceptance model- Four longitudinal field studies. Management Science, 46, 186–204. doi:10.1287/mnsc.46.2.186.11926 Venkatesh, V., Morris, M. G., Davis, G., & Davis, F. D. (2003). User acceptance of information technology - toward to unified view. Management Information Systems Quarterly, 27(3), 425–478. Vrangbæk, K., Østergren, K., Winblad-Spångberg, U., & Okkels, B. (2006). Patients’ reactions to free choice of hospitals in Norway, Denmark and Sweden. Bergen: Norwegian School of Economics and Business Administration. Vrangbæk, K., & Østergren, K. (2004). The introduction of choice in Scandinavian hospital systems. Arguments and policy processes in the Danish and the Norwegian case (Working paper 5). Bergen: Stein Rokkan Centre for Social Studies. Wager, K., Lee, F., & Glaser, J. (2005). Managing healthcare information systems: A practical approach for healthcare executives. San Francisco, CA: Jossey-Bass. Wager, K., Lee, F., White, A., Ward, D., & Ornstein, S. (2000). Impact of an electronic medical record system on community-based primary care practices. The Journal of the American Board of Family Practice, 13(5), 338–348. Wagner, E. H., Austin, B. T., & Korff, M. V. (1996). Improving outcomes in chronic illness. Managed Care (Langhorne, Pa.), 4, 12–25. Walczak, S., Brimhall, B., & Lefkowitz, J. (2006). Nonparametric decision support systems in medical diagnosis: Modeling pulmonary embolism. International Journal of Healthcare Information Systems and Informatics, (1): 2.

340

Walley, P. (2003). Designing the accident and emergency system: Lessons from manufacturing. Emergency Medicine Journal, 20(2), 126–130. Wang, Y. (2007). Integration of data mining with game theory. [Boston, MA: Springer]. Journal of International Federation for Information Processing, 207/2006, 275–280. Wang, H., Zheng, H., & Hu, J. (2008). Poisson approach to clustering analysis of regulatory sequence. International Journal of Computational Biology and Drug Design, 1(2), 141–157. doi:10.1504/IJCBDD.2008.020206 Waring, T., & Maddocks, P. (2005). Open source software implementation in the UK public sector: Evidence from the field and implications for the future. International Journal of Information Management, 25(5), 411–428. Wear, D., Kokinova, M., Keck-McNulty, C., & Aultman, J. (2005). Pimping: Perspectives of 4th year medical students. Teaching and Learning in Medicine, 17, 184–191. Weber, R. P. (1990). Basic content analysis (2nd ed.). Newbury Park, CA. Weber, R., Knaup, P., Knietig, T., Hauz, R., Merzweiler, A., Mludek, V., et al. (2001). Object-oriented business process analysis of the cooperative soft tissue sarcoma trial of the German Society for Paediatric Oncology and Haematology (GPOH). In Proceedings of the 10th World Congress on Medical Informatics (MEDINFO 2001) (pp. 58-62). Weerakkody, V., & Currie, W. (2003). Integrating business process reengineering with information systems development: Issues and implications. In van der Aalst, W. M. P., ter Hofstede, A. H. M., & Weske, M. (Eds.), Business process management (BPM2003) (pp. 302–320). Berlin, Germany: Springer-Verlag. Weick, K., & Sutcliffe, K. (2003). Hospitals as cultures of entrapment. California Management Review, 45(2), 73–84. Weijters, T., & van der Aalst, W. M. P. (2001, October 25-26). Process mining: Discovering workflow models from event-based data. In B. Kröse et al. (Eds.), Proceedings 13th Belgium-Netherlands Conference on Artificial Intelligence (BNAIC’01), Amsterdam, The Netherlands (pp. 283-290).

