Pediatric Epidemiology 9783318061222, 3318061220

Table of contents :
Cover
Front Matter
Contents
Introduction
Ethical Issues in Paediatric Epidemiology
Abstract
Ethical Questions of Epidemiological Research on Children
The Vulnerability of Children as Probands
Between Health Research and Health Care – Theoretical Basis
(Institutionalised) Children in the View of Science. A Historical Retrospective of a Vulnerable Group of Probands
Good Epidemiological Practice – Ethical Guidelines for Epidemiological Research
References
Epidemiological Studies of Child Maltreatment: Difficulties and Possibilities
Abstract
Introduction
Definitions and Definition Problems
Research Sources
Population-Based Surveys
Specific Topics
Conclusions
References
Legislation on Genetic Testing in Different Countries
Abstract
Introduction
The European Union
Germany
Switzerland
France
The Netherlands
The United Kingdom
Portugal
The United States of America
Conclusions
Acknowledgement
References
The Dilemma Associated with Incidental Findings
Abstract
Incidental Findings and their Influence on Cohort Studies
The Usefulness versus the Risks of Incidental Findings for Participants
Frequency and Medical Relevance of Incidental Findings
Possible Strategy for Dealing with Incidental Findings in Cohort Studies
References
Challenges and Opportunities in Conducting Research in Developing Countries
Abstract
Background
Relevance of Research to Maternal and Child Health in Developing Countries
Prioritizing Research Areas and Topics
Translating Evidence into User-Friendly Communication Product
Conclusions and Recommendations
References
How to Recruit a Representative Sample and How to Look for It?
Abstract
Introduction
Ways to Achieve a Representative Gross Sample
Measures to Achieve Representative Response
Assessment of and Adjustment for Representativeness
References
The Epidemiology of Global Child Health
Abstract
The Child Survival Revolution
Determinants of Child Health and Survival
Equity in Global Child Health
The Need for Data
Future Challenges
References
How to Deal with Proxy-Reports
Abstract
Definitions
Proxy Perspectives
Proxy Selection
Quality of the Proxy-Report
Proxy-Reports and Missing Data
Proxy-Reports and Information Bias
Proxy Reports and Confounding
Self-Reports
Conclusions
References
Biology at a Young Age Differs from Biology at Later Ages: Developmental Aspects of Growth and Body Functions in Children and Young Adults
Abstract
Introduction
Age Determination in Children
Chronological Age Periods in Children
Gross Anatomy: Organ and Tissue Growth in Children
Body Size Measures
Longitudinal Growth Patterns
Pubertal Development
Age-Related Biochemical and Physiological Normative Reference Ranges in Children
References
Basic Epidemiology, Statistics, and Epidemiology Tools and Methods
Abstract
Hypotheses
Study Design
Data Sources
Statistical Analyses
Graphics
Hypothesis Testing and p Values
Statistical Software
Data Protection
The Concept of Standardization – What SDS Values Are For
References
How to Deal with Confounding
Abstract
Measures of Effect versus Measures of Association
Terminology
Criteria for a Confounder
Visualization of Confounding Using a DAG
Actions Against Confounding during Data Analysis
Examples of Confounding in Pediatric Epidemiology
Conclusion
References
Author Index
Subject Index

Citation preview

Pediatric and Adolescent Medicine Editor: W. Kiess Vol. 21

Pediatric Epidemiology Editors

W. Kiess C.-G. Bornehag C. Gennings

Pediatric Epidemiology

Pediatric and Adolescent Medicine Vol. 21

Series Editor

Wieland Kiess

Leipzig

Pediatric Epidemiology Volume Editors

Wieland Kiess Leipzig Carl-Gustaf Bornehag Karlstad Chris Gennings New York, NY 21 figures, 7 in color, and 23 tables, 2018

Basel · Freiburg · Paris · London · New York · Chennai · New Delhi · Bangkok · Beijing · Shanghai · Tokyo · Kuala Lumpur · Singapore · Sydney

Pediatric and Adolescent Medicine Founded 1999 by Martin O. Savage, London

Wieland Kiess

Carl-Gustaf Bornehag

Hospital for Children and Adolescents Department of Women and Child Health University Hospitals University of Leipzig Liebigstraße 20a DE–04103 Leipzig (Germany)

Department of Health Sciences Karlstad University SE–651 88 Karlstad (Sweden)

Chris Gennings Department of Environmental Medicine and Public Health 17 E 102 St, Floor 3, Room D3-134 New York, NY 10029 (USA)

Library of Congress Cataloging-in-Publication Data Names: Kiess, W. (Wieland), editor. | Bornehag, Carl-Gustaf, editor. | Gennings, Chris, editor. Title: Pediatric epidemiology / volume editors, Wieland Kiess, Carl-Gustaf Bornehag, Chris Gennings. Other titles: Pediatric and adolescent medicine ; v. 21. 1017-5989 Description: Basel ; New York : Karger, 2018. | Series: Pediatric and adolescent medicine, ISSN 1017-5989 ; vol. 21 | Includes bibliographical references and index. Identifiers: LCCN 2017049678| ISBN 9783318061222 (hard cover : alk. paper) | ISBN 9783318061239 (electronic version) Subjects: | MESH: Epidemiologic Methods | Child Health Classification: LCC RJ106 | NLM WA 950 | DDC 614.4083--dc23 LC record available at https://lccn.loc.gov/2017049678

Bibliographic Indices. This publication is listed in bibliographic services, including Current Contents® and Index Medicus. Disclaimer. The statements, opinions and data contained in this publication are solely those of the individual authors and contributors and not of the publisher and the editor(s). The appearance of advertisements in the book is not a warranty, endorsement, or approval of the products or services advertised or of their effectiveness, quality or safety. The publisher and the editor(s) disclaim responsibility for any injury to persons or property resulting from any ideas, methods, instructions or products referred to in the content or advertisements. Drug Dosage. The authors and the publisher have exerted every effort to ensure that drug selection and dosage set forth in this text are in accord with current recommendations and practice at the time of publication. However, in view of ongoing research, changes in government regulations, and the constant flow of information relating to drug therapy and drug reactions, the reader is urged to check the package insert for each drug for any change in indications and dosage and for added warnings and precautions. This is particularly important when the recommended agent is a new and/or infrequently employed drug. All rights reserved. No part of this publication may be translated into other languages, reproduced or utilized in any form or by any means electronic or mechanical, including photocopying, recording, microcopying, or by any information storage and retrieval system, without permission in writing from the publisher. © Copyright 2018 by S. Karger AG, P.O. Box, CH–4009 Basel (Switzerland) www.karger.com Printed on acid-free and non-aging paper (ISO 9706) ISSN 1017–5989 e-ISSN 1662–3886 ISBN 978–3–318–06122–2 e-ISBN 978–3–318–06123–9

Contents

VII

1 16 30 41 60 71 85 97 105

113 143

152 153

Introduction Kiess, W. (Leipzig); Bornehag, C.-G. (Karlstad); Gennings, C. (New York, NY) Ethical Issues in Paediatric Epidemiology Rotzoll, M.; Willer, M. (Halle-Wittenberg) Epidemiological Studies of Child Maltreatment: Difficulties and Possibilities Janson, S. (Karlstad) Legislation on Genetic Testing in Different Countries Rössler, F.; Lemke, J.R. (Leipzig) The Dilemma Associated with Incidental Findings Hiemisch, A.; Kiess, W. (Leipzig) Challenges and Opportunities in Conducting Research in Developing Countries Khan, M.I.; Memon, Z.A.; Bhutta, Z.A. (Karachi) How to Recruit a Representative Sample and How to Look for It? Hoffmann, R.; Gösswald, A.; Houben, R.; Lange, M.; Kurth, B.-M. (Berlin) The Epidemiology of Global Child Health Persson, L.Å. (London) How to Deal with Proxy-Reports Genuneit, J. (Ulm) Biology at a Young Age Differs from Biology at Later Ages: Developmental Aspects of Growth and Body Functions in Children and Young Adults Söder, O. (Stockholm) Basic Epidemiology, Statistics, and Epidemiology Tools and Methods Vogel, M.; Poulain, T.; Jurkutat, A.; Spielau, U.; Kiess, W. (Leipzig) How to Deal with Confounding Genuneit, J. (Ulm) Author Index Subject Index

V

Introduction

Paediatric epidemiology deals with issues related to children’s and adolescents’ health, diseases that occur at an early age, and socio demographics and relations between children’s health and various environmental conditions. Paediatric epidemiology combines paediatrics and epidemiology, epidemiology in paediatrics, and paediatrics analyzed by using epidemiological methods. The editors of this volume are grateful to the publisher – in particular to Dr. Thomas Karger and Gabriella Karger, Basel – for making the publication of this rare book possible and for having supported the concepts of paediatric and adolescent health in general over the years. Currently, there are not many books dealing with the special aspects of epidemiology in the paediatric population. Yet, ethical, developmental and societal aspects of paediatric epidemiology differ quite substantially from those in general epidemiology dealing with research in the adult population. It therefore does not come as a surprise that a chapter in this book deals with ethical issues in paediatric epidemiology, another one explores biology at a young age in relation to body surface, water content, relative fat mass and age-specific behaviors, while yet another part of the book deals with the dilemma faced in light of incidental findings in children’s cohort studies. Classical topics of epidemiology such as how to recruit representative samples, how to deal with confounding variables, and how to deal with genetic information are the core areas of the book. It is mandatory to devote some thought to the fact that quite often in paediatric epidemiology one collects data from a representative(s) of the individual (child) rather than from him/herself. Detection of abuse in paediatric epidemiological cohorts not only poses ethical dilemmas but also elicits legal responses. Legal and ethical aspects are also to be taken into account when one is to carry out epidemiological studies and research on cohorts in low-income countries. Last but not the least, this volume undoubtedly will add to our understanding of global trends in children’s health and health issues. The editors feel privileged to have been able to work with outstanding and wellknown authors from around the world who have truly done a remarkable job to deliver their chapters meticulously and contribute generously in an exemplary manner to the book. We wish the volume success and believe that it will encourage the conduct of further research in the field of child and adolescent health, most importantly

VII

by raising interest and creating awareness in the causes, prevalence and trends of frequently occurring civilization diseases. It is our sincere hope that this book will not only serve as a textbook for paediatric epidemiology but also as a key reference for those embarking on paediatric cohort studies and epidemiological studies involving the pediatric population. Wieland Kiess, Leipzig, Germany Carl-Gustaf Bornehag, Karlstad, Sweden Chris Gennings, New York, USA

VIII

Kiess · Bornehag · Gennings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

Ethical Issues in Paediatric Epidemiology Maike Rotzoll · Manuel Willer Institute for History and Ethics of Medicine, Medical Faculty, Martin-Luther-Universität Halle-Wittenberg, Halle-Wittenberg, Germany

Abstract In the centre of reflection on paediatric epidemiology, the most vulnerable set of people are the persons used for test, that is, the probands. Children are at particular risk of having their rights breached owing to their inability to express their consent, and because of their position within family structures and social structures outside the family. This can be shown with a historical example through the experiments carried out between 1947 and 1971 on children of various ages in youth care homes. On the basis of ethical guidelines and national and international agreements on research ethics, effective epidemiological research not only in the ethical, but also in the scientific sense, should be the aspiration. The realization of knowledge in epidemiology within the framework of public health raises the question – particularly with regard to research on children – as to what extent epidemiologists themselves are responsible for how the public receive their scientific knowledge. Here, there is a need for an ethically conscious theoretical foundation and anchoring of epi© 2018 S. Karger AG, Basel demiology between health research and health care.

Ethical Questions of Epidemiological Research on Children

To provide effective and, not least, ethical health care for a population requires knowledge, based on evidence, of the origins and spread of diseases and of the efficacy and safety of treatment. In order to be able to usefully act preventively and interventively, group-specific data are needed. Within the framework of the theoretical and normative premises of evidence-based medicine, studies, as sources of evidence-based knowledge, contribute in the sense of medical progress to the improvement of the health of children. It would appear necessary, then, in the sense of the principles of non-maleficence and beneficence, to involve children in research as probands [1], in

order to properly study not only the physical development in childhood, but also the specific psycho-social conditions associated with childhood and youth. Often it is not possible to elicit the influence of specific factors on the development of children by means of studies with children as proband groups; illnesses too that occur only in childhood or occur differently in childhood when compared to adults, can be investigated only by carrying out research on or involving children. There are numerous questions apart from those related to the clinical area, which involve children specifically and cannot be answered by a mere transfer of results from research on adults. It is thus in the interest of children and young people that studies are carried out on minor probands; at the same time, children and young people are a vulnerable group and must be protected in very special ways from abuse. In the constitution of the World Health Organization (WHO), health is defined as “a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity.” This definition takes social and cultural factors into consideration, not just as possibly negative influencing factors in illness. The latter would mean understanding health to be merely the absence of illness. According to the WHO definition, health, especially that of children, means the result(s) of a complex interplay of factors; apart from the absence of disease, the other factors are suitable protection, freedom as far as possible from anxiety, and access to health care. Epidemiology, therefore, necessarily includes researching the origins and spread of diseases in the group of minors and the social and cultural factors, from the family context to health care, that have an influence on the health of children. Children are impacted greatly by these factors in a special way, they being a particularly vulnerable group. Research on or with children thus implies that ethical challenges and questions have to be encountered by the researcher. From our point of view, what is of most serious concern are questions related to the ability of children to provide consent to be part of a research process and the need for protection. Because of numerous inadequate legal regulations on epidemiological research with children, there exists an ethical need for self-obligation.

The Vulnerability of Children as Probands

Because of their lack of ability to provide consent, children as probands are especially vulnerable to being misused by researchers [2, 3]. The ability to be able to provide informed consent is related to their ability to discern and understand all aspects of the research process. And for children and young people, these are qualities that are yet in the developing stages. So the decision to include children in research cannot be based on legal majority or legal competence. Minors, incompetent children and young people can be regarded as those with an ability to provide consent, depending on their individual stages of development and the gravity of the operation involved; their will should neither be ignored nor consciously transgressed. Fundamentally, the ability

2

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

to provide consent is the prerequisite for participation in a study. According to the (German) Pharmaceuticals Law [Arzneimittelgesetz], §§ 40–42a, since 1976, the legal custodian of children decides on the participation of children who are unable to provide consent in interventional clinical research projects; the custodian is entrusted with the responsibility of protecting the interests of the children and to save them from abuse. In addition, probands who are minors must each be enlightened according to their level of individual degree of maturity; a refusal from the child’s end must be respected in all cases. The Ethics Commission decides on the general admission of a research project involving minors and must be consulted before embarking on the project. Epidemiological studies that are not covered by all these conditions in Germany require, according to the (Muster-) Berufsordnung für die in Deutschland tätigen Ärztinnen und Ärzte1 15, that they be examined and advised upon by the Ethics Commission. In research on patients who are unable to give consent, a series of ethical questions must be administered to them. Protection of the dignity of man, which is anchored in the German constitution, must be the first goal. An offence against the dignity of a human being occurs when a human being is instrumentalised. So as to see to what extent probands are exposed to the danger of instrumentalisation within a study, as a rule, 3 different kinds of researches on human beings are distinguished. Research that accompanies the care of patients and thus is of advantage to the patient does not generally represent a breach of the prohibition of instrumentalisation. This is to be distinguished from the so-called group use research, the results of which are useful to a group similar to that of the probands (with regard to basic illness, gender, age, etc.) but not to the probands themselves. Such research is, for example, when the validity of a diagnostic early detection method for a certain illness is examined using a patient diagnosed with that same illness (this is true for children, too, since the AMG change of law in 2004). To what extent this type of group use research is ethically legitimate is much debated. From the ethical point of view, the argument that this is instrumentalisation of a patient is difficult to refute. Attempts to justify such a transgression of basic individual rights with the great usefulness for a group of people are likewise to be repudiated from an ethical standpoint. The well-being of the proband outweighs, in every case, the usefulness for a group. If the well-being of the proband is not affected, or only slightly affected, by a study, such a study can, in individual cases, be justified, but 1 Inter alia: Spiegel Online 2/2/2016, “...in Frontal 21 sprechen sie erstmals über ihr Schicksal: Kinder als Versuchskaninchen;” Hamburger Abendblatt 24/9/2016, “Wurden Medikamente an Bremer Heimkindern getestet?;” NDR Schleswig-Holstein Magazin 11/10/2016, “Medikamentenversuche: Opfer wollen Aufklärung;” ARD/MDR Fakt exklusiv 18/19/2016, Medikamentenversuche an Kindern in westdeutschem Heim; WDR 19/10/2016, “Medikamententests auch an Essener Heimkindern;” Bild online 19/19/2016, “Medizin-Skandal enthüllt: Heimkinder für Medikamenten-Tests missbraucht;” WDR 20/10/2016, “Bethel räumt Medikamenten-Versuche ein;” Westfalenblatt 03/11/2016, “Medikamentenversuche an Heimkindern: Studie zur Aufklärung im Gesundheitsausschuss des Landtags angekündigt;” Giessener Nachrichten 16/12/2016, ”Giessener Klinik bis in 70er in Kinderversuche verstrickt?”.

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

3

the ethical problem of principle has not been solved by this. A third form of research is that which is basically useful for other groups than those being examined. All types of basic research belong to this category, as this does not lead to an improvement in the care of the probands themselves or of a comparable group. Research useful for outside groups cannot be justified when using patients unable to give consent, as this is a clear case of instrumentalisation. Although epidemiological research is to be strongly distinguished from clinical research for the abovenamed reasons, the principle of protection of the individual dignity of the proband is valid here too. For paediatric epidemiology, it is important to observe whether and to what extent probands can benefit from the research, and what the other interests are in such research, be they purely scientific or even economic. In studies on probands who are unable to give consent, there can be no justification from an ethical standpoint for even a slight risk of a serious transgression of the dignity of the proband. Research projects can be distinguished, however, by another criterion too. If one takes the degree of damage or the endangering of the proband into consideration, then various forms of research in epidemiology (gathering data from medical records, interviews, questionnaires, and so on, up to and including physical examination within the framework of the study, and invasive interventions) may be distinguished. Among the possible risks, apart from stress and anxiety, are the breach of privacy, the breach of confidentiality, and those risks associated with even a minor intervention such as taking a blood sample. Although the research methods in epidemiology show rather minimal risks as a rule [2], the evaluation of the risk of damage in the context of paediatric epidemiology may deviate from the research on consenting patients. Independently of this, differences in the evaluation of a risk may remain as minimal [2]. To be able to undertake an evaluation of a study, an evaluation orientated to probands, the criterion of vulnerability should be taken into consideration for a more differentiated evaluation of studies. From an ethical point of view, one must ask which criteria must be met in the concrete case in order to do justice to the vulnerability aspect of children. The vulnerability of children results from various factors, among which is the inability to consent. Children often do not have the social means at their disposal to enable them to assert their own interests in hierarchically formed social contexts. In relation to parents [1], physicians or researchers, there is a clear asymmetry of power and knowledge. Children are dependent in many ways, and have only a few possibilities at their disposal to actively influence this. From this results not only the danger of forcing children to participate against their will in studies, or to feel forced, but also questions may arise with regard to the private sphere of children and young people, which is particularly needful of protection, also with regard to parents. It is thus paramount in the case of young people able to consent that the confidentiality of information given is preserved and clearly communicated not only to the parents, but also to the proband.

4

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

Between Health Research and Health Care – Theoretical Basis

Epidemiology as the interface between social sciences and natural sciences [4] plays a key role in the context of health research and health training. The exact kind of role is a matter for debate within epidemiology, ranging from neutral scientific discipline to active shaper of health training [5, 6]. Does epidemiology see itself as a neutral discipline making efforts to gather data, or should it contribute actively to health training on the basis of a normative target? While classical definitions of epidemiology and ethical guidelines emphasise that epidemiology should actively perceive its role in health training and health policies [6], there are also some arguments against such a position. Thus, scientific objectivity is endangered when contributing towards shaping health training is to be a task of epidemiology [6]; also, active intervention and shaping of public health lie outside the competence of researchers [4]. In the area of research on children, this conflict arises clearly. Children are more strongly subject to measures of prevention and health maintenance. Their possibilities of escaping concrete measures are strongly limited owing to their limited co-determination. Paediatric epidemiology has a special responsibility here. While the discussion hitherto on the role of epidemiology has started with the latter’s own understanding, it seems cogent to ask about the implications of these various roles for the probands and other population groups. At the core of these thoughts, there is the concept of disease and causality, as implicit and – frequently – insufficiently reflected theoretical basis [4]. Especially in research on vulnerable groups such as children, it is central from an ethical point of view not theoretically reflect on these basic assumptions. Rather, reference must be made to the specific social, political, and also cultural realities of life for these groups, and this must be integrated into research. Complex diseases with a large number of known (or unknown, for that matter) possible causes are indeed cases where epidemiology can often only examine partial aspects [5]. The question about those aspects that are in the focus of research can decisively shape the results. When medical and physiological reasons are in the foreground, for instance, possible socio-psychological aspects may be excluded under certain circumstances. In such a case, there is often a dubious reduction to a purely physiological and functional concept of illness. This is all the more serious, as, according to the WHO definition of illness, the actual disease value is determined by the perception of the patient. Frequently, children are not involved in the process of determining illness, or only to a minor extent and in the case of smaller children, this is simply not possible. This means that it is determined externally, not by those affected themselves, whether the disposition examined is an illness or not. The question of the position of paediatric epidemiology between health research and health care deals with a fundamental problem. While the ethics of public health and epidemiology, as well as medical ethics, often operate with more pragmatic, rather than theoretically supported, principles, the individual theoretical and practical

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

5

routes of access to epidemiological research are based implicitly on certain value ideas and theories. Frequently remarked on are the primarily utilitarian theories forming the background of ethical discussions and thoughts in the areas of public health and epidemiology [7]. This suggests that public health and epidemiology as its core discipline pose the question of maximising general use, that is, a general maximising of health, but here already the objection finds consideration that maximising general use cannot justify any breach of individual rights [8]. As far as conflicts occur in paediatric epidemiology between general use and individual rights, it is certainly called for to take the vulnerability of children into consideration. In ethical consideration of the legitimacy of studies and reflecting on the possible dangers for probands emanating from the study itself or its consequences, vulnerability should most certainly be in the centre. Part of the responsibility of paediatric epidemiology is formed by a consciously theoretical fundament that does not ignore the possible consequences of a study within the framework of health care and training. Epidemiology, apart from the knowledge of the spread and the causes of diseases within a defined group, also has a normative goal, which might best be adumbrated with the catchword “prevention.” The need for effective prevention is heightened by the raising of the average age. This begins already in childhood. But not only the prevention of illnesses throughout the course of life but also possible measures aimed at furthering the child’s motor skills, intelligence, and so on are coming increasingly into the focus of epidemiological research. In so doing, there is the danger that social demands are perceived as normative standards and paediatric epidemiology is faced with the task of delivering knowledge enabling these social norms to be implemented in the sense of (ideologically influenced) health training. The border between justified worry about the health of a child and an “optimization” according to social demands is very porous. The transition between purely descriptive epidemiological studies and such studies that see the minors involved as “ill” or “healthy” is blurred; the latter can have an influence on probands and their social environment. If concrete medical interventions result, or if such interventions (e.g., taking blood samples) are a part of the study design in any case, then a graduation and transition of epidemiological and clinical research appear possible. This renders a look at the history of research with children quite apart from the purely epidemiological context, for such a retrospection shows that children, especially those institutionalised, were not only frequently desirable objects of research, but also that scandals involving children let to the codification of ethical rules, and especially to the necessity of informed consent. At the same time it becomes clear that such ethical codes could not always protect the vulnerable group of institutionalised children. This, in turn, lets the necessity for informed consent on the part of the proband him- or herself appear as something to be held in especially high esteem, and not lightly questioned. On the contrary, the question poses itself of how this can be better integrated into day-to-day research, so as to become a matter of routine.

