International Review of Research in Mental Retardation, Volume 33: Developmental Epidemiology of Mental Retardation and Developmental Disabilities

International Review of RESEARCH IN MENTAL RETARDATION Developmental Epidemiology of Mental Retardation and Development...

Author: Richard C. Urbano | Robert M. Hodapp | Laraine Masters Glidden

45 downloads 1004 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

International Review of RESEARCH IN MENTAL RETARDATION Developmental Epidemiology of Mental Retardation and Developmental Disabilities VOLUME 33

International Review of

RESEARCH IN MENTAL RETARDATION EDITED BY

LARAINE MASTERS GLIDDEN

DEPARTMENT OF PSYCHOLOGY ST. MARY’S COLLEGE OF MARYLAND ST. MARY’S CITY, MARYLAND

Board of Associate Editors Philip Davidson UNIVERSITY OF ROCHESTER SCHOOL OF MEDICINE AND DENTISTRY

Elisabeth Dykens VANDERBILT UNIVERSITY

Michael Guralnick UNIVERSITY OF WASHINGTON

Richard Hastings UNIVERSITY OF WALES, BANGOR

Linda Hickson COLUMBIA UNIVERSITY

Connie Kasari UNIVERSITY OF CALIFORNIA, LOS ANGELES

William McIlvane E.K. SHRIVER CENTER

Glynis Murphy LANCASTER UNIVERSITY

Ted Nettelbeck ADELAIDE UNIVERSITY

Marsha M. Seltzer UNIVERSITY OF WISCONSIN-MADISON

Jan Wallander SOCIOMETRICS CORPORATION

Developmental Epidemiology of Mental Retardation and Developmental Disabilities A Volume in

International Review of

RESEARCH IN MENTAL RETARDATION VOLUME 33 EDITED BY

Richard C. Urbano VANDERBILT KENNEDY CENTER VANDERBILT UNIVERSITY NASHVILLE, TENNESSEE

Robert M. Hodapp JOHN F. KENNEDY CENTER FOR RESEARCH ON HUMAN DEVELOPMENT AND PEABODY COLLEGE VANDERBILT UNIVERSITY NASHVILLE, TENNESSEE

AMSTERDAM • BOSTON • HEIDELBERG • LONDON NEW YORK • OXFORD • PARIS • SAN DIEGO SAN FRANCISCO • SINGAPORE • SYDNEY • TOKYO Academic Press is an imprint of Elsevier

Academic Press is an imprint of Elsevier 525 B Street, Suite 1900, San Diego, California 92101-4495, USA 84 Theobald’s Road, London WC1X 8RR, UK

This book is printed on acid-free paper. Copyright ß 2007, Elsevier Inc. All Rights Reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the Publisher. The appearance of the code at the bottom of the first page of a chapter in this book indicates the Publisher’s consent that copies of the chapter may be made for personal or internal use of specific clients. This consent is given on the condition, however, that the copier pay the stated per copy fee through the Copyright Clearance Center, Inc. (www.copyright.com), for copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale. Copy fees for pre-2007 chapters are as shown on the title pages. If no fee code appears on the title page, the copy fee is the same as for current chapters. 0074-7750/2007 $35.00 Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone: (þ44) 1865 843830, fax: (þ44) 1865 853333, E-mail: [email protected]. You may also complete your request on-line via the Elsevier homepage (http://elsevier.com), by selecting ‘‘Support & Contact’’ then ‘‘Copyright and Permission’’ and then ‘‘Obtaining Permissions.’’ For information on all Academic Press publications visit our Web site at www.books.elsevier.com ISBN-13: 978-0-12-366233-0 ISBN-10: 0-12-366233-8 PRINTED IN THE UNITED STATES OF AMERICA 07 08 09 9 8 7 6 5 4 3 2 1

Contents

Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ix xi xiii

Section I. Introduction Developmental Epidemiology of Mental Retardation/Developmental Disabilities: An Emerging Discipline Robert M. Hodapp and Richard C. Urbano I. II. III. IV. V. VI.

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Epidemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Developmental Epidemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Issues Specific to MR/DD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advances and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 4 8 11 16 19 20

Section II. Methodological Issues and Perspectives Record Linkage: A Research Strategy for Developmental Epidemiology Richard C. Urbano I. II. III. IV. V. VI.

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Past and Present Linked Datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Biases and Prejudices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Embarking on a Linked‐Data Research Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Workflow of Linkage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

27 29 31 34 38 51 51

contents

vi Second‐Order Linkage and Family Datasets Shihfen Tu, Craig A. Mason, and Quansheng Song I. II. III. IV. V. VI.

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . What Is a Second‐Order Linkage and What Are Its Benefits? . . . . . . . . . . . . . . . . . . . . Overview of Data Linkage Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How to Conduct a Second‐Order Data Linkage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Challenges of Building a Second‐Order Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53 53 56 58 70 76 77

Incorporating Geographical Analysis into the Study of Mental Retardation and Developmental Disabilities Russell S. Kirby I. II. III. IV. V.

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Medical Geography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mental Retardation and Developmental Disabilities Research . . . . . . . . . . . . . . . . . . . . Research Agenda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79 80 84 85 88 88

Statistical Issues in Developmental Epidemiology and Developmental Disabilities Research: Confounding Variables, Small Sample Size, and Numerous Outcome Variables Jennifer Urbano Blackford I. II. III. IV. V.

Statistical Challenges in Developmental Disabilities Research. . . . . . . . . . . . . . . . . . . . . Confounding Variables: How to Match on More Than Two Variables . . . . . . . . . . . Small Sample Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Numerous Outcome Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Closing Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

93 94 103 111 118 118

Economic Perspectives on Service Choice and Optimal Policy: Understanding the Effects of Family Heterogeneity on MR/DD Outcomes Stephanie A. So I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II. What Is Economics?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

121 121

contents III. IV. V. VI.

vii

The Relationship Between Economics and Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Economics of MR/DD Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Econometric Models and Estimation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

128 133 138 144 146

Section III. Populations Public Health Impact: Metropolitan Atlanta Developmental Disabilities Surveillance Program Rachel Nonkin Avchen, Tanya Karapurkar Bhasin, Kim Van Naarden Braun, and Marshalyn Yeargin‐Allsopp I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II. Background of Developmental Disabilities Surveillance in Metropolitan Atlanta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III. Prevalence Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV. Findings from Epidemiology Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Public Health Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VI. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

149 151 157 164 164 187 187

Using GIS to Investigate the Role of Recreation and Leisure Activities in the Prevention of Emotional and Behavioral Disorders Tina L. Stanton‐Chapman and Derek A. Chapman I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II. Neighborhood Influences: Effects on Childhood Emotional/Behavioral Disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III. The Importance of Neighborhood‐Related Leisure Activities . . . . . . . . . . . . . . . . . . . IV. Geographical Information Systems: Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Using GIS to Study Emotional and Behavioral Disorders: Promises and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VI. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

191 194 196 198 200 207 207

The Developmental Epidemiology of Mental Retardation and Developmental Disabilities Dennis P. Hogan, Michael E. Msall, and Julia A. Rivera Drew I. Epidemiological Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II. Models for Understanding Mental Retardation and Developmental Disabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

213 216

contents

viii

III. Data and Measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

226 229 241 243

Evolution of Symptoms and Syndromes of Psychopathology in Young People with Mental Retardation Stewart L. Einfeld, Bruce J. Tonge, Kylie Gray, and John Taffe I. II. III. IV. V. VI.

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Australian Child to Adult Development (ACAD) Study . . . . . . . . . . . . . . . . . . . Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

247 248 250 251 252 262 263

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contents of Previous Volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

267 279

Contributors

Numbers in parentheses indicate the pages on which the authors’ contributions begin.

Rachel Nonkin Avchen (149), Centers for Disease Control and Prevention, National Center on Birth Defects and Developmental Disabilities, Atlanta, Georgia 30333 Tanya Karapurkar Bhasin (149), Centers for Disease Control and Prevention, National Center on Birth Defects and Developmental Disabilities, Atlanta, Georgia 30333 Jennifer Urbano Blackford (93), Vanderbilt Kennedy Center, Department of Psychiatry, Vanderbilt University, Nashville, Tennessee 37203 Kim Van Naarden Braun (149), Centers for Disease Control and Prevention, National Center on Birth Defects and Developmental Disabilities, Atlanta, Georgia 30333 Derek A. Chapman (191), Department of Epidemiology, Virginia Commonwealth University, Richmond, Virginia 23298 Julia A. Rivera Drew (213), Department of Sociology, Brown University, Providence, Rhode Island 02912 Stewart L. Einfeld (247), School of Psychiatry, University of New South Wales, NSW 2031, Australia Kylie Gray (247), Monash University Centre for Developmental Psychiatry, Victoria 3168, Australia Robert M. Hodapp (3), Vanderbilt Kennedy Center, Department of Special Education, Vanderbilt University, Nashville, Tennessee 37203

ix

x

contributors

Dennis P. Hogan (213), Population Studies and Training Center, Brown University, Providence, Rhode Island 02912 Russell S. Kirby (79), Department of Maternal and Child Health, School of Public Health, University of Alabama at Birmingham, Birmingham, Alabama 35294 Craig A. Mason (53), College of Education and Human Development, The University of Maine, Orono, Maine; and Maine’s University Center for Excellence in Developmental Disabilities, The University of Maine, Orono, Maine 04469 Michael E. Msall (213), University of Chicago Pritzker School of Medicine, Kennedy Mental Retardation Center and Institute of Molecular Pediatrics, Comer Children’s and LaRabida Children’s Hospitals, Chicago, Illinois 60637 Stephanie A. So (121), Vanderbilt Kennedy Center, Department of Economics, Vanderbilt University, Nashville, Tennessee; and Vanderbilt Kennedy Center, Department of Pediatrics, Vanderbilt University, Nashville, Tennessee 37235 Quansheng Song (53), Maine’s University Center for Excellence in Developmental Disabilities, The University of Maine, Orono, Maine 04469 Tina L. Stanton‐Chapman (191), Department of Curriculum, Instruction, and Special Education, Curry School of Education, University of Virginia, Charlottesville, Virginia 22904 John TaVe (247), Monash University Centre for Developmental Psychiatry, Victoria 3168, Australia Bruce J. Tonge, (247), Monash University Centre for Developmental Psychiatry, Victoria 3168, Australia Shihfen Tu (53), College of Education and Human Development, The University of Maine, Orono, Maine; and Maine’s University Center for Excellence in Developmental Disabilities, The University of Maine, Orono, Maine 04469 Richard C. Urbano (3, 27), Vanderbilt Kennedy Center, Department of Pediatrics, Vanderbilt University, Nashville, Tennessee 37203 Marshalyn Yeargin‐Allsopp (149), Centers for Disease Control and Prevention, National Center on Birth Defects and Developmental Disabilities, Atlanta, Georgia 30333

Foreword

This volume, Developmental Epidemiology of Mental Retardation and Developmental Disabilities, guest-edited by Richard C. Urbano and Robert M. Hodapp, continues an important, albeit recent, tradition for the International Review of Research in Mental Retardation series. Since I began my editorship in 1997, six volumes have been organized around five diVerent themes: autism, language, personality/motivation (two volumes), neurotoxicity, and now the current volume on developmental epidemiology. As did the guest editors of the other theme-oriented volumes, Rick Urbano and Bob Hodapp have worked diligently and masterfully, successfully attracting an impressive array of experts to contribute to Volume 33. As they have written in their preface, the field is ready for this volume because developmental epidemiology is cutting edge, using technical tools that did not exist even a few years ago. Nonetheless, the utilitarian purpose of developmental epidemiology is ancient. Research scientists and clinicians have been interested in prevention of mental retardation for as long as we have been able to identify individuals who have it. However, we now have sophisticated methods for measuring, storing, and accessing data about variation at earlier points in time and relating that variation to mental or physical health outcomes at later times. These methods, discussed in the current volume, have resulted in a surge of interest in developmental epidemiology. The media helped to publicize its use in its frenzy to understand the reason for the increase in autism diagnoses. Thus, this volume is timely. Rick Urbano and Bob Hodapp, who proposed and guided it, have done a service to their colleagues. I take this opportunity to thank them personally for their prescience and persistence.

xi

xii

foreword

I am grateful as one of their served colleagues, but also as the series editor who found them to be conscientious, devoted, and tenacious guest editors. LARAINE MASTERS GLIDDEN SERIES EDITOR

Preface

Since it was begun by Norman Ellis in 1966, the International Review of Research in Mental Retardation has included over 30 volumes produced by three series editors and several editors of individual thematic volumes. Until now, however, no single volume has been devoted exclusively to epidemiology or to developmental epidemiology. Moreover, from among over 250 chapters over four decades, only a few, sporadic chapters of the International Review have been devoted exclusively to epidemiology (Boussy & Scott, 1993; Fryers, 1993). Beyond its uniqueness, the present volume is also timely in ways that are important to the field of Mental Retardation/Developmental Disabilities (MR/DD). As virtually every chapter illustrates, until recently a fully fledged field of developmental epidemiology of MR/DD was not possible. The technological and theoretical advances were simply lacking, making less feasible specific studies. As a result, one needs to describe the many technical–theoretical advances as well as the best research examples in this area. The current volume is therefore divided into three discrete sections. In Section I an introductory chapter describes the ‘‘emerging’’ nature of the field of Developmental Epidemiology of MR/DD. In Section II, consisting of the volume’s next five chapters, contributors review the various methodological, statistical, and theoretical advances that make possible studies in the developmental epidemiology of mental retardation. This section is then followed by the four chapters of Section III. Although employing the latest methods and procedures, contributors of Section III’s articles are more focused on their epidemiological findings per se. As befits a perspective that may be less familiar to most International Review readers, we begin with an introduction. In ‘‘Developmental Epidemiology of Mental Retardation and Developmental Disabilities,’’ Richard Urbano and Robert Hodapp reflect upon the exact meaning of the term ‘‘developmental epidemiology of mental retardation.’’ By way of deconstructing that term’s xiii

xiv

preface

three main components (epidemiology, developmental epidemiology, mental retardation), they highlight the many provocative twists involved in applying a developmental epidemiological perspective to persons with mental retardation. In addition, by comparing developmental epidemiological work as applied to two populations—those with MR/DD versus those with child psychiatric conditions—they illustrate the types of findings that seem imminent in the disabilities field. A general overview, a state-of-the-art description of the present, a peek into the future—this chapter illustrates a subfield at the cusp of providing critical information about both individuals with disabilities and their families. In ‘‘Record Linkage: A Research Strategy for Developmental Epidemiology,’’ Richard Urbano then tackles the entire issue of linking together various administrative records. Simply stated, it is one thing to have access to records of all of the births, deaths, hospitalizations, or other events in a town, state, or country; it is another—far more informative—thing to link these records together. But how, exactly, does one go about linking a person’s birth records to that same person’s hospitalization or other records? More generally, how does research using ‘‘somebody else’s data’’ compare to smaller scale studies in which the researcher collects his or her own data? Even once one has acquired such large, population-based datasets, what are the steps in the process of going from several very large, raw datasets to cleaned, linked sets that can be analyzed using SPSS, SAS, STATA, and other statistical packages? Building upon Urbano’s chapter, Shihfen Tu, Craig Mason, and Quansheng Song discuss ‘‘Second-Order Linkage and Family Datasets.’’ Second-Order Linkage also uses large-scale administrative databases, but this time the linkage goes beyond the individual. For example, using the mother’s social security number and other identifiers, one might be able to construct from a state’s Birth Records (and Marriage Records) various families of children born to the same mother. By performing such secondorder linkage, researchers are now able to use large-scale administrative records to examine questions involving family genetics, family SES, and parental education levels, health, birth defects, and other issues. Tu, Mason, and Song describe common problems and oVer innovative solutions for the many complicated questions involved in second-order, family linkage work. Russell Kirby follows with his chapter, ‘‘Incorporating Geographical Analysis into the Study of Mental Retardation and Developmental Disabilities.’’ In line with the epidemiologist’s focus on ‘‘person, place, and time,’’ Kirby explores issues relating to place—the ways in which where one lives aVects one’s health, well-being, and chances in life. Although a burgeoning subfield within developmental epidemiology, ‘‘geographic information systems’’ (GIS) have yet to be systematically applied in the

preface

xv

study of disabilities. Yet given the ties of disabilities to certain communities (Fujiura & Yamaki, 2000) and the recent interest in social factors related to health and health care among populations with developmental disabilities (Graham, 2005), the time seems ripe to explore the many advantages of spatially-based statistical analyses of large populations. Jennifer Blackford follows with her chapter on ‘‘Modern Statistical Methods Applied to Developmental Epidemiology.’’ Although most of us are familiar with analyses using standard statistical tests, Blackford relies on the latest statistical and computing advances to address three troublesome issues. The first concerns matching: How does one match on more than a few variables at one time? Noting the usual advice to ‘‘just pick two’’ matching variables, Blackford introduces propensity scores as a technique to match on more than a few variables simultaneously. The second issue concerns the ways in which permutation testing can help oVset the limitations involved in using small samples. This ‘‘small-sample problem’’ is ubiquitous in disability studies, particularly when examining rarely occurring conditions. Third, Blackford extends beyond Bonferroni corrections to address the issues involved in multiple (often correlated) outcome variables. Although specifically addressed to developmental epidemiologists examining persons with disabilities, this chapter will be helpful to all researchers examining behavioral issues in groups with MR/DD. The final chapter in Section II is by Stephanie So, ‘‘Economic Perspectives on Service Choice and Optimal Policy: Understanding the EVects of Family Heterogeneity on MR/DD Outcomes.’’ Although economics is, to most of us, a less well-understood discipline. So illustrates the ways in which economic principles can apply to MR/DD. Her best example involves what might be called ‘‘family centeredness.’’ Across many disability fields (most prominently, early childhood special education), numerous researchers, service providers, and policymakers debate what constitutes a family-centered practice. Few such professionals rely on economic analyses. By analyzing with large-scale epidemiologic data variables such as family resource allocation, competing family values, opportunity costs, and individual family trade-oVs, we may be able to determine which practices might be best, for which families or types of families. In performing such analyses, so implicitly asks whether many of our ideas about family-centered practices might be missing the mark. As the first chapter within the third, ‘‘findings’’ section, it is a pleasure to introduce Rachel Avchen, Tanya Bhasin, Kim Van Naarden Braun, and Marshalyn Yeargin-Allsopp’s ‘‘Public Health Impact: Metropolitan Atlanta Developmental Disabilities Surveillance Program.’’ Quite simply, the CDC’s MADDS study is among the best available, with findings for over a decade

xvi

preface

on mental retardation and developmental disabilities (Yeargin-Allsopp, Murphy, Oakley, & Sikes, 1992). More importantly for our purposes, Avchen, Bhasin, Braun, and Yeargin-Allsopp illustrate the care and attention that this group has brought to basic epidemiological issues such as case definition and case finding. Their results are among the best in the field, telling us both how often various conditions occur and which child, parent, or other factors predispose children to such conditions. Examining ‘‘Using GIS to Investigate the Role of Recreation and Leisure Activities in the Prevention of Emotional and Behavioral Disorders,’’ Tina Stanton-Chapman and Derek Chapman next illustrate how developmental epidemiological principles can be extended in another way. From Bronfenbrenner (1979) on, developmentally oriented researchers have been intrigued by the ecologies of childhood, the ways in which children live within families, which exist within neighborhoods, towns, states, and countries. Yet how such nested circles aVect children with disabilities has only begun to be examined. Returning to concerns expressed by Kirby in his chapter on geographical analysis, Stanton-Chapman and Chapman apply the perspective of GIS to the ways that parks and recreational programs can prevent emotional–behavioral disorders. Stanton-Chapman and Chapman have been instrumental in bringing geographical analyses into developmental epidemiological studies within the MR/DD field. Dennis Hogan, Michael Msall, and Julia Drew, in their chapter ‘‘The Developmental Epidemiology of Mental Retardation and Developmental Disabilities,’’ come at the developmental epidemiology of MR/DD from yet another perspective. Relying on 1997–2003 data from the National Health Interview Survey, these authors use large-scale interview data to determine issues related to prevalence, family situations, and indicators of both health care and health-care service delivery. From a diVerent perspective, their findings corroborate several of the social predictors of disability found by the CDC and other surveillance data. As the last chapter in Section III, Stewart Einfeld, Bruce Tonge, Kylie Gray, and John TaVe examine the ‘‘Evolution of Symptoms and Syndromes of Psychopathology in Young People with Mental Retardation.’’ Having worked for several decades using large-scale data from the Australian Child to Adult Longitudinal Study, Einfeld, Tonge, and their colleagues are among the leaders in the burgeoning field of dual diagnosis. Along with others, Einfeld and Tonge (1992) have developed maladaptive behavior instruments that have been specifically designed for persons with disabilities (see Dykens, 2000 for a review). In addition, these authors have produced one of the best large-scale, longitudinal studies that inform us about the amount, types, and changes over time in maladaptive behavior/psychopathology among children with MR/DD.

preface

xvii

In closing, we hope that this volume imparts a sense of developmental epidemiology’s methodological, statistical, and theoretical advances, as well as the best of those studies using such techniques. For 40 years, only sporadic attention has been paid to the developmental epidemiology of MR/DD. Our hope—and prediction—is that such topics will grow exponentially in importance in the decades to come. RICHARD C. URBANO ROBERT M. HODAPP ACKNOWLEDGMENTS

We would like to thank Marisa Sellinger and Kara Newman for their help in producing this volume, and Dr. Laraine Glidden for her support and encouragement. This volume was supported in part by NICHD grant numbers R03 HD050468 and P30 HD15052 to Vanderbilt University.

References

Boussy, C. A., & Scott, K. G. (1993). Use of data-base linkage methodology in epidemiologic studies of mental retardation. International Review of Research in Mental Retardation, 19, 135–161. Bronfenbrenner, U. (1979). The ecology of human development. Cambridge, MA: Harvard University Press. Dykens, E. M. (2000). Psychopathology in children with intellectual disability. Journal of Child Psychology and Psychiatry, 41, 407–417. Einfeld, S. L., & Tonge, B. J. (1992). Manual for the developmental behavioural checklist: Primary carer version. Sydney, Australia: School of Psychiatry, University of New South Wales. Fryers, T. (1993). Epidemiological thinking in mental retardation: Issues in taxonomy and population frequency. International Review of Research in Mental Retardation, 19, 97–133. Fujiura, G. T., & Yamaki, K. (2000). Trends in demography of childhood poverty and disability. Exceptional Children, 66, 187–199. Graham, H. (2005). Intellectual disabilities and socioeconomic inequalities in health: An overview of research. Journal of Applied Research in Intellectual Disabilities, 18, 101–111. Yeargin-Allsopp, M., Murphy, C. C., Oakley, G. P., & Sikes, R. K. (1992). A multiple-source method for studying the prevalence of developmental disabilities in children: The Metropolitan Atlanta developmental disabilities study. Pediatrics, 89, 624–629.

Section I Introduction

This page intentionally left blank

Developmental Epidemiology of Mental Retardation/Developmental Disabilities: An Emerging Discipline* ROBERT M. HODAPP VANDERBILT KENNEDY CENTER, DEPARTMENT OF SPECIAL EDUCATION VANDERBILT UNIVERSITY, NASHVILLE, TENNESSEE

RICHARD C. URBANO VANDERBILT KENNEDY CENTER, DEPARTMENT OF PEDIATRICS VANDERBILT UNIVERSITY, NASHVILLE, TENNESSEE

I.

INTRODUCTION

Whether one thinks in terms of airplanes taking oV or flowers blossoming forth, the field of developmental epidemiology of mental retardation/ developmental disabilities (MR/DD) is poised on the edge of new advances, new approaches, and new findings. In many respects, the field seems poised to do things that could hardly be imagined a few years ago. There is an excitement, level of interest, and vitality that could only be imaged a few short years ago. In part, this vitality arises from joining more established research traditions to the latest technical advances. As the chapters in this special issue illustrate, several interesting epidemiological studies already exist, and some groups have successfully employed developmental epidemiological methods for a decade or more (Boussy & Scott, 1993). Recently, however, more sophisticated statistical techniques have been developed for developmental epidemiological studies in developmental disabilities (Blackford, this issue), and new computer advances—in both software development and hardware storage and speed—allow for studies that link across and within various types of records (Urbano, this issue). As a result, increasing numbers of *Authors’ Note: This research was supported in part by NICHD grant numbers R03 HD050468 and P30 HD15052 to Vanderbilt University. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

3

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33001-7

4

Robert M. Hodapp and Richard C. Urbano

studies examine populations of entire countries or regions (Einfeld, Tonge, Gray, & Taffe, this issue) and utilize various large‐scale national and international surveys (Hogan, Msall, & Drew, this issue). Economic and other sophisticated models and theories are also at the ready (So, this issue). Still, despite these preliminaries, one must consider the field of developmental epidemiology—at least as applied to individuals with MR/DD—as only beginning to emerge. Compare developmental epidemiological work in two fields, developmental disabilities and child psychiatry. In studies of the developmental epidemiology of child psychiatric disorders, Costello and Angold (2006) have examined over time the onset, development, and contributors of child psychiatric problems. Tremblay (2004), for example, has examined the trajectory of children’s physical aggression over development, showing that highly aggressive children already were significantly more likely to show diYcult temperaments by 5 months of age (Tremblay et al., 2004). Similarly, various studies have used diverse methodologies to determine whether the prevalence rates of conduct disorder, depression, and autism were rising over time. The general conclusion is that conduct disorder does appear to be rising over successive birth cohorts, whereas neither child‐adolescent depression nor autism are rising, at least not beyond what might be expected given changes in either definition or case‐finding techniques (Costello, Foley, & Angold, 2006). Throughout these studies, there is a sophistication and maturity not yet seen among most studies using developmental epidemiology to study developmental disabilities. The process of getting to there from here—to a more sophisticated, mature research enterprise from the beginnings of a field—is the topic of this chapter. We begin with a backward parsing of the term itself, first describing the basic principles of epidemiology more generally, then how the addition of the term ‘‘developmental’’ complicates and makes more interesting the basic research enterprise. We then tackle issues involved in applying developmental epidemiology to persons with mental retardation, before ending with several advances and new directions. Our purpose throughout is to illustrate to researchers within and outside of developmental disabilities just what might be gained by this new flight into the developmental epidemiological sky. II.

EPIDEMIOLOGY

Although various definitions of the field could be presented, we present below two definitions of the field of epidemiology: Epidemiology is the study of the distribution and determinants of health‐related states or events in specified populations, and the application of this study to control health problems (Last, 1995; Yeargin‐Allsopp & Boyle, 2002, p. 113).

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

5

An emerging concept of epidemiology presents this discipline as the study of health and disease as a full spectrum across the human life span with a population approach, including etiological factors, phenomenology, comorbidities, and the uses and outcomes of clinical care (Mezzich & Ustin, 2005, p. 656).

In many ways, these two definitions might be characterized as more (Yeargin‐Allsopp & Boyle, 2002) and less traditional (Mezzich & Ustin, 2005). Still, despite their diVerences, they nevertheless illustrate several of the main tenets of epidemiology. Four of these tenets are described as follows. A.

A Focus on Populations

Epidemiology examines outcomes from a population perspective. In many epidemiological studies, geographic populations are the unit of interest, whether these involve populations of a country, state, city, or neighborhood. But as epidemiology is concerned with the occurrence of illness in populations, the concept of population can be interpreted as any group at risk: females, children, or persons with Down syndrome. Because the concept of risk within a population is essential both in defining and in interpreting epidemiological studies and results, enormous attention is paid to the gender, ethnic, racial, familial, socioeconomic, urban‐suburban‐rural, and other characteristics of the sample under study. Contrast this population‐based strategy to the approach usually adopted in psychological studies. In most such studies, researchers examine small numbers (20–40) of children or adults. Such small numbers of subjects, while acceptable for answering many types of ‘‘main eVects’’ questions, are usually inadequate to examine various interaction eVects. Such studies also generally involve samples of convenience, individuals who, on the basis of advertisements or word of mouth, volunteer to participate in any given study. Of particular concern to epidemiologists (with their population‐based focus) is whether these small samples truly represent the larger population in terms of their subject or family characteristics. If not, then epidemiologists cannot later infer from this sample the actual risk of occurrence of illness or other health‐related events in the larger population. B.

A Focus on Health‐Related Outcomes

Many epidemiological studies observe the occurrence of illnesses, deaths, hospitalizations, or other health‐related outcomes. Within this focus on health‐related outcomes, researchers expend large amounts of energy and thought to providing exact definitions of a case. Recently, for example, researchers have worked hard to define low birthweight in newborns, as well as

6

Robert M. Hodapp and Richard C. Urbano

to establish exact diagnostic criteria for disease states such as pneumonia, AIDS, or cancer, or for conditions such as intellectual disabilities or autism. Two further issues–complications arise when discussing epidemiology’s focus. Historically, epidemiology’s focus on illness and disease states ensured that its concerns were more or less concrete: How many cases of, for instance, pneumonia occur in a particular population, or within a specific age, gender, or geographic location? In many ways, one of the allures of epidemiology is that, in contrast to the more subjective measures used in much psychological research, epidemiological studies focused on generally easy‐to‐measure, concrete outcomes. In recent years, however, such distinctions may be fading. As noted in Mezzich and Ustin’s (2005) definition, epidemiology has increasingly been concerned with ‘‘the study of health and disease as a full spectrum.’’ Such a full spectrum, health‐and‐disease focus allows epidemiologists to include as outcomes subjective ratings of one’s health, well‐being, and feelings of depression. In one sense, such opening up of epidemiological studies is useful—the field may have been guilty of an overly rigid focus on diseases in past years. At the same time, however, now coming to the fore are all of the tricky measurement issues inherent in any diYcult‐to‐measure construct. A second issue concerns outcomes that, while they may relate to health and disease, are not exactly health outcomes. For example, many large‐scale epidemiological studies examine such things as divorce or employment. Strictly speaking, these outcomes might better come under the purview of demography or economics. But in recent years, some epidemiologists have called for a closer alliance between epidemiology, with its health–disease focus, and demography, with its focus on other characteristics of a population (Susser & Bresnahan, 2001). Although by no means all epidemiologists would consider, for example, divorce and its predictors or sequelae as a proper topic of study within epidemiology, many would. C.

A Focus on Causes or Probable Causes of Such Health‐Related Outcomes

Epidemiological studies are designed to describe, explain, and predict the occurrences of the outcomes; the ultimate goal is to connect outcomes with their predictors. The epidemiologist, however, is not focused specifically on the outcome of a particular individual. Instead, epidemiologists think about outcomes in population terms: how to reduce risk across the population so that the proportion of cases in a population diminishes. Thus, predictors may include individual risk behaviors (e.g., riding in a car without seatbelts) or more broadly based social risk factors [e.g., low socioeconomic status (SES), lack of access to health care]. In all cases, epidemiology has as its

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

7

goal connections between outcomes and predictors that provide clues as to possible cause(s) of the final outcome. Such clues sometimes relate to already‐specified disease mechanisms, sometimes to less well understood societal processes. A good example might involve one’s SES. Many conditions and diseases occur more often in low SES individuals. Within this sample of low‐income people, an epidemiologist would test hypotheses to identify which specific characteristics might indicate a direct pathway to the outcome. The epidemiologist might examine diet, environment, low education, or lack of health insurance for each’s relationship to the cause or progression of a disease. In this regard, it seems useful to distinguish between risk indicators and risk mechanisms (Rutter, Pickles, Murray, & Eaves, 2001). Risk indicators involve variables—such as low SES—that are simply correlated with bad outcomes. It is necessary (but not suYcient) to identify such risk indicators. The truly interesting studies take the next step, moving from an indicator to the exact risk mechanisms that are operative. In relation to diVerent outcomes, low SES might serve as a proxy for poor nutrition, a lack of health care, or households in which children do not have books to be read to them or in which parents work two jobs, not allowing enough time to interact with their children.

D.

A Focus on Intervention, Public Health, and Public Policy

Epidemiology is not solely an academic enterprise. Throughout, the field’s focus on determining amounts and correlates of health–disease outcomes has as its goal the prevention, amelioration, or treatment of those diseases. One might recall the U.S. Safety Council’s ‘‘Back to Sleep’’ advertisement program to get parents to sleep their infants on their backs, thereby reducing sudden infant death syndrome (SIDS). Other, population‐wide advertisements, inducements, and information‐sharing campaigns have all been tried, either to help eliminate diseases or to encourage practices that foster better health (e.g., exercise, nutritious eating). Most such campaigns arise from results derived from epidemiological studies. Again, the potential connections of epidemiology to demography must be noted. Just as studies of heart disease, diabetes, and obesity have led to prescriptions about not smoking, diet, and exercise, so too might public policy campaigns arise from epidemiological studies that examine what have traditionally been considered as more ‘‘social ills.’’ Thus, studies of such phenomena as divorce, school dropout, or unemployment might lead to policies designed to aVect individual behaviors or local, state, or national policy. Although not, strictly speaking, within the purview of epidemiology,

8

Robert M. Hodapp and Richard C. Urbano

such phenomena can be treated similarly to the more health‐related outcomes of interest in traditional epidemiological studies. III.

DEVELOPMENTAL EPIDEMIOLOGY

As a subspecialization of the broader field of epidemiology, developmental epidemiology focuses more intensively on changes in humans, time (both individual and historical), and the changes into and out of problem conditions. In many ways, developmental epidemiology makes more dynamic the original epidemiological focus on predictors and correlates of disease states. Before describing developmental epidemiology itself, it is important to consider the field’s ties to developmental psychopathology. As defined during the 1980s by Dante Cicchetti (1984, 1993) and Sroufe and Rutter (1984), developmental psychopathology goes beyond psychiatry, child psychiatry, or child clinical psychology. In a well‐known definition, Sroufe and Rutter (1984) defined developmental psychopathology as ‘‘the study of the origins and course of individual patterns of behavioral maladaptation . . .’’ (p. 18). In this sense, it is not enough to examine cross‐sectionally diVerent‐ aged children with a specific condition, or to compare children with and without a Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM‐IV) diagnosis. Instead, interest centers on the entry into and out of pathology, as well as on the predictors of such pathology at diVerent ages, levels of functioning, and groups or genders. Like the addition of ‘‘developmental’’ to the field of epidemiology, then, a focus of developmental psychopathology added a dynamic, changing component to child psychopathology. What, then, are the main tenets of a developmental epidemiological approach? We propose the following. A.

Appreciating Developmental Issues

In ways similar to the distinction between developmental psychopathology and either child clinical psychology or child psychiatry, the diVerences between developmental epidemiology and epidemiology somewhat consist of diVering emphases. For example, most epidemiologists would consider themselves interested in examining the number and type of children’s mental or physical health problems, and might search for predictors of such problems. Such studies might even occur at diVerent points during the childhood years. In contrast, developmental epidemiologists highlight the notion of constant, ongoing change throughout childhood (and even over the adult years). In this sense, developmental epidemiologists are adding a

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

9

second layer of development. Thus, ‘‘developmental epidemiology can be seen as concerned with the interaction between two developmental processes: of the organism (the child) and of the disease’’ (Costello & Angold, 2006, p. 49). One good example of such simultaneous developments might involve childhood depression. It is now fairly well established that postpubertal girls are at high risks for depression, whereas depression is less often seen in pre‐ or postpubertal boys or in prepubertal girls. But what, exactly, makes postpubertal girls more prone to depression? To what extent are adolescence’s many hormonal, physical, or social changes involved, and what might the eVects be of unsupportive families or other diYcult environments? Certain of these environments have already been implicated in the onset of depressive episodes. Simmons and Blyth (1987) found that children (both sexes) are most prone to symptoms of depression during their major growth spurt in adolescence (called the period of ‘‘peak height velocity’’ or PHV), particularly when faced with stressful events. Within American schools, most children transition from middle school to high school between eighth and ninth grades, which just happens to be the period of PHV for most girls (but few boys). Although one should not ascribe all cases of depression in adolescent girls to changing schools during this critical period, such changes of school (known to be a particularly stressful time for many children) may be a contributing or enhancing factor. The point is that developmental epidemiology takes seriously the child’s own development. By doing so, one gains a greater appreciation for the many physical, mental, hormonal, social, educational, and other normative changes that children experience as they develop. These normal developmental changes provide the larger context within which the types, amounts, and predictors of various problems and conditions can be examined. B.

Identifying the Developmental Nature of Risk Factors

By fully employing the perspective of developmental epidemiology, one also begins to appreciate the changing nature of environmental influences at diVerent ages and periods of development. Essentially, a particular aspect of the environment may have a specific eVect at one period during development and diVerent (or no) eVects at other developmental periods. Consider the eVect of maternal depression on child behaviors. When mothers are depressed during the child’s first year of life, motor development seems most aVected. Later, during the child’s second year, language but not motor development seems most aVected (Hay, Pawlby, Angold, Harold, & Sharpe, 2003). Since presumably maternal behaviors during depressive episodes are similar at the two age‐periods, a mother’s depression and her

10

Robert M. Hodapp and Richard C. Urbano

behaviors to the infant while depressed vary in their specific eVects on diVerent‐aged children. Developmental epidemiologists highlight several distinctions involving exposure to one or another ‘‘toxic event’’ and later health or mental health outcomes. They distinguish between the age of first exposure to some environmental event, the time since that exposure, the length of exposure, and the amount (dose) of exposure. When examining mental health outcomes of child abuse, experiencing a depressed mother, or growing up in extreme poverty, such diVerent event characteristics seem more or less important for diVerent outcomes. Just as children themselves are developing and changing over the childhood years, so too do the eVects of identical events diVer based on their onset age, length of exposure, intervening years, or dosage. C.

Appreciating the Bidirectionality of Influence on Child, Parent, and Family Outcomes

Historically, developmental psychologists have been ‘‘systems thinkers,’’ conceptualizing from general systems theory (Bertalannfy, 1968) to the phenomenon of human development (SameroV, 1995). Inherent in systems thinking is the idea of the interactions–transactions within and among diVerent levels of the system. In humans, communication occurs across levels both within and outside of the developing child. In this vein, developmental theorists from the late 1960s on have considered parents, siblings, families, schools, and neighborhoods as comprising the many environments amenable to developmental analyses. Bell (1968) emphasized the ways in which infants and caregivers mutually aVected one another; SameroV and Chandler (1975) focused on such interactions over time (i.e., transactions); and Bronfenbrenner (1979) noted the various ecologies of childhood, the fact that children reside within families, which themselves live in neighborhoods, towns, and countries (and that each level of the surrounding environment potentially aVects others). Each perspective embodies for developmentalists the more general notion that various levels of a system interact with one another. In research using developmental epidemiology, then, children comprise one—but not the only—potential outcome of interest. While the child’s entry into, say, depression would constitute one outcome of interest, the eVects of depressed children on their parents and families would constitute other outcomes. One might examine as outcomes maternal or paternal depression, divorce in the couple, or a family’s changing income, employment patterns, or moving. Some of these outcomes might relate to physical or mental health (e.g., health of the mothers, father, or siblings), others to marriage, economic, educational, or other issues. All such other‐than‐child

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

11

outcomes would come under the purview of developmental perspectives that are more interactional, transactional, and ecological. IV.

ISSUES SPECIFIC TO MR/DD

Until now, we have discussed developmental epidemiology with little regard to children with MR/DD. We have simply assumed that both the epidemiological and the developmental terms of developmental epidemiology can simply be transported to children with mental retardation. Unfortunately, the field of MR/DD itself contains a host of tricky issues, many of which complicate developmental epidemiological work. Some of these problems are shared with developmental epidemiological studies within child psychiatry and clinical child psychiatry, whereas others pertain solely to the MR/DD field. We now discuss three such issues. A.

Mental Retardation, Type of Mental Retardation, and the Problem of Caseness

Although all epidemiological studies struggle with the problem of ‘‘caseness,’’ this problem seems particularly salient when considering MR/ DD. Simply put, what constitutes a ‘‘true case’’ of a person with mental retardation? For several decades, the so‐called ‘‘three‐factor definition’’ has been used to define and diagnose individuals with mental retardation, with the three factors involving: significantly subaverage intellectual functioning . . ., concurrent deficits or impairments in present adaptive functioning . . .,

and

the onset is before age 18 years (American Psychiatric Association,

2000, p. 49). The first factor involves deficits in intellectual functioning. Using appropriate, standardized psychometric tests [e.g., the latest Stanford–Binet, Wechsler, Kaufman, or other intelligent quotient (IQ) tests], individuals are considered to fall within the mental retardation range when their IQs are at 70 or below. Due to errors of measurement, most diagnostic manuals allow for some leeway in this ‘‘IQ‐70 criterion,’’ usually up to IQ‐75. Still, significant intellectual impairment constitutes one criterion for a diagnosis of mental retardation. But such intellectual deficits should also involve a ‘‘real‐world’’ component. The second criterion therefore involves deficits in adaptive behavior.

12

Robert M. Hodapp and Richard C. Urbano

Deficits in adaptive behavior involve a lessened ability to perform daily activities required for personal and social suYciency. In the newly revised Vineland Adaptive Behavior Scales, Sparrow, Balla, and Cicchetti (2005) examine three domains: communication, or communicating one’s needs to others; daily living skills, or performing such tasks as eating, dressing, grooming, and toileting; and socialization, or following rules and working and playing with others. Although other diagnostic manuals (e.g., American Association on Mental Retardation, 2002) propose diVerent adaptive domains, recent diagnostic manuals include adaptive as well as intellectual impairments for a diagnosis of mental retardation. The third criterion involves onset during the childhood years. Unlike both the intellectual and adaptive criteria, childhood onset has received little criticism. To most professionals, mental retardation should be diVerentiated from problems associated with Alzheimer’s or other degenerative diseases, adult‐ onset traumatic brain damage, or other diseases or conditions that occur during the adult years (Hodapp, Maxwell, Sellinger, & Dykens, in press). In contrast to this three‐factor definition, most epidemiological studies diagnose mental retardation using an ‘‘IQ‐only definition’’ (usually, with IQ <70). In a well‐known study, Yeargin‐Allsopp, Murphy, Oakley, and Sikes (1992) used only the IQ‐criterion (i.e., IQ <70) to identify those Atlanta schoolchildren considered to have mental retardation. Were Yeargin‐ Allsopp et al. (1992) to have added adaptive deficits, fewer children would have been diagnosed, as children with IQ’s below 70 but who showed higher levels of adaptive behavior would not have qualified for the MR diagnosis. Even compared to depression, conduct disorder, or other hard‐to‐define conditions, then, controversy continues to surround which individuals should be considered to have (and not to have) mental retardation. In reviewing this issue, Leonard and Wen (2002) concluded that ‘‘Taxonomy in this field is particularly diYcult because professionals and consumers come from a range of backgrounds and have diVerent purposes such as advocacy, education, medical care, and service provision’’ (p. 120). Although there may ultimately be no agreed‐upon answers regarding true cases of persons with mental retardation, one recent change may lessen the confusion. This change involves the movement within the mental retardation field to examine individuals with known etiologies of mental retardation. Instead of examining individuals with mild, moderate, severe, or profound mental retardation—whose mental retardation results from many causes— researchers have recently examined groups with specific etiologies separately. Over the past 20 years, the numbers of studies have increased exponentially on disorders such as Williams syndrome, fragile X syndrome, Prader–Willi syndrome, and fetal alcohol syndrome (Hodapp & Dykens, 2004). For many of these syndromes, studies have identified etiologically related strengths and

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

13

weaknesses, proneness to specific psychiatric disorders, and times of more rapid or slowed cognitive, linguistic, and social development (Dykens, Hodapp, & Finucane, 2000 for a review). The rise of etiology‐based studies is among the most interesting advances within the entire MR/DD field (Hodapp, 2004). Although to date most etiology‐based studies have not been performed with developmental epidemiology in mind, important epidemiological findings are already arising concerning various etiological groups. Consider two examples. In a study concerning Down syndrome, Yang, Rasmussen, and Friedman (2002) examined the deaths of over 17,000 individuals with Down syndrome across the United States, over the period from 1983 through 1997. Overall, the median age at death increased from 25 years in 1983 to 49 years in 1997. The largest increase in age at death occurred in the early 1990s, and few diVerences were noted from region to region. The authors speculated that increased survival rates during the 1983–1997 period may relate to the lessening of institutionalization (especially of children) and to better medical practice, particularly to the more timely provision of cardiac surgery for children with Down syndrome. The second example concerns Prader–Willi syndrome, a disorder involving an abnormality on chromosome 15 (either a deletion or a disomy that involves receiving two chromosome 15s from the mother). Along with temper tantrums and obsessions–compulsions, this disorder is characterized by hyperphagia (extreme overeating) and resulting obesity and obesity‐ related health problems (Dykens, 1999). Recently, Whittington et al. (2001) examined early death and morbidity in one UK health region (Anglia and Oxford). The incidence of Prader–Willi syndrome was 1 per 22,000 births and the death rate was over 3% per year (Butler et al., 2002), with most cases of early death related to complications of obesity (e.g., type II diabetes, respiratory, and circulatory problems; Whittington et al., 2001). Although preliminary, such etiology‐based studies feature several advantages over studies examining heterogeneous groups with mental retardation. As opposed to studies of heterogeneous mental retardation, studies examining genetic disorders allow the researcher clear genetic criteria for who is (and is not) a case. Many genetic syndromes have corresponding ICD‐9 and other diagnostic codes, which allow for research that utilizes hospital, state administrative, and other large‐scale databases. Many controversies are thereby avoided, and relatively ‘‘clean’’ research groups are generated. B.

Disentangling Age Versus Level of Functioning

Another concern specific to populations with mental retardation is the discrepancy between the individual’s chronological age (CA) and level

14

Robert M. Hodapp and Richard C. Urbano

of functioning. In most groups within the normal range of intelligence, the child’s chronological and mental ages (MAs) are roughly the same. Even when considering children who are depressed, conduct disordered, or schizophrenic, little attention need be paid to lower mental versus CAs. When dealing with populations with mental retardation, however, diVerences exist (by definition) between the child’s CA and MA, or other level‐of‐ functioning age‐equivalent measure. So far, such issues have arisen mainly in terms of purely psychological domains. Thus, much debate has centered around appropriate control or contrast groups in psychological studies involving individuals with diVerent syndromes (Hodapp & Dykens, 2001). Is it better to match or equate on MA or on CA? Particularly, given the possibility that children with certain genetic disorders may show ‘‘spared’’ (or CA‐level) performance in certain domains, such issues of MA versus CA‐ matches have become increasingly complex. At least until now, such questions have been limited to studies of cognitive, linguistic, adaptive, and other domains of psychological functioning. But the discrepancy between higher chronological and lower MAs also reflects more general problems in performing any research with individuals with mental retardation. One issue concerns measurement. To what extent can children or adults with mental retardation accurately report how depressed they feel (a diagnostic criterion for depression), or about their confused or distorted thinking (one of the criteria for schizophrenia; Dykens, 2000)? How well can they report on their perceived quality of life, attitudes, happiness, or other subjective states? One might even wonder about the degree to which certain children or adults can report on presenting symptoms of physical diseases. More generally, however, issues persist concerning the degree to which persons with MR/DD actually experience certain ‘‘cognitive‐related’’ problems, if their experiences are the same as those of nondisabled individuals, and whether identical criteria should be used to diagnose such problems. Partly for these reasons, several researchers in the area of dual diagnosis— involving both mental retardation and mental illness—have developed caregiver‐report instruments of maladaptive behavior‐psychopathology that have been specifically designed and normed on children and adults with mental retardation (Aman & Singh, 1994; Einfeld & Tonge, 1995; Reiss, 1988). In slightly diVerent ways, each instrument measures depressed mood, distorted or confused thinking, aggressive behavior, and other emotional or behavioral problems. In making DSM‐like psychiatric diagnoses, clinicians often give less weight to the self‐reports of depressed aVect and more weight to such vegetative signs as over‐ or undersleeping and eating (Dykens, 2000 for a review). Although some headway has been made, the existence, measurement, and experiences of mental and

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

15

physical symptoms continue to pose diYcult problems for studies of persons with MR/DD. C.

Groups with Mental Retardation: Outcome or Population?

Should groups with mental retardation constitute the outcome of predictor variables or are they instead the population to be studied? In studies so far, persons with mental retardation are probably more often the outcome. In those studies, such variables as low birthweight, APGAR scores, no prenatal care, low maternal education, older maternal age, or the child’s environment have been used to predict to later child outcomes. These outcomes have ranged widely across studies, from mental retardation or developmental disabilities (Chapman, Scott, & Mason, 2002; Kochanek, KabacoV, & Lipsitt, 1990); to specific language impairments (Stanton‐Chapman, Chapman, Bainbridge, & Scott, 2002; Tomblin, Smith, & Zhang, 1997); to placement in special education classes or identification as being in need of special education services (Andrews, Goldberg, Wellen, Pittman, & Struening, 1995; Goldberg, McLaughlin, Grossi, Tytun, & Blum, 1992). Many of these studies have used hospital, state, or other administrative records—often of many thousands of individuals—to identify the types and strength of early predictors on later developmental outcomes (Boussy & Scott, 1993; Redden, Mulvihill, Wallander, & Hovinga, 2000). Although it is essential to have studies that examine children who later come to have disabilities, it is also necessary to examine outcomes in clearly specified etiological groups. As noted earlier, studies of physical and mental health are in their infancy in Down syndrome, Prader–Willi syndrome, Williams syndrome, and other of the 750þ genetic mental retardation syndromes. Beyond health, we know little of these children’s specific needs as they get older. Even in Down syndrome, a disorder well known for higher prevalence rates of heart defects, Alzheimer’s disease, leukemia, and other physical diseases (Cohen, 1996; Roizen, 2003; Roizen & Patterson, 2003), knowledge is limited beyond the basics. For virtually all of these diseases, the field knows little about the timing or course of the disease, the eVectiveness of diVerent treatments and interventions, or how often (or when) such individuals are hospitalized (Hodapp, Urbano, & So, 2006). Other issues remain totally unexamined, particularly those involving connections between health‐disease and family SES, place of residence (rural‐urban‐suburban), parent education, or family health history. In pondering the application of developmental epidemiology to mental retardation, then, we see that many diYcult issues exist. Few developmental epidemiologists have tackled such issues head on, or, if they have, their discussions have only sporadically been informed by similar debates within

16

Robert M. Hodapp and Richard C. Urbano

the larger field of MR/DD research. We now turn to several recent advances and their potential for aiding the emergence of developmental epidemiology in MR/DD.

V.

ADVANCES AND FUTURE DIRECTIONS

Given the lack of knowledge about many basic issues in MR/DD, the potential seems almost unlimited for approaches based on developmental epidemiology. But beyond the need for a more widespread, mature science, several recent advances must be noted. Over the next few decades, these advances should allow developmental epidemiological work to flourish. A.

Advance #1: Increasing Existence of and Ability to Link Large‐Scale Databases

As each of us proceeds through life, we accumulate a set of records about both ourselves and our transactions. Although available records vary in diVerent countries, states, and localities, the following are just some of the types of information that we create: birth, immunization, healthcare, school, psychological testing, income, tax, professional licensing, ownership, warranty, purchases, military service, marriage, divorce, travel, and death. Given appropriate ethical safeguards, many of these records are available to be linked and can therefore be used to examine various topics in the developmental epidemiology of mental retardation. Before using such records, however, one must first perform several functions. The most important of these functions involves linkage, or the merging of records. As discussed in Urbano (this issue), data‐linkage involves four discrete steps. The researcher must first acquire and ‘‘clean’’ the data, searching for mistakes, deciding on how to handle missing cases or variables within a case, and making certain that the same answers are treated identically from one dataset to another. Second, matching must be performed such that two or more of the same person’s records are linked together. Third, decisions must be made to decide if two or more linked records actually belong to the same person. Finally, the now‐linked records must be transformed into SPSS, SAS, or other ‘‘user‐friendly’’ language so that most social scientists could analyze these data. Although much simplified in this description, linking together thousands or even millions of individual records takes countless hours (usually weeks or months) and forms the backbone to any large‐scale records study.

DEVELOPMENTAL EPIDEMIOLOGY OF

B.

MR/DD

17

Advance #2: Ability to Perform First‐ and Second‐Order Linkage Between and Within Different Datasets

Although data linkage involves general techniques that have been known for many years (Baldwin, Acheson, & Graham, 1987; Howe & Lindsay, 1981; Newcombe, 1988), recent advances extend the use of both individual and family wide records. In individual or first‐order linkage, diVerent records can be linked on the same individual. Thus, a child’s Birth Record and record from the birth hospitalization (i.e., Hospital Discharge Record) can be linked to examine the eVects on length of hospitalization of low birthweight or low APGAR scores. Similarly, one can link together diVerent instances of the same type of record for the same individual (e.g., the person’s many diVerent hospital visits). One might even combine linking procedures such that a child’s Birth Record can be linked to that child’s Hospital Discharge Records, allowing for a longitudinal analysis of the birth predictors of later hospitalizations. In addition to linking within and across record‐types for a single individual, one might also link records across related individuals. By using the mothers’ and fathers’ Social Security Numbers as linking variables, entire families can be identified and linked together. These linked records of diVerent family members—which has been called ‘‘second‐order linkage’’ (Tu & Mason, 2004)—allows one to order horizontally all children in a family. One might determine the eVects (positive or negative) that children with disabilities have on their nondisabled siblings, or whether certain events (hospitalization, death, divorce) aVect mothers, fathers, or siblings, both at the time of their occurrence or years later. By employing first‐ and second‐order linkage techniques, one begins to approach the kinds of data necessary for developmental epidemiological analyses. Consider the many diVerent issues involved in determining the eVects (good or bad) of growing up with a sibling with disabilities. Using linked administrative records for an entire state, the researcher can examine as subjects many hundreds of families of children with a particular condition (e.g., Down syndrome, spina bifida). Using such large samples of linked administrative data, the eVects of the child with disability can be examined on nondisabled siblings who are older versus younger; same or diVerent gender; first, last, or middle born; or who are separated in age by varying numbers of years. Due to the diYculty in teasing apart such influences in small‐scale studies, each of these family structure variables remain little examined within the sibling field (Hodapp, Glidden, & Kaiser, 2005). Other issues pertain to the eVects of diVerent characteristics of ‘‘toxic events.’’ Recall the finding that, for diVerent child psychiatric disorders and diVerent events during childhood, the most important aspect of the

18

Robert M. Hodapp and Richard C. Urbano

harmful event involved the event’s timing, duration, length of time since it occurred, or intensity (Costello & Angold, 2006). With data available on deaths, hospitalizations, divorce, and other discrete events, the timing and length of time since the event can all be examined for children with mental retardation and their families. Increasingly, then, linked, large‐scale datasets allow for examinations that embody many of the basic principles of research in developmental epidemiology. C.

Advance #3: Mixing Basic Research with Public Health‐Intervention Perspectives

Like many ‘‘mixed’’ disciplines, developmental epidemiology features both basic and applied perspectives. Basic researchers are interested in questions relating to the interplay of development and environment, with the hope of identifying and understanding those risk mechanisms leading to specific disease states at specified developmental periods (Costello et al., 2006). Witness the interest in the timing, duration, time since, and intensity of toxic events in the developmental psychopathology of child psychiatric conditions. At the same time, however, more applied researchers emphasize what has been called ‘‘public health epidemiology’’ (Earls, 1979). These researchers and interventionists focus more on surveillance, screening, and intervention. In a recent review, Costello, Egger, and Angold (2005) examined the impact of diseases in terms of ‘‘disability‐adjusted life‐years,’’ or DALYs, in young adults aged 15–44 years (data were unavailable for children alone). In line with their interests in the burdens caused by psychiatric illnesses, they noted that all but 1 of the 10 conditions producing the highest amount of DALYs either consisted of or was related to a psychiatric disorder. Among these top 10 were such DSM‐disorders as depression, schizophrenia, bipolar disorder, and obsessive–compulsive disorder (OCD); among problems strongly related to psychiatric conditions were alcohol use, road and traYc accidents, and self‐inflicted injuries. Such measurements of the burden of a disease could also be performed for families of such individuals, as well as for service‐ delivery systems and for patterns of service usage (Costello et al., 2005). In examining developmental epidemiological studies in developmental disabilities, one finds that some of the basic questions have been addressed, far fewer of the applied questions. Thus, while a fair number of studies examine the prenatal, birth, and socioeconomic markers of mental retardation, specific language impairment (SLI), and certain other disorders, many fewer have examined the burden of disease, service‐delivery, and service usage. As before, the focus has almost exclusively fallen on precursors of various disabilities, not on what happens to those individuals who have a particular

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

19

disorder. This issue, what might be called ‘‘precursor versus sequelae,’’ is particularly noteworthy for mental retardation‐developmental disabilities. Simply stated, disability conditions do not travel alone. Thus, approximately 50% of infants with Down syndrome are born with congenital heart defects, a disorder that, while treatable, is also related early on with high levels of hospitalization for pneumonia, bronchitis, and bronchiolitis (So, Urbano, & Hodapp, submitted for publication). High rates of heart and circulatory problems are also found in Williams syndrome (Pober & Dykens, 1996), and obesity is seen at highly increased rates among persons with mental retardation in general (Rimmer & Yamaki, 2006), as well as with Down syndrome (Prasher, 1995) and Prader–Willi syndrome (Whittington et al., 2001). In essence, we need to increase our understanding of the public health epidemiology of developmental disabilities, both for those who may develop and for those who already have these disorders. Badly needed is sustained research and clinical attention to screening, prevalence–incidence rates in diVerent groups and locations, and the eYcacy of a wide variety of public health, educational, and medical interventions. The ‘‘fix it’’ side of developmental epidemiology seems in need of much more work. VI.

CONCLUSIONS

The field of developmental epidemiology of mental retardation is at an interesting point. Although less well developed than comparable studies in child psychopathology and other disciplines, the field seems ready to burst forth, to realize—over the next few decades—a potential that until now has remained only partially fulfilled. Fully realizing this potential would seem to entail three types of studies. First, we need greater amounts of basic information. To quote a well‐known phrase, ‘‘Epidemiology counts’’ (Freedman, 1984). In the mental retardation field, however, counting has not always been the norm. Instead, for too long, researchers and practitioners within the MR/DD field have relied on general senses of problems, on clinical hunches, on longstanding experience untempered by actual numbers. But given the rise of a formal subfield as well as of technical and technological advances, we predict that both better and more complete counting will soon occur for a wide variety of medical and nonmedical events. To give only a few examples, recent studies have examined mortality and morbidity in Prader–Willi syndrome (Whittington et al., 2001), of changing life expectancies in Down syndrome (Yang et al., 2002), and of early hospitalizations of the children themselves and of divorce in families of persons with Down syndrome (So et al., submitted for publication; Urbano & Hodapp, submitted for publication). More, better, and more complete counting seems imminent.

20

Robert M. Hodapp and Richard C. Urbano

Beyond counting per se, we predict that future studies will improve in their understandings of the mechanisms that lie at the heart of academically oriented developmental epidemiology. Which early markers or mechanisms predispose individuals with certain genetic etiologies (or with mental retardation in general) to later health problems? Which demographic characteristics of parents, siblings, or families make more likely better or worse family coping? Given that a certain ‘‘toxic event’’ does predispose to a negative outcome within individuals with MR/DD or their families, is it the timing, time since, dosage, or duration of that event that seems most detrimental? Do such patterns of predisposition mirror those found among nondisabled individuals or their families, or is something diVerent occurring within individuals with disabilities and/or their families? All are answerable questions in the very near future. Finally, one must appreciate the surveillance—public health aspects of work on developmental epidemiology of MR/DD. In recent years, the highest‐profile debate in this regard concerns whether the rate of autism is rising (Fombonne, 2003 for a discussion), but other issues are also apparent. And, as in each of these topics, one can examine whether certain environments predispose to later problems using MR/DD as either the outcome or the population. We end this chapter—and begin this volume—with a note of cautious optimism. To us, developmental epidemiology of MR/DD, while not yet fully developed, is a field that is poised, primed, and ready to go forward. New findings are arising, more researchers are working on these issues, and increasing numbers of databases allow for the examination of epidemiological questions within the MR/DD population. New statistical techniques are being developed every year, even as computing power multiplies exponentially. Although one may be partial to any number of metaphors, the field of developmental epidemiology of mental retardation is moving forward, advancing, and progressing. REFERENCES Aman, M. G., & Singh, N. N. (1994). Aberrant behavior checklist—community supplementary manual. East Aurora, NY: Slosson Educational Publishers. American Association on Mental Retardation (2002). Mental retardation: Definition, classification, and systems of supports (10th ed.).Washington, DC: Author. American Psychiatric Association (2000). Diagnostic and statistical manual of mental disorders (4th ed., Text Revision). Washington, DC: Author. Andrews, H., Goldberg, D., Wellen, N., Pittman, B., & Streuning, E. (1995). The prediction of special‐education placement from birth certificate data. American Journal of Preventive Medicine, 11, 55–61.

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

21

Baldwin, J. A., Acheson, E. D., & Graham, W. J. (1987). Textbook of medical record linkage. Oxford, New York, Toronto: Oxford University Press. Bell, R. Q. (1968). A reinterpretation of direction of eVects in studies of socialization. Psychological Review, 75, 81–95. Bertalannfy, L. V. (1968). General system theory: Foundations, development, application. New York: Braziller. Blackford, J. U. (this volume). Statistical issues in developmental epidemiology and developmental disabilities research: Confounding variables, small sample size, and numerous outcome variables. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 93–120). New York: Elsevier. Boussy, C. A., & Scott, K. G. (1993). Use of data‐base linkage methodology in epidemiologic studies of mental retardation. International Review of Research in Mental Retardation, 19, 135–161. Bronfenbrenner, U. (1979). The ecology of human development. Cambridge, MA: Harvard University Press. Butler, J. V., Whittington, J. E., Holland, A. J., Boer, H., Clarke, D. J., & Webb, T. (2002). Prevalence of, and risk factors for, physical ill‐health in people with Prader‐Willi syndrome: A population‐based study. Developmental Medicine and Child Neurology, 44, 248–255. Chapman, D. A., Scott, K. G., & Mason, C. A. (2002). Early risk factors for mental retardation: Role of maternal age and maternal education. American Journal on Mental Retardation, 107, 46–59. Cicchetti, D. (1984). The emergence of developmental psychopathology. Child Development, 55, 1–7. Cicchetti, D. (1993). Developmental psychopathology: Reactions, reflections, projections. Developmental Review, 13, 471–502. Cohen, W. I. (Ed.) (1996). Health care guidelines for individuals with Down syndrome (Down syndrome preventive medical check list). Down Syndrome Quarterly, 1(2), 1–10. Costello, E. J., & Angold, A. (2006). Developmental epidemiology. In D. Cicchetti & D. Cohen (Eds.), Developmental psychopathology, Vol. 1, Theory and method (2nd ed., pp. 41–75). New York: John Wiley & Sons. Costello, E. J., Egger, H., & Angold, A. (2005). 10‐year research update review: The epidemiology of child and adolescent psychiatric disorders: I. Methods and public health burden. Journal of the American Academy of Child and Adolescent Psychiatry, 44, 972–986. Costello, E. J., Foley, D. L., & Angold, A. (2006). 10‐year research update review: The epidemiology of child and adolescent psychiatric disorders: II. Developmental epidemiology. Journal of the American Academy of Child and Adolescent Psychiatry, 45, 8–25. Dykens, E. M. (1999). Prader‐Willi syndrome. In H. Tager‐Flusberg (Ed.), Neurodevelopmental disorders (pp. 137–154). Cambridge, MA: MIT Press. Dykens, E. M. (2000). Psychopathology in children with intellectual disability. Journal of Child Psychology and Psychiatry, 41, 407–417. Dykens, E. M., Hodapp, R. M., & Finucane, B. M. (2000). Genetics and mental retardation syndromes: A new look at behavior and interventions. Baltimore, MD: Paul H. Brookes Publishers. Earls, F. (1979). Epidemiology and child psychiatry: Historical and conceptual development. Comprehensive Psychiatry, 20, 256–269. Einfeld, S. L., & Tonge, B. J. (1995). The developmental behavioural checklist: The development and validation of an instrument to assess behavioural and emotional disturbance in children and adolescents with mental retardation. Journal of Autism and Developmental Disorders, 25, 81–104.

22

Robert M. Hodapp and Richard C. Urbano

Einfeld, S. L., Tonge, B. J., Gray, K., & Taffe, J. (this volume). Evolution of symptoms and syndromes of psychopathology in young people with mental retardation. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 247–265). New York: Elsevier. Fombonne, E. (2003). Epidemiological surveys of autism and other pervasive developmental disorders: An update. Journal of Autism and Developmental Disorders, 33, 365–382. Freedman, D. X. (1984). Psychiatric epidemiology counts. Archives of General Psychiatry, 41, 931–933. Goldberg, D., McLaughlin, M., Grossi, M., Tytun, A., & Blum, S. (1992). Which newborns in New York City are at risk for special education placement? American Journal of Public Health, 82, 438–440. Hay, D. F., Pawlby, S., Angold, A., Harold, G., & Sharpe, D. (2003). Pathways to violence in the children of mothers who were depressed postpartum. Developmental Psychology, 39, 1083–1094. Hodapp, R. M. (2004). Behavioral phenotypes: Going beyond the two‐group approach. International Review of Research in Mental Retardation, 29, 1–30. Hodapp, R. M., & Dykens, E. M. (2001). Strengthening behavioral research on genetic mental retardation disorders. American Journal on Mental Retardation, 106, 4–15. Hodapp, R. M., & Dykens, E. M. (2004). Studying behavioral phenotypes: Issues, benefits, challenges. In E. Emerson, C. Hatton, T. Parmenter, & T. Thompson (Eds.), International handbook of applied research in intellectual disabilities (pp. 203–220). New York: John Wiley & Sons. Hodapp, R. M., Glidden, L. M., & Kaiser, A. P. (2005). Siblings of persons with disabilities: Toward a research agenda. Mental Retardation, 43, 334–338. Hodapp, R. M., Maxwell, M. A., Sellinger, M. H., & Dykens, E. M. (in press). Persons with mental retardation: Scientific, clinical, and policy advances. In T. Plante (Ed.), Abnormal psychology in the 21st century. New York: Praeger/Greenwood Publishers. Hodapp, R. M., Urbano, R. C., & So, S. A. (2006). Using an epidemiological approach to examine outcomes aVecting young children with Down syndrome and their families. Down Syndrome: Research and Practice, 10, 83–93. Hogan, D. P., Msall, M. E., & Drew, J. A. (this volume). The developmental epidemiology of mental retardation and developmental disabilities. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 213–245). New York: Elsevier. Howe, G. R., & Lindsay, J. (1981). A generalized iterative record linkage computer‐system for use in medical follow‐up studies. Computers and Biomedical Research, 14, 327–340. Kochanek, T. T., KabacoV, R. I., & Lipsitt, L. P. (1990). Early identification of developmentally disabled and at‐risk preschool children. Exceptional Children, 56, 528–538. Last, J. M. (1995). A dictionary of epidemiology. New York: Oxford University Press. Leonard, H., & Wen, X. (2002). The epidemiology of mental retardation: Challenges and opportunities in the new millennium. Mental Retardation and Developmental Disabilities Research Reviews, 8, 117–134. Mezzich, J. E., & Ustin, T. B. (2005). Epidemiology. In B. J. Sadock & V. A. Sadock (Eds.), Comprehensive textbook of psychiatry (8th ed., Vol. 1, pp. 656–672). Philadelphia: Lippincott Williams and Wilkins. Newcombe, H. B. (1988). Handbook of record linkage: Methods of health and statistical studies, administration and business. Oxford, UK: Oxford University Press. Pober, B. R., & Dykens, E. M. (1996). Williams syndrome: An overview of medical, cognitive, and behavioral features. Child and Adolescent Psychiatric Clinics of North America, 5, 929–943.

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

23

Prasher, V. P. (1995). Overweight and obesity amongst Down’s syndrome adults. Journal of Intellectual Disabilities Research, 39, 437–441. Redden, S. C., Mulvihill, B. A., Wallander, J., & Hovinga, M. A. (2000). Applications of developmental epidemiological data linkage methodology to examine early risk for childhood disability. Developmental Review, 20, 319–349. Reiss, S. (1988). Reiss screen for maladaptive behavior. Chicago, IL: International Diagnostic Systems, Inc. Rimmer, J. H., & Yamaki, K. (2006). Obesity and intellectual disability. Mental Retardation and Developmental Disabilities Research Reviews, 12, 22–27. Roizen, N. J. (2003). The early interventionist and the medical problems of the child with Down syndrome. Infants and Young Children, 16, 88–95. Roizen, N. J., & Patterson, D. (2003). Down’s syndrome. The Lancet, 361, 1281–1289. Rutter, M., Pickles, A., Murray, R., & Eaves, L. (2001). Testing hypotheses on specific environmental causal eVects on behavior. Psychological Bulletin, 127, 291–324. SameroV, A. J. (1995). General systems theories and developmental psychopathology. In D. Cicchetti & D. Cohen (Eds.), Manual of developmental psychopathology. Vol. 2. Risk, disorder, and adaptation (pp. 659–695). New York: Wiley. SameroV, A. J., & Chandler, M. (1975). Reproductive risk and the continuum of caretaker casualty. In F. D. Horowitz, M. Hetherington, S. Scarr‐Salapatek, & G. Siegel (Eds.), Review of child development research (Vol. 4, pp. 187–244). Chicago: University of Chicago Press. Simmons, R. G., & Blyth, D. A. (1987). Moving into adolescence: The impact of pubertal change and school context. Hawthorne, NY: Aldine de Gruyter. So, S. A. (this volume). Economic perspectives on service choice and optimal policy: Understanding the effects of family heterogeneity on MR/DD outcomes. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 121–146). New York: Elsevier. So, S. A., Urbano, R. C., & Hodapp, R. M. (submitted for publication). Hospitalizations for infants and young children with Down syndrome: Evidence from inpatient person‐records from a statewide administrative database. Sparrow, S. S., Balla, D. A., & Cicchetti, D. V. (2005). Vineland adaptive behavior scales (2nd ed.). Minneapolis, MN: AGS Publishing. Sroufe, L. A., & Rutter, M. (1984). The domain of developmental psychopathology. Child Development, 55, 17–29. Stanton‐Chapman, T. L., Chapman, D. A., Bainbridge, N. L., & Scott, K. G. (2002). Identification of early risk factors for language impairment. Research in Developmental Disabilities, 23, 390–405. Susser, M., & Bresnahan, M. (2001). Origins of epidemiology. Annals of the New York Academy of Sciences, 954, 6–18. Tomblin, J. B., Smith, E., & Zhang, X. Y. (1997). Epidemiology of specific language impairment: Prenatal and perinatal risk factors. Journal of Communication Disorders, 30, 325–344. Tremblay, R. E. (2004). The development of human physical aggression: How important is early childhood? In L. A. Leavitt & D. B. Hall (Eds.), Social and moral development: Evidence on the toddler years (pp. 221–238). New Brunswick, NJ: Johnson & Johnson Pediatric Institute. Tremblay, R. E., Nagin, D. S., Seguin, J. R., Zoccolillo, M., Zelazo, P. D., Boivin, M., et al. (2004). Physical aggression during early childhood: Trajectories and predictors. Pediatrics, 114, E43–E50.

24

Robert M. Hodapp and Richard C. Urbano

Tu, S. F., & Mason, C. A. (2004). Organizing population data into complex family pedigrees: Application of second‐order data linkage to state birth defects registries. Birth Defects Research, Part A: Clinical and Molecular Teratology, 70, 603–608. Urbano, R. C. (this volume). Record linkage: A research strategy for developmental epidemiology. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 27–52). New York: Elsevier. Urbano, R. C., & Hodapp, R. M. (submitted for publication). Divorce in families of children with Down syndrome: A population‐based study. Whittington, J. E., Holland, A. J., Webb, T., Butler, J., Clarke, D., & Boer, H. (2001). Population prevalence and estimated birth incidence and mortality rate for people with Prader‐Willi syndrome in one UK health region. Journal of Medical Genetics, 38, 792–798. Yang, Q., Rasmussen, S. A., & Friedman, J. M. (2002). Mortality associated with Down’s syndrome in the USA from 1983 to 1997: A population‐based study. Lancet, 359, 1019–1025. Yeargin‐Allsopp, M., & Boyle, C. (2002). Overview: The epidemiology of neurodevelopmental disorders. [M. Yeargin‐Allsopp & C. Boyle (Eds.), Special issue on ‘‘The Epidemiology of Neurodevelopmental Disorders’’]. Mental Retardation and Developmental Disabilities Research Reviews, 8(3), 113–116. Yeargin‐Allsopp, M., Murphy, C. C., Oakley, G. P., & Sikes, R. K. (1992). A multiple‐source method for studying the prevalence of developmental disabilities in children: The Metropolitan Atlanta Developmental Disabilities Study. Pediatrics, 89, 624–629.

Section II Methodological Issues and Perspectives

This page intentionally left blank

Record Linkage: A Research Strategy for Developmental Epidemiology* RICHARD C. URBANO VANDERBILT KENNEDY CENTER, DEPARTMENT OF PEDIATRICS VANDERBILT UNIVERSITY, NASHVILLE, TENNESSEE

I.