Compilation of References

Weinstein, R. (2005). RFID: A Technical Overview and Its Application to the Enterprise. IT Professional Magazine, 7(3), 27-33.Whiting, R. (2004), MIT = RFID + Rx. Information Week, 988, 16. Weiser, M. (1993b). Some computer science issues in ubiquitous computing. Communications of the ACM, 36(7), 75–84. doi:10.1145/159544.159617 Weiser, M. (1993a, October). Hot topics: ubiquitous computing. IEEE Computer. Westin, A. F., & Gelder, V. (2005, August). Building privacy by design in health data systems. Center for Social and Legal Research: A report by The Program on Information Technology, Health Records and Privacy of the Center for Social and Legal Research. Retrieved January 29, 2008, from http://www.amia.org/inside/initiatives/ healthdata/2006/ehrrept9-6-05_westin.pdf White, H. (1990). Connectionist nonparametric regression: multilayer feedforward networks can learn arbitrary mappings. Neural Networks, 5(6), 535–549. doi:10.1016/08936080(90)90004-5 Wikipedia. (2007a). Information retrieval. Retrieved from http://en.wikipedia.org/wiki/Information_retrieval. Wikipedia. (2007b). Wikipedia. Retrieved from http:// en.wikipedia.org/wiki/Wikipedia.

Windows, S. (2005). Version 14.0.2005. Chicago: SPSS Inc. Wong, H. J., & Legnini, M. W. (2000). The diffusion of decision support systems in healthcare: are we there yet? Journal of Healthcare Management, 45(4), 240–254. Wong, H. K. (2007). Consumerism and quality standards for subvented social services: a case study of Hong Kong. International Journal of Public Administration, 30(4), 441–458. doi:10.1080/01900690601158287 Wooldridge, M., & Jennings, N. (1995). Intelligent software agents: Theory and Practice. [USA.]. The Knowledge Engineering Review, 10(2), 115–152. Wooton, R. (2001). Telemedicine and developing countries–successful implementation will require a shared approach. Telemedicine and Telecare, 7(Suppl 1), 1–6. doi:10.1258/1357633011936589 Wootton, R. (2001). Recent Advances: Telemedicine. British Medical Journal, 323, 557–560. World Health Organization. (2006a). Estimates of child and adult mortality and life expectancy at birth by country. Retrieved January 17, 2007, from http://www.who.int/ healthinfo/statistics/mortlifeexpectancy/en/index.html

Willems, D. (2001). Balancing rationalities: Gatekeeping in health care. Journal of Medical Ethics, 27, 25–29.

World Health Organization. (2006b, 6th June 2006). Introduction to the ‘Making Pregnancy Safer’ initiative. Retrieved October 1, 2007, from http://w3.whosea.org/ pregnancy/introf.htm

Wimmer, M. (2002). Integrated service modelling for online onestop government. Electronic Markets, 12(3), 149–156.

Wright, D. (1998). Telemedicine and developing countries. A report of study group 2 of the ITU development sector. J Telemedicine Telecare.

Winblad-Spångberg, U. (2003). From decision to practice: The role of the physician in implementing patient choice in Swedish health care [Från beslut till verklighet: Läkarnas roll vid implementeringen av valfrihetsreformer i hälsooch sjukvården]. Uppsala: Institutionen för folkhälso- och vårdvetenskap, Uppsala Universitet.

Xiaohaua, H. (2005). Temporal rule induction for clinical outcome analysis. International Journal of Business Intelligence and Data Mining, 1(1), 122–136.

Windmeijer, F. A. G., & Silva, J. S. (1997). Endogeneity in count data models: an application to demand for healthcare. Journal of Applied Econometrics, 12, 281–294. doi:10.1002/(SICI)1099-1255(199705)12:3<281::AIDJAE436>3.0.CO;2-1

Yackel, T. (2001). How the open source development model can improve medical software. Medinfo, 10, 68–72. Yang, B. H., & Rhee, S. (2000). Development of the ring sensor for healthcare automation. Robotics and Autonomous Systems, 30, 273–281. doi:10.1016/S09218890(99)00092-5

341

Compilation of References

Yang, Y., & Liu, X. (1999). A re-examination of text categorization methods. In Proceedings of the 22nd Annual International ACM Conference on Research andDevelopment in Information Retrieval (SIGIR’99) (pp. 42-49).

Zaki, M. J., Parthasarathy, S., Ogihara, M., & Li, W. (1997). New parallel algorithms for fast discovery of association rules. Data Mining and Knowledge Discovery, 1(4), 343–37.

Yellowlees, P. M., Marks, S. L., Hogarth, M., & Turner, S. (2008). Standards-Based, Open-Source Electronic Health Record Systems: A Desirable Future for the U.S. Health Industry. Telemedicine and e-Health, 14(3), 284-288.

Zeller, A., & Krinke, J. (2005). Essential open source toolset: Programming with Eclipse, JUnit, CVS, Bugzilla, Ant, Tcl/Tk and more. Chichester, England: John Wiley & Sons.