6

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

(Institutionalised) Children in the View of Science. A Historical Retrospective of a Vulnerable Group of Probands

In the year 2016, a scandal shook the worlds of experts and the public. At least until into the 1970s, medications were tested on children in the Federal Republic of Germany – children in youth institutions and in psychiatric institutions. The many press reports since February 20161, beginning with a report in the ZDF broadcast Frontal 21 on 2 February 2016 (“Medikamententests in Heimen. Kinder als Versuchskaninchen”) refer to a study by the pharmacologist Sylvia Wagner [9]. As she was looking through the Deutsche Medizinische Wochenschrift, she found around 50 pertinent publications from within the period 1945–1975. The articles freely state what sort of collective was used to test medications or inoculations, for example, “96 children at an infants’ and small children’s home were divided into 2 groups” [9]. The author found no indications of any declaration of consent by the persons responsible. Using these first results and starting with the names of the authors, she consulted further sources. In sum, she found 33 inoculation experiments with small children in homes between 1947 and 1970, and 13 experiments, primarily with neuroleptics and sexual hormones, with older children in homes or psychiatric institutions. These experiments on minors have an ethical dimension, not difficult to see upon examining the dates involved. The Nuremberg Codex of 1947 does not seem to have stopped any of them, although the first rule there states that research on human beings may be undertaken only on people capable of consent after they have been informed and have agreed. Not a few of the studies belong to the post-1964 period, after the Declaration of Helsinki. These manifests of ethics were not legally binding and the laws concerning medications were still in development. Not until the medications law of 1976 in the Federal Republic (the first was passed in 1961 and amended in 1964) was it required to provide proof of effectivity and innocuousness. § 40 deals in detail with the protection of human beings in clinical assessments, particularly the protection of minors by means of written consent of the legal guardian following elucidation. Here too it is laid down that probands shall not be “kept in an institution on court orders or the orders of a responsible authority” [10]. The approval of innoculations, however, has been legally regulated since 1896 [9]. In any case, the experiments described by Sylvia Wagner took place not only in cooperation with the producing firms but also in agreement with state authorities, for example, the Senator for Health in Berlin, the Robert Koch Institute or Federal Health Office, and some were financed by the German Research Council (DFG) [9]. Whether similar experiments were carried out after the law came into force in 1978 or after the introduction of Ethics Commissions, especially in the 1980s, has not yet been researched. Besides the experiments using medications on institutionalised children, Sylvia Wagner’s paper and the media echo a further aspect of scandalisation: an assumed connection with National Socialist crimes against humanity, a connection produced

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

7

by means of personal continuities. For example, Hans Heinze (1895–1983) is mentioned [11]; in the post-war period, he was the director of the Psychiatric Section for Children and Young People in Wunstorf near Hanover, and responsible there for experiments with medications. It is well-known that Heinze, President of the then newly founded Association for the Psychiatry of Children and Young People from 1941, was one of those mainly responsible for the National Socialist movement on “children’s euthanasia.” After the war and a time as a POW, he succeeded in founding a post-war career in the Federal Republic. Sylvia Wagner, then makes presuppositions in her paper, which were obviously taken up by the press: 1 The Nuremberg Codex and its history – the experiments on human beings in the concentration camps of the Nazis – ought to have exercised influences on research practice in the post-war years, as the Codex is a prominent example of the effect of historical events on ethical debate and practice – even then when it had become clear that the “lesson” had by no means been learnt always and everywhere [12]. 2 Research on institutionalised children in the Federal Republic of Germany shows at least continuity in personnel from National Socialist research on human beings, or more generally continuity with National Socialism. But does the reference to the extremism of deregulated research under National Socialism, together with the focus on the de facto existing continuity of person, not obstruct the view of more complex structural contexts over longer periods of time? The narrative of the criminal research in the concentration camps during the NS period, the Nuremberg Codex as the reaction to that, and as origin of “informed consent” and the protective guidelines in place today for human probands – does all these go deep enough [13]? What about the previous history of the research described by Sylvia Wagner? The figure of a skandalon in connection with research on human beings, on minors, too, and a resulting need for ethical rules is by no means limited to the period following the Second World War – it can be found, together with ethical discussions, in the late 19th century. A prominent example is the “Neisser case.” The dermatologist Albert Neisser (1855–1916) had “inoculated” patients with a new type of syphilis serum in 1892. Among the patients, of whom some later developed syphilis, were some prostitutes, some of whom were underage, thus belonging to a socially marginalised and vulnerable group of people [14]. Additionally, they had been involved in the research without being told and without getting their consent. Shortly after the scandal, the trial, and the public reaction to the event, the Prussian Ministry for Education and Medical Matters issued a directive permitting experimentation on humans only after detailed explanation to and after receiving consent from the probands [14]. Important effects of this directive are not known – the public debate, however, continued. A further influential scandal was the so-called Lübecker Totentanz (Lübeck Dance of the Dead), which involved non-institutionalised children. The whole matter was

8

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

actually an accident in the course of inoculation owing to impure material, using a method that had in principle already been introduced, the BCG inoculation against tuberculosis, with which, outside Germany, some 150,000 children had been inoculated by 1928 [15]. In Lübeck too, the parents had consented. Nevertheless, the accident, connected with a method of inoculation still seen as new, influenced the debate on research on human beings [16]. A little later, 1931, the Reich Ministry of the Interior issued remarkable guidelines that may be seen as the first document with such far-reaching requirements, nationally and internationally. The guidelines were published on the broadest basis, so that the expert public must have known of them [17]. Following a fundamental commitment to the need for experiments on human beings (“medical science needs the experiments on humans”), the guidelines point out the particular “obligations of the physician towards probands.” They then distinguish – as far as is known, for the first time – between the attempt at a cure with (potential) benefits for the proband and the scientific experiment in the narrower sense, in which such benefits are not in the foreground [17]. For experiments focusing on cures, it is established that a particularly careful assessment should be carried out in the case of children that a social position of neediness should not be exploited, that special care must be taken with living infectious agents – the rules clearly show the influence of the previous scandals. Perhaps the most important rule to distinguish between the 2 types of experiments is this: the attempted cure requires only the informed consent of the person involved or their representative; the scientific experiment requires the consent from the proband himor herself without fail. The guidelines of 1931 were not legally binding, and no punishment was threatened [17]. Little is known of their effects – but one cannot say that they remained unnoticed under National Socialism. Two cases of their application by a high-ranking NS functionary are documented. Hans Reiter (1881–1969), after 1933 President of the Imperial Ministry of Health, gave his opinion in 1937 on the planned experiment with a new measles serum by the Robert Koch Institute, which was to be tested on convalescent children. Experiments on healthy children, thus Reiter, were unacceptable “under all circumstances.” “Why no self-test?,” he noted sarcastically in the margin. Again in 1937, Reiter expressed his opinion on an experiment planned by a physician at the Women’s Clinic in Halle, Hamann, to be carried out on pregnant women with eclampsia. This was permitted only after the subsequent agreement of the director had been filed – as in prescribed in the guidelines [18, 17]. Further mention, for example, in publications on experiments on humans, is not currently known, but this point needs further research. Quite clearly the guidelines had no influence on the extreme form of deregulated research in the concentration camps. This ought not to be surprising, for the researchers used this ”historical chance” precisely because of the “unlimited possibilities” – Hans Reiter too obviously had no compunctions about participating in research on Buchenwald prisoners in 1941 [19]. The protection of probands was clearly only

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

9

meant for parts of the population now. Other groups were ascribed more the status of experimental animals, whose welfare, however, played no role, while dealing with experimental animals was regulated in the new animal welfare laws [20, 19]. This extreme research in National Socialism and its influences on the ethical debate are relatively well investigated [21, 22]. The “guidelines” were known at the Nuremberg trials; they were used as an exculpating argument for the German medical professionals. They appeared to document the ethical standards of this profession, in contrast to the 23 defendants. This helped to shape an image of a small number of criminal perpetrators versus a large number of physicians who acted ethically. Andrew Ivy, the expert for the prosecution, emphasised the supposedly legally binding character of the “guidelines,” with regard to the negative evaluation of experiments on humans worldwide in the public eye [17]. The historical research has hitherto concentrated primarily on the extreme, but much less so the daily practice of medical research, for example, in the university clinics, which could be more essential in answering the question of whether, and  how, ethical standards were taken into consideration in everyday research, whether they were discussed, and what lessons could be learnt from all these for today’s research. The historical research is, at the moment, changing the direction of its view from event to routine. An example of this is the book by the British historian Paul Weindling, which has the programmatic title “From Clinic to Concentration Camp” [23]. A bridge between the human experiments in the concentration camps and the research in the everyday clinic is formed by a contribution contained in the book, and dealing with the research department at Heidelberg, by Carl Schneider (1891–1946), one of the leading Nazi psychiatrists, who documented the then current research practice on the basis of psychiatric patient files of children [24]. The goal of the research department, established at the Heidelberg Psychiatric Clinic in the middle of a small university town, was to scientifically distinguish inherited and acquired “imbecility.” In the years 1943 and 1944, 52 children and young people were taken to the adult station for about 6 weeks. These underage probands were subjected to a comprehensive programme of investigation according to the then-current research. Twenty One of the children were subsequently taken to the Eichberg Institution near Eltvile/Wiesbaden and there murdered. In nearly all files, a table of contents of the file shows not only the comprehensive research programme at the cutting edge of contemporary science, but also that the murder of the children was conceptually included in the plan (point 16) [25, 26]. This research was by no means “pseudo-science.” On the contrary, the programme was designed according to the newest developments [27]. In the post-war period, the argument of “pseudo-science” was often used to draw a line between the criminal research of the NS period and the “good science” thereafter, as though this “good science” could be a guarantee for the ethical attitude of scientists [13]. When the research in Heidelberg (just as in Görden) made use of the ethically unlimited possibilities of

10

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

a specific historical situation with regard to the life of the probands, and this during war and in connection with the NS murders of patients, it nonetheless represented, at the one and the same time, “normal” research on a good scientific level in the everyday clinic. The Reich guidelines were clearly ignored here. A structural momentum involving the probands becomes visible too: most of the children involved were from a home and thus belonged to a particularly vulnerable patient group – the group of institutionalised children, which was obviously frequently used for human experiments still in the post-war period. The Nuremberg Codex served as a delimitation to the criminal research of the NS period but was also regarded by researchers as a “hindrance” to necessary research on human beings – not only in the Federal Republic of Germany (as were  considered the Reich guidelines of 1931, see [17], pp 42–45). Following the Declaration of Geneva by the General Assembly of the World Medical Association in 1948, from 1953 discussions within this Association about their own ethical manifesto were taking place; a year-long intensive debate on 2 groups of probands was provoked, prisoners and institutionalised children. In 1962, a draught for the Ethical Manifesto was published in the British Medical Journal, where it was stated that no experiments on institutionalised children not in the care of their parents should be carried out. In the concluding discussions, the reservations of American scientists played a particular role. They had carried out numerous inoculation experiments in the 1950s, and they thought that experiments on institutionalised children were indispensable because of the standardised conditions – thus they themselves had not held to the central point of the Nuremberg Codex [28]. The research in the United States and the fact that the “guidelines” in Germany had already been formulated in 1931 went to show that research on home children in the Federal Republic of Germany in the 1970s cannot be explained only with continuities from the NS period – a structural ethical problem of science itself is the case here. In the final redaction of the Declaration of Helsinki in 1964, children in institutions are no longer mentioned; research on those unable to give consent, including children with the consent of their guardians, is possible.

Good Epidemiological Practice – Ethical Guidelines for Epidemiological Research

The basis for a responsibly perceived role in research, also in health training and health policies, is a corresponding way of sensitizing to the questions discussed above in health training and further training (as this was already required with reference to research in the Reich guidelines of 1931). Independently of the ethical competence within epidemiology, the establishment of Research Ethics Commissions appears desirable for epidemiological studies too, which do justice to their own ethical questions in contrast to clinical research [8].

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

11

Ethical bases for epidemiological research have been demanded since the 1970s [29]. Besides questions of data protection, the choice of method, and the design of studies, questions of the rights of probands or the position of probands within the study [5], and the suitable and responsible way to deal with patients have all been discussed [30]. Not the least, there is an increasing uncertainty in the population at large about the partly contradictory results of epidemiological research, and a controversy that is present within the discipline about the task and placement of epidemiology within the context of health research and enlightenment has contributed to the formulation of ethical fundaments for epidemiological research in several places since the beginning of the 1990s [31]. The ethical guidelines first worked out for epidemiological research (Industrial Epidemiology Forum 1989, Council for International Organizations of Medical Sciences (CIOMS) 1990, International Epidemiological Association 1990, International Society for Environmental Epidemiology 1996, [31]) reacted to numerous ethical questions in the context of epidemiological research. However, there was criticism [31] that important aspects had not been taken into consideration. Thus, neither the value of ethics in training nor the role of epidemiologists in the context of health policies had been adequately discussed in the guidelines. In addition, there was insufficient evaluation of individual topics of specialised epidemiological research (e.g., molecular research), and the increasing importance of data protection guidelines in the context of changing information and communication technologies. In order to react to the dynamic development of the field, there was a demand already at the end of the 1990s for new or extended guidelines [31]. Especially in view of the ethical questions posed above with regard to the vulnerability of children as probands, it would pay to discuss whether and what form vulnerability in general and with reference to children as probands in particular should be recognised in the guidelines. Guidelines have an important function in the legitimation and (self-)definition of independent disciplines and professions [31]. As far as specific ethical questions are valid for paediatric epidemiology, these should be recognisable in the guidelines, on the one hand, to establish obligatory standards for research and on the other hand to lend expression to the self-image of epidemiology as a discipline. Besides existing guidelines of national epidemiological societies and associations, in 2008 the International Ethical Guidelines for Epidemiological Studies was developed by the CIOMS in cooperation with the WHO. The guidelines of the CIOMS refer, in various contexts, to the specific conditions of research on children and other vulnerable groups. Guideline 14 explicitly reveals the requirements for research with children. Basically, this should be without an alternative, that is, not replaceable by research on persons capable of consent. In addition, children, within the framework of their ability to understand, should be personally convinced of their participation; in every case, a refusal must be respected with no ifs or buts. In view of the minimal risks to the participants, the guidelines of the CIOMS permit studies of use to third parties, for example, when the influence of environmental factors during childhood on illnesses in the adult are to be investigated.

12

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

However, in this it seems problematic that there is no explanation of what an acceptable risk is, that is, the criterion of minimal harm is not formulated by the study participants themselves but by those conducting it. In 2004, the Deutsche Gesellschaft für Epidemiologie (German Society for Epidemiology, formerly Deutsche Arbeitsgemeinschaft Epidemiology, German Epidemiology Cooperative) the currently valid re-working of the Leitlinien und Empfehlungen zur Sicherung von “Guter Epidemiologischer Praxis” (Guidelines and Recommendations for Ensuring “Good Epidemiological Practice”), which had first appeared in 1999. These guidelines do not deal explicitly with vulnerable groups. Only guideline 1 refers explicitly to the ethical dimension: “epidemiological investigations must be carried out in harmony with ethical principles and respect the dignity and rights of man” [32]. According to the guidelines, ethical principles result from different legal foundations, as ethical duties are valid beyond the legal aspect for epidemiological practice. The basic requirement of respecting the dignity of the individual, however, often runs into complex considerations in practice. For this reason, the guideline prescribes the obligatory consultation of an Ethics Commission, which is to be carried out on the basis of the Checkliste zur ethischen Begutachtung epidemiologischer Studien (Deutsche Arbeitsgemeinschaft Epidemiologie 1999). This checklist makes the ethical requirements for epidemiological studies explicit and emphasises in particular the need for obtaining informed consent as a condition for participation in a study. Apart from cases in which, because of the type of data collecting the explicit consent of all participants cannot be obtained, the checklist does not, however, deal with any other cases of lack of consent. This is problematic, particularly from the viewpoint of paediatric epidemiology. The existing dilemma of having to carry out research on vulnerable participants unable to give consent, but at the same time having to fulfill the obligations of the special need for protection on the part of children, is not taken up either in the checklist or in the guidelines. In guideline 2 [32], it is established that the “population groups to be investigated [...] (must be motivated) with regard to the research question” [32]. In research with vulnerable groups, the question of to what extent the study could be carried out on other non-vulnerable groups must be considered central. The guidelines aim at a use/risk consideration, but as shown above, because of the lack of ability on the part of children to provide consent, the acceptance of even slight risks must appear questionable from an ethical point of view. To take into consideration the respect of vulnerability as a principle in judging the suitability of a group of probands means emphasising the fundamental principle of human dignity in the tendentially utilitarian question of the general use of a study. From the ethical viewpoint, in addition, guideline 11 seems especially relevant, which deals with the communication of research results and their implementation within the framework of public health measures. The guideline requires open and transparent communication of results and methods. Additionally, necessary consequences arising from the research need to be responsibly formulated and communicated. It is obvious that this is only partly achievable in studies on children; so it is the responsibility of paediatric epidemiology to work out measures for health training and elucidation in the context of child welfare.

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

13

References 1 Spriggs M, Caldwell PH: The ethics of paediatric research. J Paediatr Child Health 2011;47:664–667. 2 Leikin S: Ethical Issues in Epidemiologic Research with Children; in Coughlin SS, Beauchamps TL (eds): Ethics and Epidemiology. New York, Oxford, 1996, pp 199–218. 3 Council for International Organizations of Medical Sciences: International Ethical Guidelines for Epidemiological Studies. Geneva, 2009. 4 Neumeyer-Gromen A, Bräunlich A, Zeeb H, Razum O: Theorie und Praxis der Epidemiologie. Teil I: Systematik theroetischer Grundlagen der Epidemiologie als zentrale Fachdisziplin von “Public Health.” Prävention und Gesundheitsforschung 2006; 3: 190– 197. 5 Gross D: Zwischen Gesundheitsforschung und Gesundheitserziehung. Der Rollenkonflikt des Epidemiologen und seine ethischen Implikationen; in Gross D (ed): Zwischen Theorie und Praxis 2: Ethik in der Medizin in der Lehre, Klinik und Forschung. Würzburg, Könighausen & Neumann, 2002, pp 277– 300. 6 Weed DL, Mink PJ: Roles and responsibilities of epidemiologists. Ann Epidemiol 2002;12:67–72. 7 Buchanan A, Daniels N, Brock D, Wikler D: From Chance to Choice: Genetics and Justice. Cambridge, Cambridge University Press, 2000. 8 Schröder P: Public-Health-Ethik in Abgrenzung zur Medizinethik. Bundesgesundheitsblatt – Gesundheitsforschung – Gesundheitsschutz 2007; 50: 103– 111. 9 Wagner S: Ein unterdrücktes und verdrängtes Kapitel der Heimgeschichte. Arzneimittelstudien an Heimkindern. Sozial Geschichte Online 2016;19:61– 113. 10 Gesetz zur Neuordnung des Arzneimittelrechts. http://www.bgbl.de/xaver/bgbl/start.xav?startbk= Bundesanzeiger_BGBl&jumpTo=bgbl176110.pdf. vol 24, August 1976. 11 Schmidt-Langels D, Langels O: Das lange Leiden nach dem Kinderheim. SPON 2/2/2016:http://www. spiegel.de/gesundheit/diagnose/medikamententests-in-deutschland-das-lange-leiden-nachdem-kinderheim-a-1075196.html. 12 Beauchamp TL: In the Shadow of Nuremberg: Unlearned Lessons from the Medical Trial; in Rubenfeld S, Bendict S (eds): Human Subjects Research after the Holocaust. Cham, Heidelberg, Springer, 2014, pp 175–193. 13 Brody H: The Origins and Impact of the Nuremberg Doctor’s Trial; in Rubenfeld S, Bendict S (eds): Human Subjects Research after the Holocaust. Cham, Heidelberg, Springer, 2014, pp 163–173.

14

14 Elkeles B: The German Debate on Human Experimentation between 1880 and 1914; in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 19–33. 15 Bonah C, Menut P: BCG Vaccination around 1930 – Dangerous Experiment or Established Prevention? Practices and Debates in France and Germany; in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 111–127. 16 Nadav D: The “Death Dance of Lübeck”: Julius Moses and the German Guidelines for Human Experimentation, 1930; in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 129– 135. 17 Roelcke V: The use and abuse of medical research ethics. The German Richtlinien/guidelines for human subject research as an instrument for the protection of research subjects – and of medical science, ca. 1931–61/64; in Weindling P (ed): From Clinic to Concentration Camp. Reassessing Nazi Medical and Racial Research, 1933–1945. London/New York, Routledge, 2017, pp 33–56. 18 Hinz-Wessels A: Das Robert Koch-Institut im Nationalsozialismus. Berlin, Kadmos, 2008. 19 Bruns F: Medical Ethics and Medical Research on Human Beings in National Socialism; in Rubenfeld S, Bendict S (eds): Human Subjects Research after the Holocaust. Cham, Heidelberg, Springer, 2014, pp 39–50. 20 Roelcke V: Sulfonamide Experiments on Prisoners in Nazi Concentration Camps: Coherent Scientific Rationality Combined with Complete Disregard of Humanity; in Rubenfeld S, Bendict S (eds): Human Subjects Research after the Holocaust. Cham, Heidelberg, Springer, 2014, pp 51–66. 21 Weindling P: “No Mere Murder Trial”: The Discourse on Human Experiments at the Nuremberg Medical Trial; in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 167–180. 22 Schmidt U: The Nuremberg Doctors’ Trial and the Nuremberg Code; in Schmidt U, Frewer A (eds): History and Theory of Human Experimentation. The Declaration of Helsinki and Modern Medical Ethics. Stuttgart, Steiner, 2007, pp 71–116.

Rotzoll · Willer

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

23 Weindling P: From Clinic to Concentration Camp. Reassessing Nazi Medical and Racial Research, 1933–1945. London/New York, Routledge, 2017. 24 Rotzoll M, Hohendorf G: Murdering the Sick in the name of Progress? The Heidelberg Psychiatrist Carl Schneider as a brain Researcher and “Therapeutic Idealist”; in Weindling P (ed): From Clinic to Concentration Camp. Reassessing Nazi Medical and Racial Research, 1933–1945. London/New York, Routledge, 2017, pp 163–182. 25 Roelcke V: Human Subjects Research during National Socialist Era, 1933–1945; in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 151–166. 26 Hohendorf G, Rotzoll M: Medical Research and National Socialist Euthanasia: Carl Schneider and the Heidelberg Research Children from 1942 until 1945; in Rubenfeld S, Bendict S (eds): Human Subjects Research after the Holocaust. Cham, Heidelberg, Springer, 2014, pp 127–138.

27 Roelcke V: Psychiatrische Wissenschaft im Kontext nationalsozialistischer Politik und “Euthanasie.” Zur Rolle von Ernst Rüdin und der Deutschen Forschungsanstalt für Psychiatrie/Kaiser-Wilhelm-Institut; in Kaufmann D (ed): Geschichte der Kaiser-WilhelmGesellschaft im Nationalsozialismus. Bestandsaufnahme und Perspektiven der Forschung, Vol. 1/1. Göttingen, Wallstein, 2000, pp 114–148. 28 Lederer SE: Research without borders: The origins of The Declaration of Helsinki, in Roelcke V, Maio G (eds): Twentieth Century Ethics of Human Subject Research. Historical Perspectives on Values, Practices, and Regulations. Stuttgart, Steiner, 2004, pp 199–217. 29 Susser M, Stein Z, Kline J: Ethics in epidemiology. Ann Am Acad Pol Soc Sci 1978;437:128–141. 30 Sass HM: [Ethics in epidemiology]. Gesundheitswesen 1993;55:119–126. 31 Weed DL, Coughlin SS: New ethics guidelines for epidemiology: background and rationale. Ann Epidemiol 1999;9:277–280. 32 Deutsche Gesellschaft für Epidemiologie (DGEpi): Leitlinien und Empfehlungen zur Sicherung von Guter Epidemiologischer Praxis (GEP). https://dgepi.de/fileadmin/pdf/leitlinien/GEP_mit_Ergaenzung_GPS_Stand_24.02.2009.pdf. July 2008.

Prof. Dr. Maike Rotzoll Institute for History and Ethics of Medicine, Ruprecht-Karls-Universität Heidelberg Im Neuenheimer Feld 327, Raum 107a DE–69120 Heidelberg (Germany) E-Mail [email protected]

Ethical Issues in Paediatric Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 1–15 (DOI: 10.1159/000481319)

15

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

Epidemiological Studies of Child Maltreatment: Difficulties and Possibilities Staffan Janson Karlstad University, Karlstad, Sweden

Abstract The prevalence of child maltreatment in different countries and within different groups of children and families has been difficult both to estimate and to compare. Reasons for the wide variation in incidence and prevalence include differences in how maltreatment is defined, varying quality of the sources used, non-uniform construction of surveys and validity problems. Research on child maltreatment also has some specific difficulties compared to many other areas of epidemiological research, as the perpetrators seldom will acknowledge their actions even in anonymous surveys and some victims cannot describe what has happened due to exposure at an early age, severe brain injuries or reluctance to disclose violence from perpetrators on whom they are dependent. A large part of the variation in prevalence remains unexplained and some might be due to methodological artefacts. There is consequently a need to strive towards common and operational definitions of maltreatment and to work with representative samples. This chapter discusses common definitions, which are internationally agreed upon and recent international developments of surveys, which can be applied in research with parents and children. Adolescents in particular are judged as the most reliable sources. The chapter also brings up possibilities and problems with official resources like mortality registers, hospital reports, agency reports and police reports. Some specific topics like neglect, sexual abuse and cultural-geographical differences are discussed more in detail. The chapter © 2018 S. Karger AG, Basel ends with a discussion on ethical issues.

Introduction

Since the modern re-discovery of child maltreatment, boosted by the sentinel papers of Henry Kempe et al. [1] in the 1960s, our knowledge about this worldwide problem has increased substantially through epidemiological surveys and qualitative research. We have got a fair understanding about the extent of the phenomenon as well as trends, at least in industrialized countries [2, 3]. We also have quite firm knowledge about the devastating impact on children’s health and development, with adverse psychological,

somatic and social consequences that affect childhood and that has long-lasting effects into adulthood and old age [2]. There is also a widespread agreement that in order to make progress in the prevention and reduction of child maltreatment, it is important for policy-makers to be informed about its scope and characteristics. Policy-makers also need knowledge about whether information about maltreated children is reaching the attention of school teachers, hospital staff, police departments and social services or alternative agencies that are in the position to help and respond. As policy-makers make changes, provide training and raise awareness, they also want to know if their reforms are changing the patterns they originally observed [4]. The prevalence in different countries and within different groups of children and families has been difficult both to estimate and to compare, as noted in the WHO report on child maltreatment in the world from 2006 [5]. Reasons for this wide variation in incidence and prevalence include differences in how maltreatment is defined, varying quality of the sources used, non-uniform construction of surveys and validity problems. Research on child maltreatment also has some specific difficulties compared to many other areas of epidemiological research, as the perpetrators seldom will acknowledge their actions even in anonymous surveys and some victims cannot describe what has happened due to exposure at an early age, severe brain injuries or reluctance to disclose violence from perpetrators on whom they are dependent. A large part of the variation in prevalence remains unexplained and some might be due to methodological artefacts. There is a need to strive towards common and operational definitions of maltreatment and to work with representative samples.