INTRODUCTION

As certain as death and taxes, the events of our lives are recorded in public and private databases. Federal, state, county, and city government maintain databases required by statute or rules and regulation. The information recorded in these databases ranges from standard vital records of births, deaths, marriages, and divorces to school achievement, attendance, and special services, taxes, home ownership, licensures, civil and criminal court encounters. Beyond government, private industry creates detailed records about us and our interactions with them. Banks, credit card companies, credit reporting bureaus, insurance companies, utilities, supermarkets, gas stations, telephone companies, and entertainment providers (Amazon, Blockbuster, Tivo, AOL, Microsoft) all maintain massive databases. In thousands of diVerent places, and over long periods of time, data collectors are recording the raw data needed for the study of individuals. Yet, while the amount and diversity of data available from individual databases are impressive, the potential benefit from combining information is even more impressive. Data may be combined from multiple databases to increase the number of variables covered. Data may also be combined within databases to create clusters of individuals or time profiles within individuals. By linking the information on individuals between and within databases, *Author’s note: I would like to thank Theresa Urbano and Jennifer Blackford for their support and assistant with this chapter and Marisa Sellinger and Kara Newman for their assistance with the final edits. This research was supported in part by NICHD grant numbers R03 HD050468 and P30 HD15052 to Vanderbilt University. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

27

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33002-9

28

Richard C. Urbano

profiles can be created covering extended periods of time and broad sets of characteristics. For example, a within database linkage could create pregnancy histories by linking birth records based on mother identifiers. Pregnancy profiles with number of live births, spacing between pregnancies, complications of deliveries, and congenital anomalies can be created from standard birth records spanning multiple years. The time and location information recorded in linked databases makes them fit well with the person‐place‐time framework of developmental epidemiology. While there is great potential for linked databases to be used in developmental epidemiology research, there are also many challenges in creating developmental profiles from information stored in administrative databases. In psychology, for example, where many MR/DD researchers were trained, there is a strong bias against relying on data collected by others, or so‐called secondary data. Even the term ‘‘secondary data’’ is derogatory. In the best tradition of the basic sciences, in psychology the preferred research paradigm is to control the entire research enterprise from inspiration to planning, execution, and publication. In addition, it is not a simple process to convert other people’s data into a form appropriate for creating developmental profiles. Such profiles require that one draw data from multiple, diverse sources and content knowledge from diVerent areas; exercise political and diplomatic skills; and display technical expertise in informatics and statistics. Since this constellation of knowledge and skills is not often found in a single person, using linked datasets often requires a team eVort. Like any team, linkage teams have abilities beyond the individuals but also require explicit management strategies and structured communications. This model stands in contrast to the more typical configurations of principal investigator, research assistants, and graduate students. My goal in this chapter is to make a compelling argument for record linkage as a core methodology for developmental epidemiology. Simply stated, to provide the data for insights into human development, one needs to link data from multiple sources collected over extended periods of time. Without the explicit application of record linkage methodologies, our ability to describe and explain the processes influencing human development will be severely restricted. Thus far in this chapter, I have argued that a wealth of interesting data exists. We now need to change how we view these data, to create interdisciplinary research teams that can eVectively use them. Next, I will briefly cover the history of record linkage, describe recent trends that enhance the potential of record linkage, and address the characterization of ‘‘somebody else’s data’’ as ‘‘secondary data.’’ I will then explain why record linkage projects need formal project management and sketch out the elements of such a management strategy. Finally, I will provide a primer

29

RECORD LINKAGE

with the major steps required to produce individual and family profile datasets ready for analysis with standard statistical packages.

II.

PAST AND PRESENT LINKED DATASETS

Record linkage involves the processes for finding and joining matching records between and within datasets. Record linkage can be referred to by many names, for example, deduplification, unique record identification, list cleaning, biomedical record linkage, and genealogy research. While the names reflect diVerences in the purpose of the linkage and the content area using the data, the processes used are essentially the same.

A.

Historical Examples

Probably the oldest example of record linkage is the construction of family trees from such public and private vital event records as births, deaths, marriages, and divorces. The genealogies of the initial Latter Day Saints settlers and their descendents are the world’s largest database of its kind. These genealogy records have been converted into electronic form and stored in a computer database by the Utah Population Database Project (Mineau, 2005). They have been used to study such conditions as preeclampsia (Esplin et al., 2001) and prognostic factors and survival in familial melanoma (Florell et al., 2005). In scientific studies, John Snow, the father of modern epidemiology (Frerichs, 2006), is famous for his study of the 1854 cholera outbreak in London, England. Briefly, Snow identified the source of the epidemic from a simple dataset created by linking persons who died from cholera to their household locations and water source. In the mid‐twentieth century, using state‐of‐the‐art mainframe computers, various researchers conducted the first large‐scale studies linking administrative datasets (Baldwin, Acheson, & Graham, 1987; Howe & Lindsay, 1981; Newcombe, 1988). To compensate for the limited processing power and permanent data storage available at the time, they developed analytical tools and tricks of the trade; many of these are still relevant today. While the computing resources available today are several orders of magnitude more powerful than those of the 1970s, the volume and complexity of electronic data available have kept pace with the technology gains. Fortunately, as long as computing power continues to double every 18 months [otherwise referred to as Moore’s law (1965)], raw computing power will keep pace with demand.

30 B.

Richard C. Urbano Recent Developments

Several recent developments have stimulated increased interest in computer databases and record linkage. Even compared to a few decades ago, virtually every aspect of our lives is saved in electronic records. Federal, state, county, and city governments have long maintained extensive record systems. At all levels of government, there have been eVorts to improve eYciency and reduce costs. Computerization of manual systems has been one strategy for eVecting change in government information management systems. Another force driving the move to computerization is the desire of governments to improve the quality of services. Interactive web sites allow the delivery of personalized services. For example, it is now possible to use the web to order a birth certificate, renew a professional license, examine property tax assessment data, reserve a cabin at a state park, review restaurant inspections, or find the names and addresses of registered sex oVenders. To support these eYciency and quality improvement initiatives, over the last decade paper records have been migrated to computerized databases. Another recent development involves healthcare and electronic health records. One would think that the healthcare industry would have computerized databases to manage physical and behavioral health delivery. Although healthcare providers have developed information systems for managing patient information and billing, comprehensive electronic health records are the exception rather than the rule. The Health Insurance Portability and Accountability Act (HIPAA; United States Department of Health and Human Services, 2003) took a major step forward by mandating procedures for certain billing and insurance related electronic transactions. HIPAA has established rules for health information systems by requiring healthcare providers to use standard coding systems, protect personally identifiable health information, and implement security and disaster recovery procedures for computer systems. However, HIPAA does not require that comprehensive, electronic health records be created. Rather, it only defines how these records are to be handled when they do exist. A major initiative of the United States federal government, the National Health Information Infrastructure (NHII; United States Department of Health and Human Services, 2006), is designed to make electronic health records ubiquitous and sharable across systems. NHII planners have chosen to implement a national health data acquisition system by creating regional data share units, Regional Health Information Organizations (RHIOs). Two functioning regional healthcare data‐sharing projects serve as models for the RHIOs. The Regenstrief Institute in Indianapolis, Indiana has implemented a centralized shared database system. The San Diego Medical Information Network Exchange has implemented a distributed data‐sharing network. In 2005, the United States Department of

31

RECORD LINKAGE

Health and Human Services, OYce of National Coordinator (ONC) awarded $17.5 million in contracts to ‘‘advance nationwide interoperable health information technology’’ (United States Health and Human Services, 2006). In addition to the federal HNII initiative, the Centers for Disease Control and Prevention (CDC, 2006) has created the Public Health Information Network (PHIN) for managing and sharing public health data. In the private sector, health maintenance organizations (HMOs) and insurance companies are creating health profiles of their clients. For example, Blue Cross Blue Shield of Tennessee has implemented online access for healthcare providers to claims data for TennCare (Tennessee’s Medicare waiver program) enrollees. In a similar way, Healthways Inc., mines insurance claims data, and for a per client fee, provides health management services to high‐risk individuals with diabetes, hypertension, and asthma. Every place healthcare is provided will eventually contribute to our electronic health records. Beyond the government and healthcare data systems, electronic records exist for nearly everything that we do. These electronic records are a valuable resource for creating developmental profiles to study the risk and protective factors influencing the developmental process. Unfortunately, there is no single central repository for this information. In fact, for the most part the information landscape has islands of information that do not communicate or even recognize that the others exist. Electronic record linkage provides a methodology for creating individual profiles from multiple, disconnected sources. III.

BIASES AND PREJUDICES

In spite of the potential information that can be extracted from existing electronic databases, a mindset exists against using data collected by others, somebody else’s data. Others’ data are described as ‘‘secondary,’’ not ‘‘primary,’’ and thus less valued. This mindset, arguably, comes from two perceptions about valued research. First, research is hard work. Not doing the hard work is cheating. Using data collected by someone else skips an important, time‐consuming step in the research process. Thus, it is not as valued. Second, control is lost when using somebody else’s data. Someone else made all the decisions. Someone else managed the data collection endeavor. The details of who, what, when, and how individuals were sampled and observed were decided by someone else. Their decisions may have been correct for them, but they are not necessarily the ones I would have made. While both perceptions strongly influence what is valued in research and how we conduct our own research, they also limit the scope of acceptable research designs. To the extent that research questions can be framed as focused studies on small numbers of individuals, it is appropriate that the

32

Richard C. Urbano

preferred research paradigm is one in which, start‐to‐finish, the researcher is in control. Even I would agree that, in situations of focused questions and limited time periods, complete investigator control of the research process constitutes a desirable standard of practice. A.

Longitudinal Research

Although ‘‘primary data’’ may be preferable in some cases, as the focus of research expands in time, behaviors, and population covered, the complete control research paradigm becomes less feasible to implement. For example, longitudinal studies of human development are noteworthy in their diYculty and expense. Thus, few longitudinal studies have been attempted that cover the developmental period from birth to 21 years of age. Some famous exceptions from psychology and health are Terman (Elder, Pavalko, & Clipp, 1993; University of North Carolina Population Center, 2004), the Wisconsin Longitudinal Survey (Hauser & Willis, 2005; University of Wisconsin–Madison, 2006), Glidden (2000), and the Framingham health study (United States Department of Health and Human Services, 2002). In most other instances, however, adjustments to the full prospective longitudinal design provide less expensive and more practical alternatives. Partly as a response, developmental psychologists and developmental epidemiologists have created ways to perform longitudinal studies in shorter periods of time. For example, the calendar time needed to complete a developmental study can be minimized using a cross‐sectional or cross‐ sequential design (Schaie, 1965). These designs include three dimensions: study year, birth cohort, and age at observation. Table I presents a sample TABLE I A COMPARISON OF CROSS‐SECTIONAL, LONGITUDINAL, CROSS‐SEQUENTIAL DESIGNS

AND

Birth cohort (year) Study year 1996 1997 1998 1999 2000

1996

1997

1998

1999

2000

[0] [1] [2] [(3] 4

0 1 2 (3

0 1 2

0) 1

0)

A grid of birth cohort, study year, and age at observation showing cross‐sectional, cross‐ sequential, and longitudinal sampling designs for studies of birth to 3‐year olds. Longitudinal study samples are enclosed in ‘‘[ ].’’ The cross‐sectional studies are enclosed in ‘‘( ).’’ The cross‐sequential design is the combination of the 1999 and 2000 rows of the table.

RECORD LINKAGE

33

layout with five study years and five birth cohorts. The 1996 ‘‘Birth Cohort’’ column shows the sampling design for a longitudinal study covering birth to three years of age. The 1999 and 2000 ‘‘Study Year’’ rows show two cross‐ sectional studies of 1996 through 2000 ‘‘Birth Cohorts,’’ with observations also made at each year from birth to 3 years of age. The 1999 cross‐sectional study observes birth cohorts from 1996 through 1999, and the 2000 cross‐ sectional study observes birth cohorts from 1997 through 2000. While the longitudinal study will take 4 years to complete, both cross‐sectional studies compress calendar time by measuring four age groups in the same calendar year. The data from diVerent age groups are combined to describe development. Although such cross‐sectional designs are helpful, they too have their limitations. Specifically, historical changes could be confounded with age diVerences. Events aVecting age groups diVerentially could appear to be changes in the course of development. An alternative treatment of the data uses the 1990 and 2000 cross‐sectional datasets to yield a cross‐sequential design. Historical eVects and birth cohort factors can be assessed, while the time to complete is reduced to 2 years. Complex statistical analyses are required to reconstruct the longitudinal data (Masche & van Dulmen, 2004). While cross‐sectional and cross‐sequential designs improve eYciency of prospective, investigator‐controlled developmental studies, they more importantly oVer a conceptual bridge to a broader spectrum of research strategies. Cross‐sectional designs gain eYciency by collecting multiple age samples at the same time. In the same vein, time to collect data can be reduced by using somebody else’s data. Using data already collected from large populations over long periods of time is eYcient in two ways. First, the time to collect data is compressed. The data have already been collected. Second, data collection costs are somebody else’s. The data are already collected. Our secondary data were somebody else’s primary data. B.

How Do Others Use SED?

Additional support for the usefulness of somebody else’s data comes from other disciplines. Public health, demography, and economics all routinely use population data collected by others. Public health surveillance, the monitoring of important indicators of health and wellness, exemplifies how somebody else’s data constitutes a part of an overall research strategy. For example, information on births is collected by birthing centers and reported to public health. The birthing centers create birth vital records by reviewing medical records, surveying patients, and interviewing attending health professionals. Much of the data used for public health surveillance comes from ongoing data collection activities. For example, designated reportable conditions are observed, recorded, and reported by healthcare providers; these include infectious diseases (measles, mumps, malaria, polio, HIV/AIDS,

34

Richard C. Urbano

tuberculosis, sexually transmitted diseases) and, since the events of 9/11, indicators of potential bioterrorism (CDC, 2005). Cancer and immunization registries also use data and reports from healthcare providers. Birth defects, Traumatic Brain Injury, and Crash Outcome Data System (traYc accident monitoring) rely heavily on hospital discharge data. With these surveillance data, along with information obtained from direct sampling, vital statistics data systems, and an extensive suite of surveys, the CDC assesses population health status and the eVectiveness of health promotion activities. Further, public health strategies provide a powerful paradigm for explaining anomalous events detected by surveillance eVorts. Once an event has been determined to be atypical, ‘‘shoe leather epidemiology’’ is used to study the causes, correlates, and predictors of the event. In the case of an outbreak of salmonella, for instance, a standard protocol would be implemented (CDC, 2001). Investigators will identify potential places of infection, aVected and nonaVected individuals are interviewed, and biological samples obtained. In this case, then, a combination of ‘‘secondary’’ and ‘‘primary’’ data has been examined. Somebody else’s data were used to detect an event in a population, which was then followed by primary data collection to hone in on the causes and processes associated with that event. In conclusion, secondary data have an important place in a comprehensive research strategy. Administrative datasets are valuable resources, as they cover long periods of time and large populations of individuals. As they compress calendar time, such databases are particularly important for life‐span developmental research. Further, population‐based datasets allow the study of rare conditions. For example, in a condition that occurs 1 per 1000, we would reasonably expect to observe 1000 aVected individuals in a dataset with 1,000,000 people. As would any other data sources, datasets recorded by others should be considered valuable sources of information.

IV.

EMBARKING ON A LINKED‐DATA RESEARCH PROJECT

Compared to ‘‘typical’’ social science research projects, a research project using linked datasets diVers both in the composition of the research team and in the complexity of the data processing tasks required. Specialists are needed in both content area knowledge and technical computing skills. A.

Primary Data Research

In order to highlight the diVerences in the two types of projects, let us begin by examining a typical, primary data gathering research project.

RECORD LINKAGE

35

In most primary data gathering projects, the structure is hierarchical. The major professor heads up the team, which consists of a research director, research assistants, and graduate students (sometimes undergrads as well). This major professor is generally the principal investigator (PI) on the project and, even if unfunded, the acknowledged leader of the research team. Although the research director and research assistants may be ‘‘expert’’ on subjects or testing, and doctoral or postdoctoral students may be most advanced in terms of specific research questions or analyses, the major professor is clearly advanced overall. In content of the research, questions asked, approach taken, and even in the analyses and presentation of results, a single person is the acknowledged leader. B.

Linked Data Research

Contrast this hierarchical, single‐person approach to a research project using linked datasets. Although one person might conceivably have all the knowledge, skills, and time needed to implement a record linkage project, such a situation seems unlikely. More often, a multiperson, multidisciplinary team will be needed. ‘‘Content people’’ will focus most on which specific research questions can and should be asked, whereas technical specialists will devote their time and energies to all of the many tasks involved in linking the various datasets. In all ways, then, the organizational structure is less hierarchical and the knowledge to be used in the project much more dispersed across diVerent individuals. 1. TEAMWORK

As might be imagined, such a division of labor presents both advantages and disadvantages. The primary advantage of a team with members from several disciplines is synergy. By working together, the team can accomplish more than the sum of the members working independently. However, for synergy to occur the team members need to interact and share ideas. The basis of productive interaction is active communication among team members. While the research project objectives provide a common point of reference, each discipline brings a diVerent perspective to the project. Team members need to have at least a basic‐level understanding of the diVerent roles and contributions of team members. Further, team members need to appreciate and compensate for their discipline‐specific vocabularies. It is often diYcult to see the level of miscommunication that can occur with what are discipline‐specific terms. For example, to an epidemiologist surveillance describes a suite of tools for monitoring events to detect anomalies. However, to others surveillance is more akin to spying.

36

Richard C. Urbano

2. TASK COMPLEXITY

While communication challenges are a barrier to team functioning, the complexities of a record linkage present another barrier. Without an appreciation for the multiple interrelated tasks that need to be successfully completed to accomplish the record linkage portion of the project, it is easy to become frustrated with the process. A basic understanding by team members of the record linkage workflow helps minimize frustration. Further, active communication from the record linkage subteam (possibly one person) will also help minimize frustration. In addition to the management challenges created by multidisciplinary teams, the number of interrelated tasks in record linkage adds complexity. Record linkage processes do not proceed in one direction from one step to the next. The results of one step in the process may require repeating prior steps. An error detected during data cleaning could require changing the meta data and reloading the original data. For example, if the date of last menstrual period in a birth record was coded in year–month order instead of month–year as stated in the documentation, dates created from the original data would be incorrect. To correct the error, changes would need to be made to the meta data and load programs, and the process would begin again. Also, to‐be‐linked datasets are not all available or complete at the same time. New datasets and additions to existing datasets can require repeating prior linkages steps. For example, hospital discharge data are released 6–12 months beyond the end of the reporting year. If we are following the hospitalization of a birth cohort, each new hospitalization dataset would need to be linked to the birth cohort. Further, to‐be‐linked datasets with large numbers of records and many variables may require hours or days of processing, even on the fastest computers. Multiple versions of the original, cleaned, linked, and analysis datasets are created over the length of the project. Thus, management, backup, and recovery of computer files are nontrivial activities. C.

Project Management

Record linkage is a complex process that only succeeds when multiple interdependent tasks function correctly. Even small errors can cost days or weeks of eVort to correct. Linkage projects do not run themselves. One needs to pay close attention to details, monitor outcomes, and frequently assess progress. A written project management plan provides the basis for measuring outputs, assessing progress, and making adjustments. EVective management plans vary widely in level of detail and degree of automation. Simple paper and pencil systems anchor one end of the spectrum of management strategies, while complex, highly automated strategies anchor the other end. The size of the project, number of team members, number of datasets to be

RECORD LINKAGE

37

linked, budget, granting agency requirements, and technical expertise are all considerations in choosing a project management strategy. 1. SMALL PROJECTS

A project timeline and log may provide suYcient documentation and management tools for a project team with two or three members working on a small number of datasets. The project timeline, known as a Gantt Chart, provides a visual, time‐based presentation of project events. For instance, project activities, planned reviews, milestones with deliverables, and the anticipated completion date can be visualized in several ways: drawn on a sheet of graph paper with the time axis along the bottom, written on a printed calendar, entered into a shared MS Outlook calendar or one of the free web‐based calendar programs. Each representation of the timeline provides a visual prompt for project expectations and monitoring progress. Used in conjunction with the timeline, the project log provides a permanent chronological record of project activities. Even with a small project completed over a short period of time, it is unlikely that every detail will be remembered. A written log provides a permanent record to dig through to find the missing detail. Like any good audit trail, once an activity is entered into the log it should not be removed. Corrections are entries with a reference to the entry they fix. While the team will decide on the specific contents and format of the log, there should be enough detail recorded to reconstruct the sequence of important events, decisions, and outputs. The log should be stored in a single, safe location, accessible by all team members. Paper and ink logs need to be kept in a safe, accessible place, like a desk drawer, file cabinet, or bookcase. Electronic logs should be saved on a shared, always‐on‐computer that is accessible by team members. 2. LARGER PROJECTS

With larger teams, more tasks and possibly more complex relationships among tasks, larger projects need a more explicit management strategy. With a project that may take months or years to complete, a simple project timeline and log are inadequate. Three elements of the project need to be analyzed and monitored. A detailed analyses of tasks and the relationships among tasks provide the functional framework for the project. Estimates of the minimum and maximum time to complete each task determine the duration of the project. Project cost estimates are calculated from manpower and resource estimates. A visual mapping of project elements, known as a PERT chart, can support project management. A PERT chart is a compact graphical display with time on the horizontal dimension, circles with task

38

Richard C. Urbano

names, estimated manpower, dollars and time to complete drawn in circles over the time dimension, and lines between the circles showing the relationships between tasks. V.

WORKFLOW OF LINKAGE

With our management strategy chosen and tools selected and installed, we are ready to begin record linkage. There are six major steps in process, including (1) acquiring the data, (2) storing the data, (3) selecting linking variables, (4) examining record pairs, (5) matching, and (6) generating analysis datasets. A.

Acquire Data

1. FIND DATA SOURCES

Your research questions will guide where to look. For several reasons, government agencies are the first place to look for data related to questions of human development. Research on human development is consistent with the mission of many government sources. Research findings may inform policy decisions at little or no incremental cost to the agency. Government agencies are likely to have data with person‐specific identifiers. Researchers from legitimate public and private organizations may obtain datasets at little or no costs. In contrast, private‐sector data may not contain person identifiers, can be expensive, and may require extended negotiation to attain. 2. ASK FOR THE DATA

The procedures for obtaining government datasets vary. While many datasets are available via a simple request process, others require a formal research proposal. In addition, researchers must first receive Institutional Review Board approval from their institution, as well as the Institutional Review Board approval from the agency managing the data. State or federal statutes or rules and regulations further restrict access to some datasets. Release is likely to be restricted where datasets contain sensitive, confidential information. Hospital discharge, HIV/AIDS, and educational records are likely to be restricted. For these restricted datasets, a formal contractual agreement may be required to establish the research institution as an agent of the data management agency. The Family Education Rights and Privacy Act (FERPA) (United States Department of Education, 2002) further restricts release of school‐generated data without individual consent. Release of health and healthcare records are also controlled by the Health Insurance Portability and Privacy Act (United States Department of Health and Human Services, 2003).

RECORD LINKAGE

39

In thinking about datasets, one ideally wants data with individual identifiers for each record. Government electronic datasets mandated by statute or rules and regulations are most likely to have one or more highly unique individual identifiers (names, social security number, insurance numbers, tax identification numbers, unique internal identifiers, etc.). Other datasets may be more diYcult to get, and may not have or may not be released with personal individual identifiers. While any person may authorize the release of their information for a specific research project, it is not practical to gain individual consent from large numbers of people. This approach defeats the major advantage of using population‐wide data. When the primary purpose of a research initiative is to examine low‐prevalence conditions, it is even more problematic to obtain consent from large numbers of individuals who could not participate in the research. B.

Store the Data

1. PHYSICALLY READ DATA

Although reading the data should be relatively straightforward, it is often daunting because of physical damage to the media or arcane data representations. With a current vintage PC, reading datasets provided on CD or DVD is usually trouble free. Newer computers are capable of physically reading a wide range of standard CD/DVD recording formats. Most programming languages, data management tools, and statistical packages are capable of handling fixed format or delimited text files. Although reading the data should be straightforward, it often proves daunting. First, verify that the CD, DVD, or tape is actually readable. To verify that the data are readable, produce a directory listing and count the records in each file. A simple directory listing will verify that the directory structure is intact. Counting the records in each file will determine if the records in the dataset are readable. The directory listing and record count need to agree with the manifest of files and record counts provided with the dataset. A checksum, a single number calculated from all the contents of a dataset, may be provided with the dataset. The integrity of the dataset can be verified by comparing the checksum value calculated from the data received with the value provided with the dataset. If the two numbers agree, the dataset has not changed from when it was created. With the datasets successfully read, they need to be organized into variables and records. 2. CREATE SUPPORT FILES

The datasets obtained for linking are raw data without intrinsic meaning. At one level, they are merely long sequences of characters. In order to transform raw data into usable information, one needs to create a detailed

40

Richard C. Urbano

description of the contents and characteristics of the information in each dataset. This description is referred to as meta data, or data describing data. Meta data are stored in tables where the rows represent variables in the dataset and the columns represent the characteristics of the datasets. At a minimum, the meta data tables need to have columns to describe the general characteristics of the dataset. For a Tennessee births file for birth year 2005, the following columns are likely: dataset source name (‘‘TN Birth Vital Records’’), file name (‘‘bir05.txt’’), dataset creation date (‘‘2006‐01‐15’’), time period covered in dataset (‘‘2005‐01‐01:2005‐12‐31’’), geographical coverage area (‘‘Tennessee’’), human readable description of the dataset (‘‘2005 Calendar Year Tennessee Birth final data’’). Further, for each variable (field) in the dataset there will be a row in a meta data table columns for the characteristics of the field. For example, the birth certificate number which includes the year of birth and a sequential number would have the following characteristics: record type (‘‘birth’’), variable name (‘‘CertNumber’’), data type (‘‘text’’), input format string (‘‘#########’’), output format string (‘‘TN####‐#####’’), field starting position (22), field width (9), missing values (blank, ‘‘999999999’’). In addition to the basic description of the dataset, the meta data table should have information about editing rules and formatting data for export to other systems. The meta table could have the following additional attributes: output short name—Cname output long name—‘‘Child Name’’ variable labels (human readable and short)—Cname or ‘‘Child Name.’’

Comprehensive meta data tables have all the information needed to specify how to read, format, calculate, validate, and export data. Meta data driven programs can automatically generate import files for database management programs, statistical packages, interactive forms for data editing, printing reports, creating PDF files, or publishing web pages. The next record linkage step creates the instructions for loading datasets into a database management system. Data are stored in database tables with columns for variables and rows for entities. Database tables are defined using a data definition language (DDL). Most database systems use a standard query language (SQL)‐based DDL. However, each database management package adds slightly diVerent features to the standard SQL DDL. Table definitions for a specific database management system can be generated programmatically. Appropriate DDL instructions can be constructed with a little program, often called a script, using meta data and designation of the target database. Scripts can create import instruction, variable and

RECORD LINKAGE

41

value label specifications, and definitions of missing values for statistical packages. Creating DDL scripts from meta data has two advantages. First, the process of creating the tables for the data repository is completely defined by the meta data, the target database system, and generating program. Second, the process is reproducible. Running the script with the same meta data and target system will produce the same table definitions. Finally, changes to meta data can be used to automatically generate modified DDL definitions. 3. EFFICIENT STORAGE OF DATA

A useful database management system is characterized by eYcient storage and retrieval, flexible security options, and comprehensive data backup and recovery facilities. A wide range of packages with these characteristics are available. 4. CREATE CODE TABLES FOR EACH DATA SOURCE

How data are represented will vary from source to source and possibly from file to file from the same source. For example, gender may be represented as (male, female), (m, f), (boy, girl, in utero, ambiguous, unknown), (0, 1, or 1, 2). The coding schemes used for each data source will be described in the meta data for the source. The actual codes should be stored in a separate database tables with columns for data source name, source sequence indicator (this could be date or version number for periodically released data), variable name, variable class (gender, race, state, hospital ID), source code value, value description, master numeric code value, master short text description, master full descriptions. 5. TRANSFER DATA FROM SOURCE DATASETS

Datasets obtained from other organizations come in a form and format chosen by them. In order to use these data for linkage, they need to be converted from their original form into a standardized format that can be loaded into the project database management system, referred to as a data repository. Two common file formats, fixed width and comma separated value, have little or no internal meta data and thus require the most eVort to transfer from the original datasets into the project database repository. Fixed width text files have records that appear as long strings of characters with one or more characters between records. Field definitions are not part of the file; thus field definitions must be manually entered into the meta data tables. Text files with single or double quotation marks around character fields, and delimiter characters [tabs, commas, or pipes (‘j’)] between fields are called

42

Richard C. Urbano

delimited files. The internal structure of delimited text files eliminates the tedious, error‐prone process of separating long strings of characters into fields. To have a consistent processing for dealing with text files, fixed width text file records should be parsed into fields and saved as delimited records. Unlike text file datasets, many datasets come with internal documentation of their data. Thus, database or statistical package datasets require less eVort since most of the needed meta data already exist. 6. CHECK AND CLEAN THE DATA

For each source file, run descriptive statistics on each variable. Review the descriptive statistics for anomalies. The observed values in the file should not have unexpected or atypical values, code values should map to the master code table, and they should conform to the data types from the meta data table, that is, date, integer, and decimal have the appropriate format. Close attention should be paid to potential missing values. From this inspection pass on the data, procedures for converting the source data fields will be created. Every dataset has some anomalies with and between variables. In the first cleaning pass, individual variables are checked for anomalous values. In this pass, the relationships between variables can be checked, variables converted into more convenient forms, and editing rules applied. For example, a mother with an age of 5 in the birth file or a marriage date greater than the divorce date in the divorce file would be suspect, as would an address listed as Florida but with a Tennessee zip code. Particular attention should be paid to likely linkage variables such as dates, names, addresses, telephone numbers, and certificate numbers. Repeat the load and clean processing until no anomalies are detected. Then the data stored in the database repository are ready. 7. SECURE THE DATA

Since creating backups takes time, adds cost to the project, and may interfere with ongoing project activities, there needs to be a balance between the total cost of backups, the cost of recovery, and the potential risks. While automated, periodic backups of all project files allow for full recovery, it may not be practical or needed. A backup that takes longer than a few hours at night will aVect day‐to‐day activities. In contrast to full backups, more eYcient incremental backups save only those files which have changed since the last backup. Further eYciency can be gained by excluding from the periodic backup those files which do not change or can be easily reconstructed. For example, the original to‐be‐linked datasets do not change, thus it is adequate to have a single backup copy stored in a safe place.

RECORD LINKAGE

43

The second line of defense consists of cyber and physical security measures. Cyber security involves limiting access to electronic data using firewall programs that control network traYc and scrambling electronic data using encryption procedures. With the source datasets safely stored in our favorite database management system, copied to permanent, removable media (tape, DVD, CD), and securely stored, we are ready to link the data. C.

Select Linking Variables

A single variable or a set of multiple variables can be used to link datasets. For simplicity, we will refer to both single and multivariable sets as the linkage variables. The best linkage variables have many unique values, are recorded with few errors in a consistent format, have few missing values, and are uniquely associated with individuals or entities to be linked. Further, between‐dataset linkage requires that the linkage variables are recorded in all datasets to be linked. Few variables by themselves are uniformly good linkage variable candidates. Government‐assigned identification numbers are among the most commonly available, unique identifiers: social security, birth certificate, driver’s license, passport number, tax identification, and military identification numbers. However, none of these variables are universally available within or between datasets. ‘‘Social security number,’’ which was originally intended to be used only for Social Security Administration purposes, now appears in many datasets. Most children born in the United States have social security numbers assigned at birth, thus their information can be used on their parent’s United States tax filings. However, not all children living in the United States were born here. Some parents opt out of the early social security number assignment. Since it takes several months to receive the assigned social security numbers, childhood records created during the first year of life will likely lack social security numbers. Birth vital statistics records have social security numbers added before the final birth year file is created. Thus, social security number is the first candidate as a linkage variable. Biometric measures are highly unique to individuals, and other demographic variables are generally available, both are most often rejected as linking variables. Although biometric measures like finger prints are unique to individuals, they are not typically part of available datasets. Commonly available demographic variables like gender and race/ethnicity have few unique values, and many nonmatching records will have the same value. Some variables make poor linkage variables in their raw form but can be transformed into useful variables by themselves or when combined with

44

Richard C. Urbano

other variables. Names (self, parents, grandparents, maiden) and addresses available in many datasets are prime examples of variables that may be transformed from poor to very good linkage variables. Addresses collected close in time make very specific linking variables. However, there is little consistency in the recording of addresses. Thus, addresses need to be transformed into a standard format before they can be used as linking variables. Name recording is almost as variable as addresses. Recording full names in a single field with or without punctuation presents the most diYcult problem. For example, a name field might have ‘‘Dr. Doctor, John Jay Jr.’’ Parsing this string into first, last, middle, title, suYx is diYcult, since the name components could appear in any order. Even when names appear in a standard format, various misspellings, transposition of first and last name, cultural diVerences in name order, and dropping of suYxes aVect the recorded name. Oriental names and hyphenated surnames are particularly problematic. Phonetic coding of names, hidden Markov structure analysis, and string‐based comparison rules reduce transform names into more usable forms. D.

Examine Record Pairs

The result of linkage variable selection is a set of from 1 to 10 or so variables that exists in the pairs of datasets to be linked. An example set of linkage variables taken from a birth and a divorce dataset is presented in Table II. The number of potential comparisons is large even with modest numbers of records. For example, to compare a birth cohort with 80,000 (8 104) records and inpatient hospitalization admissions with 900,000 (9 105) records would have 7.2 1010 pairs of records. If we can compare 72,000 (7.2 104) pairs of records per second, it would take 106 seconds or 11.6 days to compare all record pairs. An order of magnitude improvement in the time taken to make each comparison would reduce the total time to compare all records pairs to 1.2 days. EYciency can be improved by reducing the number of pairs of records that are actually compared or increasing the rate at which records are compared. We will look at three strategies: blocking, comparison eYciency, and parallel processing. 1. BLOCKING

In most problems, the number of true matches between the first, master dataset and the second, detail dataset is a small proportion of the total records in the detail dataset. Thus, making all of the possible pair‐wise comparisons is very ineYcient.

45

RECORD LINKAGE

BIRTH

AND

TABLE II DIVORCE DATASET VARIABLES

Birth dataset

Divorce dataset

Primary linkage variables

Primary linkage variables

DadDOB DadNameGiven DadNameLast DadRace DadSSN MomDOB MomNameGiven MomNameLast MomSSN MomRace Soundex

DOBHusband NameFirstHusband NameLastHusband RaceHusband SSNHusband DOBWife NameFirstWife – SSNWife RaceWife Soundex

Secondary linkage variables

Secondary linkage variables

ChildDOB ChildNameLast CongenitalAnomaliesInd DadAge DadEducation LiveBirthsTotal MomAge MomEducation MomMarried MomResCountyCode

ResidenceCountyHusband MarriageDate Children ChildrenUnder18

Nonlinkage variables

Nonlinkage variables

AbnormalConditionsInd AlcoholUse APGAR1Minute APGAR5Minute BirthOrder BirthPlaceCityCode BirthPlaceCode BirthPlaceCountyCode BirthPlaceType BirthPlurality BirthWt BirthWtUnit ChemicalSubstanceAbuse ChildNameFirst

ID (Divorce file) StateFileNumber ResidenceStateHusband ResidenceStateWife PlaceMarriageCounty PlaceMarriageState SeparatedDate DecreeDate DecreeYear DecreeType DecreeCounty MarriageNumberHusband MaritalStatusPreviousHusband MarriageNumberWife (continued)

46

Richard C. Urbano TABLE II (Continued)

Nonlinkage variables

Nonlinkage variables

ChildNameGenID ChildNameGiven ChildNameMiddle ComplicationsInd DadHispanic DadNameGenID DataYr DeathInd DeliveryMethodsInd GestationWeeksEstimated ID (unique record id) LastDeliveryInterval LastDeliveryType LastLiveBirthInterval LastLiveBirthMoYr LastNormalMensesDate LastTerminationDate LastTerminationInterval LiveBirthsDead LiveBirthsTotal LiveBirthsLiving MedicalRiskFactorsInd MomBirthStateCode MomHispanic MomMailingAddress MomMailingCity MomMailingState MomMailingZip MomNameMaiden MomResAddress MomResCity MomResCityCode MomResCityLimits MomResStateCode MomTransferred MomTransferredFromFacCode MomWtGain ObstetricProceduresInd OOSBirthCertNum PregnanciesTotal PrenatalCareBeganMo PrenatalCareBeganPregnancyMo PrenatalVisits Terminations TobaccoNum TobaccoUse

MaritalStatusPreviousWife MarriageEndedDeathHusband MarriageEndedDivAnnulHusband MarriageEndedDivAnnulWife MarriageEndedDeathWife MarriageDuration SeparationInterval AgeHusband AgeWife CustodyNumberChildren CustodyHusband CustodyWife CustodyJoint CustodyOther CourtType

Linkage variables from birth and divorce datasets.

RECORD LINKAGE

47

Grouping records on common attributes, blocking, is designed to bring together a subset of individuals who are more similar than individuals in the full dataset. For example, names could be a good blocking variable. Since names may have minor misspellings or transposed letters, name blocking uses phonetic representation of names. A widely used phonetic encoding procedure, Russell Soundex, uses a simple mapping scheme: Keep the first letter of the name, drop all vowels, map the remaining nonvowel letters into the numbers 0–9 with letter likely to be confused in English assigned the same number. The soundex for Urbano is U615. Names are not the only linking variables that could be used for blocking. Any variable or set of variables that is likely to bring together similar individuals can serve as a blocking factor. For example, birth year alone or birth year‐gender‐race are eVective blocking factors. 2. INDEXING

In addition to using blocking, eYciency can be improved by indexing. An index based on social security numbers is much like a card catalog. Each catalog card is labeled with a single social security and lists all the locations where information relating to the social security number is stored. Because of the way indexes are organized, only a small number of index cards need to be accessed to find a specific social security number and its related data. In the birth file, there is only one entry per social security number. In the hospitalization file, there would be zero or more records per person. If we estimate that 10% of children are hospitalized on average once in any year, in Tennessee we would expect to have make 640,000,000 comparisons [80,000 (births per year) 8000 (estimated hospitalizations)]. With indexing, less than 1.5 million comparisons would be needed. 3. EFFICIENT COMPARISONS

Another strategy for improving eYciency is to reduce the time to compare individual variables. Computers are more eYcient comparing integer numbers than strings of characters, shorter strings compared to longer strings, and fixed length strings are more eYcient than variable length strings. For example, phonetic coding of names converts variable length names into fixed length strings of four to six characters. To change comparison of phonetic codes from string to integer, the 2000 or so unique phonetic codes could be represented by the integers from 1 to 2000. Anytime character variables are being compared for complete agreement, the unique string values can be mapped into a set of integers. For example, a list of unique surnames from to‐be‐linked files could be created and integer values assigned.

48

Richard C. Urbano

4. PARALLEL PROCESSING

Parallel processing is a simple idea. Divide the to‐be‐linked files into pieces. Simultaneously link all the pieces. Then put the individual linked pieces back together. Solve the problem for each part and put the piece back together for the final solution. To accomplish these block comparisons, an executive program breaks the original files into blocks, hands out the pairs of blocks to separate processors, monitors the completion of the block comparisons, and then puts the results from the separate comparison back together. E.

Match Records

Once records have been arranged for pair‐wise comparisons, one needs to make the comparisons and determine which pairs match. At the simplest level, linkage variables in a pair of records agree or not. With a set of N linkage variables, a pair‐wise comparison produces a set of N agrees and do not agrees. Based on the set of agree/do not agree results, record pairs are classified as matching, not matching, or too‐close to call. Two general strategies are used to classify records pairs, probabilistic, or deterministic. 1. PROBABILISTIC

A mathematically sophisticated way of dealing with the results from multiple comparisons, probabilistic linkage, assigns positive weights for agreements and negative weights for disagreements (Felligi and Sunter, 1969; Jaro, 1989; Newcombe, 1967). A single number is calculated by weighted sum of the comparison results. To calculate the weight for each linking variable two probabilities are estimated: (1) the probability that two linking variables have the same value in true matched pairs (m) and (2) the probability that two linking variable have the same value in true unmatched pairs (u). From the formula for the weight, w ¼ log(m/u), it can be seen that u controls the importance of a variable. For example, a variable like gender which should have a very high m value gets a small weight because if has a u value close to .5. 2. DETERMINISTIC

The main alternative strategy to probabilistic linkage is called deterministic linkage. Rather than assigning a number to pairs of records representing the degree to which the linking variables in the pair agree, a hierarchical set of decision rules is used. Deterministic and probabilistic linkage have the same objectives: (1) identify true matches as matching, (2) true nonmatches as not matching, and (3) minimize the number of ‘‘too close to call’’ pairs. Deterministic linkage uses decision rules that are applied in order, to pairs of

RECORD LINKAGE

49

records with matching pairs marked as matches with the rule number. Once a pair of records is matched, both records are marked as linked and removed from the pool of potential matches. 3. WHICH STRATEGY TO USE?

Neither strategy is a uniformly better choice. Probabilistic linkage has a strong mathematical basis and produces a single weight to judge the agreement between a pair of records. However, the weights used in probabilistic linkage are derived from two estimated quantities, M and U. U is estimated from the observed frequencies of variables and specific variable values in the to‐be‐linked datasets. M is take from prior research or trial linkages of the datasets. In contrast, deterministic linkage relies on exact comparisons and a hierarchical set of rules. While the selection and ordering of the rules are based on the characteristics of the data, the selection process is ad hoc. EVective deterministic rule sets use variables or variable sets with high specificity in the initially evaluated rules. Thus, it is very likely that variables that would have had large probabilistic weights are given priority. For example, Gomatam, Carter, Ariet, and Mitchell (2002) found deterministic linkage to be a better choice for the linkage of dataset with reliable, unique individual identifiers. Boussy and Scott (1993) found that combinations of three or four identifiers in a deterministic linkage produced very high estimates of sensitivity and specificity. Next, the ‘‘quality’’ of the linked data needs to be assessed. Quality in this context means the degree of agreement between observed matches and an accepted standard. For this assessment, a 2 2 table is created, Standard (Match/No Match) by Observed (Match/No Match). Since there is no agreed upon gold standard to which observed matches can be judged, a proxy procedure needs to be used. One procedure draws a random sample of records from one of the to‐be‐linked datasets. Then a manual linking of the datasets is done. The results of the manual linking serve as the standard, and this standard is compared to the observed results. Another procedure simulates a population dataset with known characteristics and applies the planned linkage strategies to the population. Since the true match/no match status is known from the simulated data, the simulated data serve as the standard and are compared to the observed matching results (see chapter by Tu et al. in this issue for a discussion on this method). F.

Generate Analysis Datasets

With datasets linked, analysis of the data can start, almost. Data stored in the project database repository need to be converted into a form usable by one’s statistical package, referred to as an analysis dataset. This dataset needs

50

Richard C. Urbano

to have the correct subset of variables and records from the repository needed for statistical analysis. The variables need to be in the appropriate format, that is, number, date, text, with informative variable label and value labels, and missing values assigned. In short, much of the tedious dataset preparation work done by statistical analysis prior to actual analysis should be done using existing meta data. Ideally, the analysis dataset should be ready for analysis. There are two general approaches to creating the analysis dataset. The programs to create the analysis dataset may reside with the database procedures or with the statistical software. If a database program is used, the database analyst writes a program to generate two files in a format that can be used by the statistical program. One of the files will have the data. The second file will have the instructions for loading the data into the statistical package. Since the meta data are available to the database program, the instruction will be able to describe variable labels, variable types (number, text, date), value labels, and missing values. By having the program run as part of the database, security is maintained. The database program can only access and output the data with the permissions assigned by the database analyst. Any sensitive variables or combinations of variables can be hidden or encrypted before being written to the statistical data file. With only one exit point from the database, and one set of programs to create the analysis dataset, it is easier to identify and correct problems. In contrast, the statistical software program approach transfers the data directly from the database without using intermediate files. For example, to get data from a database using SPSS, one would select File!Open Database!New Query. At this point, one can select an existing database connections or create a new one. Before the data are brought into SPSS, one can point and click to selectively import variables, limit retrieved cases, and change text variables to numeric. While this second approach is more eYcient in terms of disk storage requirements, it has several disadvantages. First, since the program is written in the language of a particular statistical package, each statistical package will need its own program. Second, while the statistical analyst can write the procedures to ‘‘connect’’ to the database repository and extract data, the database analyst still has to configure the database to limit access, protect privacy, and ensure security. Thus, most of the database programming that would go into a database driven export program (the first option) must also be created for a statistical data access program. Finally, statistical database access programs provide limited control of the import procedure. Most significantly, meta data which are easily accessible to a database program are usually not available to the statistical software. Thus, data may be loaded in the wrong data type, without variable or value labels, and without missing values assigned.

51

RECORD LINKAGE

Although both approaches to creating an analysis dataset have the same end result, they take diVerent paths and use diVerent resources. For the second, statistical program approach to work, the statistical analyst needs a basic understanding of database access procedures and programming skills. In either case, the database analyst needs to configure the database for exporting data. Thus, depending on the personnel available and their skills, either approach will work. VI.

CONCLUSIONS

Although calendar time constitutes the central dimension upon which human development progresses, it is a limiting factor for developmental researchers. Even if we constrain our interest to short portions of the life span, there are few opportunities to observe events in real time. Creative cross‐sectional and cross‐sequential research designs have been invented to compress time by combining short‐term longitudinal and cross‐sectional data. In this chapter, I have argued that the combined use of somebody else’s data and record linkage can also eVectively compress time and eYciently study full life‐span development. The electronic recording and linkage of life events allows us to identify important risk and protective factors that influence human developments, within the academic lifetime of a developmental researcher. With the aVordable technology available at the desktop, developmental epidemiologists can evaluate hundreds of variables, over millions of people, across long periods of time. In the best tradition of epidemiology, we can then study in detail the processes by which risk factors aVect children’s development. Valuing ‘‘SED’’ and using record linkage technology will lead the way to a better understanding of human development and help us to understand, treat, and develop eVective policies for children with MR/DD and their families. REFERENCES Baldwin, J. A., Acheson, E. D., & Graham, W. J. (1987). Textbook of medical record linkage. New York: Oxford University Press. Boussy, C. A., & Scott, K. G. (1993). Use of data base linkage methodology in epidemiological studies of mental retardation. International Review of Research in Mental Retardation, 19, 135–161. Centers for Disease Control and Prevention (2001). Updated guidelines for evaluating public health surveillance systems. MMWR, 50(30), 646. Centers for Disease Control and Prevention (2005). Summary of notifiable diseases—United States, 2003. MMWR, 52(54), 1–88.

52

Richard C. Urbano

Centers for Disease Control and Prevention (2006). Public health information network (PHIN). 2006, from http://www.cdc.gov/phin/. Elder, G. H., Jr., Pavalko, E. K., & Clipp, E. C. (1993). Working with archival data: Studying lives. Newbury Park, CA: Sage Publications. Esplin, M. S., Fausett, M. B., Fraser, A., Kerber, R., Mineau, G., Carrillo, J., et al. (2001). Paternal and maternal components of the predisposition to preeclampsia. New England Journal of Medicine, 344, 867–872. Felligi, I. P., & Sunter, A. B. (1969). A theory for record linkage. Journal of the American Statistical Society, 64, 1183–1210. Florell, S. R., Boucher, K. M., Garibotti, G., Astle, J., Kerber, R., Mineau, G., et al. (2005). Population‐based analysis of prognostic factors and survival in familial melanoma. Journal of Clinical Oncology, 23(28), 7168–7177. Frerichs, R. R. (2006). John Snow. 2006, from http://www.ph.ucla.edu/epi/snow.html. Glidden, L. M. (2000). Adopting children with developmental disabilities: A long term perspective. Family Relations, 49, 397–405. Gomatam, S., Carter, R., Ariet, M., & Mitchell, G. (2002). An empirical comparison of record linkage procedures. Statistics in Medicine, 21, 1485–1496. Hauser, R. M., & Willis, R. J. (2005). Survey design and methodology in the health and retirement study and the Wisconsin longitudinal study. In Population and developmental review (pp. 209–235). New York: The Population Council, Inc. Howe, G. R., & Lindsay, J. (1981). A generalized iterative record linkage computer system for use in medical follow‐up studies. Computers and Biomedical Research, 14, 327–340. Jaro, M. A. (1989). Advances in record‐linkage methodology as applied to matching the 1985 census of Tampa, Florida. Journal of the American Statistical Association, 89, 414–420. Masche, J. G., & van Dulmen, M. H. (2004). Advances in disentangling age, cohort, and time eVects: No quadrature of the circle but help. Developmental Review, 24, 322–342. Mineau, G. (2005). Utah population database project. 2006, from http://www.huntsmancancer. org/group/sharedFacilities/ccsg/UPDB.jsp. Moore, G. E. (1965). Cramming more components onto integrated circuits. Electronics, 38(8), from http://download.intel.com/research/silicon/moorespaper.pdf/. Newcombe, H. B. (1967). Record linking: The design of eYcient systems for linking records into individual and family histories. American Journal of Human Genetics, 19, 335–359. Newcombe, H. B. (1988). Handbook of record linkage: Methods for health and statistical studies, administration and business. New York: Oxford University Press. Schaie, K. W. (1965). A general model for the study of developmental problems. Psychological Bulletin, 64, 92–107. Tu, S., Mason, C. A., & Song, Q. (this volume). Second‐order linkage and family datasets. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 53–78). New York: Elsevier. UnitedStatesDepartmentof Education(2002). Legislative history of majorFERPAprovisions.2006. United States Department of Health and Human Services (2002). Framingham heart study. 2006, from http://www.ed.gov/policy/gen/guid/fpco/pdf/ferpaleghistory.pdf. United States Department of Health and Human Services (2003). Medical privacy—National standards to protect the privacy of personal health information. 2006, from http://www. hhs.gov/ocr/hipaa/. United States Department of Health Human Services (2006). HHS awards contracts to advance nationwide interoperable health information technology. 2006, from http://www. dhhs.gov/news/press/2005pres/20051006a.html. University of North Carolina Population Center (2004). The Lewis Terman study at Stanford University. From http://www.cpc.unc.edu/projects/lifecourse/terman.

Second‐Order Linkage and Family Datasets SHIHFEN TU AND CRAIG A. MASON COLLEGE OF EDUCATION AND HUMAN DEVELOPMENT THE UNIVERSITY OF MAINE, ORONO, MAINE; AND MAINE’S UNIVERSITY CENTER FOR EXCELLENCE IN DEVELOPMENTAL DISABILITIES THE UNIVERSITY OF MAINE, ORONO, MAINE

QUANSHENG SONG MAINE’S UNIVERSITY CENTER FOR EXCELLENCE IN DEVELOPMENTAL DISABILITIES THE UNIVERSITY OF MAINE, ORONO, MAINE

I.

INTRODUCTION

While most linked datasets bring together information solely on individuals, creating linked family datasets for family and sibling research is possible. In this chapter, we will address the issues involved in creating databases so that family relationships can be readily determined and followed. We will also discuss the benefits and limitations of family‐based datasets and the implications of such methods in research.

II.

WHAT IS A SECOND‐ORDER LINKAGE AND WHAT ARE ITS BENEFITS?

Researchers are increasingly interested in linking information from multiple sources as a means of creating larger, more comprehensive pictures of individual health and development. Such databases are valuable for examining a host of issues in the areas of health, education, and disabilities (Kerber & Slattery, 1997; McLennan, Kotelchuck, & Cho, 2001; Sanders & Horn, 1998; Saunders & Heflinger, 2004; Thompson et al., 2003). In the chapter by Urbano, readers were given an introduction to the technology currently available for linking databases. In this chapter, we will discuss how INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

53

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33003-0

54

Shihfen Tu et al.

to create such multisource datasets so that the information they contain can easily be organized into families or groups of connected individuals. We refer to this technique as a second‐order linkage (Tu & Mason, 2004). Datasets that are organized on the family level are particularly helpful in studying the eVect of genetic and environmental factors on health and disability related issues (Czene, Lichtenstein, & Hemminki, 2002; Gilger et al., 1998). Such datasets allow researchers and health oYcials to automatically draw connections between related individuals. When one person is found to have a particular condition, a second‐order database organized around family structures makes it possible to readily ascertain whether other related individuals also have the same condition. Moreover, if information, such as addresses, is also incorporated into a database, then further geographical analyses (Kirby, this issue) can help identify environmental factors that may potentially be associated with the condition in question by examining a family’s migration pattern over time. Finally, data from multiple sources and multiple family members can provide a more comprehensive picture of the fluid sociodemographic changes families may be experiencing over many years. As one can imagine, there are many potential benefits and challenges in building a second‐order database. In the remainder of this section, we will describe some of the possible benefits. Toward the end of this chapter, we will discuss some of the challenges one faces in this work. A.

Enhanced Tracking Capacity for Public Health Officials

Because a second‐order database connects data from related individuals across multiple sources and maintains all information over time, it has many potential benefits. First, it enhances a public health oYcials’ ability to identify and track individuals. As is often the case in an unlinked, single purpose database established by government agencies (e.g., newborn hearing screen database or electronic birth defect registry), information in a record becomes outdated with time. People change their names, addresses, marital status, or educational level. Unfortunately, it is often rare that any of the original information gets updated. At the same time, there are many important reasons why public health oYcials and researchers may need the most up‐to‐date information for tracking purposes. For example, health oYcials and researchers may need to track individuals to ensure continuing service, to monitor health status, or to evaluate the eVectiveness of an early intervention program. However, it is labor intensive and cost prohibitive to consistently update the information of each record in a database. By creating a second‐order database, updated information that is already being collected from other sources can be used to facilitate this eVort at a relatively low cost.

SECOND‐ORDER LINKAGE

55

To illustrate this point, consider the following example. A newborn screened negative for hearing loss. However, the same child was born with a certain risk factor, which puts him at risk for developing hearing loss later. According to the current protocol recommended by the Centers for Disease Control and Prevention, the parent of this child should be contacted on a regular basis so that the parent may have the child’s hearing evaluated periodically to ensure early intervention, should the hearing loss develop later (CDC National EHDI Goals, 2005). If the child’s family moves, it may be diYcult to track him. However, a second‐order database in which records are linked across related individuals and across multiple sources (e.g., screening data, immunization records, birth certificates) can potentially provide a solution. For example, a public health oYcial can potentially obtain the child’s current address through the newborn hearing screen record of his younger sibling (should one exist) or immunization data of an older sibling. In essence, the multiple sources and multiple connections provided by a second‐order database improve the health oYcials’ ability to update existing information, which in turn enhances their ability to identify and track individuals, and to reduce loss‐to‐follow‐up. B.

Identification of Families and Communities for More Focused Research

Another major benefit of this type of second‐order integrated developmental database is to provide researchers a relatively quick way to create a longitudinal database for studying family factors—genetic, environmental, and sociodemographic—at a population level. Such integrated databases can be very beneficial for conducting research in public health and epidemiology. Here we will present some examples of the applications of second‐order integrated developmental databases. 1. POPULATION‐BASED GENETIC RESEARCH

Perhaps the most obvious application of a population‐based database that identifies family units is in the study of genetically based disorders. In this regard, the Utah Population Database Project is a prime example. It contains electronic copies of the genealogies of the initial Latter Day Saints settlers and their descendents. It includes detailed pedigrees that have been linked to a variety of other public health databases, such as cancer registries, and birth and death certificates. It has been the basis for a large volume of published research on topics such as prostate cancer, preeclampsia, and breast cancer (Boucher & Kerber, 2001; Esplin et al., 2001; Neuhausen et al., 1999). Similar systems have been established using other populations (Cerhan et al., 1999; Hemminki & Czene, 2002; Rudan et al., 2003).

56

Shihfen Tu et al.

2. PSYCHOSOCIAL RESEARCH

In addition to providing appropriate cases for examining the genetic basis of various conditions, a second‐order database can also serve as a tool for conducting family‐based psychosocial research. Using linked population‐ based databases, researchers have previously identified socioeconomic and demographic factors (e.g., maternal education, marital status) on various behavioral or educational outcomes (Hollomon, Dobbins, & Scott, 1998; Qin, Agerbo, & Mortensen, 2003). With the capacity to capture the fluidity of the family configuration, a second‐order database presents a unique opportunity to further examine these psychosocial risk factors within families. For example, one can study the eVect of birth weight variation within a family as a result of exposure to changing psychosocial risk factors over time, such as parents’ education. It also becomes possible to examine eVects among siblings or across multigenerations. 3. MATERNAL AND CHILD HEALTH RESEARCH

Recognizing this potential, some researchers in maternal and child health epidemiology have begun to move beyond simply linking individual child data, to also linking health and birth‐related datasets based on mother’s information as well (Armstrong, 2003; Kotelchuck, 2003; Leiss, 2003). For example, health oYcials in states such as Washington and Massachusetts have been linking birth and death records with hospitalization records (Armstrong, 2003; Kotelchuck, 2003). In cases of birth defects or fetal death, the linked database provides important additional information on mothers’ pregnancy history. Moreover, as a database grows over time, it will become increasingly possible to identify the records of females appearing as both a mother and a child in the same system (Armstrong, 2003). Such database will provide a valuable source of data on maternal health from birth to parenthood. III.

OVERVIEW OF DATA LINKAGE METHODOLOGY

In order to understand the techniques and issues involved in creating a second‐order linkage, a review of some key general aspects of data linkage methodology is helpful. Those readers who are familiar with the methodology of data linkage may choose to skip this section. Data linkage involves pairing one or more records belonging to a specific individual in one database with one or more records belonging to that same individual in a diVerent database. The linkage is done by matching records based on a set of identifying fields, such as names or dates of birth, that are common to both databases. This record matching can be based on either a deterministic or probabilistic protocol.

SECOND‐ORDER LINKAGE

A.

57

Deterministic Linkage

In order to be paired through a deterministic linkage, records in the two databases are required to exactly match on all identifying fields used. Deterministic matching is most eVective and eYcient when it can be done using the fewest number of identifying fields, containing high quality and highly discriminating data. For example, gender is typically not a particularly useful identifier because it does a poor job discriminating between unique individuals: for any random person, half of the population will share his or her gender. In contrast, social security number is highly discriminating in that theoretically each person’s social security number is unique. However, if data are of poor quality, a highly discriminating field is still of limited use. For example, a database may contain social security numbers, but if they are often wrong, it may be a poor field to use—particularly in a deterministic linkage that requires all identifying fields to exactly match. Alternatively, a linkage based on too few identifiers may not be able to diVerentiate between diVerent individuals. For example, a match based on First Name and Last Name may be able to uniquely identify Zbigniew Brzezinski, but may be ineVective at uniquely identifying Maria Gonzalez. Adding additional identifying fields will increase the ability to diVerentiate individuals, but will also increase the likelihood that two records for the same individual will not exactly agree on all identifiers and thus not match. When data are of particularly high quality, or when common, preexisting unique identifying numbers (IDs) exist across multiple datasets, deterministic matching can be a quick and eYcient method of data linkage. However, in some situations, this approach can have significant drawbacks. For example, identifiers, such as names or middle initials, may be misspelled (e.g., ‘‘Johm L. Smith’’) or missing (e.g., ‘‘John Smith’’); or, an alternative name (e.g., ‘‘Jonathon L. Smith’’) may be used. In each of these cases, a deterministic match would yield a result of ‘‘no match,’’ even though records for the same individual exist in both datasets. Such nonlinks may potentially result in a systematic bias in the linked records. For instance, ethnic groups that have unfamiliar names or letter combinations, or that may be more likely to use nontraditional spellings would be at a higher risk of not matching in a deterministic linkage. For these cases, a probabilistic linkage may oVer a viable alternative solution. B.

Probabilistic Linkage

Unlike a deterministic linkage, a probabilistic linkage does not require an exact match of all identifiers from the source datasets in order to conclude that two records belong to the same individual. It statistically estimates an

58

Shihfen Tu et al.

index of the likelihood or odds that two records belong to the same individual, even if they disagree on some fields. As with a deterministic protocol, various characteristics of the data will impact one’s ability to perform a probabilistic match. As will be discussed later, some of these are particularly relevant when performing a second‐ order linkage such as organizing data into family groupings. The first of these is the quality of the data, defined as the accuracy and/or reliability of information contained in the identifying fields. If there are many incomplete or inaccurate data entries in a field (i.e., poor data quality), then correct matches will begin to approach a random process—or simply become impossible. While both a deterministic and probabilistic protocol will be impacted by data quality, all else being equal, a probabilistic linkage gives greater weight to higher quality fields. The second factor impacting probabilistic matching is the frequency of field values. The more common the value in a field, the greater the odds that two records will erroneously be matched. Hence, a match based on the name of ‘‘Szapocznik’’ is more likely to reflect a correct match than one based on the name ‘‘Smith.’’ While the concept is similar to a field’s ability to discriminate individuals in a deterministic match, in a probabilistic protocol, possible matches that are based on rare values in a field are given greater weight than are possible matches that are based on common values in that field. The third factor impacting probabilistic matching is the actual number of matches that exist across the two databases. The greater the number of individuals in one database that also appears in the other database, the greater the probability of a correct match. In the most extreme case, if two databases have no individuals in common, the probability of a linkage across the databases must be zero no matter how high the quality of the data are or how closely two records appear to match. Computationally, the probabilistic protocol is a much more complicated and as such, expensive and time‐consuming strategy than is the deterministic protocol. However, probabilistic linkage provides an alternative to deterministic linkage when it is important to minimize the number of overlooked matches due to small inconsistencies in the data. IV.

HOW TO CONDUCT A SECOND‐ORDER DATA LINKAGE

Regardless of whether one uses a deterministic or probabilistic approach, the product of the data linkage project is typically a table or a set of related tables in which data regarding an individual are connected across multiple sources. For example, linkage of a metabolic screening registry with a birth defects registry would provide data regarding both types of information in a

SECOND‐ORDER LINKAGE

59

single source. Such data linkages can help provide more eYcient, timely, and comprehensive care for children and families, as well as provide opportunities for public health research that would not be possible from either source alone. However, linkage protocols are typically only concerned with connecting individuals across systems. While such protocols may provide a more comprehensive picture regarding specific individuals, they are limited in not allowing one to readily examine patterns of data across groups of related individuals. For example, the previously described linkage may identify an individual with a particular combination of a metabolic disorder and birth defect, but in general, the linkage would not be done in a manner in which it would be possible to easily identify siblings, parents, or cousins of that individual in order to determine whether other relatives had either or both of these disorders. In fact, in most cases, the linkage would not be done in a manner in which it would be possible to easily determine whether many of these persons even existed. A second‐order linkage corrects both of these limitations. A.

Establishing ID Numbers

The first step in creating a second‐order linkage is to identify the records of individuals who are connected in some particular way (e.g., genetically or geographically). The second step is to find a way to group the records of these connected individuals in one dataset so that they can be easily linked to the records of these same individuals in other datasets. Ultimately, one needs to generate unique ID for individuals in the dataset, which also allow one to group connected individuals in that same dataset. In this section, we will discuss two approaches: creating a unique Family ID versus the use of an expanded ID table. 1. FAMILY ID

Perhaps the simplest way to organize data based on families is to assign a unique ‘‘Family ID’’ to each family. For instance, if mother’s information is available, then every child with the same mother may be assigned an identical Family ID. The Family ID would be included in each person’s individual record and would serve as an additional field that could be used for searching, grouping, or aggregating records. Consequently, it would be fairly easy to search for family units with the same mother. The limitation of the Family ID is that it is based on the implicit assumption that families are largely static. A Family ID is restricted in that it does not readily incorporate the various types of families present in today’s society. For example, a mother can get divorced, get remarried, change her last name, and have another child with her new partner. In such cases, the

60

Shihfen Tu et al.

mother’s information would not be identical for all children who were born to her. A static Family ID would not be able to capture the dynamics of these cases. Similarly, if one wished to search for children who live in the same household, a static Family ID would not be able to capture some relationships, such as stepsiblings. In short, a more flexible solution is required that can capture the complex fluidity of families, including blended families, step/half‐siblings, extended and multigenerational families. 2. EXPANDED ID TABLE

a. Background. Ironically, the key to capturing the complexity and fluidity of families is not to tie individuals to families, but rather to tie individuals to other individuals. As described in the following section, this can be accomplished through the use of an expanded ID table. An ID table contains ID numbers uniquely identifying individuals in the database, along with identifying fields used to diVerentiate records.1 Typically, in a child‐oriented database (e.g., birth certificate or birth defects registry), the names of the child’s mother and father serve only as additional potential identifiers for linking the records. The parents themselves along with their names and information do not appear in the database as unique individuals with their own separate records. As such, the records of children who were born to the same parent are treated as entirely distinct records and cannot be easily connected. An example of a child‐oriented database can be seen in Table I. In this case, two children, Zachary Abrams and Tyler Abrams, share the same mother, although her name appears slightly diVerent in the two records. Consequently, there is no readily available means for identifying them as siblings. In practice, this is an oversimplification, as the siblings may not share the same last name and their mother’s name may in fact change. Moreover,

AN EXAMPLE

OF A

TABLE I TYPICAL CHILD‐ORIENTED DATABASE

Last name

First name

Date of birth

Mother’s last name

Mother’s first name

Mother’s date of birth

Abrams Abrams

Tyler Zachary

3/21/2003 7/9/2001

Abrams Abrams

Catherine Katherine

12/18/1974 12/18/1974

1

In reality, there are multiple ways in which one can design an ID table that includes varying degrees of identifying information. For simplicity in our presentation, we will assume the ID table contains ID numbers and all identifying fields used in data linkage.

61

SECOND‐ORDER LINKAGE

as the database grows over time, a child will ultimately mature and have children, and consequently appear in some records as a parent herself. A traditional linkage would not allow for the possibility of connecting a record of an individual as a child and a record of his or her oVspring. b. Interrelated IDs. The solution is to have a parent assigned an ID number in exactly the same manner and nature as the child. An ID table entry for a child includes the child’s name, but lists each parent’s ID number, instead of the parent’s name. Rather than simply appear as an additional field in the child’s record, the name and information of a parent is stored in his or her own unique, separate record, identified by his or her own unique ID number. Reflecting this, the heart of a second‐order database is a central ID table that contains identifying information and unique ID numbers for all individuals that appear in any field in any record. All individual information contained in the database is linked through this common set of ID numbers. As illustrated in Table II, if a mother has multiple children in a database, these children will each have their own unique ID, but will share a common IDMother even if her name changed over time. In this example, Zachary Abrams and Tyler Abrams share a common IDMother, which can be used to group them. Similarly, if a father’s name and information is available, an ID can also be assigned to him and all of his children can be connected as a result. c. Multiple Records. Name changes and spelling variations can be incorporated by allowing unique individuals to have multiple entries in the ID table, corresponding to diVerent variations of information, possibly at diVerent times. For example, in Table II, the mother of Zachary and Tyler Abrams appears twice in the ID table, as both Catherine and Katherine. A logical question emerges as to why multiple variations of identifiers are all stored and maintained in the database—particularly, if one becomes aware that one variation (e.g., first name of ‘‘Catherine’’) is in fact wrong or has changed and is now out‐of‐date. However, this strategy oVers several benefits. To start, one is not placed in a position where one must determine which

AN EXAMPLE

OF A

TABLE II DATABASE WITH INTERRELATED IDS

ID

Last name

First name

Date of birth

IDMother

IDFather

4146709184 4146709184 7608392589 451633756

Abrams Abrams Abrams Abrams

Catherine Katherine Tyler Zachary

12/18/1974 12/18/1974 3/21/2003 7/9/2001

4146709184 4146709184

322608529

62

Shihfen Tu et al.

competing information is ‘‘correct.’’ In addition, as described in the next section, it becomes possible to exactly recreate original records. This strategy can also be helpful in situations where information in the record (e.g., last name) is known to have changed, but where future data linkages may be based on older data or databases in which the name was not updated. d. Record IDs. By replacing the string value of a mother or father’s name with an ID number, it may become impossible to exactly recreate the original record. For example, based on the ID number that appears in the IDMother field, it is impossible to determine whether the actual maternal name that appeared on Zachary’s record was Catherine or Katherine. This problem can be addressed by including an additional ID number—not an ID number that uniquely identifies a person, but rather an ID number that uniquely identifies a specific record in the ID table. An example is provided in Table III through the Record IDMother field. Record IDMother essentially serves as a second ID number—uniquely identifying a specific combination of identifiers for a specific individual, rather than simply diVerentiating unique individuals. By adding this second ID, it becomes clear that not only do Zachary and Tyler have the same mother, but her name appeared as ‘‘Katherine’’ in Tyler’s record and as ‘‘Catherine’’ in Zachary’s. B.

Cascading Algorithms for Establishing Families

As described later, the expanded use of related IDs and record IDs, in combination with various search algorithms, allows one to capture the complex fluidity of family structures. 1. WITHIN AN IMMEDIATE FAMILY

In addition to the parent–child relationship, there are two general relationships that can be established within an immediate biological family. These are illustrated in Fig. 1, and correspond to full‐siblings, in which both parents are the same across diVerent individuals, and potential full‐siblings, in which one parent matches for two children, but one or both of the children is missing information on the other parent. At a minimum, these children would be half‐ siblings, although they may in fact be full‐siblings depending on whether they also share the unknown parent. An algorithm can determine full‐sibling relationships for a child in question by first identifying her biological parents (#1 Paths) and then identifying any other individuals in the database sharing the same pair of mother and father IDs (#2 Paths). An algorithm can determine potential full‐sibling relationships by identifying any individuals in the database who share either the same mother or father ID (Path #3) and are missing the ID for their other parent (Path #4).

AN EXAMPLE

OF A

DATABASE

TABLE III INTERRELATED IDS

WITH

AND

RECORD IDS

Record ID

ID

Last name

First name

Date of birth

IDMother

Record IDMother

694977 588610 184243 410342

4146709184 4146709184 7608392589 451633756

Abrams Abrams Abrams Abrams

Catherine Katherine Tyler Zachary

12/18/1974 12/18/1974 3/21/2003 7/9/2001

4146709184 4146709184

588610 694977

IDFather

Record IDFather

322608529

467951

64

Shihfen Tu et al. Unknown

Mother

4

3

Father

1

Potential full-sibling

2

1

Child in question

2

Fullsibling

FIG. 1. Sibling relationships in an immediate family.

Mother

1

Father

1

Child in question

2

Fullsibling

Stepmother

4

3

Half sibling

Unknown

5

6

Potential half-sibling

FIG. 2. Sibling relationships in a blended family.

2. WITHIN A BLENDED FAMILY

There are also two general relationships that can be established within a blended family (i.e., family in which all children do not share both parents). These are presented in Fig. 2. In a half‐sibling relationship, either the mother matches across diVerent individuals, but the father does not (maternal half‐ siblings), or the father matches across diVerent individuals but the mother does not (paternal half‐siblings). In either case, both parents must be known, or the relationship would revert to one of a potential full‐sibling, which again would at a minimum be a half‐sibling relationship. In addition, a more tenuous relationship to establish is the potential half‐sibling. This scenario occurs when (1) one child is missing information on one parent, (2) the nonmissing parent is diVerent for the two potential half‐siblings, and (3) the nonmissing parent for the first child had children with the opposite‐sex parent for the second child.

SECOND‐ORDER LINKAGE

65

An algorithm can determine half‐sibling relationships for a child in question by first identifying his or her biological parents (#1 Paths) and then identifying any individuals sharing one of those parents’ IDs (Path #2), but having a diVerent ID for his or her other parent (Path #3). Establishing potential half‐sibling relationships is slightly more complicated. One must first identify stepparents (or nonmarital equivalents). If marriage information is available in a database, one can begin by identifying all ID numbers for any person, other than the parents of the child in question, who was also married to one of his or her parents (Path #4). Alternatively, if one has previously determined half‐sibling relationships, one can identify all ID numbers for parents of these half‐siblings. Once ID numbers for stepparents (or nonmarital equivalents) have been identified, one can search the database for individuals who have one of these IDs as parent (Path #5) and for whom the other parent is unknown (Path #6). A few points are worth noting in determining relationships in blended families—both conceptually and computationally. First, in identifying stepparents (or nonmarital equivalents), neither the use of marriage records nor the half‐sibling parent data will fully capture all such relationships. Marital records alone will miss families with unmarried parents, and half‐sibling parent data will only identify stepparents who in fact had a child with the person’s own parent. Second, depending on the nature of the data—particularly the degree to which parent data are missing—the number of ‘‘potential’’ half‐siblings may be large and predominantly cases that are in fact not half‐siblings. Consequently, establishing this type of relationship may not be of practical use. Nevertheless, this does illustrate the potential for modeling the outward extension of an immediate family. 3. WITHIN AN EXTENDED FAMILY

Through an iterative feedback of ID numbers, it also becomes possible to identify individuals in multigenerational extended families. As extended families incorporate information from successive generations, the number of possible relationships grows exponentially, consequently in this section we will focus on three. These are presented in Fig. 3. In a grandparent relationship, parent information regarding a parent of the child in question is available. In an aunt/uncle relationship, siblings of a parent can be determined through a manner similar to that in which a child’s own siblings are identified. Finally, cousins are identified as children of an identified aunt/uncle. Algorithms for establishing these relationships become increasingly complex and both resources and time consuming—although the actual process itself is somewhat iterative. One can determine grandparent relationships for a child in question by first identifying his or her biological parents (Path #1) and then identifying the parent ID numbers for these parent records

66

Shihfen Tu et al. Grand mother

Grand father

3

Aunt (nonbiological)

2

Uncle

4

Cousin

Cousin

3

2

Mother

Father

1

Child in question

Fullsibling

FIG. 3. Relationships in an extended family.

AN EXAMPLE

OF A

TABLE IV MULTIGENERATIONAL RELATIONSHIP

IN A

DATABASE

(#2 Paths). This iterative process is also illustrated in Table IV, which shows a snapshot of data for three individuals within an ID table. In Table IV, the successive linkage between ID and IDMother across the three records connects the child with the mother and the grandmother.

SECOND‐ORDER LINKAGE

67

Once ID numbers for grandparents have been identified, one can search the database for individuals who have one or both of these IDs as a parent (#3 Paths in Fig. 3) in order to identify aunt/uncles or half‐aunt/uncles, respectively. ID numbers for aunt/uncle relationships can then be used to search for cousins who have one of these IDs as a parent (Path #4). From this, it should be clear that the process could be extended backward to include more distant relatives, as well as extended forward to include related individuals in subsequent generations such as nephews and nieces. Obviously, in order to completely capture all members within these extended families, their records must exist in the database to begin with. C.

Database Structure

The shift from storing parent information as fields within a child’s record to parents having their own individual records that can grow and change over time has major implications in the design and programming of a database. One immediate consequence is the need to allow individuals to have multiple values in fields that may otherwise have had only a single value. For example, a typical unlinked birth certificate registry can be designed so that every individual—meaning every child—has a single entry. Every field, every piece of information will have one value for each individual. This simple database could consist of a single table or flat file, possibly labeled ‘‘Birth Information,’’ containing all information from a birth certificate, with each child corresponding to one record. If this same birth certificate registry is reorganized as a second‐order database, this simplicity and ease is lost. With this change, it becomes advantageous to use a relational database consisting of multiple tables based on patterns and qualities of data, which are linked together based on an individual’s ID number. The manner in which multiple tables within a database are linked through common identifiers is described by one of several successively complex normal forms, which typically serve as the basis for the structure underlying a relational database (Date, 1981). The first, and most fundamental normal form, pertains to the overall structure or shape of tables in a database. In order to satisfy the first normal form, each field within a table should contain a single piece of information and there should be no repeating fields or groups of fields. As a result, all records should contain the same number of fields corresponding to the same amount of information. For example, given a child can have more than one birth defect, a table in a birth defects registry could easily violate the first normal form by storing this information in a string field that was simply a listing of all birth defects identified for a child. In contrast, by storing birth defect information in a separate table, with each record corresponding to a

68

Shihfen Tu et al.