Yin, R.K. (1999). Enhancing the quality of case studies in health service research. Health Service Research, 34(5), Part II, 1209-1224.

Zhang, J., & Fang, X., L. S., & Olivia, R. (2007). Online consumer search depth: theories and new findings. Journal of Management Information Systems, 23(3), 71–95. doi:10.2753/MIS0742-1222230304

Yoo, S. (2006). Ubiquitous Health, Industrial Health. Korea & World Affairs, 5, 13–18. Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8, 338–352.

342

Zurada, J., & Lonial, S. (2005). Comparison Of The Performance Of Several Data Mining Methods For Bad Debt Recovery In The Healthcare Industry. The Journal of Applied Business Research, 21(2), 37–53.

343

About the Contributors

Joseph K. H. Tan, PhD, is the Wayne C. Fox Professor of E-Business Innovation & E-Health, McMaster University, Hamilton, Ontario, Canada. Previously, Dr. Tan had served as Professor and Head, Information System and Manufacturing (ISM) Department, Wayne State University, and as Acting Chair of the Masters in Health Administration (MHA) Program, Department of Healthcare & Epidemiology, Faculty of Medicine, the University of British Columbia. Currently, as the Editor-in-Chief, the International Journal of Healthcare Information Systems & Informatics (IJHISI), Professor Tan sits on various journal advisory and editorial boards as well as on numerous organizing committees for local, national, and international meetings and conferences. Professor Tan is well published and his research, which has enjoyed significant support in the last 21 years from local, national, and international funding agencies and other sources, has also been widely cited and applied across a number of major disciplines, including health care informatics and clinical decision support, health technology management research, human processing of graphical representations, ergonomics, health administration education, telehealth, mobile health, and e-health promotion programming. His hobbies include writing and editing books, book chapters, and journal articles; working on collaborative grant projects; engaging in philosophical discussions with colleagues and peers; and reading his son’s work. *** Xiangyang Li, PhD, is currently Associate Professor with the Department of Industrial and Manufacturing Systems Engineering of the University of Michigan – Dearborn. He received a Ph.D. degree with research in information security from Arizona State University, a M.S. degree in systems simulation from the Chinese Academy of Aerospace Administration, and a B.S. degree in automatic control from Northeastern University, China. Dr. Li’s research interests include health system engineering, quality and security of information systems, knowledge discovery and engineering, human machine studies, and system modelling and simulation. He has been focusing on applying system and multidisciplinary approaches to the improvement and management of complex enterprises. His research has been extensively published in a variety of journals and conferences including knowledge engineering, information assurance, human computer interaction, design, etc. Dr. Li has been well recognized in a set of projects involving national and regional health service institutions, federal agencies, and industries. He is the senior member of the Institute of Industrial Engineers, the member of IEEE, the Association for Computing Machinery, and the Chinese Association for Systems Simulation, and Academic Advocate to the Information Systems Audit and Control Association (ISACA). Aside the research and educational activities, he enjoys sports, reading and travelling. Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