Definitions and Definition Problems

Definitions of child maltreatment have been difficult to operationalize universally and there are differing standards in terms of legal, research and clinical perspectives. Legal definitions are based on cultural and relative social norms, which reduce consistency across cultures and geographical areas. Definitions of maltreatment from an epidemiological perspective are generally broader than legal definitions but also represent objective attempts to operationalize acts of maltreatment [6]. The scope of child maltreatment is generally defined to encompass physical and sexual abuse, emotional maltreatment, and exposure to intimate partner violence and neglect of a person under 18 years of age by an adult on whom the child is dependent. The WHO definition is as follows [5]: Child maltreatment includes all types of physical and/or emotional ill treatment, sexual abuse, neglect, negligence and commercial or other exploitation that result in actual or potential harm to the child’s health, survival, development or dignity in the context of a relationship of responsibility, trust or power. The very similar but shorter Swedish definition according to the “Swedish committee on child abuse and related issues” from 2002 is as follows:

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

17

Child maltreatment is when an adult person exposes a child to physical or psychological violence, sexual assault or humiliation, or neglects the basic needs of the child. Although it is short, this definition covers the problem area. It specifically does not discuss whether the maltreatment was intended or not, as this would lead to a number of other problems. For instance, how intent is actually delineated and whether acts that are part of a planned punishment or sudden unplanned outbursts of rage with severe consequences can be seen as maltreatment, are debatable questions. The above definition also does not take into consideration whether there is societal/cultural acceptance of corporal punishment of children. Although some understanding of cultural customs is necessary in practical work with children, this should not enter into definitions regarding research on incidence and prevalence. However, both of the above-mentioned issues will have great impact on comparisons between countries, particularly when research is built upon agency and police reports. Also, in surveys on parental attitudes and behaviour, we will often run into some of the following problems: • Is there a line of demarcation between harsh parenting and corporal punishment? At what point a child is identified as maltreated is fundamental to understanding the limitations of data estimating the epidemiology of child maltreatment. • Do some researchers only publish more severe types of maltreatment, but not what they may consider mild forms of corrections, like a slap on the head or even spanking of the buttocks? • Will parents all over the world give the same answers to specific questions, or will they look upon some questions as inappropriate or even provoking towards their parenthood? Will they not consider childhood to include all those up to 18 years of age, for example, if they would allow their daughter marry at the age of 15 or younger? • If a parent punishes his/her child as a means of preventing the child from hurting himself/herself or others – is that considered maltreatment or not? In many countries, it would not be reported, while it may certainly be reported in northern Europe, particularly in the Nordic countries. Other definitional problems are how one defines psychological abuse, sexual assault, neglect (passive and active), humiliation, witnessing violence, solitary or repeated violence and multiple violence. The best, but not perfect way, to overcome these problems is to let children themselves answer very specific questions about maltreatment behaviour separated from questions about attitudes (see more below).

Research Sources

Mortality Registers Mortality registers are normally of high quality in the industrialised countries, but there may be only one principal diagnosis registered and no contributing diagnoses. This is particularly troublesome when studying background factors and other possible

18

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

associations in cases of child homicide. Under such circumstances, the researcher has no other recourse than to read each individual patient record to gather the necessary data. In several countries in the western world, special Child Death Review Teams check every child mortality case, cases where murder has been suspected or proven or cases where a proper diagnosis has not been reached [7]. Sweden is known for its ongoing population-based registry, which is possible due to a unique personal identification number used for all official purposes. As an example, national cohort studies on mortality and mental health outcomes among children formerly involved in the child welfare system have given a deeper understanding of the manifestations of childhood trauma and the impact of maltreatment [8]. Inpatient Registers and Outpatient Health Registers Inpatient registers are normally of a higher quality than outpatient registers, if the latter exists at all. Before using any health registers one has to check their quality and completeness with the national or regional authority that handles them. In a country such as Sweden, the inpatient registers have been of very high quality for decades. However, the shift from ICD-9 to ICD-10 (International classifications of Diseases) was introduced gradually in the late 1990s, making it difficult to compare certain diagnoses over longer periods. Other countries have introduced ICD-10 later and it is important to know when this shift occurred for countries involved in comparative studies. A reluctance to register maltreatment diagnoses unless the health staffs are absolutely sure that a child has been abused further complicates register studies on child abuse. Diagnoses of maltreatment may be more accurate in countries like the Netherlands where there is no mandatory reporting to the social services and where multi-professional Child Abuse and Neglect Teams work to support the families through voluntary actions. Hospital data from different western countries has shown no decline in maltreatment-related injuries or fatalities [4]. Agency Registers and Out-of-Home Placements These registers differ very much between countries, both in their total coverage and in what they report [3]. A large problem is that social service registers may be nationwide, but quite often regional and sometimes even local and may be run by private institutions. Before working with such registers, one must check their quality, coverage and how detailed the data is. Processing such data will not usually give a true picture of prevalence, but the results may nonetheless be of great importance for decision makers. Such organisations may encompass community institutions involved with children, such as schools, mental health agencies, NGOs and child protection agencies. Data from these agencies can also be of importance in comparison with self-report data in order to demonstrate the number of abuse cases that have gone unreported over the years.

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

19

Another essential problem with agency registers and police registers is that many incidents of abuse or neglect are never admitted or reported. Estimates indicate that between 50 and 80% of all victims of maltreatment are not known to the child protection services. There are vast differences even between adjacent countries. As an example, the rate of substantiated physical abuse in Canada is two and a half times that of the United States [9]. National or Regional Registers on Factors Like Income and Social Position These registers are normally used to study social distributions with respect to abuse within a larger population. In countries where every person has a unique personal identity number, it is relatively easy to link different registers, allowing for powerful analyses of complex research questions. Before embarking on a register study, it is wise to check: • If there are unique personal identifications • If different forms of abuse are registered • If data quality differs between different registers • If important changes have been made in how data has been registered and when • If specific subgroups of people are not included or omitted from the registers Police Reports Police reports on child maltreatment are highly dependent on national laws and the existing thresholds as to how serious cases must be in order to be reported. When using such data one has to be aware of these facts and how well the report system is handled by the police. In the Nordic countries, all professionals working with children have been mandated to report not only obvious maltreatment cases but also suspected cases of child abuse and neglect to the social services since the 1980s. The social services then report severe cases to the police department when there is reason to believe that a crime has been committed. If all countries had mandatory reporting and handled these reports in the same way, international comparisons would be possible. However, great variation exists even between the European countries. In Sweden, police reports for child maltreatment have steadily increased and now also include many cases of neglect, probably due to direct reports from professionals outside the social services. An international public would therefore assume that child maltreatment is more common in Sweden than in other countries, while the opposite has actually been shown through self-report surveys among children, and severe cases of abuse are very few.

Population-Based Surveys

Parental Reports of Attitudes and Behaviour Most pervious population-based surveys have been directed towards adult survivors of child maltreatment, either through telephone interviews or postal questionnaires. As known from all retrospective studies, responses from adult participants are subject

20

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

to memory biases and reflect what may have happened decades ago rather than the current situation. Self-reports from adolescents, on the other hand, provide a more current view on the scope of the problem and respondents’ memories are less affected by a long delay [4]. A drawback may be that adolescents are too close to the events to have acquired a more objective perspective. A recent systematic review of childhood maltreatment assessments in populationrepresentative studies since 1990 [10] discusses several important topics concerning population surveys. In the introduction, it is stated that although causality cannot be inferred from cross-sectional surveys, it has recently been argued that representative community-based surveys have an important role to play in understanding child maltreatment and such surveys allow the study of relevant health outcomes that may be undocumented in administrative medical and social services databases. In addition, such studies allow for the exploration of research questions that are potentially difficult to address with child samples due to ethical and reporting requirements. However, population samples usually have as a drawback the fact that they are limited to persons with fixed household addresses and do not reach persons in prisons or institutions as well as other marginalised groups that may have been heavily exposed to maltreatment in childhood. This exclusion may give rise to underestimation of the true incidence of maltreatment as well as weaker associations between maltreatment and adverse outcomes. One of the world’s most well-known survey instruments aimed towards parents is the Conflict Tactic Scale (CTS), which since its creation in the 1970s has been revised and developed continuously [11]. It is: • Currently the world’s most accepted and used scale for interpersonal violence with more than 600 reviewed papers. There is a specific scale for the parent-child-relationship (upbringing). • Starts from the assumption that conflicts are unavoidable and asks about conflict solution techniques, from verbal consensus to severe violence. Asking in this way, from non-provoking questions and slowly trickling down to questions surrounded by taboo or strong emotions has been proved to work well. • Quantification from zero to >10 times a year. • Does not inquire about attitudes and emotions associated with the conflict solution techniques. • Can be administered as personal interview, telephone interview or questionnaires. The scale has been criticized for not putting the violence into a greater range of circumstances such as family life conditions, economy and isolation. It has few questions concerning neglect, emotional abuse and no questions about sexual abuse. Although retrospective self-reports generally include more detailed information of maltreatment than administrative reports, it has been shown empirically that they may miss violent incidents that have been officially reported. This appears to be true not only for events in early childhood but even later on. The reason may be not to awake unpleasant memories.

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

21

Prospective studies are important to elucidate causality but miss events that happen after the end of the study period. This may partially be solved by longstanding prospective studies, with the downside that the results may reflect a non-current societal situation. Reliance on a single method to identify experiences of childhood maltreatment often overlooks many cases. The maximal number of maltreatment cases is normally identified by using a combination of available methods, where the prospective methods seem to be most comprehensive. The most severe cases are, however, likely to be identified by both prospective and retrospective methods [6]. The majority of the world’s countries have no data on the occurrence of child abuse and have no official mechanism for receiving and responding to reports of child abuse or neglect. In 1996, the UN Secretary General called for the creation of a global study on children and violence that would address violence against children in homes and schools, but when the WHO report on violence against children was published in 2006, this work had not yet been started [5]. However, with support from the international child maltreatment report, the WHO required that studies on violence against and maltreatment of children should be carried out in every country, and data on child abuse should be collected and reported from all countries. This is part of each state’s responsibility to fulfill their obligation to the Convention on the Rights of the Child. While this chapter was written, WHO Europe has published a short practical handbook – Measuring and Monitoring National Prevalence of Child Maltreatment [12] – with the basic aim to support the creation of a surveillance system to measure and monitor child maltreatment across the European countries. The handbook suggests community-based surveys on prevalence as the most appropriate method in setting up a child maltreatment surveillance system and proposes the use of 3 established maltreatment questionnaires: the ICAST, the JVQ (see more below) or the ACE-IQ, that is, The Adverse Childhood Experiences International Questionnaire, developed by the Centers for Disease Control and Prevention at the Kaiser Permanente in San Diego 1995. The handbook also introduces a Short Child Maltreatment Questionnaire (one page) for countries lacking funds for bigger surveys. With UNICEF support, The International Society for the Prevention of Child Abuse and Neglect began the development in 2004 of an international survey through repeated Delphi rounds with experts from 31 countries. The survey was basically modelled after the CTS [11], but also from Juvenile Victimization Questionnaires [13] and the WorldSAFE questionnaire [14]. The parental version was finally tested in 7 countries in Asia, Latin America and in Russia and subscales showed high internal consistency except for neglect and sexual abuse subscales [15]. A child version (ICASTC) was successively developed using the same type of methodology and has been tested in a number of countries. The ICAST-C is now a multi-national, multi-lingual, consensus-based survey instrument available in a number of languages for international research to estimate child victimization. In this way, international comparisons

22

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

of prevalence can be done in order to set national and international priorities and garner support for programs and policy development aimed at child protection [16]. Serial surveys repeating the same questions at different points in time are of great value. In Sweden, parental self-report studies according to the CTS model have been performed in 1980, 2000, 2006 and 2011 [17]. Children’s Personal Experiences Asking children about their experiences and perspectives requires approaches that may differ considerably from those used successfully in adults. Particularly for young children, common survey or interview methods have limited applicability, as they are not in tune with the child’s level of psychological and emotional development, and therefore may give limited information about what they have experienced. Children may also be afraid to disclose their experiences in interviews and even in anonymous surveys, out of loyalty to a caregiver or fear of repercussions. Despite these limitations, it is vital to obtain children’s personal experiences and perspectives in order to understand the scope and extent of the problem. As previously mentioned, surveys among self-reporting adolescents can provide current and accurate information that carries less risk of memory bias. Specifically focused studies can also provide accurate information on underserved populations as well as knowledge of peer violence. Lower socioeconomic status is commonly associated with lower levels of participation in survey studies. However, the US national survey of children’s exposure to violence in 2014 indicated that those youths for whom parental consent was  refused for the interview came from households with more educated parents,  healthier children, higher income and less school or neighbourhood violence or from families with younger school children [18]. Conversely, immigrant parents were overrepresented among those who refused participation in the national Swedish (personal experience). Systematic deviations in response rates such as these must be taken into account when analysing the data and discussing the findings. Interestingly, a low response rate does not necessarily increase the bias of a sample. There are studies that have shown little association between response rate and the size of non-response bias [19, 20]. Sampling Methods and Sampling Size WHO Europe [12] suggests a 2-stage sampling process where first a subset of schools is selected via cluster probability sampling and thereafter a randomised sampling of school classes for the appropriate ages. This is actually the way it has been performed in Sweden since 1995. When calculating the sample size, a number of factors have to be considered: • Estimated prevalence of the problem • An acceptable error margin (normally 5%)

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

23

• Precision level • Clustering of data • Estimated non-response rate Prevalence estimates should ideally be based on findings from previous studies with similar populations.

Specific Topics

Neglect Within the field of maltreatment research, scientific studies regarding child neglect have been clearly underrepresented, and researchers have pointed out a “neglect of neglect.” There are many reasons for this. Historically it has been easier to observe and diagnose physical abuse. In addition, there is no common consensus about the definition or definitions of neglect. Child neglect may, in fact, be a composite of different types of omission on the part of the caregiver, or of unmet needs seen from the child’s perspective. These deficiencies can also be seen along a continuum of severity, frequency and chronicity, and may have varying impact depending on the child’s age and individual characteristics. The term neglect includes, but is not limited to, the following: • Neglect of basic needs such as nutrition and shelter • Medical neglect, where a child’s medical needs with regard to accessing medical or dental services, preventive health services or treatment with prescribed medicines are not met • Emotional neglect, where a child’s needs for love, attention and communication are not met by the caregiver. Exposing the child to violence between adults in the home may be seen as a form of both emotional violence and emotional neglect • Educational neglect, where the caregiver fails to ensure that the child attends school or support the child’s academic performance. A meta-analytic review from 2013 looked into studies between 1980 and 2007 of 13 independent samples of physical neglect and 16 independent samples of emotional neglect, both with almost 60,000 participants [21]. The overall estimated prevalence was 163/1,000 for physical neglect and 184/1,000 for emotional neglect, with no apparent gender differences. The following important research problems were pointed out: • The influence of research design on the prevalence of physical neglect was more pronounced than on the prevalence of emotional neglect. Studies on physical neglect in “low-resource” countries were conspicuously absent. • The use of validated instruments yielded a significantly higher prevalence for physical neglect than the use of non-validated instruments. • The combined prevalence of different forms of physical neglect was lower when one or 2 questions were used than when 3 or more questions were used. There was a significant increase of reported prevalence with an increasing number of questions.

24

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

• The combined prevalence in studies using convenience samples was significantly higher than that of studies with randomized samples For studies of emotional neglect • No differences in the reported prevalence were found between studies that reported on witnessing domestic violence only and studies that used a more comprehensive definition of emotional neglect • Interviews gave higher prevalence than questionnaires • Studies with a low to moderate response rate gave significantly lower prevalence rates of emotional neglect than studies with a high response rate. Emotional neglect may be more difficult to measure than physical neglect, as the construct of emotional neglect may be more open to personal interpretation. To overcome this problem, it is probably important to use multiple, behaviourally specific questions to rule out at least part of the subjectivity [21]. Child Sexual Abuse Child sexual abuse (CSA) is common throughout the world, and must be dealt with as one of the ways in which children are exposed to maltreatment. Studies of sexual abuse are fraught with a number of difficulties. The overall estimated prevalence of CSA is 127/1,000 in self-report studies and 4/1,000 in informant studies (agencies, official organs). This massive difference may partly be explained by the fact that most informant studies are based on reports of CSA during the last year (i.e., 1-year prevalence), while most self-reports rely on longer periods, often reporting lifetime prevalence. Another important reason for this discrepancy is that many informant studies probably miss most of the offences due to underreporting [21]. Self-reported CSA is more common among female (180/1,000) than among male participants (76/1,000). The lowest reported rates for both girls and boys have been found in Asia and the highest for girls in Australia and for boys in Africa. Girls are probably much more often exposed to sexual abuse, but it is probably also so that men are more reluctant to disclose CSA, especially in countries with a more traditional view of men as aggressors rather than victims. The fairly low CSA rates for both genders in Asia seem to be consistent with the idea that abuse experiences are less often disclosed in collectivistic cultures, and this also has to be kept in mind when studies are performed in western states with fairly large subculture populations. As in other maltreatment studies, evidence points at the use of multiple behaviourally specific questions instead of single-item labelled questions to obtain more accurate results. The use of behaviourally specific questions about CSA also diminishes the risk that the participants’ subjective perceptions and definitions will affect their interpretation of “sexual abuse,” a potential drawback of self-report studies [22]. Cultural-Geographical Differences between Countries As mentioned above, child physical abuse is a widespread global phenomenon, affecting the lives of millions of children all over the world. Recent meta-analyses on

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

25

cultural-geographical differences in child abuse [23] show extraordinarily great differences in reported prevalence of physical abuse per country, which seems to reflect how studies were performed rather than reality. The highest combined prevalence rates were found in studies using broad definitions of child abuse, as for instance is the case in the Scandinavian countries, where a box on the ear is registered as abuse, while this and even spanking on the buttocks are looked upon as normal parental behaviour in most countries in the world. High prevalence rates are also reported from studies concerning the whole childhood period and studies in which young adults have been the respondents. When performing studies in different countries it is therefore important to: • Prepare the prevalence study based on qualitative studies that demonstrate how children, adults, professionals and governmental bodies view what is child abuse and what is not. Results from such a study can give specific additional questions that can be added to an already well-known and validated questionnaire • Clearly state for which period in life the study concerns • Clearly delineate the groups of people that will be invited to answer the questionnaire • Be aware of that prevalence figures usually are higher in studies using more detailed questions. Validity Problems In their analysis of 54 representative population studies from 39 countries, Hovdestad et al. [10] found evidence for reliability and/or validity of the childhood maltreatment assessments in only 7 studies. Despite the availability of well-established checklists of life events, they are seldom used and the psychometric properties of nearly all measures are uncertain. A further complication is that maltreatment in childhood usually is of multiple types and single-item measurements are associated with underreporting [24]. Widom and Shephard [25] compared retrospective self-reports of early child maltreatment with official court and police records. When using severe/very severe violence subscales, individuals who were physically abused according to official records reported significantly higher rates of abuse than those who were not registered in official records. There was, however, a substantial group of physically abused individuals who underreported – almost 40%. Whether these people did not report because of embarrassment, a wish to protect parents, a sense of having deserved the abuse, a conscious wish to forget the past, or lack of confidence in or rapport with the interviewer, is not known. Some respondents might have been too young at the time of the abuse to remember it correctly and it is important to realise that what we remember from early childhood may be heavily dependent on information told us later in childhood, constructed by a parent, or both. On the other hand, when using a minor violence subscale, there was a very high rate of false positives. This means that the evaluation method has a direct influence of answers given [25].

26

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

Web-Based Surveys It has been argued that research participants may prefer to disclose victimization using a computer rather than in a discussion with an interviewer. The utility of online surveys in the area of child maltreatment is however uncertain, with potential risks for sampling bias, and up to now there is no indication that online surveys create more accurate estimates in population characteristics [26]. Ethics Involving children in research, in which children provide information that may result in risks for themselves or others, requires careful consideration of whether children have the capacity to understand informed consent. Research involving children in industrialised countries normally has to be approved by a research ethics committee, and co-operative research between an industrialised country and a low-income country should be approved by ethical research committees in both countries. Children should also be advised as to whom they may contact in case they become upset, experience traumatic memories or are worried for any other reason. Instruments such as the ICAST-C may not always be used safely and ethically in some countries where there are policies and laws that might compromise confidentiality or limit respect for the autonomy of the child providing information. Wherever an instrument is used, the investigators need to carefully develop their protocol with respect to recruitment, participation, consent, incentives and provision of child protection within the context of legal, social, and medical systems where the study is performed. The majority of researchers and policy makers have found that the benefits outweigh the problems of collecting contemporaneous child maltreatment data as evidenced by institutional review board-approved school surveys and household surveys around the world. One must be aware of children’s cognitive abilities, potential recall bias and, in the case of neglect, children’s specific needs. Adolescent respondents have demonstrated sufficient maturity to complete even long questionnaires and very few adverse reactions have been reported. It is obvious that some children may find it stressful to complete questionnaires about violence, but the findings are inconsistent and some children may even benefit from the surveys if they are followed up with possibilities for consultation or counselling. Several techniques are also available to increase comfort and privacy for children and adolescent responders. It is important to explain the survey carefully and to inform about confidentiality and each individual’s right to withdraw from the study [27].

Conclusions

Most researchers agree that maltreatment data can be collected from children, adolescents and parents with approaches that are accurate, methodologically robust, legal and ethical [27]. However, research regarding child maltreatment has a fairly short

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

27

history, with evidence-based methodology going back only as far as the 1970s, and epidemiological studies of children’s experiences started as late as the 1990s. Since then many well-validated instruments about children’s behaviour have become available, and after the WHO-report on child maltreatment in 2006, The International Society for the Prevention of Child Abuse and Neglect has developed epidemiological survey tools for parents and children, with support from UNICEF, that can be used worldwide. As mentioned above, WHO Europe has also in the autumn of 2016, published a handbook for measuring and monitoring national prevalence of child maltreatment. References 1 Kempe CH, Silverman FN, Steele BF, Droegemueller W, Silver HK: The battered-child syndrome. JAMA 1962;181:17–24. 2 Gilbert R, Spatz Widom C, Browne K, Fergusson D, Webb E, Janson S: Child maltreatment – burden and consequences in high income countries. Lancet 2009;373:68–81. 3 Gilbert R, Fluke J, O’Donnell M, Gonzalez-Izquierdo A, Brownell M, Gulliver P, Janson S, Sidebotham P: Child maltreatment: variation in trends and policies in six developed countries. Lancet 2012; 379: 758– 772. 4 Jud A, Fegert JM, Finkelhor D: On the incidence and prevalence of child maltreatment: a research agenda. Child Adolesc Psychiatry Ment Health 2016;10:17. 5 Pinheiro PS: World Report on Violence against Children. Geneva, UN Publishing Services, 2006. 6 Shaffer A, Huston L, Egeland B: Identification of child maltreatment using prospective and self-report methodologies: a comparison of maltreatment incidence and relation to later psychopathology. Child Abuse Negl 2008;32:682–692. 7 Fraser J, Sidebotham P, Frederick J, Covington T, Mitchell EA: Learning from child death review in the USA, England, Australia, and New Zealand. Lancet 2014;384:894–903. 8 Vinnerljung B, Hjern A, Lindblad F: Suicide attempts and severe psychiatric morbidity among former child welfare clients: a national cohort study. J Child Psychol Ppsychiatr 2006;47:723–733. 9 Fallon B, Trocmé N, Fluke J, MacLaurin B, Tonmyr L, Yuan YY: Methodological challenges in measuring child maltreatment. Child Abuse Negl 2010; 34: 70–79. 10 Hovdestad W, Campeau A, Potter D, Tonmyr L: A systematic review of childhood maltreatment assessments in population-representative surveys since 1990. PLoS One 2015;10:e0123366.

28

11 Straus MA, Hamby SL, Finkelhor D, Moore DW, Runyan D: Identification of child maltreatment with the Parent-Child Conflict Tactics Scales: development and psychometric data for a national sample of American parents. Child Abuse Negl 1998; 22: 249– 270. 12 Measuring and Monitoring National Prevalence of Child Maltreatment: A Practical Handbook. Copenhagen, WHO Regional Office for Europe, 2016. 13 Finkelhor D, Hamby SL, Ormond R, Turner H: The Juvenile Victimization Questionnaire: reliability, validity, and national norms. Child Abuse Negl 2005; 29:383–412. 14 Sadowski LS, Hunter WM, Bangdiwala SI, Munoz SR: The world studies of abuse in the family environment (WorldSAFE): a model of a multi-national study of family violence. Inj Control Saf Promot 2004;11:81–90. 15 Runyan DK, et al: The development and piloting of the ISPCAN Child Abuse Screening Tool-Parent version (ICAST-P). Child Abuse Negl 2009; 33: 826– 832. 16 Zolotor AJ, Runyan DK, Dunne MP, Jain D, Peturs HR, Ramirez C, Volkova E, Deb S, Lidchi V, Muhammad T, Isaeva O: ISPCAN child abuse screening tool children’s version (ICAST-C): instrument development and multi-national testing. Child Abuse Negl 2009;33:833–841. 17 Janson S, Långberg B, Svensson B: Physical Punishment of Children Banned Since 30 Years: The Swedish Experience. Chapter 19: Global Pathways to Abolishing Physical Punishment. New York, Routledge Publication, 2011. 18 Finkelhor D, Turner HA, Shattuck A, Hamby SL: Prevalence of childhood exposure to violence, crime, and abuse: results from the national survey of children’s exposure to violence. JAMA Pediatr 2015;169: 746–754. 19 Curtin R, Presser S, Singer E: Changes in telephone survey nonresponse over the past quarter century. Public Opin Q 2005:69:87–98.