Child Information IDChild

First Name

Last Name

DOB

IDBirth Hospital

Hospital

43563 43564

Susan Cameron

Schwartz York

1/15/2003 11/4/2003

16 23

General Hospital Central Hospital

Reported Birth Defects IDChild

ID BD

Birth Defect

ICD9

Date of Diagnosis

43563 43563 43563 43564 43564

1 2 3 1 2

Aortic valve stenosis Cleft palate without cleft lip Down syndrome Aortic valve stenosis Cleft palate without cleft lip

746.3 749 758 746.3 749

9/26/2003 1/15/2003 2/6/2003 12/9/2003 11/4/2003

FIG. 4. An example of a database conforming to the first normal form.

single birth defect for a single child, the tables presented in Fig. 4 satisfy the first normal form. Information in these tables is linked by the child’s ID number: IDChild. Subsequent normal forms involve the use of keys, which are fields that uniquely identify a specific record in a table (e.g., a unique ID number). Keys can be single fields or combinations of multiple fields—referred to as concatenated keys—that uniquely identify individual records. A concatenated key is seen in the table Reported Birth Defects presented in Fig. 4 because the key is based on a combination of both IDChild and IDBD. Together, these uniquely identify each record in that table. When a key consists of multiple fields, second normal form requires that all information in a record pertain to all fields used in the key. For example, the table Reported Birth Defects in Fig. 4 violates the second normal form because two fields—BirthDefect and ICD9—correspond solely to one part of the key, IDBD. They do not pertain directly to IDChild. To satisfy the second normal form, this information would need to be moved into a separate table linked to Reported Birth Defects through IDBD. In contrast, DateofDiagnosis does pertain to both IDBD and IDChild because the diagnosis date is unique to both a specific child and a specific birth defect for that child. Figure 5 illustrates how these tables can be organized in order to satisfy the second normal form. The third normal form requires that all fields in a table pertain to the information represented in that table’s key. This is violated when one or

69

SECOND‐ORDER LINKAGE

Child Information IDChild First Name 43563 43564

Susan Cameron

Last Name

DOB

IDBirth Hospital

Hospital

Schwartz York

1/15/2003 11/4/2003

16 23

General Hospital Central Hospital

Reported Birth Defects IDChild

ID BD

Date of Diagnosis

43563 43563 43563 43564 43564

1 2 3 1 2

9/26/2003 1/15/2003 2/6/2003 12/9/2003 11/4/2003

Birth Defects ID BD

Birth Defect

ICD9

1 2 3

Aortic valve stenosis Cleft palate without cleft lip Down syndrome

746.3 749 758

FIG. 5. An example of a database conforming to the second normal form.

more fields contain information about another, nonkey field. An example of this violation is seen in the Child Information table in Fig. 5. Specifically, this table has a single key—IDChild. To satisfy the third normal form, all other fields should relate to IDChild. However, Hospital is a string field containing the name of the hospital referenced by IDBirthHospital. As such, in order to satisfy the third normal form, the Hospital field should be stored in a separate table, linked to Child Information through IDBirthHospital. Figure 6 illustrates how these tables can be organized in order to satisfy the third normal form. As a general rule, a database should be designed to satisfy at least the third normal form, although additional, even more complex normal forms exist and continue to be developed. Fortunately, the trade‐oV oVered by this increased complexity is a naturally evolving historical or developmental database, which can track changes occurring within families over time. For example, parent education is standard information on birth certificates. If a parent has multiple children whose records are stored in the dataset, changes in parent education over time can be tracked. By linking the datasets on a second‐order level, one can now examine the same parent’s education level for every one of his/her children whose birth record exists in the database. For example, one would be able to contrast outcomes for children of teenage mothers who went on to continue school versus those whose mothers did not continue with their education. Without a second‐order linkage, these are treated as separate, independent records and such information is lost.

70

Shihfen Tu et al.

Child Information IDChild First Name 43563 43564

Last Name

DOB

IDBirth Hospital

Schwartz York

1/15/2003 11/4/2003

16 23

Susan Cameron

Hospital Information IDBirth Hospital

Hospital

16 23

General Hospital Central Hospital

Reported Birth Defects IDChild

ID BD

Date of Diagnosis

43563 43563 43563 43564 43564

1 2 3 1 2

9/26/2003 1/15/2003 2/6/2003 12/9/2003 11/4/2003

Birth Defects ID BD

Birth Defect

ICD9

1 2 3

Aortic valve stenosis Cleft palate without cleft lip Down syndrome

746.3 749 758

FIG. 6. An example of a database conforming to the third normal form.

V.

THE CHALLENGES OF BUILDING A SECOND‐ORDER DATABASE

While second‐order linkage expands the ability of public health oYcials and researchers to enhance services and conduct research, the benefits do come with potential challenges. This section discusses these issues, starting with several significant technical diYculties associated with programming, before addressing several key legal and ethical issues. A.

Computational Demands

First, the computational demands required for a second‐order linkage are significantly greater than what is required for a traditional linkage. A second‐ order linkage requires sequentially linking multiple individuals—for example, to identify full‐siblings requires one to link a child to two parents, and then link these parents to a sibling, while identifying cousins requires one to link a child to parents, parents to grandparents, grandparents to an uncle or

SECOND‐ORDER LINKAGE

71

aunt, and an uncle or aunt to a cousin. Given these multiple sequential associations, it is important that a linkage protocol match as many records as possible. Consequently, a probabilistic matching protocol, with its increased programming requirements, will likely be necessary, rather than a more straightforward deterministic match. Similarly, the algorithms used to establish familial relations add further complexity to a database and the execution of these algorithms can require multiple iterative passes through the dataset. The result can be reduced speed and performance for queries. Finally, as described in Section IV.C, the number of tables required in a second‐order database can increase rapidly, also reducing speed and increasing complexity. While the reduced speed of operation can be addressed with upgraded hardware, increased complexity requires having access to skilled programmers. B.

Difficulty in Evaluation

A second challenge facing any data linkage project is critically evaluating the overall quality of the linkage process itself. Even in a traditional linkage, assessing the quality of the record matching protocol can be challenging. One common strategy is to manually or ‘‘hand‐link’’ a subset of records and compare these with the results obtained through the electronic linkage. However, this approach assumes that the manual linkage correctly matches records, an assumption that may not be correct. A person performing a manual link may filter or scan through records in a way that he or she overlooks the correct match, or may select what is perceived as the closest match simply because a stronger match was overlooked. Similarly, a person conducting a manual link may overestimate the uniqueness of two similar, but not identical records, and erroneously conclude that they are a correct pairing; or may underestimate the uniqueness of two records and erroneously conclude that they are not a match. Ultimately, hand matching does not guarantee that two records do or do not belong to the same individual; it simply reflects one person’s opinion. This situation is further complicated when performing a second‐order linkage. Estimating the degree to which a linkage protocol correctly pairs records for individuals is challenging enough—estimating the degree to which a protocol not only pairs records for individuals but also identifies siblings, half‐siblings, or cousins is dramatically more diYcult. A solution we have proposed elsewhere (Tu, Song, & Mason, 2003) is to use simulated population data as a means of evaluating the overall quality of a data linkage protocol, as well as identifying specific areas or types of matches that are problematic.

72

Shihfen Tu et al.

POpulation Database Simulator (PODS) is a program created by the authors to generate simulated multigenerational population databases. It was written using MS SQL and Visual Basic. With PODS, users first define the size of the initial population and the number of years this simulated population will ‘‘grow’’ (e.g., 100 years). In addition, users can define a broad range of demographic parameters such as the migration rates (both in‐ and out‐), birth and death rates (including accidental death rates), marriage and divorce rates, etc. To allow for the fine‐tuning of population behavior, the parameters can be defined using a series of operators, including all major math functions, equality/inequalities, and logic rules. For example, users may determine varying pregnancy rates based on both age and marital status. Once the process begins, the simulation proceeds in 1‐year increments, with each simulated ‘‘person’’ randomly experiencing various ‘‘life‐events.’’ Basically, simulated individuals will be born, attend school, have children, get married, get divorced, and die, as well as move away from or immigrate into the simulated population. Throughout this process, user‐defined health records are generated reflecting the status of an individual at that time. Given data are created in a lifelike process, all relevant identifiers will change appropriately over time. For example, in 1 year, a simulated unmarried woman may have a child. A birth certificate would be generated for the child, who would likely have the mother’s last name. In a subsequent year, the same woman may marry and have a second child. A birth certificate for this child would more likely have the father’s last name. Consequently, simulated health records would reflect the real‐life phenomenon of siblings having the same mother, but diVerent last names for themselves and their mother. Using this program, a simulated population is created, and datasets corresponding to the data being linked are generated for these simulated individuals, with predetermined rates of errors and missing values. A linkage protocol can then be performed on the simulated datasets. Because the data are created by the user, one in fact knows which records should match, as well as all family relationships. One can then assess the degree to which the protocol in question accurately matched records and recreated family relationships. Furthermore, the parameters used in generating the simulated data can be changed and the entire procedure run again in order to determine the impact of specific types of missing data or data quality issues. C.

Linkage Creep

A third, subtler challenge is a result of changes in identifying fields over time. For example, the names of many individuals will change multiple times

SECOND‐ORDER LINKAGE

73

over several years. Such name change can become a particular issue for second‐order data linkage, which benefits from having data over as extended a time period as possible. With increasing variations in an individual’s identifiers comes an increasing likelihood of linkage creep. To illustrate linkage creep, consider a protocol that only matches two records if they have a probability of .90 or more of belonging to the same person. If a prior linkage found that the probability Record A and Record B belonged to the same person was .90, one would have concluded that it was a correct match and assigned both records the same ID number. A new linkage may now find the probability that Record B and a new record, Record C, belong to the same person is also .90. Again, one would conclude that this is a correct match and assign both records the same ID number. Records A, B, and C would now all have the same ID number and all be seen as belonging to the same person. However, given the sequential linking of A to B and B to C, the probability that A and C belong to the same person is likely to be less than .90. As a result, the .90 criteria are not satisfied for the A–C pairing, and these two records should arguably not be given the same ID number. A verification of inconsistencies in the dataset should flag these records as potentially problematic. How does one address this inconsistency? In some cases, the three (or more) records do correspond to the same individual, while in other instances the records reflect two or more diVerent individuals. The issue is further complicated by the fact that depending on sources for future linkages, the same inconsistency may reemerge repeatedly in the future. One solution is to use a running temporal filter when evaluating for this type of inconsistency. For example, records may be compared only to other records that are dated within 5 years of each other. Alternatively, we would recommend manually evaluating these occurrences and adding an Override field to the ID table. This field flags specific records in the ID table by containing the same value for those records that must be assigned the same ID number and diVerent values for those records that must not be given the same ID number. All other records would have a null value in the Override field and are allowed to match—or not match— with any other record, including those with non‐null values in the Override field. For example, three records with the same value in the Override field indicate they all must be given a single, common ID number. Alternatively, two could be given the same value and a third given a diVerent value in the Override field, indicating that the former two belong to the same person with the same ID number, while the third must belong to a diVerent person with a diVerent ID number. An algorithm should evaluate for inconsistencies resulting from Override field entries and require correction by the user.

74 D.

Shihfen Tu et al. Concerns Regarding Privacy and Confidentiality

One of the advantages of a second‐order family dataset is the ability to derive new information by linking existing datasets. Information that is ‘‘buried’’ in the unlinked or even traditionally linked datasets can be more visible in a second‐order dataset. For example, by linking a birth certificate database and a birth defect registry on the family level instead of the individual level, a public health oYcial can find the siblings or other relatives of a child who has a specific birth defect. As noted previously, while second‐ order linkage provides a good opportunity to study factors influencing birth defects (e.g., genetic or environmental factors), some will be concerned regarding privacy and confidentiality. Given justified concern over the possibility of intentional, or even more likely, unintentional abuse or misuse, it is critical that any work in this area addresses privacy and confidentiality concerns. A minimum framework one must examine in a second‐order family dataset—or in any data linkage project—consists of two federal privacy acts safeguarding health information (HIPAA) and educational records (FERPA). 1. HEALTH INFORMATION

Many of the examples discussed in this chapter involve accessing databases of health records. If a second‐order family database is created by linking one or more health datasets from diVerent sources, then it is important to determine whether any of the datasets comes from a covered entity under the Health Insurance Portability and Accountability Act (HIPAA) of 1996. If the answer is yes, then it is important to follow the HIPAA guidelines with its restrictions on accessing individually identifiable information. However, there is exemption built into HIPAA to allow public health authorities to conduct vital and necessary activities. Under HIPAA, a public health authority is ‘‘authorized by law to collect or receive such information for the purpose of preventing or controlling disease, injury, or disability, including, but not limited to, the reporting of disease, injury, vital events such as birth or death, and the conduct of public health surveillance, public health investigations, and public health interventions; or, at the direction of a public health authority, to an oYcial of a foreign government agency that is acting in collaboration with a public health authority’’; (U.S. Department of Health and Human Services, OYce for Civil Rights, 2003, p. 1). A public health entity may disclose protected health information for the public health activities and purposes based on HIPAA. It is important to determine whether the agency that is engaged in data linkage is deemed as a ‘‘public health authority.’’

SECOND‐ORDER LINKAGE

75

But HIPAA only sets the lowest standard one must follow; there may be other state laws or regulations that create conflicts with the data linkage activities. Consequently, it is wise to consult with one’s oYce of the state attorney general, particularly if working with state agencies. Having the approval of the state attorney general’s oYce can oVer clarity and confidence in a project, and thus help to alleviate concerns with other agencies. If the group conducting the linkage is not, by itself, a public health authority, but collaborates with and provides a service to a public health authority, it may be worth establishing a formal relationship as an agent of the state or local agency. Provided with appropriate memorandum of agreement, the group may also be considered exempt from HIPAA for public health purposes. Again, consultation with the state attorney general’s oYce is vital in clarifying these issues. 2. EDUCATIONAL RECORDS

Sometimes, health records and educational records are linked for a specific purpose. For example, one may link a birth defects registry with school or special education records to evaluate the eVect of an early intervention program. In such cases, since educational records are involved, linkage activities must also comply with the Family Educational Rights and Privacy Act (FERPA) of 1974, regardless of whether or not a second‐order linkage is carried out. In general, the rules for FERPA are more restrictive than HIPAA. For example, FERPA includes no public health surveillance exemption for accessing identifiable records. There are, however, exceptions to this FERPA restriction for organizations ‘‘conducting studies for, or on behalf of, educational agencies or institutions for the purpose of developing, validating, or administering predictive tests, administering student aid programs, and improving instruction, if such studies are conducted in such a manner as will not permit the personal identification of students and their parents by persons other than representatives of such organizations and such information will be destroyed when no longer needed for the purpose for which it is conducted’’ (U.S. Department of Education, Family Policy Compliance OYce, 2002, p. 7). This significantly increases the challenge of conducting a data linkage project. At a minimum, one must work closely with the Department of Education in one’s state, who must be engaged at no less than a full, active, equal partner. Finally, as noted previously, these legal standards are the minimum requirements one must follow. Research based on such datasets must comply with the corresponding Institutional Review Boards standards. Ultimately, one must struggle with the question of to what degree the organization of existing information into a second‐order database does or does not constitute the

76

Shihfen Tu et al.

creation of new knowledge beyond that which is already known. As such, continued development of strategies for linking sensitive data in ways that maximally enhances security, privacy, and whenever possible, anonymity will be valuable.

VI.

SUMMARY

In this chapter, we have discussed the principles for creating a second‐ order dataset that make it possible to organize information in a database based on families. Such datasets provide a valuable opportunity to conduct truly unique types of research (e.g., genetic, psychosocial) among siblings and across multiple generations. One can also create a longitudinal dataset within a relatively short period of time by performing a second‐order data linkage. The key to creating a second‐order dataset is to establish IDs for all individuals in a record and to establish IDs for all records in a dataset. By doing so, one can identify various relationships within diVerent types of families (e.g., blended family and extended family) and thus capture the fluidity of the modern family structure. One of the benefits of a second‐order dataset is enhanced identification and tracking capacity, which is particularly important to public health oYcials. Because the information is merged from diVerent sources and is organized on the family level, public health oYcials can quickly identify any groups of related individuals sharing a common health concern or characteristic such as a developmental disability or a birth defect. Furthermore, enhanced tracking capacity makes it easier for health oYcials to ensure uninterrupted services to those who are eligible. While the benefits of a second‐order data linkage are many, there are challenges as well. First, it imposes a high demand on computing and programming capacity. It also accentuates the diYculties (e.g., evaluation and linkage creep) seen in more typical data linkage projects based on individuals alone. In this chapter, we have oVered some possible solutions to these challenges, but clearly more work on these methods needs to be done. Finally, there are concerns regarding privacy and confidentiality. Since it is very likely that a second‐order linkage would involve linking health and/ or educational records, it is critical that the guidelines of privacy laws (e.g., HIPAA and FERPA) are closely followed. In closing, we believe that even though there are challenges associated with building a second‐order database, the potential benefits in the areas of service, public health surveillance, and research make such data linkage activity a worthwhile endeavor.

77

SECOND‐ORDER LINKAGE

REFERENCES Armstrong, R. (2003, March). LifeTrac—LifeTime resource, access and care: Building a pregnancy history database. Presentation at the Centers for Disease Control and Prevention, maternal and child health epidemiology program webcasts (March 5, 2003): Linking records for public health research and practice—Lessons from maternally linked birth records. Transcript retrieved from http://www.uic.edu/sph/cade/mchepi/meetings/march2003/index. htm Boucher, K., & Kerber, R. (2001). Measures of familial aggregation as predictors of breast cancer risk. Journal of Epidemiological Biostatistics, 6, 377–385. Centers for Disease Control and Prevention (CDC), National Early Hearing Detection and Intervention (EHDI) Goals (June 24, 2005). Retrieved from http://www.cdc.gov/ncbddd/ ehdi/nationalgoals.htm Cerhan, J. R., Parker, A. S., Putnam, S. D., Chiu, B. C., Lynch, C. F., Cohen, M. B., et al. (1999). Family history and prostate cancer risk in a population‐based cohort of Iowa men. Cancer Epidemiology, Biomarkers & Prevention, 8, 53–60. Czene, K., Lichtenstein, P., & Hemminki, K. (2002). Environmental and heritable causes of cancer among 9.6 million individuals in the Swedish family‐cancer database. International Journal of Cancer, 99(2), 260–266. Date, C. J. (1981). An introduction to database systems (3rd ed.). Boston: Addison‐Wesley. Esplin, M. S., Fausett, M. B., Fraser, A., Kerber, R., Mineau, G., Carrillo, J., et al. (2001). Paternal and maternal components of the predisposition to preeclampsia. New England Journal of Medicine, 344, 867–872. Family Educational Rights and Privacy Act (Educational Rights and Privacy Act 1974). Section 513 of Public Law, 93–380. Gilger, J. W., Pennington, B. F., Harbeck, R. J., DeFries, J. C., Kotzin, B., Green, P., et al. (1998). A twin and family study of the association between immune system dysfunction and dyslexia using blood serum immunoassay and survey data. Brain and Cognition, 36(3), 310–333. Health Insurance Portability and Accountability Act (1996). Public Law, 104–191. Hemminki, K., & Czene, K. (2002). Attributable risks of familial cancer from the family‐cancer database. Cancer Epidemiology, Biomarkers & Prevention, 11, 1638–1644. Hollomon, H. A., Dobbins, D. R., & Scott, K. G. (1998). The eVects of biological and social risk factors on special education placement: Birth weight and maternal education as an example. Research in Developmental Disabilities, 19(3), 281–294. Kerber, R. A., & Slattery, M. L. (1997). Comparison of self‐reported and database‐linked family history of cancer data in a case‐control study. American Journal of Epidemiology, 146(3), 244–248. Kirby, R. S. (this volume). Incorporating geographical analysis into the study of mental retardation and developmental disabilities. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 79–91). New York: Elsevier. Kotelchuck, M. (2003, March). The Massachusetts pregnancy to early life longitudinal (PELL) linkage project: Overview, challenges & opportunities. Presentation at the Centers for Disease Control and Prevention, maternal and child health epidemiology program webcasts (March 5, 2003): Linking records for public health research and practice— Lessons from maternally linked birth records. Transcript retrieved from http://www.uic. edu/sph/cade/mchepi/meetings/march2003/index.htm

78

Shihfen Tu et al.

Leiss, J. (2003, March). Improving the validity of maternally‐linked birth records using external validation. Presentation at the Centers for Disease Control and Prevention, maternal and child health epidemiology program webcasts (March 5, 2003): Linking records for public health research and practice—Lessons from maternally linked birth records. Transcript retrieved from http://www.uic.edu/sph/cade/mchepi/meetings/march2003/index.htm McLennan, J. D., Kotelchuck, M., & Cho, H. (2001). Prevalence, persistence, and correlates of depressive symptoms in a national sample of mothers of toddlers. Journal of the American Academy of Child & Adolescent Psychiatry, 40(11), 1316–1323. Neuhausen, S. L., Farnham, J. M., Kort, E., Tavtigian, S. V., Skolnick, M. H., & Cannon‐ Albright, L. A. (1999). Prostate cancer susceptibility locus HPC1 in Utah high‐risk pedigrees. Human Molecular Genetics, 13, 2437–2442. Qin, P., Agerbo, E., & Mortensen, P. B. (2003). Suicide risk in relation to socioeconomic, demographic, psychiatric, and familial factors: A national register‐based study of all suicides in Denmark, 1981–1997. American Journal of Psychiatry, 160, 765–772. Rudan, I., Smolej‐Narancic, N., Campbell, H., Carothers, A., Wright, A., Janicijevic, B., et al. (2003). Inbreeding and the genetic complexity of human hypertension. Genetics, 163, 1011–1021. Sanders, W. L., & Horn, S. P. (1998). Research findings from the Tennessee value‐added assessment system (TVAAS) database: Implications for educational evaluation and research. Journal of Personnel Evaluation in Education, 12(3), 247–256. Saunders, R. C., & Heflinger, C. A. (2004). Integrating data from multiple public sources: Opportunities and challenges for evaluators. Evaluation, 10(3), 349–365. Thompson, J. R., Carter, R. L., Edwards, A. R., Roth, J., Ariet, M., Ross, N. L., et al. (2003). A population‐based study of the eVects of birth weight on early developmental delay or disability in children. American Journal of Perinatology, 20(6), 321–332. Tu, S., & Mason, C. A. (2004). Organizing population data into complex family pedigrees: Application of a second‐order data linkage to state birth defects registries. Birth Defects Research Part A (formerly, Teratology), 70, 603–608. Tu, S., Song, Q., & Mason, C. A. (2003, November). Use of simulated population data in evaluating data linkage. Poster session presented at the 131st Annual Meeting of American Public Health Association, San Francisco. U.S. Department of Education, Family Policy Compliance OYce: Legislative History of Major, FERPA Provisions (2002). Report retrieved from http://www.ed.gov/policy/gen/guid/fpco/ pdf/ferpaleghistory.pdf U.S. Department of Health and Human Services, OYce for Civil Rights: Disclosures for Public Health Activities [45 CFR 164.512(b)(1)(i)] (2003). Report retrieved from http://www.hhs. gov/ocr/hipaa/publichealth.pdf

Incorporating Geographical Analysis into the Study of Mental Retardation and Developmental Disabilities RUSSELL S. KIRBY DEPARTMENT OF MATERNAL AND CHILD HEALTH, SCHOOL OF PUBLIC HEALTH UNIVERSITY OF ALABAMA AT BIRMINGHAM, BIRMINGHAM, ALABAMA

I.

INTRODUCTION

Epidemiologists focus on the distribution, determinants, and service delivery aspects of disease in humans. Researchers often stratify demographic and socioeconomic characteristics, such as race/ethnicity, gender, age, marital status, educational attainment, and income, to discern patterns of disease incidence or prevalence. Temporal patterns of disease occurrence provide clues concerning environmental or infectious disease etiologies. The spatial component of disease occurrence is the oft‐forgotten member of the proverbial descriptive epidemiologic trinity of ‘‘person, place, and time’’ characteristics (Lilienfeld & Stolley, 1994). The purpose of this chapter is to explore the use of geographic information systems (GIS) and spatial statistical methods in addressing unanswered questions in the field of mental retardation and developmental disabilities research. The chapter begins with a brief overview of the field of medical geography, examining some of the emerging themes and methods and describing their potential relevance for research on the etiology, health care, and service delivery for mental retardation and developmental disabilities. Next, we explore the utility of GIS and spatial analytical techniques for this work, primarily utilizing examples drawn from studies of other diseases due to the paucity of specific applications in this area. The chapter concludes with recommendations for the development of a research agenda for spatial analysis of factors associated with mental retardation and related developmental disabilities, and their treatment and care. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

79

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33004-2

80

Russell S. Kirby II.

MEDICAL GEOGRAPHY

Medical geography applies the theory and methods of the scientific discipline of geography to the study of human health, disease, and health care (Meade & Earickson, 2000). As the field has evolved since the mid‐twentieth century, two major traditions have emerged (Mayer, 1982, 1984). One branch of this emerging subdiscipline has concerned itself with geographical epidemiology, while the other has focused on health systems planning. The terms ‘‘geographical epidemiology,’’ ‘‘spatial epidemiology,’’ and ‘‘disease ecology’’ are used somewhat interchangeably, although the terms mean slightly diVerent things. ‘‘Spatial epidemiology’’ generally refers to the application of spatial statistical methods to the analyses of a set of cases with a specific health condition or disease. ‘‘Disease ecology’’ has traditionally focused on the study of patterns of infectious disease in relation to interaction between humans and the environment. ‘‘Geographical epidemiology’’ refers more broadly to the study of spatial aspects of health and disease. Many practitioners of medical geography focus either on geographical epidemiology or health services planning and research, leaving opportunities for crosscutting innovations linking the spatial epidemiology of disease with patterns of treatment utilization and systems of health care. With the advent of GIS, rapid methods development has occurred in the arena of spatial statistics. The past decade has seen the publication of a number of useful texts and monographs (Bailey & Gatrell, 1995; Elliott, Wakefield, Best, & Briggs, 2000; Lawson, 2001; Waller & Gotway, 2004). As GIS software becomes more aVordable and comes into wider use, applications are evolving in virtually all areas of epidemiology and health services research. Maheswaran and Craglia (2004) provide an overview of practical applications of GIS in public health.

A.

Disease Mapping

On the surface, disease mapping may appear to be a simple proposition. However, a number of decisions must be made in order to take data about the locations of persons with a disease or health condition and transform it into a map. Several resources provide overviews of the general issues and available methodologies (Jerrett et al., 2003; Lawson & Williams, 2001; Moore & Carpenter, 1999). The researcher must first decide what to map: (1) spatial distribution of incidence (new cases occurring/diagnosed in the population at risk during a specific time period), (2) spatial distribution of prevalence (all cases existing in a defined population at a specific point or period in time), or (3) spatial

INCORPORATING GEOGRAPHICAL ANALYSIS

81

distribution of health care providers involved in care of persons with this disease and/or health services utilization among persons with this disease. Researchers must then choose to create maps using choropleth or isopleth mapping techniques. Most health data are mapped as choropleth maps, using the following general approach. Each record is grouped into one of many small areas (e.g., ZIP Codes, census tracts, minor civil divisions, counties, or states) covering the region under study. Each of these small areas is assigned a color, shading, or other classification based on its value on the variable being mapped. Isopleth maps, on the other hand, display interpolated lines of equal value for a given phenomenon across the region of interest. An isopleth map can be created by determining the geographical or geometric center of each small area (e.g., ZIP Code) and assigning the value for that area to that specific point, then fitting lines of equal value across the resultant spatial distribution of point locations. Common nonhealth examples of isopleth maps include topographic maps displaying elevation above sea level and maps showing the spatial distribution of daily high temperatures across the United States. To give an example for mental retardation research, we might be interested in mapping the population prevalence of mental retardation by ZIP Code area for a state or metropolitan area. A choropleth map might classify the ZIP Codes into categories with population prevalence greater than 5%, from 3% to 5%, from 2% to 3%, from 1% to 2%, and from 0% to 1%. The comparable isopleth map might display lines of equal prevalence for 5%, 4%, 3%, 2%, and 1%. Regardless of mapping method, decisions must be made concerning which spatial smoothers to use (Bernardinelli & Montomoli, 1992; Clayton & Kaldor, 1987; Pickle & Su, 2002); whether to map the data using ZIP Codes, census tracts or block groups, or other political geographical units (Krieger et al., 2002a,b); how to classify the data into meaningful categories; and how to display the results eVectively. The resulting map will display areas of higher and lower values, but does not demonstrate the results of a statistical test of the significance of these observations. Is it more likely that a local value represents a statistical elevation in the rate than that it occurred by chance? The researcher must also remember that many possible maps can be created from the same data, and no single map is the ‘‘best’’ map; it all depends on the purpose for which the map is created. One method long in use is to convert the data to a probability map (Choynowski, 1959; Rushton & Lolonis, 1996). The resultant map displays the probability that the value at any given location in the area studied exceeds the expected value based on empirically based assumptions concerning the underlying distribution. While this approach answers the question of whether there are areas with higher‐than‐expected rates, additional maps must also be prepared to display the actual data.

82

Russell S. Kirby

Another option is the class of methods for local indicators of spatial association (LISA), initially developed by Anselin (1995), and popularized in the exploratory spatial data analysis program GeoDa (Anselin, Syabri, & Kho, 2006). The LISA measure, as operationalized in GeoDa, calculates Moran’s I as a statistic that measures the likelihood that the value at each location on the map is statistically higher or lower than that expected at adjacent locations. In its original formulation, Moran’s I (first proposed in 1950) provided a general measure of the degree of departure from spatial randomness for the distribution of a variable of interest. LISA provides a methodology for identifying local deviations within the overall spatial distribution of a variable. LISA can be applied both to map point data or data measured over small areas. In considering patterns in the association between two variables across small areas, such as a rate of a disease or health condition and a risk factor, researchers should also control for the possibility that the values for adjacent areas are spatially correlated. This phenomenon is also known as ‘‘spatial autocorrelation.’’ One method for controlling for the eVect of spatial autocorrelation that has recently come into increasing use is geographically weighted regression for ecological data (Fotheringham, Brundson, & Charlton, 2002). B.

Identification of Clusters in Space and Time

Do cases of a disease or health condition occur jointly in space, time, or space and time? While there is limited research specific to developmental disabilities and mental retardation (Brewster, Kirby, & Canino, 1998), methods to test hypotheses related to disease clustering have evolved over at least the past half century, and have examined a wide array of diseases and health outcomes, including cancer, birth defects, and adverse pregnancy outcomes (Jacquez, Grimson, Waller, & Wartenberg, 1996a; Jacquez, Waller, Grimson, & Wartenberg, 1996b; Lawson & KulldorV, 1999). Analyses of the spatial distribution of disease or health outcomes include global and local tests. These analyses may also consider the role of environmental exposures across an entire region or study area. Specific locations of emission of environmental exposures or toxic substances are referred to as ‘‘point‐sources’’ of exposure. Statistical methods concerned with the distribution of outcomes around point‐sources are termed focused tests, or tests to identify clustering around a predetermined point of location of risk (Lawson & Waller, 1996). In a recent paper, Jacquez et al. (2005b) provided definitions and examples of global, focused, and local tests of spatial clustering. Global tests examine a spatial distribution to discern a spatial pattern (e.g., Moran’s I). These tests can determine if there is a spatial structure to the geographical distribution of cases within a study region, but not the

INCORPORATING GEOGRAPHICAL ANALYSIS

83

specific locations with potentially larger numbers of cases, or higher incidence or prevalence rates. Local tests can identify clustering and spatial aggregation within a study region. These methods can also identify outliers (high or low), clusters of high or low values, and areas with statistically higher or lower than expected numbers, cases, or rates. One of the more widely used classes of local tests is the LISA (Anselin, 1995; Anselin et al., 2006). Another method in wide use is KulldorV’s spatial scan statistic (SaTScan) (KulldorV, 1997). Scan statistics are a class of statistical tests that can detect clusters of events across time, space, or space‐time. The SaTScan procedure gradually scans a window of potentially varying size across time, space, or space and time, retaining the number of observed and expected observations at each location. The window yielding the maximum likelihood represents the most likely cluster, and a p‐value is assigned to this cluster. A recent example of the application of SaTScan in a multilevel modeling framework examined spatial clustering of mental disorders across neighborhoods in Malmo¨, Sweden (Chaix et al., 2006). Focused tests are designed to identify clustering in relation to specific point‐sources or locations of interest. These methods are especially useful in environmental epidemiology, where there is a specific point‐source of exposure and concerns that health risks are associated with proximity to that location (Lawson & Waller, 1996). As researchers generate environmental hypotheses concerning associations between releases of industrial by‐ products and aspects of human development processes, these tests may come into wider use. C.

Units for Spatial Analysis

Prior to mapping or spatial analysis, health data must be assigned to geographical units. The question is: Which spatial units should the researcher use to display the patterns in the data? In the United States, data can be aggregated by political or administrative units and measured across states, counties, cities, and other minor civil divisions; by census geography into census tracts, block groups or block; or by postal geography into ZIP Codes or ZIPþ4 subunits. In choosing among these options, data quality must guide decision making. If ZIP Codes are the only units reported for each case or study subject, data can be presented at this scale, but care must be taken to ensure that boundaries are synchronized for cases (numerator) and underlying population (denominator) in any analyses. For example, in a community where a large ZIP Code has recently been divided into two or three new ZIP Codes, one of which retains the original ZIP Code, for several years after this

84

Russell S. Kirby

change occurs it is unlikely that accurate counts of health events will be available for each of the new ZIP Codes. Until stable and reliable reporting is demonstrated, the researcher will be well advised to combine the three areas into a single areal unit. If complete addresses are available, the researcher can geocoded addresses to latitude–longitude coordinate pairs, and, on the basis of the geocoded results, reassign each record to political units, census tracts, and ZIP Codes (Krieger, Waterman, Lemieux, Zierler, & Hogan, 2001; Krieger et al., 2002b). Since most mail is delivered correctly even when addressed with incorrect ZIP Codes, this option should be considered. County of residence is also misreported frequently enough that reclassification of records based on geocoding should be considered if complete addresses are available on each record. As a basic rule of thumb, the finer the geographical scale, the more noise and random error in the resulting pattern. Not only will the range of values across the areal units be greater, but the standard errors associated with each measure increase as well. Gregario, DeChello, Samocuik, and KulldorV (2005) suggest that, in most cases, it will not be worth the eVort to prepare data for analyses to detect spatial clusters across areal units smaller than census tracts. Krieger et al. (2002a) also demonstrate that scale makes a diVerence in the results of spatial analyses of health data. D.

Spatial Statistics

Space does not permit an extensive overview of spatial statistics methods and their potential applications in medical geography and public health. Several useful texts and reviews provide this information and identify future directions for methodological development (Elliott & Wartenberg, 2004; Waller & Gotway, 2004). A key issue for researchers to consider involves the decision to analyze data as a point distribution (based on the specific latitude–longitude coordinates assigned to each case or study subject), or ecologically (aggregated into geo‐referenced areal units). A second issue is whether to focus exclusively on the spatial pattern of cases or to model covariates. If the latter choice is made, the researcher must also choose whether to utilize multilevel modeling methods or, if suYcient data are available, to analyze each covariate at the same scale as the study subjects. III.

MENTAL RETARDATION AND DEVELOPMENTAL DISABILITIES RESEARCH

Although examples exist of population‐based analyses of specific mental health conditions (Scully, Owens, Kinsella, & Waddington, 2004), relatively few studies have examined spatial patterns of mental retardation

INCORPORATING GEOGRAPHICAL ANALYSIS

85

or developmental disabilities. Several impediments may have hindered the application of spatial epidemiology and GIS to this subject matter. Specifically, there are comparatively few programs for population‐based surveillance or registration of these conditions (Rice, Schendel, CunniV, & Doernberg, 2004). While the Centers for Disease Control and Prevention (CDC) has maintained a program in the metropolitan Atlanta area since the 1980s (Bhasin, Brocksen, Avchen, & Van Naarden Braun, 2006; Yeargin‐ Allsopp, Murphy, Oakley, & Siles, 1992; Yeargin‐Allsopp et al., 2003), and CDC now funds a network of projects for surveillance of autism and related disabilities (Rice et al., in press), much more must be done. Even when these programs do exist, spatially specific locational data are not always available. Administrative data sources, such as special education, frequently do not make address information available to researchers. Addressing a broader question concerning research across the age‐ spectrum, Weich (2005) speculated on why previous research has generally not identified significant spatial variation in rates of mental health or mental disorders, even when evidence suggests this should occur. Are we studying the problem at the wrong scale, examining the wrong exposures or risk factors, measuring the wrong outcomes, applying the wrong operational models, asking the wrong questions? The answers remain unclear, particularly given that (at present) most population‐based research in mental health and mental retardation utilizes secondary data, employs cross‐sectional study designs measured across administrative units of geography, and utilizes compositional measures of space. As more analyses are conducted of spatial patterns of intellectual disability, care must be taken to examine the link between poverty and service dependency. Metzel (2005) convincingly demonstrated the interconnectivity between these two phenomena in a social, historical, and geographical study of the role of a voluntary service organization in metropolitan Baltimore, Maryland during the twentieth century. In the modern ‘‘post‐asylum’’ era of treatment for mental health issues, a wide array of factors may influence spatial patterns of prevalence. Researchers should also consider the utility of qualitative and mixed‐methods research as well as contextual analysis (Smith, 2005). IV.

RESEARCH AGENDA

Establishment and maintenance of population‐based surveillance programs and/or registries for developmental disabilities and mental retardation are an absolute necessity for research into spatial aspects of the incidence and prevalence of these conditions. Generally, the focus should be

86

Russell S. Kirby

on prevalence, as most developmental disorders have complex etiologies that may involve perinatal or genetic factors and have variable age at presentation or clinical diagnosis. Surveillance or registry data should include suYciently detailed address information or spatial data to permit mapping and spatial analysis. However, as noted by Jacquez (1998), while maps can be the end product of an analysis, at the current state of knowledge concerning the etiology of developmental disabilities in childhood GIS should be viewed more as a tool for hypothesis generation, especially in the arena of population genetics and environmental epidemiology. Researchers should always ensure that hypotheses are plausible rather than becoming enticed by easily generated statistical results showing ‘‘significant’’ correlations or associations. Methodologies for assessing spatial patterns of environmental exposures are rapidly evolving. Advances in computing capacity and environmental modeling permit ever more sophisticated estimates of exposure accounting for groundwater patterns, atmospheric conditions, wind speed and direction, and related variables. A major gap involves the limited number of monitoring stations or locations collecting data. As this situation improves, mathematical models to estimate exposure levels across an entire region of interest will provide ever more accurate localized estimates of exposure. Due to the behavioral component, the outcomes of many forms of environmental exposure will remain diYcult to quantify at the appropriate scale. As one example, if a researcher posits that risk of autism is associated with in utero exposure to mercury or other heavy metals, without direct biological measures taken during the pregnancy or detailed food frequency questionnaire data, it will be diYcult to quantify the level of exposure across individual study subjects. A recent ecological study highlights these issues. Using administrative data of environmental mercury exposure and autism measured at the county level, Palmer, Blanchard, Stein, Mandell, and Miller (2006) noted the need both for more direct measures of exposure and for population‐ based assessment of the prevalence of developmental disabilities. Assessment of residential exposures remains complicated, as researchers frequently do not obtain full residential histories on their study subjects (Jacquez, Goovaerts, & Rogerson, 2005a). Limited data exist, but anecdotally it appears that many families migrate to states or metropolitan areas suggested by others as better communities for services for their children with autism and other developmental disabilities. Not only should this migration be documented, but its impact on hypotheses concerning the etiology of these disorders must also be assessed. In child health research, GIS applications have focused primarily on its utility as a tool to map and display results. GIS could also enhance study

INCORPORATING GEOGRAPHICAL ANALYSIS

87

designs and study methods in environmental epidemiology and health services research, and facilitate the translation of study results to the community (Miranda & Dolinoy, 2005; Miranda, Dolinoy, & Overstreet, 2002). GIS applications that permit the public to access, query, and investigate neighborhood‐level data empower communities to identify and own their problems and concerns, and to generate community‐based solutions. Accessibility of health services for families with children aVected by mental retardation and other developmental disabilities represents another research arena with a great deal of promise. Although basic methods for analysis of geographical accessibility are well developed (Pearce, Witten, & Bartie, 2006; Rushton, 1999), these methods have only been applied in limited ways for the study of children with developmental disabilities (Fulcher & Kaukinen, 2005; Talen & Anselin, 1998). Longley (2005) has proposed that GIS be employed more eVectively in health and human services delivery, by better utilization and presentation of geodemographic data. GIS also holds considerable potential for improving the eVectiveness of population‐based interventions through better understanding of community networking, targeting of programs and services (Caley, 2004). However, GIS applications are much better at generating maps of spatial patterns (e.g., areas of unmet need, higher densities of clients) than at modeling accessibility and spatial interaction in the specific context of treatment and service delivery (Parker & Campbell, 1998). As GIS science evolves, better methods for incorporating the time dimension into spatial modeling will become available, including ‘‘temporal GIS’’ or ‘‘space‐time intelligent systems’’ (Jacquez et al., 2005a). Temporal GIS applications expand on traditional GIS through its ability to collect, process, manage, and analyze data measured both across space and time. As the use of GIS and spatial analysis in mental retardation research becomes more widespread, some basic caveats must remain foremost in our minds. Although progress has been made, greater linkages between cartographic theory, spatial statistical methods, and epidemiology are needed, and multidisciplinary collaborations remain a necessity (Kirby, 1996). We must always remember that the ‘‘points’’ we locate on our maps represent actual children and families, whose privacy we are charged to protect. Although some methods have been proposed to disassociate spatial locations from specific observations (Armstrong, Rushton, & Zimmerman, 1999), these methods are neither foolproof nor necessarily scientifically neutral (Kwan, Casas, & Schmitz, 2004). However, we must not allow these issues to promote inertial tendencies that prevent any research whatsoever from being conducted.

88

Russell S. Kirby V.

CONCLUSIONS

Spatial analysis and GIS hold considerable promise in mental retardation/ developmental disabilities research. Ultimately, when more eYcient surveillance systems are implemented, these tools may assist in untangling the etiology of mental retardation and developmental disabilities. For now, however, GIS will be more useful for hypothesis generation and for developing more eYcient patterns of service delivery and treatment, at least in health care systems designed to serve the entire population. REFERENCES Anselin, L. (1995). Local indicators of spatial association—LISA. Geographical Analysis, 27, 93–115. Anselin, L., Syabri, I., & Kho, Y. (2006). GeoDa: An introduction to spatial data analysis. Geographical Analysis, 38, 5–22. Armstrong, M. P., Rushton, G., & Zimmerman, D. L. (1999). Geographically masking health data to preserve confidentiality. Statistics in Medicine, 18, 497–525. Bailey, T. C., & Gatrell, A. C. (1995). Interactive spatial data analysis. Essex, England: Longman Scientific and Technical. Bernardinelli, L., & Montomoli, C. (1992). Empirical Bayes versus fully Bayesian analysis of geographical variation in disease risk. Statistics in Medicine, 11, 983–1007. Bhasin, T. K., Brocksen, S., Avchen, R. N., & Van Naarden Braun, K. (2006). Prevalence of four developmental disabilities among children aged 8 years—Metropolitan Atlanta Developmental Disabilities Surveillance Program, 1996 and 2000. MMWR Surveillance Summary, 55(1), 1–9. [Erratum in: MMWR Morbidity and Mortality Weekly Report, 55, 105–106]. Brewster, M. A., Kirby, R. S., & Canino, C. (1998). Final report: Adverse reproductive outcomes in Pulaski County for years 1980 through 1990. Arkansas Reproductive Health Monitoring System in collaboration with Arkansas Department of Health. Atlanta, GA: Agency for Toxic Substances and Disease Registry. Caley, L. M. (2004). Using geographic information systems to design population‐based interventions. Public Health Nursing, 21, 547–554. Chaix, B., Leyland, A. H., Sabel, C. E., Chauvin, P., Ra˚stam, L., Kristersson, H., et al. (2006). Spatial clustering of mental disorders and associated characteristics of the neighbourhood context in Malmo¨, Sweden, in 2001. Journal of Epidemiology and Community Health, 60, 427–435. Choynowski, M. (1959). Maps based on probabilities. Journal of the American Statistical Association, 54, 385–388. Clayton, D. G., & Kaldor, J. M. (1987). Empirical Bayes estimates of age‐standardized relative risks for use in disease mapping. Biometrics, 43, 671–681. Elliott, P., & Wartenberg, D. (2004). Spatial epidemiology: Current approaches and future challenges. Environmental Health Perspectives, 112, 998–1006. Elliott, P., Wakefield, J. C., Best, N. G., & Briggs, D. J. (2000). Spatial epidemiology: Methods and applications. New York: Oxford University Press.

INCORPORATING GEOGRAPHICAL ANALYSIS

89

Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically weighted regression: The analysis of spatially varying relationships. West Sussex, England: John Wiley and Sons. Fulcher, C., & Kaukinen, L. (2005). Mapping and visualizing the location of HIV service providers: An exploratory spatial analysis of Toronto neighborhoods. AIDS Care, 17, 386–396. Gregario, D. I., DeChello, L. M., Samocuik, H., & KulldorV, M. (2005). Lumping or splitting: Seeking the preferred areal unit for health geography studies. International Journal of Health Geographics, 4, 6. Jacquez, G. M. (1998). GIS as an enabling technology. In A. Gatrell & M. Lo¨yto¨nen (Eds.), GIS and health (pp. 17–28). Philadelphia: Taylor and Francis. Jacquez, G. M., Grimson, R., Waller, L., & Wartenberg, D. (1996a). The analysis of disease clusters. Part 2: Introduction to techniques. Infection Control and Hospital Epidemiology, 17, 385–397. Jacquez, G. M., Goovaerts, P., & Rogerson, P. (2005a). Space‐time intelligent systems: Technology, applications, and methods. Journal of Geographical Systsem, 7, 1–5. Jacquez, G. M., Kaufmann, A., Meliker, J., Goovaerts, P., AvRuskin, G., & Nriagu, J. (2005b). Global, local and focused geographic clustering for case‐control data with residential histories. Environmental Health: A Global Access Science Source, 4, 4. Jacquez, G. M., Waller, L., Grimson, R., & Wartenberg, D. (1996b). The analysis of disease clusters. Part 1: State of the art. Infection Control and Hospital Epidemiology, 17, 319–327. Jerrett, M., Burnett, R. T., Goldberg, M. S., Sears, M., Krewski, D., Catalan, R., et al. (2003). Spatial analysis for environmental health research: Concepts, methods, and examples. Journal of Toxicology and Environmental Health, 66, 1783–1810. Kirby, R. S. (1996). Toward congruence between theory and practice in small area analysis and local public health data. Statistics in Medicine, 15, 1859–1866. Krieger, N., Waterman, P., Lemieux, K., Zierler, S., & Hogan, J. W. (2001). On the wrong side of the tracts? Evaluating the accuracy of geocoding in public health research. American Journal of Public Health, 91, 1114–1116. Krieger, N., Chen, J. T., Waterman, P. D., Soobader, M. J., Subramanian, S. V., & Carson, R. (2002a). Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: Does the choice of area‐based measure and geographic level matter? The public health disparities geocoding project. American Journal of Epidemiology, 156, 471–482. Krieger, N., Waterman, P., Chen, J. T., Soobader, M. J., Subramanian, S. V., & Carson, R. (2002b). Zip code caveat: Bias due to spatiotemporal mismatches between zip codes and US census‐defined geographic areas—the Public health disparities geocoding project. American Journal of Public Health, 92, 1100–1102. KulldorV, M. (1997). A spatial scan statistic. Communications in Statistics Theory and Methods, 26, 1481–1496. Kwan, M. P., Casas, I., & Schmitz, B. C. (2004). Protection of geoprivacy and accuracy of spatial information: How eVective are geographical masks? Cartographica, 39, 15–28. Lawson, A. B. (2001). Statistical methods in spatial epidemiology. New York: Wiley. Lawson, A. B., & KulldorV, M. (1999). A review of cluster detection methods. In A. Lawson, D. Bohning, E. LesaVre, J. F. Viel, & R. Bertollini (Eds.), Advanced methods of disease mapping and risk assessment for public health decision making (pp. 99–110). London: Wiley. Lawson, A. B., & Waller, L. A. (1996). A review of point pattern methods for spatial modelling of events around sources of pollution. Environmetrics, 7, 471–487. Lawson, A. B., & Williams, F. L. R. (2001). An introductory guide to disease mapping. New York: Wiley. Lilienfeld, D. E., & Stolley, P. D. (1994). Foundations of epidemiology (3rd ed.). New York: Oxford University Press.

90

Russell S. Kirby

Longley, P. (2005). Geographical information systems: A renaissance of geodemographics for public service delivery. Progress in Human Geography, 29, 57–63. Maheswaran, R., & Craglia, M. (2004). GIS in public health practice. Boca Raton, FL: CRC Press. Mayer, J. D. (1982). Relations between two traditions of medical geography: Health systems planning and geographical epidemiology. Progress in Human Geography, 6, 216–230. Mayer, J. D. (1984). Medical geography: An emerging discipline. Journal of the American Medical Association, 251, 2680–2683. Meade, M., & Earickson, R. (2000). Medical geography (2nd ed.). New York: Guilford Press. Metzel, D. S. (2005). Places of social poverty and service dependency of people with intellectual disabilities: A case study in Baltimore, Maryland. Health & Place, 11, 93–105. Miranda, M. L., & Dolinoy, D. (2005). Using GIS‐based approaches to support research on neurotoxicants and other children’s environmental health threats. NeuroToxicology, 26, 223–228. Miranda, M. L., Dolinoy, D., & Overstreet, M. A. (2002). Mapping for prevention: GIS models for directed childhood lead poisoning prevention programs. Environmental Health Perspectives, 110, 945–953. Moore, D. A., & Carpenter, T. E. (1999). Spatial analytical methods and geographic information systems: Use in health research and epidemiology. Epidemiologic Reviews, 21, 143–161. Palmer, R. F., Blanchard, S., Stein, Z., Mandell, D., & Miller, C. (2006). Environmental mercury release, special education rates, and autism disorder: An ecological study of Texas. Health & Place, 12, 203–209. Parker, E. B., & Campbell, J. L. (1998). Measuring access to primary medical care: Some examples of the use of geographic information systems. Health & Place, 4, 183–193. Pearce, J., Witten, K., & Bartie, P. (2006). Neighbourhoods and health: A GIS approach to measuring community resource accessibility. Journal of Epidemiology and Community Health, 60, 389–395. Pickle, L. W., & Su, Y. (2002). Within‐state geographic patterns of health insurance coverage and health risk factors in the United States. American Journal of Preventive Medicine, 22, 75–83. Rice, C., Baio, J., Van Naarden Braun, K., Doernberg, F., Meaney, F. J., & Kirby, R. S., for the ADDM Network (in press). A public health collaboration for the surveillance of autism spectrum disorders (ASD). Paediatric and Perinatal Epidemiology. Rice, C., Schendel, D., CunniV, C., & Doernberg, N. (2004). Public health monitoring of developmental disabilities with a focus on the autism spectrum disorders. American Journal of Medical Genetics, 125C, 22–27. Rushton, G. (1999). Methods to evaluate geographic access to health services. Journal of Public Health Management and Practice, 5, 93–100. Rushton, G., & Lolonis, P. (1996). Exploratory spatial analysis of birth defect rates in an urban population. Statistics in Medicine, 15, 717–726. Scully, P. J., Owens, J. M., Kinsella, A., & Waddington, J. L. (2004). Schizophrenia, shizoaVective and bipolar disorder within an epidemiologically complete, homogeneous population in rural Ireland: Small area variation in rate. Schizophrenia Research, 67, 143–155. Smith, P. (2005). OV the map: A critical geography of intellectual disabilities. Health & Place, 11, 87–92. Talen, E., & Anselin, L. (1998). Assessing spatial equity: An evaluation of measures of accessibility to public playgrounds. Environment and Planning A, 30, 595–613.

INCORPORATING GEOGRAPHICAL ANALYSIS

91

Waller, L. A., & Gotway, C. A. (2004). Applied spatial statistics for public health data. New York: John Wiley and Sons. Weich, S. (2005). Absence of spatial variation in rates of common mental disorders. Journal of Epidemiology and Community Health, 59, 254–257. Yeargin‐Allsopp, M., Murphy, C. C., Oakley, G. P., & Siles, K., Metropolitan Atlanta Developmental Disabilities Study StaV (1992). A multiple‐source method for studying the prevalence of developmental disabilities in children: The Metropolitan Atlanta Developmental Disabilities Study. Pediatrics, 89, 624–630. Yeargin‐Allsopp, M., Rice, C., Karapurkar, T., Doernberg, N., Boyle, C., & Murphy, C. (2003). Prevalence of autism in a US metropolitan area. Journal of the American Medical Association, 289, 49–55.

This page intentionally left blank

Statistical Issues in Developmental Epidemiology and Developmental Disabilities Research: Confounding Variables, Small Sample Size, and Numerous Outcome Variables JENNIFER URBANO BLACKFORD VANDERBILT KENNEDY CENTER, DEPARTMENT OF PSYCHIATRY VANDERBILT UNIVERSITY, NASHVILLE, TENNESSEE

I.

STATISTICAL CHALLENGES IN DEVELOPMENTAL DISABILITIES RESEARCH

During the past decade, researchers have provided great contributions to our understanding of developmental disabilities. Scientists have made significant advances in the areas of phenotyping, intervention programs, and family research. These advances have resulted in increased knowledge about developmental disabilities and have provided new research directions. We still have much to learn. However, there are some constraints in the discovery of new knowledge that are inherent in studies that examine rarely occurring groups, compare groups, or include confounding variables. Fortunately, statisticians have made great strides in thinking about many of these issues. Areas such as clinical trials, epidemiology, genomics, and imaging now routinely deal with problems of matching, confounding variables, small sample sizes, and numerous outcome variables. Computer advances allow some of these new methods to be available in standard, readily available, software packages. These new methods are not terribly hard or complicated. Typical, practicing social scientists with some background in basic statistics (like the t‐test) can perform these new techniques.

INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

93

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33005-4

94

Jennifer Urbano Blackford

This chapter introduces ‘‘state‐of‐the‐art’’ statistical techniques for developmental disability researchers. As an introduction, these new methods cannot be covered in exhaustive detail. However, this chapter will try to provide a conceptual understanding, appropriate context, and practical example for each method. Specifically, this chapter will cover three interrelated issues: confounding variables, small sample size, and numerous outcome variables. Three new statistical methods will be presented. Propensity scores will provide a new method for controlling for confounding variables. Permutation tests will illustrate a new method for calculating ‘‘exact’’ p‐values, even when your sample size is very small. Finally, multivariate permutation tests (MPTs) will be introduced as a method for simultaneously analyzing numerous outcome variables while controlling for Type I error. These methods are all new, statistically powerful techniques that can be used by the typical developmental disability researcher. Use of these new methods can bypass some of the typical limitations inherent in developmental disability research, providing an opportunity to advance scientific knowledge in this field. II. A.

CONFOUNDING VARIABLES: HOW TO MATCH ON MORE THAN TWO VARIABLES

The Problem with Confounding Variables

Developmental disability research often asks questions about how a group with a certain disability diVers from another group, often referred to as a control or contrast group. The control group may be a group of people without a disability or a group with a diVerent disability. A confounding variable is a variable that is statistically related to both the predictor variable and the outcome variable. Prior to hypothesis testing, it is critical to ensure that the only diVerence between the two groups is the one of interest, for example, disability status or type of disability. When the two groups are also diVerent on other measures, these group diVerences may ‘‘confound’’ the results. That is, the ‘‘confounding’’ variable may be responsible for the observed relationship between the predictor and outcome variable. Consider a study designed to examine group diVerences in communication skills between children with and without autism. The two groups are selected to diVer on developmental disability status, but they also diVer on developmental age. Both developmental age and having autism may be related to communication skills. If there is a significant group diVerence on communication, it could be the result of autism or it could be caused by the confounding variable, developmental age. Confounding variables often result in a misinterpretation of a study’s findings.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

95

Confounding variables present a challenge for both traditional social science research and population‐based research methods. The relatively new field of developmental epidemiology seeks to answer questions about developmental disabilities at a population level. For relatively low‐frequency events, such as developmental disability, epidemiological methods, and large observational datasets, provide a unique opportunity to explore questions that are unanswerable with smaller datasets. Confounding variables are especially problematic when analyzing large observational datasets using epidemiological methods. First, the data are typically collected retrospectively, making it diYcult to control for potentially confounding variables in advance. Second, the large datasets are usually observational (as opposed to experimental or quasi‐ experimental), which increases the chance that an observed relationship between two variables is influenced by a third variable. The presence of confounding variables is a critical issue for both small and large studies. Let us consider a practical example. A researcher is interested in examining the impact of having a child with Down syndrome on a sibling’s sense of well‐being. Maternal age is a well‐known risk factor for having a child with Down syndrome (Erickson, 1978; Hook & Lindsjo, 1978; Penrose, 1933). Older mothers are more likely to have other children, leading to an increased likelihood of later birth order for children with Down syndrome. Gender and birth order, of both child and sibling, may influence the sibling’s sense of well‐being. Thus, each of these variables might be a confounding factor. I will explore traditional and new methods for dealing with confounding variables below using this example. B.

Traditional Methods: Random Assignment, Matching, and Covariate Analysis

Traditionally, three tools are used for controlling confounding variables in research studies: random assignment, matching, and covariate analysis. Random assignment refers to the random assignment of people to groups. Random assignment is the gold standard for ensuring that two groups are similar, regardless of whether all possible group diVerences are measured (Fisher, 1966). With random assignment, one can assume that any diVerences between people will be randomly distributed across the groups. Thus, random assignment is a very good method for balancing groups on the observed and unobserved individual characteristics that are not related to the outcome variable. However, random assignment is not always feasible. A researcher cannot randomly assign a child to have a developmental disability. Similarly, one cannot assign a sibling to be raised with a child with a developmental disability. Other methods are needed since random assignment is not applicable.

96

Jennifer Urbano Blackford

A second method for controlling confounding variables is matching. Matching is the selection of participants into the control or contrast groups that are similar on specific individual characteristics to the groups with developmental disabilities. This strategy balances groups a priori on potentially confounding variables. Because it is diYcult to match participants on more than a few variables, matching is useful only for a small number of variables. For example, if you wanted to match on five variables and each of the five variables had only two categories (e.g., male/female), then you would have 32 (25) diVerent categories to match on. In the example, the researcher might want to match on child gender (male/female), sibling gender (male/female), and child birth order (first/not first). Even when the categories for each variable are limited to two, this combination of 2 2 2 results in eight diVerent categories for matching. Finding the right combinations for each of the eight cells, especially for the children with Down syndrome, would be a daunting task. The issue of matching for large datasets is slightly diVerent. Instead of matching to create the best groups a priori, the researcher usually tries to construct the best control group post hoc, from the dataset of possible participants. Matching on more than two variables is an issue for large datasets as well as small datasets. Even with large numbers of participants for matching, finding participants for each of the multiple categories is often impossible. A third method for controlling confounding variables is covariate analysis. Covariate analysis is an analytic method that statistically controls for the eVects of confounding variables. Covariate analysis can be particularly useful because the analysis is done post hoc and does not require all potentially confounding variables (covariates) to be identified in advance during participant recruitment. Covariate analyses can be performed whether matching was used to construct the groups. In the example mentioned earlier, child gender, sibling gender, and child birth order could all be included as covariates in the statistical analysis. Ideally, one would include all potentially confounding variables as covariates. However, practical constraints usually limit the possible number of covariates. First, there should be at least 10 participants for each variable included in the analysis. An analysis with 10 covariates would require a minimum of 100 participants. Next, the analysis of covariance (ANCOVA) has two specific assumptions that are often not met. One assumption is that the covariate is related to the outcome variable. The other assumption is that the relationship between the covariate and outcome variable is the same for each group, called ‘‘homogeneity of regression.’’ Homogeneity of regression is often not met, especially when there are substantial diVerences between the groups.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

97

In summary, if confounding variables are not accounted for, it is likely that statistical results will be misinterpreted. The traditional approaches have limitations and become increasingly diYcult when there are more than two confounding variables. Given these constraints, the practical approach is usually to pick the two or three most important variables and use one of the available methods. This approach often makes it impossible to include all confounding variables in the model, thereby producing results with limited validity. C.

New Method: Propensity Scores

The use of a relatively new method, propensity scoring, can provide statistical control for multiple potentially confounding variables (Rosenbaum & Rubin, 1983). Although propensity scores have begun to receive attention as an important method in biomedical studies, clinical trials, program evaluation, and outcomes research, propensity scores have yet to be used in developmental disability research. Propensity scores are a method for creating a single score, or variable, from a group of confounding variables. Because they can represent a large number of variables with a single score, propensity scores have wide application. The propensity scores can then be used in place of multiple confounding variables in traditional approaches to confounding variables such as matching and covariate analyses. The use of a single variable is advantageous because it reduces the complexity of the analyses and the need for large numbers of study participants. A propensity score is the conditional probability that given that person’s observed scores on all of the confounding variables, a person will be in a particular group. Conditional probability is the likelihood of one event given the occurrence of another event. Returning to the example used earlier, a sibling’s propensity score is the likelihood that the sibling is in group A (child with Down syndrome) instead of group B (child without Down syndrome) given the sibling’s gender, child’s gender, and child’s birth order. A propensity score is a single number that is used to represent an individual’s score with respect to a group of covariates and the individual’s group membership (e.g., child with or without Down syndrome). One of the strengths of a propensity score approach is that the score is calculated without using any information about the outcome measure. This independence between the propensity score and the outcome measure makes it a good proxy for random assignment to groups (Yanovitzky, Zanutto, & Hornik, 2005). Propensity score methods can be applied to any two groups at one time. Analyses with more than two groups need a separate propensity score to

98

Jennifer Urbano Blackford

represent each group comparison. For example, an analysis with three groups will have three separate propensity scores: group 1 versus group 2, group 1 versus group 3, and group 2 versus group 3. Each of the three propensity scores would be used together in subsequent analysis, compared to using only a single propensity score in analyses between two groups. Having more than two groups for comparison does not mean that researchers need to perform multiple separate analyses. However, they will have extra work in computing several propensity scores. The mechanics of computing a propensity score are relatively straightforward. Standard statistical tests, such as descriptive statistics, t‐test, logistic regression, and ANCOVA, are the only tools needed. To calculate a propensity score, one first selects all potential confounding variables that are important for the specific population and research question. This variable selection should be based on prior research findings or theory and should not be driven by empirical relationships found by preliminary statistical analyses of the data. Only variables that are expected to be related to both group assignment (e.g., sibling with or without Down syndrome) and the outcome variable, but not caused by either, should be included. Gender and birth order are examples of appropriate variables. Next, determine whether you have enough subjects to compute propensity scores. Propensity score computation uses logistic regression, a standard statistical method, so you can use the rule‐of‐thumb for that analysis. I recommend having 5–10 participants in the group of interest (e.g., sibling with Down syndrome) for each confounding variable. Propensity scores can be calculated with fewer observations, but scores will be less reliable than those calculated from larger samples. Then, determine how diVerent the two groups are on this initial set of variables, using traditional statistical methods to test for group diVerences, such as a t‐test or chi‐square. If the groups are equal on each of the variables, there is no need to use the propensity score or any other method of control. If there is a group diVerence on any of the variables, you must control for the diVerences. The degree of diVerence (e.g., t‐statistic value) between the groups can later serve as a benchmark to determine the eVectiveness of the propensity score. To compute the propensity score, perform a logistic regression analysis using a standard software package. A logistic regression is a statistical analysis that uses one (or more) predictor variables to predict a categorical outcome variable. In the logistic regression, the outcome variable is a binary variable representing group assignment (e.g., sibling with or without Down syndrome). The predictor variables in the analysis are the confounding variables for which one wants control. A logistic regression analysis uses an empirically derived formula (i.e., weighting of predictor variables) to best

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

99

discriminate between the two groups of the outcome variable. This formula can be applied to each individual’s values on all of the predictor variables to get an individual ‘‘predicted score.’’ The predicted score represents the probability that the individual would be in the group of interest given the predictor variables scores. For example, the predictor variables of child gender, sibling gender, and child birth order predict the outcome variable of sibling disability group. A predicted score of .60 means that, given a specific child’s gender, the sibling’s gender, and child’s birth order, there is a 60% chance that the sibling has Down syndrome. These individual predicted scores, and not the overall logistic regression results, are used for the propensity score calculation. It is easy to compute these individual predicted scores using standard logistic regression analyses in basic statistical software packages. It does not matter whether the overall regression model was significant because only the individual scores are needed. The individual predicted scores are the propensity scores and represent the probability of being in the group of interest given. Once you have calculated the propensity scores, they can be used either in propensity score matching or as a single covariate in covariate analyses. It is important to inspect your propensity score data prior to analysis, just as you would for any statistical analysis. The visual inspection of propensity score data should focus on two areas: variability and overlap. First, use the distribution of the propensity scores to evaluate variability. The distribution refers to the number of times each value in the sample occurred. These numbers can be plotted with the values from the smallest to the largest along the horizontal axis of the plot. The plotting can be done by hand or by using the descriptive or plot function in a statistical package. The variability of the distribution can be evaluated by visual inspection. The propensity scores for each group should show some variability or ‘‘spread.’’ A distribution that looks like a bell‐shaped curve has good variability. A distribution in which all scores have the same value does not have good variability. Kurtosis of the distribution, a measure of peakedness, can be used to quantify variability. In most statistical packages, a kurtosis value of 0 indicates a normal distribution. Kurtosis values greater than 5 suggest a peaked distribution, one that lacks variability. If either group does not have suYcient variability in the propensity score, as demonstrated by the distribution or kurtosis, the covariate analysis may not adequately control for the confounding variables. Matching can still be used but it may be diYcult to select appropriate matches. Next, inspect the propensity scores from the two groups for overlap. The distributions for the groups should have some overlap; that is, there should be some of the same propensity score values in each group. Overlap can be

100

Jennifer Urbano Blackford

determined by visual inspection of side‐by‐side box and whisker plots. Box plots can be produced using the descriptive statistic function in a statistical software package. Specify that the plots should be produced ‘‘by group’’ to get side‐by‐side plots. A box plot has a vertical rectangular area that represents the data values between the lower and upper quartiles (25th and 75th percentiles). The whiskers, or lines extending from the rectangles, represent the extreme lower and upper values. The rectangular area from one box plot should overlap with the rectangle for the second box plot. Specifically, if the lower and upper quartiles for the first group are .10 and .50, respectively, overlap would be demonstrated if either the lower or upper quartiles of the second distribution are between .10 and .50. If the distributions do not overlap, further analyses, such as matching on propensity scores, may be problematic. If there is no overlap, you will not be able to use the propensity scores for matching. However, you may still be able to use the propensity scores in an ANCOVA. The calculation of propensity scores can be illustrated with the previous example. Calculate the propensity scores using logistic regression. The outcome variable in the logistic regression would be whether the child has Down syndrome (yes/no). The predictor variables would be child gender, sibling gender, and child birth order. The predicted scores, which constitute the output from the statistical package, will be the individual propensity scores. Remember, these scores represent the probability that the child has Down syndrome given the child gender, sibling gender, and child birth order. Now, the three confounding variables are represented by a single variable. Next, examine the propensity score distributions for each group. Both distributions look like a bell‐shaped curve, which is good and suggests variability. The kurtosis values are 0.98 and 2.6, respectively. Both of these values are less than 5, which confirms suYcient variability. Next, box plots by group are produced using statistical software. The vertical rectangle (i.e., box plot) from one group does overlap with the rectangle from the second group. Looking at the quartiles, the lower and upper quartiles for the first group are .15 and .67, respectively. The upper and lower quartiles for the second group are .58 and .75, respectively. Thus, there is overlap between the two groups occurring between .58 and .67. Now that the propensity scores have been inspected for variability and overlap, the single propensity score is ready to be entered into an ANCOVA as a single covariate. Once propensity scores have been calculated and inspected for variability and overlap, they can be used in several ways. The propensity scores can be used for matching. For example, one could match the two groups, siblings of children with Down syndrome versus without Down syndrome on their propensity scores. Alternatively, one could enter the propensity score as a single covariate into an ANCOVA.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

101

Matching is one approach that can utilize propensity scores. Using traditional methods, matching had to be done on all of the potentially confounding variables simultaneously in order to provide the necessary control. Matching on more than two variables can be tricky, if not impossible. With propensity scores, participants from one group can be matched to participants in the second group on just the single propensity score. Matching on propensity scores can be used in any situation in which almost all of the participants in the group of interest can be matched on propensity score to a participant in the control group. Matching on propensity scores is most ideal when you have a large dataset (N > 100), a small group of participants with a developmental disability (<25% of whole sample), and a large pool of potential participants (>75%) to be selected for a control group. In this situation, one would calculate propensity scores for each individual and then use a matching strategy to select the best control group from the large available group of potential control participants. Matching on propensity scores is not useful for matching participants as they enter a research study because you must have data on all of the potential participants before you can calculate the propensity scores; this method of matching can only be used retrospectively. To use propensity scores for matching participants from one group to participants from a second group, you must have a systematic methodology for matching. It will be rare to have exact matches so the matching methodology provides a guide for determining what constitutes ‘‘a match.’’ There are diVerent matching methods and research is being conducted to determine which matching methods are superior. A very good and computationally simple matching approach is to randomly order the participants in the two groups, for example, children with and without Down syndrome. Match the first child with Down syndrome to a child without Down syndrome having the nearest propensity score. Remove both children from their respective group and repeat until all of the children with Down syndrome have a match in the group of children without Down syndrome. Propensity scores can also be used as a controlling variable, or covariate, in an ANCOVA. With this approach, the propensity score would be included as a single covariate in the statistical analyses. All of the potentially confounding variables are represented by a single score. This method is especially useful in two situations. First, covariate analysis with propensity scores is ideal when one has already collected data on research participants. The propensity score method is used to generate a single variable that can be used as a covariate in later analysis. Second, covariate analysis is ideal if there is only a small overlap between the propensity score distributions for the two groups. When the distributions do not overlap by at least 50%, matching can be diYcult because there are not many participants with

102

Jennifer Urbano Blackford

similar propensity scores across the two groups. Using the propensity score as a covariate does not require overlapping distributions. D.

Applications

Propensity scores provide a method for using a single value, a propensity score, to represent an entire group of potentially confounding variables. It is no longer necessary only to address one or two confounding variables due to constraints with the traditional approaches of matching and covariate analyses. The propensity score provides an advantageous method for minimizing the bias that comes from nonrandom assignment without requiring a larger sample size or reducing statistical power. For a typical research study in which participants for the two (or more) groups are recruited simultaneously, I recommend a hybrid approach. First, decide which one or two variables are most important to remain similar for the groups. As the participants are being selected, attempt as much as possible to match participants on these two variables. Then, when all of the data have been collected, compute propensity scores for the two groups and use these propensity scores as covariates in the statistical analyses. For large scale research projects in which data are being examined retrospectively, I sometimes recommend a diVerent approach. First, identify the group that you are interested in, for example, a group of children whose siblings have Down syndrome. This group will likely be small. Next, identify the potentially confounding variables that are available in the dataset and use these to compute propensity scores for the entire dataset. Use a matching strategy to select children to be in the control group based on their propensity scores. Table I provides a guide for when to use propensity scores to control for potentially confounding variables given diVerent sample sizes and diVerent numbers of confounding variables. For additional tutorials on propensity scores, see articles by Yanovitzky et al. (2005) or D’Agostino (1998). For a more advanced discussion, see Rosenbaum and Rubin (1983). For discussions on matching methods see Bergstralh, Kosanke, and Jacobsen (1996) or Rosenbaum (1998). There are some limitations to the use of propensity scores. First, propensity scores cannot be used prospectively to select study participants into groups. Propensity scores can only be used after the study data have been collected. Second, propensity score analyses require a minimum sample size. For each potentially confounding variable, one should have 5–10 study participants in the group of interest. Finally, if multiple groups are being compared, using propensity scores becomes a more intensive process because one must compute propensity scores separately for each two group

103

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

WHEN

TO

TABLE I USE PROPENSITY SCORES: PARTICIPANT SELECTION POTENTIALLY CONFOUNDING VARIABLES

BY

NUMBER

OF

Number of potentially confounding variables Participant selection type

1

2

Prospective

Matching on single variable

Retrospective

Matching on single variable

Matching or propensity scores as covariates (if matching could not be accomplished) Matching using propensity scores

3þ Matching on 1–2 variables followed by propensity scores as covariates Matching using propensity scores

comparison. Multiple group comparisons also results in an increased number of propensity scores that will be included as covariates in the subsequent ANCOVAs. III. A.