About the Contributors

Yung-wen Liu, Ph.D., is an Assistant Professor in the Department of Industrial and Manufacturing Systems Engineering. Dr. Liu has been working on healthcare research/projects for eight years. The research/projects include OR capacity and patient flow evaluation, pelvic pain studies, sexual satisfaction analysis for treated prostate cancer patients, multi-staged disease progress modeling, and racial/ ethnic disparity in healthcare quality and costs. The methods applied to the research/projects include statistical analysis, simulation, meta-analysis, cost-effectiveness analysis, stochastic and optimization modeling. The projects were done/ are being worked through the collaboration with Mercy Hospital of Buffalo, University of Washington School of Medicine, Seattle Cancer Care Alliance, Henry Ford Health System, and John D. Dingell VA Medical Center. Some research results were published in the journals such as Journal of Urology, Clinical Journal of Pain, and Healthcare Issue in IIE Transactions. His research in disease progress modeling won the best student paper award in 2006 Society for Health System Annual Conference. He teaches statistics and simulation courses and helps students on healthcare related senior design projects at the University of Michigan-Dearborn. Umit Topacan is a MA student in the department of Management Information Systems at Bogazici University, Istanbul, Turkey. He received his BS in Computer Education and Educational Technologies from the same university. His research interests include technology management, information technology adoption, health information services and service development. He has work experiences both of academia as a teaching and research assistantship; and private sector as IT consultant. Nuri Basoglu is an Associate Professor in Department of Management Information Systems, Bogazici University, Istanbul, Turkey. His research interests are socio-technical aspects of IS, customer-focused product development, information technology adaptation and wireless service design, Intelligent adaptive human computer interfaces, information systems strategies. He has published articles in journals such as ‘Technology Forecasting and Social Change’, ‘Journal of High Technology Management’ and ‘Technology in Society’, ‘International Journal of Services Sciences’. Dr. Basoglu received his BS in Industrial Engineering, Bogazici University in Turkey, MS and Ph.D. in Business Administration, Istanbul University. Tugrul Daim is an Associate Professor of Engineering and Technology Management at Portland State University. He is published in many journals including “Technology in Society”, “Technology Forecasting and Social Change”, “Int’l J of Innovation and Technology Management”, “Technology Analysis and Strategic Management”, and “Technovation”. Dr.Daim received his BS in Mechanical Engineering from Bogazici University in Turkey, MS in Mechanical Engineering from Lehigh University in Pennsylvania, another MS in Engineering Management from Portland State University and a Ph.D. in Systems Science-Engineering Management from Portland State University. George E. Heilman is an Associate Professor of Management Information Systems at WinstonSalem State University’s School of Business and Economics, where he teaches information technology and accounting courses in the undergraduate, MBA and MHA programs. He holds undergraduate degrees from Purdue University, masters degrees in business administration and public affairs from Indiana University, and a Ph.D. in Computer Information Systems from the University of Arkansas. He has published a number of discipline-based and pedagogical articles and book chapters dealing with a variety of technology, finance and healthcare issues.

344

About the Contributors

Monica Cain is Associate Professor of Economics at Winston-Salem State University’s School of Business and Economics. She did her graduate work at Wayne State University, where she earned an M.S. in Community Health Services from the School of Medicine in 1995 and a Ph.D. in Economics from the College of Liberal Arts in 2002. Dr. Cain teaches undergraduate economics and graduate healthcare management courses. She has published several articles in the areas of community health and public healthcare management. Russell S. Morton is an Associate Professor of Management Information Systems (MIS) at WinstonSalem State University. He received a PhD in MIS from the University of Kentucky in 1996. He started at WSSU in the fall of 1997 after teaching one year at the University of Indiana Southeast in New Albany, Indiana. He received a teaching award from the University of Kentucky and the first annual Business and Economics Award for Outstanding Achievement in Teaching at WSSU. His research has appeared in the International Journal of Electronic Business, the International Journal of Healthcare Information Systems and Informatics and the Journal of Computer Information Systems. He is the State Volunteer Chair for The Rocky Mountain Elk Foundation, a national habitat conservation organization, and serves on the board of the Huntsville Historic Preservation Society. Dr. Morton received a BSBA with emphasis in Labor Relations and MBA with emphasis in Information Systems from the University of Colorado at Denver. He was a long time employee of Caterpillar Tractor Company in Denver, Colorado before leaving to pursue his PhD. Teemu Paavola, PhD, has worked in teleoperator business development, in academic life, and as a member of technology strategy group of Sonera Plc. He is currently Managing Director and CEO of LifeIT Plc, a consulting company owned by a Finnish health district and private parties, such as TietoEnator Plc. Dr. Paavola holds a research fellow position at Seinajoki Central Hospital, and he has contributed to numerous articles on information technology management, including the textbooks Tietotekniikan linkki liiketoimintaan (1999), Linking IT to Business - A Tale of Discovering IT Benefits (2001), Clinical Knowledge Management (2005), Managing Worldwide Operations and Communications with Information Technology (2007), Encyclopedia of Portal Technologies and Applications (2007), Exploring IT System Benefits in Health Care (2008) and Medical Informatics (2009). Peter Stone is Professor of Maternal Fetal Medicine in the University of Auckland. His postgraduate training was in Britain, gaining a Doctor of Medicine based on Doppler studies in fetal growth restriction from the University of Bristol. After working in Wellington at the University of Otago for 11 years, where he set up the maternal fetal medicine service he moved to Auckland in 1998. He has been involved in Obstetrics , Maternal and Fetal Medicine and Women’s Health throughout his professional career. He has been a member of a number of Ministerial advisory groups most recently on the screening advisory groups for HIV and Down Syndrome as well as being a member of the National Screening Advisory Group for the Director General of Health He is part of the ISTAR group which brought Mifepristone into New Zealand. He is a councillor for the RANZCOG and is currently Chair of the New Zealand Training and Accreditation Committee of RANZCOG. Research interests include fetal welfare assessment, ultrasound studies of the cervix in pregnancy, and early pregnancy development including implantation and trophoblast deportation. Other research interests include teaching quality improvement. Currently he is developing an ultrasound teaching programme for the Pacific in association with the RANZCOG and the Pacific Women’s Health Research Development unit set up in his Department in Middlemore Hospital. 345