Janson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

20 Peress M: Correcting for survey nonresponse using variable response propensity. J Amern Stat Ass 2010; 105:1418–1430. 21 Stoltenborgh M, Bakermans-Kranenburg MJ, van Ijzendoorn MH: The neglect of child neglect: a metaanalytic review of the prevalence of neglect. Soc Psychiatry Psychiatr Epidemiol 2013;48:345–355. 22 Stoltenborgh M, van Ijzendoorn MH, Euser EM, Bakermans-Kranenburg MJ: A global perspective on child sexual abuse: meta-analysis of prevalence around the world. Child Maltreat 2011;16:79–101. 23 Stoltenborgh M, Bakermans-Kranenburg MJ, Ijzendoorn MH, Alink LRA: Cultural-geographical differences in the occurrence of physical abuse? A meta-analysis of global prevalence. Int J Psychol 2013; 48:81–94.

24 Kessler RC, Davis CG, Kendler KS: Childhood adversity and adult psychiatric disorder in the US national comorbidity survey. Psychol Med 1997; 27: 1101–1119. 25 Widom CS, Shepard RL: Accuracy of adult recollection of childhood victimization: part 1. Physical abuse. Psychol Assess 1996;8:412–421. 26 Bethlehem J: Selection bias in web surveys. Int Stat Rev 2010;78:161–188. 27 Tonmyr L, Hovdestad WE, Draca J: Commentary on Canadian child maltreatment data. J Interpers Violence 2014;29:186–197.

Prof. Staffan Janson Department of Public Health, Karlstad University SE–651 88 Karlstad (Sweden) E-Mail [email protected]

Janson Child Maltreatment Epidemiology

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 16–29 (DOI: 10.1159/000481320)

29

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

Legislation on Genetic Testing in Different Countries Franziska Rössler · Johannes R. Lemke Institute of Human Genetics, University of Leipzig Hospitals and Clinics, Leipzig, Germany

Abstract Genetic testing has become part of routine diagnostics in an increasing number of medical conditions. At the same time, genetic testing methods have entered a high-throughput era, allowing both diagnostics and medical research to address more comprehensive and complex questions. And outside the usual healthcare system, direct-to-consumer genetic testing has gained importance in addressing issues on life style or ancestral origin. Still, legislation in terms of genetic testing seems to lag behind in many countries. We aim to review legal aspects regulating genetic testing and handling genetic test results in several countries and point out similarities and differences. © 2018 S. Karger AG, Basel

Introduction

Genetic testing has become an increasingly important part of medical diagnostics as well as medical research. Through several providers and laboratories, it is offered to the broad public within diagnostic and research settings but also as direct-to-consumer genetic testing, providing consumers with information on, for example, life style or ancestral origin. Both ethical and legal issues were raised making genetic testing an exceptional and very controversial topic in human medicine. In the past years, several Western European countries such as Switzerland and Germany have passed laws directly designed for issues related to genetic testing, focusing on the protection of the human individual with its personal genetic information. In this study, we focus on general regulations across Europe as well as on specific legislation on genetic testing in 6 European countries and in the United States. For individual countries, the distribution of responsibilities in the context of genetic examination and testing as well as differences between diagnostic and pre-symptomatic, so called predictive, testing are discussed first. Second, the performance of genetic

analysis and regulations concerning the handling of genetic samples are examined. Third, we focus on how genetic counselling is organized in each country focusing on how the results of genetic testing are communicated and how protection of genetic data is realized.

The European Union

The European Union (EU) with its 28 member states does not directly address genetic testing by law. In principle, the European Parliament and the European Council establish directives that, according to the “Treaty on the Functioning of the European Union,” require to be implemented in the national laws of the member states [1]. Nonetheless, responsibilities in genetic testing as well as provision of genetic counselling are not specifically appealed to and rely on the legislation of the member countries. With regard to in vitro testing being conducted, the general safety of genetic tests as expressed on the European market is addressed in Directive 98/79/EC from 1998 [2]. Furthermore, Directive 98/44/EC on the legal protection of biotechnological inventions applying to the conditions under which genetic tests can be patented was adopted in 1998 and implemented by all EU countries [3]. Concerning data security, Directive 95/46/EC of the European Parliament and the European Council governs the processing of personal data and is as well applicable to genetic data [4]. The Council of Europe with its 47 member states establishes conventions and treaties that are open for signature to member and non-member states of the EU. In 1997 the Convention for the Protection of Human Rights and Dignity of the Human Being with regard to the Application of Biology and Medicine was opened for signature and has been signed by 35 European states. It represents the first legally binding international agreement aiming at prevention of misuse of advances in biology and medicine to “preserve human dignity, rights and freedoms.” The convention covers major fields such as predictive genetic testing, regulations of consent and privacy as well as medical research but, for instance, excludes insurance, employment and criminal matters [5]. Discrimination of an individual on the basis of genetic qualities is prohibited under article 11. According to articles 11–14, genetic tests should be performed only for medical purposes and accompanied by “appropriate” genetic counselling. Eventually, 29 of the signing states have ratified and implemented the determined principles into national law in total. Six of the ratifying states maintain special reservations such as France or Switzerland. A total of twelve member states have neither signed nor ratified the Convention for different reasons including Belgium, Germany and the United Kingdom [6]. In 2008, an additional protocol to the convention was opened for signature by the Council of Europe Steering Committee on Bioethics focusing on prior information and consent to genetic testing for medical purposes and concrete indications for genetic counselling. It supplements the convention by diagnostics, disease predisposition or carrier testing [5].

Legislation on Genetic Testing in Different Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

31

Quality assurance for genetic services is appealed to in article 5 and it is stated that genetic testing should only be performed if there is a scientifically proven “clinical utility” at least (article 6). Interestingly, article 7 states that “a genetic test for health purposes may only be performed under individualized medical supervision.” This requirement would basically prohibit direct-to-consumer sales and require service delivery through medical practitioners only [7]. Until now the additional protocol has been signed by 9 EU member states including France, Iceland and Portugal but only ratified by 4 of them (i.e., Moldova, Montenegro, Norway and Slovenia) [8]. The European Commission as an executive organ of the EU funded the platform EuroGentest to “harmonize the process of genetic testing, from sampling to counseling, across Europe” and provides helpful information to genetic laboratories, healthcare professionals and the public. The webpage gives an overview on accreditation and certification issues and lists practice guidelines as well as workshops, tools and a quality management system for laboratories. For physicians, the platform provides a variety of sources leading to Ethical-Legal Papers of different European countries, Patient’s Rights Legislations in Europe and guidance on genetic counselling. EuroGentest has also developed a series of leaflets in 15 different languages containing general information for patients about genetics and genetic testing [9].

Germany

In 2009, the Federal parliament of Germany passed the Human Genetic Examination Act (GenDG) comprising the fields of medical diagnostics and determination of descent as well as insurance and employment sectors. It is aimed at determining the requirements for genetic examinations and genetic analyses and preventing discrimination based upon genetic characteristics. The duty of the state “to protect human dignity and to ensure the individual right to self-determination via sufficient information” is highlighted in the first paragraph. A diagnostic genetic examination in Germany can be carried out by any physician on the basis of written and informed consent of the index person. Yet predictive genetic examinations as well as genetic counselling are to be performed by medical specialists in the field of human genetics or specifically trained medical doctors. Under paragraph 3, a “genetic examination” is defined as any examination directed at genetic analysis or any prenatal risk assessment measures. Hence, not the testing method but the purpose of testing is decisive. Genetic analysis of a biological sample is allowed only in the course of a genetic examination by the responsible medical doctor or by “a person or institution commissioned by the responsible medical doctor” [10]. Still patients purchasing direct-toconsumer tests from abroad will not be penalized [11]. Biologic samples obtained for genetic testing may be used only for the initial purpose and must be destroyed immediately afterwards.

32

Rössler · Lemke

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

After determination of diagnostic results, genetic counselling has to be offered to the tested individual and this is a legal requirement. With regard to predictive as well as prenatal genetic testing, genetic counselling is required both before and after the analysis. The corresponding test results may be communicated only by the responsible medical doctor or genetic counselor directly to the patient. If a person or institution was authorized to conduct the analyses, the results may be disclosed only to the person who commissioned it. Results of genetic examinations and analyses must be retained and destroyed after a period of ten years by the responsible medical person [10].

Switzerland

The Federal Act on Human Genetic Testing of 2004 provides the legal basis of genetic testing in Switzerland. Germany’s GenDG has been designed very similar to this legal text in purpose and scope. It is focused on the protection of human dignity and personality, quality assurance and the avoidance of abusive genetic testing or usage of genetic data. This legal declaration comprises guidelines underlying genetic testing in medical diagnostics, determination of descent as well as in insurance and employment matters [12]. Genetic examinations as well as the order of diagnostic genetic tests can be conducted by any Suisse physician on the basis of legal consent. However, when it comes to predictive testing, only a geneticist or a specifically trained medical doctor is allowed to arrange it. Also, in Switzerland, the genetic sample is only to be used for its initial purpose and to be destroyed right after. Comparable to Germany, test results are to be disclosed to the index person through genetic counselling. Concerning predictive and prenatal genetic testing, genetic counselling is necessary before and after the analysis. At the moment, a total revision is in process and a new law is intended to be passed in 2017. In this context, regulations are in progress, for instance, concerning directto-consumer genetic testing. Inter alia a medical doctor or a pharmacist shall be involved in genetic testing for features like athleticism or ethnic origin. Prospectively, also non-hereditary genetics (i.e., characterization of tumors) shall be included [13].

France

In France, the civil law is determined by the Civil Code, which contains a special chapter named “Examination of the genetic characteristics of a person and of the identification of a person by his or her genetic fingerprints” (Articles 16–10 to 16–13). According to the French Public Health Code, either the diagnosis of genetic disease, the characteristics of one or more genes potentially causing a developing disease or the adaption of the medical care of a person must be the aim of testing. By this, genetic testing for the purpose of “gaining information” is not allowed [11, 14]. Additionally, the revised French

Legislation on Genetic Testing in Different Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

33

Bioethics Law of 2011 has implemented the prohibition of genetic testing outside the conditions of the Public Health Code and the corresponding penalties, one year’s imprisonment and a fine of EUR 15,000 are put down in the French criminal code. Genetic examination in France in general is allowed only for medical or scientific research purposes on the basis of informed consent. Only senior physicians can order genetic testing and are as well responsible for written informed consent of the index person. Yet asymptomatic or predictive testing can be ordered only by multidisciplinary teams that are declared and recognized by the French health institutions. Those teams comprise a geneticist or a physician expert in the disease, a genetic counselor and a psychologist. Usually an interval of about one month is required between the first interview and taking of a genetic sample to enable the patient to think about the consequences of genetic testing. Interestingly, in the special case of a prenatal testing, a pre-agreement on the issue of the pregnancy after the genetic test is required. No genetic test will be performed if no parental agreement on medical termination of the pregnancy in case a positive test result exists in advance. The Public Health Code appeals to quality assurance of laboratories and training of scientists, that is, laboratories need to get a specific authorization delivered for 5 years [11]. In France, there is as well a strong division between symptomatic and predictive testing. A diagnostic test represents the simplest case with only one genetic sample of the index person required and sent to the laboratory. For predictive testing, 2 different samples taken on different days are to be sent to the laboratory. With regard to prenatal testing, usually the analysis of both parents and the fetus is required at the same time. Genomic DNA will be stored for at least 10 years after the performance of the genetic test. Yet a clear directive on this time span does not exist. Nonetheless, the index person is allowed to ask for the sample to be destroyed at any time. Genetic counseling is mandatory for all types of genetic testing in France and has to be performed before testing by the ordering physician or a genetic counselor working with the physician. The ordering physician finally receives the result and is obliged to inform the patient about it during an appointment. All documents containing genetic information are usually kept in the patient’s file in the laboratory and the clinic and will be accessible by the responsible biologists and physicians. Finally, genetic data will be stored for at least ten years and, in principle, genetic diagnosis is allowed to appear in official follow-up reports.

The Netherlands

Professional medical practice standards and patient’s rights in the Netherlands are stated in the Dutch Civil Code. The Dutch Medical Treatment Contracts Act, as part of the Civil Code, applies to all contracts concerning medical services offered by a healthcare provider with a focus on strengthening a patient’s legal position.

34

Rössler · Lemke

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

Diagnostic as well as predictive genetic tests in the Netherlands can be ordered by any medical doctor. Still before starting any medical intervention, health-care providers have to provide sufficient information about indication, the proposed treatment, alternatives, prognoses, risks and possible side effects according to the Medical Treatment Contracts Act. On the basis of the Dutch Act on population screening, offering and practicing direct-to-consumer genetic testing for the detection of cancer risk factors as well as risk factors of non-treatable or non-preventable diseases, without a license, are against the law and a punishable offence. According to article 7, the responsible Dutch Minister can refuse providing the obligatory license if a test appears scientifically unsound, not in accordance with professional medical practice standards or if the expected benefit is not in balance with the potential health risks [15, 16]. In the context of genetic testing, there is no time limit for storing the obtained genetic samples in the Netherlands. This also applies to genetic information but, in clinical practice, genetic data is often stored for 3 generations of the concerning family. Provision of genetic counselling with regard to diagnostic as well as predictive tests is not covered by Dutch law. Though the Medical Treatment Contracts Act contains quality norms on how to deal with confidential patient data in general it does not contain any exceptional rules for the sharing of genetic information opening the possibility to share genetic test results between doctors without legal permission of the patient.

The United Kingdom

UK’s constitution is not contained in a written document but is to be found in statutes passed by Parliament and in the common law that existed over the centuries as a result of the decisions made in the courts [17]. But now no statute directly relating to genetic testing can be found. Most laws partly touching the topic are designed for consumer protection, for example, on medical devices, advertisements or data protection in general. In 2010, the Human Genetics Commission, which advised the government on new developments in human genetics, had set some guidelines in a Common Framework of Principles to “safeguard the interests of consumers and their families” and designed voluntary accreditation schemes for testing undertaken in laboratories. The UK Government however abolished the Human Genetics Commission as part of a budget-cutting and reform effort [18]. Nonetheless, in clinical practice in the United Kingdom, only consultant physicians do order genetic analysis. There is no official difference made between diagnostic and predictive testing, but medical doctors in general are likely to order predictive testing via a clinical genetics service. In 2004, the UK Human Tissue Act primarily concerned with the use of biological samples was passed. According to this Act, genetic analysis of any human tissue without legal consent of the donor is considered a criminal act [19]. Genetic samples

Legislation on Genetic Testing in Different Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

35

practically will be stored during a time span of mutual agreement between a physician and his or her patient. With regard to genetic counselling, there are no specific regulations. Even if it takes place, it does not necessarily have to be conducted by a geneticist. Nonetheless, in daily clinical practice, samples will not be tested unless counselling has taken place. The results of a genetic test will be communicated to the physician and clinical geneticist under whose clinical geographical area the patient lives. All incoming results will be scanned and entered permanently on the Clinical Data Repository where they are accessible to any user. Yet, an audit trail for every access to any record does not exist.

Portugal

Portugal has signed and ratified the European Convention on Human Rights and Biomedicine without any reservations and, according to the Portuguese Constitution, the convention has “superior force” over any national legislation. In the national legislation Law n°12/2005 of 2005 on personal genetic information, health information is of particular interest for genetic diagnostics. Inter alia comprises the principle of nondiscrimination of an individual with a genetic disease or a certain genetic heritage as well as restrictions for insurance policies and employers demanding genetic information of a person. According to the law, a diagnostic genetic test has to follow “the general principles that regulate all other health care interventions or services.” Still predictive genetic testing can be requested only by a medical geneticist on the basis of written and informed consent. With regard to pre-symptomatic testing of severe, late-onset diseases with no cure or effective treatment, a previous psychological and social evaluation must be provided [20]. Before obtaining a genetic sample, legal consent of the index person must be documented including the purpose of the collection and the duration of storage. Portugal clearly highlights the duty of the government to regulate the offer and performance of genetic tests in order to avoid their production by national or foreign laboratories not having “the support of a proper and multidisciplinary medical team”. Through this the government is held responsible for accreditation, certification and licensing of all Portuguese laboratories performing genetic tests [21]. Law n°12/2005 states that any genetic test results are to be delivered in the context of a medical consultation only to the index person. Medical records in general can be consulted only by a physician responsible for the patient’s treatment and sharing of any health information is allowed only under a patient’s written authorization. In the context of storage of health and genetic data, a separation from the remaining personal information has to be guaranteed through the establishment of different levels of access [20].

36

Rössler · Lemke

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

The United States of America

There are 3 federal agencies in the United States controlling certain areas of genetic testing. The Centers for Medicare and Medicaid Services regulate all laboratory testing, except research, performed on humans based on the Clinical Laboratory Improvement Amendments of 1988 [22]. The US Food and Drug Administration is responsible for regulating safety and effectiveness of genetic testing as medical devices under the Federal Food, Drug and Cosmetic Act but to date has regulated only a relatively small number of genetic tests sold as kits to laboratories. The Federal Trade commission is limited to how tests are advertised and to ensuring that advertisement is not misleading [23]. It has to be taken into account that there are different specific regulations on genetic testing in each federal state of the United States. Diagnostic testing can be ordered by every physician, but the legal framework varies. A major issue is the way in which surreptitious genetic testing is done because there is no federal law prohibiting private companies from offering DNA testing to consumers of various biological samples without requiring the consent of the tested individual. The level of protection from surreptitious genetic testing varies from state to state and inter alia depends on where a person lives, where the sample is analyzed and how the state’s law is interpreted [24]. In the United States, the storage of DNA samples is done only if paid for. Otherwise, genetic samples will be destroyed right after processing. In principle, genetic counselling is required in the context of genetic testing but the realization of a post-test assignment depends vastly on the individual result. As  a consequence, results are allowed to be communicated, for instance, via telephone. Genetic test results will be sent back to the ordering hospital but not necessarily to the ordering doctor. The genetic information will then be scanned and permanently accessible to any doctor in the corresponding hospital. The Health Insurance Portability and Accountability Act of 1996 is aimed at protecting patient privacy by restricting the sharing of medical information and defining it as “Protected Health Information.” In 2013, the Privacy Rule was modified according to the Genetic Information Nondiscrimination Act passed in 2008 by the US Congress to “restrict the access of issuers of health insurance and employers to individuals’ genetic information, as well as to prohibit genetic discrimination” [25]. The Freedom of Information Act of 1966 was the first US law allowing citizens to access Federal documents upon request. So concerning research documents, there is currently no specific exemption for the genomic information of participants registered in those federal databases. For this reason, the National Human Genome Research Institute, maintaining several databases containing genomic information, established a special data sharing policy allowing access to their individual research data only to researchers upon request.

Legislation on Genetic Testing in Different Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

37

Table 1. Comparison of regulations on genetic testing in different countries State

Who can order genetic testing? diagnostic

predictive

How long can DNA be stored after genetic testing is completed?

Germany

Any physician

Medical geneticist or specialized physician

Storage is not allowed, but can be performed in case of separate consent

Switzerland

Any physician

Medical geneticist or specialized physician

Storage is not allowed, but can be performed in case of separate consent

France

Consultant physician Multidisciplinary team

At least 10 years

The Netherlands

Any physician

Without limit

The United Kingdom

Consultant physician Consultant physician

According to mutual agreement

Portugal

Any physician

Medical geneticist

Not legally regulated

The United States

Any physician

Any physician

Not legally regulated

Any physician

Special Regulations in US Law Enforcement The so-called Combined DNA Index System of the US Federal Bureau of Investigation is used to search the National DNA Index System containing DNA profiles submitted by forensic laboratories. Names and personally identifiable information are not stored at the Index System, so analysts in the laboratories sharing the DNA profiles contact each other in case of a match [26]. In 2013, the US Supreme Court framed the rule that law enforcement may collect DNA samples from all suspects arrested for a crime stating “when officers make an arrest supported by probable cause to hold for a serious offense and bring the suspect to the station to be detained in custody, taking and analyzing a cheek swab of the arrestee’s DNA is, like fingerprinting and photographing, a legitimate police booking procedure” [27].

Conclusions

Comparing different countries worldwide, it becomes obvious that the ethical approaches to genetic testing as well as their legal bases are very different and inconsistent. In many countries, major legal issues of genetic testing remain unclearly regulated. Even in nations directly addressing genetic testing for medical purposes within their legislations, the growing market of direct-to-consumer tests purchasable by the public remains vastly unregulated, suggesting that existing laws have already become outdated [7]. The direct comparison on issues related to genetic testing between countries reveals significant differences. Table  1 addresses a few such issues, showing that all  countries have in common that genetic testing can exclusively be ordered by

38

Rössler · Lemke

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

physicians. Even most countries restrict predictive genetic testing only to specialized physicians. However, regarding what happens to the genetic testing sample after testing is completed, the legislations of the individual countries set strikingly different strategies ranging from discarding the DNA sample to storing it unlimited, and when it comes to the handling of genetic reports, many countries appear to not have regulations at all. Most likely, these differences or even uncertainties in regulation do not simply refer to specific national attitudes to genetic testing. They rather illustrate that current legislation in many countries no longer meet the requirements of the current role, extent and impact of genetic testing. Counteracting this growing discrepancy remains a major task in the years to come.

Acknowledgement The authors thank Christel Depienne (Strasbourg, France), Bobby Koeleman and Iris de Lange (Utrecht, The Netherlands), Sanjay Sisodiya (London, UK) and Heather Mefford (Seattle, USA) for their support and participation in helpful discussion.

References 1 The Treaty on the Functioning of the European Union. Off J Eur Union 2012;326. 2 Directive 98/79/EC of the European parliament and of the council of 27 October 1998 on in vitro diagnostic medical devices. Off J Eur Communities 1998; 331:1–37. 3 Directive 98/44/EC of the European Parliament and of the Council of 6 July 1998 on the legal protection of biotechnological inventions. Off J Eur Communities 1998;213:13. 4 Directive 95/46/EC of the European Parliament and of the Council of 24 October 1995 on the protection of individuals with regard to the processing of personal data and on the free movement of such data. Off J Eur Communities 1995;281:31–50. 5 Varga O, et al: Definitions of genetic testing in European legal documents. J Community Genet 2012;3:125–141. 6 Council of Europe, T. O. Convention for the protection of Human Rights and Dignity of the Human Being with regard to the Application of Biology and Medicine: Convention on Human Rights and Biomedicine. Eur Treaty Ser – No. 164, 1997. 7 Grimaldi KA, et al: Personal genetics: regulatory framework in Europe from a service provider’s perspective. Eur J Hum Genet 2011;19:382–388.

8 Council of Europe, T. O. Additional Protocol to the Convention on Human Rights and Biomedicine concerning Genetic Testing for Health Purposes. Counc Eur Treaty Ser – No. 203, 2008. 9 EuroGentest. http://www.eurogentest.org/index. php?id=138. 10 German Federal Parliament. Human Genetic Examination Act (Genetic Diagnosis Act – GenDG), 2009. 11 Borry P, et al: Legislation on direct-to-consumer genetic testing in seven European countries. Eur J Hum Genet 2012;20:715–721. 12 Federal Assembly of the Swiss Confederation. Federal Act on Human Genetic Testing, 2004. 13 Federal Assembly of the Swiss Confederation. Revision of the Federal Act on Human Genetic Testing, 2016. 14 Journal officiel de la République française. Code de la santé publique, 1953. 15 van der Maas PJ: Evaluatie Wet op het bevolkingsonderzoek, Evaluation Act on population screening. ZorgOnderzoek Ned, 2000. 16 Van Hellemondt RE, Hendriks AC, Breuning MH: Regulating the use of genetic tests: Is Dutch law an example for other countries with regard to DTC genetic testing? amsterdamlawforum, 2011.

Legislation on Genetic Testing in Different Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

39

17 Courts and Tribunals Judiciary. The justice system and the constitution. https://www.judiciary.gov.uk/ about-the-judiciary/the-judiciary-the-governmentand-the-constitution/jud-acc-ind/justice-sys-andconstitution/. 18 Genomeweb: UK Budget Cuts Whack Genetics Commission. https://www.genomeweb.com/sequencing/ uk-budget-cuts-whacks-genetics-commission, 2010. 19 Human Tissue Authority. Human Tissue Act, 2004. 20 EuroGentest. Patient Rights in the EU – Portugal. Eur Ethical-Legal Pap N° 13 2008. 21 The Portuguese Prime Minister, de SL Law n. o 12/2005 of 26 January, Personal genetic information and health information, 2005. 22 Centers for Medicare and Medicaid Services. Clinical Laboratory Improvement Amendments. https://www. cms.gov/Regulations-and-Guidance/Legislation/ CLIA/index.html?redirect=/clia/ (Accessed: August 30, 2016).