SMALL SAMPLE SIZES

The Problem with Small Sample Size

Small sample sizes present a second set of statistical challenges. Existing statistical wisdom is the minimum number of data observations per group is 30. Thirty is a magic number because an important statistical theorem, the central limit theorem, states that a data distribution will approach normality as the sample size increases. In social science research, the distribution will usually approach normality even when the sample size is only moderately large, for example, when the sample size is 30 or more (Hayes, 1988). Designing every study with at least 30 observations per group is ideal, however, it is often diYcult to achieve that number. For the purposes of this discussion, sample sizes fewer than 30 will be considered ‘‘small’’ and those with more than 30 observations will be considered ‘‘moderately large.’’ Practically, small is best exemplified by 10–15 observations and large as greater than 50. Researchers in developmental disabilities often conduct studies with fewer than 30 participants per group because it can be diYcult and costly to recruit participants with disabilities. Even very large epidemiological studies often end up with relatively small sample sizes once the dataset is restricted to persons with developmental disabilities, especially if the analyses involve additional constraints. For example, examining children with siblings or

104

Jennifer Urbano Blackford

children with divorced parents may greatly restrict the sample size. Small sample size, fewer than 30 observations, is a common problem for many diVerent types of developmental disability research. Small sample sizes are problematic due to both their frequent violations of normality and their limited statistical power. The most frequently used statistical tests, parametric tests, assume that three properties are faithfully kept in the study sample: normality, homogeneity of variance, and independent and identical distributions (i.i.d.). See Table II for a detailed list of the assumptions. Researchers may often assume that their data distributions are normal and therefore do not require diVerent statistical methods. Micceri (1989) demonstrated that commonly used data types are often not normally distributed, even when sample sizes are large. Micceri collected 440 large sample (N > 400) data distributions from authors of published research and major test publishers. Data included ability, achievement, and psychometric measures. He examined various characteristics of the distributions and found that only 15% of the distributions were normally distributed. The majority of the distributions (49%) had one heavy tail, indicating skewness. This finding illustrates the ubiquity of nonnormal data even with large sample sizes and should emphasize the need to find appropriate statistics for nonnormally distributed data. In addition to the normal distribution issues inherent in small sample sizes (N < 30), research designs with small sample sizes also have lower statistical power than larger sample sizes, even when the test parameters are respected. Statistical power is a function of sample size and eVect size, the size of the true diVerence. To have a significant finding with a small sample size requires a larger eVect size than that required for a larger sample. Another way to think about this issue is that for a given eVect size, all other things being equal, a significant result is more likely with 40 observations than with 20 observations. With very large eVect sizes, sample size is not an issue. However, most developmental disability research explores eVect sizes that are in the small to medium range, where sample size is a very relevant issue. Statistical analyses seek to maximize the possibility of finding a true statistically significant diVerence, while protecting against erroneous findings. Analytic methods that are especially suited to small sample sizes include nonparametric tests and permutation tests because these methods do not require distributional normality. B.

Traditional Method: Nonparametric Tests

Nonparametric tests, also known as distribution‐free tests, do not make assumptions about the normality of the data distribution, so they are

TABLE II COMMON PARAMETRIC TEST ASSUMPTIONS, IMPORTANCE, Assumption

AND

SOLUTIONS

Description

Importance

Solution

Normality

Data from each population must be normally distributed

Less important, can be violated without a large impact; violations of kurtosis (peakedness of distribution) are worse than violations of skewness (asymmetry of distribution)

Homogeneity of variance

All populations being tested must have the same variance

More important, especially problematic with unequal sample sizes

Independence of observations

Observations must be independent of one another

Nonindependence has serious impact on the significance level and power of t‐test and ANOVA

Use samples of 30 or larger to utilize central limit theorem; use transforms to normalize the data; use nonparametric tests that do not assume normality; use permutation tests that do not assume normality Using the same sample size in each group can help because variance shifts are more likely to occur when the groups are of diVerent sizes; use a correction for unequal variances (e.g., Welch’s); use permutation tests that have no variance assumptions Use a test for dependent measures or a measure that does not have an assumption of independence

106

Jennifer Urbano Blackford

appropriate tests when the data distribution is nonnormal or unknown. Nonparametric tests assume that we do not know information about the parameters of the data distribution such as the mean and standard deviation. This is why they are called ‘‘nonparametric.’’ The chi‐square, Mann– Whitney U test, and the sign test are examples of nonparametric tests. Nonparametric tests are alternatives to their parametric test counterparts. For example, the Mann–Whitney U test is the nonparametric counterpart to the t‐test. Nonparametric tests are most commonly used for the analysis of categorical or ordinal data, although they are well suited for small sample size analyses. In first approaching a dataset, it is important to examine the distribution of the data, especially when sample sizes are small. Under certain circumstances, the nonnormal distribution may still allow use of a parametric test (e.g., t‐test), particularly if the sample sizes are equal, samples sizes are fairly large (N > 25–30), and the tests are two‐tailed (Glass, Peckham, & Sanders, 1972; Sawilowsky & Blair, 1992). However, both Type I error and Type II error of parametric tests are impacted when sample sizes are small (N < 30), sample sizes are not equal, or one‐tailed tests are used. Let us briefly review Type I and Type II errors. Type I error is the chance of having a statistically significant diVerence when there is not a true diVerence in the population. When a researcher specifies an alpha level for an analysis, the researcher is explicitly stating what level of Type I error is acceptable for the statistical analysis. The alpha level, also called the nominal alpha, is typically set at 5%. It is expected that the Type I error rate for any given statistical test will be the same as the nominal alpha. When one or more of the assumptions are violated, Type I error is likely to be inflated; that is, there will be more errors than expected and the diVerence between the two groups will appear to be significant when actually it is not. Inflated Type I error can lead to reports of eVects that are not real. Violations of the parametric test assumptions can also increase Type II error. Type II error is the probability of not finding a significant result when there is a true diVerence. As Type II error increases, statistical power decreases. Decreased statistical power makes it possible to miss true diVerences. Thus, both inflated Type I and Type II errors can lead to faulty conclusions about the research findings. But nonparametric tests have more statistical power than parametric tests when sample sizes are small and data distributions are not normal; that is, nonparametric tests are more likely to correctly discover true diVerences between groups (Blair & Higgins, 1981; Sawilowsky & Blair, 1992; Tanizaki, 1997). When data come from a normally distributed population, the Mann– Whitney U test, a nonparametric test for independent groups, was shown to be 95% as powerful as the t‐test (ChernoV & Savage, 1958; Hodges &

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

107

Lehmann, 1956). Nonparametric tests have advantages over parametric test when sample sizes are small or not normally distributed. Nonparametric tests have some limitations. First, while nonparametric tests do not require normality, they do assume other properties of the dataset. The Mann–Whitney U test requires homogeneity of variance and independent observations, just like the t‐test. Failure to meet assumptions of nonparametric tests can impact Type I error, just as with parametric tests (Blair, 1981). The second limitation is that the nonparametric test often answers a question that is diVerent from the question answered by a parametric test. For example, the Mann–Whitney U test tells you the likelihood that an observation randomly selected from one group has a larger value than an observation randomly selected from the other group. Alternately, the t‐test tells you about the mean diVerence between the participants in the two groups. Thus, nonparametric tests are appropriate when the variance between the groups is similar and when the answer provided by the nonparametric test matches the research question. C.

New Method: Permutation Tests

Permutation tests are a powerful, flexible statistical method that can be used where other approaches commonly fail. Permutation tests are known by many names: exact tests, randomization tests, and rerandomization tests. A permutation test is an exact method for determining the p‐value, or statistical significance, of any statistic. A test is exact ‘‘if the actual probability of making a Type I error is exactly a for each and every one of the possibilities that make up the hypothesis’’ (Good, 2000). Permutation tests are not diVerent statistical tests. Standard statistics, like a t‐value, are still used. The diVerence is in the calculation of the p‐value. With traditional statistical methods, the p‐value is based on a standard set of tables which are derived from theoretical distributions that include assumptions (e.g., normality) about the data. When these assumptions are violated, the p‐value from the standard table is likely to be inaccurate. In contrast, the permutation test approach calculates an exact p‐value for the statistic (e.g., t‐statistic) given your observed data. This p‐value is accurate even when standard statistical assumptions are violated. Specifically, the p‐values generated from a permutation test are based on an empirical distribution of the test statistic for the observed data that represents all possible arrangements of the data under the null hypothesis. Stated another way, for a permutation test, a test statistic distribution is created from the repeated random rearrangement of the data observations. This test statistic’s distribution is an empirical distribution that represents all possible statistic values under the condition that the null hypothesis (i.e., no group

108

Jennifer Urbano Blackford TABLE III PERMUTATION TEST METHOD Permutation test step

1. Determine your research question 2. State your null hypothesis 3. Choose a test statistic

4. Compute the test statistic for the observed data 5. Rearrange the observations

6. Compute the test statistic for the new arrangement 7. Continue to rearrange the observations and compute the test statistic many times, saving the test statistic values each time

8. Calculate an exact p‐value by counting the number of values from the rearranged samples that equal or exceed the original statistic and dividing by the number of rearranged samples

WITH

EXAMPLE Example

Test for group diVerences in sociability for children with or without mental retardation There is not a diVerence between the two groups on sociability I select the t‐test as my statistic because I want to test for group diVerences in the mean sociability scores The t‐value for the observed data is 1.65 Group assignment is randomly shuZed across the 20 children; children with mental retardation may now be assigned to without mental retardation group The t‐value for the first rearrangement is .98 The t‐values over 20 rearrangements are: .98, .23, .30, 1.6, .22, .86, .15, .34, .56, .75, .32, .08, .90, .76, .30, 1.8, .20, .45, and .85. These data comprise the empirical distribution. In practice you would do a large number of rearrangements—typically 1,000–10,000 One of the 20 t‐values is greater than the original t‐value of 1.65. Thus, the exact p‐value is .05 or 1/20

diVerences) is true. The actual observed statistic value is then compared to the empirical test distribution to determine statistical significance. Statistical significance is still calculated in the traditional way. For a two‐tailed test, if the observed test statistic is greater than ±2.5% of scores in the empirical distribution, then the result is statistically significant. The actual computation for the permutation test method follows these general steps: select the test statistic, generate the empirical distribution, and calculate an exact p‐value. There are eight specific steps to performing a permutation test. Each step, with a practical example, is provided in Table III. As shown in the example, each time the group status (with/without mental retardation) is randomly shuZed across all study participants, the observed relationship between group status (with/without mental retardation) and the outcome measure (sociability) is removed. Group status is now

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

109

randomly shuZed across sociability scores, disrupting any naturally occurring relationship between mental retardation and sociability. This random pairing of group status and sociability represents one permutation sample that might exist under the null hypothesis of no group diVerences. Each time the group status is randomly shuZed, the test statistic value (t‐value) is saved. This process is repeated many times. The many random samples created represent all possible permutations of group status and sociability expected under the null hypothesis. The test statistics from each of the random samples are used to create an empirical test distribution, or a distribution of the test statistics one would expect to find under the null hypothesis. This empirical test distribution is used to determine statistical significance of the observed test statistic. The major benefit of using a permutation test approach is that permutation tests can be used with almost any type of data: small or large sample sizes; normal or nonnormal distributions; homogeneous or heterogeneous variance among groups; and continuous, ordinal, or categorical data. With all of these data types, permutation tests can be used to calculate an accurate and exact p‐value, even when standard statistical assumptions are violated. Since permutation tests are not a diVerent type of test, just an exact way to determine p‐value, any kind of statistic can be used: parametric (e.g., t‐test), nonparametric (e.g., Mann–Whitney U test), or novel (e.g., median divided by median absolute deviation). In summary, permutation tests are not new statistics but are simply a new method for determining statistical significance. Permutation tests are almost free of assumptions or the conditions that must be met in order for the results to be valid. Although permutation tests do not require normality of the data distribution or homogeneity of variance, they do have a few assumptions. First, observations across participants must be independent; that is, the data from one participant does not influence the data from another participant. Second, observations in the sample must be exchangeable. Observations are exchangeable if under the null hypothesis, no group diVerences, all possible pairings of study participant with group membership (e.g., sibling with/without Down syndrome) are possible and valid. These assumptions are usually easily met. Permutation tests are not a new method. They were first introduced in the 1930s by Pitman (1937) and later by Fisher (1966), but they did not immediately gain widespread usage. Permutation tests require large numbers of iterations (e.g., 5000) for calculation, making them nearly impossible to perform by hand or with older computers having slow processors and limited memory. The concept of the permutation tests was very appealing, but only recently aVordable high‐speed processors and memory have made the computation of permutation tests practical. Standard desktop computers

110

Jennifer Urbano Blackford

are now powerful enough to perform permutation tests in relatively manageable running times. This increased computer capacity has resulted in a renewed interest in permutation tests. D.

Applications

Permutation tests provide a versatile tool that is applicable in most situations. The flexibility, combined with the conceptual simplicity, makes permutation tests very appealing to both statisticians and nonstatisticians. Advances in computer availability and processing speed makes permutation tests possible and practical. When sample sizes are large, permutation tests and parametric statistical methods will have the same results and parametric tests may be more readily accepted. However, for small sample sizes, permutation tests should be the tool of choice. As with any other statistical analysis, you should select your analytic approach and stay with that choice. Do not perform a permutation test only following a nonsignificant p‐value from a standard parametric test. Instead, decide if the permutation test is most appropriate given your data. If you decide to use the permutation test, it should be your only statistical analysis method. Table IV provides a guide to which tests are appropriate for various sample sizes, distributions, and group variance situations. Although the use of permutation tests is relatively new, software to perform these analyses is available. Common statistical packages may have options within traditional statistical analyses to request an exact test. For example, with SAS (SAS software, 2003) you can request an exact test for a

RECOMMENDED TYPE

OF

TEST

TABLE IV SAMPLE SIZE, DISTRIBUTION,

BY

AND

VARIANCE

Sample size (per cell)

Between groups variance

Very small (<10)

Normal distribution Homogeneous variance

Permutation

Heterogeneous variance Nonnormal distribution Homogeneous variance Heterogeneous variance

Small (11–30)

Moderately large (30–50)

Large (>50)

Parametric

Parametric

Permutation

Parametric or permutation Permutation

Parametric

Parametric

Permutation Permutation

Permutation Permutation

Nonparametric Permutation

Nonparametric Permutation

Note: A permutation test is always appropriate. Parametric and nonparametric tests are recommended above because the results are likely to be similar to the permutation tests and are likely to be more familiar to reviewers.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

111

contingency table, correlation, or logistic regression. The R statistical software (R Development Core Team, 2005) currently has a package (‘‘exactRankTests’’) for computing exact rank tests, one sample tests, two sample tests, and two sample paired tests. SPSS (SPSS for Windows, 2005) has an additional package for conducting exact tests called SPSS Exact Tests. There are also several stand‐alone software packages that specialize in exact tests. The StatXact software package (StatXact version 7, 2005) performs a wide variety of permutation tests and also provides procedures (PROCs) that can be used within the SAS software. Resampling Stats (Resampling Stats, 2003) computes permutation tests and has both a stand‐alone version and companion versions for Excel and Matlab. For simple analytic designs (e.g., t‐test), it is possible to write a short custom program to perform a permutation test, only requiring basic programming skills and a standard statistical package. For programming steps, examples and flowcharts for various statistical tests, see Edgington (1995). The application of permutation tests to more complex designs, such as complex multivariate methods and factor analyses, have yet to be developed. There are several limitations to permutation tests. First, researchers need to be able to use statistical software packages in order to perform permutation tests. They cannot be done in Excel or by hand. Second, at this time permutation tests methods are not available for more complex multivariate statistical analyses. Finally, the unfamiliarity of audiences and reviewers with permutation tests may present a challenge. However, there are increasing numbers of chapters and articles describing the benefit of this method. Some peer‐ reviewed research articles have been published using permutation tests. In the near future, permutation tests should achieve widespread acceptance. To learn more about various nonparametric tests see Nonparametric statistics for the behavioral sciences (Siegel & Castellan, 1988). For more advanced reading on permutation tests see Permutation tests: a practical guide to resampling methods for testing hypotheses (Good, 2000) or Randomization tests (Edgington, 1995). IV. A.

NUMEROUS OUTCOME VARIABLES

The Problem with Analyzing Numerous Outcome Variables

Data collection is often the most expensive and time‐consuming part of a research project. Data collection can be especially diYcult for developmental disability research, where participants can be hard to find, recruit, and retain. Because each participant is so valuable, there is often a desire to collect as much data as possible. Obtaining countless data for each subject is

112

Jennifer Urbano Blackford

a useful strategy, but it results in large amounts of data on few participants. This issue is also present in very large epidemiological studies, where there are often a vast number of measures. The analysis of large amounts of data, especially with small sample sizes, is problematic for several reasons. The main concern is the multiple testing problem, which occurs when multiple statistical tests are performed to answer a single research question. When a nominal alpha level is set, the researcher is stating the amount of Type I error that is acceptable for that single test. For example, a nominal alpha of 5% indicates that the researcher will accept a 5% chance that a statistically significant result is not true. When multiple tests are performed, the actual value for Type I error increases arithmetically with each test. For example, the Type I error for performing t‐tests on 5 outcome variables would be 25% (5% 5 tests) and not the expected 5%. With increased numbers of tests, the actual value for Type I error quickly exceeds the specified alpha level. To prevent Type I errors and misinterpretation of the data, it is necessary to control for Type I error. A second issue involves research designs with many related measures (e.g., EEG, imaging) or serial measurement (e.g., time‐series on biological measurements), both of which often result in substantially more outcome variables than participants. MPTs provide a powerful new statistical tool for analyzing numerous outcome variables, while addressing the multiple testing problem and allowing for more outcome variables than participants. This versatile tool extends the permutation tests described earlier to a multiple outcome variable design. B.

Traditional Methods: Independent Tests, Data Reduction, Single‐Step Corrections

There are several traditional tools for analyzing multiple outcome variables: independent variable analysis, data reduction, and single‐step corrections (e.g., Bonferroni). The first approach to analyzing numerous outcome variables is to analyze each of the outcome variables separately. With this method, the significance for each separate test is reported as though it were an independent test of the research question. The main concern with this method is that multiple tests for any single research question are not independent. These multiple tests are often referred to as a ‘‘family’’ of tests. When conducting a ‘‘family’’ of tests, it is important to use the right alpha level. As described earlier, if there are five tests in your family of tests for a specific research question, the Type I error rate is 25% (5% 5 tests) and not the typical alpha of 5%. Having multiple tests that result in an inflated Type I error is a problem often referred to as the ‘‘multiple testing’’ or ‘‘multiple comparison’’ problem.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

113

A second approach to multiple outcome variable analysis is to reduce the data from many variables to just a few variables, also known as data reduction. Data from many variables can be combined into a single variable, guided either by theory (e.g., creating an average of multiple items to create a single test score) or by empirical data (e.g., factor analysis). Empirical data reduction methods (e.g., factor analysis) often require very large sample sizes. Data reduction eVectively reduces the number of outcome variables to be analyzed. It is useful when the variable of interest is best represented by the combined variable and not the individual variables. For example, on a common measure of childhood behavior, such as the Child Behavior Checklist, most researchers are interested in the aggregate measure of externalizing behavior and not the individual items on the measure. There are two concerns with data reduction approaches. First, important detail may be lost when data are reduced to a single variable. Losing detail is of particular concern when data analyses are exploratory because there is no prior knowledge about which particular variables may be important. Reducing the data to a single variable in exploratory analyses may obscure critical findings. Second, when data reduction is guided by empirical findings, it may be more diYcult to replicate the results because empirical data reduction in another sample would likely result in a diVerent single outcome variable. Analytic results from these diVerent outcome variables are likely to diVer as well due to naturally occurring diVerences between samples, even when selected from the same population. When decisions about data analyses are made using results from prior analyses on the same data, the probability of finding a result that fails to replicate is increased. A third traditional approach for analyzing multiple outcome variables is to first analyze all of the outcome variables separately. Next, a single‐step correction method is applied to the results to control for Type I error. The most common single‐step correction method is the Bonferroni, where the value of alpha is adjusted relative to the number of tests performed. For example, if the nominal alpha level is 5% and there are 10 tests, the Bonferroni‐corrected alpha would be 0.5% (5%/10). This method is eVective in controlling Type I error. However, this approach is conservative; that is, the overall alpha level for all of the tests is usually less than 5%. A conservative Type I error correction results in decreased statistical power. The uses of data reduction or single‐step correction methods are both acceptable approaches for the analysis of many outcome variables. However, neither approach is ideal. Information loss or study‐specific findings are likely results of data reduction methods. Statistical power is reduced with single‐step corrections. A relatively new statistical tool, MPTs, provides an alternative method. This new strategy has greater statistical power and does not require data reduction, while controlling for Type I error.

114 C.

Jennifer Urbano Blackford New Method: Multivariate Permutation Tests

MPTs are an extension of permutation tests. Although permutation testing has been around for a while, MPT (Blair & Karniski, 1993; Blair, Higgins, Karniski, & Kromrey, 1994; Pesarin, 2001; Westfall & Young, 1993) and multivariate stepwise permutation tests (Blair & Karniski, 1994; Troendle, 1995; Westfall & Young, 1993) were only introduced a decade ago. MPTs address all of the concerns found in methods that are more traditional; they control Type I error, do not require data reduction, and have strong statistical power. MPTs provide an excellent solution to the problems inherent in analyzing numerous outcome variables because: (1) they can be used with small sample sizes, (2) they can be used with numerous outcome variables, (3) there are no distributional assumptions, (4) the underlying correlation structure of the outcome variables is preserved in the test and does not impact the validity of the results, and (5) they can be used when there are more outcome variables than observations. These characteristics make MPT an extremely powerful and widely applicable statistical method. The MPT creates an empirical test distribution by rearranging observed data, just as with a permutation test. The main diVerence is that with the MPT there are multiple outcome variables, not just a single outcome variable. In order to maintain the correlations among variables, all of the outcome variables are treated as a single unit and rearranged together. There are eleven basic steps to performing an MPT. Table V illustrates each step along with a practical example. For illustrative purposes, consider a scenario in which the researcher has collected measures of EEG activity, resulting in data for 132 voxels per child. The researcher wants to compare the brain activity of 10 children with autism to 10 children with Down syndrome. Since the study is exploratory, there is no theory to guide data reduction. Instead, the researcher wants to compare the two groups at each of the 132 data points. Since there are more data points than participants, traditional statistics are not possible. In addition, if the 132 tests were performed separately, the uncorrected alpha level would be 100%. The Type I error when using a Bonferroni‐corrected alpha would be conservative, that is, less than the nominal alpha. A conservative Type I error reduces the likelihood of finding true between group diVerences. Compared to the permutation tests, there are three additional aspects to the MPT. First, the data are rearranged as a group of variables instead of each variable being rearranged independently. Rearranging all outcome variables together maintains the natural underlying correlational structure of the data. For example, in time‐series data there may be natural correlations between adjacent time points. It is important to maintain these

TABLE V MULTIVARIATE PERMUTATION TEST METHOD Permutation test step 1. Determine your research question 2. State your null hypothesis 3. Choose a test statistic

115

4. Compute the test statistic for each of the dependent variables for the observed data 5. Rearrange the observations

6. Compute the test statistic for each of the dependent variables for the new arrangement 7. Continue to rearrange the observations and compute the test statistic many times for all of the dependent variables 8. Select a multivariate statistic to represent all of the dependent variables and save this statistic for the empirical test distribution. An example of a multivariate statistic is tjmaxj, which is the maximum absolute t‐value out of a group of t‐values. Another is tjsumj which is the sum of all the (absolute) t‐values in a group

WITH

EXAMPLE Example

Test for group diVerences between children with autism and children with Down syndrome on all 132 data points, which represent voxels There is no diVerence between the two groups on any of the 132 data points I select the t‐test because I want to test whether the average value of brain activity at each time point is diVerent for children with autism compared to children with Down syndrome I compute the t‐value for each of the 132 variables. For illustrative purposes, I will just show the t‐values for the first 5 variables: 1.65, 2.26, 1.95, .36, and .24 Group assignment is randomly shuZed assignment across children, without changing the relationships among variables; that is, all 132 variables stay the same and only group assignment changes. Children with autism may now be assigned to the children with Down syndrome group The t‐values for the first 5 variables this time are: .65, .98, .12, .39, and .53 The data are rearranged 20 times resulting in t‐values for all of the variables for each rearrangement The tjmaxj multivariate statistic is used to represent the largest t‐value from all of the 72 data points for each rearrangement. The tjmaxj for the first rearrangement was .98. The tjmaxj values for all 20 rearrangements are: .98, .23, .30, 1.6, .22, .86, .15, .34, .56, .75, .32, .08, .90, .76, .30, 1.97, .20, .45, and .85. These data comprise the empirical distribution. In practice you would do a large number of rearrangements—typically 1,000–10,000 (continued)

TABLE V (Continued )

116

Permutation test step

Example

9. Sort the observed statistic values from all of the dependent variables and employ either a single‐step, step‐up, or step‐down testing procedure (see chapter for definitions). For a two‐tailed test, use the absolute values 10. For the first test, calculate an exact p‐value by counting the number of values from the rearranged samples that equal or exceed the original statistic and dividing by the number of rearranged samples. If you are using a single‐step procedure, you can calculate exact p‐values for all dependent variables now and your testing is complete. If you are using a step procedure, continue to the next step 11. If the previous test was significant, proceed with testing the next largest test statistic. For the step‐down‐ testing process, each subsequent test is based on a subset of the original empirical distribution. Rearranged data for the variables already tested are removed by deleting the t‐values associated with those variables from the data generated in step 7. Continue with step 8 to test the next value. This process continues until a test statistic is not statistically significant or the last variable is tested

A step‐down testing procedure is used, sorting the absolute t‐values from largest to smallest

None of the 20 t‐values is greater than the largest absolute original t‐value of 2.26, the t‐value corresponding to the second outcome variable. Thus, the exact p‐value is .00 or 0/20

To continue testing the remaining outcome variables, return to the 132 (voxels) 20 (rearrangements) data matrix created in step 7 and delete the t‐value for variable 2 (the variable just tested) in each of the 20 rearrangements. The new matrix is 131 (voxels) 20 (rearrangements). Continue with step 8 and select the tjmaxj for each rearrangement from the 131 remaining outcome variables to create the empirical distribution. Test the next largest t‐value, 1.95, against that distribution and calculate an exact p‐value

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

117

relationships because maintaining such relationships can substantially increase statistical power in the multivariate analysis. The second addition is in the selection of a value for the empirical test distribution. With the MPT, each rearrangement results in multiple test statistics (e.g., one t‐test for each outcome variable). However, the final empirical distribution still needs to be composed of only one value from each rearrangement. It is necessary to select a single test statistic from each rearrangement. The selection of one value is commonly done using a multivariate statistic. One example of a multivariate statistic is tjmaxj, which is the maximum absolute t‐value out of a group of t‐values. Another commonly used multivariate statistic is tjsumj, or the sum of all of the absolute t‐values in a group. The largest absolute value is best when the eVect is expected to be isolated to one or a few outcome variables. An isolated eVect is frequently seen when analyzing data for a time‐ or region‐specific eVect in large datasets. The sum of the absolute t‐values should be used when testing for an overall eVect of the outcome variables. The third addition is that a stepwise method is applied to the process of significance testing for the group of outcome variables. Single‐step methods, like Bonferroni, test the significance of all outcome variables at once using an adjusted alpha. Multistep methods sort all of the variables and then test the significance of one variable at a time. After the first variable is tested, the test criterion is modified for the remaining variables. Then the next variable is tested and the process continues iteratively. The main diVerence between a single‐step and multistep method is the test criteria used for determining significance. The test criterion is constant with a single‐step method but with the multistep method, it changes iteratively. The multistep method has increased statistical power across the group of significance tests because the criterion is modified with each variable tested. Multistep methods can be either step‐up (starting with the least significant variable) or step‐down (starting with most significant variable). Both methods can be applied to MPTs. Research has shown that Type I error rates are similar for both methods (Blair & Karniski, 1994; Blair, Troendle, & Beck, 1996) and power is slightly increases with step‐up methods (Blair & Karniski, 1994; Blair et al., 1996). D.

Applications

MPTs are new statistical methods that provide exact p‐values for analyses of multiple outcome variables. MPTs have no assumptions about the data distribution, variance, or the underlying correlation structure of the data; however, the permutation test assumption of exchangeability still holds. Type I error is controlled for at a family level and statistical power is higher

118

Jennifer Urbano Blackford

than that for other methods. MPTs can be used when the number of outcome variables is more than the number of observations. In summary, MPTs provide a very flexible and powerful analytic tool for multivariate analyses. This method is especially well suited for exploratory analyses with numerous outcome variables. As with any technique, MPTs have limitations. MPTs are more challenging to perform than the traditional methods. However, there are increasing numbers of software programs that can perform MPTs. The SAS statistical software (SAS software, 2003) has a procedure, PROC MULTTEST, which can perform multivariate stepwise permutation tests for a variety of basic statistical procedures. Other programs include StatExact (StatXact version 7, 2005) and NPC 2.0 (NPC 2.0, 2001). An MPT for correlations program (Yoder, Blackford, Waller, & Kim, 2004) is available at http://kc. vanderbilt.edu/Quant/Programs/Programs.htm. Finally, one could write a custom program to perform MPTs using the steps outlined in Table V. The other limitations for MPTs are the same as the permutation test limitations. MPTs are not available for complex multivariate analytic methods. Audiences and reviewers may be unfamiliar with this statistical technique. For more information on the MPTs, see Resampling‐based multiple testing: Examples and methods for P‐value adjustment (Westfall & Young, 1993) and Multivariate permutation tests: With applications in biostatistics (Pesarin, 2001). V.

CLOSING REMARKS

Today’s researchers in the fields of developmental epidemiology and developmental disabilities are fortunate. Such new strategies, such as propensity scores, permutation testing, and MPTs, enable researchers to overcome statistical challenges that have been frustrating and have plagued studies for years. These new statistical approaches are flexible, practical, and easy to perform using current computer technology and available software. Use this chapter as the key to opening a door to new research avenues and exciting new discoveries. REFERENCES Bergstralh, E., Kosanke, J., & Jacobsen, S. (1996). Software for optimal matching in observational studies. Epidemiology, 7, 331–332.

STATISTICAL ISSUES IN DEVELOPMENTAL EPIDEMIOLOGY RESEARCH

119

Blair, R. C. (1981). A reaction to ‘‘Consequences of failure to meet assumptions underlying the fixed eVects analysis of variance and covariance.’’ Review of Educational Research, 51, 499–507. Blair, R. C., & Higgins, J. J. (1981). A note on the asymptotic relative eYciency of the Wilcoxon rank‐sum test relative to the independent means t test under mixtures of two normal distributions. British Journal of Mathematical and Statistical Psychology, 34, 124–128. Blair, R. C., & Karniski, W. (1993). An alternative method for significance testing of waveform diVerence potentials. Psychophysiology, 30, 518–524. Blair, R. C., & Karniski, W. (1994). Distribution‐free statistical analyses of surface and volumetric maps. In R. W. Thatcher, M. Hallett, T. ZeYro, E. R. John, & M. Huerta (Eds.), Functional neuroimaging: Technical foundations (pp. 19–28). San Diego: Academic Press. Blair, R. C., Higgins, J. J., Karniski, W., & Kromrey, J. D. (1994). A study of multivariate permutation tests which may replace Hotelling’s T2 test in prescribed circumstances. Multivariate Behavioral Research, 29, 141–163. Blair, R. C., Troendle, J., & Beck, R. W. (1996). Control of familywise errors in multiple endpoint assessments via stepwise permutation tests. Statistics in Medicine, 15, 1107–1121. ChernoV, H., & Savage, I. R. (1958). Asymptotic normality and eYciency of certain nonparametric test statistics. The Annals of Mathematical Statistics, 29, 972–994. D’Agostino, R. B., Jr. (1998). Propensity score methods for bias reduction in the comparison of a treatment to a non‐randomized control group. Statistics in Medicine, 17, 2265–2281. Edgington, E. S. (1995). Randomization tests (3rd ed.). New York: Marcel Dekker, Inc. Erickson, J. D. (1978). Down syndrome, paternal age, maternal age and birth order. Annals of Human Genetics, 41, 289–298. Fisher, R. A. (1966). The design of experiments (8th ed.). New York: Hafner. Glass, G., Peckham, P., & Sanders, J. (1972). Consequences of failure to meet assumptions underlying the fixed eVects analysis of variance and covariance. Review of Educational Research, 42, 237–288. Good, P. I. (2000). Permutation tests: A practical guide to resampling methods for testing hypotheses (2nd ed.). New York: Springer. Hayes, W. L. (1988). Statistics. Fort Worth: Holt, Rinehart and Winston, Inc. Hodges, J. L., Jr., & Lehmann, E. L. (1956). The eYciency of some nonparametric competitors of the t‐test. The Annals of Mathematical Statistics, 27, 324–335. Hook, E. B., & Lindsjo, A. (1978). Down syndrome in live births by single year maternal age interval in a Swedish study: Comparison with results from a New York State study. American Journal of Human Genetics, 30, 19–27. Micceri, T. (1989). The unicorn, the normal curve, and other improbable creatures. Psychological Bulletin, 105, 156–166. NPC 2.0 (2001). Ponte San Nicolo. Italy: Methodologica. Penrose, L. S. (1933). The relative eVect of paternal age and maternal age in mongolism. Journal of Genetics, 27, 219. Pesarin, F. (2001). Multivariate permutation tests: With applications in biostatistics. New York: Wiley. Pitman, E. J. G. (1937). Significance test which may be applied to samples from any population I and II. Journal of the Royal Statistical Society Series, 4, 119–130, 225–232. R Development Core Team (2005). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Resampling Stats (2003). Arlington, VA : Resampling Stats, Inc. Rosenbaum, P. R. (1998). Multivariate matching methods. Encyclopedia of Statistical Sciences, 2, 435–438.

120

Jennifer Urbano Blackford

Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal eVects. Biometrika, 70, 41–55. SAS software (2003). Cary, NC: SAS Institute, Inc. Sawilowsky, S. S., & Blair, R. C. (1992). A more realistic look at the robustness and type‐II error properties of the t‐test to departures from population normality. Psychological Bulletin, 111, 352–360. Siegel, S., & Castellan, N., Jr. (1988). Nonparametric statistics for the behavioral sciences (2nd ed.). New York: McGraw‐Hill, Inc. SPSS for Windows (2005). Chicago: SPSS, Inc. StatXact version 7 (2005).Cambridge, MA: Cytel Inc. Tanizaki, H. (1997). Power comparison of non‐parametric tests: Small‐sample properties from Monte Carlo experiments. Journal of Applied Statistics, 24, 603–632. Troendle, J. F. (1995). A stepwise resampling method of multiple hypothesis‐testing. Journal of the American Statistical Association, 90, 370–378. Westfall, P. H., & Young, S. S. (1993). Resampling‐based multiple testing: Examples and methods for P‐value adjustment. New York: Wiley. Yanovitzky, I., Zanutto, E., & Hornik, R. (2005). Estimating causal eVects of public health education campaigns using propensity score methodology. Evaluation and Program Planning, 28, 209–220. Yoder, P. J., Blackford, J. U., Waller, N. G., & Kim, G. (2004). Enhancing power while controlling family‐wise error: An illustration of the issues using electrocortical studies. Journal of Clinical and Experimental Neuropsychology, 26, 320–331.

Economic Perspectives on Service Choice and Optimal Policy: Understanding the Effects of Family Heterogeneity on MR/DD Outcomes* STEPHANIE A. SO VANDERBILT KENNEDY CENTER, DEPARTMENT OF ECONOMICS VANDERBILT UNIVERSITY, NASHVILLE, TENNESSEE; AND VANDERBILT KENNEDY CENTER, DEPARTMENT OF PEDIATRICS, VANDERBILT UNIVERSITY NASHVILLE, TENNESSEE

I.

INTRODUCTION

Other chapters in this issue have explored the nature and usefulness of developmental epidemiology in the world of mental retardation and developmental disabilities (MR/DD) research. How, then, does a chapter on economics fit into this schema? The answer is simple: economic research is complementary to other disciplines that investigate outcomes. MR/DD researchers can use economic perspectives and methods to translate the findings of their research into policy. II.

WHAT IS ECONOMICS?

Economics is a social science. The problems that it studies are all driven by the fundamental realization that resources are scarce, but people have competing uses for them. The contribution of economics comes from its studies of how people make decisions to allocate those scarce resources to one use or another. *Author’s Note: Funding for this research by the Vanderbilt Kennedy Center’s Nicholas Hobbs Society and NICHD grant RO3HD 050468 is gratefully acknowledged by the author. The author would like to thank Dr. John P. Conley, Dr. Robert M. Hodapp, and Dr. Richard C. Urbano for comments on previous versions of this chapter. The views expressed here are solely the responsibility of the author and should not be interpreted as reflecting the views of Vanderbilt University or of any other person associated with the Nicholas Hobbs Society or NICHD. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

121

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33006-6

122

Stephanie A. So

Resources may be defined in various ways, depending on the research question of interest. Resources include time, income, land, breathable air, and so on. In studying (1) the behaviors of people as they decide to use these resources and (2) the outcomes and implications of these decisions, economics can be and is an enormous field. Like any social science, economics includes many subfields. Some subfields specialize on the sets of decisions they examine, while other subfields arise because of a common interest in the outcomes associated with many diVerent decisions. Labor economists, for example, study the decisions to work (supply work) and the decisions to hire workers (demand work). They study the outcomes of those decisions such as the employment rates, wage rates, and productivities by industry or company. They study all of the factors that aVect how people make these decisions (the existence of unions, the structure of unemployment benefits, and so on). This installed base of knowledge about why and how people work helps to develop or analyze policies that are aimed at employment outcomes (e.g., the American Disabilities Act, hours restrictions on medical residents). The central issue in labor economics is how people perceive the value for diVerent uses of time. Policies and circumstances aVect the valuation of these alternative uses of time. The notion that time is scarce, particularly for parents, is a theme that we will develop later in this chapter as we examine the economic decisions that aVect the developmental and health outcomes for children with MR/DD. At the same time, researchers who are interested in the employment outcomes and transitions for adults with MR/DD should be aware that there may be much of interest to them in the labor economics literature. Other economic subfields concentrate on issues surrounding given outcomes of interest. Health economics and the economics of education are examples of these. Health economists are fundamentally interested in how decisions to allocate resources aVect health outcomes. Studies within health economics may vary widely, according to the decision maker. Some studies examine the behaviors and outcomes of patients and their families, because these are the agents who demand care in health markets. Other studies model the decisions of providers (physicians, therapists, and so on) who supply care. Some of the most important questions about providers are how they respond to incentives, such as reimbursement policies, treatment norms, more motivated versus less motivated patients, and regulation. Because we have a third‐party payer system in this country, still other studies model the incentives of payers, such as private or government insurers, and the eVects that their decisions will have on health services use and outcomes. In each case, the central question is how the decisions of these actors in the health care market combine to aVect service use and health outcomes.

ECONOMIC PERSPECTIVES

123

Many of these studies have parallel and/or related issues to questions about disability services and outcomes. While the economics of disability services has received less attention to date, interest in disability research is growing as economists begin to explore the distinctions and connections between health and disability, as well as their eVects on other economic decisions, outcomes, and policy‐relevant behaviors. A common interest in outcomes also unites the economics of education. These studies ask about all the inputs associated with educational outcomes. The particular research focus for these decisions may vary with: the diVerences (and relatedness) between educational settings such as public and private schools; the choice and eVectiveness of inputs that are selected and provided in schools, homes, neighborhoods, or before birth (as in the nature versus nurture hypothesis); or the eVects of policy. When evaluating the eVects of policy, or any factor, on choices and outcomes, the eVects may be direct and indirect. These eVects may be conceptually meaningful to the researcher in understanding the mechanisms by which outcomes may be improved. For example, Head Start may have a direct eVect on educational outcomes, say test scores, through its special education services. At the same time, there may be substantial indirect eVects of the program on the same outcome of interest. Head Start may also improve educational outcomes by providing nutritious meals: the meals may improve child health, improve the abilities for children to attend while in school, and translate into higher test scores. Empirical evidence on the strengths of these diVerent eVects may be tremendously helpful in thinking about the net benefits and costs of a policy. MR/DD researchers primarily interested in the developmental and educational outcomes of children may already be familiar with many of the studies and research methods developed and used in the economics of education literature. Finally, there are areas of economic research, such as public economics or the theory of risk and insurance, which add more universal insights about how people make decisions. The lessons and implications from these models cut across many types of decisions, outcomes, and policy. On the technical side of economics, there is econometrics, dedicated to the development and improvement of the statistical techniques that economists need to analyze all of these decisions and outcomes in the data. A.

Common Misperceptions About Economics, Economists, and the Importance of Costs

In an ideal world, with limitless resources, every problem could be solved by devoting as many resources as were needed to the best uses available. In this idealized world, every patient could receive the most eVective treatments

124

Stephanie A. So

and services and every student could receive the finest inputs to education. In the real world, however, resources are limited and economic arguments invariably intrude on allocation decisions. People perceive economic interests to be behind managed care organizations, government insurance programs, and payers or politicians who require proof that programs are cost eVective. Through guilt by association, economics often is perceived, at best, as an unimaginative and inflexible realm of inquiry or, at worst, as a capricious and arbitrary set of activities somehow related to the counting of costs and benefits. It is true that costs and benefits play important roles in economics. After all, how people perceive costs and benefits, throughout all studies of economics, shape the behaviors of people as they decide to allocate resources. Economists are keenly interested in understanding these abstract issues (conceptualized as the eVects of incentives on choice), as well as in understanding how specific changes in incentives aVect outcomes. It is also true, however, that faced with the (relatively few) economic studies that catch the public eye, the careful attention that economists pay to diVerent kinds of incentives will likely be lost in the reportage. Instead, the public’s attention will be drawn to the highly situation‐specific decision, like how the government is going to spend your tax money or whether the insurance companies are going to add a specific health benefit that you care about. When these studies are interpreted out of academic context and without reference to the scientific method (the assumptions, the data, the research methodology, the limitations to interpretation), it is easy to understand why the public is led to believe that economists are fixated on dollars and cents. B.

True Relationship Between Costs and Economics

As illustrated earlier, there are economic choices (i.e., behavioral decisions that govern resource allocation) in the framework of every overt decision: the decision to work, the decision to take one’s child to the physician, and the decision to track down the details of the economic study reported in the newspaper. Readers of newspaper articles have many things to do with their time. Learning more about economics is costly. Those who have weighed the costs and benefits and choose not to learn any more about economics likely will never do so unless something changes. This may be despite economists’ enduring certainty that they have delivered the most thorough and illuminating of explanations in their academic journals. And so the outcome persists that readers will tend to misinterpret what economics can contribute. Could we improve outcomes in the economic literacy of newspaper readers? Perhaps we could, but only if we change the fundamental trade‐oVs that face them. We could lower the costs of choosing to learn economics by writing

ECONOMIC PERSPECTIVES

125

more interesting and accessible articles in fields outside of academic economics. Alternatively, we might expect readers to choose to learn more about economics if their situation changes so that they perceive a higher benefit to learning economics. Predicting this change in behavior (and subsequent outcomes) could be made in as systematic a way as predicting that an avid sports reader will be more likely to turn to the arts section when her new love interest reveals his passion for the ballet. Here we come to the two abstract and key potential roles for costs in economic studies. Costs are related to economic decisions, generally, as either (1) a measure of the outcomes we observe (e.g., when one decides to go to the doctor, the action results in an expenditure) or (2) as an important incentive that influences the decision and thus the outcome of interest (e.g., the price of the doctor may help to predict how much health care one uses and thus how healthy one is as a result). Particular costs, such as the price of a commodity or the value of an asset, are of no interest unless they are important to the economic model in one of those two ways. Thus, the labor economist may study how work decisions vary between women and men without ever checking on the exact cost of business attire and the health economist may study the likelihood that hospitals will specialize in cardiac surgery without ever seeing the net income statements of the hospitals involved. Most likely, they are focused on understanding the role of far more important incentives that shape the diVerences in the respective economic decisions and outcomes that result. C.

Economics Research Strategy

The newspaper example contains the essence of the economic research strategy. Regardless of the economic decision under study, economists model the behavior of decision makers, or agents and how this behavior results in a choice. Depending on the decision of interest, agents might be consumers (on the demand side) or firms (on the supply side), but each agent faces a choice that results in an outcome. Consumers decide whether and how much of a good or service to buy. Firms decide what and how much to produce. The choice may be observed as a quantity that is purchased or sold or, downstream, as the eVect on or a change in health outcomes. Agents may be countries which are trying to decide the terms of a trade agreement, politicians who are deciding on a voting strategy, parents who are trying to determine which school district to use, or school oYcials who are trying to figure out how to put their resources to best use. In each case, the model specifies the objective that the decision maker wishes to maximize with his decision and the trade‐oVs that are faced in making the decisions. The trade‐oVs arise because the agent faces scarce resources. These scarce resources, and whatever else limits the agent’s decisions, are called

126

Stephanie A. So

constraints. Once the problem is framed, decisions and the outcomes of those decisions are examined as they relate to the model’s parameters. In classical labor economics, the worker may want to maximize his personal happiness, so personal welfare is the objective function. Both leisure (not working) and income (being able to purchase goods) make him happy and so he must decide on how to allocate his time between leisure activities and work. Time has been called the scarcest of resources: it is impossible to manufacture any more hours in the day. A wage rate is a parameter of the model that influences how he will evaluate the trade‐oV between devoting an hour to leisure versus work. Thus, as the wage rate changes, we might observe diVerent workers’ decisions about how hard to work. Some workers may face a more severe problem if they have no other means of support. If those workers do not work at all, they will not survive. An economic model for these workers might reflect their circumstances by modeling an additional constraint. Informally, the additional constraint might be specified so that it restricts their choice set by ruling out an option (e.g., they may not choose a life of leisure). Choices always remain. (Otherwise, it is no longer an economics problem.) Even these workers will vary in how much they decide to work after they have satisfied their survival constraint. In other research, So (2002) has argued that health and disability must be modeled more explicitly in economic models of family labor decisions. Rather than adopt the traditional economic model that distinguishes clearly between labor and leisure as the primary two alternatives to time, So showed that caregiving activities devoted to chronic conditions ought to be considered as a third distinct activity. The rewards and costs that characterize caregiving, under more adverse circumstances, may be conceptually very diVerent and empirically distinguishable from the rewards and costs that characterize typical caregiving activities. So’s model makes the economic argument that decisions and outcomes of families who are faced with these challenges may diVer strikingly from the predictions that one might make about the same families under the assumption that they do not face any within household demands to care for someone with a chronic illness or disability (So, 2002). Furthermore, policies predicated on the idea that families balance only typical trade‐oVs may require serious reevaluation when applied to these caregiving families. Like the workers who face an extra starvation constraint, families who have caregiving requirements may not have the same flexibility to devote resources to other activities as typical economic theory might suggest. Even when the decision maker is not an individual, economic models continue to use the structure of agent, objective, and constraints to study decisions and outcomes. For example, health care organizations may wish to maximize profits, but in doing so must choose between diVerent combinations

ECONOMIC PERSPECTIVES

127

of quantity and quality of care provided. In health economics, firms that face one set of constraints may opt to serve many patients while forgoing the most cutting‐edge technologies, whereas firms that face other constraints may serve a select few patients but provide the most luxurious sets of services possible. These firms face multiple constraints as well: on the kinds of services they can oVer as regulated by the government, by the kinds of market conditions that make diVerent choices possible, and, ultimately, by their financial conditions. As economists have learned from building and testing these models, they have found that outcomes emerge, on average, in ways that are consistent with the idea that agents behave as if they are maximizing their objectives given their constraints. If the price of heating oil goes up, the average person will consume less by turning the thermostat down. These behaviors lead to new questions about new outcomes. Will they find new sources of heat so that the temperature will go back up? Will they wear sweaters instead so that the temperature remains low? Temperature may not be the most important outcome for some people, and so it is that more complicated economic models are required when the outcome is produced via more complex mechanisms. The important concern in more complicated models is that the models produce testable results which are then examined in the data. If diVerent families behave diVerently to an increase in price, what does that tell us about the agents’ decision processes, their objectives, their constraints, or their outcomes? In each of these diVerent situations, economic analysis of the outcomes that we observe help us to formulate the correct questions about the circumstances of the agents themselves, just as we would be led to ask questions if we observed a sudden increase in cases of hypothermia in certain parts of town when heating oil prices went up. If we can identify that outcomes change as constraints or characteristics change, then we can learn something about both the decision process and its consequences. The central issue of empirical economic studies is to seek results that strengthen, test, refute, or add to economic theories about why people behave the way that they do. As empirical evidence on how behaviors that shape choices are understood, and as the outcomes that result from those choices are understood, then economics helps us to understand how to use policy to aVect outcomes. Policies are merely special cases of the sets of circumstances and incentives that guide economic behavior. Economics asks, in order to achieve an objective, what policy should we implement (taxes, regulation, incentives to increase competition)? Then, the discipline of the science begins again. How will this change the incentives to reallocate resources? What outcomes will result? In the context of economics as a behavioral science, the empirical methods and results produced by economic studies improve our ability to answer these kinds of questions.

128 D.

Stephanie A. So Newspaper Readers and Children with MR/DD Outcomes: The Connection

The main point of this chapter is to explore the idea of economics as a behavioral science as it relates to health policy generally and MR/DD policy specifically. Our argument is the following. Even if economists believe they have provided fundamental insights in their articles, if people do not choose to read the articles, then there may be (in the starkest case) absolutely no benefit transmitted from that academic service to those outcomes. Similarly, services that are oVered to children with MR/DD outcomes, even if they are thought to be eVective by experts, will not change outcomes if children do not use them. We realize that the problem, in reality, is much more complex than this. The rest of the chapter elaborates on some of these issues.

III.

THE RELATIONSHIP BETWEEN ECONOMICS AND POLICY

Let us begin by considering one of the classic examples of the ‘‘law of unintended consequences.’’ The basic idea, again, is that people respond to incentives. It is irresponsible and often self‐defeating to make policy on the assumption that people will not change their behavior in order to maximize their best interests under the new incentives. In other words, in framing the alternatives of diVerent policies, economists must consider the possibility of unexpected dynamic responses to policy shifts, not just the simple first order static eVects that were intended. Beginning in the 1960s, federal regulators mandated increasingly rigorous safety standards on automobile manufacturers. Seatbelts were required to be installed in every car; later the regulations included safety glass, crumple zones, airbags, and many other smaller improvements. On its face, lowering the probability that any particular auto accident results in death or serious injury seems like very sensible public safety policy. One would expect that the cost–benefit analysis would be straightforward: the costs of improvements versus the savings one might recoup from, say, halving the number of deaths per accident. However, this naı¨ve prediction ignores how drivers respond to driving safer cars. Each driver makes choices about how safely to drive. Single people without families are more inclined to drive after a night out at the bar. If an executive is late to a meeting, he might drive a little faster and not come to a complete stop at a light before turning right. On the other hand, if our children are in the car with us, we take far fewer risks. The point is that

ECONOMIC PERSPECTIVES

129

people weigh the costs and benefits of reckless driving depending on their circumstances. One of the major costs of reckless driving is the chance of being injured or killed in an accident. Improvements in car safety lower these potential costs. Thus, when car safety regulations were put into eVect, one would expect to see more reckless driving and an increase in the number of accidents as a result. How does improved automobile safety aVect the outcome of vehicular fatalities? On the one hand, there are more accidents. On the other, each accident has a smaller probability that it results in a fatality. Because the total number of deaths is a product of these two eVects, without looking at the data, it is unclear if mandated increases in auto safety will, in fact, save lives. What this example tells us is that there is a potentially important unintended consequence of the policy. Determining its degree of importance is an empirical question. Peltzman (1975) executed just such a study and found both of these oVsetting eVects present. Fortunately, he found that the safety eVect outweighed the reckless driving eVect so, on net, vehicular fatalities decreased. He also discovered another unintended consequence: the increase in reckless driving resulted in an increase in the number of pedestrian deaths. Thus, how one feels about further increases in car safety might depend on how one gets around. If one walks, takes the bus, or rides a bicycle, one might want to mandate steel spikes that come out of the steering wheel and point directly at the driver’s chest. Doing so might just save the lives of more pedestrians. Another leading example comes from education policy. When Brown v. Board of Education was settled in 1954, the Supreme Court turned down the argument that schools which were segregated but equal in all physical respects were able to oVer the same opportunities to students of all races. The principle of the law was well reasoned and the mandatory busing and other policy responses that followed were all well‐intentioned eVorts to increase educational opportunities and outcomes for disadvantaged children. However, the immediate eVect of these policies was that wealthier parents withdrew their children from public schools and chose private education instead. Since then, public schools, especially in larger cities, seem to have been in a state of continual crisis. Some of the most recent research in the economics of education has argued for the importance of at least two factors that improve educational outcomes for children. The first is the peer group a child has in the school (Gaviria & Raphael, 1997; Robertson & Symons, 1996). Children from any educational and social background do better when they share a classroom with other children who do well in school, do not disrupt class, and set a good example.

130

Stephanie A. So

The second is teacher quality (Hanushek, 1997). A motivated and experienced teacher simply does a better job compared to the alternative. Perhaps surprisingly, something that in itself seems to have very little eVect is increased spending per pupil (Gaviria & Raphael, 1997; Hanushek, 1996). One need only compare the spending in New York City or Chicago to that in rural school districts in Montana to see the plausibility of this result. Here we see the crux of the problem: we cannot force parents to send their children to public schools nor force the best teachers to teach in them. Wealthy parents are willing to pay to buy their children a peer group of smart children in order to increase their child’s opportunities. Such schools benefit from providing such a peer group and so oVer scholarships to poor but smart children. On average, the wealthy and the smart will tend to find alternatives to troubled public schools, while the poor and the less able will be left behind. As the quality of the peer group at public schools declines, other parents who are less wealthy will reach a point where they also feel they need to seek an alternative. An unintended consequence of the desegregation policies was to lower the average quality of the peer group in public schools. Teachers also tend to prefer to teach smart, well‐behaved children, so schools (public or private) that oVer classrooms full of such children have their choice of the best teachers available. Thus, teaching talent becomes concentrated at the best schools, which deepens the problem for public schools. Is it still possible that mandatory busing improved school outcomes for some children? Of course, it might be that the fully segregated schools that minority children were forced to attend before the 1950s were even worse than the partly integrated schools that they subsequently attended. However, it is equally plausible that the newly integrated schools did not do as good a job as they did before Brown v. Board of Education. To find out if minority children’s educational outcomes improved and, if so, by how much, is again an empirical question. Here, the correct comparison is that of the desegregated children’s outcomes to the counterfactual of what the outcomes would have been if no such policy had taken place. It is in the disentangling of the direct eVect of the policy and the indirect (sometimes unintended) eVects that econometrics is useful. We will discuss some of these econometrics problems later. What can be seen from the immediate discussion is that agents acting in their own self‐interests may respond to changes in policy in ways that potentially may more than oVset the intended and socially beneficial first order eVects. Are there solutions to this problem? Economic theory helps us to understand the ways that people will respond to policy changes. Econometric analysis takes us a bit further by helping us to understand the magnitude of these eVects.

ECONOMIC PERSPECTIVES

A.

131

Economics, Health Policy, and Health Outcomes

Incentives matter in health policy as much as they do anywhere else. An obvious example is the fundamental dilemma of how to price health care. We would like people to balance the costs and benefits of their health care consumption. If people use too little, they miss days at work, feel miserable, and may ignore conditions that are ultimately very expensive to treat in the later stages. If they consume too much, they bankrupt the health care system. This creates a basic tension to which there really is no satisfactory solution. If we charge people the full price for physician visits or other services, they will underutilize services and ignore conditions that could be treated cheaply if services are delivered early. If we let them use services for free, some people will use them just to talk to someone. Variations on these trade‐oVs exist for any service. A dilemma exists if we would like to insure people against health care costs that exceed their ability to pay, but still give them incentives not to overuse the system. The implementation of co‐payments by insurers is a crude way to try to balance these competing objectives. How to set these co‐payments in a way that optimally balances these objectives is, again, an empirical question. One obvious problem that arises is that one size could never fit all. Poor people should have lower co‐payments than rich people since a given co‐payment has a larger disincentive eVect on the poor as well as a lessening of the insurance dimension by a larger amount. Co‐payments are an example of what economists call a ‘‘second‐best’’ solution. This means that the problem is so constrained that first‐best, socially optimal policies are not available. It would be impractical and perhaps illegal to charge every patient a diVerent co‐payment based on the patient’s income and on how essential the doctor believed the service was. Thus, we have to look for the best alternative policy that does not violate the institutional and economic constraints that policy makers face. Weighing patients’ benefits gets trickier still when economists consider the reality that a patient may have more than one condition. In recent papers, such as those by Dow, Philipson, and Sala‐i‐Martin (1999) and Peltzman (2002), a slightly more subtle application of the car safety story is applied to health care. These papers examine the incentive eVects that arise when risks to mortality are reduced in only one medical area. For example, as technology has reduced the risk of dying from infectious diseases, people may devote more of their resources to other activities. They may travel more freely, eat more indulgently, and increase their risks of diabetes instead. The empirical estimation of these oVsetting behaviors across mortality risks are of great current interest. The preceding examples contain lessons about the diYculties involved in estimating the eVects from a key variable such as changes in policy incentives

132

Stephanie A. So

or in response to new technologies. In health economics, two analogous problems are the following: (1) to estimate the eVect of a service on health outcomes and (2) to estimate the eVect of incentives on service use. As the reader will suspect, these research questions are intertwined with many related decisions that take place within particular family settings. To illustrate the last point, consider a final example from the economics of obesity. Chou, Grossman, and SaVer (2004) reported that two‐thirds of the increase in adult obesity between the mid‐1980s and 1999 were explained by the rapid growth in the per capita number of fast food and full‐service restaurants. However, they argue that rapid increases in the supply of services are usually in response to market demand. Anderson, Butcher, and Levine (2003) provide strong evidence that obesity in children may be explained by demand‐side issues. Their findings show that increases in average maternal work hours account for as much as one‐third of the growth in obesity among some children. Finding the right policy on how to treat the epidemic of childhood obesity will require significant social science input about the within‐home decisions that try to balance these competing objectives of child health and family work. B.

Overarching Lesson: The Economics of Services, Policy, and Outcomes

It is not enough to show that services (such as the academic economic literature or packaged salads in fast food restaurants) are available and could improve outcomes. It is also risky to mandate that services be used in a certain way or in a certain combination. As long as people are free to follow their own interests, they will find ways to select alternatives that work best for them, given their circumstances, even if these services do not yield the best possible outcome for them. It is understandable that researchers, who work for years to show the eVectiveness of a service through painstaking randomized controlled trials, and providers, who follow evidence‐based best practices, may become frustrated with the general public when they do not choose to consume these services. One of the main insights from economics is that we cannot force people to behave as policy intends. Rather than rail against the constraints that people face which in turn cause them to choose less than medically optimal services, public policy must take these behaviors into account and, instead, try to change the incentives that people face. No matter how eVective a service is determined to be in a controlled setting, it is no good unless the services ultimately are chosen by the clients—or their families. Economics can help researchers in other fields to learn what they

ECONOMIC PERSPECTIVES

133

need to know about behavior in order to improve policy. In other words, it is critical to approach public health questions from a social science as well as a biological science standpoint.

IV.

ECONOMICS OF MR/DD OUTCOMES

At first, it may seem strange to consider the outcomes of children with MR/DD from this standpoint. We might be able to provide incentives for diabetes patients to eat better, for obese patients to exercise, and for cancer patients to quit smoking. If the client is an adult, we might be able to tailor these interventions to be more eVective according to the incentives that face the type of patient (rich or poor, male or female, old or young) that we are treating (Ensor & Cooper, 2004). Children in general, including children with MR/DD, however, either do not or may not be allowed to make decisions about their own service use and outcomes. In these cases, economic theory clearly dictates that the parents should be studied because they are the ones who make decisions about which services to consume. In this chapter, we consider how parents aVect service use (thus outcomes) in two important ways. They decide on the kinds of inputs that the family will provide within the household in addition to which services and supports the family will purchase from outside. Family choices of service‐mix clearly will aVect the outcomes of children with MR/DD. Economic studies of within‐family decisions began with a series of seminal papers by Gary Becker (Becker, 1991; Becker & Tomes, 1979). Becker was the first to articulate that families have finite endowments of money, time, energy, and other resources. Parents are faced with decisions about how to allocate these resources among all their competing uses: a child with a disability, other children, a spouse, careers, household production, and so on. Economists have studied how these decisions may be made jointly by couples and how the opportunity costs of the alternatives shape choices and outcomes. For example, one reason that it is more common for mothers to stay at home to care for young children is that the market wage of the father is usually higher. Thus, the value of the forgone opportunity required to stay at home, in terms of lost wages, is typically smaller for the mother than the father. The interesting empirical question is how child outcomes are aVected by a stay‐at‐home mother compared to the alternative. From this perspective, if parents decide not to provide a service to a child with MR/DD, we are never able to conclude that these parents are intentionally shortchanging the child or are unaware of the benefits of the service. While these might be explanations, it is plausible that the parents are making

134

Stephanie A. So

rational trade‐oVs and have chosen a diVerent way to balance all of their demands. For example, a single parent might be unable to take a child to speech therapy sessions because the money from the second and third jobs is needed at home to support the other four children. How to reconcile this choice must begin with understanding the relationships between family circumstances, service choice, and outcomes. A.

Family Economics Research Strategy: Families, Services, and Outcomes

How, then, can economics be used to improve outcomes for children with MR/DD? We propose to consider two logically separate problems. The first problem is probably more familiar: to find what combination of services most improve outcomes, for given family and child‐specific characteristics. Note that this is diYcult or impossible to address in many traditional randomized control trials. The empirical objective is to determine how any given combination of family circumstances (income level, access to health insurance, marital status, and so on) interact with the eVectiveness of various service combinations. It is necessary to turn to large, population‐based datasets in order to have enough variation and repetition of family structure to get statistically significant answers to these questions. Some services for children with MR/DD may only be productive if parents match those services with household time. For example, behavioral therapy, special help in reading, and some kinds of physical therapy benefit from reinforcement at home. We would expect such services to be more eVective when parents have more time and energy for them. It follows that two parent households, households with a nonworking mother, households with fewer children, without an unhealthy grandparent living within the home, or with a young and vigorous grandparent would benefit more. The common thread here is that such households would have a lower opportunity cost of time. Many services may not require as significant or sustained complementary family time inputs. Examples may include surgery and other types of acute medical services and therapies in clinic delivered by highly specialized providers. We would expect to see that family demographics have very little impact on outcomes in these cases. Services that give extra time to the family, in contrast, may have a greater impact on families with high opportunity costs of time. Day care services might free a parent to work or take care of household chores during the day so that more time is available when the child comes home. In contrast, these services would not have as great a benefit for children who live in a household with adequate parental time.

ECONOMIC PERSPECTIVES

135

There are other kinds of potential complementarities between services and family characteristics. Children with non‐English speaking parents might respond better to speech therapy when they have siblings in the house than when they do not. Educated parents might be more eVective at managing more complex service arrangements or be better able to find combinations of services that are mutually supportive. The point is that we should not expect a given treatment regimen for a child with MR/DD to have the same eVect, independent of that child’s family circumstances. Best practices require taking family circumstances into account when recommending a set of support services. At a formal level, this first problem calls for using family demographic data and child characteristics to predict health (including developmental) outcomes. The second problem might be less familiar: to find out what combination of services are actually chosen, for given family and child‐specific characteristics. At a formal level, this type of study calls for using family demographic data and child characteristics to predict the likelihood that a given service is chosen by a family for their child. As we have said before, simply knowing what we would like a parent to do for his or her child is not enough. In order to make policy that actually helps children with MR/DD, we also need to know how and why a parent chooses a service for the child. This is a pure social science question. Parents have many objectives, including providing for all children, saving for retirement, getting promoted, and many others, in addition to the welfare of their child with MR/DD. These parents, like other parents in some ways and unlike others in potentially identifiable ways, must find the best possible way to allocate their limited resources. Ignoring this balancing act is like attempting to hold back the sea. What is called for is policy ju‐jitsu. We must use the momentum of self‐interest to steer the parent in the right direction. DiVerent types of families face significantly diVerent costs for using or providing diVerent services. Most obviously, the opportunity cost of time will play a role. Given the current delivery system, many services require that a child be accompanied by an adult. This practice can be especially diYcult for a family in which all parents work outside the home. Such families also may have relatively little time to investigate treatment options or to try to get their child accepted for certain services. Thus, parents may not take advantage of what are considered to be highly beneficial programs for their children. Poor or less educated parents also may be less likely to come in for important follow‐up visits. Studies that examine the ways in which diVerent families systematically choose outcomes can be used to understand how to array the services so that families might more easily choose the services that will be best for their children.

136 B.

Stephanie A. So MR/DD Demographic and Population‐Based Data

For both of the analyses suggested in the previous sections, it is necessary to use large‐scale datasets to get meaningful answers. Large‐scale datasets, such as population‐based surveys or administrative data that capture all the activity in a given area, contain enough observations to allow the researcher to exploit the natural variation in the population’s characteristics. These sources of data are particularly important when the research question of interest is to examine the eVects of family characteristics, such as marital status, family size (number of siblings), maternal education, and so forth on child outcomes and service use. In longitudinal data, it may be possible to observe the actual changes in family structure as marriages form and dissolve or children are added. In cross‐sectional data, if samples are large enough, it may be possible to lessen unobserved heterogeneity by comparing similar types of families. In truth, the outcomes of children with MR/DD largely have been overlooked by economists as they have concentrated on the eVects of families and their circumstances on the educational, health, and later employment, wealth, and family outcomes of children generally. Thus, there is very little to report in the existing literature on MR/DD outcomes directly. Nonetheless, there exists a large literature about the roles that family and service characteristics play in determining outcomes (see, for example, the surveys contained in the Handbook for health economics; Cuyler & Newhouse, 2000). Economic studies have estimated the eVects of policy interventions like State Child Health Insurance Plans on early vaccinations and the eVects of numerous other programs, services, technologies, and family inputs, on educational and health outcomes (Joyce & Racine, 2005). For the reasons argued in the first part of the chapter, special attention has been given to the interactions of family characteristics with these policies and services in the production of these outcomes. The models, techniques, and approaches used in these studies may easily be applied to determine if similar relationships between families, services, and outcomes exist for children with MR/DD. For example, questions of the eVectiveness of parents’ use of therapies that aVect the outcomes of children with MR/DD are not structurally diVerent from studies that isolate the eVects of prenatal care on infant birthweight (Conway & Deb, 2005; Evans & Lien, 2005). C.

Preliminary Evidence of Family Effects on MR/DD Outcomes

In MR/DD research, the emergence of family and/or demographic variables as potential risk factors for outcomes or service use, in the epidemiological sense, opens the door between researchers with content expertise in the

ECONOMIC PERSPECTIVES

137

condition or the services and those with content expertise in economic and econometric research in education and health. For example, Urbano and Hodapp (submitted for publication) examine the correlates of divorce in families of children with Down syndrome (DS) compared to families of children with typically developing children. In their study, the presence of a child with DS was not clearly linked with a higher probability of divorce than in families with typically developing children. When divorce occurred in families of children with DS, however, the families tended to be in rural areas and formerly headed by fathers who had attained lower levels of education. Family poverty and low parental education have been linked with both poorer health outcomes and divorce. Urbano and Hodapp’s (submitted for publication) ongoing work involves asking richer questions of the data about family circumstances such as how rural outcomes and family income interact. One nice feature of their study design is that divorce is unlikely to cause DS to emerge in a child. As a result, they are plausibly able to compare the diVerences in risks of divorce in families of children with DS and families of children with typically developing children, by concentrating on the potential interactions of each child’s characteristics with these demographic features of location and education. One might ask how divorce aVects subsequent service use and outcomes compared to service use and outcomes of children of married parents, both across and within these groups. So, Urbano, and Hodapp (submitted for publication) analyzed population‐ based data on the inpatient outcomes of a birth cohort of young children with DS. Two distinctly diVerent subpopulations emerged with respect to health, some with repeated hospital use, and others with no hospital use after birth at all. Those who used many hospital services experienced hospitalizations primarily in their first year of life. Urbano and Hodapp’s (submitted for publication) data suggest that there may be an association between the timing of divorce and the presence of a child with DS. Further research is required to examine the hypothesis that children with DS who are healthy have no higher risk for divorce than typically developing children, but children with DS who are unhealthy, particularly those who live in areas or families that make it diYcult to access care or reap the benefits of care, are at higher risk for divorce. In a somewhat analogous economic study, Reichman, Corman, and Noonan (2003) investigated the eVect of poor infant health on the probability of divorce. This study built on a small literature, using the National Health Interview Survey, that showed that children with poor health had higher risks of parental divorce. However, in previous studies, infant’s health was assumed to be exogenous. Reichman et al.’s (2003) study investigated the possibility that there may be alternative explanations that caused both the infant’s poor health and divorce. For example, if the parents who divorce

138

Stephanie A. So

tend not to be interested in investing in long‐run propositions, then it may be that parents who tend to end their marriages are also parents who tend not to use prenatal care. In that case, the benefits to divorce and child health are systematically diVerent across types of parents and there would be a bias in previously estimated eVects of divorce on health. Reichman et al.’s study used instrumental variables to measure infant health in a new dataset, the Fragile Families and Child Wellbeing Study. They found that, even after addressing the potential biases in the data, the finding of poor infant health on increased parental divorce was robust. Poor child health significantly reduced the probability that the parents would stay together for parents of all socioeconomic status, with the largest eVect for parents with the lowest socioeconomic status. If the children with severe health problems from birth are at higher risk for divorce compared to both typically developing children and healthy children with DS, particularly if they are in low income or in rural areas, as these studies might suggest, then children with DS who are born with congenital health problems may suVer from a quadruple dose of poor health, family instability, low income, and whatever negative eVects that DS itself may confer. So many diYculties in early life are especially troubling given the cumulative eVects, in economic terms, of early failures to accumulate human capital (Grossman, 1972). Compared to both typically developing children and children with DS who do not have health problems, the downstream outcomes for these children may show dramatic disparities as they age that may predispose these children to have relatively very poor adult outcomes. In these as in any economic studies, we will need to separate the direct eVects of a policy, service, or family characteristic, from the indirect eVects in the estimation strategy. First, we will want to estimate the direct and indirect eVects of family circumstances on MR/DD outcomes. Second, we will want to estimate the direct and indirect eVects of family circumstances on MR/DD service choice. The combined results potentially could shape the policy debate dramatically. For this reason and others, it is important to be sure that the estimation strategies are sound. We turn next to a brief discussion of econometrics and the most common problems faced in the economic analysis of large datasets. V.

ECONOMETRIC MODELS AND ESTIMATION TECHNIQUES

In trying to understand the eVects of family, economic, and child‐specific variables on outcomes and services, economic studies typically start with a regression framework. To produce consistent and reliably estimated eVects,

ECONOMIC PERSPECTIVES

139

the regression model must be correctly specified and the assumptions of the regression model should be satisfied. However, it is rarely the case that any economic model is estimated as a simple linear regression. Real world choices are rarely that straightforward; in addition, the forms of the data are never ideal. Thus, diYculties arise, both conceptually and empirically, when the research aims to move beyond documenting correlations in the data to try to infer the causal explanations between independent variables and outcomes. Many problems arise from the fact that the data are not generated within an experimental framework. People are not randomly assigned to conditions; instead, they play some role in choosing their own conditions. Economists, sensitized to the possibilities of choice, will often treat income, marital status, education levels, and even family size as choice variables. This conceptual approach raises the possibility that there are underlying variables, not directly observed in the data, that jointly aVect outcomes, services, and observed characteristics. Broadly speaking, designers of estimation strategies in large social science datasets must be vigilant about possibilities that behavior and choices by agents will cause the assumptions in the regression model to be violated. In response to these challenges, econometricians have developed sophisticated techniques to deal with the threats to validity caused by potential model misspecification. In any given model specification, certain assumptions are included about the underlying relationship. Choice of variables and the ways in which they enter the model have an impact on the relationships that it is possible to explore (or miss) and, of course, on the eventual interpretation of the estimates themselves. The functional forms and the measures used impose conditions that may or may not be appropriate. For example, a study that uses average income in the community as a proxy for household income is not able to ask whether a household in a community with greater variation in income (i.e., inequities) will experience diVerent outcomes than a household in a more homogeneous neighborhood. (This specification may be appropriate, however, for certain kinds of research question where only the average income matters.) Similarly, economists are sensitive to the kinds of measures that are used for household income or wealth. Depending on the question, annual earned income, savings, program benefits (such as social security or retirement benefits), insurance status, access to loans in credit markets, and even perceptions of future income may each be important determinants of decisions to purchase services and outcomes. For example, for most typically developing children, sending a child to a public versus a private college is an objective that shapes and is shaped by many earlier family resource decisions. Why would the prospect of being a caregiver of a child with special needs, late in life, do any less?

140

Stephanie A. So

Later, we discuss three other problems with model specification that are of concern because they are known to bias estimates or change the properties of the estimators. Familiarity with the ways that these problems are addressed by econometric techniques will help in the design of future large‐scale studies that examine MR/DD outcomes. Many of these statistical issues are complex and a full review will not be attempted here. A.