About the Contributors

Emma Parry isClinical Director of MFM at Auckland District Health Board. Her main interests over the last few years have been induction of labour rates and the techniques used to achieve labour induction. HerMD thesis is titled ‘Induction of labour: How, why and when?’. She has also been involved in looking at Obstetricians views on induction of labour and caesarean section. She is a trained sub-specialist in Maternal-fetal medicine. Her clinical interests focus around high risk pregnancy and complex multiple pregnancy. She is leading a team of subspecialists in New Zealand setting up a MFM Network. David Parry is a Senior Lecturer in the Auckland University of Technology School of Computing and Mathematical Sciences New Zealand. His PhD thesis was concerned with the use of fuzzy ontologies for medical information retrieval. He holds degrees from Imperial College and St. Bartholomew’s Medical College, London, Auckland University of Technology and the University of Otago, New Zealand. His research interests include internet-based knowledge management and the semantic web, health informatics, the use of Radio Frequency ID in healthcare and information retrieval. Phurb Dorji is consultant in charge of the perinatal service Jigme Dorji Wangchuck National Referral Hospital, Thimphu, Bhutan. He has extensive experience in the development of services for the care of pregnant women in developing nations and a continuing interest in telemedicine. He wasrecently an invited speaker at the Asia and Oceania Federation of obstetrics and Gynaecology Conference. Jongtae Yu is a doctoral student in Business Information Systems at Mississippi State University. He holds a Masters degree in Marketing from the University of Alabama and MBA degree from Pukyong National University in South Korea. He worked as assistant researcher at KIOS research center in South Korea and was responsible for an economic evaluation and cost-benefit analysis for financial losses caused by damages from public development projects. His research interests include health care management information systems, telecommunications, and cross cultural global information systems. His current research projects include the identification of inhibitors and enablers in adopting digital multimedia broadcasting (DMB) and the adoption of ubiquitous banking services, which is expected to be the future banking service. Chengqi Guo is an assistant professor of Computer Information Systems at the University of West Alabama, Livingston Alabama. He is also a doctoral student of Business Information Systems in the College of Business & Industry at Mississippi State University in Starkville, MS. He holds a Masters degree in Operations & Management Information Systems from Northern Illinois University. He has years of industrial experience as a business consultant and IS analyst working for large state owned company in China and Fortune 500 firm in United States. His research interests lie in virtual ventures, business telecommunication, mobile commerce, healthcare information systems, and cross cultural research. Some of his current research projects include ubiquitous banking service adoption, online social network, Web 2.0, and RFID implementation.

346

About the Contributors

Mincheol Kim is an associate professor in the Department of Management Information Systems at Cheju National University in South Korea. He was awarded a PhD degree in Operation Research & Management Information Systems from Korea University and a Masters degree in Health Policy & Management from Seoul National University in South Korea. He worked as researcher in Marketing Research Team at SK Telecom in South Korea and was responsible for marketing research and demand forecasting in Telecommunications Industry. His current research interests include ubiquitous healthcare, telecommunications and healthcare management. His current research projects include the entry strategy for telecommunication company and adoption of ubiquitous banking services. Roy Rada is a Professor of Information Systems at the University of Maryland Baltimore County. Previously, he was Boeing Distinguished Professor of Software Engineering at Washington State University, Editor of Index Medicus at the National Library of Medicine, and Professor of Computer Science at the University of Liverpool. Rada has worked as a consultant on computer-supported diagnosis in pathology and radiology, led a team developing medical informatics standards, developed online training material for doctors, and consulted with insurance companies and hospital networks about compliance with government regulations related to information systems. Rada’s educational credentials include a Ph.D. from University of Illinois in Computer Science and a M.D. from Baylor College of Medicine. He has authored hundreds of scientific papers. His first journal article appeared in 1979 in “Computers and Biomedical Research” and described a novel coding system for medical problem statements. Since having been treated for cancer in 2003, he has devoted considerable effort to the role of information systems, particularly patient online support groups, in health care.