23 National Human Genome Research Institute. Regulation of Genetic Tests. https://www.genome. gov/10002335/regulation-of-genetic-tests/. 24 Strand NK: Shedding privacy along with our genetic material: what constitutes adequate legal protection against surreptitious genetic testing? AMA J Ethics 2016;18:264–271. 25 Feldman EA: The genetic information nondiscrimination act (GINA): public policy and medical practice in the age of personalized medicine. J Gen Intern Med 2012;27:743–746. 26 Federal Bureau of Investigations. Combined DNA Index System. https://www.fbi.gov/services/laboratory/ biometric-analysis/codis#CODIS-Overview. 27 Supreme Court of the United States. Maryland v. King, 133 S. Ct. 1958, 2013.

Prof. Johannes R. Lemke Institute of Human Genetics, University of Leipzig Philipp-Rosenthal-Strasse 55 DE–04103 Leipzig (Germany) E-Mail [email protected]

40

Rössler · Lemke

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 30–40 (DOI: 10.1159/000481321)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

The Dilemma Associated with Incidental Findings Andreas Hiemisch · Wieland Kiess Hospital for Children and Adolescents, Department of Women and Child Health, University Hospitals, Leipzig University, Leipzig, Germany

Abstract Incidental findings are a well-known and common phenomenon in cohort studies. They can have a crucial effect on the quality of the collected data. On the other hand, they can also have very serious consequences for the individual participant of a study. The challenge for the principal investigator is to find the best way between these two motives. For the success of the study an adequate strategy of dealing with incidental findings should be established prior to the start of the study. In the following chapter we would like to offer our support in dealing with this challenge as well as to pro© 2018 S. Karger AG, Basel mote a better understanding of incidental findings.

Only in the last 2 decades have incidental findings become a topic in the scientific world. Even the anthology “Ethics and epidemiology” by Coughlin and Beauchamp from 1996 has no mention of this issue at all [1, 2]. Yet, since then it has aroused everincreasing attention and is the root of many discussions. In the 1970s of the last century, medical diagnostics was more than revolutionized by the introduction of high-definition, three-dimensional imaging procedures such as MRI, CT, or sonography. Due to continuous development, these techniques are continuing to become more efficient and more precise as well as more affordable. Thus, these procedures have been utilized for scientific research purposes and therewith brought completely new opportunities to medical research project in general.

However, diagnostic broadness and high precision, both highly valued characteristics of modern imaging in a clinical environment, have brought a whole new range of challenges to scientific research. Results that are unrelated to the actual focus of the screening and also that are abnormal would be delivered frequently. Results in the field of clinical diagnostics, which are known as incidental findings, are setting new challenges for scientists. In the hospital environment, these findings are still appreciated and used for enabling in the early detection of illnesses. For the scientist, however, they mean having to think beyond the context of the study. The superior, meta individual aim of the study is not the sole focus anymore; instead, the individual concerns of a single subject constitute the aim of this study. A serious scientific analysis of the problem of incidental findings has just started at the beginning of the current millennium. Previously various descriptions and individual case studies of diverse, in clinical diagnostic appearing incidental findings have been published [3–5]. There are, nevertheless, very few papers on the theoretical problem itself including multidimensional analysis and approach [2, 6–8]. Furthermore, the term “incidental finding” is still being used very inconsistently. Neither a coherent nor a binding definition exists. In the global literature, the attempt can be made to find a fractional consensus to characterize incidental findings and can be summoned into 3 main conditions [9–14]: 1. Incidental findings are discovered during the course of a medical study. 2. They potentially impact the health or reproductive capabilities of the individual research participant. 3. They are discovered in the course of conducting research but are beyond the aims of the study. Although this consensus is being established, it is not uncontested. There are consistent controversial discussions regarding condition 1, as the very term “incidental findings” has its routes in the clinical diagnostic and it is still used broadly [15]. Also, incidental findings are entirely possible without any diagnostic intention, for example, during the making of an anatomical map [2]. With regards to condition 2, we would like to suggest the supplement that incidental findings “potentially impact the health, development or reproductive capabilities of the individual research participant.” The entire international literature on incidental findings is referring very little, if at all, to children as participants. Therefore, their specific requirements are not being incorporated. Since the book at hand is primarily about paediatric research, we feel it is an endorsement strongly needed. The third condition too leaves a lot of room for discussion. The extent to which we can even speak of an incidental finding while we are, for instance, finding an abnormality during a blood examination of a healthy research participant, despite specifically analyzing the explicit parameter, remains unanswered. Another question is to what extent it really is an unintended random result? It is by now known that incidental findings are discovered fairly frequently in studies that use modern imaging

42

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

techniques. That is why these results can at the most be characterized as random with regards to one single participant. In view of the population as a whole, no incidental findings will appear anymore, as we can now even go as far as to state that we nowadays expect the occurrence of incidental findings [8, 16]. It is apparent that even the definition of the term is far from a universal settlement. Vaguer still is the attempt to investigate the medical relevance of incidental findings. Also comparable are the legal aspects a principal investigator could encounter while dealing with incidental findings. Even sparser is the availability of data and existence of guidelines on incidental finings when one takes minors and participants who are unable to give their consent for participation in a study. Hence, for the scientist, an ethically responsible handling of incidental findings is further impacted with great uncertainty [8]. The following chapters aim to provide a general summary of the latest scientific research results on incidental findings to illustrate problematic aspects in dealing with incidental findings and to present possible practical approaches, specifically taking into account the legal and ethical responsibility for minor research participants.

Incidental Findings and their Influence on Cohort Studies

One of the favorite study designs in epidemiology is cohort studies. The cohort design have been increasingly applied worldwide for more than 70 years in terms of various differing characteristics, outcome variables and target groups [17]. In Europe alone, there are nearly 80 birth cohort studies registered at this moment in time [18, 19]. In addition, there are just as many cohorts comprising children and adolescents including newborn babies. The number of cohort projects with adult participants is even immensely higher. It can be assumed that the reason for this could especially be the cohort studies, which with their broad applied spectrum of diagnostic procedures could generate a high number of incidental findings [8, 20]. However, it is this area in which there are comparatively few scientific examinations on frequency and medical relevance, which will be subject of discussion in the following chapters. Yet, incidental findings have a specific relevance for cohort research. Cohort projects are by definition long-term studies. Often there is no fixed end point at all; rather the appearance of the outcome variables and the answer of the hypothesis define the time frame. Some designs are even “lifetime – cohorts” from the beginning. In any case, the aim is to accompany the participant as uninfluenced as possible until the appearance of defined outcome variables and to consistently perform measurements [21]. Though this succeeds, it happens only incompletely even with highest possible efforts and best methodical planning. The participation in a study alone as well as in singular tests influences the participants in their further thinking and subsequent behavior. For instance, the pupils of a school class in the German LIFE child cohort study were questioned and tested on lifestyle-associated

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

43

diseases. In follow-ups, the teachers were summoned for the motivation for the repeated participation of their class. The predominant response was the changed, more health-conscious behavior from the onset of the study [22]. This implied an occasional distinct differentiation from its natural course; in the worst case, there was even the absence of an outcome variable for long term studies. So could, for example, the participation in a study that examines the connection between the use of nicotine and the appearance of heart attacks and the cognitive focus on the issue lead to the participant quitting its use or even not start at all. Moreover, some participants answer with social desirability in mind and will not admit to smoking. This effect, named after the Hawthorne experiments, describes the influence due to almost every single test on the individual, resulting in falsification on both current test results as well as future ones [23, 24]. Principal investigators have to take this effect into account and design the study methodically so that the Hawthorne effect is as slight as possible. In correlation with the topic of this chapter, this also means that notably the acknowledgment of incidental findings offers a particularly high potential for the Hawthorn effect. The participant is compelled to deal with the incidental finding that leads to an above-average cognitive focus on the contents of the study. Even with seemingly primarily mundane variables, for example, weight, blood pressure or blood sugar anomalies will result in a consultation with a doctor to add further diagnostics and for the participant to undergo potential interventions. Depending on the subjective awareness of the potential danger of the incidental finding the participant is likely to change his health-related behavior. Thus, as an object of a study, the examinee deviates far from its natural course. Consequently, this should result in an exclusion of the cohort. Particularly no feedback of an incidental finding should be made considering the Hawthorn effects’ influence on the quality of the research result. In an Australian study, it is shown that 94.4% of respondents would give away blood and histoid samples to scientific research if they would get access to the specific individual information relevant to their health gained in the analysis. This study also shows that 83.4% of respondents would be interested to acquire information on hereditary genetic diseases or potential genetically health risks and that 70% of respondents would want to be informed of incidental findings in the biological samples even if these had no direct impact on their health [25]. In the Study of Health in Pomerania (SHIP), representative for the population, with participants aged between 20 and 79 years, whole body MRI scans alongside other medical procedures were carried out and an additional systematic examination of the motivation for the voluntary participation in the research program was performed [8, 26]. In preparation, verbal or written conversations were held in an attempt to clarify and explain the scientific background and the total absence of an individual profit. Nevertheless, at the end of the examination, 97% of the respondents stated that they took part in the examination due to their personal interest to clarify health-related doubts and maintain their overall health, 85% stated that they wanted to support science, 40% wanted to settle a medical condition and still 14% of the examinees were hoping to substitute preventive

44

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

checkups with the research screening (Multiple choices possible). Only 1.8% stated that the financial allowance made available as a reason for motivation to go through the medical examination. To evaluate and to individually analyze the answers, additional interviews were executed in the following weeks. During these interview sessions, none of the respondents could describe the stated scientific interest any further [8]. Hence, only the interest in personal health with mostly unobtainable expectations could be taken as a motivation to participate in a medical study. This effectively demonstrates how the willingness to participate is vitally influenced by the acknowledgement of the results and feedback, particularly on incidental findings. A consequent strategy without any feedback could therefore lead to a distinct bias in the representative nature of the cohort. A principal investigator has to ultimately decide on a balanced strategy to secure the quality of the study while considering the commitment to success of the project, not least taking into account the moral responsibility.

The Usefulness versus the Risks of Incidental Findings for Participants

In the subjective view of the participant of a medical study, incidental findings are always recognized as something positive. Incidental findings are considered a preventive method to detect illnesses before symptoms even occur. According to this naive way of thinking, it is perceived that the chances of healing would be much higher if the finding is discovered earlier. Only through persistent public health efforts in the last decades this health consciousness and behavior has developed. Physical fitness and prevention and early detection of diseases are very much in vogue right now. The urge to be physically checked up is so strong that even long travels to the medical centre or going through exhausting tests are willingly accepted. Personal reasons related to focus and interest in health commonly outweigh even financial compensations [8]. Experiences from the LIFE Adult Cohort Study [27] showed that participants were surprisingly willing to even pay to take part in the study. The consciousness of parents regarding the health of their children has also decisively changed over the years, although this is dependent on their social and economic status [28]. The idea of getting a thorough health check-up for their child is also the predominant reason for letting them take part in a medical study [22]. The participants can rarely differentiate between a scientific institute and a health care centre as Erdmann [8] were able to show. The actual relevance of incidental findings for the health of the individual is discussed in a later part. On the other hand, if an incidental finding occurs in the scientific research environment of principal investigators (PIs), they too often cannot tell which value the finding will have for the individual. The less experience they have in this specific field the more likely they are to decide on informing the participant. His decision will be based on his heartfelt ethical and moral obligation towards the proband and also on

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

45

Table 1. Positive aspects versus potential risks in the decision-making process whether or not to inform the individual about an incidental finding (for references refer to the text) Positive aspects 1 Early detection of life-threatening illnesses 2 Moral and ethical responsibility of the PI for the proband, more so if proband is a child 3 Legal security for the PI Potential risks 1 Medical risks (damage through following diagnostics, insufficient average standards, uncertainty of relevance of results, lack of treatment options and more) 2 Psychological risks (strain after transit of incidental finding, waiting time until being informed and more) 3 Financial risks (worse conditions for specific to the individual insurances, loss of earnings, privately funded medical treatments and more) 4 Health political risks (overloaded GP’s, high expenses for health insurance and more)

the fear of long-lasting damage that they would not be able to inform the individual. The lack of clarity of possible legal consequences in case the PI decides not to inform the participant is also part of the decision-making process (Table 1). Often neither the principal investigator (PI) nor the participant realize the whole array of risks an incidental finding can entail and can also understand the considerable and lasting effect on the proband and his environment. Ordinarily a participant who was informed of an incidental finding, for example, in the form of a space occupying lesion, will immediately consult his general practitioner (GP) or a specialist. In order to clarify the participant’s condition, the consulted doctor will again decide to perform further diagnostic measures. Every examination, however, is associated with risks of complications and unwanted side effects. Even routinely used methods such as drawing blood from the body can lead to injuries of the nerves or strong bleeding when it is not done properly. These risks are amplified when, for example, children are involuntarily examined; when they retaliate to subject themselves to such examinations they are many times put on anaesthetics. In this context, there are often underestimated dangers because of the mostly uncritically used computed tomographies. In an analysis, Pearce et al. [29] were able to show that in children under the age of 15 who had to undergo a CT diagnostic, the risk of a brain tumour or leukaemia tripled in the 10 years following the procedure. With the necessity for invasive diagnostics (such as bronchoscopy or laparoscopy), the risks for complications increase. In the end the ambiguous mass is often an artifact, e.g. because of a for the purpose of research appropriate, specific MRI weighting. In such a case, the damage of the following diagnostics outweighed the benefits of knowing the incidental finding. This field, however, is negatively affected by more medical difficulties. In many areas of medical diagnostics (e.g., blood parameter, blood pressure, ECG, lung function and many more), the range of what is classed as “normal” or average differs greatly

46

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

between adults and children. Frequently, due to the lack of studies on healthy children, there are no reliable age-specific standards at all. Thus, an adequate assessment early on of an abnormal result on a child is nearly impossible for the PI. Consequently, this increases the as incidental findings classified and therefore to be clarified results. Often the pathologic relevance is uncertain even if there is a normal range. The appropriate literature does not show a health impairing effect, for example, on 5 times increased vitamin B12 serum concentration. This gets still more complicated when a genetic analysis is part of the research project. In most areas, there is still very unclear geneillness-coherence and furthermore, there are no safe assertions about the actual occurrence of the disease even if there is a distinct relation (genetic penetrance). Also, ethically difficult are cases in which a genetic disease definitely occurs in the later life of the proband, verified by the analysis, such as the “Chorea Huntington”, but treatment options do not exist. The participant now knows he will be affected by a fatal disease. What is not known, however, is when this will be, and the participant is in no way able to influence the disease. From the moment the person becomes aware of any illness in his or her body, the knowledge about the future illness will drastically reduce the quality of life, possibly even cause psychological damage in the person. A second area of concern is the potential psychological strain that the tested person is likely to develop after hearing about the incidental finding. As previously mentioned, the main motivation for participants to take part in a study is commonly a personal interest in their own health. Thus, after being examined, the participants expect either a positive or a negative result. In an after-study-survey of the SHIP cohort, over 30% of the questioned participants stated that they found that the uncertainty about a potential finding invariably built up stress in them [8]. The results were sent out by postal service 4 weeks after they partook in the examination. Another period where stress built up was the time between the awareness of the incidental finding to the time when a consultation was done with the GP. A non-medical professional can rarely judge the urgency to clarify a finding. This tension increases drastically when the consultation with a specialist about the finding (e.g., Friday afternoon) does not reveal precise information regarding the finding. Furthermore, the degree of the psychological strain is very much defined by the kind of potential diagnosis. Just a suspicion alone raised as a result of an examination about the presence of a tumour can crucially influence the life of the individual in a negative way as well as be responsible for lasting psychological trauma. These symptoms can appear even after the negative confirmation of the suspicion. If there is a tumour suspicion when the participant is a child, then the world of the family falls apart. Hence, from our point of view, a professional conversation (possibly counseling) and ongoing support are vitally important, considering the enormity of the finding. The third area of risk includes the potential financial consequences that incidental finings can cause. Once a proband gains knowledge of a suspicious result or goes through a medical examination, for example, on the suspicion of a tumour, the proband has to make a decision if he or she wishes to conclude a health-, life-, accident-

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

47

or income protection insurance. This incidental finding can lead to a considerable increase in fees or in the worst case to the total refusal of the insurance with corresponding lifelong financial implications. By this children who take on the role as participants are the worst affected because such policies can usually be arranged only at the legal age, which comparatively prolongs the gap of care. Additionally, loss of earnings, private health care bills and even just the transport charges to and from the specialists can take families to the verge of their financial limits. A fourth area of concern is the political risk that incidental findings can imply. Big cohort studies often include 10,000 and more participants. There are also projects that are carried out with more than 200,000 people [30]. If in projects of this size all incidental findings are uncritically reported back to the participants without any reservation, then a majority of them would naturally seek a GP or specialist to clarify the results obtained. This would lead to the utter congestion of the medical care system. Furthermore, the financial expenditure for the necessary diagnostic methods would hardly be sustainable by the insurance companies. Facing the significant risks incidental findings can entail, the procedure of sharing the feedback should be carefully and critically planned out before the start of the study. The individual consequences for the participant can be tremendous. The PI is obliged, according to the Declaration of Helsinki [31], to keep any damage away from the volunteering proband. Children as participants are thereby always under special protection.

Frequency and Medical Relevance of Incidental Findings

The previous sections demonstrated the influence incidental findings can have on a research project as well as the individual consequences for the participant. Yet, there are additional questions that require answers: Do incidental findings happen so frequently that it is considered worthy for the PI to investigate upon them further? Is there medical relevance at all to these incidental findings? What are the possible consequences? And can one, from a medical point of view, simply ignore incidental findings? Over the last 7 decades there have been multiple publications from the routine life of clinical diagnostics on incidental findings [3]. With the establishment of modern imaging modalities such as computer- or magnetic resonance tomography in the 1970s, a strongly increasing interest to publish was observed. On the other hand, it can also be assumed that these examination techniques especially led to more frequent incidental findings [15, 32, 33]. Unfortunately, only a very few surveys with a methodically good approach have focused primarily on incidental findings. In most articles, they are merely a byproduct of the actual study and at best the frequency and the types of findings are described. Often, they even are just individual case reports. Methodically valuable analysis stems almost exclusively from cohort studies.

48

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

Subsequently selected results from examinations from the past few years will be briefly presented. The relative indication of incidental findings (in percent) is regarding the total number of examinations. In clinical diagnostic as well as medical research projects studies on the brain, the spine or the heart are often the focus of imaging. Quattrocchi et al. [34] found in a population-based cohort study with 3,000 lumbar/spinal MR images on adults also 2,060 (68.6%) extra spinal unexpected findings and classified them according to the organ systems where they occurred (Vascular system [54], Kidney [742], Uterus [365], Ovaries [222], Prostate [30], Bowel [358], Liver [44], Spleen [2], Lymphatic system [38], Adrenal glands [1], Abdominal-Pelvic Fluid [204]). In addition, the accidental findings were classified according to their expected relevance as clinically inimportant findings – no further work-up indicated (1,721 = 57.3%), likely unimportant findings, incompletely characterized (265 = 8.8%) and potentially important (74 = 2.5%). For instance 11 aortic aneurysms, 38 enlarged lymph nodes, 2 uterine solid lesions, 15 bladder wall thickenings or prostate lesions, 1 adrenal glands solid lesion as well as 7 colorectal wall thickenings suspected for cancer were classified as potentially important/of relevance. Rutherford et al. [35] examined 161 adult patients with chronic kidney disease via cardial MRI. In total, they identified 102 (63.3%) non-cardial incidental findings on 95 patients of which 15 (9.3%) were suspected to be malignant. Dunet et al. [50] (2016) were able to detect in a meta-analysis including 12 studies with a total of 7,062 patients aged between 0.5 and 93 years for cardial MRI a prevalence of minor and major incidental findings with 17% (9–26%) and 12% (7–18%) respectively. Tan et al. [51] (2015) analyzed 1,597 abdominal multi-phasic, multi-detector computed tomography of healthy grown-up potential kidney donors (aged between 18 and 74 years). As a result of this analysis, 2015 (131.8%) suspicious findings on a total of 1,195 (74.9%) patients could be detected. In 17.3% of participants, these findings were classified as incidentalomas. Real malignant neoplasias could be found in 3 (0.18%) of the 1,597 participants in the clinical diagnostic performed later. A scientific processing of incidental findings on brain MRI was done in the prospective Rotterdam Cohort Study. A total of 2,000 adults aged 45–97 years were divided into 3 age groups: (a) 45–59, (b) 60–74 and (c) 75–97 and were analyzed. Most commonly identified were asymptomatic brain infarcts (a: 4.0%, b: 6.8%, and c: 18.3%), meningioma (a: 0.5%, b: 1.0%, and c: 1.6%) as well as aneurysm (a: 1.7%, b: 1.8%, and c: 1.6%), although for the first mentioned, a definite age dependency was recognized. Three cases resulted in urgent intervention: one case of malignant primary brain tumour (low grade glioma), one case of multiple cerebral metastases and one case of a large, chronic subdural hematoma in an otherwise asymptomatic person after a minor head trauma 4 weeks prior to the time of intervention. Two cases led to surgical intervention: the subdurale hematoma and a 12-mm aneurysm of the medial cerebral artery [36]. Altogether 3.2% of the participants were referred to a consultant. For three fourths of these participants again a further wait was decided after the consultation with the specialist, and the remaining 0.8% had an intervention take place. However, the analysis of the Rotterdam Cohort Study did not finish there. The regular follow-up examinations of the

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

49

study also included process assessments of previously discovered incidental findings. The majority of potential meningiome and aneurysms that were either untreated or not confirmed stayed unaltered during the average after-care period of 48–60 months [37]. The study group of Illes et al. [6] investigated the occurrence of incidental findings in brain MRI depending on the age and gender of the participants. Unexpected results were observed in 47% of the participants. Hereby, practically double the number of findings (60%) were detected in the group of over 60 year olds than in the group of younger adults (34%). Of those detected, 6.6% were recommended to seek further medical clarification. Remarkably, of the results that needed clarification, three-fourths were classified urgent (one cavernous angioma and 2 arterio-venous malformations). Among those over 60 years of age, on the other hand, none were classified urgent. Gender differences to the disadvantage of male participants could be recognized only in the higher age groups (81% male to 41% female). In another interesting and extensive study, the connection between the analyzed region of the body and the imaging technique was tested. In order to do this, 1,426 scans of research participants aged between 3 and 97 were analyzed. Chart 2 summarizes the results. Thus, incidental findings are often found in abdominal CT, followed by thorax CT and brain MRI. Sonography and nuclear medicine imaging, on the other hand, seem to have a rather small potential for unexpected results. There were no significant gender differences. However, the dependence on age became obvious once more. Regarding an age group younger than 40 years (reference), the odds ratio rose for the 40–65 year olds to 4.1% and for the over 65 year olds to 9.7%. Another characteristic feature of the study is the tracing of the confirmed findings. Thirty-five (2.5%) of the total 1,055 (74%) suspicious results were rated as “need clarification” and it was recommended to the participant to see their GP. Of those 35 results 27 were again referred to a specialist. For 5 participants, a non-invasive diagnostic procedure including meanly blood tests followed. Six participants had to undergo invasive diagnostic (bronchoscopy, biopsy and similar) each without further therapeutic provisions. For 8 participants, the exclusion diagnostic resulted in surgical intervention, 2 others received radiotherapy (renal cell carcinoma, liver metastasis) and a further 2 participants received medicinal treatment. IT was established in retrospect that from a total of 35 participants who had been referred to the clinicians for further clarification incidentally detected pathologies, six (0.4%) actually benefited from the incidental detections of the finding. For 24 participants, the benefit was rated “vague”. Yet, these participants too had to forgo invasive and risky tests such as biopsy, laparoscopy, CT, PET, and other similar tests that often identified suspicious tumours as enlarged nymph nodes [38] (Table 2). Through rapid technological progress, the use of modern imaging techniques is getting ever more cost effective, which in return implies that there would be an increased utilization of these techniques. The technological progress makes the devices also even higher defining and faster; a deeper tissue penetration is achieved and makes a versatile combination of methods possible. Only a few years ago, 1,5 Tesla MRI devices were state of the art, while currently, 3 Tesla MRI devices are being established in many areas. Even MRIs with 7 Tesla are already being used [32]. Because of the

50

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

Table 2. General Frequency of IFs in Imaging Research [38] Imaging

Number of Total examinations number of IFs

Number of IFs/ examination

Number (%) of Number of examinations Ifs receiving action (% of with any IF examinations)

CT Thorax Abdomen/pelvis All other

361 207 48

419 266 29

1.16 1.29 0.60

198 (54.8) 126 (60.9) 12 (25.0)

10 (2.8) 19 (9.2) 1 (2.1)

MRI Head All other

231 83

136 30

0.59 0.36

99 (42.9) 16 (19.3)

5 (2.2) 0

Sonography

120

15

0.13

11 (9.2)

0

Plain film radiography

257

154

0.60

100 (38.9)

0

Nuclear medicine

119

6

0.05

5 (4.2)

0

1,426

1,055

0.74

567 (39.8)