Omitted Variables

Problems arise with inference from regression estimates when the model does not include the correct set of variables. Including irrelevant variables reduces the eYciency of the estimates. However, excluding relevant variables introduces potentially more serious problems. In the case of an omitted independent variable, simple estimates of eVects will be biased because the expected value of the disturbance term varies with the omitted independent variable and is not constant. The situation is exacerbated when an included independent variable, such as the one whose eVect is of policy relevance, is correlated with the variable that has been left out. In that case, the estimated eVect of the included variable will capture some of the impact of the omitted variable. The size and direction of the bias depend on the nature of the relationship between the omitted and included variables. The most obvious example of this result (and, unfortunately, one that was prevalent in the early literatures) is when researchers asked the eVect of race on outcomes without including measures of socioeconomic status, such as income. In time periods or regions where race and income were strongly correlated, patterns of outcomes (or service use) that might be strongly aVected by income and family circumstances were instead attributed to race. Of course, this greatly alters the policy implications. How marital status aVects family choices sometimes can be interpreted in the same way. The direct eVect of marriage on outcomes may be reduced once family investments of time are included. For example, certain outcomes of children of single parents who receive time through another means, such as a grandparent or a reading clinic, may be similar to outcomes of children in households with two parents. On the other hand, if diluted income is the causal link in the eVects of divorce on outcomes, then providing financial support for services to children in all low‐income families, whether single or two‐parent households, may be well advised. Researchers who use family structure and demographic variables to investigate within‐household decisions must be particularly concerned about omitted variables related to family characteristics. If they are not, they risk attributing eVects to family structure that might lead the field to believe there

ECONOMIC PERSPECTIVES

141

are strong reasons to support one family structure over another. In fact, much of the economic research on families and educational outcomes has evolved through those stages. Whereas there was great early interest in family size, birth order, and marital status, the eVect sizes of many of these structural characteristics have been adjusted downward, as studies have used increasingly more sophisticated econometric techniques (Nechyba, McEwan, & Older‐Aguilar, 1999). This finding has been interpreted as a reflection of the idea that all families try their best, in their own way, to improve child outcomes (Grossman, 1972). As results such as these become available, new studies focus more intensely on the particular incentives or constraints of potential causal pathways. While the results may be less dramatic, the more precisely estimated eVects can be used more meaningfully to inform policy about which services to deliver to whom. In economic studies, the best way to avoid the identification errors associated with omitted variables is to use economic reasoning to determine the best set of variables to include. Variables are preselected by articulating theoretical relationships. A variable that happens to have large explanatory power, but cannot be justified on the basis of theory, is looked on with suspicion: including it when it is a proxy for some relevant omitted variable will not automatically solve the statistical problem of the omitted variable and may introduce new problems. In studies that involve service use and outcomes for a particular service, people with expertise in all content areas should collaborate on determining the set of variables to be examined. How economic studies deal with omitted variables depends on the consequences that they may have, statistically, within the regression framework. For example, sometimes the omitted variable is correlated with an explanatory variable (as in the previous example) and with the outcome. In this case, the simple ordinary least squares estimator is biased even asymptotically. This is because the process of assigning credit to the independent variable for variation in the dependent variable always erroneously attributes some of the variation of the dependent variable. A common approach to this contemporaneously correlated problem among variables is to use the instrumental variables approach. If a new independent variable, called an instrument, can be found that is correlated with the original variable and yet contemporaneously uncorrelated with the disturbance, the estimate will be consistent. Economic studies have grappled with this issue as they have attempted to measure the direct eVect of maternal education on health services use. The concern in these studies is that there is an omitted characteristic about a mother (say, innate ability) that causes her to choose her level of education

142

Stephanie A. So

and also the level of services that she demands for her child. Currie and Moretti (2003) use instruments for maternal education (perhaps related to the time period in which she was merely a student) that are unrelated to the level of services that she consumes as a mother. A fairly robust finding from these studies is that the level of maternal education itself continues to play a significant role in child health outcomes. These findings are encouraging for those who support educational interventions to mothers, regardless of previous educational backgrounds. B.

Endogeneity

Endogeneity is another common model misspecification problem and is a special case of the correlation between variables problem described previously. Endogeneity occurs whenever variables are determined simultaneously. For example, in cross‐sectional data, families are observed who experience both poverty and divorce. Divorced families experience lower income as a direct consequence of divorce; on the other hand, greater stress from low income increases the probability of divorce. In statistical terms, a change in the disturbance term changes all of these variables simultaneously, which leads to identification issues. As a consequence, the ordinary least squares estimator is biased, even asymptotically. When the assumptions of the linear regression model are violated in this way, an alternative estimator or estimation strategy is necessary. As an alternative to finding longitudinal data in which to observe income before and after divorce, econometric studies typically employ maximum likelihood estimation or two‐stage least squares. The two‐stage least squares estimator is the most widely used and easily understood because it is a special case of the instrumental variables approach already described. For example, if the research question is how income might predict divorce, then one would like to find an instrument for income that is correlated with income but not with the decision of that family to divorce. A two‐stage least squares approach might use the predicted income of the household, based on all of its other demographic characteristics, as an instrument for actual income. In removing the household’s actual income as the explanatory variable, and using predicted income (for a typical family in that situation, but not that particular family) in its place, the explanatory income variable is no longer tied to the disturbance on that family’s divorce term. As in the ordinary instrumental variables case above, the predicted value for income is a good instrument because it is highly correlated with the variable (actual family income) for which it is acting as an instrument. However, there is no reason to expect that predicted income is itself correlated with the family’s decision to divorce.

ECONOMIC PERSPECTIVES

C.

143

Selection Bias

One of the most serious problems encountered in statistical analysis is selection bias. When groups are not randomly assigned, individuals may self‐ select. If a patient seeks care, then she has revealed herself to be willing to do at least some of the work. Selection bias may be at work whenever it is possible that the fact that one observes people in a group may indicate that they are not representative of the population at large. Exit polls may not reflect the popular sentiment (and vice versa). Average treatment eVects in controlled samples may not be applicable to the population that chooses to use the service. Labor studies of the wages and hours data that are collected from people who work would not be accurate reflections of the hours and wages people would be willing to accept, unless the analysis takes into account the responses of people who would like to work but failed to find employment. Econometric studies often deal with this problem by explicitly modeling the choice of self‐selection. In doing this, the study is designed to use as much of the data as are available. Formally, this means that the first stage of analysis is to model the probability that a person will choose to participate in the group. As an example, the desire to work (known as the work participation choice) is modeled before wages and hours are ever considered. In order to model work participation, data are used on all potential workers, both employed and unemployed. The models diVerentiate between those people who choose to be unemployed and those people who are involuntarily unemployed. Similar studies were proposed earlier in this chapter, with respect to the service choices that people make. Many surveys ask about unmet need for services. Using economics to understand selection bias issues, more knowledge can be gained to understand if people do not use services for insurance reasons, for lack of availability reasons, or from their own readings of their particular circumstances. To restate the statistical lesson of social science selection bias more generally, epidemiologists know how important it is to think about the interpretation of average treatment eVects in out‐of‐sample populations. The same is true in economic studies of health services use for similar, albeit social science reasons. The eVects of service utilization may not translate to out‐of‐sample populations because they may be mitigated by the very characteristics that caused people not to choose the services in the first place. However, newly available population‐based data can help to alleviate concerns about selection bias. With careful study design and statistical interpretation, combining economic and epidemiological research may yield insights not only about the eVectiveness of services, but about how people might be helped to choose the services that would most benefit them.

144

Stephanie A. So

A final note concerns the econometric issues when studies select subjects, knowingly, on the basis of the value of a variable of interest. For example, a study may examine the outcomes only of families most at risk. Suppose a study selects only families who live in poverty. In statistical terms, the data come from a truncated distribution because observations from the upper income tail in the distribution of error terms are eliminated systematically. Not only is the expected value of the error term nonzero, it will also vary from observation to observation. The solution for dealing with truncated or censored data usually involves maximum likelihood techniques. VI.

CONCLUSIONS

In order to improve child outcomes, service guidelines and best practices must be based on far more than the individual child’s disabilities, health, preferences, or other characteristics. Even the best purely scientific (albeit not social‐scientific) professional advice, tailor‐made to the child’s own conditions or severity level, may not be suYcient because the child is not the one who makes the service choices. Parents and caregivers do. As a result, parents’ decisions must be understood in the broader context of families’ constraints, needs, and choices between competing alternatives. Instead of viewing each family’s individual circumstances as a reason to make exceptions and work around a fixed set of recommendations, those involved in policy setting and service delivery eventually must accept the reality that diVerent families are best served diVerently. Thus, the goal of research should be to understand the ways in which care should be arranged to provide, consistently, the best sets of services customized to the family’s circumstances. Let us be perfectly clear. Even in the special case where one cares about no one else’s outcomes except the child’s, this is the only practical way to think about delivering services. Economics provides a systematic framework in which to think about these decisions across diverse family conditions. Just as economists are accustomed to studying how diVerent families decide how, when, and where to work, which neighborhoods to move to and how much to pay in taxes, whether to obtain private health insurance or accept some form of governmental support, or how much to accumulate in savings/dissavings now that later will aVect their well‐being in retirement, so too can researchers in MR/DD use economics to understand the interactions between families, disability and health services, and the short‐ and long‐term outcomes of all involved. The core intuition and starting point for this chapter is that services cannot be useful if they are not chosen voluntarily. Even mandates will not

ECONOMIC PERSPECTIVES

145

work perfectly in the face of unintended consequences. Obviously, families try to do the best that they can. The overwhelming evidence from economic research is that families do try to do the best they can for their children, but in diVerent ways. This is exactly what economics predicts. True economic research on outcomes and services research deepens our understanding of why certain outcomes and services are chosen. The results inform how incentives (policy) may improve these decisions. It should be obvious now that economic research goes far beyond counting the costs and benefits, in a strictly descriptive fashion, to include careful discussions of how families in general, and families of children with special needs specifically, aVect their children’s outcomes. Economic studies of actual service choice are important because these expose not only the decisions of the families, but also of the providers. Service delivery systems that do not find the clients who will be most benefited by them are an indication that the provision of services must be improved. When services are not chosen, it is always appropriate to think about how to change the incentives that face families in order to help the families to choose more of the services. In other words, change does not always depend on having to change the intrinsic characteristics of the clients themselves. Alternatively, services that are oversubscribed (as measured by long waiting lists and so on) indicate that rebalancing is needed by the system, whether through greater incentives to existing providers to provide more care, increases in the numbers of providers allowed in the system, changes in prices, or some other mechanism. Either way, understanding how families choose which services will help explain how to provide the most appropriate services to maximize outcomes. Whether studying families, providers, or anyone else involved in the care of children with MR/DD, econometric techniques may be translated easily to empirical studies of outcomes and service use. Statistical problems arise when the data are nonexperimental and when values and variables are related because they are linked by agents’ fundamentally consistent (nonindependent) behaviors, attitudes, barriers, environments, and past decisions. However, many sophisticated techniques have been developed in response to the statistical challenges that these underlying behaviors pose to the analysis. The empirical literature already contains a huge installed base of related evidence and hypothesis tests that researchers may use to estimate the true eVects of families’ incentives on service use and outcomes. As the economic literature has evolved, it has sought more and more content knowledge from people who know about the actual services and decision makers. As the empirical literature in MR/DD outcomes evolves, we hope that its researchers will seek more and more interaction with economists and econometricians.

146

Stephanie A. So REFERENCES

Anderson, P. M., Butcher, K. F., & Levine, P. B. (2003). Maternal employment and overweight children. Journal of Health Economics, 22, 477–504. Becker, G. S. (1991). A treatise on the family. Cambridge, MA: Harvard University Press. Becker, G. S., & Tomes, N. (1979). An equilibrium theory of the distribution of income and intergenerational mobility. Journal of Political Economy, 87, 1153–1189. Chou, S. Y., Grossman, M., & SaVer, H. (2004). An economic analysis of adult obesity: Results from the behavioral risk factor surveillance system. Journal of Health Economic, 23, 565–587. Conway, K. S., & Deb, P. (2005). Is prenatal care really ineVective? Or, is the ‘Devil’ in the distribution? Journal of Health Economics, 24, 489–513. Currie, J., & Moretti, E. (2003). Mother’s education and the intergenerational transmission of human capital: Evidence from college openings. Quarterly Journal of Economics, VCXVIII, 1495–1532. Cuyler, A. J., & Newhouse, J. P. (2000). Handbook of health economics. Amsterdam: North‐Holland. Dow, W., Philipson, T., & Sala‐i‐Martin, X. (1999). Longevity complementarities under competing risks. American Economic Review, 89, 1358–1371. Ensor, T., & Cooper, S. (2004). Review article: Overcoming barriers to health services on the demand side. Health Policy and Planning, 19, 69–79. Evans, W. N., & Lien, D. S. (2005). The benefits of prenatal care: Evidence from the PAT bus strike. Journal of Econometrics, 125, 207–239. Gaviria, A., & Raphael, S. (1997). School‐based peer eVects and juvenile behavior. San Diego, CA: University of California. Grossman, M. (1972). On the concept of health capital and the demand for health. Journal of Political Economy, 80, 223–255. Hanushek, E. A. (1996). School resource and student performance. In G. Burtless (Ed.), Does money matter? The eVect of school resources on student achievement and adult success Washington, DC: Brookings Institution Press. Hanushek, E. A. (1997). Assessing the eVect of school resources on student performance: An update. Educational Evaluation and Policy Analysis, 19, 141–164. Joyce, T., & Racine, A. (2005). Chip shots: Association between the state children’s health insurance programs and immunization rates. Pediatrics, 115, e526–e534. Nechyba, T., McEwan, P., & Older‐Aguilar, D. (1999). The impact of family and community resources on student outcomes: An assessment of the international literature with implications for New Zealand. Wellington, New Zealand: Ministry of Education. Peltzman, S. (1975). The eVects of automobile safety regulation. Journal of Political Economy, 83, 677–725. Peltzman, S. (2002). OVsetting behavior and medical breakthroughs. University of Chicago (Working Paper No. 177). Reichman, N. E., Corman, H., & Noonan, K. (2003). EVects of child health on parents’ relationship status. National Bureau of Economic Research (Working Paper No. 9610). Robertson, D., & Symons, J. (1996). Do peer groups matter? Peer group versus schooling eVects on academic attainment. Discussion Paper No. 311. London School of Economics Centre for Economic Performance, London, England. Retrieved from: http://cep.lse.ac.uk/pubs/ download/DP0311.pdf So, S. A. (2002). Jointly determined labor supply for married couples. A review of the theory on the added worker eVect. Disability Research Institute. So, S. A., Urbano, R. C., & Hodapp, R. M. (submitted for publication). Hospitalizations for infants and young children with Down syndrome: Evidence from person‐records from a statewide administrative database. Urbano, R. C., & Hodapp, R. M. (submitted for publication). Divorce in families of children with Down syndrome: A population‐based study.

Section III Populations

This page intentionally left blank

Public Health Impact: Metropolitan Atlanta Developmental Disabilities Surveillance Program* RACHEL NONKIN AVCHEN, TANYA KARAPURKAR BHASIN, KIM VAN NAARDEN BRAUN, AND MARSHALYN YEARGIN‐ALLSOPP CENTERS FOR DISEASE CONTROL AND PREVENTION, NATIONAL CENTER ON BIRTH DEFECTS AND DEVELOPMENTAL DISABILITIES, ATLANTA, GEORGIA

I.

INTRODUCTION

Sixty years ago, on July 1, 1946, CDC was founded in an eVort to provide service and consultation to combat communicable diseases throughout the United States. At its inception, CDC was an abbreviation for the Communicable Disease Center, and there were fewer than 400 employees, most of whom were engineers and entomologists with expertise in malaria. Today, the acronym CDC represents the Centers for Disease Control and Prevention and the agency, as one of the 13 major operating components of the Department of Health and Human Services, is comprised of over 15,000 employees from a wide spectrum of disciplines. CDC continues to head public health eVorts to prevent and control infectious and chronic diseases, injuries, workplace hazards, disabilities, and environmental health threats through surveillance and epidemiologic research. CDC has consistently been at the forefront of investigating birth defects and developmental disabilities with almost 40 years of experience in surveillance and epidemiologic research in this area. The Metropolitan Atlanta Congenital Defects Program (MACDP) was initiated in 1967 as the first population‐based birth defects surveillance system in the United States, but no comparable surveillance system existed for developmental disabilities. In 1979, a request was made of CDC for data regarding the prevalence of *The findings and conclusions in this chapter are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

149

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33007-8

150

Rachel Nonkin Avchen et al.

mental retardation and cerebral palsy and a pilot study was planned and implemented. After the success of this small pilot study, which aimed to quantify the prevalence of severe mental retardation in school‐age children in one county of metropolitan Atlanta, the Metropolitan Atlanta Developmental Disabilities Study (MADDS) was launched in 1984. This was the first United States prevalence study of multiple developmental disabilities. The MADDS methodology evolved into the Metropolitan Atlanta Developmental Disabilities Surveillance Program (MADDSP) beginning in 1991. MACDP and MADDSP serve as the gold standard models for monitoring birth defects and developmental disabilities in the metropolitan Atlanta area and across the United States. The public health framework for developmental disabilities activities at CDC has three core components: surveillance, epidemiology, and prevention (Fig. 1). Surveillance represents the ongoing ability for monitoring and routine reporting of prevalence for select developmental disabilities as well as for examining temporal trends in the prevalence of these conditions. Developmental disabilities surveillance activities provide a population‐based case series of children with select conditions for the epidemiologic investigation of risk or protective factors and clues to the causes of conditions. Subsequently, findings from epidemiologic studies can help to facilitate

FIG. 1. Three core components of the public health framework for developmental disabilities activities at CDC.

PUBLIC HEALTH IMPACT

151

planning and implementation of health intervention and prevention initiatives by influencing changes in health education, policy, and practice. Surveillance activities can capture changes in health policy practice, illustrating the cyclical influence from one component of the model to the next. The ultimate goals of these activities are to decrease the occurrence and improve both short‐ and long‐term consequences of developmental disabilities. This brief history of developmental disabilities activities within CDC is intended to provide background on the beginnings of MADDS and MADDSP. The next section will detail the methods for developmental disabilities surveillance used at CDC, followed by a detailed compilation of results from surveillance reports and epidemiologic studies on developmental disabilities using MADDS and MADDSP data. The chapter concludes with a discussion of the public health impact of MADDS and MADDSP on the epidemiology and public health practice of developmental disabilities.

II. A.

BACKGROUND OF DEVELOPMENTAL DISABILITIES SURVEILLANCE IN METROPOLITAN ATLANTA

Metropolitan Atlanta Developmental Disabilities Study

MADDS was initiated in 1984 due to growing concern about the lack of prevalence data on developmental disabilities in the United States. It was the first population‐based epidemiologic program to monitor multiple developmental disabilities in school‐aged children conducted in the United States. The study was funded by the Agency for Toxic Substances and Disease Registry (ATSDR) through a cooperative agreement between the CDC and the Georgia Department of Human Resources. The goal of MADDS was to devise methods for ascertaining children with select developmental disabilities so that prevalence of these disabilities could be systematically monitored. The study aimed to establish the prevalence of five developmental disabilities in 10‐year‐old children: cerebral palsy, epilepsy, hearing impairment, mental retardation, and visual impairment. The other goals of this study were to relate the five developmental disabilities to service needs in the metropolitan area and to generate hypotheses about risk factors for these conditions. The latter goal of this study will be discussed in greater detail in Section V of this chapter. The MADDS study area encompassed the five metropolitan Atlanta counties, which included Clayton, Cobb, DeKalb, Fulton, and Gwinnett. The study population was composed of 10‐year‐old children who (1) were born in the study area between 1975 through 1977 and still resided there between 1985 through 1987, (2) were born in the study area between 1975

152

Rachel Nonkin Avchen et al.

through 1977 but did not reside there during the study years (1985–1987), or (3) were not born in the study area but resided there during the study years. Ten‐year‐old children were chosen as the target population for MADDS because developmental disabilities are generally recognized by the time a child is school‐aged, including mild disabilities. Second, from the literature, the prevalence of mental retardation was thought to peak by 10 years of age (Yeargin‐Allsopp, Murphy, Oakley, & Sikes, 1992). 1. CASE DEFINITIONS

Cerebral palsy was defined as a group of nonprogressive disorders occurring in young children in which abnormalities of the brain cause impairment of motor function. The impairment of motor function may result in paresis, involuntary movement, or incoordination. Motor disorders that are transient, disorders that result from progressive disease of the brain, and motor disorders due to spinal cord abnormalities are not included (Murphy, Yeargin‐Allsopp, Decoufle, & Drews, 1993). Records were reviewed for any indication of cerebral palsy including cerebral palsy from physical findings (e.g., spasticity, athetoid movements). MADDS included children with congenital and acquired cases of cerebral palsy and central nervous system disorders such as hydrocephaly and microcephaly. For the purpose of the study, children with myelomeningocele were excluded unless there was evidence of another neurologic process that resulted in physical findings consistent with a diagnosis of cerebral palsy. A child was classified as having cerebral palsy by a qualified professional based on the description of the physical findings that was available in the child’s record. In cases where a determination could not be made a classification of cerebral palsy, not otherwise stated was given. The classification of cerebral palsy subtypes was established based on the Little Club Classification (MacKeith, 1959). Epilepsy was defined as two or more epileptic seizures diagnosed by a physician. Clusters of seizures (two or more) that occurred within a 24‐hour period were considered a single seizure. Children who only had simple febrile seizures were not included as having epilepsy. Children with febrile status epileptics were included as having epilepsy only if they otherwise qualified on the basis of afebrile seizures. Hearing impairment was defined as a bilateral pure‐tone hearing loss that averaged 40 decibels (dB) or worse unaided in the better ear at frequencies of 500, 1000, or 2000 Hz (normal speech range; Roeser, 1988). The diagnosis of a hearing loss was accepted only from a physician or professional in the field of audiology. Mental retardation was defined as an intelligent quotient (IQ) of 70 or less on the most recent psychometric test performed by a psychometrist;

PUBLIC HEALTH IMPACT

153

this included children with sensory problems such as blindness or deafness. It was reported that most children (85%) ascertained by this survey had a psychometric test (Yeargin‐Allsopp et al., 1992). If a child with Down syndrome did not have an IQ score they were considered to have mental retardation. While many facilities use the definition provided by the American Association of Mental Deficiency which includes adaptive functioning, the definition used by MADDS was based on IQ alone. Adaptive functioning was not included for case definition because inconsistencies in the types and uses of adaptive instruments among the diVerent school systems was evident (Yeargin‐Allsopp et al., 1992). Visual impairment (legal blindness) was defined as follows: (1) a measured visual acuity of 20/200 or worse in the better eye with correction, (2) a description of visual acuity that reflected 20/200 or worse (e.g., light perception only), or (3) a statement by a trained person, such as ophthalmologist or optometrist, that the child was ‘‘blind.’’ This definition represents children with legal blindness and not children with milder forms of visual impairment. 2. METHODOLOGY

Children ascertained by MADDS were identified using a multiple source approach by reviewing records from educational, medical, and social service providers for children with developmental disabilities. All records from any participating source were reviewed and pertinent information was abstracted to ensure a complete developmental and health profile for each child. In addition to information obtained from source records, maternal/ child demographic data were obtained from the child’s birth certificate. Linkage with MACDP also allowed MADDS investigators to collect additional information on structural and chromosomal anomalies. Educational sources used for this study consisted of nine public school systems and psychoeducational and state school programs. The Department of Education sources were used due to the mandate by the Federal Education for All Handicapped Children Act (PL 94–142) to identify and educate all children with developmental disabilities (US Congress, 2005). While this act mandates the identification of all children with special educational needs, the investigators of this study acknowledged that children with milder forms of the disabilities, such as mild cerebral palsy, might not receive special educational services through the public school systems. Other sources used to ascertain children for MADDS included medical facilities such as hospitals specialized in treating children with developmental disabilities, pediatric clinics, Georgia Department of Human Resources, and other public and private agencies. Children were identified from one of these facilities using relevant disability codes from the International

154

Rachel Nonkin Avchen et al.

Classification of Diseases, 9th edition‐clinical modification (Puckett, 1993). The records were reviewed for any child with a flagged ICD‐code, and abstractors collected all behavioral, developmental, and pertinent medical information from the child’s source file. MADDS investigators found a majority (95%) of the children ascertained for the study were identified at an educational source. This finding stressed the importance of educational source records for identifying children with developmental disabilities. Yet, information from medical and clinical sources were also critical in many instances to determine final case status, particularly for cerebral palsy and epilepsy. The findings from MADDS further illustrated that surveillance of developmental disabilities in school‐aged children was possible. B.

Metropolitan Atlanta Developmental Disabilities Surveillance Program

Due to the success of MADDS, CDC initiated MADDSP in 1991. MADDSP is an active surveillance program that was established to monitor the occurrence of cerebral palsy, hearing loss (previously termed hearing impairment), mental retardation, and vision impairment in children 3–10 years of age in the five county metropolitan Atlanta area (Clayton, Cobb, DeKalb, Fulton, and Gwinnett). In 1996, autism spectrum disorders were added to MADDSP. While epilepsy was included in MADDS, it was not included in MADDSP because surveillance of epilepsy was deemed too resource‐intensive; it required record review at over 20 electroencephalogram (EEG) laboratories. Therefore, surveillance for epilepsy did not facilitate timely surveillance of the other disabilities. To date, MADDSP estimates the prevalence of the five disorders in the metropolitan Atlanta area and serves as a model program for federally funded state programs monitoring developmental disabilities across the United States. The main objective of MADDSP is to provide regular and systematic monitoring of prevalence for select developmental disabilities according to various demographic characteristics of children and their mothers. MADDSP data were used to measure the progress of Healthy People 2000 prevention of mental retardation objectives, and the surveillance data will continue to be used to evaluate the Healthy People 2010 objectives for the prevention of mental retardation and early identification of children with autism spectrum disorders. MADDSP uses a similar methodology to that employed by MADDS. MADDSP ascertains children who have one or more of the five developmental disabilities. The age cohort was expanded from only 10‐year‐old children to include children 3–10 years of age. This age range was originally

PUBLIC HEALTH IMPACT

155

chosen because the lower age boundary corresponds to the beginning of the age span covered by Part B of the Individuals with Disabilities Education Act (IDEA) (US Congress, 2005), and the upper age boundary is consistent with the age at which most children served under IDEA should have received special education services (Boyle et al., 1996). However, as of the 1996 surveillance year, only 8‐year‐old children are actively monitored by MADDSP. This decision was made in an eVort to ascertain the most complete number of children with one of the five developmental disabilities and to maximize the timeliness of surveillance reports. Further, previous surveillance data indicated that prevalence rates were stable by age 8 for all the disabilities monitored. MADDS demonstrated that a multiple source methodology was critical for obtaining comprehensive information on children with developmental disabilities, and for this reason, MADDSP also reviews records from nine public school systems, psychoeducational programs, medical facilities, pediatric clinics, Georgia Department of Human Resources, and other public and private agencies serving children with developmental disabilities. Since 1991, children identified through MADDSP have also been linked to MACDP and birth certificate data. These linkages provide additional information regarding congenital and structural abnormalities as well as birth and maternal characteristics. MADDSP method relies on the ability to use multiple sources of administrative data to determine the prevalence of the disabilities and to build on surveillance data to identify hypotheses that can be tested using case‐ control methods. Because there was no examination of children for MADDS or MADDSP, having access to records at sources where large numbers of children are evaluated in the community for developmental disabilities is critical for complete ascertainment of children with developmental disabilities (to the extent possible). For MADDS and MADDSP, the data sources mentioned above are divided into two categories: (1) educational sources: public school special educational and psychoeducational sources and (2) health sources: medical facilities, pediatric clinics, Georgia Department of Human Resources, and other public and private agencies serving children with developmental disabilities with educational sources being the major source of cases. Like MADDS, a similar percentage of cases were identified from educational data (Boyle et al., 1996; Yeargin‐Allsopp et al., 1992). Further, approximately 6% of cases were identified uniquely through health sources for MADDSP (Bhasin, Brocksen, Nonkin Avchen, & Van Naarden Braun, 2006). These data underscore the importance of access to educational records for obtaining as complete a count as possible of children with developmental disabilities.

156

Rachel Nonkin Avchen et al.

1. CASE DEFINITIONS

The case definitions for MADDSP were refined where appropriate subsequent to MADDS and were derived from national standards for each of the disabilities. The current definitions are as follows. Autism spectrum disorders are defined as a constellation of behaviors indicating social, communicative, and behavioral impairment or abnormalities. The essential features of autism spectrum disorders are: (1) impaired reciprocal social interactions, (2) delayed or unusual communication styles, and (3) restricted or repetitive behavior patterns. Confirmed autism spectrum disorder cases include children that display behaviors (as described on a comprehensive evaluation by a qualified professional) that are consistent with the diagnostic criteria listed in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM‐IV‐TR) for any of the following conditions: autistic disorder, pervasive developmental disorder‐not otherwise specified (PDD‐NOS, including atypical autism), or Asperger’s disorder; child disintegrative disorder and Rett’s disorder are not currently included as conditions monitored by MADDSP. Cerebral palsy continues to be defined as a group of nonprogressive, but often changing, motor impairment syndromes secondary to lesions or anomalies of the brain arising at any time during brain development, and motor impairment function may result in paresis, involuntary movement, or incoordination. Like MADDS, children with postnatally acquired cerebral palsy are eligible as cases in MADDSP, but children with motor disorders that are transient, result from progressive disease of the brain, or are due to spinal cord abnormalities/injuries are not considered cerebral palsy cases for MADDSP. Confirmed cases of cerebral palsy include children diagnosed as having cerebral palsy from a qualified physician, or children identified by another qualified professional as having this disability on the basis of physical findings noted in source records. MADDSP considers physicians, physical therapists, occupational therapists, nurse practitioners, or physician’s assistants to be a qualified professional. Final case determination is made by medical staV aYliated with MADDSP. Hearing loss remains defined as a measured, bilateral, pure‐tone hearing loss at frequencies of 500, 1000, and 2000 Hz averaging 40 dB or more, unaided, in the better ear. In the absence of a measured, bilateral hearing loss, children meet the case definition if their source records include a description, by a licensed or certified audiologist or qualified physician, of a hearing loss of 40 dB or more in the better ear (e.g., profound sensorineural hearing loss). Severity is defined on the basis of hearing impairment levels as measured in the better ear as follows: (1) moderate, a hearing loss of

157

PUBLIC HEALTH IMPACT

40–64 dB; (2) severe, a hearing loss of 65–84 dB; and (3) profound, a hearing loss of 85 dB. Mental retardation is defined using only an IQ of less than or equal to 70 as was the criterion for MADDS. In the absence of an IQ score and in the context of testing, MADDSP accepts a written statement by a psychometrist that a child’s intellectual functioning falls within the range for severe or profound mental retardation for case ascertainment. The severity of mental retardation is defined according to the following International Classification of Disease, Ninth Edition, Clinical Modification (1993) categories: (1) mild, an IQ ¼ 50–70; (2) moderate, an IQ ¼ 35–49; (3) severe, an IQ ¼ 20–34; and (4) profound, IQ < 20. Vision impairment is defined as a measured visual acuity of 20/70 or worse, with correction, in the better eye. In the absence of a measured visual acuity, a child is considered a case if a source record includes a functional description, by a qualified physician or vision professional, of visual acuity of 20/70 or worse (e.g., light perception only) or a statement by a qualified physician or vision professional that the child has low vision or blindness. Severity of visual impairment is defined using both the WHO categories for low vision (20/70 to better than 20/400) and blindness (worse than 20/400) and the United States categories for low vision (20/70 to better than 20/200) and blindness (worse than 20/200). C.

MADDS Follow‐Up (MADDSFU) Study of Young Adults

The MADDS Follow‐Up Study of Young Adults (MADDS Follow‐Up) was conducted from 1997 to 2000 and followed a subset of children originally identified by MADDS. At the time of the follow‐up study the population was 21–25 years of age. The MADDS Follow‐Up consisted of an extensive in‐person or telephone interview that obtained information about daily functioning, competitive employment, educational attainment, living arrangements, financial and transportation assistance, living arrangements, marital status, and participation in daily leisure activities. The MADDS Follow‐Up provided an unprecedented opportunity to examine the outcomes of young adults with developmental disabilities identified in childhood using various factors and outcomes related to transition into young adulthood. III.

PREVALENCE ESTIMATES

The prevalence estimates derived from MADDS and MADDSP are considered administrative prevalence estimates. Administrative prevalence in these systems reflects ascertainment of cases from data sources that provide

158

Rachel Nonkin Avchen et al.

services to children with developmental disabilities such as educational and clinical sources. Data are collected by actively reviewing such administrative records for pertinent information to inform case status. This surveillance methodology does not rely on formal reporting of conditions to the state health department; in fact, most developmental disabilities are not reportable conditions. Instead, this methodology extends beyond special education data to incorporate necessary information from other medical and service provisions. This methodology provides a closer approximation of the true population prevalence than does reliance on one administrative database alone. Only a small percentage of children, particularly those with milder disabilities, are thought not to be identified by MADDS or MADDSP because these children may not be receiving special education, diagnostic, or treatment services.

A.

MADDS Prevalence Summary

Table I provides a summary of the prevalence findings for cerebral palsy, epilepsy, hearing impairment, mental retardation, and vision impairment overall and by sex and race for 10‐year‐old children who were residing in the five county metropolitan Atlanta area between 1985 and 1987. The numerator consists of children ascertained by MADDS who have one or more of the studied disabilities. The denominator used to calculate the prevalence of these developmental disabilities was from the Georgia intercensal population estimates provided by the Georgia OYce of Planning and Budget (Georgia Department of Human Resources, 1985). The intercensal estimates were reported in 5‐year age groups (0‐ to 4‐year olds, 5‐ to 9‐year olds, 10‐ to 14‐year olds). In order to determine the prevalence of epilepsy and hearing impairment, the number of 10‐year‐old children residing in the five counties of metropolitan Atlanta was established from the percentages of 10‐year‐old children (specific for county, race, and sex) in the 10‐ to 14‐year‐old group from the 1980 census. As indicated in Table I, a majority of the children ascertained by MADDS had mental retardation. With the exception of vision impairment, a higher prevalence of cerebral palsy, epilepsy, hearing impairment, and mental retardation was found among black and male children as compared to white children and females; however, the diVerence in magnitude varied by disability. The greatest disparity between black and white children and males and females was observed among children with mental retardation. The MADDS data were also used to establish the prevalence of other epilepsy syndromes including Lennox–Gastaut and infantile spasms. In a prevalence study conducted by Trevathan, Murphy, and Yeargin‐Allsopp

PREVALENCE

TABLE I CEREBRAL PALSY, EPILEPSY, HEARING IMPAIRMENT, MENTAL RETARDATION, AND VISUAL IMPAIRMENT, IN 10‐YEAR‐OLD CHILDREN IN METROPOLITAN ATLANTA, MADDS, 1985 THROUGH 1987

OF

Cerebral palsy a

PE (95% CI) Total Racec White Black Sex Male Female a

b

Epilepsy a

Hearing impairment b

PE (95% CI)

b

PE (95% CI)

Mental retardation a

b

Vision impairment

PE (95% CI)

PEa (95% CI)b

2.3 (2.0, 2.6)

6.0 (5.5, 6.5)

1.1 (0.9, 1.4)

12.0 (11.3, 12.7)

0.7 (0.5, 0.9)

2.1 (1.7, 2.5) 2.7 (2.1, 3.3)

5.7 (5.1, 6.4) 6.4 (5.6, 7.3)

1.0 (0.8, 1.3) 1.3 (0.9, 1.7)

7.4 (6.7, 8.1) 19.7 (18.3, 21.3)

0.8 (0.6, 1.0) 0.5 (0.3, 0.9)

2.7 (2.3, 3.2) 1.2 (0.9, 1.6)

6.7 (6.0, 7.5) 5.2 (4.6, 6.0)

1.2 (0.9, 1.6) 1.0 (0.7, 1.3)

13.8 (12.7, 14.9 10.1 (9.2,11.1)

0.9 (0.6, 1.2) 0.7 (4.0, 1.1)

Prevalence estimate. 95% confidence interval. c White includes white Hispanic and black includes black Hispanic. b

a

160

Rachel Nonkin Avchen et al.

(1997), a prevalence estimate of 0.3/1000 was found for Lennox–Gastaut. A majority of these children (91%) had mental retardation and 39% had infantile spasms. All of the 17% of children with profound mental retardation identified through MADDS had Lennox–Gastaut. This study also found that children with Lennox–Gastaut were more likely to have at least one of the other developmental disabilities tracked by MADDS. Another study conducted by Trevathan, Murphy, and Yeargin‐Allsopp (1999) assessed the epidemiologic profile of infantile spasms among 10‐year‐ old children in metropolitan Atlanta using the data collected by MADDS. It was found that 0.2/1000 10‐year‐old children had infantile spasms and 83% also had mental retardation. Further, 56% of children with infantile spasms had profound mental retardation while 50% had Lennox–Gastaut. B.

MADDSP Prevalence Summary

Table II indicates the average annual prevalence estimates for cerebral palsy, hearing loss, mental retardation, and vision impairment among 8‐year‐old children in metropolitan Atlanta during 1991–1994. Prevalence estimates for cerebral palsy, hearing loss, mental retardation, and vision impairment in children 3–10 years for study years 1991 through 1994 have been reported elsewhere (Boyle et al., 1996; Mervis, Boyle, & Yeargin‐ Allsopp, 2002; Van Naarden Braun, Decoufle, & Caldwell, 1999; Winter, Autry, Boyle, & Yeargin‐Allsopp, 2002). Table III summarizes the prevalence of those disabilities and autism spectrum disorders for the 1996 and 2000 study years. Overall, race‐ and sex‐specific estimates are provided. Prevalence estimates were calculated using, as the denominator, the number of 8‐year‐old children who resided in the five county metropolitan Atlanta area during the specified study year of interest according to the bridged‐race intercensal population estimates determined by the (National Center for Health Statistics, 2005). During the study period, 1991–1994, MADDSP ascertained 477 children who met the case definition for cerebral palsy, hearing loss, mental retardation, and vision impairment in 1991, 564 in 1992, 597 in 1993, and 644 in 1994. According to the bridged‐race intercensal population estimates there were 126,796 children 8 years of age residing in metropolitan Atlanta. Ninety‐five percent confidence intervals were computed based on the exact binomial method (Fleiss, 1981). Two race categories (white and black) represented the majority of children in these analyses. The average annual prevalence of hearing loss and mental retardation during this period was consistent with the prevalence findings for these disabilities in MADDS. However, a higher prevalence of cerebral palsy was observed during this period than was previously found in MADDS,

161

PUBLIC HEALTH IMPACT

TABLE II AVERAGE ANNUAL PREVALENCE OF CEREBRAL PALSY, HEARING LOSS, MENTAL RETARDATION, AND VISION IMPAIRMENT AMONG 8‐YEAR‐OLD CHILDREN BY RACE AND SEX, METROPOLITAN ATLANTA, MADDSP, 1991–1994

Total Race White, non‐Hispanic Black, non‐Hispanic Sex Male Female

Cerebral palsy

Hearing loss

Mental retardation

Vision impairment

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

3.2 (2.9, 3.5)

1.2 (1.1, 1.5)

12.5 (11.9, 13.1)

1.1 (0.9, 1.3)

3.2 (2.8, 3.6)

1.1 (0.9, 1.4)

8.9 (8.2, 9.6)

1.2 (0.9, 1.5)

3.5 (3.0, 4.1)

1.5 (1.2, 1.9)

19.3 (18.1, 20.5)

1.2 (0.9, 1.6)

3.4 (3.0, 3.9) 2.9 (2.5, 3.4)

1.5 (1.2, 1.8) 1.0 (0.8, 1.3)

14.8 (13.9, 15.7) 10.3 (9.5, 11.1)

1.2 (1.0, 1.5) 1.0 (0.8, 1.3)

a

Prevalence estimate. 95% confidence interval.

b

3.2/1000 children versus 2.3/1000 children, respectively. The prevalence of vision impairment was also slightly higher than was found for children with visual impairment in MADDS. However, this increase is likely due to the change in case definition of vision impairment, which was limited to children with legal blindness (visual acuity of 20/200 or worse) in MADDS and expanded to include children with a visual acuity of 20/70 or worse in MADDSP. The race and sex patterns observed in MADDSP (1991–1994) are similar to those found by MADDS. As found previously by MADDS data, a preponderance of black and male children were aVected with mental retardation as compared to white and female children. A higher prevalence of cerebral palsy was also observed among black and male children; however, the magnitude of diVerence was smaller than that observed for mental retardation. The race and sex diVerence among children with sensory impairment was minimal. Prevalence estimates decreased from 1996 to 2000 for all the developmental disabilities except autism spectrum disorders; the estimates for 2000 were more in line with previous prevalence reports than those from 1996 for cerebral palsy, hearing loss, mental retardation, and vision impairment (Table III). Further, the sex–race pattern for these disabilities was similar to previous MADDS and MADDSP reports, which might suggest that the influence of certain social or biologic mechanisms have not changed over time.

TABLE III PREVALENCE OF AUTISM SPECTRUM DISORDERS, CEREBRAL PALSY, HEARING LOSS, MENTAL RETARDATION, AND VISION IMPAIRMENT AMONG 8‐YEAR‐OLD CHILDREN BY RACE AND SEX, METROPOLITAN ATLANTA, MADDSP, 1996 AND 2000 ASD

Total Racec White, non‐Hispanic Black, non‐Hispanic Sex Male Female a

CP

HL

MR

1996

2000

1996

2000

1996

2000

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

4.2 (3.6, 5.0)

6.5 (5.8, 7.3)

3.6 (3.0, 4.3)

3.1 (2.6, 3.7)

1.4 (1.1, 1.9)

4.6 (3.7, 5.7)

7.9 (6.7, 9.3)

3.3 (2.5, 4.2)

2.7 (2.0, 3.6)

4.0 (3.1, 5.1)

5.3 (4.3, 6.4)

4.1 (3.1, 5.2)

6.7 (5.6, 8.0) 1.8 (1.2, 2.5)

11.0 (9.7, 12.4) 2.0 (1.5, 2.7)

3.8 (2.9, 4.8) 3.5 (2.7, 4.4)

1996

VI 2000

2000

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

PEa (95% CI)b

1.2 (0.9, 1.6)

15.5 (14.2, 16.8)

12.0 (11.0, 13.1)

1.4 (1.0, 1.8)

1.2 (0.9,1.6)

1.5 (1.0, 2.2)

0.9 (0.5, 1.4)

9.8 (8.4, 11.3)

6.8 (5.6, 8.1)

1.4 (0.9, 2.0)

1.1 (0.6, 1.6)

4.1 (3.1, 5.2)

1.3 (0.8, 2.0)

1.4 (0.9, 2.0)

22.7 (20.4, 25.2)

16.5 (14.7, 18.4)

1.4 (0.8, 2.1)

1.5 (1.0, 2.1)

3.6 (2.8, 4.5) 2.6 (2.0, 3.4)

1.2 (0.8, 1.9) 1.6 (1.1, 2.3)

1.4 (1.0, 2.0) 1.0 (0.6, 1.5)

19.1 (17.1–21.1) 11.8 (10.2, 13.5)

14.0 (12.5, 15.7) 10.0 (8.7, 11.4)

1.6 (1.0, 2.2) 1.2 (0.7, 1.8)

1.5 (1.0, 2.1) 0.9 (0.5, 1.4)

Prevalence estimate. 95% confidence interval. c White includes white non‐Hispanic only and black includes black non‐Hispanic only. ASD, autism spectrum disorders; CP, cerebral palsy; HL, hearing loss; MR, mental retardation; VI, vision impairment. b

1996

PUBLIC HEALTH IMPACT

163

In comparison to previous MADDSP study years and the 2000 study year, the greatest magnitude of increase in prevalence was observed among children with mental retardation in 1996. This increase is partly attributed to the higher number of children participating in special education programs for children with intellectual disability in 1996. In 1996, using the special education data, a greater percentage (9.5%) of children were in classes for mild, moderate, severe, and profound intellectual disability compared with the percentage of children in these programs in 2000 (8.6%) (Bhasin et al., 2006). As stated in a recent publication of this population, it is evident that the prevalence data obtained by MADDSP partly reflects the trends observed in special education placement of children with disabilities. In comparison to the 1996 study year, a higher prevalence of autism spectrum disorders was observed in the 2000 study year. At the time of this publication, extensive analyses were being conducted to evaluate plausible diVerences in the estimates. Race/ethnicity, sex, cognitive functioning, previous classifications, previous diagnoses, and special education exceptionality are factors being explored. Final analyses for this comparison between study years are currently in progress and thus, conclusions are not available at this time. Similar to the findings for MADDS and previous MADDSP study years, a greater proportion of black children were found among children with mental retardation and cerebral palsy in the 1996 and 2000 study years. A greater proportion of black children were also observed among children with sensory impairment in the 2000 study year; this race pattern diVered slightly from what was observed in 1996 for this group of children. In comparison to the other disabilities, a greater proportion of white children were aVected with autism spectrum disorders. Study results also show that male children were more likely to be aVected with autism spectrum disorders, cerebral palsy, hearing loss, mental retardation (only in 2000), and vision impairment. The higher proportion of males aVected with these disabilities is well known (Drews, Yeargin‐Allsopp, Murphy, & Decoufle, 1994; Fombonne, 1999, 2003; Gillberg & Wing, 1999; Mervis et al., 2002; Murphy et al., 1993; Van Naarden Braun et al., 1999). The predominance of males with mental retardation and vision impairment can partly be attributed to X‐linked conditions such as Fragile X and ocular albinism (Mervis et al., 2002; Yeargin‐Allsopp, Drews, Decoufle, & Murphy, 1995). The higher male‐to‐female ratio that has been consistently reported for autism spectrum disorders could also be partly due to X‐linked conditions like Fragile X since many children with autism spectrum disorders also have mental retardation. However, other possible reasons for this sex diVerence might be attributed to factors related to recognition and diagnosis (i.e., higher functioning males might come to the attention of a health care professional earlier than higher functioning females).

164

Rachel Nonkin Avchen et al. IV.

FINDINGS FROM EPIDEMIOLOGY STUDIES

Not only do population‐based surveillance programs allow for reporting of prevalence estimates, but they provide a framework for etiologic and descriptive studies. MADDS and MADDSP data have been used for many fruitful epidemiologic analyses and both datasets are linked to external database systems such as birth certificate, census, and MACDP databases. A number of special epidemiologic studies have also been carried out by collecting additional information on cases identified by MADDS and MADDSP, and in some instances MADDS and MADDSP cases were compared to matched controls. Following section reviews prevalence and epidemiologic studies that have been conducted using MADDS, MADDSP, and MADDS Follow‐Up data. Three tables (Tables IV–VI) were created to summarize the research conducted to date. The tables are organized by surveillance system and include a reference for each study and information about developmental disability(s) focus, surveillance period, population, major findings, and conclusions. V.

PUBLIC HEALTH IMPACT

The studies described throughout this chapter provide information on morbidity and mortality associated with developmental disabilities across the life span. These studies reflect the more than 20‐year history of CDC surveillance and research activities that have contributed to the understanding of developmental disabilities in key areas across the life stages including prenatal, perinatal, and postnatal risk factors; sociodemographic risk factors; and transition of children with developmental disabilities into young adulthood. In addition to the specific policy and public health implications of MADDS, MADDSP, and MADDS Follow‐Up Study data, the methodology developed through these systems is being replicated across the United States as part of the Autism and Developmental Disabilities Monitoring Network (ADDM) and Centers for Autism and Developmental Disabilities Research and Epidemiology (CADDRE), which have the potential for public health impacts far beyond metropolitan Atlanta. A.

Prenatal Risk Factors

Although the cause of a developmental disability is often unknown, studies have identified several prenatal risk factors for adverse developmental outcomes (Yeargin‐Allsopp, Murphy, Cordero, Decoufle, & Hollowell, 1997). Studies conducted by MADDS and MADDSP have examined a number of

SUMMARY References

OF

TABLE IV PUBLICATIONS USING MADDS DATA

DD focus

Surveillance period

CP, HI, MR, VI, EP

1985–1987

10‐year olds

Number of observed deaths was greater than the number of expected deaths regardless of the number of disabilities Observed‐to‐expected mortality ratio was 3:1 In general, the magnitude of mortality ratios was directly related to various measures of DD severity, except for isolated MR and cardiovascular problems

The specific underlying causes of death among deceased cohort members included some that were the putative cause of the developmental disability (e.g., a genetic syndrome) and others that could be considered intercurrent diseases or secondary health conditions (e.g., asthma)

Yeargin‐Allsopp et al. (1992)

CP, HI, MR, VI

1985–1987

10‐year olds

Prevalence of CP was 2.3/1000 children Prevalence of HI was 1.1/1000 children Prevalence of MR was 12.0/1000 children Prevalence of VI was 0.7/1000 children About 95% of the children with one or more of these four disabilities were initially identified through the school systems

A multiple source case ascertainment method based on extant records is much less costly than conducting medical and psychological assessments on populations of children. In addition, this method made it possible to estimate accurately the ‘‘administrative prevalence’’ of the DD tracked by MADDS

Drews et al. (1992)

Blindness

1985–1987

10‐year olds

Overall prevalence of blindness was 6.8/10,000 children Prevalence varied by gender and race and ranged from a low of 1.8/10,000 black girls to a high of 8.8/10,000 black boys Retinopathy of prematurity was most common cause of blindness About 66% of children with blindness also had a comorbid DD tracked by MADDS

The low prevalence of blindness among black girls and the frequent occurrence of blindness with other disabilities are noteworthy

MADDS Decoufle and Autry (2002)

Population

Major findings

Conclusions

165

(continued )

TABLE IV (Continued ) DD focus

Murphy et al. (1993)

CP

1985–1987

Winter et al. (2002)

CP

1985–1987, 1991–1994 (MADDSP)

Murphy, Trevathan, and Yeargin‐Allsopp (1995)

EP

1985–1986

166

References

Surveillance period

Population

Major findings

Conclusions

10‐year olds

Prevalence of CP was 2.3/1000 children Prevalence of CP was higher among boys and black children CP was acquired postnatally in 16% of the children and these children were more likely to be black males Spastic CP was the most common subtype (88%) About 75% of children with CP also had one of the other four DD tracked by MADDS

Multiple source case ascertainment is a useful method to obtain the administrative prevalence of CP among school‐age children. However, this method may be limited in its ability to estimate mild cases of CP and to accurately determine CP subtypes and severity

3‐ to 10‐year olds

There was a modest increase in the overall prevalence of congenital CP from 1.7 to 2.0/1000 1‐year survivors during the period from 1975 to 1991. This trend was primarily attributable to a slight increase in CP in infants of normal birthweight CP rates in moderately low and very low birthweight infants did not show consistent trends

There has been a modest increase in the prevalence of CP in 1‐year survivors born from 1975–1991. This increase was seen only in infant survivors of normal birthweight

10‐year olds

Prevalence of EP was 6.0/1000 children Capture recapture showed prevalence of EP could be as high as 7.7/1000 There was a greater prevalence of EP among boys; the prevalence did not vary appreciably by race 40% of children with EP had their first seizure at 1 year of age Partial seizures, including secondarily generalized seizures, were the most common seizure type (58%) About 3.5% of children with EP also had one of the other DD tracked by MADDS

An accurate estimate of the public health burden of childhood EP and determination of possible risk factors for idiopathic EP both depend on conducting complete community‐based case ascertainment and obtaining detailed clinical data

EP

1985–1987

10‐year olds

Prevalence of Lennox–Gastaut syndrome was 0.26/1000 children 91% of the children with Lennox–Gastaut syndrome also had MR; 17% of the profound MR cases had Lennox–Gastaut syndrome 39% of the children with Lennox–Gastaut syndrome also had infantile spasms Lennox–Gastaut syndrome accounts for 4% of all childhood EP Children with severe Lennox–Gastaut syndrome were more likely to have at least one of the DD tracked by MADDS

Lennox–Gastaut syndrome accounts for only 4% of all childhood EP, yet is a significant contributor to childhood morbidity

Trevathan et al. (1999)

EP

1985–1987

10‐year olds

The cumulative incidence of infantile spasms was 2.9/10,000 live births Half of the children with infantile spasms had cryptogenic IS The age‐specific prevalence of infantile spasms was 2.0/10,000 among 10‐year‐old children 83% of children with infantile spasms had MR; 56% profound MR. Among children with profound MR, 12% had a history of infantile spasms DD outcome did not diVer between those with cryptogenic spasms versus symptomatic infantile spasms 50% of those with infantile spasms had Lennox–Gastaut syndrome

Infantile spasms are rare in the general population. Yet, a significant percentage of all children with profound MR and severe childhood EP syndromes in the general population have a history of infantile spasms

Kuenneth et al. (1996)

EP

1985–1987

10‐year olds

Mothers of children with EP had more previous live births and more adverse reproductive outcomes including spontaneous abortions, very low birthweight, and infants with birth defects than mothers of children without EP Risk of EP was especially strong for a mother of child with birth defects Birth defects that were reported most frequently were CNS defects and Down syndrome

There are a few strong risk factors for childhood EP. Our results suggest that women who gave birth to a child with EP are likely to have a reproductive history characterized by adverse outcomes

167

Trevathan et al. (1997)

(continued)

TABLE IV (Continued ) DD focus

Surveillance period

Population

Drews et al. (1994)

HI

1985–1987

10‐year olds

Prevalence of HI was 1.1/1000 children Etiology could not be determined for 55% of the cases of HI 74% of the cases of HI were diagnosed after age 2 About 25% of the HI cases had a comorbid DD tracked by MADDS

Methods for early identification of children with hearing loss need to be improved

Murphy, Yeargin‐Allsopp, Decoufle, and Drews (1995)

MR

1985–1987

10‐year olds

Prevalence of MR was 12.0/1000 children Prevalence of mild MR was 8.4/1000 children Prevalence of severe MR was 3.6/1000 children Prevalence of MR was higher among blacks and males Children with severe MR were more likely to have one of the other DD tracked by MADDS

The MR prevalence rates reported in this study may reflect social and demographic characteristics unique to metropolitan Atlanta. Caution should be applied when comparing results across MR prevalence studies due to likely diVerences in case definitions, case ascertainment, time periods, age categories, and social and demographic composition of study populations

Drews et al. (1996)

MR

1985–1986

10‐year olds

Maternal smoking during pregnancy was associated with slightly more than 50% increase in the prevalence of idiopathic MR (OR ¼ 1.6) Children whose mothers smoked at least one pack a day during pregnancy had more than a 75% increase in the occurrence of idiopathic MR

Our data suggest that maternal smoking may be a preventable cause of MR

Decoufle and Boyle (1995)

MR

1985–1986

10‐year olds

Maternal education was the strongest predictor for having a child with MR Relative to children of white mothers with 12 years of education, children of black mothers, except those whose mothers had 16 or more years of education, were at increased risk

For isolated MR, maternal educational level was the most important predictor from among seven sociodemographic variables examined. There was a significant race– education interaction that indicated a steeper gradient in risk among white mothers than among black mothers

168

References

Major findings

Conclusions

169

Williams and Decoufle (1999)

MR

1985–1987

10‐year olds

There was not an elevated odds of having children with MR among mothers <17 years (adjusted OR ¼ 0.43) Children with isolated MR were more likely to be black, male, and of a higher birth order There was an elevated odds of codevelopmental MR among black children whose mothers were >30 years that was not attributed to Down syndrome

Educational level measured at time of delivery may not adequately characterize the ultimate educational attainment of teenaged mothers. Codevelopmental retardation among children of older black mothers could have resulted from multiple risk factors operating separately or synergistically

Decoufle et al. (1993)

MR

1985–1987

10‐year olds

48% of women worked while pregnant Risk for having children with MR increased among women with low level white collar occupations, especially service occupations There was a strong positive association between children with MR and maternal employment in the textile and apparel industries

Most comparisons yielded OR that were not indicative of unusual risks, but we did find lower than expected risks among children of teachers and health‐care professionals. We also found a strong, positive association between MR and maternal employment in the textile and apparel industries

Yeargin‐Allsopp et al. (1995)

MR

1985–1986

10‐year olds

There were elevated OR for mild MR among black children compared to white children after controlling for socioeconomic factors including sex, maternal age and education, birth order, and economic status

Five sociodemographic factors accounted for approximately half of the excess prevalence of mild MR among Black children. Possible reasons for the residual diVerence are discussed

Mervis et al. (1995)

MR

1985–1986

10‐year olds

Low birthweight children as a whole had an increased odds for MR (OR ¼ 2.8) Low birthweight children had a greater odds of severe MR than mild MR Normal birthweight children that were born preterm also had an elevated odds for MR

This was one of the first population‐based study of the association between low birthweight and MR conducted in the United States. Although the data document the magnitude of the association between low birthweight and MR, they do not shed any new light on the etiological mechanisms involved

(continued)

TABLE IV (Continued ) References

DD focus

Surveillance period

Population

Major findings

Conclusions

MR

1985–1986

10‐year olds

Boys were more likely than girls to have MR Older mothers were more likely than younger mothers to have a child with MR accompanied by another neurologic condition Other neurologic conditions were more common with severe MR than with mild MR High birth order, black maternal race, and low maternal education were associated with a higher prevalence of isolated MR

These findings suggest that sociodemographic risk factors for MR vary according to the presence of other neurologic conditions and that subdivisions based on medical or physical criteria may be useful in epidemiologic studies of MR

Yeargin‐Allsopp et al. (1997)

MR

1985–1987

10‐year olds

The majority of identified cases of MR (78%) did not have a known cause Of the cases where cause was determined, prenatal insults were present in 12%, perinatal causes in 6%, and postnatal causes in 4% of cases

Intensive use of public health prevention strategies can reduce the number of children who receive a MR diagnosis

170

Drews et al. (1995)

CP, cerebral palsy; DD, developmental disabilities; EP, epilepsy; MR, mental retardation; OR, odds ratio; HI, hearing impairment; VI, vision impairment.

SUMMARY References

OF

TABLE V PUBLICATIONS USING MADDSP DATA

DD focus

Surveillance period

Population

CP, HL, MR, VI

1991

3‐ to 10‐year olds

Prevalence of MR varied by age, race, and sex and ranged from 5.2/1000 children to 16.6/1000 children Prevalence of CP was 2.4/1000 children Severe MR accounted for one‐third of all cases Prevalence of CP was higher among black children (3.1/1000 children) than among white children (2.0/1000 children) Prevalence of moderate to severe HL was 1.1/1000 children Prevalence of HL was higher among black males than among children in the other race and sex groups Prevalence of VI was 0.8/1000 children

MADDSP data will be used to direct early childhood intervention eVorts to reduce the prevalence of these four DD. MADDSP data also are being used to measure progress toward the year 2000 national objectives for the prevention of serious MR

Centers for Disease Control and Prevention (1996)

CP, HL, MR, VI

1991

3‐ to 10‐year olds

4.5% of children in MADDSP had at least one DD attributable to a postnatal cause 3.5% of the cases of MR and 12.4% of the cases of HL were caused by postnatal insults Bacterial meningitis and child battering were the most common postnatal causes of DD The prevalence of two or more DD was more than twofold higher for those with postnatally acquired DD than for DD attributable to other causes

Surveillance for DD should include information about the underlying causes. Knowledge about cause can be used to inform and design public health prevention eVorts, because most postnatally acquired DD is preventable

Ashley‐Koch et al. (2001)

CP, HL, MR, VI

1991–1993

3‐ to 10‐year olds

Children with sickle cell disease had increased risk for DD (OE ¼ 3.2, p < .0001), particularly MR (OE ¼ 2.7, p ¼ .0005) and CP (OE ¼ 10.8, p < .0001) This risk was confined to DD associated with stroke (OE ¼ 130, p < .0001)

Intervention to prevent strokes in children with sickle cell disease is crucial to preventing a DD

MADDSP Boyle et al. (1996)

Major findings

Conclusions

171

(continued)

TABLE V (Continued )

172

References

DD focus

Surveillance period

Population

Major findings

Conclusions

Decoufle et al. (2001)

CP, HL, MR, VI

1991–1994

3‐ to 10‐year olds

7.2% of children with a birth defect had a serious DD 17.8% of children with DD had a birth defect Birth defects that originated in the nervous system and chromosomal defects resulted in the highest prevalence ratios (PR) for a subsequent DD PR were lowest for isolated birth defects Regardless of the severity of the defect or whether defects of the nervous system, chromosomal defects, or ‘‘other syndromes’’ were counted—PR for any DD monotonically increased with the number of coded birth defects per child or the number of diVerent birth defect categories per child

These data highlight the possible early prenatal origins of some DD and suggest that both the number of coded birth defects present and the number of anatomic systems involved are strongly related to functional outcomes

Van Naarden Braun et al. (2005)

CP, HL, MR, VI

1991–1994

3‐ to 10‐year olds

Recurrence risk estimates for MR, CP, HL, and VI ranged from 3% to 7% and were many times greater than the baseline prevalence for each disability The RR for MR was eight times greater than the baseline prevalence Isolated MR was highly concordant between siblings with MR Demographic, SES factors, and birthweight were not significantly associated with recurrence risks for MR

Further research is needed to investigate the roles of genetic and environmental factors on the recurrence of DD, particularly isolated mild MR

Van Naarden Braun et al. (2003)

BD, MR, SED, SDD, SLD, SLI

1991–1994, 1991–1998, Special Education Database of Metropolitan Atlanta (SEDMA)

3‐ to 10‐year olds

Approximately 147 infants screened were positive for a metabolic or endocrine disorder and were at risk for MR if left untreated using the MADDSP and newborn screening program database linkage. Yet, only three children were identified with MR Approximately 216 children would be expected to have MR if their metabolic disorder was left untreated using the SEDMA and newborn screening program database linkage. Yet, nine children were identified with a less severe DD from this linkage

Although children found in MADDSP or SEDMA have a low occurrence of DD attributable to these metabolic or endocrine disorders, our finding of any cases of DD of varying severity attributable to a metabolic or endocrine disorder suggests a need for ongoing population‐based monitoring of the long‐term developmental outcomes of children identified through newborn screening programs

173

Centers for Disease Control and Prevention (2004)

CP, HL, MR, VI

1991–1994

5‐ to 10‐year olds

Estimated lifetime costs in 2003 in dollars are expected to total $51.2 billion for persons born in 2000 with MR, $11.5 billion for persons with CP, $2.1 billion for persons with HL, and $2.5 billion for persons with VI

The costs associated with DD in the United States highlight the need for strategies to reduce the prevalence of these conditions and prevent development of secondary conditions

Yeargin‐Allsopp et al. (2003)

ASD

1996

3‐ to 10‐year olds

Prevalence of ASD was 3.4/1000 children The male–female ratio was 4:1 68% of children with IQ or developmental test results had cognitive impairment 40% of children with autism were identified only at educational sources

The rate of ASD found in this study was higher than the rates from studies conducted in the United States during the 1980s and early 1990s, but it was consistent with those of more recent studies. Schools were the most important source for information on black children, children of younger mothers, and children of mothers with less than 12 years of education

DeStefano et al. (2004)

ASD, MR

1996

3‐ to 10‐year olds

There was no significant association between the MMR vaccine and ASD The overall distribution of ages at time of MMR vaccination among children with ASD was similar to that of matched control children

Similar proportions of case and control children were vaccinated by the recommended age (18 months) or shortly after and before 24 months of age, the age atypical development is usually recognized in children with autism. Vaccination before 36 months was more common among case children than control children, especially among children 3–5 years of age, likely reflecting immunization requirements for enrollment in early intervention programs

Winter et al. (2002)

CP

1985–1987 (MADDS), 1991–1994

3‐ to 10‐year olds

There was a modest increase in the overall prevalence of congenital CP from 1.7 to 2.0/1000 1‐year survivors during the period from 1975 to 1991. This trend was primarily attributable to a slight increase in CP in infants of normal birthweight CP rates in moderately low and very low birthweight infants did not show consistent trends

The reason for an increase in CP in heavier birthweight infants is unclear. Additional trends in CP rates may emerge as monitoring of CP continues. With more years of data, we may better understand the impact of recent medical interventions on the risk of CP and other DD

(continued)

TABLE V (Continued ) DD focus

Surveillance period

Population

Scher et al. (2002)

CP

1991–1992

3‐ to 10‐year olds

Twins were at an approximately fivefold increased risk of fetal death, sevenfold increased risk of neonatal death, and fourfold increased risk of CP compared with singletons Twins from growth–discordant pairs and twins whose cotwin died were at increased risk of both mortality and CP

This is the largest population‐based study of CP to date. The overall rates of death or CP were higher in twins than singletons, although small twins generally did better than small singletons. Cotwin death was a strong predictor of CP in surviving twins. This risk was the same for same‐ and diVerent‐sex pairs, and observed across the birthweight spectrum

Boyle et al. (2000)

CP

1991–1992

3‐ to 10‐year olds

There was no association between exposure to magnesium sulfate and CP risk (OR ¼ 0.9, CI ¼ 0.3, 2.6)

Several ongoing randomized clinical trials of magnesium and CP may shed more definitive light on this relation

Schendel et al. (1996)

CP, MR

1986–1988

3‐ to 5‐year olds

There was no association between prenatal magnesium sulfate exposure and infant mortality (adjusted RR ¼ 1.02, CI ¼ 0.8, 1.25) Among Atlanta‐born survivors, those exposed to magnesium sulfate had a lower prevalence of CP and MR than those not exposed

A reduced risk for CP, and possibly MR, among very low birthweight children is associated with prenatal magnesium sulfate exposure. The reduced risk for childhood CP or MR does not appear to be due to selective mortality of magnesium sulfate‐exposed infants

Centers for Disease Control and Prevention (1997)

HL

1991–1993

3‐ to 10‐year olds

Average annual prevalence of HL was 1.1/1000 children Only 8% of children had their HL diagnosed during their first year of life The mean age of earliest known diagnosis was 2.9 years

This study highlights the public health intervention opportunity for universal newborn hearing screening programs to identify children with HL at younger ages. Interventions to reduce the occurrence of communication disabilities associated with HL are most successful if aVected children are identified early, ideally during the first few months of life

174

References

Major findings

Conclusions

175

Van Naarden Braun et al. (1999)

HL

1991–1993

3‐ to 10‐year olds

Prevalence of HL was 1.1/1000 children The highest rate of HL was among male, black children (1.4/1000 children) 90% of case children with a known type of loss had sensorineural loss The mean age of earliest diagnosis was 2.9 years overall, 3.6 years for moderate HL, and 2.4 years for severe‐profound HL 30% of case children had another neurodevelopmental condition, most frequently MR

MADDSP is currently the only ongoing source of information on the prevalence of serious HL among children in the US that is not derived from a one‐time hearing screen. Further research to examine reasons for racial disparities among rates for children with HL may be fruitful

Van Naarden Braun and Decoufle (1999)

HL

1991–1993

3‐ to 10‐year olds

Overall prevalence of bilateral congenital sensorineural HL was 5.3/10,000 children Prevalence of bilateral congenital sensorineural HL was 6.6/10,000 children with birthweight 2500–2999 g Prevalence of bilateral congenital sensorineural HL was 12.7/10,000 children with birthweight 1500–2499 g Prevalence of bilateral congenital sensorineural HL was 51.0/10,000 children with birthweight <1500 g

The elevated RR among children weighing less than 3000 g may have implications for future newborn hearing screening criteria

Centers for Disease Control and Prevention (1999)

MR

1991–1994

3‐ to 10‐year olds

The expected number of children with MR attributable to one of six metabolic disorders barring treatment was 148 Two children were identified by MADDSP as having MR associated with one of the six underlying metabolic disorders under study

Newborn screening is highly eVective in reducing the burden of MR associated with studied metabolic disorders. However, screening for metabolic disorders does not ensure complete detection of aVected infants Surveillance for DD among children with metabolic disorders would facilitate eVorts to determine the eVectiveness of treatment and metabolic control

(continued)

TABLE V (Continued )

176

References

DD focus

Surveillance period

Population

Major findings

Conclusions

Mervis and Boyle (2002)

VI

1991–1994

6‐ to 10‐year olds

Prevalence of VI was 10.7/10,000 children Nearly two‐third of children with VI had a comorbid DD There were no significant genders or racial diVerences

The common presence of coexisting disabilities emphasizes the importance of multidisciplinary services. The inclusion of case ascertainment sources other than VI classes is critical to ensure an accurate prevalence rate and unbiased description of children with VI

Mervis et al. (2000)

VI

1991–1993

3‐ to 10‐year olds

Prenatal etiologies were identified in 43% of the children 38% of the prenatal etiologies were genetic Perinatal etiologies were found in 27% of the children Prenatal etiologies were more common in children with isolated VI Perinatal and postnatal etiologies were more common in children with multiple disabilities Children with prenatal etiologies had less severe VI than those with perinatal or postnatal insults

The results of this study have implications for targeting VI prevention eVorts. An important focus should be the prevention of VI due to retinopathy of prematurity (ROP)

ASD, autism spectrum disorders; BD, behavioral disorders; CI, 95% confidence interval, CP, cerebral palsy; DD, developmental disabilities; IQ, intelligence quotient; MR, mental retardation; OD, odds ratio; OE, observed:expected ratio; HL, hearing loss; RR, risk ratio; SDD, significant developmental delay; SED, severe emotional disorder; SLD, specific learning disability; SLI, speech and language impairment; VI, vision impairment.

SUMMARY References

OF

TABLE VI PUBLICATIONS USING MADDS FOLLOW‐UP STUDY DATA

DD focus

Surveillance period

Population

CP, HL, MR, VI, EP

1997–2000

21‐ to 25‐year olds

Young adults with a history of DD were less likely to be involved in tobacco use, substance abuse, and sexual activity. Areas of concern included below normal body mass index, lack of HIV/AIDS and sex education, and preventive healthcare services for women, and victimization

Despite some healthy lifestyle indicators, health gaps may place young adults with a history of DD at risk for poor health and quality of life

Van Naarden Braun et al. (in press)

CP, HL, MR, VI, EP

1997–2000

21‐ to 25‐year olds

For young adults with isolated impairment, activity limitations are not probable outcomes; this is not the case for those with severe MR and/or multiple impairments

The type and extent of activity limitations vary by impairment characteristics. The conceptual framework of the International Classification of Functioning (ICF) provides a useful tool for testing hypotheses to pinpoint areas of intervention and its application would strengthen future investigations within the epidemiology of DD

Van Naarden Braun et al. (2005)

CP, HL, MR, VI, EP

1997–2000

21‐ to 25‐year olds

The results underscore the diVerences in leisure lifestyles by impairment type and severity. Activity limitations, educational attainment, and the acquisition of adult social roles were found to be significant predictors of leisure activity

This study emphasizes the importance of improving daily activities, increasing attendance of postsecondary school, opportunities for competitive employment, and participation in impairment‐related programs to help increase the number and scope of types of leisure activities for young adults with DD

Van Naarden Braun et al. (2006)

CP, HL, MR, VI, EP

1997–2000

21‐ to 25‐year olds

Attaining adult social roles varies by impairment type and severity Experiencing activity limitations partially mediate the relationship between impairment and adult social roles Attending postsecondary education increases the likelihood of attaining markers of adulthood

Intervention to reduce activity limitations and to develop strategies to increase attendance in postsecondary education may increase the likelihood for the acquisition of adult social roles among young adults with childhood impairment

MADDSFU Rurangirwa et al. (2006)

Major findings

CP, cerebral palsy; DD, developmental disabilities; EP, epilepsy; HL, hearing loss; MR, mental retardation; VI, vision impairment.

Conclusions

178

Rachel Nonkin Avchen et al.

potential prenatal risk factors for developmental disabilities including prenatal magnesium sulfate exposure (Schendel, Berg, Yeargin‐Allsopp, Boyle, & Decoufle, 1996), twinning (Scher et al., 2002), maternal smoking (Drews, Murphy, Yeargin‐Allsopp, & Decoufle, 1996), infection (Mervis, Yeargin‐ Allsopp, Winter, & Boyle, 2000), reproductive risk factors (Drews et al., 1996), and low birthweight (Van Naarden Braun & Decoufle, 1999). Population‐based data on cerebral palsy collected in metropolitan Atlanta were used to examine a hypothesis from several other epidemiologic studies, which suggested that magnesium sulfate used for the treatment of preeclampsia or preterm labor might lower the risk of cerebral palsy in very low birthweight children. Two Atlanta studies examined this hypothesis and found that the risk for cerebral palsy was reduced in very low birthweight babies born after exposure to magnesium sulfate. The study investigators suggested this finding should be confirmed in a randomized clinical trial (Boyle, Yeargin‐ Allsopp, Schendel, Holmgreen, & Oakley, 2000). The Randomized Clinical Trial of the Beneficial EVects of Antenatal Magnesium Sulfate concluded that contrary to the original hypotheses, antenatal magnesium sulfate was associated with worse, not better, perinatal outcome in a dose–response fashion (Mittendorf et al., 2002). Overtime, there has been a secular trend toward an increase in the number of multiple births. The question of whether this increase has been associated with a related increase in morbidity or mortality is an important public health question. Because of the large MADDSP population‐based dataset, cases of twins with cerebral palsy were included in a large international dataset of more than 25,000 twins covering a 10‐year period. This combined dataset was used to examine demographic and clinical factors associated with morbidity (cerebral palsy) and mortality (fetal or neonatal death). It was found that from 1980 to 1989 there was an increasing trend toward twin gestations, an increasing proportion of unlike sex pairs, and increasing maternal age in twin gestations. The study reported an increased risk of fetal and neonatal death as well as an increased risk of cerebral palsy in twins as compared to singletons (Scher et al., 2002). Prenatal risk factors for mental retardation that are preventable are of great public health interest since many known risk factors for mental retardation are genetic and primary prevention is not easily achieved. In a paper that examined the relationship between maternal smoking during pregnancy and idiopathic mental retardation, it was found that smoking during pregnancy increases by 50% the likelihood of mental retardation in a child, and there seems to be a dose response to this relationship (Drews et al., 1996). This translates into a 35% attributable risk of mental retardation in children of mothers who smoke. This finding has tremendous implications for the

PUBLIC HEALTH IMPACT

179

general population, especially for populations at high risk (e.g., teenaged and young adult women). Unlike smoking, the prevention potential related to low birthweight is not as great. MADDS data confirmed that a birthweight 2500 g increased the risk of mental retardation by almost threefold (Mervis, Decoufle, Murphy, & Yeargin‐Allsopp, 1995). The risk was highest in children born at birthweights <1500 g, and the risk was higher for severe mental retardation than for mild mental retardation. However, prevention of low birthweight remains elusive. In a publication that examined etiologies of vision impairment, prenatal factors contributed more to vision impairment than perinatal or postnatal factors (Mervis et al., 2000). It was also noted that prenatal etiologies were increasingly common as birthweight increased, and 56% of children weighing 2500 g had a prenatal etiology. From a prevention perspective, however, it is important to note that 61% of the prenatal etiologies were genetic, and retinopathy of prematurity accounted for 70% of the vision impairment in children born weighing <1500 g. Further, retinopathy of prematurity was identified as the most common known cause of vision impairment in 10‐year‐old children (Drews, Yeargin‐Allsopp, Murphy, & Decoufle, 1992). Similar to vision impairment, birthweight is an important contributor to the risk of hearing loss (Van Naarden Braun & Decoufle, 1999). The overall percentage of moderate to profound bilateral sensorineural hearing loss that was attributable to children weighing <2500 g at birth was 18.9% and was 9.4% for children weighing <1500 g at birth. From an epidemiologic perspective, little is known about the risk of epilepsy related to a maternal history of previous adverse reproductive outcomes. Using a case‐control methodology, it was found that mothers with epilepsy had a history of more spontaneous abortions, previous very low birthweight infants, and more children with birth defects. The risk for birth defects was highest for central nervous system defects and Down syndrome (Kuenneth, Boyle, Murphy, & Yeargin‐Allsopp, 1996). The linkage of MACDP to MADDSP allows CDC investigators to examine the overlap between the occurrence of birth defects and developmental disabilities. It was found that 7.2% of children with a birth defect had a developmental disability and 17.8% of children with a developmental disability had a birth defect (Decoufle, Boyle, Paulozzi, & Lary, 2001). B.