347

348

Index

A

B

accident and emergency departments (A&E) 242 active server pages (ASP) 46 activity diagram 278, 279, 281, 282, 283, 284 Act on Specialized Medical Care 95 acute ischemia , 302, 303, 304, 305, xv acute physiology and chronic health evaluation (APACHE) 85, 163, 167, 173, 178, 181, 183 administrative support 56 agent language mediated activity model (ALMA) 200 American Health Information Management Association 67, 71 American Library Association 15, 32 American Telemedicine Association 1 anaesthetization stages 97 analytical hierarchy process (AHP) 1, 2, 3, 4, 5, 8, 9, 10 antipathetic behaviour 258 application program interfaces (API) 46 Area Under the Roc Curve (AUC) 305 arrhythmia 303, 304, 305 artificial intelligence (AI) 157 artificial-joint surgery 95, 97, 98, 99 artificial neural networks (ANN) 84, 85, 86, 88, 90, 91, 92 association mining 257, 259, 262 Association of County Councils in Sweden 130, 131, 136 asynchronous transfer mode (ATM) 36 attendings 153, 154, 156 attribute-value-class-set (AVC-set) 291, 292 attribute-value-square-set (AVS-set) 291

basic e-Medicine service (BEMS) 44, 45, 46, 47, 48, 49, 50, 51, 52 Basic Health Unit (BHU) 167 Bayesian networks 305, 310 big five model 156 binary splitting 290 body mass index (BMI) 106 broadband multimedia network (BMN) 36, 37, 41, 42, 43, 44 business process reengineering (BPR) 94, 96

C capacity limits 131 categorical dependent variable 290 causality 259, 260, 263, 265, 270, 271 Centre for Evidence Based Medicine 16 class diagram 276, 279, 280, 281, 282, 284 classification and regression trees (CART) 289, 290, 292, 293, 300 clinical data , 302, xv clinical pathways 247 command response protocol 203 composite linked birth file 87 computer-base patient records system (CBPR) 177 computerised physician order systems 278 computerized physician order entry (CPOE) 68, 76, 82 Consumer Health Informatics 126, 128, 129, 140 continuous dependent variable 290 cost-benefit analysis 238 customer relationship management (CRM) 127

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Index

D database management system (DBMS) 45, 291 data mining 114, 256, 258, 289, 290, 299, 300 decision-support 55 decision trees , 288, 289, 290, 291, xiv, 299, 300 diabetes 106, 107, 110, 111 diffuse symptom 236, 238 digital data network (DDN) 41, 42 disease management , 102, 103, 104, x, 106, 107, 108, 109, 110, 111 dynamic causal mining (DCM) 257, 258, 259, 260, 261, 262, 265, 266, 267, 269, 270 dynamic causal rule 257, 262, 263, 266

E ebusiness 163 e-commerce systems 198 economic development 162 ED information systems (EDIS) , 241, 242, 243, 244, 245, xiii, 247, 248, 251, 252 efficient healthcare 232 e-health 126, 128, 129, 140, 142, 143, 162, 163, 165, 166, 170, 171, 172, 173 electrocardiogram (ECG) , 2, 10, 302, 303, 304, 305, 306, 307, 308, 309, xv electronic books (eBooks) 23 electronic healthcare record (EHR) 67, 69, 73, 74, 76, 79, 80, 178, 308, 310 electronic health data capture 55 electronic logistic information system 2 electronic mail (e-mail) 57, 146, 148, 160, 170, 171, 180, 181, 234, 288 electronic medical record (EMR) , 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, viii, 177, 178, 179, 183, 193 emergency department information systems (EDIS) 69 emergency departments (ED) , 241, 242, 243, 244, 245, 246, xiii, 247, 248, 250, 251, 252 emergency obstetric care (EMOC) 167, 173 emergency rooms (ER) 242 empowerment of patients 233 enterprise resource planning (ERP) 74

e-patient network 233 e-patients 232, 233, 239 EPI-MEDICS project , 302, 303, 304, 305, 307, 308, xiv, 309 ethnography 232, 234 EU Privacy Directive 116 evidence-based medicine (EBM) 15, 32, 33 extensible markup language (XML) 307, 308