Total

35 (2.5)

obtained enhancement of detailed resolution, more unexpected modifications are being recorded, which causes an increase in incidental findings in equal measures [39]. Following this trend, the SHIP cohort study, representative for the population in Pomerania [26] uses whole-body MRIs with contrast-enhanced heart-MRI, MR angiography and for women, MR-mammography on healthy volunteering adults since 2007. The analysis of 2,500 scans resulted in a rate of incidental findings of 53.2% (1,330). Of those, 904 had an apparent medical relevance and 787 (31, 4%) were confirmed with the participant: 9 results were classified “urgent.” After localisation, the abdominal organs (6.8%), the urinary passages (6.8%) and the skeletal system (6.0%) were most commonly affected [40]. Far less yielding is the scientific literature with regards to frequency and relevance of incidental findings during childhood, which may be routed in the prevailing ethnical conventions in medical research on children and juveniles. Especially referring to modern imaging techniques, 3 fundamental restrictions can be conducted. Minors are not or only partly able to consent. Additionally, children have no perceptible interest to voluntarily take part in a study. Second, the operationality is limited, especially with very young children, as they simply do not lay still for long durations of time. Then in a clinical setting, well-established anaesthesia procedure is hardly justifiable in the context of scientific research. Third, because of the health implications, examinations including ionizing radiation, for example, CT, on infants are also inacceptable. For this reason, it is commonly practiced to analyze only those scans that were done during medical care or only when older children are included in studies. The group around Ortega et al. [41] analyzed 524 CT scans, which were performed in emergency departments because of head injuries. Overall, 137 (26.2%) had an incidental

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

51

finding. The most common one was sinus opacification with an air fluid level, which resulted often – especially for young children – in treatment with antibiotics. In a further retrospective examination 225 conventional and functional brain MRI from neurological healthy children aged between 0 months and 18 years were processed. At this, incidental findings were made in 47 (21%) cases, the most frequent one being chronic and acute sinusitis (22 + 7 cases). Moreover, focal white matter lesion of uncertain etiology (5 cases), tonsillar ectopia (3 cases) arachnoidal cysts (3 cases), venous angioma (2 cases), mega cisterna magna (2 cases) as well as one case each of ventricular asymmetry, pineal cyst, hypoplasia pons, petrous apex lesions and cerebellar tonsil lesion uncertain etiology were described. The latter was the only case classified as “requires urgent clarification” (0.4%). Seventeen findings with uncertain health implications were also confirmed (8%). Coherences to age and gender were not found [42]. In the Philadelphia Neurodevelopment Cohort Study [43] based on the population, 1,400 brain MRIs with 3 Tesla MRIs were taken from children and adolescents aged from 8 to 23 years. One hundred forty-eight scans showed incidental findings (10.6%) that were categorized into the following entities: pineal cyst (34 cases), other cyst (19 cases), Cavum septum pellucidum (16 cases), other ventricular abnormalities (33 cases), vascular abnormalities (36 cases) and cerebellar abnormalities (33 cases). After expert assessment, 12 of the findings were referred to further clarification (0.8%). There was no coherence between incidental finding and age. Gender differences could be noticed only in an increased prevalence of cavum septum pellucidum in boys. Due to the different methodical approaches, target variables, devices, resolutions, picture section and emphasis, individual studies can only partially be compared or meta analyzed. The different kinds of reading procedures, the experience of the analyst as well as the criteria to define incidental findings also have to be considered. All in all it can be established that incidental findings are a very common phenomenon in imaging methods. Their potential increases with age, so that especially mature adults – quoting the described studies: every 2nd participant – have to expect an incidental finding. But for up to 25% of examined children and adolescents unexpected abnormalities were found. Frequency is dependent on the kind and localization of the imaging technique. Abdominal and thoracic scans via CT or MRI resulted in the most incidental findings, followed by brain MRI. Sonography, on the other hand, seemingly produces very few secondary findings. Considering the frequency of incidental findings in the described publications, their medical relevance according to the necessary interventions was comparatively low and was in most cases under 1%. In these cases however, the health risks indicated by the incidental finding were potentially dramatic. Another section that the scientific literature accredits with at least the same great potential for incidental findings is the genetic diagnostics. In years gone past, this area has notably developed rapidly so that numerous gene-illness-connotations exist. Unfortunately, a gene can be present in different variations. Only very few (pathological) variations end up in a specific illness, where other factors such as penetrance, expressiveness and methylation also play a role. It is decisively dependent on these factors,

52

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

despite the pathological potential of the variation, if and to which degree the illness will phenotypically be formed [44]. This multidimensional interference makes it particularly difficult to make a valid analysis regarding an unexpected result. Jamuar et al. [45] tried to get a little closer to the core of this topic by examining 56 defined genes of the blood of each 377 individuals. They found a total of 41,607 gene variations. A specific analysis of the variations regarding chromosomal coordinate(s), protein change, coding/non-coding regions, type of mutation (frame shift, nonsense, nonsynonymous, splicing, and synonymous) followed. As a result of this analysis it was found that 14 of the 41,607 variations were pathogenic. After a subsequent literature research, only 4 of the variations were rated pathogenic and 2 as likely pathogenic. In the end, the prevalence of real incidental findings in the examined cohort study was valued at 1.6%. At the same rate, as the gene diagnostic methods got improved, the cost of the analysis was reduced. This accelerated the trend to whole genome (next generation) sequencing in cohort studies in the last couple of years. The human DNA with around 23,000 genes implies an immensely high potential for unexpected findings [46]. From a scientific, ethical and legal point of view, overcoming such an estimation is currently one of the biggest challenges for genetic biobanks [25]. Apart from the modern imaging procedures and genetic analysis, the number of systematic investigations on incidental findings, especially on children and adolescents is very manageable. In the population-based German LIFE Child Cohort Study [47], such data regarding unexpected anomalies from other diagnostic procedures was scientifically processed. The assessment was a sub-sample from 969 healthy children aged from 3 months to 18.9 years. These children were run through an age-specific examination program. By far the most anomalies were found in the blood analytics. Age-specific measurements from up to 132 parameters took place (liver, kidney, heart, metabolism, bones, inflammation, thyroids, allergy- and hormone parameter, haemogram). Almost every single participant had at least one and the majority even more than one anomalies. How can this high number of findings be explained? As previously discussed, there is a huge shortfall in age-specific standards, especially for children. For the researcher every unexpected finding results in the challenge to judge it appropriately. Through an expertise process, only 2.2% of the findings in the described study were classed as acutely harmful and were reported back. A very similar result was shown in the urine analytic; of 48.2% suspicious findings, 0.75 were classed hazardous. These were most commonly symptom-free leukocyturias, proteinurias, and erythrocyturias. The third most commonly (22%) detected were incidental findings during sonographies of the kidneys, thyroids and brains (infants). Here almost a third (6.7%) were rated acutely dangerous. For other examinations, such as echocardiography, spirometry or bodily examinations, incidental findings became a lot rarer but more often they were then classified as acutely hazardous (Fig. 1). In 966 cases (99.7%), at least one anomaly was found. Of these cases, according to the assessment of the experts, 63 (6.5%) were rated to be acutely dangerous and were reported back to the parents with the recommendation to seek further advice. A

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

53

0.7%

Urinanalysis

48.2%

3.3% 4.5%

Electrocardiography

2.2%

Blood analysis

97.1%

6.7%

Ultrasound

22.7%

0% 0%

Spiroergometry

3.7% 4.3%

Cardiovascular echography Physical examination

1.5% 2.5%

Spirometry

1.5% 2.0% 0

20

40

60

80

100 %

Fig. 1. Frequency of incidental findings depending on the method of examination in comparison to the cases classified as relevant to health [20].

subsequent consultation with the paediatrician took place in 34 cases. For the remaining 29 cases, the parents stated that they could not see a reason for further investigation, since the child was symptom free. In 11 cases, the paediatrician could not make a diagnosis, as the anomaly could not be verified anymore. Eighteen children were diagnosed by their paediatrician but no intervention followed. Two children were treated with medication (one case of hypothyrosis and one case of chronic myeloid leukaemia) and in one case (hernia inguinal) an operation followed. Thus, for 3 participants (0.3%), a specific therapeutic consequence could be derived [20]. The example shows that there is a different potential for incidental findings depending on the different methods of examinations used. However, independent of the method of examination, it rarely results in the need for action. Rarer still is the need for acute therapeutic intervention. Nevertheless, for the individual case, these could be of life-saving consequences. The challenge for the PI therefore is to judge the unexpected finding appropriately to identify realistic dangers for health, development or fertility of the participant and protect him or her from the potentially negative consequences of a false positive result.

Possible Strategy for Dealing with Incidental Findings in Cohort Studies

After we have shown the general and also serious relevance incidental findings can have for both, the research project and the participant, the question that arises is how to strategically proceed in cohort studies. The challenge is to regard to perspectives of

54

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

Table 3. Conditions for successful incidental findings management Demands on part of the project – A principal investigator with sufficient incidental findings experience and medical knowledge to be able to judge emergencies or study doctor – Team of experts to estimate to medical relevance of the incidental finding depending on used methodology – Key holder Demands on part of the participant – Thorough informative consultation with the participant/legal guardian to inform of incidental findings, their risks and usefulness – Agreement to be informed of incidental findings in the informed consent – Expression of confidentiality towards the GP or pediatrician

the researcher, the clinical expert, the ethical advisory board, the politician for health and not least the perspective of the participant in equal measures. Hence, from a medical, ethical and research quality motivation a explicitly specific selection should be made to only report back findings to those probands for whom an acute danger for long lasting damage to their health, development or fertility exists. Scientific analysis and publications on this particular issue are distinctively rare. With the in the following designed model, we are referring to the experiences in the 2 cohort studies “SHIP” [8, 26] and “LIFE Child” [20, 47], both of which are intensively dedicated to the subject, actively applying incidental findings management and are scientifically supervising it [8, 20]. The model outlines one possible strategic of dealing with incidental findings. Albeit, the actions are extensive and require a few conditions (Table 3). The overall goal is to challenge the above-mentioned justice under special consideration for underage participants. In detail the model outlines the following actions. If an incidental finding occurs during the analysis of an examination result, an initial assessment by the principal investigator (PI) takes place to determine whether this is a life threatening result, for example, signs of an acute myocardial infarction on the ECG. In this case, the result would be reported back to the participant instantly with the strong recommendation to either see an expert or go straight to the nearest accident and emergency department – depending on the urgency of the situation. In extremely acute cases, the emergency doctor should be called for. These actions alone already require 2 conditions. First, the PI has to have sufficient medical knowledge to be able to judge the urgency of a result adequately. Alternatively, this job could be delegated to a study-doctor with appropriate qualifications. Second, the participant must have explicitly agreed in the informed consent that he be informed should a suspicious finding occur, as most legislations document the individual’s right to nescience. However, in case the result is not acutely life threatening, the PI assesses the finding once more with reference to his experience or by comparing it to a positive/negative list. Such a chart stems from precedents in other cohort studies – already evaluated incidental findings. When a finding coincides with a precedent on the

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

55

negative list, it is not a relevant result and the proband will not be informed. In longitudinal studies, those abnormal results can, where necessary, be pursued and assessed in the follow-ups. Though, when an incidental finding coincides with a precedent on the positive list, a relevant health risk can be assumed. The proband or his legal guardian will be informed with the recommendation to see his GP or paediatrician. Furthermore, the GP or paediatrician will be informed about the abnormal result by the study-doctor At this point, we never speak about a diagnosis but solely about a suspicious result. For one, it is not for the doctor to make such an assessment and for another it usually takes further medical examinations by the GP or paediatrician to verify and give a precise diagnosis. We also suggest that the proband or legal guardian be not informed directly about abnormal results. Experience shows that non-medical professionals often consult the Internet immediately, which in return almost always leads them to the worst diagnosis. This is hugely traumatizing for the families concerned. It is much more tactful to use sentences such as “do not worry for now, it is merely an abnormal result as part of a scientific study. Please see your GP/paediatrician. He will discuss and explain everything for you and if needed do further examinations.” The participant has to agree in the informed consent to be informed about these results too. To what extent a legal guardian has the right to nescience regarding results of the child in his care is legally not clarified. In this case, they could convene to contingency rights if there is a relevant health threat for an underage proband. However, because of the legal uncertainties, those probands who do not agree to be informed about suspicious findings will usually be excluded from cohort studies. In any case, a thorough informative consultation about the potential occurrence of incidental findings as well as the possible consequences for the participant and her family should take place beforehand (ref. Table 1). The informed consent should document these points in writing. A precise consultation can also reduce the fear of incidental findings [48]. Furthermore, a written release of confidentiality is needed to convey the abnormal result to the GP or paediatrician. This can also be arranged during the initial clarification. At best, the feedback of an incidental finding happens directly on the day of examination as long as the participant is still at hand. The data is commonly still unpseudomized at this point and the proband or legal guardian can be individually catered for. When the results are available only in the following days or assessed later, there is generally no access to the name of the participant anymore. The personal data is kept separately from the examination results. Then the key holder, who for his part has no access to the examination results and does not regularly belong to the research team, has to de-pseudomize (re-identify) the data. This ensures adequate anonymity. Should an incidental finding match neither the positive nor the negative list, it has to be classed as a new precedent. For this, an assessment by the expert according to the used methodology will follow. For an ECG examination, the expert would be, for example, a cardiologist, or for an MRI examination, the expert would be depending on the result, for example, a radiologist or neurologist or oncologist. Optimally an

56

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

8QH[SHFWHG LQFLGHQWDOILQGLQJ

Decision-making panal

Principal investigator

Principal investigator

,QWHUGLVFLSOLQDU\ clinical DGYLVRU\ERDUG

List of positives and negatives

Professional H[SHUWLVH SRVVLEO\PXOWLSOH

Ethnical DGYLVWRU\ERDUG

3ULQFLSDOLQYHVWLJDWRU Decision-making process

,PPHGLDWHH[SHUW assessment

not LQFOXGHG

no result

(WKQLFDOYDOLGD WLRQDQGVXSSRUW

3UHFHGHQW Result

Consequence

Acutely life threatening

1RWDFXWHO\ life threatening

‡ LQVWDQWDQHRXV  IHHGEDFNWRWKH  SUREDQGRUSDUHQW ‡ UHFRPPHQGDWLRQ  WRVHHVSHFLDOLVW  LPPHGLDWHO\ ‡ SRVVLEO\DPEXODQFH  HPHUJHQF\GRFWRU

1RVXVWDLQDEOH KD]DUGWREH H[SHFWHG

‡ QRIHHGEDFNWRWKH  SUREDQG ‡ SRVVLEO\QHZDV  VHVVPHQWDWQH[W  IROORZXSWHVWIRU  ORQJWHUPVWXGLHV

8UJHQWVXVSLFLRQ RIORQJODVWLQJ GDQJHU

‡ IHHGEDFNWRWKH  SDUWLFLSDQWOHJDOFDUHU ‡ UHFRPPHQGDWLRQWR  VHH*3RUSDHGLDWULFLDQ ‡ UHSRUWRIWKHLQFLGHQWDO  ILQGLQJWRWKH*3RU  SDHGLDWULFLDQ

Fig. 2. Flowchart of how to strategically deal with incidental findings in cohort studies.

interdisciplinary team makes the assessment. Alternatively, the experts can individually assess the result in a multi-stage process. Validations are often made on the basis of result constellations (for instance, blood results, ECG and MRI to assess a cardial MRI abnormality). This approach is particularly suitable for assessments and often obtains a high success rate [49]. The result is the estimation of the potential of a longlasting hazard for the participant. The result will be documented on the positive/ negative list and can then be used as reference for similar cases in the future. However, in some very rare cases, a distinct estimation is not possible, despite the multiple expert assessments. In this case, we recommend consulting an ethical advisory board to generate a non-medical evaluation of the relevance of potential consequences for the participant. In the LIFE Child Cohort Study, such an ethical advisory board consists of non-medical professionals such as vicars, lawyers, bankers, economy managers and other similar professionals [47]. These professionals also have to think analytically and make decisions for other people in their professional lives. Experience proves this to be particularly established for an ethical advisory board. The decision of the ethical advisory board subsequently results in another precedent and will be included in the positive/negative list. The clinical advisory board as well as the ethical advisory board work closely with the principal investigator and discuss their decisions with him. The principal investigator takes a central role on the whole model. Hence, he should possess a significant experience on the subject’s incidental findings. This

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

57

depends on resources as the entire experience process does not have to be run through once more. Exceptions of incidental findings are signs of violence or abuse on children. For this case, an individual strategy with a local child protection authority or the ethics commission should be stipulated prior to the start of the study. The model described here has empirically already been successfully tested and to the current status optimized [22]. Nevertheless, in individual cases, adjustments to the individual circumstances and available resources for the planned cohort study may be necessary.

References 1 Coughlin SS, Beauchamp TL (eds): Ethics and Epidemiology. New York, Oxford University Press, 1996. 2 Hoffmann M, Schmücker R: Die ethische Problematik der Zufallsbefunde in Populationsbasierten MRT-Studien. Münster, Westfälische WilhelmsUniversität, 2011. 3 Vara P, Niemineva K: Small-cystic degeneration of ovaries as an incidental finding in gynecological laparotomies. Acta Obstet Gynecol Scand 1951; 31: 94– 107. 4 Hertz M, Rubinstein ZJ, Shahin N, et al: Crossed renal ectopia: clinical and radiological findings in 22 cases. Clin Radiol 1977;28:339–344. 5 Weisberg LA: Incidental focal intracranial computed tomographic findings. J Neurol Neurosurg Psychiatry 1982;45:715–718. 6 Illes J, Rosen AC, Huang L, et al: Ethical consideration of incidental findings on adult brain MRI in research. Neurology 2004;62:888–890. 7 Shalowitz DI, Miller FG: Disclosing individual results of clinical research: implications of respect for participants. JAMA 2005;294:737–740. 8 Erdmann P: Zufallsbefunde aus Bildgebenden Verfahren in Populationsbasierter Forschung: Eine Empirisch-Ethische Untersuchung [Zugl.: Greifswald, Univ., Diss., 2014]. Münster, Mentis, 2015. 9 Wolf SM: Introduction: the challenge of incidental findings. J Law Med Ethics 2008;36:216–218. 10 Clarke AJ: Managing the ethical challenges of nextgeneration sequencing in genomic medicine. Br Med Bull 2014;111:17–30. 11 Enquete Kommission. Schlussbericht der EnqueteKommission Recht und Ethik der Modernen Medizin. Wiesbaden, VS Verlag für Sozialwissenschaften, 2002. 12 Kumra S, Ashtari M, Anderson B, et al: Ethical and practical considerations in the management of incidental findings in pediatric MRI studies. J Am Acad Child Adolesc Psychiatry 2006;45:1000–1006.

58

13 Presidential Commission for the Study of Bioethical Issues, ed. Anticipate and Communicate: Ethical Management of Incidental and Secondary Findings in the Clinical, Research, and Direct-to-Consumer Contexts. Washington, DC, http://www.bioethics.gov, 2013. 14 Stroud K, O’Doherty KC: Ethically sustainable governance in the biobanking of eggs and embryos for research. Monash Bioeth Rev 2015;33:277–294. 15 Berland LL: The American College of Radiology strategy for managing incidental findings on abdominal computed tomography. Radiol Clin North Am 2011;49:237–243. 16 Rangel EK: The management of incidental findings in neuro-imaging research: framework and recommendations. J Law Med Ethics 2010;38:117–126. 17 Dawber TR, Meadors GF, Moore FE Jr: Epidemiological approaches to heart disease: the Framingham Study. Am J Public Health Nations Health 1951; 41: 279–286. 18 Nybo Andersen AM, Casas M: Birthcohorts.net, 2017, http://www.birthcohorts.net/. 19 Kogevinas M, Vassilaki M: Enrieco: Environmental Health Risks in European Birth Cohorts, 2017, http://www.enrieco.org/ (accessed March 26, 2017). 20 Quante M, Bruckmann S, Wallborn T, et al: Managing incidental findings and disclosure of results in a paediatric research cohort – the LIFE Child Study cohort. J Pediatr Endocrinol Metab 2015;28:75–82. 21 Grimes DA, Schulz KF: Cohort studies: marching towards outcomes. Lancet 26;359:341–345. 22 Hiemisch A: Incidental Findings – Ethical Risk for Research or Harmless Side Effects? Leipzig, Lecture at the annual convention of the German Society of Pediatrics and Adoloscent Medicine (DGKJ), 2014. 23 Franke RH, Kaul JD: The hawthorne experiments: first statistical interpretation. Am Soc Rev 1978; 43: 623–643. 24 McCarney R, Warner J, Iliffe S, et al: The hawthorne effect: a randomised, controlled trial. BMC Med Res Methodol 2007;7:30.

Hiemisch · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

25 Fleming J, Critchley C, Otlowski M, et al: Attitudes of the general public towards the disclosure of individual research results and incidental findings from biobank genomic research in Australia. Intern Med J 2015;45:1274–1279. 26 Volzke H: Study of Health in Pomerania (SHIP). Concept, design and selected results]. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 2012;55:790–794. 27 Loeffler M, Engel C, Ahnert P, et al: The LIFE-AdultStudy: objectives and design of a population-based cohort study with 10,000 deeply phenotyped adults in Germany. BMC Public Health 2015;15:691. 28 Brähler E, Kiess W, Schubert C, Kiess J (eds): Gesund und Gebildet: Voraussetzungen für Eine Moderne Gesellschaft; mit 29 Tabellen. Göttingen, Vandenhoeck & Ruprecht, 2012. 29 Pearce MS, Salotti JA, Little MP, et al: Radiation exposure from CT scans in childhood and subsequent risk of leukaemia and brain tumours: a retrospective cohort study. Lancet 2012;380:499–505. 30 German National Cohort (GNC) Consortium: The German National Cohort: aims, study design and organization. Eur J Epidemiol 2014;29:371–382. 31 World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA 2013;310:2191–2194. 32 Kempe L: Bildgebende verfahren: bilder werden tiefer, schärfer, schneller. Dtsch Arztebl 2014;111:A-1914. 33 Wolbarst AB: Looking Within: How X-ray, CT, MRI, Ultrasound, and Other Medical Images are Created, and How They Help Physicians Save Lives. Berkeley, University of California Press, 1999. 34 Quattrocchi CC, Giona A, Di Martino AC, et al: Extra-spinal incidental findings at lumbar spine MRI in the general population: a large cohort study. Insights Imaging 2013;4:301–308. 35 Rutherford E, Weir-McCall JR, Patel RK, et al: Research cardiac magnetic resonance imaging in end stage renal disease – incidence, significance and implications of unexpected incidental findings. Eur Radiol 2017;27:315–324. 36 Vernooij MW, Ikram MA, Tanghe HL, et al: Incidental findings on brain MRI in the general population. N Engl J Med 2007;357:1821–1828. 37 Bos D, Poels MM, Adams HH, et al: Prevalence, clinical management, and natural course of incidental findings on brain MR images: the population-based Rotterdam Scan Study. Radiology 2016;281:507–515.

38 Orme NM, Fletcher JG, Siddiki HA, et al: Incidental findings in imaging research: evaluating incidence, benefit, and burden. Arch Intern Med 2010; 170: 1525–1532. 39 De Cocker LJ, Lindenholz A, Zwanenburg JJ, et al: Clinical vascular imaging in the brain at 7T. Neuroimage 2016;18. pii: S1053–S8119. 40 Hegenscheid K, Seipel R, Schmidt CO, et al: Potentially relevant incidental findings on research wholebody MRI in the general adult population: frequencies and management. Eur Radiol 2013;23:816–826. 41 Ortega HW, Vander Velden H, Reid S: Incidental findings on computed tomography scans in children with mild head trauma. Clin Pediatr (Phila) 2012;51: 872–876. 42 Kim BS, Illes J, Kaplan RT, et al: Incidental findings on pediatric MR images of the brain. AJNR Am J Neuroradiol 2002;23:1674–1674. 43 Gur RE, Kaltman D, Melhem ER, et al: Incidental findings in youths volunteering for brain MRI research. AJNR Am J Neuroradiol 2013;34;2021–2025. 44 Rosenberg LE, Rosenberg DD: Human Genes and Genomes: Science, Health, Society. London, Academic Press, 2012. 45 Jamuar SS, Kuan JL, Brett M, et al: Incidentalome from genomic sequencing: a barrier to personalized medicine? EBioMedicine 2016;5:211–216. 46 McLaughlin HM, Ceyhan-Birsoy O, Christensen KD, et al: A systematic approach to the reporting of medically relevant findings from whole genome sequencing. BMC Med Genet 2014;15:134. 47 Poulain T, Baber R, Vogel M, et al: The LIFE child study: a population-based perinatal and pediatric cohort in Germany. Eur J Epidemiol 2017;32:145–158. 48 Shaw RL, Senior C, Peel E, et al: Ethical issues in neuroimaging health research: an IPA study with research participants. J Health Psychol 2008; 13: 1051– 1059. 49 Phillips JP, Cole C, Gluck JP, et al: Stakeholder opinions and ethical perspectives support complete disclosure of incidental findings in MRI research. Ethics Behav 2015;25:332–350. 50 Dunet V, Schwitter J, Meuli R, et al: Incidental extracardiac findings on cardiac MR: systematic review and meta-analysis. J Magn Reson Imaging, DOI: 10.1002/jmri.25053. 51 Tan N, Charoensak A, Ajwichai K, et al: Prevalence of incidental findings on abdominal computed tomography angiograms on prospective renal donors. Tansplantation, DOI: 10.1097/TP.0000000000000486.