Perinatal Risk Factors

Newborn metabolic screening is one of the largest public health prevention programs targeting health outcomes for infants. Yet, the long‐term developmental status of babies screened and treated had not been evaluated on a

180

Rachel Nonkin Avchen et al.

large population until MADDSP data were used to evaluate the outcomes of children born from 1981 through 1991 (Centers for Disease Control and Prevention, 1999; Van Naarden Braun, Yeargin‐Allsopp, Schendel, & FernhoV, 2003). Results showed that screening does not assure developmental disabilities will be prevented in all children. More systematic ongoing surveillance for developmental conditions in children who screen positive and are diagnosed with a metabolic or endocrine disorder is critical for evaluating this public health prevention program. There have been inconclusive findings regarding the influence of infant survival on the prevalence of cerebral palsy. MADDS and MADDSP data were used to examine the impact of survival and birthweight on the prevalence of cerebral palsy overtime, from 1975 to 1991. The prevalence of cerebral palsy increased over this period for children born <1500 g. Although there was a modest increase in the prevalence of cerebral palsy in 1‐year survivors, this increase was only seen in infant survivors of normal birthweight (Winter et al., 2002). C.

Postnatal Risk Factors

Postnatal causes of developmental disabilities are substantially less than prenatal and perinatal causes and vary by disability from 4% in children with mental retardation to about 16% in children with cerebral palsy (Centers for Disease Control and Prevention, 1996; Murphy et al., 1993; Yeargin‐Allsopp et al., 1997). Although mortality from sickle cell disease has long been appreciated, morbidity related to developmental disabilities in children with sickle cell disease has not been. Using data from MADDSP, it was found that children with sickle cell disease had an increased risk for developmental disabilities associated with stroke. It was concluded that aggressive interventions are needed to prevent stroke in these children (Ashley‐Koch, Murphy, Khoury, & Boyle, 2001). Just as it is important to identify risk factors for developmental disabilities, it is also important to note the lack of an association for certain hypotheses with public health implications. For example, there has been considerable controversy as to whether receipt of the MMR vaccine increases the risk of autism. Using MADDSP data, the vaccination histories of children with and without autism were compared to see if there were diVerences in the age distribution of the MMR vaccine that might suggest a relationship between timing of the vaccination and the onset of autism. The study concluded that there was no relationship between the MMR vaccine and autism, which corroborates results of other published epidemiologic studies that examined this relationship (Chen, Landau, Sham, & Fombonne, 2004; Dales, Hammer, & Smith, 2001; DeStefano, Bhasin, Thompson, Yeargin‐Allsopp, & Boyle,

PUBLIC HEALTH IMPACT

181

2004; Fombonne, 2003; Kaye, del Mar Melero‐Montes, & Jick, 2001; Taylor et al., 1999). To investigate the risk for recurrence of a developmental disability for families with at least one child with a developmental disability, birth certificate data was linked longitudinally to identify all siblings born to the same mother between 1981 and 1991. These data were linked to MADDSP data to evaluate the risks and factors associated with recurrence of developmental disabilities. Recurrence risks across all developmental disabilities identified by MADDSP ranged from 3% to 7%, which was much higher than the general population prevalence (Van Naarden Braun, Autry, & Boyle, 2005). This risk varied by disability, and the greatest risk for recurrence was for children with mental retardation, more specifically for isolated mild mental retardation. D.

Sociodemographic Risk Factors

In addition to the overall health of the mother and biologic conditions that infer risk for developmental disabilities, there are a number of sociodemographic factors that are also important predictors for developmental disabilities. These include maternal occupation (Decoufle, Murphy, Drews, & Yeargin‐Allsopp, 1993), education (Decoufle & Boyle, 1995), age (Williams & Decoufle, 1999), and median family income (Bhasin & Schendel, 2006). MADDS data were the first to examine the possible association between maternal occupation and mental retardation using a large population‐based surveillance cohort. Approximately 48% of pregnant women worked in the mid‐1970s (Decoufle et al., 1993). A case‐control method was used to interview mothers of children with and without mental retardation. There was a strong positive association between maternal employment in the textile and apparel industries and mild mental retardation, and there was a low risk for mental retardation in children of mothers in education and health care professions. However, because of limitations in the data, the authors cautioned against overinterpretation of the findings. The association between children’s mean IQ scores and their mother’s level of education has been reported in the literature as early as the 1930s. MADDS data were used to further examine this relationship in 10‐year‐old children in metropolitan Atlanta born in 1975 and 1976 (Decoufle & Boyle, 1995). MADDS analyses employed the concept of isolated versus nonisolated disability and defined the terms to mean having a developmental disability in the absence (isolated) or presence (nonisolated) of another serious neurologic condition. This was an important concept in that it provided the ability to subdivide children with a developmental disability in a way that was possibly more meaningful than the traditionally applied

182

Rachel Nonkin Avchen et al.

level of severity, which is recognized as having limitations particularly when examining racial groupings for mental retardation. When mental retardation was examined, it was found that maternal education (at the time of the child’s birth) was inversely related to mental retardation without other serious neurologic conditions (i.e., isolated mental retardation). Additionally, low maternal education was the strongest predictor for nonisolated mental retardation. Another important finding from this study was that black children born to mothers with any level of education, except 16 years, had an increased risk of isolated mental retardation compared to white children. Another MADDS publication continued the examination of mental retardation using this grouping and found that other neurologic conditions were more common with severe mental retardation than with mild mental retardation and older mothers were more likely to have a child with mental retardation with neurologic conditions (not accounted for by Down syndrome) (Drews, Yeargin‐Allsopp, Decoufle, & Murphy, 1995). This study also corroborated results from Decoufle and Boyle (1995) that showed black race and low maternal education were associated with a higher prevalence of isolated mental retardation. The finding that a higher prevalence of mental retardation occurs in black children has generated much controversy overtime. MADDS data were used to inform this debate by examining a number of potential confounders or eVect modifiers of this relationship using a logistic regression model that adjusted for relative factors including sex, maternal age, birth order, maternal education, and economic status (Yeargin‐Allsopp et al., 1995). These factors were shown to contribute to about half of the elevated risk for mild mental retardation in black children compared to white children. However, the risk remained elevated after controlling for the various factors and possible explanations for the remaining diVerence were described in the paper. In addition to race and maternal education, maternal age is another potentially important risk factor for mental retardation. Analyses from Williams and Decoufle (1999) focused on the potential relationship between maternal age and mental retardation and results supported findings from other studies, which showed that older maternal age is associated with mental retardation with neurologic conditions. Yet in the MADDS data, this association was due to the presence of Down syndrome in white but not black mothers. Children of teenaged mothers were not at an increased risk for mental retardation with or without neurologic conditions. MADDSP data were used to look at the relationship between sociodemographic factors and autism spectrum disorders. A case‐control study was conducted to examine the relationship of children with autism spectrum

PUBLIC HEALTH IMPACT

183

disorders based on the absence (isolated) or presence (nonisolated) of a coexisting MADDSP disability or birth defect. Findings showed that markers of higher social class (higher maternal education and higher median family income) were significantly associated with isolated autism spectrum disorders but not nonisolated autism spectrum disorders. Also when the association was further examined by source of ascertainment (educational versus health sources), children with autism spectrum disorders identified from only health sources had mothers with higher education and median family income than children identified at educational sources only. These findings suggest that future studies should consider phenotypic subgroups of children with autism spectrum disorders and must consider potential ascertainment bias (Bhasin & Schendel, 2006). E.

Transition of Children with Developmental Disabilities into Young Adulthood

As children with developmental disabilities reach young adulthood, supportive services such as rehabilitation, special educational, and specialized pediatric care often cease. Without these support systems, young adults with developmental disabilities may experience new problems with daily activities, exacerbations of existing problems, and/or diYculties in acquiring employment, independent living, and/or maintaining a healthy lifestyle. The new social roles of young adulthood, coupled with the vulnerabilities created as a result of declining support systems, underscore the need to examine the consequences of developmental disabilities among individuals in this age group. Each analysis conducted using MADDS Follow‐Up Study data demonstrated that there is significant variability in daily functioning (Van Naarden Braun, Yeargin‐Allsopp, & Lollar, in press), obtaining an adult social role (Van Naarden Braun, Yeargin‐Allsopp, & Lollar, 2006), participating in a diversity of leisure activity (Van Naarden Braun, Yeargin‐Allsopp, & Lollar, 2005), and maintaining a healthy lifestyle (Rurangirwa, Van Naarden Braun, Schendel, & Yeargin‐Allsopp, 2006) among young adults with developmental disabilities identified in childhood by type and severity of impairment. For young adults with an isolated developmental disability, limitations in daily activities, diYculties in acquiring normative markers of adulthood, and restrictions in leisure activities need not be expected outcomes. By contrast, unfortunately, for young adults with severe mental retardation, the presence and number of activity limitations can be predicted by the presence of impairment. However, despite extensive diYculties in daily functioning, young adults with severe mental retardation are participating in impairment‐related vocational and educational programs

184

Rachel Nonkin Avchen et al.

(e.g., supported employment, day activity center) as well as a range of leisure activities. Further analyses with the MADDS Follow‐Up data found that activity limitations partially mediate the relationships between impairment and acquiring adult social roles, attaining markers of independence, and participating in leisure activity. Therefore, intervention to reduce activity limitations and to develop strategies to increase attendance in postsecondary education may increase the likelihood for the acquisition of adult social roles among young adults with childhood impairment. In addition to these two types of interventions, increasing opportunities for competitive employment, and participation in impairment‐related programs may help increase the number and scope of types of leisure activities for young adults with developmental disabilities. F.

Replication of MADDSP Methodology: ADDM/CADDRE Surveillance Network

Due to public concern over the rising prevalence of autism in the United States, the Children’s Health Act of 2000 mandated CDC to establish surveillance and research programs to address the prevalence and causes of autism and related developmental disabilities (US Congress, 2000). Under the provisions of this act, NCBDDD established the ADDM and CADDRE network. Figure 2 illustrates the sites in the ADDM and CADDRE network. All funded sites are either located in a state health department or university serving as an agent of their respective state department of health. Investigators also collaborate with their respective state Department of Education in an eVort to collect surveillance data, which in turn may be useful for educational planning for students with autism spectrum disorders and other developmental disabilities. The mission of the ADDM and CADDRE Surveillance Network is to understand the magnitude and characteristics of the population of children with autism spectrum disorders and related developmental disabilities to inform science and policy. Together, the ADDM and CADDRE network will estimate prevalence and population characteristics of children with autism spectrum disorders living in a population‐base of over 600,000 children in the United States. One of the strengths of the ADDM and CADDRE Surveillance Network data is that the majority of participating sites are replicating the MADDSP methodology to conduct surveillance. This network would likely not have come to fruition without the longstanding history, experience, and influence of years of results from MADDS and MADDSP. The ADDM and CADDRE network completed data collection for 2000 and 2002 surveillance years and is currently collecting data for 2004 and 2006. In addition to surveillance, other CADDRE aims include

Washington Montana

North Dakota

Vermont

Minnesota Michigan

Oregon

Wisconsin

South Dakota

Idaho

Ohio

Utah

D

Illinois Indiana Colorado

California

ork

Pennsylvania

Iowa

Nebraska

Nevada

New Y

Michigan

Wyoming

Kansas

Missouri

Kentucky

West Virginia Virginia

Maine New Hampshire Massachusetts Rhode Island Connecticut New Jersey Delaware Maryland

arolina

New Mexico

Arkansas

Oklahoma

North C

South Carolina

Alabama Georgia

Mi

ssi

ssi

pp

i

Arizona

Tennessee

Texas

Louisiana

Alaska

Florida

Hawaii Legend ADDM CADDRE MADDSP, gold standard for ADDM and CADDRE

FIG. 2. ADDM and CADDRE network.

U.S. Virgin Islands

186

Rachel Nonkin Avchen et al.

implementation of a multisite collaborative epidemiology study of autism and related developmental disabilities and site‐specific, investigator‐initiated studies on autism spectrum disorders. G.

Policy Implications

MADDS prevalence for hearing impairment (later referred to as hearing loss) from the 1980s found that approximately 1 in every 1000 children is born with some hearing impairment. Unfortunately, 74% of cases of hearing impairment were diagnosed after age 2 (Drews et al., 1994) and only 8% of children had their hearing loss diagnosed during their first year of life (Centers for Disease Control and Prevention, 1997; Van Naarden Braun et al., 1999). Unlike some of the other developmental disabilities, there is extensive literature about the importance of early intervention on hearing loss in terms of improved communication and language skills. In the early 1990s, there was interest in looking at the age of first identification of children with hearing loss in order to promote early intervention. MADDSP data showed that among all children with congenital hearing loss, the mean age at earliest known diagnosis was 2.9 years and less than 10% of all case children had their hearing loss diagnosed <1 year of age (Van Naarden Braun et al., 1999). The mean age at diagnosis for children with moderate hearing loss was 3.6 years compared with 2.4 years for children with severe or profound hearing loss (Van Naarden Braun et al., 1999). These data contributed to the initiation of programs to promote early screening of newborns to assess hearing function, preferably before hospital discharge. Today, newborn hearing screening is conducted in every state throughout the United States, and about 90% of all newborns are screened. There are early hearing and detection intervention (EHDI) programs at every state, which promote early infant hearing screening, timely follow‐up evaluations, and early intervention services. In addition to state eVorts, this national public health initiative is supported by a wide range of federal agencies, advocacy groups, the public, and professional organizations such as the American Academy of Pediatrics. Early identification and intervention help children with hearing loss maximize their communication and language development. MADDSP data have also been used to determine the economic impact of developmental disabilities. An economic evaluation of the costs for four of the MADDSP developmental disabilities was performed and showed that the costs for each disability were substantial and varied by type and severity of developmental disability (Centers for Disease Control and Prevention, 2004). Estimated lifetime costs based on 2003 dollars are expected to total $51.2 billion for persons born in 2000 with mental retardation, $11.5 billion

187

PUBLIC HEALTH IMPACT

for persons with cerebral palsy, $2.5 billion for persons with vision impairment, and $2.1 billion for persons with hearing loss. The lifetime cost associated with having one of these developmental disabilities is significant and has implications for the aVected individual, their family, and society.

VI.

CONCLUSIONS

The establishment and ongoing implementation of MADDS and MADDSP have used a public health framework for understanding developmental disabilities. In addition to providing ongoing estimates of prevalence that are useful for policy, prevention, and intervention planning, MADDSP continues to facilitate epidemiologic studies to investigate risk and other factors associated with developmental disabilities. The studies presented in this chapter demonstrate the utility of MADDS and MADDSP data to address important public health issues. Further, such activities will continue to inform gaps, which will help identify future research and policy priorities.

REFERENCES American Psychiatric Association (2000). Diagnostic and statistical manual of mental disorders (4th ed., Text Revision). Washington, DC: American Psychiatric Association. Ashley‐Koch, A., Murphy, C. C., Khoury, M. J., & Boyle, C. A. (2001). Contribution of sickle cell disease to the occurrence of developmental disabilities: A population‐based study. Genetics in Medicine, 3(3), 181–186. Bhasin, T. K., & Schendel, D. (2006). Sociodemographic risk factors for autism in a US metropolitan area. Journal of Autism and Developmental Disorders. Sep. 2 [Epub ahead of print]. Bhasin, T. K., Brocksen, S., Nonkin Avchen, R., & Van Naarden Braun, K. (2006). Prevalence of four developmental disabilities in 8‐year‐old children: Metropolitan Atlanta Developmental Disabilities Surveillance Program, 1996 and 2000, Morbidity and Mortality Weekly Report, 55(SS01), 1–9. Erratum in: Morbidity and Mortality Weekly Report. 2006 February; 54(4): 105–106. Boyle, C. A., Yeargin‐Allsopp, M., Doernberg, N. S., Holmgreen, P., Murphy, C. C., & Schendel, D. E. (1996). Prevalence of selected developmental disabilities in children 3–10 years of age: The Metropolitan Atlanta Developmental Disabilities Surveillance Program, 1991. Morbidity and Mortality Weekly Report Surveillance Summaries, 45(SS‐2), 1–14. Boyle, C. A., Yeargin‐Allsopp, M., Schendel, D. E., Holmgreen, P., & Oakley, G. P. (2000). Tocolytic magnesium sulfate exposure and risk of cerebral palsy among children with birth weights less than 1,750 grams. American Journal of Epidemiology, 152(2), 120–124. Centers for Disease Control and Prevention (1996). Postnatal causes of developmental disabilities in children aged 3–10 years—Atlanta, Georgia, 1991. Morbidity and Mortality Weekly Report, 45(6), 130–134.

188

Rachel Nonkin Avchen et al.

Centers for Disease Control and Prevention (1997). Serious hearing impairment among children aged 3–10 years—Atlanta, Georgia, 1991–1993. Morbidity and Mortality Weekly Report, 46(45), 1073–1076. Centers for Disease Control and Prevention (1999). Mental retardation following diagnosis of a metabolic disorder in children aged 3–10 years—metropolitan Atlanta, Georgia, 1991–1994. Morbidity and Mortality Weekly Report, 48(17), 353–356. [Erratum in Morbidity and Mortality Weekly Report, 2000 March 3, 49(8), 162.] Centers for Disease Control and Prevention (2004). Economic costs associated with mental retardation, cerebral palsy, hearing loss, and vision impairment—United States, 2003. Morbidity and Mortality Weekly Report, 53(3), 57–59. Chen, W., Landau, S., Sham, P., & Fombonne, E. (2004). No evidence for links between autism, MMR and measles virus. Psychological Medicine, 34, 543–553. Dales, L., Hammer, S. J., & Smith, N. J. (2001). Time trends in autism and in MMR immunization coverage in California. Journal of American Medical Association, 285, 1183–1185. Decoufle, P., & Autry, A. (2002). Increased mortality in children and adolescents with developmental disabilities. Paediatric and Perinatal Epidemiology, 16(4), 375–382. Decoufle, P., & Boyle, C. A. (1995). The relationship between maternal education and mental retardation in 10‐year‐old children. Annals of Epidemiology, 5(5), 347–353. Decoufle, P., Murphy, C. C., Drews, C. D., & Yeargin‐Allsopp, M. (1993). Mental retardation in ten‐year‐old children in relation to their mothers’ employment during pregnancy. American Journal of Industrial Medicine, 24(5), 567–586. Decoufle, P., Boyle, C. A., Paulozzi, L. J., & Lary, J. M. (2001). Increased risk for developmental disabilities in children who have major birth defects: A population‐based study. Pediatrics, 108(3), 728–734. DeStefano, F., Bhasin, T. K., Thompson, W. W., Yeargin‐Allsopp, M., & Boyle, C. (2004). Age at first measles‐mumps‐rubella vaccination in children with autism and school‐matched control subjects: A population‐based study in metropolitan Atlanta. Pediatrics, 113(2), 259–266. Drews, C. D., Yeargin‐Allsopp, M., Murphy, C. C., & Decoufle, P. (1992). Legal blindness among 10‐year‐old children in metropolitan Atlanta: Prevalence, 1985 to 1987. American Journal of Public Health, 82(10), 1377–1379. Drews, C. D., Yeargin‐Allsopp, M., Murphy, C. C., & Decoufle, P. (1994). Hearing impairment among 10‐year‐old children: Metropolitan Atlanta, 1985 through 1987. American Journal of Public Health, 84(7), 1164–1166. Drews, C. D., Yeargin‐Allsopp, M., Decoufle, P., & Murphy, C. C. (1995). Variation in the influence of selected sociodemographic risk factors for mental retardation. American Journal of Public Health, 85(3), 329–334. Drews, C. D., Murphy, C. C., Yeargin‐Allsopp, M., & Decoufle, P. (1996). The relationship between idiopathic mental retardation and maternal smoking during pregnancy. Pediatrics, 97(4), 547–553. Fleiss, J. (1981). Statistical methods for rates and proportions. New York: John Wiley and Sons. Fombonne, E. (1999). The epidemiology of autism: A review. Psychological Medicine, 29(4), 769–786. Fombonne, E. (2003). Epidemiological surveys of autism and other pervasive developmental disorders: An update. Journal of Autism and Developmental Disorders, 33, 365–382. Georgia Department of Human Resources (1985). Georgia vital statistics reports. Gillberg, C., & Wing, L. (1999). Autism; not an extremely rare disorder. Acta Psychiatrica Scandinavica, 99, 399–406. Kaye, J. A., del Mar Melero‐Montes, M., & Jick, H. (2001). Mumps, measles, and rubella vaccine and the incidence of autism recorded by general practitioners: A time trend analysis. British Medical Journal, 322, 460–463.

PUBLIC HEALTH IMPACT

189

Kuenneth, C. A., Boyle, C., Murphy, C. C., & Yeargin‐Allsopp, M. (1996). Reproductive risk factors for epilepsy among ten‐year‐old children in metropolitan Atlanta. Paediatric and Perinatal Epidemiology, 10(2), 186–196. MacKeith, R. C. (1959). Memorandum on terminology and classification of ‘cerebral palsy’. Cerebral Palsy Bulletin, 1, 27–35. Mervis, C. A., Decoufle, P., Murphy, C. C., & Yeargin‐Allsopp, M. (1995). Low birthweight and the risk for mental retardation later in childhood. Paediatric and Perinatal Epidemiology, 9(4), 455–468. Mervis, C. A., Yeargin‐Allsopp, M., Winter, S., & Boyle, C. (2000). Aetiology of childhood vision impairment, metropolitan Atlanta, 1991–1993. Paediatric and Perinatal Epidemiology, 14(1), 70–77. Mervis, C. A., Boyle, C. A., & Yeargin‐Allsopp, M. (2002). Prevalence and selected characteristics of childhood vision impairment. Developmental Medicine & Child Neurology, 44(8), 538–541. Mittendorf, R., Dambrosia, J., Pryde, P. G., Lee, K. S., Gianopoulos, J. G., Besinger, R. E., et al. (2002). Association between the use of antenatal magnesium sulfate in preterm labor and adverse health outcomes in infants. American Journal of Obstetrics and Gynecology, 186, 1111–1118. Murphy, C. C., Yeargin‐Allsopp, M., Decoufle, P., & Drews, C. D. (1993). Prevalence of cerebral palsy among ten‐year‐old children in metropolitan Atlanta, 1985 through 1987. Journal of Pediatrics, 123(5), S13–S20. Murphy, C. C., Trevathan, E., & Yeargin‐Allsopp, M. (1995). Prevalence of epilepsy and epileptic seizures in 10‐year‐old children: Results from the metropolitan Atlanta developmental disabilities study. Epilepsia, 36(9), 866–872. Murphy, C. C., Yeargin‐Allsopp, M., Decoufle, P., & Drews, C. D. (1995). The administrative prevalence of mental retardation in 10‐year‐old children in metropolitan Atlanta, 1985 through 1987. American Journal of Public Health, 85(3), 319–323. National Center for Health Statistics (2005). Available: http://www.cdc.gov/nchs/about/major/ dvs/popbridge/datadoc.htm#inter1 Puckett, C. D. (1993). The educational annotation of ICD‐9‐CM (4th ed.). Reno, Nevada: Channel Publishing, Ltd. Roeser, R. (1988). Audiometric and immittance measures: Principles and interpretation. In R. J. Roeser (Ed.), Auditory disorders in school children (2nd ed., pp. 1–34). New York: Thieme Medical Publishers. Rurangirwa, J. K., Van Naarden Braun, K., Schendel, D. E., & Yeargin‐Allsopp, M. (2006). Healthy behaviors and lifestyles in young adults with a history of developmental disabilities. Research in Developmental Disabilities, 27(4), 381–399. Schendel, D. E., Berg, C. J., Yeargin‐Allsopp, M., Boyle, C. A., & Decoufle, P. (1996). Prenatal magnesium sulfate exposure and the risk for cerebral palsy or mental retardation among very low‐birth‐weight children aged 3 to 5 years. Journal of American Medical Association, 276(22), 1805–1810. Scher, A. I., Petterson, B., Blair, E., Ellenberg, J. H., Grether, J. K., Haan, E., et al. (2002). The risk of mortality or cerebral palsy in twins: A collaborative population‐based study. Pediatric Research, 52(5), 671–681. Taylor, B., Miller, E., Farrington, C. P., Petropoulos, M. C., Favot‐Mayaud, I., Li, J., et al. (1999). Autism and measles, mumps, and rubella vaccine: No epidemiological evidence for a causal association. Lancet, 353, 2026–2029. Trevathan, E., Murphy, C. C., & Yeargin‐Allsopp, M. (1997). Prevalence and descriptive epidemiology of Lennox‐Gastaut syndrome among Atlanta children. Epilepsia, 38(12), 1283–1288.

190

Rachel Nonkin Avchen et al.

Trevathan, E., Murphy, C. C., & Yeargin‐Allsopp, M. (1999). The descriptive epidemiology of infantile spasms among Atlanta children. Epilepsia, 40(6), 748–751. US Congress (2000). Children’s Health Act of 2000. Public Law No. 106–310. Available: http:// www7.nationalacademies.org/ocga/laws/PL106_310.asp US Congress (2004). Individuals with Disabilities Education Improvement Act of 2004 (IDEA). Public Law 94–142. Available: http://www.ed.gov/policy/speced/guid/idea/idea2004.html Van Naarden Braun, K., & Decoufle, P. (1999). Relative and attributable risks for moderate to profound bilateral sensorineural hearing impairment associated with lower birth weight in children 3 to 10 years old. Pediatrics, 104(4 Pt. 1), 905–910. Van Naarden Braun, K., Decoufle, P., & Caldwell, K. (1999). Prevalence and characteristics of children with serious hearing impairment in metropolitan Atlanta, 1991–1993. Pediatrics, 103(3), 570–575. Van Naarden Braun, K., Yeargin‐Allsopp, M., Schendel, D., & FernhoV, P. (2003). Long‐term developmental outcomes of children identified through a newborn screening program with a metabolic or endocrine disorder: A population‐based approach. Journal of Pediatrics, 143(2), 236–242. Van Naarden Braun, K., Autry, A., & Boyle, C. (2005). A population‐based study of the recurrence of developmental disabilities—Metropolitan Atlanta Developmental Disabilities Surveillance Program, 1991–1994. Paediatric and Perinatal Epidemiology, 19, 69–79. Van Naarden Braun, K., Yeargin‐Allsopp, M., & Lollar, D. (2005). Factors associated with leisure activity among young adults with developmental disabilities. Research in Developmental Disabilities. Nov. 6 [Epub ahead of print]. Van Naarden Braun, K., Yeargin‐Allsopp, M., & Lollar, D. (2006). A multi‐dimensional approach to the transition of children with developmental disabilities into young adulthood: The acquisition of adult social roles. Disability and Rehabilitation, 28(15), 915–928. Van Naarden Braun, K., Yeargin‐Allsopp, M., & Lollar, D. (in press). A population‐based study of activity limitations among young adults with developmental disabilities. Journal of Developmental Epidemiology. Williams, L. O., & Decoufle, P. (1999). Is maternal age a risk factor for mental retardation among children? American Journal of Epidemiology, 149(9), 814–823. Winter, S., Autry, A., Boyle, C., & Yeargin‐Allsopp, M. (2002). Trends in the prevalence of cerebral palsy in a population‐based study. Pediatrics, 110(6), 1220–1225. Yeargin‐Allsopp, M., Murphy, C. C., Oakley, G. P., & Sikes, R. K. (1992). A multiple‐source method for studying the prevalence of developmental disabilities in children: The Metropolitan Atlanta Developmental Disabilities Study. Pediatrics, 89(4 Pt. 1), 624–630. [Erratum in Pediatrics 1992 December, 90(6), 1001]. Yeargin‐Allsopp, M., Drews, C. D., Decoufle, P., & Murphy, C. C. (1995). Mild mental retardation in black and white children in metropolitan Atlanta: A case‐control study. American Journal of Public Health, 85(3), 324–328. Yeargin‐Allsopp, M., Murphy, C. C., Cordero, J. F., Decoufle, P., & Hollowell, J. G. (1997). Reported biomedical causes and associated medical conditions for mental retardation among 10‐year‐old children, metropolitan Atlanta, 1985 to 1987. Developmental Medicine and Child Neurology, 39(3), 142–149. Yeargin‐Allsopp, M., Rice, C., Karapurkar, T., Doernberg, N., Boyle, C., & Murphy, C. (2003). Prevalence of autism in a US metropolitan area. Journal of American Medical Association, 289(1), 49–55.

Using GIS to Investigate the Role of Recreation and Leisure Activities in the Prevention of Emotional and Behavioral Disorders TINA L. STANTON‐CHAPMAN DEPARTMENT OF CURRICULUM, INSTRUCTION, AND SPECIAL EDUCATION CURRY SCHOOL OF EDUCATION, UNIVERSITY OF VIRGINIA CHARLOTTESVILLE, VIRGINIA

DEREK A. CHAPMAN DEPARTMENT OF EPIDEMIOLOGY VIRGINIA COMMONWEALTH UNIVERSITY, RICHMOND, VIRGINIA

I.

INTRODUCTION

The early recognition of emotional and behavioral problems can prevent the development of future associated issues. However, the causes of many of these problems are not easily determined (Baglin, 2004). While estimates vary, Surgeon General David Satcher reported that 16–22% of children had mental health problems (Surgeon General of the United States, 2001). Many of these children are referred to as emotionally disturbed (ED), socially maladjusted, conduct disordered, or antisocial oppositional. If their educational performance is aVected by the disorder, the children may eventually be labeled seriously ED by the school district. The Individuals with Disabilities Education Act in 1997 (IDEA, 1997) defines ED as: 1. The term means a condition exhibiting one or more of the following characteristics over a long period of time and to a marked degree, which adversely aVects educational performance. INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

191

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33008-X

192

Tina L. Stanton‐Chapman and Derek A. Chapman

a. An inability to learn which cannot be explained by intellectual, sensory, or health factors. b. An inability to build or maintain satisfactory interpersonal relationships with peers and teachers. c. Inappropriate types of behavior or feelings under normal circumstances. d. A general, pervasive mood of unhappiness or depression. e. A tendency to develop physical symptoms or fears associated with personal or school problems. 2. The term includes children who are schizophrenic or autistic. The term does not include children who are socially maladjusted, unless it is determined that they are seriously ED. Federal reports typically include the number of students with ED counted at a particular time during the school year. For the school year 2000–2001, 473,663 children were identified as students with ED in the public schools (U.S. Department of Education, 2002). Prevalence rates for school‐identified ED are estimated to range between 3% and 6% (Kaufman, 2001). As the number of children identified with ED increases each school year, teachers and administrators are becoming overwhelmed. It is, therefore, critical to identify those children who are most at‐risk for ED. Although ED has often been considered as a ‘‘child’’ problem, a separate but parallel line of enquiry has developed in epidemiology which focuses on the role that a neighborhood might play as a determinant of ED. Traditional thinking about ED has emphasized behavioral change among individuals as a means to reduce risk. Such approaches appear to have had notable successes in reducing the aggregate level of problem behavior. However, neighborhood‐ level prevention may yield similar payoVs that complement traditional, individual approaches (Sampson & MorenoV, 2000). For example, Block (1991) used a neighborhood‐level prevention system entitled ‘‘early warning system’’ to prevent gang homicides. The early warning system plotted each homicide incident and used mapping and statistical clustering procedures to allow police to identify possible neighborhood crisis areas at‐risk for gang violence incidents. With rapid dissemination of information, police could intervene in ‘‘hot spot’’ areas to pacify emerging trouble. Places could also be put under periodic surveillance to reduce opportunities for trouble to occur. The idea of ‘‘hot spots’’ suggests a neighborhood‐level response may be more eVective than policies that simply target individuals or families. By responding proactively to neighborhoods and places that disproportionately generate crimes, intervention strategies can more eYciently stave oV ‘‘epidemics’’ and their spatial diVusion (Sampson & MorenoV, 2000).

ROLE OF RECREATION

193

One neighborhood‐level intervention strategy worth considering for ED prevention involves community‐based leisure activities. This approach encourages social interaction in the hope that it will in turn lead to greater cohesion, shared expectations, and social control (Sampson & MorenoV, 2000). Such initiatives include building prosocial influences within the community and community organizing through voluntary associations. Unfortunately, there is no convincing evidence that such approaches reduce rates of ED in children. However, several hypotheses do exist as to why some children at high‐risk for ED avoid later involvement in violence and delinquency (Loeber & Stouthamer‐Loeber, 1998). One such hypothesis is called the social development model. According to the social development model, the risk of antisocial behavior in adolescence is reduced when youths encounter prosocial influences in their neighborhood, families, schools, and peer networks; when they are actively involved in prosocial institutions; when bonds to prosocial institutions develop; and when youths acquire beliefs that support positive behaviors (Catalano & Hawkins, 1996). The model covers four periods of a child’s development— birth through high school—and the social interactions children have during each period determine whether a child selects a prosocial path or an antisocial path. Positive social interactions lead to a prosocial path whereas negative social interactions lead to an antisocial path (Catalano & Hawkins, 1996). The antisocial behavior predicted includes aggressive behavior, behavior problems at home or school, juvenile delinquency, and ED. The development of geographic information system (GIS) has provided new tools for research into neighborhood‐ and community‐level factors. Advances in computing power and graphics, as well as the development of GIS‐based locational analysis models and methods (McLaVerty, 2003), provide exciting opportunities to empirically test theories of neighborhood eVects on the development of social and emotional disorders in children. Using GIS technology, researchers can include in analyses contextual factors about the neighborhood in which an individual lives, target interventions to high‐risk areas, and visualize and measure access to resources in the community. The purpose of this chapter is to discuss the importance of community‐ level factors, such as access to recreational resources, in the design of interventions for children at‐risk for ED and the application of GIS technology to this topic. First, we present evidence that neighborhood factors relate to lessened levels of childhood ED. Second, we describe what is meant by neighborhood‐based leisure activities and how these activities might decrease the prevalence of ED. Third, we define GIS and describe three basic GIS functions that relate to the study of community‐level factors.

194

Tina L. Stanton‐Chapman and Derek A. Chapman

Finally, we conclude by presenting the promises and challenges of using GIS to study emotional and behavioral disorders.

II. NEIGHBORHOOD INFLUENCES: EFFECTS ON CHILDHOOD EMOTIONAL/BEHAVIORAL DISORDERS A.

Significance of the Neighborhood

Theoretical and empirical examinations of neighborhood influences on individual development are still at a rudimentary level (Elliott et al., 1996). One possible reason for the diYculty in establishing neighborhood‐level eVects is the vague nature of the term ‘‘neighborhood.’’ In the current literature, no one has found a completely adequate operational definition for ‘‘neighborhood.’’ Individuals living in the same household may define their neighborhood as vastly diVerent in size and scope, so that one person’s definition incorporates high‐risk factors while another person’s does not (Rutter & Giller, 1983). This leads to accurate, although contradictory, impressions of a neighborhood. One definition of neighborhood, which has generally been accepted in the urban studies literature, is Lancaster’s definition. In this literature, a neighborhood is a bundle of spatially based attributes associated with a cluster of residences, sometimes in conjunction with other land uses (Lancaster, 1966). The attributes discussed in this definition consist of: structural characteristics of the residential and nonresidential buildings

such as type, scale, materials, state of repair, and density

infrastructural characteristics such as roads and sidewalks demographic characteristics of the residents (e.g., age distribution, family

composition, racial, ethnic, and religious types)

class status characteristics of the resident population (e.g., income,

occupation, education composition)

tax/public service package characteristics such as the quality of safety

forces, public schools, public administration, and parks and recreation

environmental characteristics (e.g., degree of land, air, and water

pollution)

access to major destinations of employment, shopping, and entertainment

such as parks and recreation. It must be emphasized that, while most of the attributes listed above usually are present to some extent in all neighborhoods, the quantity and composition of constituent attributes typically vary dramatically across neighborhoods within a single area (Galster, 2001). This implies that neighborhoods can be distinctly categorized by type and/or quality and the social characteristics of

ROLE OF RECREATION

195

such neighborhoods vary systematically along dimensions of socioeconomic status (SES), family structure and life cycle (e.g., female‐headed households, number of children in family), residential stability (e.g., home ownership), and racial/ethnic composition. A recent review of neighborhood eVects on child outcomes suggests that community characteristics are related to behavior problems (Leventhal & Brooks‐Gunn, 2000). They found that the strongest eVects appear to be for the adverse eVect of low‐SES neighbors on children’s externalizing problems. For example, Brooks‐Gunn, Duncan, Klebenov, and Sealand (1993) reported that, among 3‐year‐olds, low numbers of professional and managerial workers in the neighborhood were associated with higher amounts of reported behavior problems. Similarly, the presence of low‐income neighbors was associated with increased amounts of externalizing behavior problems among 5‐ to 6‐year‐old children (Chase‐Lansdale, Gordon, Brooks‐Gunn, & Klebanov, 1997). Kupersmidt, Griesler, DeRosier, Patterson, and Davis (1995) examined peer‐reported aggression and peer rejection in a sample of elementary school children. Their findings suggested that low‐SES children living in single‐ parent families were more aggressive than children living in single‐parent families in high‐SES neighborhoods. These same children (low‐SES children from single‐parent families) were more likely to be rejected by their peers than their middle‐SES comparison peers. A study by Peeples and Loeber (1994), focusing on the relationship between race and oVending in low‐SES and middle‐SES Pittsburgh neighborhoods, showed that the relationship between race and oVending was also dependent on the type of neighborhood. For 13‐ and 16‐year‐old African‐ American boys, those living in low‐SES neighborhoods were more likely to elicit delinquent and criminal behavior than those living in higher SES neighborhoods. Their delinquent behavior was also found to be more severe and committed more frequently. Based on the findings from these studies, it is clear that economic disadvantage is a factor in the development of problem behavior. What is not known from these studies is the extent to which the children’s educational performance is aVected by their problem behavior and whether any of these children were identified by the schools as having ED. However, we do know, from previous studies, that poverty itself is the best predictor of school failure (Hodgkinson, 1995). B.

Area Effects and Their Influence on Emotional and Behavioral Disorders

Neighborhood poverty exacerbates the eVects of family poverty through ‘‘social contagion’’ and ‘‘collective socialization’’ (Sampson & MorenoV, 1997). Essentially, communities are thought of as a social environment.

196

Tina L. Stanton‐Chapman and Derek A. Chapman

Negative social environments may deprive children of positive social supports and expose them to antisocial behaviors, while positive environments have the opposite eVect. For example, persistent unemployment in the neighborhood may provide models of joblessness (Wilson, 1996) and reduce children’s motivation for school achievement (Albee & Gullotta, 1996), whereas having a parent graduate from a university might promote the value of an education in young children. Thus, the belief that where people live aVects outcomes, such as education, behavior, and health, is emphasized by collective socialization and one way it is studied is by examining area eVects. Area eVects can be defined as the change in the contribution to life‐chances made by living in one area rather than another (Atkinson & Kintrea, 2001). Area eVects studies generally have examined the social isolation of the poor in ‘‘ghettos’’ which stress short‐term goals and deviant norms (Murray, 1996), a lack of role‐ models for children by the absence of a successful middle class (Wilson, 1987, 1996), and underfunded and poor‐quality services (DuVy, 2000). Typically these studies use either administrative boundaries, such as zip codes, or census tracts as a way to evaluate area eVects (Brody et al., 2001).

III. A.

THE IMPORTANCE OF NEIGHBORHOOD‐RELATED LEISURE ACTIVITIES

Recreation and Leisure Activities as an Intervention

Interventions which target neighborhood risk factors may take the form of recreation and leisure activities. Recreation is defined ‘‘as an activity that requires physical skills, rules, and either a score or end result’’ (Williams, 2004, p. 67). Recreational sports are played with a cooperative spirit with the intent of having fun. Examples of recreational sports include five‐on‐five basketball, basketball shooting game of HORSE, Putt‐Putt golf, and a homerun derby in baseball. To date, there is little intervention data on recreational activities, in contrast with other aspects of children’s play (Pellegrini & Smith, 1998; Rubin, Fein, & Vandenberg, 1983). The little research conducted has been with children who have identified ED and are case studies in nature. For instance, Williams (2004) presents a case study regarding a 13‐year‐old boy with ED who recently joined a recreational basketball team. The adolescent demonstrates hostile behavior during competition, and thus, his educational team developed an intervention plan to prevent his removal from the team. The intervention plan consisted of a five‐level system, with level 5 being unlimited competition by the child to level 1 which meant he could only shoot baskets independently outside of team play. If the boy showed no hostile behavior

ROLE OF RECREATION

197

during practice sessions, he would be able to fully participate in the weekly game. However, if he demonstrated hostile behavior during practice sessions, the coach was at liberty to reduce the boy’s playing time in the weekend game. The results showed that the boy fully participated in 6 out of 10 games and was able to remain on the recreational team for the complete season. In contrast to recreational sports, leisure activities encompass activities that include physical and/or cognitive participation (Williams, 2004). Leisurely activities usually do not include keeping score, and typically involve ‘‘standards of practice.’’ For example, while leisure rock climbers do not keep score, they adhere to a variety of ethical and safety standards established by the climbing industry. Leisurely activities also share the goals of relaxation and exercise (Williams, 2004). Examples of leisure activities include walking, running, rock climbing, dancing, visiting a park, bike riding, and exercise. Similar to recreational activities, no research has been done in terms of interventions to prevent the development of ED using leisurely activities. Work that has been conducted involved children who have already been identified as having ED. Williams (2004) describes a case study in which children with ED took part in a risk and adventure thematic unit plan at a summer camp. The thematic unit plan lasted 4 weeks and each day was structured so that the events and activities built upon the days before, and led to a grand finale during the last week. The thematic curriculum provided organization and structure to the summer camp environment and tied together cognitive, motor, and aVective skills. The curriculum was successful because the campers were oVered choices during the traditionally competitive units like soccer and basketball. Campers had the choice of competing with peers or taking part in a more leisurely activity. Children with ED frequently chose the alternatives if they knew they had diYculty with competitive environments. Thus, they avoided situations which may have led to negative behaviors during the competitive game (Williams, 2004). The findings from this research and descriptive work show that there are at least two important areas to consider when designing and/or implementing interventions for children at‐risk for ED. The first area is to promote organized, structured, and cooperative activities for children (Bay‐Hinitz, Peterson, & Quilitch, 1994; Murphy, Hutchinson, & Bailey, 1983; Nabors, Willoughby, LeV, & McMenamen, 2001; Williams, 2004); these activities can have a beneficial eVect on children’s subsequent behaviors. For example, Bay‐Hinitz et al. (1994) found that preschoolers demonstrated increases in cooperative actions and decreases in aggressive behaviors when involved in cooperative games, but demonstrated more aggression and less cooperative actions when involved in competitive games. Because organized games typically specify rules for relating to one another and discourage the use of aggressive behavior, the use of games and other organized activities may

198

Tina L. Stanton‐Chapman and Derek A. Chapman

lead to a reduction in aggressive behavior and the promotion of cooperative forms of play (LeV, Costigan, & Power, 2004). The second area to consider is providing active monitoring of children’s recreational and leisure behavior as well as encouraging children to act cooperatively (Colvin, Sugai, Good, & Lee, 1997; Roderick, Pitchford, & Miller, 1997). For example, Roderick et al. trained playground supervisors to distribute raZe tickets to those children who exhibited positive behaviors on the playground. This positive reinforcement strategy resulted in a 75% reduction in kicking and a 47% reduction in kicking and a 47% reduction in hitting on the playground. Thus, by providing adult supervision and encouraging cooperative play, professionals have an eVective method for shaping children’s behavior. B.

Community Resource Mapping

If recreational and leisurely activities are beneficial in shaping children’s behavior, then how do early interventionists find appropriate neighborhood resources? The answer to this question is community resource mapping (Trivette, Dunst, & Deal, 1997). Developing a community map involves identifying the various kinds of resources that exist in a given locale (e.g., neighborhood, village, county). This approach focuses on what communities have to oVer by identifying their existing assets and resources as well as highlighting what communities lack and the problems they face (Trivette et al., 1997). Community resource mapping also involves identifying the location of each resource, which will then serve as a source from which families can find and access services they deem important. Examples of community resources include YMCAs, parks, playgrounds, and basketball courts. To conduct a resource mapping for a given community, early interventionists complete a ‘‘capacity inventory’’ to assess the potential of citizens, organizations, and associations (Kretzmann & McKnight, 1993). The inventory collects data through a series of questions (i.e., What is the address of the family? How many miles from the family address do you want to search? What types of community resources are you searching for?). When the questions are answered, the inventory is used to create a picture or map of the capacities or assets existing in the community. IV. A.

GEOGRAPHICAL INFORMATION SYSTEMS: BASIC CONCEPTS

Geographic Information System Technology

GIS refers to a ‘‘computer system for the input, editing, storage, retrieval, analysis, synthesis, and output of location‐based information’’ (Croner, 2003, p. 76). Information that would be diYcult to associate by any other

ROLE OF RECREATION

199

means can be visually presented and analyzed using GIS. While most widely known for producing the visual displays seen in atlases and road maps, map making is actually the end result of one of many data analysis functions performed by a GIS. Every feature drawn on a map represents a row in an underlying data table containing geospatial information. Maps are simply visual representations of data in the same way that a graph or chart represents data from a research study or survey. For example, a U.S. state map (including Washington, DC) has an underlying data table with 51 rows, each of which contains the information needed to display the state boundaries. GIS represents real‐world entities graphically using three basic shapes: areas (e.g., counties, zip codes), lines (e.g., streets, bus routes), and points (e.g., schools, hospitals). These shapes are referenced by GIS software using a data format known as a shapefile. Many shapefiles are either included with GIS software or can be downloaded at no charge from websites (e.g., www.geodata.gov, www.census.gov, and www.geographynetwork.com). Since all GIS data are georeferenced to the correct location on the earth’s surface using a coordinate system (e.g., the commonly used Geographic Coordinate System measured in longitude and latitude), multiple layers of data can be seamlessly superimposed upon another in a single display. Once loaded into a GIS software program, a multitude of analysis and mapping tools are also available. A full discussion of GIS analysis tools and procedures is well beyond the scope of this chapter (see Kirby, this issue), so what follows is a brief description of three basic functions of GIS that compliment epidemiological research. A more detailed discussion of GIS and spatial analytic techniques in an epidemiological context can be found elsewhere (Lawson et al., 1999; Rushton, 2003). 1. MAPPING

One of the most basic GIS functions is the thematic display of quantitative data by area from one or more data sources. Population counts and other demographic information are included with shapefiles of most political boundaries. Through the data tables underlying these areas, researchers can also display and analyze their own georeferenced data. For example, one could create a U.S. state map shaded or colored based on the proportion of children in special education by simply making a data table with the state name and special education rate and joining it to the table associated with the state shapefile using the state name as the keyed or matching field. 2. GEOCODING

The ability to geocode, or assign a map coordinate to a real‐world location, is a second basic GIS function that is critical to epidemiological research. The coordinates associated with a particular location can be directly obtained via a global positioning system (GPS) device, which determines longitude and

200

Tina L. Stanton‐Chapman and Derek A. Chapman

latitude by computing its distance to at least 4 of the approximately 24 satellites orbiting the earth that constantly transmit their position for this purpose. However, epidemiologists typically work with larger datasets, so carrying a GPS device to each location can be impractical. More commonly, locations are assigned map coordinates using a process called address geocoding. In address geocoding, street addresses are electronically matched to a GIS street map database [e.g., the Census Bureau’s Topologically Integrated Geographic Encoding and Referencing (TIGER) database] containing the longitude and latitude of all known roads broken into small segments. This method is not as precise as a GPS device because some interpolation is required. For example, 150 Main Street would be assigned a point halfway along the even numbered side of the 100–199 Main Street segment. To the extent that houses are not evenly distributed along a given segment or in cases where even numbers are not all on the same side of the road, some degree of error is introduced using this method, which is usually negligible for most research purposes. 3. MEASURING DISTANCE

A third basic GIS function is the ability to calculate distances among the various features on a map. The simplest way to compute distance between two locations is with a straight‐line or Euclidian measurement derived from the underlying map coordinates. Euclidean distance is also commonly used to identify equidistant buVer areas around a given location (e.g., a 5‐mile radius around a clinic to denote a catchment area). It should be noted that buVer functions in a GIS are adequate for measuring walking distance but fail to account for the ease, cost, and time of travel for other means of transportation such as car or bus travel (Martin, Wrigley, Barnett, & Roderick, 2002). Researchers interested in studying actual travel time between locations can use specialized GIS software to incorporate data on traYc patterns and cost into estimates (Lovett, Haynes, Sunnenberg, & Gale, 2002; McLaVerty, 2003; Yang, George, & Mullner, 2006).

V.

USING GIS TO STUDY EMOTIONAL AND BEHAVIORAL DISORDERS: PROMISES AND CHALLENGES

As early intervention researchers shift their focus toward neighborhood‐ or community‐level contextual factors, GIS technology will play an increasing role. A number of promising research opportunities related to the study of emotional and behavioral disorders are possible using just the basic GIS

ROLE OF RECREATION

201

functions of mapping, geocoding, and distance measurement and geographic data available at little or no cost.

A.

Epidemiological Studies

Within the field of epidemiology there has been a call for a return to a more traditional focus on the cultural, social, and environmental influences on the health of populations (Kaplan, 1996; Schwartz, 1994; Susser, 1994). In order to fully understand health behaviors of individuals, broader contextual factors need to be considered. For example, prevention of unintentional injuries involves much more than educating individuals about risk‐taking behaviors. Policies and laws (e.g., seat belt and handgun laws), built environments (e.g., type of materials used to construct playground equipment and their surfaces), and familial factors (e.g., parental attitudes and degree of supervision) also play a key role in the prevention of unintentional injuries (ChristoVel & Gallagher, 1999). GIS technology can provide a useful tool to researchers for investigating the relationship between community‐level factors and developmental disabilities, including emotional and behavioral disorders. Using address geocoding and mapping functions in a GIS, any geographic layer containing socioeconomic or other community‐level variables can be joined to individual data points. For example, the percent of children living in poverty or median rent could be determined for each child’s neighborhood in a sample of children with behavioral disorders and a comparison group. The researcher would geocode the children’s addresses and overlay a map layer which included neighborhood boundaries and community‐level economic data. Fortunately, publicly available data collected by the U.S. Census Bureau in April 2000 provide scores of contextual variables from an overall sample of one in every six persons living in the United States (U.S. Census Bureau, 2002). A summary of the social and economic characteristics found in Census 2000 data are listed in Table I. These variables are aggregated across various hierarchical census geographies including states, counties, tracts, and block groups. Census tracts are small, relatively permanent statistical subdivisions of a county designed to be homogeneous with respect to demographic characteristics and averaging about 4000 persons (U.S. Census Bureau, 2001). Census block groups are statistical subdivisions of a tract containing between 300 and 3000 people with an optimum size of 1500 (U.S. Census Bureau, 2002). Map layers of all census geographies and associated summary data can be downloaded at no charge from the U.S. Census Bureau’s website (www.census.gov).

202

Tina L. Stanton‐Chapman and Derek A. Chapman TABLE I SELECTED COMMUNITY‐LEVEL VARIABLES

IN

CENSUS 2000 DATA

Social characteristics

Economic characteristics

Marital status School enrollment Educational attainment Residence 5 years ago (migration) Language spoken at home Disability Grandparents as caregivers

Labor force status Place of work Occupation, industry, and class of worker Work status in 1999 Income in 1999 Poverty status in 1999 Value of home or monthly rent paid

B.

Intervention Planning

Through the targeting of limited resources to high‐risk populations, GIS can play an important role in intervention planning. Miranda, Dolinoy, and Overstreet (2002) provide an excellent example of applying GIS models to highlight critical areas for targeted childhood lead poisoning intervention in six North Carolina counties. Demographic data from the U.S. Census Bureau and residential tax parcel data from county tax assessor oYces were combined with data from the North Carolina Childhood Lead Poisoning Prevention Program using GIS to predict statistically based lead exposure risk levels. The same type of analysis can be applied to intervention planning for emotional and behavioral disorders. For example, based on Kupersmidt et al. (1995), one may want to identify areas of the community with a high number of homes with a female head of household. An example of these data mapped at the census block group level for the City of Richmond, Virginia is provided in Fig. 1.

C.

Resource Maps

GIS is an essential tool for community resource mapping. Features, political boundaries, and resource locations can be easily symbolized, added, and removed from a map as needed. Pearce, Witten, and Bartie (2006) used GIS to investigate regional variations in accessibility to shopping, education, recreation, and health facilities in New Zealand. Figure 2 is an example of a community resource map for Broward County, Florida. GIS data pertaining to the county boundary, cities, major roads, streets, bodies of water, and parks were downloaded at no cost from the Broward County website

ROLE OF RECREATION

203

FIG. 1. Number of female‐headed households with children by census block group, City of Richmond, Virginia.

(http://gis.broward.org/). Recreational resources in Fig. 2 refer to skating facilities and skate parks, swimming pools, miniature golf locations, and YMCA centers. Addresses of recreational resources were obtained from phone book entries at www.yellowbook.com and were geocoded using E‐Z Locate Trial software downloaded from www.geocode.com.

204

Tina L. Stanton‐Chapman and Derek A. Chapman

FIG. 2. Selected recreation resources, Broward County, Florida.

D.

Community Planning

Area‐based measures of geographical access are commonly used in public health to assess community‐level needs. Measures like the physician to population ratio are used to determine health professional shortage areas. For

ROLE OF RECREATION

205

example, Susi and Mascarenhas (2002) recently used GIS to describe geographical disparities in access to dental care in rural and Appalachian areas of Ohio by mapping the ratio of dentists to population by county and zip code. If research demonstrates the eVectiveness of having a YMCA or skate park within walking distance of youths’ homes, community planners can identify areas in need of additional recreational resources. Figure 3 clearly identifies areas of the county which are not within a short distance of any of the 31 recreation resource points from Fig. 2 by displaying a ¼ and ½ mile buVer around each point. Public school locations are also shown in Fig. 3 using data provided with the Arcview 9.1 GIS software used to create the maps in this chapter. E.

Challenges

With any relatively new technology, there are always challenges associated with its use. One such challenge not unique to GIS is the quality of the underlying data. Incomplete, misspelled, or missing street addresses cannot be properly geocoded. Before geocoding, it is critical to standardize street addresses (e.g., removing special characters, using standard U.S. Post OYce abbreviations, separating apartment numbers from the street name). Similarly, the quality of data in the street map database used as a reference for geocoding aVects data accuracy. Most street map databases used in address geocoding are based on the Census Bureau’s TIGER database. However, many commercially available street map software packages include additional data cleaning and updates, which allow them to consistently outperform publicly available street map databases. A second major challenge is that the release of geospatial data has been constrained by confidentiality concerns and the protection of an individual’s identity (Croner, 2003). In general, while state statutes generally permit the release of patient‐level data without informed consent to researchers conducting certain institutional review board approved studies of public health significance. There is, however, no such provision for educational records. The Family Educational Rights and Privacy Act (FERPA) of 1974, also known as the Buckley Amendment, prohibits the release of student education records without signed consent to researchers unless the studies are being conducted for or on behalf of the educational agency or institution in order to: (1) develop, validate, or administer predictive tests; (2) administer student aid programs; or (3) improve instruction. FERPA’s legal statute citation can be found in the U.S. Code (20 USC 1232g), regulations are found in the Federal Register (34 CFR Part 99), and FERPA’s 1994 amendments are found in Public Law (P.L.) 103–382. Several recent population‐based epidemiological studies of developmental disabilities have relied predominantly on administrative data from the public

206

Tina L. Stanton‐Chapman and Derek A. Chapman

FIG. 3. Selected recreation resources with ¼ and ½ mile buVer in relation to public schools, Broward County, Florida.

school system (Chapman, Scott, & Mason, 2002; Croen, Grether, & Selvin, 2001; Drews, Yeargin‐Allsopp, Decoufle, & Murphy, 1995; Mason, Chapman, & Scott, 1999; Murphy, Boyle, Schendel, Decoufle, & Yeargin‐Allsopp, 1998; Stanton‐Chapman, Chapman, Bainbridge, & Scott, 2002; Stanton‐Chapman,

207

ROLE OF RECREATION

Chapman, & Scott, 2001). Since informed consent is not practical in large‐scale population‐based studies such as these, and most readily accessible school data are aggregated at the district or county level, innovative solutions must be found to provide researchers the data needed to investigate community‐level risk for developmental disabilities. One potential solution involves increased collaboration between researchers and Department of Education oYcials. If school oYcials routinely geocoded student addresses, rates of various exceptionalities could be computed for small areas, such as census tracts and block groups, and provided to researchers for analysis. Several methods could be employed to safeguard these data against inappropriate disclosure including temporal or spatial aggregation, smoothing, or other masking techniques (Rushton & Lolonis, 1996). Ideally, state Departments of Education or local school districts would develop web‐based online querying and mapping systems, which would permit users to query databases containing individual educational records on servers which would use masking techniques to de‐identify the data before returning results.

VI.

CONCLUSIONS

Each year the number of children who are identified in the schools as having emotional and behavioral disorders continues to rise. The literature reviewed in this chapter provides ample evidence that neighborhood‐level risk factors should be considered when interventional programming is being planned. Recreational and leisure activities have been found to be eVective in reducing behavior problems in children. For this reason, it is critical that children in at‐risk neighborhoods have access to such activities. GIS software and information available on the internet (e.g., telephone books, park information) can be used to create community resource maps. Early interventionists can use these maps to locate recreational facilities, identify area strengths and deficiencies (Trivette et al., 1997). They can use this information to inform public policy aimed at improving the quality of neighborhoods. Changing the structural characteristics of neighborhoods, such as the economic and social resources available, is likely to make the most significant changes for children and families in inner‐city communities (Brooks‐Gunn et al., 1993).

208

Tina L. Stanton‐Chapman and Derek A. Chapman REFERENCES

Albee, G. W., & Gullotta, T. P. (1996). Primary prevention’s evolution. In G. W. Albee & T. P. Gullotta (Eds.), Primary prevention works: Issues in children’s and families’ lives (Vol. 6, pp. 3–22). Thousand Oaks, CA: Sage Publications. Atkinson, R., & Kintrea, K. (2001). Disentangling area eVects: Evidence from deprived and non‐deprived neighborhoods. Urban Studies, 38, 2277–2298. Baglin, C. A. (2004). An overview of the problem. In C. A. Baglin, M. E. B. Lewis, & B. Williams (Eds.), Recreation and leisure for persons with emotional problems and challenging behaviors (pp. 1–7). Baltimore: Sagamore Publishing. Bay‐Hinitz, A. K., Peterson, R. F., & Quilitch, H. R. (1994). Cooperative games: A way to modify aggressive and cooperative behaviors in young children. Journal of Applied Behavior Analysis, 27, 435–446. Block, C. (1991). Early warning system for street gang violence crisis areas: Automated hot spot identification in law enforcement. Chicago: Illinois Criminal Justice Information Authority. Brody, G. H., Conger, R., Gibbons, F. X., Ge, X., McBride Murry, V., Gerrard, M., et al. (2001). The influence of neighborhood disadvantage, collective socialization, and parenting on African American children’s aYliation with deviant peers. Child Development, 72, 1231–1246. Brooks‐Gunn, J., Duncan, G. J., Klebenov, P. K., & Sealand, N. S. (1993). Do neighborhoods influence child and adolescent development? American Journal of Sociology, 99, 353–395. Catalano, R. F., & Hawkins, J. D. (1996). The social development model: A theory of antisocial behavior. In J. D. Hawkins (Ed.), Delinquency and crime: Current theories (pp. 149–197). Cambridge, England: Cambridge University Press. Chapman, D. A., Scott, K. G., & Mason, C. A. (2002). Early risk factors for mental retardation: The role of maternal age and maternal education. American Journal on Mental Retardation, 107, 46–59. Chase‐Lansdale, P. L., Gordon, R. A., Brooks‐Gunn, J., & Klebanov, P. K. (1997). Neighborhood and family influences on the intellectual and behavioral competence of preschool and early school‐age children. In J. Brooks‐Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. 1. Context and consequences for children (pp. 79–118). New York: Russell Sage Foundation. ChristoVel, T., & Gallagher, S. S. (Eds.) (1999). Injury prevention and public health: Practical knowledge, skills, and strategies. Gaithersburg, MD: Aspen Publishers, Inc. Colvin, G., Sugai, G., Good, R. H., & Lee, Y. (1997). Using active supervision and precorrection to improve transition behaviors in an elementary school. School Psychology Quarterly, 12, 344–363. Croen, L. A., Grether, J. K., & Selvin, S. (2001). The epidemiology of mental retardation of unknown cause. Pediatrics, 107, E86. Croner, C. M. (2003). Public health, GIS, and the internet. In J. E. Fielding, R. C. Brownson, & N. M. Clark (Eds.), Annual review of public health (Vol. 24, pp. 57–82). Palo Alto, CA: Annual Reviews. Drews, C. D., Yeargin‐Allsopp, M., Decoufle, P., & Murphy, C. C. (1995). Variation in the influence of selected sociodemographic risk factors for mental retardation. American Journal of Public Health, 85, 329–334. DuVy, B. (2000). Satisfaction and expectations: Attitudes to public services in deprived areas. London: London School of Economics. Elliott, D., Wilson, W., Huizinga, D., Sampson, R., Elliot, A., & Ranking, D. (1996). The eVects of neighborhood disadvantage on adolescent development. Journal of Research in Crime and Delinquency, 33, 389–426.

ROLE OF RECREATION

209

Galster, G. (2001). On the nature of neighborhood. Urban Studies, 38, 2111–2124. Hodgkinson, H. L. (1995). What should we call people? Race, class, and the census for 2000. Phi Delta Kappan, 77, 173–179. Kaplan, G. (1996). People and places: Contrasting perspectives on the association between social class and health. International Journal of Health Services, 26, 507–519. Kaufman, J. M. (2001). Characteristics of emotional and behavioral disorders of children and youth (7th ed.). Upper Saddle River, NJ: Merrill Prentice‐Hall. Kirby, R. S. (this volume). Incorporating geographical analysis into the study of mental retardation and developmental disabilities. In R. C. Urbano & R. M. Hodapp (Eds.), International review of research in mental retardation (Vol. 33, pp. 79–91). New York: Elsevier. Kretzmann, J. P., & McKnight, J. L. (1993). Building communities from the inside out: A path toward finding and mobilizing a community’s assets. Chicago: ACTA Publications. Kupersmidt, J. B., Griesler, P. C., DeRosier, M. E., Patterson, C. J., & Davis, P. W. (1995). Childhood aggression and peer relations in the context of family and neighborhood factors. Child Development, 66, 360–375. Lancaster, K. (1966). A new approach to consumer theory. Journal of Political Economy, 74, 132–157. Lawson, A., Biggeri, A., Bohning, D., LesaVre, E., Viel, J. F., & Bertollini, R. (Eds.) (1999). Disease mapping and risk assessment for public health. New York: John Wiley & Sons, Ltd. LeV, S. S., Costigan, T., & Power, T. J. (2004). Using participatory research to develop a playground‐based prevention program. Journal of School Psychology, 42, 3–21. Leventhal, T., & Brooks‐Gunn, J. (2000). The neighborhoods they live in: The eVects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin, 126, 309–337. Loeber, R., & Stouthamer‐Loeber, M. (1998). Development of juvenile aggression and violence. Some common misconceptions and controversies. American Psychologist, 53, 242–259. Lovett, A., Haynes, R., Sunnenberg, G., & Gale, S. (2002). Car travel time and accessibility by bus to general practitioner services: A study using patient registers and GIS. Social Science & Medicine, 55, 97–111. Martin, D., Wrigley, H., Barnett, S., & Roderick, P. (2002). Increasing the sophistication of access measurement in a rural healthcare study. Health and Place, 8, 3–13. Mason, C. A., Chapman, D. A., & Scott, K. G. (1999). Risk factors for severe emotional disabilities and emotional handicaps (SED/EH): An epidemiological perspective. American Journal of Community Psychology, 27, 357–381. McLaVerty, S. L. (2003). GIS and health care. In J. E. Fielding, R. C. Brownson, & N. M. Clark (Eds.), Annual review of public health (Vol. 24, pp. 25–42). Palo Alto, CA: Annual Reviews. Miranda, M. L., Dolinoy, D. C., & Overstreet, M. A. (2002). Mapping for prevention: GIS models for directing childhood lead poisoning prevention programs. Environmental Health Perspectives, 110, 947–953. Murphy, C. C., Boyle, C., Schendel, D., Decoufle, P., & Yeargin‐Allsopp, M. (1998). Epidemiology of mental retardation in children. Mental Retardation and Developmental Disabilities Research Reviews, 4, 6–13. Murphy, H. A., Hutchinson, J. M., & Bailey, J. S. (1983). Behavioral school psychology goes outdoors: The eVect of organized games on playground aggression. Journal of Applied Behavioral Analysis, 16, 29–35. Murray, C. (1996). The emerging British underclass. In R. Lister (Ed.), Charles Murray and the underclass: The developing debate (pp. 23–52). London: Institute for Economic AVairs.

210

Tina L. Stanton‐Chapman and Derek A. Chapman

Nabors, L., Willoughby, J., LeV, S. S., & McMenamin, S. (2001). Promoting inclusion for young children with special needs on playgrounds. Journal of Developmental and Physical Disabilities, 13, 179–190. Pearce, J., Witten, K., & Bartie, P. (2006). Neighbourhoods and health: A GIS approach to measuring community resource accessibility. Journal of Epidemiology and Community Health, 60(5), 389–395. Peeples, F., & Loeber, R. (1994). Do individual factors and neighborhood context explain ethnic diVerences in juvenile delinquency? Journal of Quantitative Criminology, 10, 141–157. Pellegrini, A. D., & Smith, P. K. (1998). Physical activity play: The nature and function of a neglected aspect of play. Child Development, 68, 577–598. Roderick, C., Pitchford, M., & Miller, A. (1997). Reducing aggressive playground behavior by means of a school‐wide raZe. Educational Psychology in Practice, 13, 57–63. Rubin, K. H., Fein, G., & Vandenberg, B. (1983). Play. In E. M. Hetheington (Ed.), Handbook of child psychology: Vol. IV. Socialization, personality, and social development (pp. 693–774). New York: Wiley. Rushton, G. (2003). Public health, GIS, and spatial analytic tools. In J. E. Fielding, R. C. Brownson, & N. M. Clark (Eds.), Annual review of public health (Vol. 24, pp. 43–56). Palo Alto, CA: Annual Reviews. Rushton, G., & Lolonis, P. (1996). Exploratory spatial analysis of birth defect rates in an urban population. Statistics in Medicine, 15, 717–726. Rutter, M., & Giller, H. (1983). Juvenile delinquency: Trends and perspectives. New York: Penguin Books. Sampson, R. J., & MorenoV, J. D. (1997). Ecological perspectives on the neighborhood context of urban poverty: Past and present. In J. Brookes‐Gunn, G. J. Duncan, & A. L. Aber (Eds.), Neighborhood poverty: Vol. II: Policy implications in studying neighborhoods (pp. 1–22). New York: Russell Sage Foundation. Sampson, R. J., & MorenoV, J. D. (2000). Public health and safety in context: Lessons from community‐level theory on social capital. In B. D. Smedley & S. L. Syme (Eds.), Promoting health: Intervention strategies from social and behavioral research. Washington, DC: National Academy Press. Schwartz, S. (1994). The fallacy of the ecological fallacy: The potential misuse of a concept and the consequences. American Journal of Public Health, 84, 819–824. Stanton‐Chapman, T. L., Chapman, D. A., & Scott, K. G. (2001). Identification of early risk factors for learning disabilities. Journal of Early Intervention, 24, 193–206. Stanton‐Chapman, T. L., Chapman, D. A., Bainbridge, N. L., & Scott, K. G. (2002). Identification of early risk factors for language impairment. Research in Developmental Disabilities, 23, 390–405. Surgeon General of the United States. (General of the United States 2001). Report of the Surgeon General’s conference on children’s mental health: A national action agenda. Available online at: www.surgeongeneral.gov/topics/cmh/defaut.htm. Susi, L., & Mascarenhas, A. (2002). Using a geographical information system to map the distribution of dentists in Ohio. Journal of the American Dental Association, 133, 636–642. Susser, M. (1994). The logic in ecological: I. The logic of analysis. American Journal of Public Health, 84, 825–829. Trivette, C., Dunst, C., & Deal, A. (1997). Resource‐based early intervention practices. In S. K. Thurman, J. R. Cornwell, & S. R. Gottwald (Eds.), The contexts of early intervention: Systems and settings (pp. 73–92). Baltimore: Paul H. Brookes. U.S. Census Bureau (2001). Introduction to Census 2000 data products. Washington, DC: U.S. Government Printing OYce. U.S. Census Bureau (2002). Census 2000 basics. Washington, DC: U.S. Government Printing OYce.

ROLE OF RECREATION

211

U.S. Congress (1997). Individuals with Disabilities Education Improvement Act of 1997. Public Law 105–17. Accessed from: http://www.ed.gov/offices/OSERS/Policy/IDEA/index.html U.S. Department of Education (2002). Twenty‐fourth annual report to congress on the implementation of the individuals with disabilities education act. Washington, DC: U.S. Government Printing OYce. Williams, B. (2004). Recreation and leisure strategies. In C. A. Baglin, M. E. B. Lewis, & B. Williams (Eds.), Recreation and leisure for persons with emotional problems and challenging behaviors (pp. 67–91). Baltimore: Sagamore Publishing. Wilson, W. J. (1987). The truly disadvantaged: The inner‐city, the underclass, and public policy. Chicago: University of Chicago Press. Wilson, W. (1996). When work disappears: The world of the new urban poor. New York: Knopf. Yang, D. H., George, R., & Mullner, R. (2006). Comparing GIS‐based methods of measuring spatial accessibility to health services. Journal of Medical Systems, 30, 23–32.

This page intentionally left blank

The Developmental Epidemiology of Mental Retardation and Developmental Disabilities DENNIS P. HOGAN POPULATION STUDIES AND TRAINING CENTER, BROWN UNIVERSITY PROVIDENCE, RHODE ISLAND

MICHAEL E. MSALL UNIVERSITY OF CHICAGO PRITZKER SCHOOL OF MEDICINE KENNEDY MENTAL RETARDATION CENTER AND INSTITUTE OF MOLECULAR PEDIATRICS, COMER CHILDREN’S AND LARABIDA CHILDREN’S HOSPITALS, CHICAGO, ILLINOIS

JULIA A. RIVERA DREW DEPARTMENT OF SOCIOLOGY, BROWN UNIVERSITY PROVIDENCE, RHODE ISLAND

I.

EPIDEMIOLOGICAL STUDIES

In this chapter, we provide an overview of developmental epidemiology as a powerful research tool with unique perspectives for investigators studying mental retardation and developmental disability. We survey the major dimensions characterizing epidemiological population studies. We review some of the major population‐based developmental epidemiology studies, and the insights they have provided in understanding mental retardation and developmental disabilities in community, school, and family contexts, and over the life course of children with mental retardation or developmental disabilities. We then provide a detailed description of the major conceptual and theoretical models of disability that have been advocated for use in population‐based studies, and their application to mental retardation and developmental disabilities. Finally, we illustrate the advantage of developmental epidemiology studies INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

213

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33009-1

214

Dennis P. Hogan et al.

of children by applying these perspectives to a nationally representative study of American children with mental retardation and developmental disabilities, also examining the relevance of functional limitations. A.

Four Dimensions of Developmental Epidemiology

A useful overview of epidemiological studies of mental retardation is found in Leonard and Wen (2002). Here we consider four principal dimensions that we use to characterize developmental epidemiology studies. The first framework involves the identification of the cases for study. For specific medical impairments that are relatively rare in the population, the units of study are children who present symptoms and are medically diagnosed. Typically, a case control population is defined, often from a population of children without the medical condition who are matched on race/ethnic and socioeconomic criteria. Examples include the Metropolitan Atlanta Congenital Defects Program (Correa‐Villasenor et al., 2003) and the record linkage study of the Western Australian Maternal and Child Health Research Institute (Leonard et al., 2005). Another strategy is to sample a population at high risk of developmental disabilities and follow them over time, distinguishing the experiences of children who develop a medical impairment from those who do not. Examples of this strategy for identifying subject and control cases are studies that identify very low birth weight (under 1250 g) infants (Msall & Tremont, 2000; Vohr et al., 2000) and the Assisted Technology Registries that in the United States use data from fertility clinics (Green, 2004). A distinctly diVerent type of study design uses a large sample of all children in a population to identify those children with mental retardation and developmental disabilities; the control group for analysis is those children in the population who did not manifest mental retardation or developmental disabilities. Examples include the birth registry cohort samples in the United Kingdom (children born in the first week of March 1945; see Hall et al., 2005) and Finland (children in two provinces of Finland born in 1966; see Taanila, Murray, Jokelainen, Isohanni, & Rantakallio, 2005). The second dimension of developmental epidemiology studies involves the use of birth cohorts in the study. Some studies are cross‐sectional, collecting information on mental retardation and developmental disabilities at one point in time (Bowe, 1995; Larson et al., 2001; Spencer, 2005). These studies sometimes attempt to discern underlying causes and risk factors by asking retrospective questions about the social and family situations, health histories, and life experiences of children with and without mental retardation and developmental disabilities. The contrasting study uses a cohort longitudinal design in which a sample of children are identified at an initial point of observation (often at birth) and are periodically followed up for the onset of disability and changing limitations in tasks and participation (Hall et al., 2005; Msall & Tremont, 2000; Taanila et al., 2005; Vohr et al., 2000).