F Free Hospital Choice 132, 135 fuzzy logic , 200, 211, 212, 302, 306, 307, xv fuzzy rule 200

G Gastrointestinal Motility Online , vii, 14, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 32 general practitioner (GP) 132, 134 GHA 107, 108 global assistance for medical equipment (GAME) 171, 172 Global Care Solutions (GCS) 218 global outbreak alert and response network (GORAN) 171, 172 Google 17, 18, 30, 31, 33 Gore, Al 16 graphical user interface (GUI) 44, 50

H hardware 205, 209, 211 Hawking, Stephen 16 head injury observation (HIO) 249, 250 head mounted display (HMD) 215 Healthcare Information and Management Systems Society (HIMSS) 67, 70, 73, 180, 192 healthcare information system (HIS) 162, 163, 164, 214 healthcare management 214, 215, 216, 217 healthcare management information systems (HMIS) 216 healthcare resource groups (HRG) 289, 294, 297 Health Enhancement Research Organization (HERO) 107, 110

349

Index

health information service (HIS) 2, 6, 8, 10 Health Insurance Portability and Accountability Act (HIPPA) , 74, 75, 79, 112, 113, x, 118, 119, 120, 121, 122, 124, 125, x Health Maintenance Organization (HMO) 103, 106, 216 health management 270, 271 health personal information (HPI) 120, 121 health status , 102, 104, x, 107, 108, 109 heart disease 106, 107 heuristics 289 Hippocratic Oath 67 human computer interface (HCI) 49 hypertension 106, 107

IT productivity paradox 96

I

least absolute deviation (LAD) 291 least square deviance (LS) 290, 291 lipid disorders 106, 107 local area network (LAN) 36, 37, 38, 39, 40, 41, 50, 51, 203 LOGIT 85

INDANA database 305 information and communication technology (ICT) 2, 8, 36, 37, 38, 41, 42, 43, 51, 52, 53, 283, 286 information entropy 290 information exchange 68 information systems (IS) 241, 175, 177, 180, 183, 185, 188, 192, 241, 242, 243, 244, 246 information technology (IT) , 36, 38, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 94, 95, 96, 99, 100, 101, 109, 126, 127, 128, 169, 140, 141, 143, 168, 170, 172, 176, ix, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, ix, 217, 227, 230 Institutional Review Board 57, 148, 237 integrated definition (IDEF) 276, 286 integrated services digital network (ISDN) 36, 41 intelligent agent framework , xii intelligent agents , 195, 198, 199, xii, 200, 204 intelligent monitoring systems 2 intelligent software agents 197 International Telecommunications Union (ITU) 37, 53, 168, 170, 173 internet , 65, 72, 73, 77, 82, 83, 146, 147, 181, 157, 180, 186, 192, 206, ix intranet 72, 206

350

J java server pages (JSP) 46, 52 Jigme Dorji Wanchuck National Referral Hospital (JDWNRH) 162, 167, 172 Joint Commission on Accreditation of Healthcare Organizations (JCAHO) 67, 70

K knowledge discovery in databases (KDD) 289, 296

L

M machine learning 289, 291, 300, 301 magnetic resonance imaging vendor 2 managed care organizations (MCO) , 102, 103, 104, x, 105, 107, 108, 109 maternity care coordination (MCC) 84, 86, 87, 88, 89, 90, 91 Medicaid 84, 86, 87, 90, 91, 92, 93 medical adverse events 232, 233 medical errors , 65, 66, ix, 68, 73, 74, 75, 77 medical imaging retrieval system 284 medical subject headings (MeSH) 17, 27 Medical University of South Carolina (MUSC) 54, 55, 56, 57, 59, 60, 61, 62, 63 metadata 207, 208 Ministry of Social Affairs and Health 97

N National Health Information Network (NHIN) 76 Nature Publishing Group (NPG) 20, 22, 30 netnography 234, 239 neural networks 84, 85, 86, 88, 90, 91, 92, 93, 304, 310

Index

neurons 88 NeuroShell2 88, 89, 90 new productivity paradox 96 new public management (NPM) 127 nonidentifiable information (NII) 114 North Carolina State Center for Health Statistics (SCHS) 87