Andreas Hiemisch Hospital for Children and Adolescents Department of Women and Child Health University Hospitals, Leipzig University Liebigstraße 20a, DE–04103 Leipzig (Germany) E-Mail [email protected]

The Dilemma of Incidental Findings

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 41–59 (DOI: 10.1159/000481322)

59

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

Challenges and Opportunities in Conducting Research in Developing Countries M. Imran Khan · Zahid Ali Memon · Zulfiqar A. Bhutta Center of Excellence Women and Child Health, The Aga Khan University, Karachi, Pakistan

Abstract Developing countries disproportionally suffer from mortality and morbidity especially for children. Research leading to scientific progress in understanding causes of death has resulted in reduction in child morbidity and mortality in developed nations. However due to resource limitations, human and otherwise, in developing countries the progress towards achieving development goals, has been minimal. The collaborations and partnerships between developed and developing nations provides multiple opportunities for research and thereby learning, leading to reduction in child mortality. The scientists in both developed and developing nations need to understand the challenges that may occur while conducting research. In this paper we provide a framework for the researchers in both developing and developed nations to try to understand and develop research agenda and ideas that could not only address child health problems in developing countries but could also result in the © 2018 S. Karger AG, Basel progress of science benefitting the world at large.

Background

Child health continues to be a public health priority worldwide [1]. There is a reduction in the child mortality over time from 12.7 to 5.9 million between 1990 and 2015; yet every day 16,000 children die due to preventable causes [2]. A major proportion of child deaths are clustered in developing countries of Southern Asia and sub-Saharan Africa [3]. A significant reduction in mortality is achievable only if health systems are equipped with effective and proven interventions to improve child survival and health in developing countries [4]. Designing cost-effective and relevant public health program requires information that is based on research in developing countries [5, 6]. Child health has recently received substantial focus and prioritization in global development goals resulting in demand for evidence on effectiveness of preventive and curative strategies to reduce child deaths [7]. However, there are multiple reasons for

evidence generation not being a priority in developing countries, non-availability of research focused funding in low- and middle-income countries being one. Additional factors are the lack of research opportunities in developing countries [8]. Emphasis on use of evidence in health policy formulation to influence change or modification has increased the demand for generating evidence [9]. The disease burden in developing countries is high and research limited due to lack or inappropriate allocation of resources [10]. Only 27% of the studies that contribute to global knowledge on child health come from developing countries. In the presence of above-mentioned challenges, there are opportunities for scientists, both in developed and developing countries to collaborate with organizations, institutions and individuals to generate information that can be used for informing country and global child health strategies [11]. However, the mutual benefit should be the mainstay of institutional collaborations and factors such as confidentiality, publication and data rights, and conflict of interest should be defined and agreed upon even at the very beginning of the collaboration discussions. In the past four decades or so, the focus on child mortality and morbidity in developing countries has diverted some resources in the conduct of research in developing countries. This has largely resulted due to the realization that a disproportionate number of children is dying in the developing countries due to preventable causes and that uptake of many cost-effective interventions such as administration of vaccines is lagging behind [7, 10, 12]. Research has also been prioritized to understand the disparity in the availability and utilization of intervention packages between the developed and developing country populations [12].

Relevance of Research to Maternal and Child Health in Developing Countries

Statistics show that majority of deaths in developing countries in women and children are due to preventable causes [13]. The scale of these interventions must follow steps of effectiveness analysis, showing feasibility. Recently, the public health introduction of effective interventions has been significant, largely due to the research in the developing countries. The research must continue to keep the momentum of evidence based strategies. The children in developed countries have much lower risk of death compared to children born in developing countries; therefore, a need to better understand causes of death, and thereby reduce mortality could benefit both developed and developing countries. Information on predictors of mortality is critical in the design and conduct of public health interventions for equitable access to quality health care [3]. Generating evidence is a critical task, as this guides policy, but conduct of research in developing countries is not without challenges [15, 16]. The challenges associated with the conduct of research should not discourage researchers in exploring opportunities that are enormous in developing countries. This chapter focusses on

Conducting Research in Developing Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

61

opportunities and challenges in designing research studies in developing countries. The challenges and processes explained in this chapter may also be applicable to other areas of public health and medicine but is described here with the special emphasis on pediatrics and child health.

Prioritizing Research Areas and Topics

The extent to which a public health problem is prevalent in a population plays an important role in setting the stage for research. Information on distribution and determinants of a health condition is usually not available [17, 18]. Estimation of global burden of disease is an important scientific activity carried out by Institute for Health Metrics and Evaluation to guide international investments in prevention and management of childhood illnesses [19]. A routine disease surveillance system either does not exist or is dysfunctional in a majority of the developing countries. They collect information from various sources and then use modelling to estimate global burden. Therefore non-availability of disease burden information becomes the first and the foremost challenging factor prioritizing the health interventions and policies in lowand middle-income countries. Data on pneumonia, a major killer of children, for example, is sparse and out of 156 studies available globally, no study reported pneumococcal meningitis incidence for Southeast Asia [20]. Similarly, the global burden of typhoid is estimated on incidence data from 7 countries only. These estimates were then generalized to the rest of the world and regions [21]. In such situations, convincing the national authorities, as well as other agencies such as World Health Organization to take measures for prevention and control of typhoid has been challenging [22]. When it comes to decision making, case fatality rate, which is a key indicator of disease burden, is not available for all developing countries with the highest burden of diarrhoea and pneumonia [23]. As estimating burden of disease may sound easy, conducting these studies pose a great challenge to the scientists in terms of obtaining sufficient funding, finding suitable laboratory facilities, selecting an appropriate study population, and providing the right interpretation of these studies to a larger population. This problem can be overcome by relying on data from other sources such as national laboratory networks, hospitals with more robust data and/or triangulating information from small-scale research studies to model the burden of a problem in a specific country or region [24]. As important it is to have burden estimates, the public health policy requires information on the factors that predict and/or prevent mortality. Cross-sectional surveys, and case-control studies are the design of choice to assess burden and associated factors with a disease [25]. Investigators should be aware of the limitations and requirements at the planning phase so that data generated is robust and could represent the whole population. For example, if prevalence of stunting is to be measured at the national level, then to conduct a survey on a representative sample that covers all

62

Khan · Memon · Bhutta

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

Table 1. Research questions and the appropriate outcome for studies conducted in developing countries The question

What is the problem? What is the priority problem?

What are potential solutions? New technology, behaviour or policy change

Packaging the impact of the potential intervention

Piloting for scale up

Can the existing intervention be implemented at scale and through appropriate platforms? Non-inferiority studies

Study Distribution and outcomes determinants

Safety and efficacy Cost-effectiveness and sustainability

Safety surveillance and impact (economic and epidemiological)

Study design

Clinical trials (phase I, II, III)

Population surveillance Economic evaluation and implementation research

Descriptive epidemiology

Clinical trials (phase IIIb, IV)

geographies of the country may be very expensive. The investigator then has to look at alternative sampling strategies by not compromising data validity. A case-control study could also provide information on stunting by highlighting the risk groups within a population, and thus help in formulating design-targeted intervention strategies [26]. Operational or implementation research is being widely used to guide program implementation of the effective community and facility intervention related to newborn and child health [27]. This research constantly guides to achieve best outcomes. Randomized controlled trials provide evidence on the effectiveness of the intervention in improving child health outcomes [28]. On the whole, research prioritization should be based on the burden of disease and use of research in improving efficiency, equity and coverage of live-saving interventions [29]. Research opportunities, as mentioned previously, are enormous in developing countries [30]. Selection of an area of research is critical in conducting research [31]. Among many factors that guide prioritization, public health impact in terms of disease frequency, death as outcome and availability of effective interventions should be given more weightage. If a disease or a public health problem is worrisome for the health managers, and is minimally understood, then the research could be focussed to understand the factors associated with the distribution by conducting an observational study. Decision on the design and practicability of research is given in Table 1. Selection of study site for epidemiological studies, and especially trials, is of utmost significance [32]. Most often the investigators pick a study site with high burden of disease. From the point of view of logistic and other operational reasons, this may be the most practical approach; however, the results of the study are questioned as to the representative nature of the study population. Picking up a site that represents a country, region and population should follow a predefined selection criteria and should be discussed with local stakeholders. The key question should be to check if the data represents the population of interest and if the policy makers will use the information for decision making [33].

Conducting Research in Developing Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

63

The selection of site is critical in view of population dynamics and mobility, since a large number of studies are conducted in the urban low socio-economic settings. A study setting that has high migration (internal and external) makes a correct estimation of the population denominator difficult and thereby calculation of rates. A more stable population with minimal migration will have low loss to follow-up, and hence will give more precise estimates [34]. Many a times, studies also include a rural arm in the study, and the same study may give results on the burden and risk factors that are different from the urban arm, making it challenging to advocate for uptake of suggested intervention [35]. Prior to the design of the research, one must understand the regulatory framework of the country or the region where the research will be conducted [36]. There are 2 aspects for importance of regulatory bodies in the research; one, the researcher will have a better understanding of approval requirements and can prepare well; and second, the regulator will have an understanding of the need for research on the selected topic in the target country and future utility for scale up. Though this is a recently emerging phenomenon, many developing countries may not have very robust regulatory mechanisms. The approval processes most often rely on World Health Organization and other United Nations agency guidelines. An early discussion with the regulatory bodies at the local level will improve trust between the researcher and the regulator. The researcher must be aware of the regulatory approval/review timelines, processes, and required supporting documents. The researchers, therefore, should allocate sufficient time in the planning phase and in the actual research so that research timelines are not affected. A reputable local partner could be very helpful in ensuring smooth discussions and approval. The population is the centre of the focus in the conduct of a study [37]. The researchers should engage the community through representatives in discussing the need for the research and how the findings of the research will impact their health. Prior to the start of the study, the researcher must collect all relevant information about the study population, and conduct regular and repeated meetings to increase awareness about the purpose of the research. Often, this may need additional times for the research to be carried out particularly for longitudinal studies. This builds a trust between the researcher and population groups, resulting in higher recruitment and retention of study participants and thus production of valid and meaningful data [38]. The research should not only be based on scientific rigor but also be founded on ethical principles [39, 40]. The key areas of ethical significance in research in developing countries are the standard of care used in research; availability of interventions proven to be useful during research; and the quality of informed consent process. A usual argument in the discussions on ethics in research is that the developing country populations are different due to their background characteristics, and therefore the ethical standards that are followed in developed countries may not be applicable in developing countries. This has no relevance, since the ethical principles should be

64

Khan · Memon · Bhutta

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

emphasized more if the study population is not well equipped to understand research processes. Local capacity in bio-ethics plays a key role in bringing in participant rights, and ensure adherence to ethical principles while conducting research in developing countries. The researcher must engage with the local ethical review boards, understand their processes and address their concerns in a timely manner [41]. The international standard for the conduct of trails is guided by good clinical practices [42]. Interpretation of informed consent should be universal, irrespective of where the research is conducted [43]. Individual autonomy and therefore consent may vary across cultures. If permission from community leaders, elders or spouses is needed prior to individual consent, this must be sought. The participants, and for that matter the communities, may not understand research concepts such as placebo, randomization and vaccine failure and this can result in confusion between the research and therapeutic context [44]. The researcher should also learn local cultural contexts to develop informed consent processes especially for females and children [45]. The language used in the consent should be commonly understood and free from technical jargons. The research information in the study should be carefully reviewed at the stages of translation, adaptation and back translation so that the message and the meaning are consistent. In some communities, a written consent may not be acceptable and hence, the researcher must discuss with the ethical review board on the feasibility of the written informed consent and if verbal consent could suffice. If written consent is still required, then the researcher should discuss with the communities the importance and application of written consent [46]. A key component of consent – free willingness to participate – may be difficult to evaluate but should be the centre of focus. It is important that the investigator devises strategies to ensure that the field staff provide enough information to the participants and comply with the different aspects of the informed consent form [47]. Legal, political and social complexities in developing countries may arise in relation to confidentiality aspects. For example, groups involved in illegal commerce and immigrants may feel threatened if they provided medical and/or demographic information [46]. During the process of obtaining informed consent, the researcher should inform participants on the precautions that would be in place to protect confidentiality as well as any limitations to ensure confidentiality and possible adverse social and psychological consequences. Research should be guided by the principle of altruism, but economic or academic interests may influence the conduct of the research [48]. Many researchers use incentives to increase compliance and participation. There is a rich body of literature on the effect of incentives on the outcome due to the resultant selection bias. The investigator/researcher should be aware of the consequences of incentive provision. Many a times, the incentive is not directly offered as cash compensation for participation, but for the time that the participant spends in the study recruitment [49]. Ethical issues do not end with the project’s ethical approval. The researchers’ compliance with the ethical standards and guidelines is associated with their compliance with the ethical boards. Once a research is approved by the ethical board, the study

Conducting Research in Developing Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

65

teams do not comply with the remaining guidelines, for example, reporting deviation from the protocol. The researcher should be convinced that these guidelines are in place to protect the rights of the participants as well as the investigator and that there is an effect on the results of the study due to non-compliance [50]. Data management, intellectual property and data rights are an important component of scientific collaboration. The investigators should allocate sufficient time in the planning of the data management activities, and how the data will be shared between the partners [51]. Data management capacity in the developing countries may not be at the same level as that in developed countries, though in the past 20 years due to large multi-country projects and research studies, some institutions have developed data management skills [52]. It is important to develop mechanisms to enhance the social value of research through communication and advocacy [53]. Through collaborative partnerships, strategies should be devised to disseminate results in appropriate languages and formats to key stakeholders, including the local community, health policy makers, healthcare providers, and international health-care organizations for larger use of the effective interventions. While it is important to generate information that guides policy, adoption of research findings into policy or practice change for a larger public health good involves steps that are more than science and fall in the fields of communication and advocacy. Adoption of a robust research design in answering a question is critical; it is also very important to consider the suitability of a design in a specific environment [54]. The engagement of stakeholders particularly policy community from selection of research topic, study design and dissemination of the results will improve likelihood of use of research to inform decision making. Formulation of a technical advisory committee representing ministries of health, planning and development departments, academic institutions, technical agencies and funding agencies could lead to acceptability of results, avoid duplication and improve uptake [55].

Translating Evidence into User-Friendly Communication Product

As researchers, the emphasis is placed on the statistical significance in association of an intervention with the outcome. These pertain to the basic steps of assessing the effectiveness of the public health interventions. Figure 1 describes the interaction among key stakeholders involved in the generation and uptake of evidence in public health decision making. The presentation of information that is largely statistical becomes challenging in convincing individuals who are based on key positions of decision making [56]. Therefore, communication products such as policy briefs should be developed keeping in view different types of target audience such as politicians and bureaucrats. Their interest is mainly financial, and they may get the political mileage by introducing a new intervention. Therefore, construction of a laboratory for a ministry of

66

Khan · Memon · Bhutta

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

Policy maker

Researcher

Fig. 1. The interaction of multiple users of evidence and their interaction in advancing public health intervention.

Evidence generation and use

Implementer

Communities

health may be more rewarding in terms of votes compared to the introduction of new vaccines in the immunization program. Therefore, for researchers to have a sustained impact on health, it is important to understand the policy process in addition to understanding the mechanism to implement a public health intervention [57]. Usually research is conducted and published, but the very communities that participated in the research are unaware about the results of the study. It should be mandatory for the researchers to share the results with communities in the language that they easily understand [58]. Generating and developing local knowledge on child health issues require robust data [59]. Such data is generated through the conduct of well-designed epidemiological studies that require qualification in conduct of research studies. Until recently, such expertise was absent or minimal in developing countries [60]. The presence of local expertise in a developing country is important not only for highlighting and including local needs in research design but also for ensuring compliance with good clinical practices. An indirect benefit of collaborative research has been the research capacity improvement of the local investigators; however, this has also resulted in the brain drain of trained individuals to the more developed world organizations.

Conclusions and Recommendations

Though there are multiple challenges of conducting pediatric and child health research in developing countries, it also offers a number of opportunities of collaborative research for the mutual benefit of the scientific and policy communities in developed and developing countries to bring sustainable changes into the lives of children who die needlessly due to preventable causes. Going forward, the key area to focus is building capacity to generate evidence that can inform local decision making. Development of local expertise in the provision of health-care research should be an integral component of any proposed research.

Conducting Research in Developing Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

67

When externally sponsored research is proposed, which falls outside the national priorities, its relevance must be justified by academic research bodies and the appropriate research Ethics Committees. Countries should set national priorities related to the provision of health care and to enhance their capacity to conduct relevant research by following ethical standards in accordance to their needs. Scientists should establish an effective system for the ethical review of research, which includes the establishment and maintenance of research ethics committees, independent of the government and sponsors. National and international sponsors of research should ensure that adequate provision is made for training in the ethics of research for professionals involved in research related to health care, and that community benefits and beneficence are ensured. Scientific publications generate knowledge, but this should be translated to userfriendly communication products that are understood by communities, politicians and bureaucrats. The needs of target audience in terms of the use of research to develop communication products other than scientific papers should be prioritized.

References 1 Liu L, Oza S, Hogan D, Perin J, Rudan I, Lawn JE, et al: Global, regional, and national causes of child mortality in 2000–13, with projections to inform post-2015 priorities: an updated systematic analysis. Lancet 2015;385:430–440. 2 Khodaee GH, Khademi G, Saeidi M: Under-five mortality in the World (1900–2015). Int J Pediatr 2015;3:1093–1095. 3 Victora CG, Wagstaff A, Schellenberg JA, Gwatkin D, Claeson M, Habicht J: Applying an equity lens to child health and mortality: more of the same is not enough. Lancet 2003;362:233–241. 4 Requejo JH, Bhutta ZA: The post-2015 agenda: staying the course in maternal and child survival. Arch Dis Child 2015;100(suppl 1):S76–S81. 5 Altbach PG: Advancing the national and global knowledge economy: the role of research universities in developing countries. Studies High Educ 2013;38: 316–330. 6 Bhutta ZA, Das JK, Walker N, Rizvi A, Campbell H, Rudan I, et al: Interventions to address deaths from childhood pneumonia and diarrhoea equitably: what works and at what cost? Lancet 2013; 381: 1417–1429. 7 Ravishankar N, Gubbins P, Cooley RJ, Leach-Kemon K, Michaud CM, Jamison DT, et al: Financing of global health: tracking development assistance for health from 1990 to 2007. Lancet 2009; 373: 2113– 2124.

68

8 Gonzalez Block MA, Mills A: Assessing capacity for health policy and systems research in low and middle income countries. Health Res Policy Syst 2003; 1: 1. 9 Hanney SR, Gonzalez-Block MA, Buxton MJ, Kogan M: The utilisation of health research in policy-making: concepts, examples and methods of assessment. Health Res Policy Syst 2003;1:2. 10 Orne-Gliemann J, Perez F, Leroy V, Newell ML, Dabis F: [A decade of child health research in developing countries]. Sante 2003;13:69–75. 11 Busse H, Aboneh EA, Tefera G: Learning from developing countries in strengthening health systems: an evaluation of personal and professional impact among global health volunteers at Addis Ababa University’s Tikur Anbessa Specialized Hospital (Ethiopia). Global Health 2014;10:64. 12 Leroy JL, Habicht JP, Pelto G, Bertozzi SM: Current priorities in health research funding and lack of impact on the number of child deaths per year. Am J Public Health 2007;97:219–223. 13 Bhutta ZA, Das JK, Bahl R, Lawn JE, Salam RA, Paul VK, et al: Can available interventions end preventable deaths in mothers, newborn babies, and stillbirths, and at what cost? Lancet 2014; 384: 347– 370. 14 Bhutta ZA, Das JK, Rizvi A, Gaffey MF, Walker N, Horton S, et al: Evidence-based interventions for improvement of maternal and child nutrition: what can be done and at what cost? Lancet 2013;382:452– 477.

Khan · Memon · Bhutta

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

15 Vargas E, Becerril-Montekio V, Gonzalez-Block MÁ, Akweongo P, Hazel CN, Cuembelo Mde F, et al: Mapping the use of research to support strategies tackling maternal and child health inequities: evidence from six countries in Africa and Latin America. Health Res Policy Syst 2016;14:1. 16 Hirose A, Hall S, Memon Z, Hussein J: Bridging evidence, policy, and practice to strengthen health systems for improved maternal and newborn health in Pakistan. Health Res Policy Syst 2015;13:S47. 17 Bhutta ZA, Zipursky A, Wazny K, Levine MM, Black RE, Bassani DG, et al: Setting priorities for development of emerging interventions against childhood diarrhoea. J Glob Health 2013;3:010302. 18 Rudan I, Gibson JL, Ameratunga S, El Arifeen S, Bhutta ZA, Black M, et al: Setting priorities in global child health research investments: guidelines for implementation of CHNRI method. Croat Med J 2008; 49:720–733. 19 Blencowe H, Vos T, Lee AC, Philips R, Lozano R, Alvarado MR, et al: Estimates of neonatal morbidities and disabilities at regional and global levels for 2010: introduction, methods overview, and relevant findings from the Global Burden of Disease study. Pediatr Res 2013;74(suppl 1):4–16. 20 O’Brien KL, Wolfson LJ, Watt JP, Henkle E, DeloriaKnoll M, McCall N, et al: Burden of disease caused by Streptococcus pneumoniae in children younger than 5 years: global estimates. Lancet 2009;374:893– 902. 21 Crump JA, Ram PK, Gupta SK, Miller MA, Mintz ED: Part I. Analysis of data gaps pertaining to Salmonella enterica serotype Typhi infections in low and medium human development index countries, 1984–2005. Epidemiol Infect 2008;136:436–448. 22 Mogasale V, Maskery B, Ochiai RL, Lee JS, Mogasale VV, Ramani E, et al: Burden of typhoid fever in lowincome and middle-income countries: a systematic, literature-based update with risk-factor adjustment. Lancet Glob Health 2014;2:e570–e580. 23 Leung DT, Chisti MJ, Pavia AT: Prevention and control of childhood pneumonia and diarrhea. Pediatr Clin North Am 2016;63:67–79. 24 Sullivan M, Verhoosel J: Statistics: Informed Decisions Using Data. Pearson, 2013. 25 Tavazzi L: Observational research as a platform for evidence-based public health policies and learning health systems. Eur Heart J 2017;38:1891–1894. 26 Levy PS: Optimum allocation in stratified random network sampling for estimating the prevalence of attributes in rare populations. J Am Stat Assoc 1977; 72(360a):758–763. 27 Peters DH, Adam T, Alonge O, Agyepong IA, Tran N: Implementation research: what it is and how to do it. BMJ 2013;347:f6753.

28 Donner A, Klar N: Design and analysis of cluster randomization trials in health research. J Glob Health 2017;7:011003. 29 Atkins S, Marsden S, Diwan V, Zwarenstein M; ARCADE consortium: North-south collaboration and capacity development in global health research in low- and middle-income countries – the ARCADE projects. Glob Health Action 2016;9:30524. 30 Arora NK, Mohapatra A, Gopalan HS, Wazny K, Thavaraj V, Rasaily R, et al: Setting research priorities for maternal, newborn, child health and nutrition in India by engaging experts from 256 indigenous institutions contributing over 4,000 research ideas: a CHNRI exercise by ICMR and INCLEN. J Glob Health 2017;7:011003. 31 de Haan S, Kingamkono R, Tindamanyire N, Mshinda H, Makandi H, Tibazarwa F, et al: Setting research priorities across science, technology, and health sectors: the Tanzania experience. Health Res Policy Syst 2015;13:14. 32 Barker PM, Reid A, Schall MW: A framework for scaling up health interventions: lessons from largescale improvement initiatives in Africa. Implement Sci 2016;11:12. 33 Atkins D, Best D, Briss PA, Eccles M, Falck-Ytter Y, Flottorp S, et al: Grading quality of evidence and strength of recommendations. BMJ 2004;328:1490. 34 Sibanda EL, Weller IV, Hakim JG, Cowan FM: The magnitude of loss to follow-up of HIV-exposed infants along the prevention of mother-to-child HIV transmission continuum of care: a systematic review and meta-analysis. AIDS 2013;27:2787–2797. 35 Breiman RF, Cosmas L, Njuguna H, Audi A, Olack B, Ochieng JB, et al: Population-based incidence of typhoid fever in an urban informal settlement and a rural area in Kenya: implications for typhoid vaccine use in Africa. PloS One 2012;7:e29119. 36 Ndebele P, Blanchard-Horan C, Shahkolahi A, Sanne I: Regulatory challenges associated with conducting multicountry clinical trials in resource-limited settings. J Acquir Immune Defic Syndr 2014; 65(suppl 1):S29–S31. 37 Woolf SH, Zimmerman E, Haley A, Krist AH: Authentic engagement of patients and communities can transform research, practice, and policy. Health Aff (Millwood) 2016;35:590–594. 38 MacQueen KM, Bhan A, Frohlich J, Holzer J, Sugarman J: Evaluating community engagement in global health research: the need for metrics. BMC Med Ethics 2015;16:44. 39 Bhutta ZA, Offringa M: Standards of Research for Clinical Trials in Low-and Middle-Income Countries; Optimizing Treatment for Children in the Developing World. Springer, 2015, pp 143–158.