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

215

A third dimension of developmental epidemiology studies involves the number of cases available for comparison. In studies based on cases identified as developmentally disabled or mentally retarded, case sizes often number 100– 300 children, with a control sample of equivalent size or larger. Studies of high‐ risk children often begin with a sample of 750–1000 children, of whom several hundred may be identified as developing mental retardation or developmental disabilities. Even very large surveys of population cohorts can produce remarkably few cases of particular medical impairments. For example, the U.K. study used a birth cohort sample of 5362 children born to married mothers; of these 111 children had mild intellectual impairment and 23 children had severe intellectual impairment by the age of 15/16 years (Hall et al., 2005). The Finland birth cohort study included 12,058 children of whom 129 individuals with intellectual disability. The power of the population sample‐based studies for the estimation of prevalence rates of particular conditions is severely limited by the challenge of producing adequate samples of children to produce estimates with small standard errors. These problems were resolved in the West Sussex Computerized Health System which pooled data for 105,760 children born between January 1982 and December 1997 to identify 293 children with cerebral palsy (Sundrum, Logan, Wallace, & Spencer, 2005). A fourth dimension of developmental epidemiology studies involves the method of data collection. Some studies use intensive clinical measurements requiring the presence of subject children at a medical facility (e.g., birth magnetic resonance imaging; see Toft et al., 1994). Other studies collect information in laboratory or school settings (such as those that assess IQ and cognitive functioning; see Taanila et al., 2005; Vohr et al., 2000). Typically, information of this sort is collected directly by observation or from instruments administered directly to the children. The measurement of cognitive and social skills in natural settings (school or home) typically uses instruments (see discussion in Section II) which are administered to the parents who report on their children’s behavior (Msall & Tremont, 2000). Information on medical outcomes, access to health services, and social participation of children are typically gathered from interviews with parents (Hogan & Msall, in press). Data collected in physician oYces are typically seen as the ‘‘gold standard’’ against which to measure population surveys of parents, but this is in part illusory. The eVective identification and coding of birth defects in surveillance systems is complex and subject to a variety of flaws (Rasmussen & Moore, 2001). Many of the instruments used for assessing practical intelligence and social skills in medical oYces are in fact administered to parents either through completion of a paper and pencil form or by health specialists in their collection of medical histories and situations of children. Sample surveys with structured interviews typically are least accurate in the specification of particular medical conditions. These sources of error are diminished by asking parents about medical impairments that have been identified

216

Dennis P. Hogan et al.

by a doctor, other health professional, or school oYcial (Hogan & Msall, in press). B.

Quality of Parent Reports

The Texas Birth Defects Registry identifies children with major abnormal structural anomalies using trained interviewers who abstract information in regular visits to medical facilities where they review birth registrations, logbooks, and hospital discharge lists (Ramadhani, Canfield, Waller, & Case, 2004). Birth mothers were interviewed using standardized 1 hour questionnaires an average of 294 days after the birth, and this information was matched to information from the Birth Defects Registry. Analysis demonstrated that the quality of maternal reports of demographics is highly reliable and that there is considerable consistency (94% and better agreement) in the measurement of gestational and nongestational diabetes. The kappa coeYcient was calculated to correct for chance concordance (a major issue in the measurement of concordance when a condition is rare). The kappa values were high for gestational and nongestational diabetes, nearly perfect for insulin use, but only moderate for seizures/epilepsy (Ramadhani et al., 2004). Survey methods that interview mothers or caregivers of children typically provide useful and reliable information for analysis, although the data are somewhat weaker for less well‐defined medical impairments. II.

MODELS FOR UNDERSTANDING MENTAL RETARDATION AND DEVELOPMENTAL DISABILITIES

Population epidemiological analyses of mental retardation and developmental disability require a conceptual framework that accounts for the various medical conditions associated with developmental disabilities, medical care, and impact on daily lives of children. A variety of frameworks have been used to describe the complex web of children’s health and well‐being. The first framework, the ‘‘medical impairment model,’’ focuses on medical diagnosis of impairments (pathophysiological processes aVecting organ system performance). This clinical/medical tradition aims for accurate diagnosis, critical analysis of laboratory indicators, and use of optimal management strategies informed by intense medical cohort studies. For example, if a child has type 1 diabetes, frequent glucose monitoring, intensive insulin replacement to avoid hyperglycemia, titration of insulin to appropriate carbohydrate nutritional intake, and regular exercise are associated with decreased long‐term renal, ophthalmic, and vascular complications (The Diabetes Control and Complications Trial Research Group, 1993). This framework

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

217

is most useful in addressing impairments interfering with a child’s daily health functioning such as breathing, gaining weight, using basic senses, and being neurologically responsive. For example, without insulin replacement the child with type 1 diabetes loses weight, is lethargic, and can become comatose. The second framework, the ‘‘developmental disability model,’’ focuses on discrepancies between an individual child’s performance and that of his/her peers using appropriate psychometric tools. This tradition quantifies delays in development or intensity of clusters of behavioral states and establishes criteria for: (1) developmental motor, cognitive, social–emotional, or adaptive disorders; (2) communicative impairments; (3) coordination and perceptual impairments; (4) autistic spectrum disorders; (5) specific learning disabilities; and (6) attention deficit hyperactivity disorders. The strength of the developmental disability model is the reliance upon comprehensive assessment of developmental and behavioral processes often involving several sessions of standardized interviewing and structured observation. In school age children, a variety of cognitive assessments (Weschler Intelligence Scale III, Kaufman Assessment Battery for Children 2, Stanford Binet Intelligence Scale V, DiVerential Ability Scales, Leiter International Performance Scale—Revised), individualized academic achievement tests (Wide Range Achievement Test‐3, Woodcock Johnson III Tests of Achievement, Peabody Individual Achievement Test— Revised, Weschler Individual Achievement Test, Kaufman Test of Educational Achievement), and parent and teacher ratings of behavior are available. These have been reviewed by Sattler and Aylward (2001) and Aylward (1993). Using a combination of psychological, achievement, and behavioral assessments, a child’s school performance diYculties can be understood in terms of learning diVerences, learning disorders, learning disabilities, cognitive impairments, attention deficits, mood disorders, or adjustment disorders. Despite their relative strengths, both the medical impairment model and the developmental disability model focus on a child’s deficits and do not adequately account for a child’s skill in performing daily living activities in natural environments at home and in the community. For example, stating that a child of age 10 years has the medical impairments of Down syndrome, short stature, hypothyroidism, myopia, and middle ear tubes for recurrent otitis media and conductive hearing loss of 30 dBs, and scores of 65 or lower on the Clinical Evaluation of Language Fundamentals 4th edition (CELF‐4 0; Semel, Wiig, & Secord, 2003) do not acknowledge that the child may be able to communicate basic needs in short sentences, have a reading vocabulary of 100 words, be able to carry on a phone conversation, and be independent in basic activities of daily living (feeding, dressing, toileting, and basic mobility). In a similar fashion, many children with Down syndrome despite having academic challenges in middle elementary school are able to learn

218

Dennis P. Hogan et al.

some basic reading decoding, even though they may struggle with reading complex information quickly and writing a book report about what they read. A variety of adaptive‐functional scales have been developed to measure children’s skills in performing daily activities (Table I). The third framework, the ‘‘biopsychosocial model,’’ combines biological, psychological, and social perspectives on a child’s health and well‐being (Stein & Silver, 1999). This model takes into account the child’s physical, behavioral, and developmental status as well as increased use of the following services compared to his/her peers: medical services (glasses, hearing aids, inhalation medications for asthma, anticonvulsants, nutrition supports), rehabilitative and compensatory services (physical therapy, occupational therapy, speech‐language therapy, alternative mobility supports, augmentative communication, robotic assistants), educational supports (Early Intervention and Special Education services), and behavior supports (counseling, stimulant medication). This model also allows for descriptions of the child’s developmental strengths as well as challenges with daily activities. In addition, it can be applied to a heterogeneous population of children with complex medical, developmental, or behavioral impairments. The weakness of this model is that individuals with myopia, seasonal rhinitis, and eczema can be described as having multiple impairments because of repeated use of medications and health services despite having readily controlled symptoms. In 1992, The American Association on Mental Retardation Ad Hoc Committee on Terminology and Classification attempted to address the limitations in both the medical impairment and developmental disability models for individuals with intellectual and adaptive disabilities (Luckasson et al., 1992). The theoretical model of mental retardation they adopted combines elements of the medical impairment and developmental disability model, and also takes into account the medical interventions of the biopsychosocial model (Fig. 1). Mental retardation was defined as significantly subaverage intellectual functioning occurring before the age of 18 years with concurrent limitations in at least 2 of 10 adaptive skill domains. It was specifically noted that adaptive and functional limitations often coexist with adaptive and functional strengths and other personal capabilities. Emphasis also included that cognitive assessments would be appropriate for the linguistic and cultural diversity of the individual. In addition, it was explicitly stated that with appropriate supports over a sustained period the life functioning of persons with mental retardation will generally improve. In 2002 further revision was undertaken (Luckasson et al., 2002). Mental retardation was defined as a disability characterized by significant limitations in intellectual functioning with concurrent strengths and limitation in conceptual, social, and practical adaptive skills and occurring before 18 years of age.

TABLE I ADAPTIVE‐FUNCTIONAL SCALES FOR CHILDHOOD DISABILITIES

Domains

Concurrent validity

IDEA‐FSTM

VABS‐2

SIB‐R

Warner Initial Developmental Evaluation of Adaptive and Functional SkillsTM Self‐care (feeding, dressing, diaper awareness) Mobility Social cognition (verbal, nonverbal play) CLAMS CAT

Vineland Adaptive Behavior Scales

Scales of Independent Behavior‐Revised Early Developmental Form Adaptive skills Problem behaviors

Disability samples

Children in early intervention, or with special health care needs

Time to administer

15 minutes

Communication Daily living Socialization Motor

PEDI

WeeFIMTM

Pediatric Evaluation of Disability Inventory

Functional Independence Measure for Children

Motor Self‐care Social function Caregiver assistance (CA) Environmental modifications (EM) Battelle WeeFIM Injury severity

Self‐care Mobility Social cognition

Vineland Social Maturity Scale 2 Kaufman ABC Peabody Picture Vocabulary Test Children with autistic spectrum disorders, mental retardation, or developmental disabilities

Chronological age Early screening profiles

Children with mental retardation or developmental disabilities

Children with brain injury, cerebral palsy, spinal injury, or juvenile arthritis

30 minutes

5 minutes

45 minutes; 15 minutes for CA/EM Scales

Battelle VABS PEDI Amount of assistance Questionnaire Children with cerebral palsy, prematurity, congenital heart, or sensory, genetic, developmental, learning, or attentional disorders 20 minutes

220

Dennis P. Hogan et al.

FIG. 1. Theoretical model of mental retardation.

With appropriate personalized supports over a sustained period of time, the life functions of persons with mental retardation would be optimized. Five explicit dimensions were included in this definition: intelligence, adaptive behavior, participation‐interactions‐social roles, physical and mental health and etiology, and ecological context. The fourth framework, the ‘‘International Classification of Functioning (ICF) model,’’ describes a child’s health and well‐being in terms of four components: (1) body structures, (2) body functions, (3) activities, and

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

221

(4) participation. Body structures are anatomical parts of the body, such as organs and limbs, as well as structures of the nervous, sensory, and musculoskeletal systems [World Health Organization (WHO), 2001]. Body functions are the physiological functions of body systems, including psychological functions such as attending, remembering, and thinking. Activities are tasks, including learning, communicating, walking, carrying, feeding, dressing, toileting, bathing, reading, preparing meals, shopping, washing clothes. Participation means involvement in community life, such as relationships, education, work, and recreational, religious, civic, and social activities. Table II illustrates the kind of mobility, manipulative, and cognitive early functional descriptors that apply to the measurement of activity and participation limitations. The ICF model also accounts for contextual factors in a child’s life, including environmental and personal factors. Environmental factors, such as policy, social, and physical facilitators, and barriers include positive and negative attitudes of others, legal protections, and discriminatory practices. Personal factors include age, gender, interests, and sense of self‐eYcacy. The strength of the ICF model is that it describes both functioning and enablement. Its weakness is that it has not been widely used with children and does not have explicit indicators for all the domains of the model. However, the model does oVer the promise of a much broader perspective with respect to children’s activities and participation (Simeonsson, Lollar, Hollowell, & Adams, 2000). Because of its emphasis on both adaptive functioning and supports, the AAMR Model of Mental Retardation is easily adapted to the ICF Framework. To illustrate the potential of this model, a variety of scenarios are described in Table III. Figure 2A illustrates how to apply the ICF model to a child with fetal alcohol syndrome who is struggling socially, behaviorally, and vocationally and is in danger of dropping out from school. Figure 2B illustrates how to apply the ICF model and the AAMR 10th edition model to a 10‐year‐old girl with Down syndrome. More recently, the Institute of Medicine proposed a ‘‘Developmental Kaleidoscope Model of Children’s Health’’ that expands on the ICF model (and its AAMR adaptation to children with mental retardation) to incorporate policies and services. This is illustrated in Fig. 3 for a 16‐year‐old boy with hemiphlegic cerebral palsy with a complex partial seizure disorder (National Research Council and Institute of Medicine, 2004). It is this multidimensional functional, participatory, and contextual emphasis of the ICF, AAMR, and Developmental Kaleidoscope models that best capture the complex variations in mental retardation and developmental disabilities in population epidemiological studies. It is the framework that motivates this study, which focuses on the diagnosis of medical conditions, the measurement of functional disabilities, and the health services needed and received.

222

Dennis P. Hogan et al. TABLE II KEY MOBILITY, MANIPULATIVE, AND COGNITIVE EARLY CHILDHOOD FUNCTIONAL DESCRIPTORS

Moving about and manipulating objects Does the child: (1) walk without help; (2) stack two blocks; (3) mark on paper with a crayon or pencil; (4) walk upstairs putting both feet on each step; (5) walk downstairs putting both feet on each step; (6) kick a ball; (7) run smoothly and change direction; (8) open a door by turning and pulling the doorknob or handle; (9) open a jar by unscrewing the lid; (10) jump over a small object; (11) throw a ball several feet; (12) unbutton the buttons on clothing; (13) catch a ball thrown from several feet away? Caring for self Does the child: (1) drink from an open cup or glass; (2) use a straw for drinking; (3) understand that hot things are dangerous; (4) use a spoon to feed him/herself; (5) take oV clothes; (6) ask to use the toilet/bathroom; (7) urinate (pee) in toilet or potty chair; (8) have bowel movements (poop) in toilet or potty chair; (9) put on underpants, shorts, pants with elastic waistband; (10) wash his/her hands? Acquiring and using information Does the child: (1) follow simple directions (come here, sit down); (2) say 1–10 words; (3) use gestures such as waving ‘‘hi’’ and ‘‘bye’’ or shaking head for ‘‘no’’; (4) respond to the request to ‘‘give me’’ an object; (5) point to a few pictures that someone names; (6) name a few pictures; (7) point to or name a few body parts on self or doll; (8) put 2 words together; (9) use own name to talk about him/herself; (10) use action words ending in ‘‘ing’’ (sleeping, eating); (11) answer ‘‘yes’’ and ‘‘no’’ correctly when asked a question; (12) understand and use words meaning more than 1 (plurals such as cats or toys)? Attending to and completing tasks Does the child: (1) look at someone who speaks to him/her; (2) listen and pay attention to a simple story being read; (3) play with toys for a few minutes; (4) change activities without getting upset? Interacting and relating to others Does the child: (1) give objects or point to objects to show them to others; (2) use names for at least two people; (3) smile or laugh when others say something funny or nice such as ‘‘good work,’’ ‘‘that’s a pretty hat’’; (4) try to please other people; (5) play a simple game with another person (hiding, chasing); (6) play ‘‘make believe’’ (feed dolls, put dolls to bed, pretend to go shopping); (7) imitate an activity you do?

In the case of the United States, studies of children with mental retardation and developmental disabilities have been hampered by inadequate data when compared to other nations. In doing this review, we decided it was essential to implement these ideas in an American study of children with mental retardation and developmental disability. In this way, we hope to illustrate how valuable the approach can be in the study of mental retardation and disability among American children, even when restricted to the limited data systems now available.

TABLE III ICF MODEL SCENARIOS Definition

Girl, 5 years

Pathophysiology

Molecular/ biochemical mechanisms interfering with function

475 g birth weight, 26‐week gestation, developmental lung injury

Body structures and body functions

Loss of organ structure or organ function

Activity (functional) strengths

Ability to perform essential activities: feed, dress, toilet, walk, talk

223

Dimension

Boy, 8 years

Girl, 12 years

Boy, 16 years

Fetal alcohol syndrome

Myotonic dystrophy with type 1 diabetes

Growth delays, chronic lung disease, asthma, microcephaly, mild mental retardation

Congenital heart disease requiring surgical repair, middle ear tubes, myopia

Needs insulin replacement to maintain glucose homeostasis, moderate mental retardation, facial muscle weakness and hypotonia

Runs, jumps rope, climbs monkey bars, toilet trained, performs basic dressing, loves to do art projects

Daily walks of two miles, loves Dr. Seuss books, plays piano

Loves swimming, sticker chart helps with food choices and glucose monitoring, can read first grade words

Hemiplegic cerebral palsy secondary to intrauterine stroke with death of cotwin Complex partial seizures, 50 dB sensorineural hearing loss, decreased muscle bulk and motor control on right side. Verbal IQ 75, nonverbal IQ 60 Reads at fourth grade level, knows time and money, loves to bowl, takes medications without supervision (continued)

TABLE III (Continued ) Dimension

Girl, 5 years

Boy, 8 years

Girl, 12 years

Boy, 16 years

Activity (functional) limitations

DiYculty in performing essential activities

Unable to snap or zipper; speech mispronunciations, struggling with remembering letters and numbers

Unable to ride two wheel bike, speech not understood by strangers

Speech diYculties in noisy environments

Participation

Involvement in community roles typical of peers DiYculty in assuming roles typical of peers

Plays indoors and outdoors with peers

Attends Special Olympics, is earning badges in Boy Scouts Not allowed on overnights because wets the bed, Special Olympic year round sports not available without parent driving 2 hours

DiYculty with nonverbal communication because of facial muscle weakness, does not know common words on signs, unable to understand correct change Swims three times per week and horseback rides on weekends Fearful of being on an adaptive swimming team because of temptation to dietary indiscretion from candy machines

224

Definition

Participation restriction

Misses school due to asthma flares, receives speech and occupational therapies

On adaptive bowling team

Family fearful to teach use of public transportation because of worry about seizures

Contextual factors: environmental facilitators

Attitudinal, legal, policy, and architectural facilitators

Contextual factors: environmental barriers

Attitudinal, legal, policy, and architectural barriers

Good hand‐washing policies at school, asthma care plan, use of stories and talking books for emerging literacy Pediatric school nurse to maximize attendance, home activities with art and board games

Use of music to enhance speech intelligibility, cuts in school budget limit computers

Adaptive swim coach willing to have all swimmers on health snack choices

Strategies to promote mature public behaviors, limited activities with peers after school

Mentoring in diabetic management by older teen with Down syndrome and type 1 diabetes mellitus

225

Peer support group for teens with epilepsy teaches importance of medical bracelet Fragmentation in vocational and young adult transition services because of combination of motor, hearing, communicative, and cognitive challenges

226

Dennis P. Hogan et al.

A

Body function and structure Lower extremity paraplegia Neurogenic bowel and bladder Hydrocephalus requiring shunting Bilateral club feet

Activities Wheel chair mobility Able manual skills Active learner/communicator Self-care independent with bladder catherization

Environmental factors Barriers to access Accommodation Physical support Emotional support

Participation Family No friends Academic difficulty Clubs Work

Personal factors Plays the drums Likes photography Red sox fan

B Body function and structure • Short stature • Hypothyroidism • Myopia • Frequent sinusitis and otitis

• • • •

Activities • Performs own self-care • Communicates basic needs in short sentences • Does best in small groups with structure

Environmental factors Community transportation Facilitator at Scouts Buddy system/mentoring Emotional support

• • • •

Participation • Attends church with family • Good friend at Scouts • Curricular modifications • Recreational accomodations

Personal factors Loves music Eager to learn sports skills Cub fan Limited access to TV

FIG. 2. (A) ICF and AAMR model for a 10‐year‐old child with spina bifida. (B) ICF and AAMR model for 10‐year‐old child with Trisomy 21.

III.

DATA AND MEASURES

To provide an epidemiology of children with mental retardation and developmental disabilities, we created a pooled dataset using cases from the focal child samples across the annual 1997–2003 National Health Interview Surveys (NHIS). The NHIS is the primary source of data about the health of the American people. The NHIS is a population‐based survey administered annually by the U.S. Census Bureau through personal household interviews

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

227

FIG. 3. The kaleidoscope model of disability. We illustrate this framework with the case of a 16‐year‐old boy with hemiplegic cerebral palsy with a complex partial seizure disorder. He was born at term and weighed 3.5 kg. At delivery there was evidence of demise of a cotwin.

Biology: vulnerability of developing central nervous system to vascular events. Developmental processes and social environment: challenges in complex academics, speech intelligibility, and adaptive skills requiring bimanual coordination. Benefits from structure and family supports. Physical environment and social capital: father is a computer scientist at a community college. Mother is a nurse. College students help with after school transportation and after school activities. Behavior: anxiety about impact of a seizure when away from home.

Child health: pediatrician frustrated in accessing appropriate young adult providers when teen turns 18. Despite well‐controlled seizures, concern about self‐management as young adult. Policy: gaps in vocational training and opportunities for higher education and community living. Waiting list for community housing is 10 years. Services: quality vocational and supportive employment programs are a scarce resource for teens with functional literacy but who have challenges in communication, adaptive skills, and health maintenance. Potential: young adult will learn both community living, job holding, and leisure management with a combined strategy of structured tasks, mentoring, and job coaching.

with the adult most knowledgeable about the health of family members. The sampling frame was stratified at the state‐level and included over‐samples for Blacks and Hispanics. The basic survey collects household

228

Dennis P. Hogan et al.

and family social and economic information and collects information about members of the household. In addition, the NHIS randomly sampled one adult and one child (referred to as the ‘‘focal child’’) from each sample household. The focal child interview provides detailed information about the child’s medical conditions, disability statuses, health care access and use, and health outcomes. The NHIS thus provides a unique opportunity to concurrently examine the health and family contexts in which children with developmental disabilities are living, health outcomes, and access to needed medical services. The overall response rate for the pooled dataset for all eligible children in the focal child sample was 81.0%, yielding a total sample of 92,573 children (National Center for Health Statistics, 2000, 2002a,b, 2003a,b, 2004). For our purposes, we selected only focal children who were between the ages of 5 and 17 at the time of the survey. [Population survey measures of medical conditions and disabilities work less well for children under age of 5 years (Hogan & Msall, in press).] After selecting for age, our analytical sample size was 65,497 children aged 5– 17 years. A.

Measure of Mental Retardation/Developmental Disabilities

Our measure of mental retardation and developmental disabilities was based on five items. Adult respondents were asked ‘‘Has a doctor or health professional ever told you that [SELECTED CHILD]’s name had: Mental Retardation?’’ Children positively identified by this question were coded as having mental retardation. Respondents were also shown a flashcard with a list of conditions and asked ‘‘Looking at this list, has a doctor or other health professional ever told you that [SELECTED CHILD] had any of these conditions?’’ If respondents reported that the selected child had Down syndrome, cerebral palsy, muscular dystrophy, or autism, we considered them to have a developmental disability. Children identified as having either mental retardation or a developmental disability were coded as yes on the overall mental retardation/developmental disability measure (N ¼ 924). B.

Measure of Functional Limitation

Our measure of functional limitation was based on 11 items covering four areas: (1) self‐care limitations; (2) mobility limitations; (3) sensory limitations; and (4) learning limitations. For self‐care limitations, adult respondents were asked if the focal child ‘‘need[ed] the help of other persons

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

229

with . . .’’ (1) bathing or showering; (2) dressing; (3) eating; (4) getting in or out of bed or chairs; (5) using the toilet, including getting to the toilet; and (6) getting around inside the home. For mobility limitations, respondents were asked, ‘‘Does [SELECTED CHILD] have an impairment or health problem that limits [HIS/HER] ability to (crawl), walk, run, or play?’’ For sensory limitations, respondents were asked, ‘‘Which statement best describes [SELECTED CHILD]’s hearing without a hearing aid: Good, a little trouble, a lot of trouble, or deaf?’’ Children who had a lot of trouble or were deaf were considered to have a sensory limitation. Respondents were also asked ‘‘Does [SELECTED CHILD] have any trouble seeing even when wearing glasses or contact lenses?’’ Children who were reported as yes on this item were also considered to have a sensory limitation. Finally, for learning limitations, respondents were asked ‘‘Has a representative from a school or a health professional EVER told you that [SELECTED CHILD] had a learning disability?’’ and whether the focal child is ‘‘LIMITED IN ANY WAY because of diYculty remembering or because [THEY] experience periods of confusion?’’ Children with an aYrmative response to either item were considered to have a learning limitation. Children with a limitation in any of these four limitations were considered to have a functional limitation for the purposes of this analysis (N ¼ 8276). IV.

RESULTS

A variety of studies have attempted to identify the prevalence of disabilities, developmental disabilities, and mental retardation. Merrick and Carmeli (2004) provide a summary of the overall rates of medical impairment disability in several populations. They report overall medical impairment rates (per 1000) of 112 in the United Kingdom, 119 from the Rochester Health Survey in the United States, 79 in 5 Scandinavian nations, and 77 in Israel. Other studies have measured limitation in daily activities, and linked these to medical conditions, producing estimates of functional limitations, finding rates of 58–120 per 1000 (Wells & Hogan, 2003). Yeargin‐Allsopp and Boyle (2002) estimate that approximately 10 per 1000 school‐age children are mentally retarded; between 20 and 24 per 1000 newly born children have cerebral palsy, and a reported range of 2–6 per 1000 for autism. However, many estimates suggest a prevalence rate of mental retardation of 3–4 per 1000 (data summarized in Leonard & Wen, 2002). Durkin (2002) uses other international studies of serious mental retardation that show a much broader possible range of prevalence, with most nations are estimated to have rates of 20–60 per 1000. This illustrates a situation in which the relatively inferior statistical system in the United States seems to produce

230

Dennis P. Hogan et al.

far less accurate information about the prevalence of developmental disabilities than can be garnered from the superior birth cohort longitudinal studies that are common in many other highly developed nations. However, the methodologies used in defining disabilities and mental retardation in diVerent national studies vary remarkably. New population estimates of the point prevalence rates of developmental disabilities, mental retardation, and functional limitation are needed to provide a more accurate portrait for American children.

A.

Population Estimates of Children with Mental Retardation and Developmental Disabilities

We are able to provide greatly improved estimates with the children recorded in the 1997–2003 National Health Interview Surveys. Table IV provides this improved information about the prevalence rates of developmental disabilities, mental retardation, and functional limitations in the population of American children aged 5–17, and the number of children impacted in 2003. The overall prevalence of mental retardation and developmental disability is estimated at 13.9 per 1000, or approximately 743,600 children aged 5–17 years. Rates of mental retardation are about 7.5 per 1000, or approximately 398,400 American children. Evidence indicates that about

TABLE IV POPULATION PREVALENCE ESTIMATES OF MENTAL RETARDATION, DEVELOPMENTAL DISABILITIES, AND FUNCTIONAL LIMITATIONS AMONG U.S. CHILDREN AGED 5–17 YEARS Condition Mental retardation/ developmental disabilities (All) Functional limitations Mental retardation Cerebral palsy Muscular dystrophy Down syndrome Autism a

Prevalence per 1000 (95% CI)

N (thousands)a

13.9 (12.9–14.9)

743.6

126.6 7.5 3.8 0.6 1.4 3.0

(124.1–129.1) (6.9–8.1) (3.4–4.2) (0.4–0.8) (1.2–1.6) (2.6–3.4)

6724.5 398.4 201.8 31.9 74.4 159.3

2003 population estimates. All numbers shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997– 2003).

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

231

1.4 children per 1000, or 74,400 school‐aged children, have Down syndrome. Estimates of cerebral palsy range from 3.4 to 4.2 per 1000. The prevalence of autism is nearly as high as that of cerebral palsy, with about 159,300 American children aVected in 2003. These point‐prevalence rates are generally within the ranges estimated for the United States based on the limited number of local American community studies and data from other nations. But they are far more precise, and provide well‐bounded estimates for the prevalence of mental retardation and developmental disabilities among American children. Limitations in activities of everyday life (functional limitations) are nearly 10 times more common than mental retardation and developmental disabilities, with an estimated prevalence rate of 127 per 1000, including some 6,724,500 children. We believe that these are the best population estimates of disability ever produced for the United States. B.

Differentials in Prevalence

Table V shows the prevalence of developmental disabilities, mental retardation, and functional limitation for younger and older children, by sex and race/ethnicity. There are statistically significant diVerences in prevalence rates by these demographic characteristics. The prevalence of mental retardation and developmental disabilities is slightly higher among older children and show dramatic diVerences by sex (16 per 1000 among boys versus 12 per 1000 among girls). There are statistically significant diVerences in the prevalence of developmental disabilities by race and ethnicity, with the primary diVerence being the much higher rates among black children (18 per 1000). Prevalence rates of disability across the family characteristics are not provided since the occurrence of disability among children increases the likelihood of poverty and may increase separation and divorce. There are considerably more children with functional limitations (N ¼ 8276) in the population, with significant diVerences in the prevalence rates across all demographic groups. Older children, males, and children who are African American have significantly higher prevalence rates of functional limitation. Table V also provides information in the prevalence of functional limitations among children with developmental disabilities. While the apparent diVerences are large, none of the diVerences are statistically significant. C.

Family Origins

The socioeconomic status characteristics of children with and without developmental disabilities and functional limitation are shown in Table VI. (These percentages diVer from Table V in that they describe the family situations of children with and without disabilities rather than prevalence

232

Dennis P. Hogan et al.

TABLE V DEMOGRAPHIC CHARACTERISTICS OF CHILDREN WITH MENTAL RETARDATION/DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONS

All kids (N ¼ 65,497) Demographic characteristics Age 5–11 years 12–17 years Sex Male Female Race/Ethnicityd Non‐Hispanic, White Hispanic Non‐Hispanic, Black Non‐Hispanic, Other

(%)

Mental retardation/ developmental disabilitiesa (NDD ¼ 924) Yes (%)

No (%)

Functional limitationsb (NFL ¼ 8276)

Functional limitations (children with MR/DD)c (NDD&FL ¼ 729)

Yes (%)

No (%)

Yes (%)

No (%)

54.5 45.6

1.3 1.5

98.7 98.6

10.8 14.9

89.2 85.1

77.5 76.7

22.5 23.3

51.2 49.0

1.6 1.2

98.4 98.9

14.8 10.4

85.2 89.6

78.6 74.9

21.4 25.1

64.2

1.3

98.7

12.9

87.1

76.3

23.7

15.5 14.9

1.3 1.8

98.7 98.2

11.0 14.2

89.0 85.8

78.9 79.8

21.1 20.2

5.3

1.4

98.6

9.8

90.2

72.4

27.6

a DiVerences by age and race in the incidence of MR/DD are statistically significant at the p < .05 level. b All diVerences between in the distribution of FL by demographic characteristics are statistically significant at the p < .05 level. c None of the diVerences in the distribution of DD and FL by demographic characteristics are statistically significant. d 15 children are missing information on race/ethnicity. All percentages shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003).

rates across family groups.) Children with mental retardation and developmental disabilities are more often from homes in which the best educated parent has less than a high school education, are 8% more likely to be in poverty, and are 10% less likely to live in a two‐parent home. Similar diVerentials are observed in the socioeconomic and family situations of children with and without functional limitations. There are no statistically significant diVerences in the family situations of children who are developmentally disabled or mentally retarded by their functional status.

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

233

TABLE VI SOCIOECONOMIC STATUS OF CHILDREN WITH MENTAL RETARDATION/DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONS

All kids (N ¼ 65,497) Socioeconomic status

(%)

Mental retardation/ developmental disabilitiesa (NDD ¼ 924) Yes (%)

No (%)

Functional limitationsa (NFL ¼ 8276) Yes (%)

No (%)

Functional limitations (children with MR/DD)b (NDD&FL ¼ 729) Yes (%)

No (%)

Education Less than HS HS graduate Some college College graduate

13.5 24.6 32.1 29.9

18.4 25.6 30.5 25.5

13.4 24.6 32.1 30.0

16.8 27.5 34.4 21.3

13.0 24.1 31.7 31.2

19.1 26.7 30.6 23.5

16.0 21.7 30.2 32.1

Income to poverty ratio Below poverty Adequate Secure

16.2 40.5 43.2

24.2 44.8 31.0

16.1 40.5 43.4

22.9 43.2 34.0

15.2 40.1 44.7

25.2 45.2 29.7

21.0 43.7 35.3

Family structure (in home) Neither parent 3.1 One parent 26.0 Both parents 70.9

4.7 34.4 60.9

3.1 25.9 71.0

4.0 34.1 61.9

2.9 24.9 72.2

5.3 36.1 58.6

2.7 28.5 68.8

a

All diVerences between children with MR/DD and other children, and between children with FL and other children in the distribution of SES are statistically significant at the p < .05 level. b None of the diVerences between children with DD and FL and other children with DD are statistically significant at the p < .05 level. All percentages shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003).

D.

Health Outcomes

Are health indicators more problematic for children with mental retardation or developmental disabilities and for children with functional limitations? This question is addressed in Table VII, which indicates that there are statistically significant diVerences in the health indicators of these children compared to other children. Children with developmental disabilities are nearly 3 times more likely to miss 11 or more school days, about 7 times more likely to spend 6 or more nights in a hospital, and 10 times more likely to be of fair or poor health status. Similar diVerences of more modest magnitude are observed when comparing children with and without functional limitations. The health situation of children with disabilities who also

234

Dennis P. Hogan et al. TABLE VII HEALTH INDICATORS FOR CHILDREN WITH DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONSa

All kids (N ¼ 65,497) Health indicators

(%)

Mental retardation/ developmental disabilities (NDD ¼ 924) Yes (%)

No (%)

Functional limitations (NFL ¼ 8276)

Functional limitations (children with MR/DD) (NDD&FL ¼ 729)

Yes (%)

No (%)

Yes (%)

No (%)

School days missed 0–5 days 82.6 6–10 days 11.5 11þ days 5.8

67.1 16.1 16.8

82.9 11.5 5.7

69.2 16.4 14.4

84.6 10.8 4.6

61.9 18.2 19.9

83.7 9.4 6.9

Hospital stays None 1–5 nights 6þ nights

97.9 1.6 0.5

92.1 4.1 3.8

98.0 1.6 0.5

94.9 3.2 1.9

98.3 1.4 0.3

90.0 5.1 4.9

98.9 0.9 0.1

Visits to ER/ED None 1 visit 2þ visits

81.8 12.3 5.9

75.4 13.4 11.2

81.9 12.3 5.9

73.0 16.3 10.7

83.0 11.7 5.2

73.2 13.7 13.1

82.7 12.4 4.9

Health status Excellent Very good Good Fair Poor

54.2 28.3 15.6 1.7 0.3

26.3 26.2 28.8 13.0 5.7

54.6 28.3 15.4 1.5 0.2

37.6 27.9 26.6 6.1 1.8

56.6 28.4 14.0 1.0 0.1

19.3 24.7 33.6 14.9 7.4

49.8 31.1 12.7 6.5 0.0

All diVerences are statistically significant at the p < .05 level. All percentages shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003). a

experience functional limitations is especially dismal—20% miss 11 or more school days, 10% have 1 or more nights in the hospital, 27% have visited the emergency room in the past year, and 22% are of fair or poor health status. Developmental disabilities linked with functional limitations greatly worsen health outcomes. E.

Health Care

Children with developmental disabilities, especially those who are functionally limited, are more likely to have a usual source of health care and are more likely to have health insurance (Table VIII). Even so, children with

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

235

TABLE VIII ACCESS TO HEALTH CARE FOR CHILDREN WITH MENTAL RETARDATION/DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONS

All kids (N ¼ 65,497) Health care access

(%)

No usual source Delayed care (cost) Unmet need (cost)

5.7 7.4 5.4

Health insuranceb Uninsured Privately insured Publicly insured

15.8 75.2 9.0

Mental retardation/ developmental disabilitiesa (NDD ¼ 924) Yes (%)

Functional limitationsa (NFL ¼ 8276)

Functional limitations (children with MR/DD)b (NDD&FL ¼ 729)

No (%)

Yes (%)

No (%)

Yes (%)

No (%)

4.3 8.3 6.0

10.3 4.1 3.6

6.8 8.4 5.8

6.7 3.4 2.1

4.1 8.5 5.9

8.2 4.6 3.1

14.5 74.3 11.2

15.9 75.2 9.0

17.3 72.7 10.0

15.6 75.5 8.8

13.5 73.9 12.6

18.0 75.6 6.4

a

All diVerences between children with MR/DD and other children, and between children with FL and other children are statistically significant at the p < .05 level. b DiVerences in the distribution of access to a usual source of health care and health insurance coverage between children with MR/DD and FL and other children with MR/DD are statistically significant at the p < .05 level. All percentages shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003).

developmental disabilities and those with functional limitations are twice as likely as other children to have a cost‐related unmet need or a delay in receiving needed medical care. F.

Description of Children by Specific Developmental Disabilities

In Table IX, we investigate diVerences in family and health by types of developmental disability. Each of these comparisons is between children with and without a particular type of medical impairment; children with more than one impairment appear in the table once for each medical impairment. Sample sizes are large for autism, cerebral palsy, and mental retardation, providing a strong basis for statistical inference. Even though the sample sizes are only 96 for children with Down syndrome and 41 for children with muscular dystrophy, the relationships are suYciently strong that many statistically significant findings (p < .05) are observed (in bold typeface). (In practice, given the low proportion of children with each type of

236

Dennis P. Hogan et al.

TABLE IX FAMILY AND HEALTH CHARACTERISTICS OF CHILDREN WITH MENTAL RETARDATION/DEVELOPMENTAL DISABILITIES, BY TYPE OF MEDICAL IMPAIRMENTa

All kids (N ¼ 65,497)

Mental retardation (NMR/DD ¼ 924, NMR ¼ 515)

Down syndrome (NDS ¼ 96)

Characteristics

Yes (%)

Yes (%)

Yes (%)

Yes (%)

Yes (%)

Parent HS or less In poverty In two‐parent family Missed 6þ school days 6þ nights in hospital 1þ visit ER/ED Health status fair/poor No usual source of care Delayed needed care (cost) Unmet need for care (cost) Uninsured

38.0

52.3

45.8

45.3

36.5

30.1

16.2 70.9

30.7 55.5

16.8 59.8

17.7 67.0

16.3 65.3

18.1 63.2

16.8

34.3

25.8

33.5

31.1

32.7

2.1

10.1

2.9

10.1

10.7

5.3

18.2

29.6

17.0

26.6

21.3

22.5

1.9

22.9

16.4

38.6

15.6

16.2

6.7

4.8

8.2

5.7

8.7

1.9

4.0

7.7

10.2

14.0

3.5

9.7

2.5

6.5

7.0

0.0

2.4

6.9

15.8

13.3

15.8

10.4

13.6

14.1

Muscular Cerebral dystrophy palsy Autism (NMD ¼ 41) (NCP ¼ 262) (NA ¼ 198) Yes (%)

Percentages in bold represent diVerences that are statistically significant at the p < .05 level. All percentages shown in the table are weighted. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003). a

developmental disability, the data for these children can be compared with the percentages for the total population of children.) Children with mental retardation are much more likely to be in homes in which the parental education is high school or less (52% versus 38% for all children). Children with autism have better educated parents than other children, with only 30% having a high school education or less. Children who are mentally retarded are much more likely to be in poverty (31% versus 16% for all children). The poverty status of children with other types of developmental disabilities does not diVer from those of all children. Children who are mentally retarded are much less likely to be in two‐parent homes

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

237

(56% versus 71% for all children). In general, children with developmental disabilities of all types less frequently live in two‐parent homes compared to all children. Children with mental retardation and each type of developmental disability are much more likely to miss six or more school days. The likelihood of staying six or more nights in the hospital is far higher for children who are mentally retarded and have cerebral palsy; they also are much more likely to visit the emergency room for care. Only 2% of all children have a health status that is classified as fair or poor. The comparable percentages are eight or more times higher for each type of medical impairment. Fair or poor health status is particularly common for children with mental retardation (23%) and children with muscular dystrophy (39%). Children with medical impairments are more likely to have a usual source of care and to have health insurance than other children. Delayed care and unmet need for care (due to cost) are more common among children who are mentally retarded and those with Down syndrome. These statistics highlight the higher risk family situations, the less favorable health outcomes, and greater unmet needs for care experienced by all American children with mental retardation and developmental disabilities. What is most striking is that the health situations of children with mental retardation are far worse than the situations of children with physical disabilities. G.

Analytic Models

For the remainder of this population‐based epidemiological analysis, we classify children according to whether they have a development disability (yes/no) or a functional limitation (yes/no). We estimate logistic regression models of the eVects of each of these disability indicators, taking into account their higher risk family situations, as well as controlling for age, sex, and race/ethnicity. In statistical terms, this provides a better indicator of the ‘‘true’’ eVects of developmental disabilities and functional limitations on indicators of health outcomes (Table X) and health care access (Table XI). These models allow a comparison of the impacts of developmental disabilities and functional limitations. Population‐based developmental epidemiology studies provide the appropriate basis to compare children with these disabilities to other children in the population. They also are well suited to explore the impact of household risk factors on health outcomes and access since the study population includes all children, and not just children who are high risk. The statistical models shown provide the odds ratios associated with membership in each category, with a coeYcient of 1.0 indicating no diVerence. Statistically significant coeYcients are in boldface. Presence of developmental disability and functional limitation each are associated with an increase in the likelihood that six or more days of school were missed and of experiencing a hospital stay longer than six nights

LOGISTIC REGRESSION RESULTS

OF

TABLE X HEALTH INDICATORS FOR CHILDREN WITH MENTAL RETARDATION/DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONS Hospitalized for 6þ nights

Explanatory variables

OR (95% CI)

Fair/poor health status

1þ ER visits

Missed 6þ school days

OR (95% CI)

OR (95% CI)

OR (95% CI)

Household income to poverty ratio Below poverty Adequate Secure (ref.)

1.2 (1.0,1.3) 1.1 (0.9,1.2) 1.0

3.1 (2.5,3.8) 1.8 (1.5,2.2) 1.0

1.3 (1.2,1.4) 1.1 (1.1,1.2) 1.0

1.3 (1.2,1.4) 1.1 (1.1,1.2) 1.0

Education Less than HS HS graduate Some college College grad. (ref.)

1.2 (1.0,1.4) 1.3 (1.1,1.5) 1.2 (1.0,1.4) 1.0

2.7 (2.2,3.4) 2.2 (1.8,2.7) 1.6 (1.3,2.0) 1.0

1.2 (1.1,1.3) 1.2 (1.1,1.3) 1.2 (1.2,1.3) 1.0

1.3 (1.2,1.4) 1.4 (1.3,1.5) 1.3 (1.2,1.4) 1.0

Family structure (in home) Neither parent One parent Both parents (ref.)

0.9 (0.7,1.2) 1.2 (1.1,1.3) 1.0

1.3 (1.0,1.6) 1.1 (1.0,1.3) 1.0

1.1 (1.0,1.3) 1.3 (1.2,1.4) 1.0

1.0 (0.9,1.1) 1.3 (1.2,1.4) 1.0

MR/DD indicator Functional limitation indicator

2.1 (1.6,2.7) 2.8 (2.4,3.1)

4.1 (3.4,4.9) 5.5 (4.9,6.2)

1.0 (0.8,1.1) 1.7 (1.6,1.8)

1.3 (1.2,1.5) 2.3 (2.2,2.4)

All models control for sex, age, and race/ethnicity, and missing data for each explanatory variable. Interactions between MR/DD indicator and FL indicator were not statistically significant in any of the models. Percentages in bold represent diVerences that are statistically significant at the p < .05 level. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003).

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

239

(Table X). Children with functional limitations are more likely to visit emergency rooms. Having a developmental disability is associated with about a 30% increase in the likelihood that six or more school days were missed compared to a 230% increase in the case of children with functional limitations. Both developmental disabilities (4.1 times) and functional limitations (5.5 times) are associated with extremely large increases in the likelihood that children are of fair or poor health status. In general, the unfavorable health outcomes of children with functional limitations are somewhat higher than the comparable figures for children with mental retardation and developmental disabilities. Children with developmental disabilities and children with functional limitations are more likely to have a usual source of care. This eVect is particularly large for children who are developmentally disabled and functionally limited (Table XI). Taking into account family risk factors, there is no diVerence in the health insurance coverage of children with mental retardation or developmental disabilities and children with functional limitations compared to children without these disabilities. Children who are developmentally disabled and have functional limitations are only half as likely as other children to lack health insurance. Taking into account family risk factors, children with developmental disabilities are as likely as other children to receive needed medical care in a timely fashion. Children with functional limitations in contrast are 2.5 times more likely to have an unmet need for care or a delay in care because of cost, even though they are equally as likely as other children to have health insurance. Family risk factors have substantial impacts on the indicators of health outcomes (Table X) and access to health care (Table XI), taking into account their somewhat greater likelihood of having developmental disabilities and functional limitations. Children in poverty are more likely than children from higher income families to have extended absences from school, lengthy hospital stays, and visits to the emergency room. Children in poverty have three times the likelihood that they are of fair or poor health status. Those children living in poverty or with less than secure incomes are much more likely to have access to health care, as are children of parents with less than a college degree. Children of parents with less than a college degree are somewhat more likely to have unfavorable health outcomes, with this impact being especially large for the likelihood of fair or poor health status. Children living in one parent families are more likely to experience negative health outcomes, but these diVerences are modest after their poorer socioeconomic statuses are taken into account. The negative eVects of risky social origins on access to health care are notable. Even taking into account their greater likelihood of being developmental disabled or functionally limited, the negative impact of poverty, low

TABLE XI LOGISTIC REGRESSION RESULTS OF ACCESS TO HEALTH CARE FOR CHILDREN WITH MENTAL RETARDATION/ DEVELOPMENTAL DISABILITIES AND FUNCTIONAL LIMITATIONS No usual source

Delayed care (cost)

OR (95% CI)

OR (95% CI)

2.6 (2.4,3.0) 2.3 (2.1,2.5) 1.0

2.9 (2.6,3.4) 2.9 (2.5,3.2) 1.0

3.9 (3.2,4.7) 3.4 (2.9,4.0) 1.0

4.8 (4.4,5.2) 4.5 (4.2,4.9) 1.0

2.8 (2.5,3.1) 1.7 (1.6,1.9) 1.4 (1.3,1.6) 1.0

1.3 (1.1,1.5) 1.4 (1.2,1.6) 1.5 (1.3,1.7) 1.0

1.7 (1.4,2.0) 1.6 (1.3,1.9) 1.5 (1.3,1.8) 1.0

3.6 (3.3,3.9) 2.3 (2.1,2.5) 1.8 (1.7,2.0) 1.0

Family structure (in home) Neither parent One parent Both parents (ref.) MR/DD indicator

1.4 (1.3,1.6) 0.9 (0.9,1.0) 1.0 1.0 (0.6,1.7)

0.9 (0.7,1.1) 1.4 (1.3,1.5) 1.0 1.0 (0.7,1.2)

0.9 (0.7,1.1) 1.5 (1.4,1.7) 1.0 0.9 (0.7,1.3)

1.3 (1.1,1.4) 1.0 (0.9,1.0) 1.0 1.0 (0.7,1.5)

Functional limitation indicator MR/DD by FL interaction

0.9 (0.8,1.0) 0.5 (0.2,0.9)

2.4 (2.2,2.6) –

2.5 (2.2,2.8) –

1.0 (0.9,1.1) 0.5 (0.3,0.8)

Explanatory variables Household income to poverty ratio Below poverty Adequate Secure (ref.) Education Less than HS HS graduate Some college College grad. (ref.)

Unmet need (cost) OR (95% CI)

All models control for sex, age, and race/ethnicity, and missing data for each explanatory variable. Model results for which the coeYcient for the interaction term between MR/DD and FL was not significant are not shown. Percentages in bold represent diVerences that are statistically significant at the p < .05 level. Source: National Center for Health Statistics. National Health Interview Surveys (1997–2003).

Uninsured OR (95% CI)

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

241

education, and non‐two‐parent family situations more often have no usual source of care, unmet need or delays in care, and to be without health insurance. It is well to remember that high‐risk family situations are at least as detrimental for quality of care and for health status as are medical impairments and disabilities in daily activities. It would be easy to neglect this key finding in a study that did not include a comparison population that represented the entire range of the experiences of American children. Again we see the advantages of developmental epidemiology for addressing health outcomes and health care of children. Taken together, the findings of this population‐based analysis of the family situations, health, and health care access of children have shown that children with developmental disabilities are disadvantaged in terms of their family situations and indicators of health. But the results show that for every indicator of health, the presence of a functional limitation has a greater negative impact than having a developmental disability. Children with mental retardation and developmental disabilities are much more likely to have a usual source of health care and health insurance compared to any other children. Children with developmental disabilities who also are functionally disabled are at particular disadvantage in health outcomes. They are less likely than any other children to lack health insurance, but 14% of these most needy children lack health insurance. Children with limitations in function are more likely to have unmet need for services or delays in services because of cost. Children with developmental disabilities are no more likely than other children to have this disadvantage in use of medical care.

V.

CONCLUSIONS

The National Health Interview Survey (NHIS) is the major study that investigates the medical impairments, disabilities, and health outcomes of the American population. This study is the major source of information about access to medical care, the use of routine medical care and special services, and the insurance status and unmet need for care. This major annual survey provides information on which government health policies are formulated. The unique design of the NHIS provides detailed information about the health situation of children (through a focus on one child in each family). Mental retardation and developmental disabilities among children are suYciently rare that general population surveys of health do not provide adequate information to conduct meaningful analysis. By combining seven waves of data collected in the NHIS from 1997 to 2003, we were able to construct a nationally representative sample of children with developmental disabilities. The merged

242

Dennis P. Hogan et al.

dataset also provides an excellent basis for studying the well‐being of children with health disabilities. With population data, it is possible to investigate the family situations of children with mental retardation and developmental disabilities. The NHIS data provides information on current family situations, but not the history of marital status and poverty prior to the birth of the child with mental retardation and developmental disabilities. With this information, we are able to characterize the family situations of these children and the extent to which social disadvantage exacerbates their health care and health outcomes. Population data permitted the examination of real‐world health outcomes among children with mental retardation and developmental disabilities. Client‐ based studies systematically underrepresent those children whose medical conditions are associated with only modest health disadvantage. (This is because such children either will not use particular medical services and thus not be included in the client population or require care so infrequently that they are unlikely to be included in any time bound sample of patients receiving care.) Another advantage of a population‐based sample is that it provides comparison populations (well children and children with limitation in functioning). We thus were able to demonstrate that functional disability has a greater detrimental impact of health outcomes than developmental disabilities. The real‐world outcomes investigated here included school days missed, use of emergency room services, and hospitalization. With population level data, we were able to identify children who lack access to health care or have unmet need for medical services. That large numbers of children with mental retardation and developmental disabilities still lack timely medical care or actually do not get needed care because their parents are unable to pay for the cost of services is clearly unacceptable and demonstrates the need for new public policies to further improve health care utilization for children with special needs. Population‐based studies of children with mental retardation and developmental disabilities are uniquely valuable to address several critical research issues. But the relatively rare prevalence of these children in the population necessitates the use of studies with extremely large sample sizes. Prospective longitudinal studies are needed to determine the family consequences of a child who is mentally retarded or developmentally disabled. One major American survey—the Early Childhood Longitudinal Study Birth Cohort—is the appropriate design for a developmental epidemiology study, but the sample size of approximately 10,000 children will yield relatively few children with mental retardation and developmental disabilities. The planned National Children’s Study will use a cohort longitudinal design for an original sample of 100,000 children and would serve as a major

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

243

new resource for research on American children that uses a developmental epidemiology approach. Our analysis has provided a strong argument for the value of population‐based epidemiological studies of children with mental retardation and developmental disabilities. At the same time, the limitations of such studies mean that they should be used as a complement to studies based on patients or clients, the other type of design for developmental epidemiology.

REFERENCES Aylward, G. P. (1993). Review of the book Advances in child neuropsychology, Volume 1 Contemporary Psychology, 38, 810. Bowe, F. G. (1995). Population estimates: Birth to 5 children with disabilities. The Journal of Special Education, 28, 461–471. Correa‐Villasenor, A., Cragan, J., Kucik, J., O’Leary, L., SiVel, C., & Williams, L. (2003). The metropolitan Atlanta congenital defects program: 35 years of birth defects surveillance at the Centers for Disease Control and Prevention. Birth Defects Research (Part A), 67, 617–624. Durkin, M. (2002). The epidemiology of developmental disabilities in low‐income countries. Mental Retardation and Developmental Disabilities Research Reviews, 8, 206–211. Green, N. S. (2004). Risks of birth defects and other adverse outcomes associated with assisted reproductive technology. Pediatrics, 114, 256–259. Hall, I., Strydom, A., Richards, M., Hardy, R., Bernal, J., & Wadsworth, M. (2005). Social outcomes in adulthood of children with intellectual impairment: Evidence from a birth cohort. Journal of Intellectual Disability Research, 49, 171–182. Hogan, D. P., & Msall, M. E. (in press). Key indicators of health and safety: Infancy, pre‐school, grade school, and middle school. In B. Brown (Ed.), Key indicators of child and youth well‐being: Completing the picture. Mahwa, NJ: Lawrence Erlbaum Associates, Publisher. Larson, S. A., Lakin, K. C., Anderson, L., Kwak, N., Lee, J. H., & Anderson, D. (2001). Prevalence of mental retardation and developmental disabilities: Estimates from the 1994/1995 national health interview survey disability supplements. American Journal on Mental Retardation, 106, 231–252. Leonard, H., & Wen, X. (2002). The epidemiology of mental retardation: Challenges and opportunities in the new millennium. Mental Retardation and Developmental Disabilities Research Reviews, 8, 117–134. Leonard, H., Petterson, B., De Klerk, N., Zubrick, S. R., Glasson, E., Sanders, R., et al. (2005). Association of sociodemographic characteristics of children with intellectual disability in Western Australia. Social Science and Medicine, 60, 1499–1513. Luckasson, R., Coulter, D. L., Polloway, E. A., Reiss, S., Schalock, R. L., Snell, M. E., et al. (1992). Mental retardation: Definition, classification, and systems of supports (9th ed.). Washington, DC: American Association on Mental Retardation. Luckasson, R., Borthwick‐DuVy, S., Buntinx, W. H. E., Coulter, D. L., Craig, E. M., Reeve, A., et al. (2002). Mental retardation: Definition, classification, and systems of supports (10th ed.). Washington, DC: American Association on Mental Retardation.

244

Dennis P. Hogan et al.

Msall, M. E., & Tremont, M. R. (2000). Functional outcomes in self‐care, mobility, communication, and learning in extremely low‐birth weight infants. Clinical Perinatology, 27, 381–401. National Center for Health Statistics (2000). Data File Documentation, National Health Interview Survey, 1998 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Center for Health Statistics (2002a). Data File Documentation, National Health Interview Survey, 1999 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Center for Health Statistics (2002b). Data File Documentation, National Health Interview Survey, 2000 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Center for Health Statistics (2003a). Data File Documentation, National Health Interview Survey, 2001 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Center for Health Statistics (2003b). Data File Documentation, National Health Interview Survey, 2002 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Center for Health Statistics (2004). Data File Documentation, National Health Interview Survey, 2003 (machine readable data file and documentation). Hyattsville, MA: National Center for Health Statistics, Centers for Disease Control and Prevention. National Research Council and Institute of Medicine (2004). Children’s health, the nation’s wealth: Assessing and improving child health. Committee on evaluation of children’s health. Board on children, youth and families, division of behavioral and social science and education. Washington, DC: The National Academies Press. Ramadhani, T. A., Canfield, M. A., Waller, D. K., & Case, A. P. (2004). Medical records vs. interview responses: A comparative analysis of selected variables for linked birth defect cases. Birth Defects Research (Part A), 70, 592–596. Rasmussen, S. A., & Moore, C. A. (2001). EVective coding in birth defects surveillance. Teratology, 64, S3–S7. Sattler, J. M., & Aylward, G. P. (2001). Assessment of children: Cognitive applications (4th ed.). San Diego, CA: Jerome M. Sattler Publisher. Semel, E., Wiig, E., & Secord, W. (2003). Clinical evaluation of language fundamentals (4th ed.). San Antonio, TX: The Psychological Corporation. Simeonsson, R. J., Lollar, D. J., Hollowell, J., & Adams, M. (2000). Revision of the international classification of impairments, disabilities and handicaps: Developmental issues. Journal of Clinical Epidemiology, 53, 113–124. Spencer, N. J. (2005). Does material disadvantage explain the increased risk of adverse health, educational, and behavioral outcomes among children in lone parent households in Britain? A cross sectional study. Journal of Epidemiology and Community Health, 59, 152–157. Stein, R. E. K., & Silver, E. J. (1999). Operationalizing a conceptually based noncategorical definition. Archives of Pediatric & Adolescent Medicine, 153, 68–74. Sundrum, R., Logan, S., Wallace, A., & Spencer, N. (2005). Cerebral palsy and socioeconomic status: A retrospective cohort study. Archives of Disease in Childhood, 90, 15–18. Taanila, A., Murray, G. K., Jokelainen, J., Isohanni, M., & Rantakallio, P. (2005). Infant developmental milestones: A 31‐year follow‐up. Developmental Medicine and Child Neurology, 47, 581–586. The Diabetes Control and Complications Trial Research Group (1993). The eVect of intensive treatment of diabetes on the development and progression of long‐term complications in insulin‐dependent diabetes mellitus. New England Journal of Medicine, 329, 977–986.

DEVELOPMENTAL EPIDEMIOLOGY OF

MR/DD

245

Toft, P. B., Lou, H. C., Kra¨geloh‐Mann, I., Andresen, J., Gu¨ttler, F., Guldberg, P., et al. (1994). Brain magnetic resonance imaging in children with optimally controlled hyperphenylalaninaemia. Journal of Inherited Metabolic Disease, 17, 575–583. Vohr, B. R., Wright, L. L., Dusick, A. M., Mele, L., Verter, J., Steichen, J. J., et al. (2000). Neurodevelopmental and functional outcomes of extremely low birth weight infants in the National Institute of Child Health and Human Development Neonatal Research Network, 1993–1994. Pediatrics, 105, 1216–1226. Wells, T., & Hogan, D. (2003). Developing concise measures of childhood activity limitations. Maternal and Child Health Journal, 7, 115–126. World Health Organization (2001). International classification of functioning, disability, and health. Geneva, Switzerland: World Health Organization. Yeargin‐Allsopp, M., & Boyle, C. (2002). Overview: The epidemiology of neurodevelopmental disorders. Mental Retardation and Developmental Disabilities Research Reviews: The Epidemiology of Neurodevelopmental Disorders, 8, 113–116.

This page intentionally left blank

Evolution of Symptoms and Syndromes of Psychopathology in Young People with Mental Retardation STEWART L. EINFELD SCHOOL OF PSYCHIATRY, UNIVERSITY OF NEW SOUTH WALES, NSW, AUSTRALIA

BRUCE J. TONGE, KYLIE GRAY, AND JOHN TAFFE MONASH UNIVERSITY CENTRE FOR DEVELOPMENTAL PSYCHIATRY, VICTORIA, AUSTRALIA

I.

INTRODUCTION

It is well established that children and adolescents with developmental disabilities have high rates of psychopathology (Einfeld & Tonge, 1996a,b; Rutter, Tizard, & Whitmore, 1970). Despite this, little research has been done to examine the course and development of psychopathology in this population over time. Longitudinal studies are essential in order to understand fully the nature and course of a problem, to examine risk and protective factors in the development or amelioration of pathology, and to thus inform the development of preventative and intervention programs. To date, only six studies have examined psychopathology in children and adolescents with mental retardation (MR) longitudinally. The first of these studies revisited a birth cohort of children born in the 1950s in Aberdeen, Scotland when they were 22 years of age (Richardson & Koller, 1996). In a sample of 221, it was found that 65% of those who had behavioral disturbance as children continued to have behavioral problems in young adulthood. Thirty‐nine percent of males had behavioral disturbance in childhood, and in adulthood this had decreased slightly to 34%. However, in females behavioral disturbance increased slightly from childhood to young adulthood (35–42%). For the females, the level of emotional problems increased from childhood to young adulthood, with the exception of those in INTERNATIONAL REVIEW OF RESEARCH IN MENTAL RETARDATION, Vol. 33 0074-7750/07 $35.00

247

Copyright 2007, Elsevier Inc. All rights reserved. DOI: 10.1016/S0074-7750(06)33010-8

248

Stewart L. Einfeld et al.

the mild range of MR. In childhood, antisocial behavior problems were found to be 3–5 times higher in males than in females, a finding which persisted into adulthood. In males, the frequency of antisocial behavior problems also increased with increasing IQ. Although the study was the first to demonstrate that the high rates of psychopathology in children with MR persisted from childhood to young adulthood, there were very few subjects with moderate or severe MR, thus limiting any conclusions that could be drawn regarding this group. Chess (1977) reviewed 44 children (5–12 years of age) with mild MR 6 years after initial assessment. At Time 1, 60% of the sample received a diagnosis of a psychiatric disorder, while at Time 2, 41% received a diagnosis. Although there was a decrease in the percentage of children diagnosed, the small sample size limits the conclusions that can be drawn from this study. Quine and Pahl (1989) reassessed 166 children 4 years after initial assessment. Of those who were classified as having ‘‘behavior problems’’ at Time 1, 76% still met this criterion at Time 2. Overactivity was the only behavior problem which decreased from Time 1 to Time 2. A study conducted in the United States (Alabama) examined psychopathology in a predominantly African‐American population of 237 adolescents (13–16 years of age) with mild MR (Wallander, Frison, & Rydvalova, 2001). Risk for psychopathology in African‐American adolescents with mild intellectual disability was investigated. The adolescents were assessed annually at three time points. High rates of psychopathology were reported, which were stable over time. Similar results were reported in a Dutch study which surveyed 968 children aged 6–18 years, with mild–moderate levels of MR. Thus far, 1 year follow‐up has demonstrated persistently high levels of psychopathology, with 71% of those who met criteria for clinical caseness still meeting criteria 1 year later (Wallander, Dekker, & Koot, 2003). Chadwick, Kusel, Cuddy, and Taylor (2005) reassessed 82 children with severe intellectual disability 5 years later when they were adolescent. They found little diVerence in rates of behavior problems between the two assessment occasions, apart from overactivity which declined significantly. II.

THE AUSTRALIAN CHILD TO ADULT DEVELOPMENT (ACAD) STUDY

In 1989/1990, a comprehensive attempt was made to identify all young people aged 4–18 years with MR who lived in a number of census regions in the Australian states of New South Wales and Victoria (Einfeld & Tonge, 1996a). These regions were local government areas, which together represented a cross section of the Australian community, particularly for social

EVOLUTION OF PSYCHOPATHOLOGY

249

class, mix of ethnic origin, and urban/rural distribution, which may be factors associated with psychopathology. The epidemiological cohort was recruited from all health, education, and family agencies that provide services to young people with MR of all levels whose families lived in the selected census districts. Children who were not living with their parents but were in institutional or small group care were included provided their parent lived within one of the census regions. These criteria ensured the inclusion of institutionalized children who tend to have a higher level of behavior disturbance (Einfeld & Tonge, 1992). Registration with regional Disability Services provided the mechanism for the provision of state‐funded services for young people and their families. Since those with IQ less than 50 (moderate to severe or profound MR) virtually always require some health, education, or welfare service, this longitudinal study was likely to have achieved a virtually complete ascertainment of this population in Australia (Einfeld & Tonge, 1996a). Consistent with other studies, some young people with mild MR blend in with the general population and do not receive any specific health, education, or welfare services. Therefore, the young people included in this study with mild MR are likely to be biased toward the lower end of the mild MR range or have medical conditions associated with MR, such as epilepsy and cerebral palsy (Einfeld & Tonge, 1992, 1996a), and/or have emotional or behavioral problems that bring them into contact with services. Therefore, the sample is likely to be representative of young people with moderate or more severe levels of MR, but is only representative of those with mild MR who have some reason to receive health, education, or welfare services for persons with MR. The longitudinal study also recruited separate cohorts of young people with MR aged 4–18 years who had Fragile X (FraX) syndrome, Williams syndrome (WS), and Prader–Willi syndrome (PWS). These groups were recruited in New South Wales from specialist genetics clinics and parents’ support associations, but not from services for children with behavioral disturbance. There was also a group of young people with Down syndrome (DS) identified within the epidemiological cohort. Another group of young people diagnosed with Autistic Disorder according to DSM‐IV criteria (American Psychiatric Association, 1994) were also included in the longitudinal study. These young people were all identified through regional autism assessment services and are likely to be representative of all children in the community who are assessed to have this condition and receive health, education, and welfare services. The participation rate of all the young people with MR identified in the census areas was 80.2%. Further descriptions of the cohort and the evidence for its epidemiological validity have been reported previously (Einfeld and Tonge, 1996a).

250

Stewart L. Einfeld et al. III.

METHODS

The study has gathered data on a broad range of potential biopsychosocial risk and protective variables including causes of MR, and measures of life events, parental mental health, and family functioning in 1991/1992 (Time 1), 1996/1997 (Time 2), 1999/2000 (Time 3), and 2003/2004 (Time 4). The major measure of psychopathology was the Developmental Behaviour Checklist (DBC; Einfeld & Tonge, 1992). The DBC has 96 items completed by the parents or other primary carers reporting problems with emotions or behavior over a 6‐month period. The instrument has high interrater reliability between parents (ICC ¼ .80), nurses (ICC ¼ .83), and between teachers and teachers’ aides (ICC ¼ .60). Test retest reliability (ICC ¼ .83) and internal consistency are high (Cronbach’s alpha ¼ 0.94). The DBC has five factor analytically derived subscales (Disruptive/Antisocial, Self‐absorbed, Communication Disturbance, Anxiety, and Social Relating). Content, criterion, construct and/or concurrent validity have been demonstrated for the total score and the subscales. The total behavior problem score (TBPS) on the DBC correlates with child psychiatrist’s rating of severity of psychopathology using Rutter’s (Cox & Rutter, 1985) definition (r ¼ 0.81, p < .001). The instrument has high criterion group validity in distinguishing psychiatric cases from noncases (t ¼ 7.8, p < .001). The Receiver Operating Characteristics (ROC) of the DBC were examined in 70 individuals for whom checklists were completed and who were also assessed by 2 of 3 child psychiatrists and 1 experienced clinical psychologist with an overall rating of severity of psychopathology. The area under the ROC curve was 92%, indicating that the DBC provides a cut‐oV with high specificity and sensitivity. The DBC depression scale (Einfeld & Tonge, 2002) is based on 10 items from the DBC with a total possible score of 20. The scale has face validity for depression and includes the items ‘‘Appears depressed, downcast or unhappy,’’ ‘‘Irritable,’’ and ‘‘Cries easily for no reason, or over small upsets.’’ The DBC attention deficit hyperactivity (ADHD) score (Einfeld & Tonge, 2002) is a six item scale, derived from items in the DBC. Items include ‘‘Becomes overexcited,’’ ‘‘Impulsive, acts before thinking,’’ and ‘‘Cannot attend to one activity for any length of time, poor attention span.’’ Increased intensity of ADHD symptoms is reflected by a higher total score on the six items. High internal consistency (Cronbach’s alpha ¼ 0.88) and construct validity of the scale has been established. For a more detailed description of this measure, see Einfeld and Tonge (1995). For the purposes of studying the possible eVects of level of MR on behavior, we assigned each child to a mild, moderate, severe, or profound level of MR using DSM‐IV degree of severity of MR criteria IQ score ranges. As part of the data we gathered from families, we obtained adaptive

EVOLUTION OF PSYCHOPATHOLOGY

251

behavior information on social competence, type of school or work place attendance, type and range of social networks, recreational interests and participation, daily living skills, and language and communication ability, which further informed our assignment to MR range. The level of MR was evaluated by viewing the reports of individual IQ assessments undertaken in the past 3 years. The cognitive assessments had been done by registered psychologists, who were employed by the various MR services agencies. When the reports were not available or out of date, level of MR was assessed by psychometric tests such as the WISC‐III (Wechsler, 1981) or Stanford–Binet (Terman & Merrill, 1960). It is possible that subjects in the upper range of MR theoretically may not meet the adaptive behavior criterion (criterion B) of the DSM‐IV definition, despite the fact that they qualified for services. Nevertheless there are only 29 subjects in this upper mild range in the epidemiological cohort of 574. Any eVect on prevalence rates caused by inclusion of these subjects would be minimal. The success of this longitudinal study is dependent on the ongoing participation of the young people and their families. Considerable eVort has been expended to keep in touch with the families and track them if they move. The participation rate of Time 1 families at Time 2 was 91%, at Time 3 was 82%, and at Time 4 was 81%. This high rate and the lack of any significant demographic, age, sex, IQ level, or psychopathology score diVerence between participants and nonparticipants strengthens confidence in the representativeness of the study findings. In this chapter, we describe changes in some parameters of psychopathology over the 11‐year period. We have selected some representative externalizing and internalizing behaviors of interest. IV.

DATA ANALYSIS

Longitudinal regressions of each of the DBC‐based dependent variables on age of entry to the ACAD study, aging during the study (both in years), gender (girl ¼ 1, boy ¼ 0), MR level (severe or profound ¼ 1, mild or moderate ¼ 0), whether the same respondent throughout completed the DBC‐P or DBC‐A (yes ¼ 1, no ¼ 0), group (epidemiological cohort or syndrome group), and interactions of group and gender with aging in the study were estimated using Stata version 9 (StataCorp, 2005). Of these independent variables, only aging during the study and interactions with it varied with time. The others are ‘‘between subject’’ variables. The reference group for each syndrome group was the epidemiological cohort except those identified as having the syndrome in question. This reference group included those with DS, except for regressions in which DS is an independent variable.

252

Stewart L. Einfeld et al. AGE SUMMARIES

TABLE I DATA COLLECTION TIME

BY

Time

n

Mean

SD

1 2 3 4

834 694 670 652

12.0 16.3 19.2 23.0

5.3 5.2 5.4 5.3

V. A.

RESULTS

Age Age summaries are presented in Table I.

B.

Aging During the Study

624 participants present at data waves 1 and 4 had been in the study for an average of 11.2 years (SD. 97, minimum 7.6, maximum 13.6). C.

Gender

The overall proportion of girls for whom the DBC was completed remained between .37 and .38 during Times 1–4. It was lower than this for the groups with autism (.18–.19) and FraX (.16–.20), about the same for PWS (.34–.39), and higher for WS (.46–.49) and DS (.57–.59). D.

MR Level

The overall proportion of young people with severe or profound MR (as opposed to mild or moderate MR) for whom the DBC was completed remained between .18 and .20 during Times 1–4. It was higher than this in the non‐DS epidemiological cohort (.28–.29), but lower in each of the syndrome groups: PWS (0), FraX (.02–.03), DS (.04–.06), WS (.07–.10), and autism (.13–.15). E.

Relationship to Respondent

The overall proportion of young people with no change in relationship to respondent for whom the DBC was completed remained between .69 and .73 during waves 1–4. It was higher for WS (.85–.88) and lower for PWS (.54–.63), but similar in other groups.

253

EVOLUTION OF PSYCHOPATHOLOGY

Mean disruptive score

0.8

Epidemiological

0.6

Autism DS PWS

0.4

FraX syndrome WS

0.2 1991/1992 1996/1997 1999/2000 2003/2004 Time FIG. 1. DBC disruptive scores.

1. EXTERNALIZING PSYCHOPATHOLOGY

DBC ‘‘disruptive’’ subscale score. Here, to illustrate the type of analysis conducted, we show results using a figure Fig. 1, a table of means Table II, and a regression Table III. In the results following, in order to save space, we show a figure with means Figs. 2–9, and describe the significant results from the regression analysis. In this and subsequent descriptions of the results regarding aging in the study, we give the p‐value for one of the three relevant coeYcients where it is appropriate. Estimates relying on more than one of these coeYcients are given without p‐values. The DBC disruptive subscale score: declines slightly by about .01 (p < .001) for boys, by less than this (.003–

.005) for girls with each year of aging in the study. For those with WS, the decline with aging is faster, by an extra .01 per year than for those in the non‐WS epidemiological cohort is not significantly diVerent between girls and boys is lower by about .15–.18 (p < .001) for those with severe or profound MR is higher by .23 for PWS (p < .001) and by .10 for autism (p < .01), lower by .18 for DS (p < .001), than for those in the epidemiological cohort without the relevant syndrome.

TABLE II MEAN DISRUPTIVE SUBSCALE SCORE Time

Epidemiological cohort

Autism

Fragile X (FraX) syndrome

Williams syndrome (WS)

Prader–Willi syndrome (PWS)

Down syndrome (DS)

1 2 3 4

0.50 0.45 0.42 0.40

0.59 0.60 0.54 0.55

0.46 0.48 0.37 0.38

0.68 0.58 0.54 0.41

0.73 0.72 0.66 0.63

0.40 0.36 0.28 0.27

TABLE III REGRESSION TABLE FOR DISRUPTIVE SUBSCALE

Age at entry Aging in study Girl Girl*aging in study Severe/profound MR Same respondent relationship Syndrome Syndrome*aging in study Const

Epidemiological cohort (E)

Autism versus non‐autism E

Down versus non‐Down E

PWS versus non‐PWS E

FraX versus non‐FraX E

WS versus non‐WS E

.002 .011 .005 .006 .159 .030

.002 .011 .001 .006 .148 .026 .098 .001 .494

.002 .011 .011 .007 .184 .026 .184 .005 .531

.003 .011 .013 .005 .155 .040 .230 .001 .515

.003 .012 .004 .007 .160 .020 .061 .006 .530

.002 .011 .0005 .007 .148 .028 .076 .011 .506

.504

Typeface code for p‐values: ns <.05 <.01 <.001. Asterisk (*) indicates interation.

256

Stewart L. Einfeld et al. 1.2 Epidemiological

Mean ADHD score

1.0

Autism

0.8 DS

0.6

PWS FraX syndrome

0.4

WS

0.2 1

2

3

4

Time FIG. 2. DBC ADHD scores.

Mean "abusive" score

0.7 Epidemiological Autism

0.5 DS PWS

0.3 FraX syndrome WS

0.1 1

2

3

4

Time

FIG. 3. DBC ‘‘abusive’’ score.

DBC attention deficit hyperactivity scale. The mean DBC scores for the attention deficit hyperactivity scale are shown in Fig. 2. The DBC attention deficit hyperactivity score:

257

EVOLUTION OF PSYCHOPATHOLOGY

1.8

Mean "tantrum" score

Epidemiological

1.4 Autism DS

1.0

PWS

0.6

FraX syndrome WS

0.2 1

2

3

4

Time FIG. 4. DBC ‘‘tantrum’’ scores.