O obstetrics 162, 173, 174 obstructive sleep apnea (OSA) 235 online patient communities 232 open source movement 162 open source software (OSS) , 162, 176, 163, 164, 165, 167, 168, 169, 170, 175, 176, 177, 178, 179, xii, 180, 181, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193 ordinary line square (OLS) 214, 221 organization for economic cooperation and development (OECD) 171 out patient card (OPCard) 46 outpatient clinic 276, 277, 278, 279, 280, 281

P paper-based system 62 paperless hospitals 209, 210 Parkinson’s disease 120 patient profiling 198, 200 Patient Safety and Quality Improvement Act 76 Patients’ Rights Act 132 payment by results (PbR) 289, 301 perceived ease of use (PEOU) 217 perceived usefulness (PU) 217 perinatal medicine 162 personal digital assistant (PDA) 23, 24, 202, 203, 204, 206, 308 personal ECG monitor (PEM) , 302, 303, 304, 305, 306, 307, 308, 309, 310, xv personal identifiable information (PII) , 112, 113, 114, x, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124 picture archival and communication systems (PACS) 69 pimping 153, 154, 156, 160

population management 109 Practice Partner Patient Record 55 preferred provider organization (PPO) 103 prescription writer 56 private market 130 process management 94 process thinking 95, 96 profile matching 199, 200 protected health information (PHI) 120 public healthcare initiative 86 public healthcare services 85, 86, 90 PubMed 15, 17, 21, 23, 34

Q questionnaire for user interaction satisfaction (QUIS) 56, 57, 62, 64

R radio-frequency identification (RFID) , 69, 80, 81, 195, 196, 197, 199, xii, 200, 201, 202, 203, 204, 209, 210, 211 random blood glucose (RBG) 250 receiver operating characteristic (ROC) curves 306 reimbursable event 238 risk status , 102, 104, x, 105, 106, 107, 108, 109 Royal Government of Bhutan (RGOB) 168, 172

S secure socket layer (SSL) 38 Seinajoki Central Hospital 94, 95, 96, 97, 99 self-care 309, 310 self organising mapping (SOM) 248, 249 semi-automatically linking 237 snapshot data analyses 96 software 197, 198, 200, 201, 202, 204, 205, 206, 210, 211 spirometry (SPR) 250 state diagram 278, 280, 281, 285 student inspired model for effective clinical teaching (SIMECT) 154, 155, 158 SWOT , 65, 66, ix, 67, 77, 78, 80, 83 sympathetic behaviour 258

351

Index

system crashes 60 system dynamics 256, 257, 258, 262, 271, 273, 274 system speed 56, 61, 62, 63 system thinking 257, 258

T taxonomy 74, 79 technology adoption 214, 216, 217 technology adoption model (TAM) 217, 218, 220 technology gap 256 telecare 2 telecommunication technologies 2 telemedicine , 1, 2, 8, 9, 10, 163, 164, 165, 172, 173, 174, 284, 286, 302, xv, 311 tele-nurse 309, 310 theory of constraints 94, 96 theory of reasoned action (TRA) 220 time-motion study 61, 64 total decision weight 8 total quality management 94, 95, 96 treatment status , 102, 104, x, 105, 106, 107, 108, 109 triability 2, 3, 6, 7, 8, 9

U ubiquitous healthcare system 214 u-health system 214, 215, 216, 217, 218, 220, 221, 222, 223, 224, 226, 227, 228 UK National Health Service (NHS) 289, 294, 295, 297, 300 unified medical language system (UMLS) 21, 26, 27, 32

352

unified modelling language (UML) , 275, 276, 277, 278, 279, 280, 281, xiv, 282, 283, 284, 285, 286, 287 U.S. Department of Health and Human Services (HHS) 70, 73, 113, 118, 121, 123, 124 user resistance 71

V vectors of alternatives 5 very small aperture (VSAT) satellite 36, 41, 42, 43, 44, 51, 168 Veterans Health Administration 67 victorian emergency medical data (VEMD) 248, 253 virtual reality 215

W walk-in clinic 133, 134 web 2.0 232 WebMD 15, 22 weight of attributes 5 wide area network (WAN) 36, 37, 38, 40, 41, 42, 43, 44, 50, 51 Wiki , vii, 14, 17 Wikipedia 16, 17, 18, 25, 34 wired local area network (WLAN) 203, 204, 206 women, infants and children program (WIC) 86, 87, 91 World Health Organisation 167, 170 write once, read many (WORM) 203

Y Y2K investments 96