Mean "kicks" score

0.8

Epidemiological

0.6

Autism

0.4

DS PWS

0.2

FraX syndrome WS

0.0 1

2

3

4

Time FIG. 5. DBC ‘‘kicks’’ score.

is lower by about .01 per year of older age at entry to the study ( p < .01) declines by .02 ( p < .001) per year of aging in the study for boys and

by about .01 per year for girls. For WS, the ADHD score declines by .04 per year for boys and .03 per year for girls is higher by .21 for both those with autism ( p < .001) and with WS ( p < .001), and lower by .36 for those with DS ( p < .001) than for those in the epidemiological cohort without the relevant syndrome.

258

Stewart L. Einfeld et al.

Mean "throws" score

0.8

0.6

Epidemiological Autism

0.4

DS PWS

0.2

FraX syndrome WS

0.0 1

2

3

4

Time FIG. 6. DBC ‘‘throws’’ scores.

Mean self-absorbed score

0.7 Epidemiological Autism

0.5 DS PWS

0.3 FraX syndrome WS

0.1 1

2

3

4

Time FIG. 7. DBC self‐absorbed scores.

2. SOME DBC DISRUPTIVE SUBSCALE ITEMS

‘‘Abusive’’ (abusive, swears at others). The mean DBC ‘‘abusive’’ scores are shown in Fig. 3. The DBC ‘‘abusive’’ item score:

259

EVOLUTION OF PSYCHOPATHOLOGY

Mean anxiety score

0.8

Epidemiological

0.6 Autism DS PWS

0.4

FraX syndrome WS

0.2 1

2

3

4

Time FIG. 8. DBC anxiety score.

Mean depression score

0.8

Epidemiological

0.6

Autism DS PWS

0.4

FraX syndrome WS

0.2 1

2

3

4

Time FIG. 9. DBC depression score.

is not related to aging for boys increases slightly (by up to .01 per year) with aging for girls ( p < .05,

p < .01 for FraX versus epidemiological cohort without FraX, and not significant for autism versus epidemiological cohort without autism)

260

Stewart L. Einfeld et al.

increases slightly with aging for those with autism ( p < .05) is low (practically zero) for those with severe or profound ID ( p < .001) is lower by .23 for DS ( p < .001), higher by .29 for PWS ( p < .001).

Tantrums (has temper tantrums, e.g., stamps feet, slams doors). The mean DBC tantrum scores are shown in Fig. 4. The DBC ‘‘tantrums’’ item score: declines with aging for boys in the study by .04 per year (p < .001), more

slowly for girls (.01), faster in WS (.06 for boys, .03 for girls), and not so noticeably in FraX (decline of .015 for boys, increase of .014 for girls) is low for those with severe or profound MR (p < .01) is lower by .45 for DS (p < .001), higher by .31 for autism (p < .001), and higher by .54 for PWS (p < .001). ‘‘Kicks’’ (kicks, hits others). The mean DBC ‘‘Kick’’ scores are shown in Fig. 5. The DBC ‘‘kicks’’ item score: declines with aging in the study (.02 per year) (p < .001), faster in WS

(.04) (p < .05)

is lower by .26 for DS (p < .001), higher by .22 for autism (p < .001).

‘‘Throws’’ (throws or breaks objects). The mean DBC ‘‘throws’’ scores are shown in Fig. 6. The DBC ‘‘throws’’ item score: is very slightly lower for those with older age at entry to the study

(.01 per year) (p < .05)

declines with aging in the study for boys by .02 per year (p < .001) and

hardly at all for girls, faster for those with autism (.05 per year for boys, .03 per year for girls) is lower by .2 for DS (p < .001), higher by .17 for autism (p < .001), and higher by .23 for PWS (p < .01) than for those in the epidemiological cohort without the corresponding syndrome. 3. INTERNALIZING PSYCHOPATHOLOGY

DBC self‐absorbed scale. The mean DBC self‐absorbed scores are shown in Fig. 7. The DBC self‐absorbed behavior subscale score: is slightly lowered by older age at entry to the study (by just under .01

per year older) (p < .001)

EVOLUTION OF PSYCHOPATHOLOGY

261

declines very slightly (by about .01) with each year of aging in the study

( p < .001), except for those with FraX syndrome. The decline is slightly slower for girls (.008 per year, though not significantly except in the analyses of autism versus non‐autistic epidemiological cohort and FraX versus non‐FraX epidemiological cohort). For those with WS, the decline with aging is faster, by an extra .01 per year than for those in the epidemiological cohort without WS is not significantly diVerent between girls and boys (although girls have marginally lower means) is higher by about .25–.28 for those with severe or profound MR ( p < .001) is higher by .25 for autism (p < .001), lower by .12 for DS (p < .001), than for those in the epidemiological cohort without these syndromes. Anxiety. The mean DBC anxiety scores are shown in Fig. 8. The DBC anxiety subscale score: declines very slightly (by about .01) with each year of aging in the study

for boys (p < .001), except for those with FraX syndrome. For those with WS, the decline with aging is faster, by an extra .01 per year than for those in the epidemiological cohort without WS. The rates of decline are slower (by .007–.008) for girls in all groups is lower by about .06–.08 for those with severe or profound MR (p < .05) is higher by .16 for autism (p < .001) and by .14 for WS (p < .001), lower by .14 for DS (p < .001), than for those in the epidemiological cohort without the corresponding syndrome. Depression. The mean DBC depression scores are shown in Fig. 9. The DBC depression scale score: is higher by .06 for girls than for boys (p < .01) remains virtually constant for boys, except in WS, as they age in the

study and increases very slightly for girls (by .003–.005 per year), but this increase is significant (p < .05) only in the analyses of FraX and WS versus their corresponding reference groups. In WS, boys’ depression declines on average by .008 per year (p < .05) is higher by .10 for autism (p < .001) and by .25 for PWS (p < .001), lower by .18 for DS (p < .001), than for those without the corresponding syndrome in the epidemiological cohort.

262

Stewart L. Einfeld et al. VI.

A.

DISCUSSION

Externalizing Behavior

Disruptive behavior was found to decline slowly with aging. At the public health level, this suggests that through the childhood and adolescent years, programs should address disruptive behavior, as well as the educational needs of children with MR. For families, there is a need to increase availability and establish eVectiveness of a number of programs of demonstrated eYcacy in treatment of disruptive behaviors in children and adolescents with MR. Such programs include those by Sanders, Mazzucchelli, and Studman (2003a,b) and Wacker et al. (1998). At the same time, the demands on services for behavioral supports should decrease somewhat for older persons. There are limits to the extent to which one can generalize from group outcomes to prognosis for any given individual. Nevertheless, it should be of some legitimate reassurance for families to know that, at least in general, disruptive behavior declines with time. It is not surprising that attention deficit hyperactivity scores decline with aging, consistent with findings in non‐MR children (Wilens, Biederman, & Spencer, 2002). However, the course of externalizing symptoms or individual externalizing behaviors is not uniform. Scores for the item ‘‘Abusive, Swears at others’’ increase over time for girls and those with autism. The items ‘‘Has temper tantrums, e.g., stamps feet, slams doors,’’ ‘‘Throws or breaks objects,’’ and ‘‘Kicks, hits others’’ all decline progressively. B.

Internalizing Behavior

The self‐absorbed score, which is associated with lower IQ, declines. This may represent a developmental phenomenon of increased mental age. The pattern of change in the anxiety subscale is of some interest. Anxiety scores decline in both genders with aging, significantly more so for boys. This is unlike the pattern of change seen in normally developing children, in whom an increase in some anxiety disorders is observed postpuberty. The reason for failure to observe this postpubertal increase in anxiety is unknown. Depression scores on the other hand, while also higher in girls, do not decline over time. Again, the expected increase in adolescence is not apparent (Costello, Egger, & Angold, 2005). C.

Down Syndrome

As has been noted elsewhere, participants with DS have less psychopathology across the whole range of symptoms and syndromes.

EVOLUTION OF PSYCHOPATHOLOGY

D.

263

Prader–Willi Syndrome

Disruptive behavior scores are higher, especially for the items ‘‘abusive’’ (Abusive, swears at others), ‘‘tantrums’’ (Has temper tantrums, e.g., stamps feet, slams doors), and ‘‘throws’’ (Throws or breaks objects), though not for ‘‘kicks’’ (Kicks, hits others). Perhaps this last behavior requires more physical movement than what comes readily for people with PWS. Anxiety is not associated with PWS, but depression is associated. E.

Williams Syndrome

WS participants have high initial scores but these decline significantly more quickly than those of the non‐WS epidemiological cohort across most externalizing and internalizing dimensions. Previous literature has noted the significantly higher anxiety scores in WS compared with other persons with MR (Dykens, 2003; Einfeld, Tonge, & Florio, 1997), and this is confirmed here. However, we are not aware of the observation previously that this elevated anxiety also declines significantly more quickly. F.

Fragile X Syndrome

FraX participants had lower initial levels of externalizing behaviors, but these were relatively stable. Internalizing behavior trends showed a mixed pattern. Scores on the ‘‘self‐absorbed’’ scale tend to persist in FraX syndrome in comparison to others whose scores decline. Scores increase in anxiety but are stable for depression and ‘‘self‐absorbed’’ scores remain steady. G.

Implications for Behavioral Phenotypes

These longitudinal data contribute to an extra dimension in delineating behavior phenotypes. The data demonstrate that behavioral characteristics in genetic disorders are not static, and further, that changes over time are quite specific to the particular disorder. This implies that the eVects of genomic lesions on behavior pathways are interacting with and impacting on other developmental processes. Thus, we can anticipate that we will need to describe not just gene‐to‐behavior pathways in a particular syndrome, but rather gene‐to‐behavior pathways in that syndrome at each life stage. REFERENCES American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: American Psychiatric Association Press.

264

Stewart L. Einfeld et al.

Chadwick, O., Kusel, Y., Cuddy, M., & Taylor, E. (2005). Psychiatric diagnoses and behaviour problems from childhood to early adolescence in young people with severe intellectual disabilities. Psychological Medicine, 35, 751–760. Chess, S. (1977). Evolution of behavior disorder in a group of mentally retarded children. Journal of the American Academy of Child & Adolescent Psychiatry, 16, 5–18. Cox, A., & Rutter, M. (1985). Diagnostic appraisal and interviewing. In M. Rutter & L. Hersov (Eds.), Child and adolescent psychiatry: Modern approaches (2nd ed., pp. 233–247). Oxford, England: Blackwell Scientific. Costello, E. J., Egger, H., & Angold, A. (2005). 10‐year research update review: The epidemiology of child and adolescent psychiatric disorders: I. Methods and public health burden. Journal of the American Academy of Child & Adolescent Psychiatry, 44(10), 972–986. Dykens, E. M. (2003). Anxiety, fears, and phobias in persons with Williams syndrome. Developmental Neuropsychology, 23(1–2), 291–316. Einfeld, S. L., & Tonge, B. J. (1992). Manual for the developmental behaviour checklist. Melbourne and Sydney: Monash University Centre for Developmental Psychiatry and School of Psychiatry, University of N.S.W. Einfeld, S. L., & Tonge, B. J. (1995). The developmental behaviour checklist: The development and validation of an instrument to assess behavioural and emotional disturbance in children and adolescents with MR. Journal of Autism and Developmental Disorders, 25, 81–104. Einfeld, S. L., & Tonge, J. (1996a). Population prevalence of psychopathology in children and adolescents with intellectual disability: I. Rationale and methods. Journal of Intellectual Disability Research, 40, 91–98. Einfeld, S. L., & Tonge, J. (1996b). Population prevalence of psychopathology in children and adolescents with intellectual disability: II. Epidemiological findings. Journal of Intellectual Disability Research, 40, 99–109. Einfeld, S. L., & Tonge, B. J. (2002). Manual for the developmental behaviour checklist (Second Edition)—primary carer version (DBC‐P) and teacher version (DBC‐T). Sydney and Melbourne: School of Psychiatry University of New South Wales and Centre for Developmental Psychiatry and Psychology, Monash University. Einfeld, S. L., Tonge, B. J., & Florio, T. (1997). Behavioural and emotional disturbance in individuals with Williams syndrome. American Journal on Mental Retardation, 102, 45–53. Quine, L., & Pahl, J. (1989). Stress and coping in families caring for a child with severe mental retardation: A longitudinal study. Canterbury, England: University of Kent. Richardson, S. A., & Koller, H. (1996). Twenty‐two years: Causes and consequences of mental retardation. Cambridge, MA: Harvard University Press. Rutter, M. L., Tizard, J., & Whitmore, K. (1970). Education, health and behaviour. London: Longmans. Sanders, M. R., Mazzucchelli, T., & Studman, L. J. (2003a). Practitioner’s manual for Stepping Stones Triple P. Brisbane, QLD, Australia: Triple P International. Sanders, M. R., Mazzucchelli, T., & Studman, L. J. (2003b). Stepping Stones Triple P: For families with a child who has a disability. Brisbane, QLD, Australia: Triple P International. StataCorp (2005). Stata statistical software. Release 9. College Station, TX: Stata Corporation. Terman, L. M., & Merrill, M. A. (1960). Stanford–Binet intelligence scale: Manual for the third revision, Form L‐M. Boston: Houghton MiZin. Wacker, D. P., Berg, W. K., Harding, J. W., Derby, K. M., Asmus, J. M., & Healy, A. (1998). Evaluation and long‐term treatment of aberrant behaviour displayed by young children with disabilities. Journal of Developmental & Behavioral Pediatrics, 19, 260–266.

EVOLUTION OF PSYCHOPATHOLOGY

265

Wallander, J. L., Frison, S., & Rydvalova, S. (2001). Risk for psychopathology in African American adolescents with mild intellectual disability. Proceeding of the 34th Annual Gatlinburg Conference 2001. Wallander, J. L., Dekker, M. C., & Koot, H. M. (2003). Psychopathology in children and adolescents with intellectual disability: Measurement, prevalence, course, and risk. In L. M. Glidden (Ed.), International review of research in mental retardation (Vol. 26, pp. 93–134). San Diego, CA: Academic Press. Wechsler, D. (1981). Manual for the Wechsler Intelligence Scale—Revised. San Antonio, TX: The Psychological Corporation. Wilens, T. E., Biederman, J., & Spencer, T. J. (2002). Attention deficit/hyperactivity disorder across the lifespan. Annual Review of Medicine, 53, 113–131.

This page intentionally left blank

Index

A AAMR model of mental retardation, 221, 226 Abusive scores, DBC, 256, 258–260 Acts, federal privacy Family Educational Rights and Privacy Act (FERPA), 75–76 Health Insurance Portability and Accountability Act (HIPAA), 74–75 Adaptive behavior, 11–12 in diagnosis of MR, 12 Adaptive‐functional scales, for childhood disabilities, 219 ADDM. See Autism and developmental disabilities monitoring network Address geocoding, 202, 203, 207 Alpha level, 106, 112 American Association on Mental Retardation, 12 Analysis of covariance (ANCOVA). See Covariate analysis ANCOVA. See Covariate analysis Antisocial behavior in adolescence, 195 problems, 247–248 Anxiety score, DBC, 259, 261 Arcview 9.1 GIS software, 207 Area eVects and childhood emotional/behavioral disorders, 197–198 studies, 198 Attention deficit hyperactivity scale, DBC scores for, 256–257 Autism, 4, 5, 86 Autism and developmental disabilities monitoring network (ADDM), 178, 184–186 267

Autism spectrum disorders definitions for, by MADDSP, 156 sociodemographic factors and, 183

B ‘‘Back to Sleep’’ advertisement program, 7 Behavioral disorders, 197–198 Behavioral disturbance, 247–248 Biometric measures, 43 Biopsychosocial model, 218 Birth cohorts and conduct disorder, 4 use of, for epidemiology studies of MR/DD, 214–214 Birth defects and developmental disabilities, 180 Blended family, relationships in, 64–65 Buckley Amendment, 207 C CA. See Chronological age CADDRE. See Centers for Autism and Developmental Disabilities Research and Epidemiology Capacity inventory, 200 CDC. See Centers for Disease Control and Prevention Centers for Autism and Developmental Disabilities Research and Epidemiology (CADDRE), 178, 184–186

index

268 Centers for Disease Control and Prevention (CDC), 55, 85, 149 public health framework components at, 151 surveillance activity for developmental disabilities at, 150–151 Cerebral palsy definitions for by MADDS, 152–153 by MADDSP, 156–157 perinatal risk factor, 180 Child‐adolescent depression, 4 Childhood emotional/behavioral disorders area eVects and their influence on, 197–198 community resource mapping for, 200 epidemiological studies of, by GIS, 203–204 intervention planning by GIS, 204 neighborhood influences on, 196–197 Child‐oriented database, 60 Child psychiatric problems, 4 Children’s physical aggression, trajectory of, 4 Children with emotional disturbance, 199 community‐based leisure activities, 195 definition by Individuals with Disabilities Education Act, 193–194 neighborhood‐level prevention system, 194 Children with MR/DD family and health characteristics of, 236 and functional limitations access to health care for, 235 demographic characteristics of, 232 health indicators for, 234 logistic regression results of health indicators for, 238 population prevalence estimates of, 230 socioeconomic status of, 233 Chi‐square test, 106 Choropleth maps, 81 Chronological age (CA), and MA in children with MR, 13–14 Clustering, disease, 82–83 focused tests for, 83 global tests for, 82–83 local tests for, 83 Coding schemes, for data source, 41 Community‐based leisure activities, for ED prevention, 195 Community‐level factors, for ED intervention, 195, 203 Community planning, 206–207 Community resource mapping, 200

Conduct disorder, 14 and birth cohorts, 4 Confounding variables, for DD research overlap evaluation of, 99 problem with, 94–95 propensity scoring method, 97–103 traditional methods covariate analysis (ANCOVA), 96 matching, 96 random assignment, 95 variability evaluation of, 99 Contextual factors, 203 Covariate analysis of confounding variables, 96 with propensity scores, 101–102 Crash Outcome Data System, 34 Cross‐sequential designs, 33 limitations of, 33 in longitudinal studies, 32–33

D Databases computerized, 30 large‐scale, 13, 16 Latter Day Saints, 29 linkage, 28 linked. See Second‐order data linkage; Second‐order linkage database simulated multigenerational population, 72 Data definition language (DDL), 40–41 Data‐linkages. See also Second‐order data linkage steps for, 16 types of, 17 Data reduction method, for numerous outcome variables, 113 Data repository, 41 Datasets, 54. See also Second‐order linkage database administrative, 34 approaches for creating analysis, 50 birth and divorce as variables in, 45–46 creating backups in, 42 cyber and physical security measures for, 43 generation of analysis, 49–51 large, 95 manual linking of, 49

index meta data in, 40 reading and verification of, 39 restricted, 38 transfer of data from source, 41–42 Datasets, linked historical examples of, 29 recent developments in, 30–31 DBC. See Developmental behaviour checklist DBC‐based dependent variables, 251 DD. See Developmental disabilities DDL scripts, from meta data, 41 Delimited files, 41–42 Depression, in children with MR, 14 Depression score, DBC, 259, 261 Designing and/or implementing interventions, areas of, 197 Deterministic linkage approach, 57 and hierarchical set of rules, 49 objectives of, 48–49 Developmental behaviour checklist (DBC), 250 anxiety score, 259, 261 attention deficit hyperactivity score, 256–258 depression scale, items in, 250 depression score, 259, 261 disruptive subscale items abusive, 256, 258–260 kicks, 257, 260 tantrums, 257, 260 throws, 258, 260 disruptive subscale score, 253 mean, 254 regression table for, 255 self‐absorbed scale, 258, 260–261 Developmental disabilities (DD) ADDM/CADDRE surveillance network, 184–186 birth defects and, 180 CDC role for, 150–151 epidemiological studies in, 3. See also Developmental epidemiology, studies of MR/DD MADDS for. See Metropolitan Atlanta Developmental Disabilities Study MADDSFU of young adults with. See MADDS Follow‐Up Study MADDSP for. See Metropolitan Atlanta Developmental Disabilities Surveillance Program

269 public health impact of MADDS and MADDSP on, 164, 178–187 recurrence risks in, 181 research for MR and. See MR/DD, outcome research risk factors perinatal, 180 postnatal, 180–181 prenatal, 178–180 sociodemographic, 181–183 transition of children with, into young adulthood, 183–184 Developmental disabilities (DD) research, statistical issues in confounding variables problem with, 94–95 propensity scoring method, 97–103 traditional methods, 95–97 numerous outcome variables MPT for, 114–118 problem with, 111–112 traditional tools for analysis, 112–113 small sample size nonparametric tests for, 104–107 permutation tests, 107–111 problem with, 103–104 Developmental disability model, 217–218 Developmental epidemiology, 4–8 in appreciating developmental issues, 8–9 and bidirectionality of influence, 10–11 in connecting outcomes and predictors, 6–7 definition of, 4–5, 8 and developmental nature of risk factors, 9–10 discipline‐specific terms in, 35 focus on divorce, 6 in identifying nature of risk factors, 9–10 and low SES individuals, 7 and populations, 6 recent advances in, 16–19 in reducing risk across populations, 6 research, statistical issues in. See Developmental disabilities research, statistical issues in shoe leather, 34 in studying child interactions, 10 Developmental epidemiology, focus of health‐related outcomes, 5–6 on causes of, 6–7

index

270 Developmental epidemiology, focus of (cont.) on intervention, public health and public policy, 7–8 on populations, 6 Developmental epidemiology, studies of MR/DD, 213 data and measure, 226–228 of functional limitation, 228–229 measure of MR/DD, 228 dimensions of case identification, 214 method of data collection, 215–216 number of cases available for comparison, 214–215 use of birth cohorts, 214 models for biopsychosocial model, 218, 220 developmental disability model, 217–218 ICF model, 220–226 medical impairment model, 216–217 parent reports, quality of, 216 results for, 229–230 analytic models, 237–241 description of children by specific developmental disabilities, 235–237 diVerentials in prevalence, 231 family origins, 231–233 health care, 234–235 health outcomes, 233–234 population estimates of children with MR/DD, 230–231 Developmental kaleidoscope model of children’s health, 221, 227 Developmental psychopathology, definition of, 8 Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM‐IV), 8 Disease clustering, 82–83 ecology, 80 incidence, spatial distribution in, 80 mapping, in medical geography, 80–82 prevalence, spatial distribution in, 80 temporal patterns in occurrence of, 79 Disruptive subscale score, DBC items abusive, 256, 258–260

kicks, 257, 260 tantrums, 257, 260 throws, 258, 260 mean, 254 regression table for, 255 Distribution‐free tests. See Nonparametric tests, for small sample size variables Divorce eVect of, on MR/DD children, 137 focus of epidemiology on, 6 Down syndrome (DS), 95 Alzheimer’s disease in patients of, 15 deaths in United States by, 13 etiology‐based studies of, 13 group at risk of, 5 heart defects in, 19 leukemia in, 15 psychopathology in, 262 DS. See Down syndrome DSM‐like psychiatric diagnoses, vegetative signs in, 14

E Early hearing and detection intervention (EHDI) programs, 187 Early warning system, 194 Econometric models and estimation techniques, for MR/DD outcomes research, 138–140 issues in endogeneity, 142 omitted variables, 140–142 selection bias, 143–144 Economics, 121–123 costs and, relationship between, 124–125 health, 122 labor, 122 misperceptions about economics, and economists, and the importance of costs, 123–124 of MR/DD outcomes, 133–134 evidence of family eVects on MR/DD outcomes, 136–138 family economics research strategy, 134–136 MR/DD demographic and population‐based data, 136 relationship between policy and, 128–130

index

271

economics, health policy, and health outcomes, 131–132 economics of services, policy, and outcomes, 132–133 research strategy, 125–127 ED. See Children with emotional disturbance Emotional/behavioral disorders, in children area eVects and their influence on, 197–198 community resource mapping for, 200 epidemiological studies of, by GIS, 203–204 intervention planning by GIS, 204 neighborhood influences on, 196–197 Endogeneity, 142 Epidemiological studies. See also Developmental epidemiology for childhood emotional/behavioral disorders, 203–204 in developmental disabilities, 3 populations in, 5 and social ills, 7 Epilepsy, definitions for by MADDS, 153 Etiology‐based studies, 13 Expanded ID table approach, for ID generation, 60–61 interrelated IDs, 61 multiple records, 61–62 record IDs, 62 Extended families, relationships in, 65–67 E‐Z Locate Trial software, 205

F Family‐based psychosocial research, 56 Family Educational Rights and Privacy Act (FERPA), 38, 74–75, 207 Family ID, for ID generation, 59–60 Family trees, construction of, 29 Federal Education for All Handicapped Children Act (PL 94‐142), 154 Federal privacy acts Family Educational Rights and Privacy Act (FERPA), 75–76 Health Insurance Portability and Accountability Act (HIPAA), 74–75 Female behavioral disturbance, 247–248 FERPA. See Family Educational Rights and Privacy Act

First‐order data linkage, 17–18 Fragile X syndrome, psychopathology in, 263 Full‐sibling relationships, 62 Functional limitation, children with MR/DD and access to health care for, 235 demographic characteristics of, 232 health indicators for, 234 logistic regression results of health indicators for, 238 measure of, 228–229 population prevalence estimates of, 230 socioeconomic status of, 233

G Gang homicides, 194 Gantt Chart, 37 Geocoding address, 202–203, 207 in geographical information systems, 201–202 GeoDa, analysis program, 82 Geographical disparities, GIS in, 207 Geographical epidemiology, 80 Geographical information systems (GIS), 80, 200–201 child health research, 86–87 entities in, 201 functions of distance measurement, 202 geocoding, 201–202 mapping, 201 in geographical disparities, 207 for MR/DD research, 84–87 in studying emotional and behavioral disorders and challenges, 207–209 and community planning, 206–207 and epidemiological studies, 203–204 and intervention planning, 204 and resource maps, 204–206 technology, 203 Geographic coordinate system, 201 GIS. See Geographical information systems GIS‐based locational analysis models and methods, 195 Global positioning system (GPS), 201–202

index

272 Government(s) agencies, as data sources, 38 electronic datasets, 39 information management systems, 30

H Health economics, 122, 127 relation between health policy, health outcomes, and, 131–132 economists, 122 policy, relation between health economics, health outcomes, and, 131–132 Health care providers, spatial distribution of, 80 Health information systems, 30 Health Insurance Portability and Accountability Act (HIPAA), 30, 74–75 Health Insurance Portability and Privacy Act, 38 Health‐related outcomes causes of, 6–7 diagnostic criteria for, 5 focus of epidemiology on, 5–6 Healthways Inc., 31 Hearing impairment definitions for by MADDS, 153 by MADDSP, 157 hearing loss, 179 HIPAA. See Health Insurance Portability and Accountability Act

I ICD‐9 code, 13 ICF model. See International Classification of Functioning model Immediate family, relationships in, 62–63 Independent variable analysis, for numerous outcome variables, 112 Indexing, in linkage workflow, 47 Individuals with Disabilities Education Act in 1997 (IDEA, 1997), 193–194 Institutional Review Board, approval for data collection from, 38

Instrumental variables approach, 141 Intelligent quotient (IQ) tests, 11 International Classification of Functioning (ICF) model, 220–226 Intervention planning, for childhood emotional/behavioral disorders, 204 IQ‐70 criterion, 11 IQ‐only definition, in diagnosis of mental retardation, 12 Isopleth maps, 81

K Kaufman test, 11 Kicks scores, DBC, 257, 260

L Labor economics, 122, 126 Labor economists, 122 Lancaster’s definition, of neighborhood, 196 Latter Day Saints, database, 29 Leisure activities, 199 Leisure activities, for ED prevention community‐based, 195 neighborhood‐related community resource mapping, 200 leisure activities as an intervention, 198–200 Linkage creep, 72–73 Linked data, quality of, 49 Linked‐data research project, 34 linked data research for task complexity in, 36 teamwork in, 35 and primary data research, 34–35 project management in, 36 large projects for, 37–38 small projects for, 37 Linking variables probabilities of, 48 selection, for linkage work flow, 43–44 LISA. See Local indicators of spatial association Local indicators of spatial association (LISA), 81–82 Logistic regression analysis, for propensity scores computation, 98

index

273

Longitudinal studies. See also Secondary data cross‐sectional designs in, 32–33 cross‐sequential designs in, 32 and databases, 32–33 dimensions in designs of, 32 in human development, 32 Longitudinal study of psychopathology in children and adolescents MR, 247 in Australia, 248–249, 262–263 data analysis, 251–261 methods, 250–251 Low‐SES neighborhoods, 197

M MA. See Mental age MACDP. See Metropolitan Atlanta Congenital Defects Program MADDS. See Metropolitan Atlanta Developmental Disabilities Study MADDS Follow‐Up (MADDSFU) Study, of young adults, 157–158, 184 prevalence and epidemiologic studies by, 177 MADDSFU. See MADDS Follow‐Up Study, of young adult MADDSP. See Metropolitan Atlanta Developmental Disabilities Surveillance Program Male behavioral disturbance, 247–248 Mann–Whitney U test, 106, 107 Mapping, of disease, 80–82 Maps choropleth, 81 and data analysis functions, 201 isopleth, 81 resource, 204–206 Markov structure analysis, 44 Matching method, for confounding variables analysis, 96 on propensity scores, 100–101 Maternal and child health research, 56 Medical geography disease mapping in, 80–82 spatial clusters, identification of, 82–83 spatial statistics in, 84 units for spatial analysis in, 83–84 Medical impairment model, 216–217 Mental age (MA), versus CA in children, 14

Mental illness, and dual diagnosis, 14 Mental retardation (MR), 81 child outcomes in, 15 definitions for by MADDS, 153 by MADDSP, 157 diVerences in CA and MA in children with, 14 and dual diagnosis, 14 etiologies of, 12–13 information needed for developmental epidemiology of, 16 ‘‘IQ‐only definition,’’ in diagnosis of, 12 maternal age and, 183 occurrence of, in black children, 182–183 prenatal risk factors for, 179 registries for, 85–86 research for DD and. See MR/DD, research theoretical model of, 218, 220 and three‐factor definition, 11 types and problems of, 11–13 variables in study of, 15 Mental retardation and developmental disabilities. See MR/DD Meta table, attributes of, 40 Metropolitan Atlanta Congenital Defects Program (MACDP), 150 Metropolitan Atlanta Developmental Disabilities Study (MADDS), 150, 151–152 case definitions for cerebral palsy, 152–153 epilepsy, 153 hearing impairment, 153 mental retardation, 153 visual impairment (legal blindness), 153 methodology for, 153–154 prevalence and epidemiologic studies by, 165–170 prevalence estimates by, 158–160 Metropolitan Atlanta Developmental Disabilities Surveillance Program (MADDSP), 150, 154–156 case definitions for autism spectrum disorders, 156 cerebral palsy, 156–157 hearing loss, 157 mental retardation, 157 vision impairment, 157

index

274 Metropolitan Atlanta Developmental Disabilities Surveillance Program (MADDSP) (cont.) policy implications, 186–187 prevalence and epidemiologic studies by, 171–176 prevalence estimates by, 160–164 MMR vaccine, 181 Models, for developmental epidemiology studies of MR/DD analytical models, 237–241 biopsychosocial model, 218 developmental disability model, 217–218 ICF model, 220–226 medical impairment model, 216–217 Moran’s I measure, 82 Mother ID, 62, 66 Mother identifiers, 28 MPT. See Multivariate permutation tests MR. See Mental retardation MR, psychopathology study in adolescents with, 247–248 in African‐American adolescents, 248 in Australia, 248–249, 262–263 data analysis, 251–261 methods, 250–251 MR/DD, 14 advances in, 16 ability to perform first‐ and second‐order linkage, 17–18 increasing ability to link large‐scale databases, 16 mixing basic research with public health‐intervention perspectives, 18–19 developmental epidemiology studies of. See Developmental epidemiology, studies of MR/DD issues in, 11 and age versus level of functioning, 13–15 and groups with mental retardation, 15–16 and type of mental retardation and problem of caseness, 11–13 progress in, 3 research, 84–85 disease mapping for, 80–82 GIS for, 85–87 spatial analysis, units for, 83–84 spatial clustering of diseases, identification of, 82–83

spatial statistics, 84 secondary data in, 28 specific issues in, 11–16 MR/DD outcome research, eVect of economic perspectives on econometric models and estimation techniques for, 138–140 family economics research strategy, 134–136 family eVects on MR/DD outcomes, evidence of, 136–138 issues in endogeneity, 142 omitted variables, 140–142 selection bias, 143–144 MR/DD demographic and population‐based data, 136 service use on, eVect of, 133–134 Multivariate permutation tests (MPT), 114. See also Permutation tests aspects of, 114, 117 limitations, 118

N National Health Information Infrastructure (NHII), 30 National Health Interview Survey (NHIS), 226, 228, 241–242 Neighborhood influences in childhood emotional/behavioral disorders, 196–197 as a determinant of ED, 194 Neighborhood‐level eVects. See Neighborhood influences Neighborhood‐level intervention, 195 Neighborhood‐level prevention, 194 Neighborhood‐related leisure activities, for ED prevention community resource mapping, 200 recreation and leisure activities as an intervention, 198–200 NHIS. See National Health Interview Survey Nominal alpha level. See Alpha level Nonparametric tests, for small sample size variables, 104–107 North Carolina Childhood Lead Poisoning Prevention Program, 204 Numerous outcome variables, for DD research

index

275

MPT for, 114–118 multiple testing problem with, 112 problem with, 111–112 traditional methods for analysis data reduction, 113 independent variable analysis, 112 single‐step correction method, 113

O OVending, relationship between race and, 197 Omitted variables, 140–142 Overactivity, in children, 248 Override field, 73

P Parallel processing, in linkage workflow, 48 Parametric test, 105, 106, 107 Parent–child relationship, 62 Parent reports, quality of, 216 Peer rejection, 197 Peer‐reported aggression, 197 Perinatal risk factors, for DD, 180 Permutation tests, for small sample size variables, 107–109. See also Multivariate permutation tests application, 110–111 benefit of, 109 limitations to, 111 Person‐specific identifiers, data from, 38 PERT chart, 37–38 Phonetic coding, of names, 47 Pneumonia, epidemiology’s focus on, 5 Point‐sources of exposure, 82 Policy consequence of, and economics, 128–130 economics of services, outcomes, and, 132–133 Population‐based epidemiologic program. See Metropolitan Atlanta Developmental Disabilities Study Population‐based genetic research, 55 Population‐based surveillance programs, 85–86 POpulation Database Simulator (PODS), 72 Population epidemiological analyses of MR/DD. See Developmental epidemiology, studies of MR/DD

Postnatal risk factors, for DD, 180–181 Potential full‐sibling relationships, 62 Prader–Willi syndrome (PWS), 13 psychopathology in, 263 Pregnancy histories, based on mother identifiers, 28 Prenatal risk factors, for DD, 178–180 Primary data, hierarchical structure of, 34 Probabilistic linkage approach. See also Family‐based psychosocial research objectives of, 48–49 Probabilistic matching protocol, 71 Project data repository, 41 Project log, 37 Project management plan, in linkage projects, 36 Project timeline, representation of, 37 Propensity scores application of, 102 computation of, 98–99 logistic regression analysis for, 98 covariate analysis with, 101–102 definition, 97 distributions, 100 limitations to, 102–103 on matching method, 100–101 visual inspection, 99–100 Propensity scoring method, for confounding variables. See Propensity scores Psychopathology, developmental, definition of, 8 Psychopathology, in adolescents with MR, study of, 247–248 in African‐American adolescents, 248 in Australia, 248–249, 262–263 data analysis, 251–261 methods, 250–251 in Down syndrome, 262 externalizing psychopathology, 262 DBC abusive scores, 258–260 DBC attention deficit hyperactivity scale, 256–257 DBC disruptive subscale score, 253–256 in Fragile X syndrome, 263 internalizing psychopathology, 262 DBC anxiety subscale score, 259, 261 DBC depression scale score, 259, 261 DBC self‐absorbed behavior subscale score, 260–261 in Prader–Willi syndrome, 263

index

276 Psychopathology, in adolescents with MR, study of (cont.) in Williams syndrome, 263 Psychosocial research, 56 Public health‐intervention perspectives, 18–19 Public health surveillance, monitoring by, 33–34

R Race, relationship between oVending and, in neighborhoods, 197 Random assignment method, for confounding variables analysis, 95 Record linkage(s), 28 communication challenges as barrier in, 36 electronic, 31 interrelated tasks in, 36 methodologies, 28 projects, project management plan in, 36 teams, 28 variables, 42, 43 names and addresses as, 44 Record linkage, workflow of data acquisition in, 38 getting approval for, 38–39 data storage in creating code tables for, 41 creating support files for, 39–41 data checking and cleaning for, 42 eYcient storage of data in, 41 reading data for, 39 securing data for, 42–43 transferring data from source datasets for, 41–42 examining record pairs for blocking in, 44, 47 comparisons in, 47 indexing in, 47 parallel processing in, 48 generation of analysis datasets for, 49–51 matching records in deterministic strategy for, 48–49 probabilistic strategy for, 48 selection of strategy for, 49 selecting linking variables for, 43–44 Recreational activities, for ED intervention, 198–200 Recreational sports, 198–199

Recreation resources, 208 Regenstrief Institute, in Indianapolis, 30 Regional Health Information Organizations (RHIO), 30 Relationship(s) in blended family, 64–65 in extended families, 65–67 in immediate family, 62–63 between race and oVending, 197 Resource maps, 204–206 Risk behaviors, of individual, 6 Risk factors, developmental nature of, 9–10 R statistical software, 111 Russell Soundex, 47

S Sample size 30, 103 San Diego Medical Information Network Exchange, 30 SAS statistical software, 118 SaTScan (spatial scan statistic) procedure, 83 Schizophrenic children, 194 School‐identified ED, prevalence rates, 194 Secondary data. See also Longitudinal studies bias against, 28 combination of primary and, 34 Second‐order data linkage, 17–18. See also Second‐order linkage database Second‐order integrated developmental database. See Second‐order linked database Second‐order linkage database applications of maternal and child health research, 56 population‐based genetic research, 55 psychosocial research, 56 benefits of, 53 enhanced tracking capacity for public health oYcials, 54–55 families and communities identification for focused research, 55–56 challenges in computational demands, 70–71 concerns regarding privacy and confidentiality, 74–76 diYculty in evaluation, 71–72

index linkage creep, 72–73 creation of. See Second‐order linkage database, creation of database structure, 67–70 techniques in, 56 deterministic linkage in, 57 probabilistic linkage in, 57–58 Second‐order linkage database, creation of, 58 cascading algorithms for establishing families within blended family, 64–65 within extended family, 65–67 within immediate family, 62 ID numbers generation, approaches for expanded ID table, 60–62 family ID, 59–60 Selection bias, 143–144 SES. See Socioeconomic status Shapefile, in GIS, 201 Sibling relationships, 64 Sign test, 106 Single‐step correction method, for numerous outcome variables, 113 Small sample sizes, for DD research nonparametric tests for, 104–107 permutation tests, 107–111 problem with, 103–104 Snow, John, 29 Social development model, 195 Social environments, negative and positive, 198 Social ills and epidemiological studies, 7 Social risk factors, 6 Social security number, 17, 43 Sociodemographic risk factors, for DD, 181–183 Space‐time intelligent systems, 87 Spatial aggregation, 83 Spatial autocorrelation, 82 Spatial clustering, of MR/DD, 82–83 units of, 83–84 Spatial distribution, of MR/DD, 80, 82–83 Spatial epidemiology, 80 Spatial statistics, in medical geography, 84 Spatial units, 83–84 SPSS, 16, 50 SPSS Exact Tests, 111 Standard coding systems, for healthcare providers, 30 Standard query language (SQL), 40 Stanford–Binet test, 11

277 Statistical clustering procedures, 194 StatXact software package, 111 Sudden infant death syndrome (SIDS), 7 Summer camp, for children with ED, 199 Synergy, advantage of, 35

T Tantrums scores, DBC, 257, 260 TBPS. See Total behavior problem score Temporal GIS. See Space‐time intelligent systems TennCare, 31 Tests, for spatial distribution of disease or health outcomes global, 82–83 local, 83 Texas Birth Defects Registry, 216 30, sample size, 103 Throws scores, DBC, 258, 260 TIGER database. See Topologically Integrated Geographic Encoding and Referencing database Topologically Integrated Geographic Encoding and Referencing (TIGER) database, 202, 207 Total behavior problem score (TBPS), 250 Type I error control by MPT, 117–118 of parametric tests, 106 Type II error, of parametric tests, 106

U United States, initiative for electronic health records in, 30–31 Units, for spatial analysis, 83–84 U.S. Safety Council, 7 US children, population prevalence estimates of MR/DD and functional limitations among, 230–231 Utah Population Database Project, 29, 55

V Vineland Adaptive Behavior Scales, domains of, 12

index

278 Visual impairment (legal blindness) definitions for by MADDS, 153 by MADDSP, 157 prenatal risk factors for, 179

W Wechsler test, 11

Williams syndrome (WS), 19 psychopathology in, 263 WISC‐III test, 251 WS. See Williams syndrome

Z Zip Codes, as spatial unit, 83–84

Contents of Previous Volumes

Volume 1

Volume 2

A Functional Analysis of Retarded Development SIDNEY W. BIJOU

A Theoretical Analysis and Its Application to Training the Mentally Retarded M. RAY DENNY

Classical Conditioning and Discrimination Learning Research with the Mentally Retarded LEONARD E. ROSS

The Role of Input Organization in the Learning and Memory of Mental Retardates HERMAN H. SPITZ Autonomic Nervous System Functions and Behavior: A Review of Experimental Studies with Mental Defectives RATHE KARPER

The Structure of Intellect in the Mental Retardate HARVEY F. DINGMAN AND C. EDWARD MEYERS Research on Personality Structure in the Retardate EDWARD ZIGLER

Learning and Transfer of Mediating Responses in Discriminating Learning BRYAN E. SHEPP AND FRANK D. TURRISI

Experience and the Development of Adaptive Behavior H. CARL HAYWOOD AND JACK T. TAPP

A Review of Research on Learning Sets and Transfer or Training in Mental Defectives MELVIN E. KAUFMAN AND HERBERT J. PREHM

A Research Program on the Psychological Effects of Brain Lesions in Human Beings RALPH M. REITAN

Programming Perception and Learning for Retarded Children MURRAY SIDMAN AND LAWRENCE T. STODDARD

Long-Term Memory in Mental Retardation JOHN M. BELMONT

Programming Instruction Techniques for the Mentally Retarded FRANCES M. GREENE

The Behavior of Moderately and Severely Retarded Persons JOSEPH E. SPRADLIN AND FREDERIC L. GIRARDEAU

Some Aspects of the Research on Mental Retardation in Norway IVAR ARNIJOT BJORGEN

Author Index-Subject Index

279

280

contents of previous volumes

Research on Mental Deficiency During the Last Decade in France R. LAFON AND J. CHABANIER

A Theory of Primary and Secondary Familial Mental Retardation ARTHUR R. JENSEN

Psychotherapeutic Procedures with the Retarded MANNY STERNLIGHT

Inhibition Deficits in Retardate Learning and Attention LAIRD W. HEAL AND JOHN T. JOHNSON, JR.

Author Index-Subject Index

Volume 3 Incentive Motivation in the Mental Retardate PAUL S. SIEGEL Development of Lateral and Choice-Sequence Preferences IRMA R. GERJUOY AND JOHN J. WINTERS, JR. Studies in the Experimental Development of Left-Right Concepts in Retarded Children Using Fading Techniques SIDNEY W. BIJOU Verbal Learning and Memory Research with Retardates: An Attempt to Assess Developmental Trends L. R. GOULET Research and Theory in Short-Term Memory KEITH G. SCOTT AND MARCIA STRONG SCOTT

Growth and Decline of Retardate Intelligence MARY ANN FISHER AND DAVID ZEAMAN The Measurements of Intelligence A. B. SILVERSTEIN Social Psychology and Mental Retardation WARNER WILSON Mental Retardation in Animals GILBERT W. MEIER Audiologic Aspects of Mental Retardation LYLE L. LLOYD Author Index-Subject Index

Volume 5 Medical-Behavioral Research in Retardation JOHN M. BELMONT Recognition Memory: A Research Strategy and a Summary of Initial Findings KEITH G. SCOTT

Reaction Time and Mental Retardation ALFRED A. BAUMEISTER AND GEORGE KELLAS

Operant Procedures with the Retardate: An Overview of Laboratory Research PAUL WEISBERG

Mental Retardation in India: A Review of Care, Training, Research, and Rehabilitation Programs J. P. DAS

Methodology of Psychopharmacological Studies with the Retarded ROBERT L. SPRAGUE AND JOHN S. WERRY

Educational Research in Mental Retardation SAMUEL L. GUSKIN AND HOWARD H. SPICKER

Process Variables in the Paired-Associate Learning of Retardates ALFRED A. BAUMEISTER AND GEORGE KELLAS

Author Index-Subject Index

Volume 4

Sequential Dot Presentation Measures of Stimulus Trace in Retardates and Normals EDWARD A. HOLDEN, JR.

Memory Processes in Retardates and Normals NORMAN R. ELLIS

Cultural-Familial Retardation FREDERIC L. GIRARDEAU

contents of previous volumes

281

German Theory and Research on Mental Retardation: Emphasis on Structure LOTHAR R. SCHMIDT AND PAUL B. BALTES

Placement of the Retarded in the Community: Prognosis and Outcome RONALD B. MCCARVER AND ELLIS M. CRAIG

Author Index-Subject Index

Physical and Motor Development of Retarded Persons ROBERT H. BRUININKS

Volume 6 Cultural Deprivation and Cognitive Competence J. P. DAS Stereotyped Acts ALFRED A. BAUMEISTER AND REX FOREHAND Research on the Vocational Habilitation of the Retarded: The Present, the Future MARC W. GOLD Consolidating Facts into the Schematized Learning and Memory System of Educable Retardates HERMAN H. SPITZ An Attentional-Retention Theory of Retardate Discrimination Learning MARY ANN FISHER AND DAVID ZEAMAN Studying the Relationship of Task Performance to the Variables of Chronological Age, Mental Age, and IQ WILLIAM E. KAPPAUF Author Index-Subject Index Volume 7 Mediational Processes in the Retarded JOHN G. BORKOWSKI AND PATRICIA B. WANSCHURA The Role of Strategic Behavior in Retardate Memory ANN L. BROWN Conservation Research with the Mentally Retarded KERI M. WILTON AND FREDERIC J. BOERSMA

Subject Index

Volume 8 Self-Injurious Behavior ALFRED A. BAUMEISTER AND JOHN PAUL ROLLINGS Toward a Relative Psychology of Mental Retardation with Special Emphasis on Evolution HERMAN H. SPITZ The Role of the Social Agent in Language Acquisition: Implications for Language Intervention GERALD J. MAHONEY AND PAMELA B. SEELY Cognitive Theory and Mental Development EARL C. BUTTERFIELD AND DONALD J. DICKERSON A Decade of Experimental Research in Mental Retardation in India ARUN K. SEN The Conditioning of Skeletal and Autonomic Responses: Normal-Retardate Stimulus Trace Differences SUSAN M. ROSS AND LEONARD E. ROSS Malnutrition and Cognitive Functioning J. P. DAS AND EMMA PIVATO Research on Efficacy of Special Education for the Mentally Retarded MELVINE E. KAUFMAN AND PAUL A. ALBERTO Subject Index

282 Volume 9 The Processing of Information from Short-Term Visual Store: Developmental and Intellectual Differences LEONARD E. ROSS AND THOMAS B. WARD Information Processing in Mentally Retarded Individuals KEITH E. STANOVICH Mediational Process in the Retarded: Implications for Teaching Reading CLESSEN J. MARTIN Psychophysiology in Mental Retardation J. CLAUSEN Theoretical and Empirical Strategies for the Study of the Labeling of Mentally Retarded Persons SAMUEL L. GUSKIN The Biological Basis of an Ethic in Mental Retardation ROBERT L. ISAACSON AND CAROL VAN HARTESVELDT Public Residential Services for the Mentally Retarded R. C. SCHEERENBERGER Research on Community Residential Alternatives for the Mentally Retarded LAIRD W. HEAL, CAROL K. SIGELMAN, AND HARVEY N. SWITZKY Mainstreaming Mentally Retarded Children: Review of Research LOUIS CORMAN AND JAY GOTTLIEB Savants: Mentally Retarded Individuals with Special Skills A. LEWIS HILL

contents of previous volumes Visual Pattern Detection and Recognition Memory in Children with Profound Mental Retardation PATRICIA ANN SHEPHERD AND JOSEPH F. FAGAN III Studies of Mild Mental Retardation and Timed Performance T. NETTELBECK AND N. BREWER Motor Function in Down’s Syndrome FERIHA ANWAR Rumination NIRBHAY N. SINGH Subject Index

Volume 11 Cognitive Development of the Learning-Disabled Child JOHN W. HAGEN, CRAIG R. BARCLAY, AND BETTINA SCHWETHELM Individual Differences in Short-Term Memory RONALD L. COHEN Inhibition and Individual Differences in Inhibitory Processes in Retarded Children PETER L. C. EVANS Stereotyped Mannerisms in Mentally Retarded Persons: Animal Models and Theoretical Analyses MARK H. LEWIS AND ALFRED A. BAUMEISTER An Investigation of Automated Methods for Teaching Severely Retarded Individuals LAWRENCE T. STODDARD

Volume 10

Social Reinforcement of the Work Behavior of Retarded and Nonretarded Persons LEONIA K. WATERS

The Visual Scanning and Fixation Behavior of the Retarded LEONARD E. ROSS AND SUSAM M. ROSS

Social Competence and Interpersonal Relations between Retarded and Nonretarded Children ANGELA R. TAYLOR

Subject Index

contents of previous volumes The Functional Analysis of Imitation WILLIAM R. MCCULLER AND CHARLES L. SALZBERG Index

283 Autonomy and Adaptability in Work Behavior of Retarded Clients JOHN L. GIFFORD, FRANK R. RUSCH, JAMES E. MARTIN, AND DAVID J. WHITE Index

Volume 12 An Overview of the Social Policy of Deinstitutionalization BARRY WILLER AND JAMES INTAGLIATA Community Attitudes toward Community Placement of Mentally Retarded Persons CYNTHIA OKOLO AND SAMUEL GUSKIN Family Attitudes toward Deinstitutionalization AYSHA LATIB, JAMES CONROY, AND CARLA M. HESS Community Placement and Adjustment of Deinstitutionalized Clients: Issues and Findings ELLIS M. CRAIG AND RONALD B. MCCARVER

Volume 13 Sustained Attention in the Mentally Retarded: The Vigilance Paradigm JOEL B. WARM AND DANIEL B. BERCH Communication and Cues in the Functional Cognition of the Mentally Retarded JAMES E. TURNURE Metamemory: An Aspect of Metacognition in the Mentally Retarded ELAINE M. JUSTICE Inspection Time and Mild Mental Retardation T. NETTELBECK

Issues in Adjustment of Mentally Retarded Individuals to Residential Relocation TAMAR HELLER

Mild Mental Retardation and Memory Scanning C. J. PHILLIPS AND T. NETTELBECK

Salient Dimensions of Home Environment Relevant to Child Development KAZUO NIHIRA, IRIS TAN MINK, AND C. EDWARD MEYERS

Cognitive Determinants of Reading in Mentally Retarded Individuals KEITH E. STANOVICH

Current Trends and Changes in Institutions for the Mentally Retarded R. K. EYMAN, S. A. BORTHWICK, AND G. TARJAN Methodological Considerations in Research on Residential Alternatives for Developmentally Disabled Persons LAIRD W. HEAL AND GLENN T. FUJIURRA A Systems Theory Approach to Deinstitutionalization Policies and Research ANGELA A. NOVAK AND TERRY R. BERKELEY

Comprehension and Mental Retardation LINDA HICKSON BILSKY Semantic Processing, Semantic Memory, and Recall LARAINE MASTERS GLIDDEN Proactive Inhibition in Retarded Persons: Some Clues to Short-Term Memory Processing JOHN J. WINTERS, JR. A Triarchic Theory of Mental Retardation ROBERT J. STERNBERG AND LOUIS C. SPEAR Index

284

contents of previous volumes

Volume 14

Volume 15

Intrinsic Motivation and Behavior Effectiveness in Retarded Persons H. CARL HAYWOOD AND HARVEY N. SWITZKY

Mental Retardation as Thinking Disorder: The Rationalist Alternative to Empiricism HERMAN H. SPITZ

The Rehearsal Deficit Hypothesis NORMAN W. BRAY AND LISA A. TURNER Molar Variability and the Mentally Retarded STUART A. SMITH AND PAUL S. SIEGEL Computer-Assisted Instruction for the Mentally Retarded FRANCES A CONNERS, DAVID R. CARUSO, AND DOUGLAS K. DETTERMAN

Developmental Impact of Nutrition on Pregnancy, Infancy, and Childhood: Public Health Issues in the United States ERNESTO POLLITT The Cognitive Approach to Motivation in Retarded Individuals SHYLAMITH KREITLER AND HANS KREITLER Mental Retardation, Analogical Reasoning, and the Componential Method J. MCCONAGHY

Procedures and Parameters of Errorless Discrimination Training with Developmentally Impaired Individuals GIULO E. LANCIONI AND PAUL M. SMEETS

Application of Self-Control Strategies to Facilitate Independence in Vocational and Instructional Settings JAMES E. MARTIN, DONALD L. BURGER, SUSAN ELIAS-BURGER, AND DENNIS E. MITHAUG

Reading Acquisition and Remediation in the Mentally Retarded NIRBHAY N. SINGH AND JUDY SINGH

Family Stress Associated with a Developmentally Handicapped Child PATRICIA M. MINNES

Families with a Mentally Retarded Child BERNARD FARBER AND LOUIS ROWITZ

Physical Fitness of Mentally Retarded Individuals E. KATHRYN MCCONAUGHY AND CHARLES L. SALZBERG

Social Competence and Employment of Retarded Persons CHARLES L. SALZBERG, MARILYN LIKINS, E. KATHRYN MCCONAUGHY, AND BENJAMIN LINGUGARIS/KRAFT Toward a Taxonomy of Home Environments SHARON LANDESMAN Behavioral Treatment of the Sexually Deviant Behavior of Mentally Retarded Individuals R. M. FOXX, R. G. BITTLE, D. R. BECHTEL, AND J. R. LIVESAY Behavior Approaches to Toilet Training for Retarded Persons S. BETTISON Index

Index

Volume 16 Methodological Issues in Specifying Neurotoxic Risk Factors for Developmental Delay: Lead and Cadmium as Prototypes STEPHEN R. SCHROEDER The Role of Methylmercury Toxicity in Mental Retardation GARY J. MYERS AND DAVID O. MARSH Attentional Resource Allocation and Mental Retardation EDWARD C. MERRILL

contents of previous volumes Individual Differences in Cognitive and Social Problem-Solving Skills as a Function of Intelligence ELIZABETH J. SHORT AND STEVEN W. EVANS Social Intelligence, Social Competence, and Interpersonal Competence JANE L. MATHIAS Conceptual Relationships between Family Research and Mental Retardation ZOLINDA STONEMAN Index Volume 17 The Structure and Development of Adaptive Behaviors KEITH F. WIDAMAN, SHARON A. BORTHWICK-DUFFY, AND TODD D. LITTLE Perspectives on Early Language from Typical Development and Down Syndrome MICHAEL P. LYNCH AND REBECCA E. EILERS The Development of Verbal Communication in Persons with Moderate to Mild Mental Retardation LEONARD ABBEDUTO Assessment and Evaluation of Exceptional Children in the Soviet Union MICHAEL M. GERBER, VALERY PERELMAN, AND NORMA LOPEZ-REYNA Constraints on the Problem Solving of Persons with Mental Retardation RALPH P. FERRETTI AND AL R. CAVALIER Long-Term Memory and Mental Retardation JAMES E. TURNURE Index Volume 18 Perceptual Deficits in Mildly Mentally Retarded Adults ROBERT FOX AND STEPHEN OROSS, III

285 Stimulus Organization and Relational Learning SAL A. SORACI, JR. AND MICHAEL T. CARLIN Stimulus Control Analysis and Nonverbal Instructional Methods for People with Intellectual Disabilities WILLIAM J. MCILVANE Sustained Attention in Mentally Retarded Individuals PHILLIP D. TOMPOROWSKI AND LISA D. HAGER How Modifiable Is the Human Life Path? ANN M. CLARKE AND ALAN D. B. CLARKE Unraveling the ‘‘New Morbidity’’: Adolescent Parenting and Developmental Delays JOHN G. BORKOWSKI, THOMAS L. WHITMAN, ANNE WURTZ PASSINO, ELIZABETH A. RELLINGER, KRISTEN SOMMER, DEBORAH KEOUGH, AND KERI WEED Longitudinal Research in Down Syndrome JANET CARR Staff Training and Management for Intellectual Disability Services CHRIS CULLEN Quality of Life of People with Developmental Disabilities TREVOR R. PARMENTER Index

Volume 19 Mental Retardation in African Countries: Conceptualization, Services, and Research ROBERT SERPELL, LILIAN MARIGA, AND KARYN HARVEY Aging and Alzheimer Disease in People with Mental Retardation WARREN B. ZIGMAN, NICOLE SCHUPF, APRIL ZIGMAN, AND WAYNE SILERMAN

286 Characteristics of Older People with Intellectual Disabilities in England JAMES HOGG AND STEVE MOSS Epidemiological Thinking in Mental Retardation: Issues in Taxonomy and Population Frequency TOM FRYERS Use of Data Base Linkage Methodology in Epidemiological Studies of Mental Retardation CAROL A. BOUSSY AND KEITH G. SCOTT Ways of Analyzing the Spontaneous Speech of Children with Mental Retardation: The Value of Cross-Domain Analyses CATHERINE E. SNOW AND BARBARA ALEXANDER PAN Behavioral Experimentation in Field Settings: Threats to Validity and Interpretation Problems WILLY-TORE MRCH Index

Volume 20 Parenting Children with Mental Retardation BRUCE L. BAKER, JAN BLACHER, CLAIRE B. KOPP, AND BONNIE KRAEMER Family Interactions and Family Adaptation FRANK J. FLOYD AND CATHERINE L. COSTIGAN Studying Culturally Diverse Families of Children with Mental Retardation IRIS TAN MINK Older Adults with Mental Retardation and Their Families TAMAR HELLER A Review of Psychiatric and Family Research in Mental Retardation ANN GATH

contents of previous volumes A Cognitive Portrait of Grade School Students with Mild Mental Retardation MARCIA STRONG SCOTT, RUTH PEROU, ANGELIKA HARTL CLAUSSEN, AND LOIS-LYNN STOYKO DEUEL Employment and Mental Retardation NEIL KIRBY Index

Volume 21 An Outsider Looks at Mental Retardation: A Moral, a Model, and a Metaprincipal RICHARD P. HONECK Understanding Aggression in People with Intellectual Disabilities: Lessons from Other Populations GLYNIS MURPHY A Review of Self-Injurious Behavior and Pain in Persons with Developmental Disabilities FRANK J. SYMONS AND TRAVIS THOMPSON Recent Studies in Psychopharmacology in Mental Retardation MICHAEL G. AMAN Methodological Issues in the Study of Drug Effects on Cognitive Skills in Mental Retardation DEAN C. WILLIAMS AND KATHRYN J. SAUNDERS The Behavior and Neurochemistry of the Methylazoxymethanol-Induced Microencephalic Rat PIPPA S. LOUPE, STEPHEN R. SCHROEDER, AND RICHARD E.TESSEL Longitudinal Assessment of Cognitive-Behavioral Deficits Produced by the Fragile-X Syndrome GENE S. FISCH Index

contents of previous volumes Volume 22 Direct Effects of Genetic Mental Retardation Syndromes: Maladaptive Behavior and Psychopathology ELISABETH M. DYKENS Indirect Effects of Genetic Mental Retardation Disorders: Theoretical and Methodological Issues ROBERT M. HODAPP The Development of Basic Counting, Number, and Arithmetic Knowledge among Children Classified as Mentally Handicapped ARTHUR J. BAROODY The Nature and Long-Term Implications of Early Developmental Delays: A Summary of Evidence from Two Longitudinal Studies RONALD GALLIMORE, BARBARA K. KEOGH, AND LUCINDA P. BERNHEIMER Savant Syndrome TED NETTELBECK AND ROBYN YOUNG The Cost-Efficiency of Supported Employment Programs: A Review of the Literature ROBERT E. CIMERA AND FRANK R. RUSCH Decision Making and Mental Retardation LINDA HICKSON AND ISHITA KHEMKA ‘‘The Child That Was Meant?’’ or ‘‘Punishment for Sin?’’: Religion, Ethnicity, and Families with Children with Disabilities LARAINE MASTERS GLIDDEN, JEANNETTE ROGERS-DULAN, AND AMY E. HILL Index Volume 23 Diagnosis of Autism before the Age of 3 SALLY J. ROGERS The Role of Secretin in Autistic Spectrum Disorders AROLY HORVATH AND J. TYSON TILDON

287 The Role of Candidate Genes in Unraveling the Genetics of Autism CHRISTOPHER J. STODGELL, JENNIFER L. INGRAM, AND SUSAN L. HYMAN Asperger’s Disorder and Higher Functioning Autism: Same or Different? FRED R. VOLKMAR AND AMI KLIN The Cognitive and Neural Basis of Autism: A Disorder of Complex Information Processing and Dysfunction of Neocortical Systems NANCY J. MINSHEW, CYNTHIA JOHNSON, AND BEATRIZ LUNA Neural Plasticity, Joint Attention, and a Transactional Social-Orienting Model of Autism PETER MUNDY AND A. REBECCA NEAL Theory of Mind and Autism: A Review SIMON BARON-COHEN Understanding the Language and Communicative Impairments in Autism HELEN TAGER-FLUSBERG Early Intervention in Autism: Joint Attention and Symbolic Play CONNIE KASARI, STEPHANNY F. N. FREEMAN, AND TANYA PAPARELLA Attachment and Emotional Responsiveness in Children with Autism CHERYL DISSANAYAKE AND MARIAN SIGMAN Families of Adolescents and Adults with Autism: Uncharted Territory MARSHA MAILICK SELTZER, MARTY WYNGAARDEN KRAUSS, GAEL I. ORSMOND, AND CARRIE VESTAL Index

Volume 24 Self-Determination and Mental Retardation MICHAEL L. WEHMEYER

288 International Quality of Life: Current Conceptual, Measurement, and Implementation Issues KENNETH D. KEITH Measuring Quality of Life and Quality of Services through Personal Outcome Measures: Implications for Public Policy JAMES GARDNER, DEBORAH T. CARRAN, AND SYLVIA NUDLER Credulity and Gullibility in People with Developmental Disorders: A Framework for Future Research STEPHEN GREENSPAN, GAIL LOUGHLIN, AND RHONDA S. BLACK Criminal Victimization of Persons with Mental Retardation: The Influence of Interpersonal Competence on Risk T. NETTELBECK AND C. WILSON The Parent with Mental Retardation STEVE HOLBURN, TIFFANY PERKINS, AND PETER VIETZE Psychiatric Disorders in Adults with Mental Retardation STEVE MOSS Development and Evaluation of Innovative Residential Services for People with Severe Intellectual Disability and Serious Challenging Behavior JIM MANSELL, PETER MCGILL, AND ERIC EMERSON The Mysterious Myth of Attention Deficits and Other Defect Stories: Contemporary Issues in the Developmental Approach to Mental Retardation JACOB A. BURACK, DAVID W. EVANS, CHERYL KLAIMAN, AND GRACE IAROCCI Guiding Visual Attention in Individuals with Mental Retardation RICHARD W. SERNA AND MICHAEL T. CARLIN Index

contents of previous volumes Volume 25 Characterizations of the Competence of Parents of Young Children with Disabilities CARL J. DUNST, TRACY HUMPHRIES, AND CAROL M. TRIVETTE Parent–Child Interactions When Young Children Have Disabilities DONNA SPIKER, GLENNA C. BOYCE, AND LISA K. BOYCE The Early Child Care Study of Children with Special Needs JEAN F. KELLY AND CATHRYN L. BOOTH Diagnosis of Autistic Disorder: Problems and New Directions ROBYN YOUNG AND NEIL BREWER Social Cognition: A Key to Understanding Adaptive Behavior in Individuals with Mild Mental Retardation JAMES S. LEFFERT AND GARY N. SIPERSTEIN Proxy Responding for Subjective Well-Being: A Review ROBERT A. CUMMINS People with Intellectual Disabilities from Ethnic Minority Communities in the United States and the United Kingdom CHRIS HATTON Perception and Action in Mental Retardation W. A. SPARROW AND ROSS H. DAY Volume 26 A History of Psychological Theory and Research in Mental Retardation since World War II DONALD K. ROUTH AND STEPHEN R. SCHROEDER Psychopathology and Intellectual Disability: The Australian Child to Adult Longitudinal Study BRUCE J. TONGE AND STEWART L. EINFELD

contents of previous volumes Psychopathology in Children and Adolescents with Intellectual Disability: Measurement, Prevalence, Course, and Risk JAN L. WALLANDER, MARIELLE C. DEKKER, AND HANS KOOT Resilience, Family Care, and People with Intellectual Disabilities GORDONGRANT, PAULRAMCHARAN, AND PETER GOWARD Prevalence and Correlates of Psychotropic Medication Use among Adults with Developmental Disabilities: 1970–2000 MARIA G. VALDOVINOS, STEPHEN R. SCHROEDER, AND GEUNYOUNG KIM Integration as Acculturation: Developmental Disability, Deinstitutionalization, and Service Delivery Implications M. KATHERINE BUELL Cognitive Aging and Down Syndrome: An Interpretation J. P. DAS Index

289 CARMICHAEL OLSON, AND GERALYN R. TIMLER Memory, Language Comprehension, and Mental Retardation EDWARD C. MERRILL, REGAN LOOKADOO, AND STACY RILEA Reading Skills and Cognitive Abilities of Individuals with Mental Retardation FRANCES A. CONNERS Language Interventions for Children with Mental Retardation NANCY C. BRADY AND STEVEN F. WARREN Augmentative and Alternative Communication for Persons with Mental Retardation MARYANN ROMSKI, ROSE A. SEVCIK, AND AMY HYATT FONSECA Atypical Language Development in Individuals with Mental Retardation: Theoretical Implications JEAN A. RONDAL Index

Volume 27

Volume 28

Language and Communication in Individuals with Down Syndrome ROBIN S. CHAPMAN

Promoting Intrinsic Motivation and Self-Determination in People with Mental Retardation EDWARD L. DECI

Language Abilities of Individuals with Williams Syndrome CAROLYN B. MERVIS, BYRON F. ROBINSON, MELISSA L. ROWE, ANGELA M. BECERRA, AND BONITA P. KLEIN-TASMAN Language and Communication in Fragile X Syndrome MELISSA M. MURPHY AND LEONARD ABBEDUTO On Becoming Socially Competent Communicators: The Challenge for Children with Fetal Alcohol Exposure TRUMAN E. COGGINS, LESLEY B. OLSWANG, HEATHER

Applications of a Model of Goal Orientation and Self-Regulated Learning to Individuals with Learning Problems PAUL R. PINTRICH AND JULIANE L. BLAZEVSKI Learner-Centered Principles and Practices: Enhancing Motivation and Achievement for Children with Learning Challenges and Disabilities BARBARA L. MCCOMBS Why Pinocchio Was Victimized: Factors Contributing to Social Failure in People with Mental Retardation STEPHEN GREENSPAN

290 Understanding the Development of Subnormal Performance in Children from a Motivational-Interactionist Perspective JANNE LEPOLA, PEKKA SALONEN, MARJA VAURAS, AND ELISA POSKIPARTA Toward Inclusion Across Disciplines: Understanding Motivation of Exceptional Students HELEN PATRICK, ALLISON M. RYAN, ERIC M. ANDERMAN, AND JOHN KOVACH Loneliness and Developmental Disabilities: Cognitive and Affective Processing Perspectives MALKA MARGALIT The Motivation to Maintain Subjective Well-Being: A Homeostatic Model ROBERT A. CUMMINS AND ANNA L. D. LAU Quality of Life from a Motivational Perspective ROBERT L. SCHALOCK Index Volume 29 Behavioral Phenotypes: Going Beyond the Two-Group Approach ROBERT M. HODAPP Prenatal Drug Exposure and Mental Retardation ROBERT E. ARENDT, JULIA S. NOLAND, ELIZABETH J. SHORT, AND LYNN T. SINGER Spina Bifida: Genes, Brain, and Development JACK M. FLETCHER, MAUREEN DENNIS, HOPE NORTHRUP, MARCIA A. BARNES, H. JULIA HANNAY, SUSAN H. LANDRY, KIM COPELAND, SUSAN E. BLASER, LARRY A. KRAMER, MICHAEL E. BRANDT, AND DAVID J. FRANCIS The Role of the Basal Ganglia in the Expression of Stereotyped, Self-Injurious Behaviors in Developmental Disorders HOWARD C. CROMWELL AND BRYAN H. KING

contents of previous volumes Risk Factors for Alzheimer’s Disease in Down Syndrome LYNN WARD Precursors of Mild Mental Retardation in Children with Adolescent Mothers JOHN G. BORKOWSKI, JULIE J. LOUNDS, CHRISTINE WILLARD NORIA, JENNIFER BURKE LEFEVER, KERI WEED, DEBORAH A. KEOGH, AND THOMAS L. WHITMAN The Ecological Context of Challenging Behavior in Young Children with Developmental Disabilities ANITA A. SCARBOROUGH AND KENNETH K. POON Employment and Intellectual Disability: Achieving Successful Employment Outcomes KAYE SMITH, LYNNE WEBBER, JOSEPH GRAFFAM, AND CARLENE WILSON Technology Use and People with Mental Retardation MICHAEL L. WEHMEYER, SEAN J. SMITH, SUSAN B. PALMER, DANIEL K. DAVIES, AND STEVEN E. STOCK Index

Volume 30 Neurodevelopmental Effects of Alcohol THOMAS M. BURBACHER AND KIMBERLY S. GRANT PCBs and Dioxins HESTIEN J. I. VREUGDENHIL AND NYNKE WEISGLAS-KUPERUS Interactions of Lead Exposure and Stress: Implications for Cognitive Dysfunction DEBORAH A. CORY-SLECHTA

contents of previous volumes Developmental Disabilities Following Prenatal Exposure to Methyl Mercury from Maternal Fish Consumption: A Review of the Evidence GARY J. MYERS, PHILIP W. DAVIDSON, AND CONRAD F. SHAMLAYE Environmental Agents and Autism: Once and Future Associations SUSAN L. HYMAN, TARA L. ARNDT, AND PATRICIA M. RODIER Endocrine Disruptors as a Factor in Mental Retardation BERNARD WEISS The Neurotoxic Properties of Pesticides HERBERT L. NEEDLEMAN Parental Smoking and Children’s Behavioral and Cognitive Functioning MICHAEL WEITZMAN, MEGAN KAVANAUGH, AND TODD A. FLORIN Neurobehavioral Assessment in Studies of Exposures to Neurotoxicants DAVID C. BELLINGER From Animals to Humans: Models and Constructs DEBORAH C. RICE

291 Individual Differences in Interpersonal Relationships for Persons with Mental Retardation YONA LUNSKY Understanding Low Achievement and Depression in Children with Learning Disabilities: A Goal Orientation Approach GEORGIOS D. SIDERIDIS Motivation and Etiology-Specific Cognitive–Linguistic Profiles DEBORAH J. FIDLER The Role of Motivation and Psychopathology in Understanding the IQ–Adaptive Behavior Discrepancy MARC J. TASSE´ AND SUSAN M. HAVERCAMP Behavior-Analytic Experimental Strategies and Motivational Processes in Persons with Mental Retardation WILLIAM V. DUBE AND WILLIAM J. MCILVANE A Transactional Perspective on Mental Retardation H. CARL HAYWOOD Index

Index

Volume 32

Volume 31

Research on Language Development and Mental Retardation: History, Theories, Findings, and Future Directions LEONARD ABBEDUTO, YOLANDA KELLER-BELL, ERICA KESIN RICHMOND, AND MELISSA M. MURPHY

The Importance of Cognitive–Motivational Variables in Understanding the Outcome Performance of Persons with Mental Retardation: A Personal View from the Early Twenty-First Century HARVEY N. SWITZKY Self-Determination, Causal Agency, and Mental Retardation MICHAEL L. WEHMEYER AND DENNIS E. MITHAUG The Role of Motivation in the Decision Making of Adolescents with Mental Retardation ISHITA KHEMKA AND LINDA HICKSON

Residential Services Research in the Developmental Disabilities Sector STEVE HOLBURN AND JOHN W. JACOBSON The Measurement of Poverty and Socioeconomic Position in Research Involving People with Intellectual Disability ERIC EMERSON, HILARY GRAHAM, AND CHRIS HATTON

292 The Influence of Prenatal Stress and Adverse Birth Outcome on Human Cognitive and Neurological Development LAURA M. GLYNN AND CURT A. SANDMAN Fluid Cognition: A Neglected Aspect of Cognition in Research on Mental Retardation CLANCY BLAIR AND MEGAN PATRICK Dietary Supplementation with Highly Unsaturated Fatty Acids: Implications for Interventions with Persons with Mental Retardation from Research on Infant Cognitive Development, ADHD, and Other Developmental Disabilities NATALIE SINN AND CARLENE WILSON

contents of previous volumes Screening for Autism in Infants, Children, and Adolescents KYLIE M. GRAY, BRUCE J. TONGE, AND AVRIL V. BRERETON People with Mental Retardation and Psychopathology: Stress, Affect Regulation and Attachment: A Review CARLO SCHUENGEL AND CEES G. C. JANSSEN Diagnosis of Depression in People with Developmental Disabilities: Progress and Problems ANN R. POINDEXTER Index

International Review of Research in Mental Retardation, Volume 33: Developmental Epidemiology of Mental Retardation and Developmental Disabilities

International Review of Research in Mental Retardation, Volume 24 (International Review of Research in Mental Retardation)

International Review of Research on Mental Retardation (International Review of Research in Mental Retardation)