Conducting Research in Developing Countries

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

69

40 Bhutta ZA: Ethics in international health research: a perspective from the developing world. Bull World Health Organ 2002;80:114–120. 41 Acosta CJ, Galindo CM, Ochiai RL, Danovaro-Holliday MC, Laure-Page A, Thiem VD, et al: Implementation of good clinical practice guidelines in vaccine trials in developing countries. Vaccine 2007;25: 2852–2857. 42 Guideline IHT: Guideline for good clinical practice. 1996. ICH www.ich.org/MediaServer.jser. 2016. 43 Regmi PR, Aryal N, Kurmi O, Pant PR, van Teijlingen E, Wasti SP: Informed consent in health research: challenges and barriers in low‐and middle‐ income countries with specific reference to Nepal. Dev World Bioeth 2017;17:84–89. 44 Tam NT, Huy NT, Thoa le TB, Long NP, Trang NT, Hirayama K, et al: Participants’ understanding of informed consent in clinical trials over three decades: systematic review and meta-analysis. Bull World Health Organ 2015;93:186–198H. 45 Hyder AA, Rattani A, Krubiner C, Bachani AM, Tran NT: Ethical review of health systems research in low- and middle- income countries: a conceptual exploration. Am J Bioeth 2014;14:28–37. 46 Van Belle G, Mentzelopoulos SD, Aufderheide T, May S, Nichol G: International variation in policies and practices related to informed consent in acute cardiovascular research: results from a 44 country survey. Resuscitation 2015;91:76–83. 47 Kao C, Aranda S, Krishnasamy M, Hamilton B: Interventions to improve patient understanding of cancer clinical trial participation: a systematic review. Eur J Cancer Care (Engl) 2017;26. 48 Zumla A, Costello A: Ethics of healthcare research in developing countries. J R Soc Med 2002 06; 95: 275– 276. 49 Priebe S, Yeeles K, Bremner S, Lauber C, Eldridge S, Ashby D, et al: Effectiveness of financial incentives to improve adherence to maintenance treatment with antipsychotics: cluster randomised controlled trial. BMJ 2013;347:f5847. 50 Nichols AS: Research Ethics Committees (RECs)/Institutional Review Boards (IRBs) and the Globalization of Clinical Research: can ethical oversight of human subjects research be standardized. Wash U Global Stud L Rev 2016;15:351.

51 Harding A, Harper B, Stone D, O’Neill C, Berger P, Harris S, et al: Conducting research with tribal communities: sovereignty, ethics, and data-sharing issues. Environ Health Perspect 2012;120:6–10. 52 Memiah P, Ah Mu T, Penner J, Owour K, NgunuGituathi C, Prevot K, et al: Bridging the gap in implementation science: evaluating a capacity-building program in data management, analysis, utilization, and dissemination in low- and middle- income countries. Popul Health Manag 2017, Sep 8. 53 Imran Khan M, Freeman AJ, Gessner BD, Sahastrabuddhe S: The need for an information communication and advocacy strategy to guide a research agenda to address burden of invasive nontyphoidal salmonella infections in Africa. Clin Infect Dis 2015; 61(suppl 4):S380–S385. 54 Ioannidis JP, Greenland S, Hlatky MA, Khoury MJ, Macleod MR, Moher D, et al: Increasing value and reducing waste in research design, conduct, and analysis. Lancet 2014;383:166–175. 55 Jao I, Kombe F, Mwalukore S, Bull S, Parker M, Kamuya D, et al: Involving research stakeholders in developing policy on sharing public health research data in Kenya: views on fair process for informed consent, access oversight, and community engagement. J Empir Res Hum Res Ethics 2015;10:264–277. 56 Ayah R, Jessani N, Mafuta EM: Institutional capacity for health systems research in East and Central African schools of public health: knowledge translation and effective communication. Health Res Policy Syst 2014;12:20. 57 Cohen BE, Marshall SG: Does public health advocacy seek to redress health inequities? A scoping review. Health Soc Care Community 2017;25:309–328. 58 Hirose A, Hall S, Memon Z, Hussein J: Bridging evidence, policy, and practice to strengthen health systems for improved maternal and newborn health in Pakistan. Health Res Policy Syst 2015; 13(suppl 1): 47(2–7). 59 Schneider M, Sorsdahl K, Mayston R, Ahrens J, Chibanda D, Fekadu A, et al: Developing mental health research in sub-Saharan Africa: capacity building in the AFFIRM project. Glob Ment Health (Camb) 2016;3:e33. 60 Lansang MA, Dennis R: Building capacity in health research in the developing world. Bull World Health Organ 2004;82:764–770.

Dr. Zulfiqar A. Bhutta, Professor & Founding Director Centre of Excellence in Women and Child Health, The Aga Khan University Stadium Road Karachi (Pakistan) E-Mail [email protected]

70

Khan · Memon · Bhutta

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 60–70 (DOI: 10.1159/000481323)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

How to Recruit a Representative Sample and How to Look for It? R. Hoffmann · A. Gösswald · R. Houben · M. Lange · B.-M. Kurth Department of Epidemiology and Health Monitoring, Robert Koch Institute, Berlin, Germany

Abstract Representativeness can be considered a quality feature of surveys. The necessity and desirability of representativeness depend on the survey’s context and goals. The German Health Interview and Examination Survey for Children and Adolescents (KiGGS) is used as a practice-oriented example in this chapter in order to illustrate measures to achieve representative results. The KiGGS study is conducted by the Robert Koch Institute and it gathers data for the German health monitoring system. It aims at the provision of data in regular sequences, for health policy actors and the scientific and general public. Representativeness is required with regard to the socio-demographic characteristics of gender, age, region, and social status, which have an influence on the health status and risk factors in a nationwide perspective. Three major components of representativeness and their implementation within KiGGS are depicted: sampling, measures to recruit participants, adapting these measures continuously, and estimating and adjusting for people not taking part in the survey. The role and impact of representativeness are illustrated with the example of obesity in Germany. The practice-oriented approach is enriched by concise reviews of theory, discussed in survey research literature. Depicted survey design options should generally be checked for their availability, appropriateness, and affordability. © 2018 S. Karger AG, Basel

Introduction

National health surveys are performed with the aim to provide policy-makers with a reliable and clear picture of how health is distributed in a given population, and what indicators contribute to or reduce opportunities to be healthy. Repeating these surveys on a regular basis makes it possible to assess regional differences and developments over time. A well-established, continuous system of health monitoring opens up the prospect of recognizing changes reliably and comparatively early.

Box 1. KiGGS at a glance Component of the health monitoring system at the RKI Collection of information about health status, risk factors and health behavior of the German minor (0–17 years) population Cross-sectional information over 3 waves until now Key aspects of KiGGS are: General health indicators1 (mental health issues, subjective health, accidental injuries) Chronic diseases1 (bronchial asthma, hay fever, neurodermitis) Health factors1 (alcohol and tobacco consumption, physical activity, sports) Extensive medical and physical examinations2 (anthropometry, BIA, blood pressure, ergometry, accelerometry) Extensive laboratory diagnostic program2 (Collection of blood and urine samples) KiGGS baseline 2003–2006 Health examination survey and self-administered questionnaires (SAQ-paper) n = 17,641 KiGGS Wave 1 2009–2012 Follow-up, telephone survey n = 12,368 KiGGS Wave 2 2014–2017 (still ongoing) Follow-up, Health Examination Survey and Self-Administered Questionnaires (SAQ-paper) → AIM: the aim of the KiGGS study is to generate comprehensive and representative data on the health of the German population aged 0–17 years for the entire Federal Republic of Germany. 1

KiGGS baseline, KiGGS Wave 1 and KiGGS Wave 2. 2 KiGGS baseline and KiGGS Wave 2.

In order to perceive the required reliable data on the population-level, sampling methods and survey procedures have to focus on achieving a sample, which is representative for the population. This is true, if the distribution of men and women, young and old, wealthy and poor, ill and healthy in the sample correlates to that in the general population. Representativeness can be seen as a quality feature of surveys and depends on the appropriateness of methods and measures applied. Especially in surveys that are part of a health monitoring system and that contribute to data assessment for political decision making, representativeness is essential. The chapter addresses the selected measures to collect representative data for children and adolescents within the German Health Interview and Examination Survey for Children and Adolescents (KiGGS). KiGGS is part of the health monitoring system in Germany, conducted by the Robert Koch Institute (see box 1). The monitoring studies are financed by the German Federal Ministry of Health as well as by the Robert Koch Institute [1]. Data help to observe health and health behavior of the population living in Germany. Furthermore, the Robert Koch-Institute aims at detecting health

72

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

25

Low Middle High

Percent

20 14

15

12 9.8

10

6.3 5 0

4.4

3

3–6

3

1.3

7.5

5.9

7–10

3.6

11–13 Age groups

5.2

14–17

Fig. 1. Obesity by age and social status.

problems in an early stage, evaluating the results of interventions and to offer this information on national and international levels [1, 2]. One exemplary result of the KiGGS baseline study that has had consequences for health policy until now is the enormous increase in the prevalence of obesity in children and adolescents in a time period over the last 20 years. The main message was that 6.3% of our children and adolescents in Germany are obese (instead of 3%, 20 years before) [3]. The prevalence of obesity depends on age, gender, and on social status (Fig. 1). We use this example to illustrate the importance of representativeness. Selected components of representativeness and their implementation within KiGGS will be deepened: sampling, measures for recruiting participants, and readjusting measures to increase participation, continuously, to achieve a well-balanced participation in all groups. Furthermore, adjustment through weighting for people not taking part in the survey as a method to compensate for non-participation will also be depicted. Specific challenges in surveying children are constantly addressed. It is important to stress that the following examples from the KiGGS study are strongly based on its specific study characteristics. Some solutions might not be prudent or accessible for other specific populations or in other countries. Available options should generally be evaluated carefully for its appropriateness. Often, decisions also need to be balanced in terms of financial and timely affordability.

Ways to Achieve a Representative Gross Sample

In order to establish effective sampling procedures, the aim of a study has to be well defined at the beginning of the survey and must be practically feasible. The goals need to be specified with respect to a target population, an available sample

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

73

frame, an algorithm for the sampling of the gross sample, and procedures to ascertain eligibility of sample members. In the following paragraphs, these 4 steps will be addressed. The KiGGS target population consists of children aged 0 to 17 registered in Germany (see info box). As fieldwork for both the KiGGS baseline study (2003– 2006) and Wave 1 (2009–2012) unfolded for 3 years each, the target population was fixed to 2004 and 2010, respectively, to which the survey results refer [4]. By nature, the target population is time related and finite in size [5]. All members of the target population need to be accessible through the sample frame, that is, a list that contains all members of the target population. The sample frame for KiGGS was chosen from local population registers, where every resident living in Germany needs to be registered by law. Some countries do not have such registers and need to rely on other sources (e.g., lists of telephone numbers) [5]. Thus, both accessibility and expected quality of lists are crucial in the decision, which population list to use. A nationwide health survey cannot include the whole population due to financial and timely constraints. Therefore, a sampling procedure must be developed that allows the assumption to have information on “all” population members, although it is only gathered from “some” [6]. Very generally speaking, the term of representativeness claims that the distribution of specific characteristics is equal in the target population and the net sample [7] (Fig. 2). An important step to establish this connection is to choose the potential participants out of the defined target population at random [8]. Statistically speaking, prospective participants need to have a known and a non-zero probability to be chosen [5]. In total, these randomly selected prospective participants constitute the gross sample. However, this procedure involves a sampling error, but it is an inevitable and even deliberate component in sample surveys [5]. KiGGS follows a 2-stage sampling procedure in order to achieve randomness on the sampling level [9]. As a first step, communities are selected at random (Primary Sample Units) and as a second step, the prospective participants are sampled within these communities. This proceeding has different reasons. One is that Germany does not have a nationwide population register, which could serve as a sampling frame for a one-stage sampling procedure. In Germany, registries are organized on community level. Furthermore, the degree of urbanization and the region influence a person’s health. Living in big, medium-sized, or small cities, in villages on the country side, in the north or in the south, or in different federal states of Germany makes a difference to the health status of the people. These aspects should therefore be well represented in the selected gross sample. First, the communities were sampled according to the distribution of inhabitants throughout the country. Sampling was based on a list of German communities stratified according to districts and the BIK classification system, which takes into account the grade of urbanization, regional population density, and administrative borders. On the whole, 167 communities

74

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

German population under 18 years (approx. n = 17,400,000) Two-staged sampling

Gross sample (n = 26,787)

Measures to increase participation

Explanation Healthy Net sample (n = 17,641)

Disease 1 Disease 2

Fig. 2. Relationship of target population, gross sample, and net sample and measures adopted to achieve a representative net sample for KiGGS.

were sampled (Fig. 3). The sampling procedure of the communities follows a protocol developed in cooperation with the Leibniz Institute for the Social Sciences (GESIS), Mannheim, Germany and is explained in detail elsewhere [10]. The general idea is that the probability of selecting a community for the sampling frame is proportional to the number of inhabitants living in communities of the same community type. For example, given that in 2004, the minor population in Germany comprised approximately 14.7 million children, if 200,000 children lived in communities of a certain type, each community of this type would have had a chance of 2.3% to be sampled (167*200,000/14,700,000). By choosing this sampling procedure, each prospective participant of a sampled community has a calculable chance to be drawn from the local registry and to be

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

75

Primary sample units

Fig. 3. Sample communities in KiGGS.

member of the gross sample. This chance is independent of any health-related features such as the social status or health behavior and health status themselves. Under these circumstances, the distribution of the characteristics in the gross sample might coincide with their distribution in the target population (e.g., for obesity). In order to illustrate the importance of this independent probability to be selected for the composition of the gross sample, a convenience sample may serve as a counterexample. For instance, a survey conductor promotes a health survey by displaying posters in public transport and offers a certain amount of money as an incentive. The gross sample cannot be described precisely in this case. The description has to stay vague: it could be that all persons who use public transport or all persons who notify the announcement or were told about it. The probability to be a member of this sample is not at random and not calculable. Even the fact that someone uses public transport depends on many different aspects: the person’s social situation, his or her age, the health status, the attitude toward matters of the

76

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

environment, having a driver’s license living in a city or not, and so on. The probability that someone takes notice of the announcement might even depend on further aspects, for instance, the offer of receiving a reward for participation. So we could not claim representativeness of this convenience sample for the whole target population. Every population list faces methodological limitations, which have to be taken into account. Coverage error may arise when the target population and the sample frame do not match. Undercoverage occurs if population members are falsely not part of the sample frame. Due to organizational considerations, addresses for KiGGS are always selected several weeks prior to the examination date [10]. Hence, only infants aged 2 months and older could be included in the gross sample. This leads to the undercoverage of the youngest infants. Overcoverage is given if persons are listed in the sample frame although they should not be listed. This may be caused by administrative restrictions, for example, persons moved to another town, but this fact is not yet recorded in the registers. Persons could also leave the country of interest, outgrow the age range or die between the drawing of the list, sampling, and the contact attempt. These persons cannot be included into the net sample; their inclusion probability into the gross sample cannot be calculated. Such cases are classified as noneligible. They can be subtracted from the gross sample without influencing the survey process with regards to representativeness. Exclusion could also stem from insufficient understanding of the survey’s intention or the incapability of providing information. On the one hand, it is necessary to exclude these persons due to the feasibility of a survey (study personnel cannot be fluid in all languages). On the other hand, there is risk of systematically biasing the gross sample. This would occur if language barriers or limited mental capacity of prospective participants would cause the exclusion and these characteristics are connected to a specific health status. Higher prevalence of obesity among migrants is observable. If the migrant population would have been underestimated, the overall prevalence would have been misrepresented. Criteria of eligibility must be defined precisely before the survey start and have to be described in detail in the study protocol. Exclusion of the prospective participants according to these criteria is conducted during the process of recruitment and should finally lead to a more “realistic” adjusted gross sample. Summarizing Info Box Many measures can and should be taken into account in order to achieve representativeness, even before participants answer questionnaires, take part in health examinations, or refuse to take part. For gaining a representative gross sample, a key component in sampling is randomness. It requires an accurate definition of the population of interest, a list that contains the population members, a protocol of how to select members from the list, and procedures of how to cope with deficiencies of these lists. Similar to the nature of the target population, population lists merely pretend

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

77

stability where, in practice, constant change is real. Nevertheless, with great leverage in the hands of the researcher, the sampling procedure can lead to a good basis for a representative net sample.

Measures to Achieve Representative Response

Throughout the last few years, the undertaking to recruit survey participants has become increasingly difficult [11]. With fewer persons taking part in a survey, the risk increases that participants do not represent the target population’s diversity. Hence, both the number of nonparticipants and the systematic differences between them and persons taking part in the survey are important. It is the goal to implement measures, which attract a heterogeneous group of prospective participants [12]. Thereby, after acknowledging the varying characteristics of places of residence in Germany (e.g., the grade of urbanization) in the sampling procedure, the focus is on the individual characteristics of age, gender, and the social status of prospective participants. The health status itself should not be the underlying reason for making the decision whether to participate or not. Though health is a very sensitive topic and often connected to cultural attributions. For instance, when obese people feel stigmatized by public debate about obesity or inappropriate wording in survey information material, they might refrain from participation. Cooperation in Theory There are comprehensive theories why people decide in favor of participation in a survey or against it [6]. The aspect of social exchange plays a central role in these theories. It is crucial for the survey conductor to establish trust that those benefits anticipated by the prospective participant will be realized [6]. The cost-benefit theory stresses the need to make benefits salient to the participant and that the benefits outweigh the costs [13]. The leverage-saliency theory resembles this approach. It emphasizes that the effect of survey features may vary between prospective participants [13, 14]. KiGGS in Practice on the Societal and Community Levels A comprehensive persuasion strategy should address different levels of communication with prospective participants and thereby use different modes with the overall goal to establish trust [6, 15]. On the macro level, the target population has to be informed about the scope and aims of the study. Various public relation measures can be applied to reach different target groups. For KiGGS, a website (http://www.kiggsstudie.de/english/home.html) provides general information to the interested public and specific information for prospective participants. Press releases for regional media inform the local population of a sampled community about the aims of the study

78

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

and announce the fact that the study will be conducted in their region. Specifically addressed information letters are sent to mayors, local health authorities, and pediatricians asking them to support the study. These and other community representatives (particularly of the migrant population) are known to have relevant influence on the decision making aspects of community members [9, 16]. An additional measure is the offer of a telephone and e-mail information service carried out by personnel well trained in communication. Overall, it is the aim to provide prospective participants with as much information as they require. Any restraints and insecurities toward the study aims and the study itself should be decreased and the public interest of the survey should be fortified [6, 8, 15]. KiGGS in Practice on the Individual Level On individual level, every prospective participant may have his or her own rationale for (non-)participation. Among others, common reasons for nonparticipation are that persons from the gross sample could not be contacted, prospective participants lack time, do not value the study, or are temporarily absent [10]. Strategies to convince prospective participants should focus on these and other reasons. As mentioned earlier, different subgroups are likely to be convinced by different reasons. Marketing theories teach us that boys and girls, younger and older children are attracted by different types of promotion and adults are responsive to other messages than their children. Usage of Different Modes in Approaching Sampled Individuals For KiGGS, prospective participants are invited with a short easily understandable invitation letter and an age-specific information brochure designed by graphic artists. This brochure contains detailed information on the study and aspects of data protection in the form of several very short articles accompanied by pictures of children or adolescents of different social and ethnic groups. The importance of objective data on the population’s health and health behavior for health policies and resource allocation is pointed out. Further, the individual benefits of participating are depicted; at least, minimization of any potential burden is intended. A key advantage of postal invitations is that they can be handled individually and no pressure can be laid on prospective participants. In case people do not react either to the invitation or to the postal reminder 2 weeks later, specially trained personnel try to contact sampled persons by phone calls (if phone numbers are available) and/or house visits. So far, nonparticipation could be unintentional, for example, due to missing literacy, and therefore other measures different from nonpersonal measures need to be invested during the contact phase [15]. These employees are trained in bringing forward arguments fitting for the specific individual, with regard to the individuals’ needs and preferences. General guidelines of communication and a catalogue of arguments and procedures to answer frequently asked questions were developed for KiGGS; these documents are constantly

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

79

revised and adapted. This catalogue tries to anticipate all sorts of concerns. It deals among others with insecurities about data privacy and potential impacts of survey results. For intensified personal measures, a contact guideline defines a certain number of phone calls, different call times during the day, and a certain number of house visits as well. By applying this method, the effort to contact and to convince each single prospective participant is designed as similar as possible and therefore standardized and comparable. The exact documentation of these contact attempts allows evaluating the effectiveness of these personal and time-intensive measures. For instance, if certain time ranges are identified in which contact chances are the best – maybe even varying for different subgroups – personnel might be allocated according to these results. The experience in the KiGGS baseline study showed that the above-described intensified measures lead to an increase of participation by 13% [16]. Benefits and Burden of Physical Examinations In the best case, examinations are interesting, challenging and fun, especially for young participants. Well-trained study nurses or physicians can help to deal with children’s fear of collecting blood samples. Especially for parents, the report on results of the different health examinations and selected laboratory parameters is a strong incentive. The duration of health examinations should not be too long [17]; an acceptable time frame would be 2–3 h for children. Time scheduling for examinations has to be adapted to time-use patterns of parents, children, and adolescents [16]. Appointments should also be offered after school and parenteral work on labor days and during weekend days, preferably on Saturdays. Incentives – Implementing a Specific Measure A common way to trigger participation and to communicate the appraisal to participants is the provision of incentives [18]. Incentives can occur in various forms according to their interpretation (equity vs. inequity), what to give (monetary vs. nonmonetary), when to give (prior or “promised”), which value the incentive should have and if different choices might be offered (different subgroups might feel attracted by different incentives) [19]. As such, incentives exemplarily stand for the role of social exchange. For examinations, different “promised” incentives according to age groups are in use during KiGGS. Children up to 10 years are offered small gifts, whereas adolescents receive a same value cash incentive. Similar response quotas across age groups in the KiGGS baseline study (data not shown) may also be partly attributable to the usage of incentives [10, 16]. Migrants – Addressing a Specific Subgroup In KiGGS, several measures have been implemented in order to include as many potential participants as possible, with a special focus on migrant groups. In the KiGGS baseline study, invitation and information material and also question-

80

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

naires were provided in 6 different languages. Study personnel are trained to apply the same standards in evaluating language proficiencies because a fair command of German was still considered essential in order to assure informed consent of participation. Mixed-Mode Approach in Questionnaires Methods of data collection are known to influence the willingness to participate. These methods should be as attractive and convenient as possible. Questionnaires should be designed in a clear, easy-to-read, and appealing manner, easily understandable with focus on topics that appear interesting to the participants, and they should not be too long. Different modes of questionnaire presentation can be offered to try to attract different subgroups of the population. Paper and web mode are both selfadministered by the participants and therefore preferable, taking the sensitivity of the health topic into account. In offering different response mode options concurrently, potential limitations of using only a single mode can ideally be amortized and lead to a mode fitting for the specific individual, with regard to its needs and preference. However, practical feasibility and financial affordability might stand in the way to conduct a survey in a mixed-mode design [15]. Summarizing Info Box Nationwide population-based surveys are characterized by the enormous diversity of their participants. Individual perceptions, needs, and concerns might influence the decision of whether to participate or not. Established trust is therefore crucial for any attempt to obtain cooperation. A mix of communication modes often helps, also to assure and support eligibility and to inform prospective participants about the survey. Therefore, the survey topic and mode of data collection are also crucial elements when thinking about survey design. A tailored approach [6] should be guided by socio-demographic characteristics and should always be preferred over a “one size fits all” approach. Importantly, the measures adopted should be disproportionately attractive between groups who have different likelihoods of participating [12].

Assessment of and Adjustment for Representativeness

After data collection is complete, it is advisable to evaluate the quality of the net sample [20], that is, to estimate whether participants represent the target population in terms of the health status and other specified characteristics like the social status, gender, and age. Implicitly, it is assumed that the net sample could lack representativeness, whereas the results should be representative. This gap crucially stems from potential limitations in building the gross sample and determination of eligibility of its members and the lack of participation.

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

81

Basic Question Approach – Further Measures against the Unknown Prospective participants who did not respond could be systematically different from actual participants. The enormous range of individuality in the target population would be narrowed to uniformity in very few population groups. In order to rule out this worst case scenario, nonparticipants are asked to answer a short questionnaire with 10–15 key indicators of interest, for example, self-measured body weight and height. Possibly against intuition, this approach is common in survey research and often nonparticipants cooperate. In the KiGGS baseline study, the short questionnaire was answered by 22% of the adjusted gross sample, so information on 89% of the adjusted gross sample was collected. Participants and nonparticipants gave fairly equal answers with regard to health information (e.g., body-mass-index) but differed on the maternal educational status [10]. Besides financial affordability, the approach is affected by the vagueness about the characteristics of the remaining 11% of the adjusted gross sample [20]. However, this is a small proportion both in absolute and relative terms. It is only one third of the amount of non-participants for which no information was available originally. Use of Sample Frame Information – A Source for Knowledge Population registries, which contain the addresses of all prospective participants, including nonparticipants, also contain information about age, gender, place of  residents, grade of urbanization, and citizenship. In other countries, even more information, for example, on occupation and educational status, is available. Nevertheless, group-specific response quotas could be calculated. In the KiGGS baseline study, these were similar across genders and age groups, but differed for children and adolescents with German (68%) and non-German citizenship (51%) [10]. At the outset, this may indicate necessity for intervention in order to  incorporate more participants of non-German citizenship living in Germany.  Yet in terms of  representativeness, misspecification of results would occur only when systematic differences between participants and nonparticipants were given. Adjustment through Weighting In the KiGGS baseline study, survey results collected from 17,641 participants should be representative for the German population under 18 years, which is roughly about 14.7 million (Fig. 2). Each participant acts in place of population members who are not included in the survey. Weighting is the process of establishing the individual factor for each participant with regards to characteristics known from population distributions. Both sampling criteria (see Ways to achieve a representative gross sample) and the personal characteristics of gender, citizenship, and educational status were regarded in the weighting procedure. For the KiGGS baseline study, the Microcensus served as external benchmark. It is an obligatory, yearly sample survey, commissioned by the Federal Statistical Office covering roughly 1%

82

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

of the population in Germany [21]. A central prerequisite of weighting is that it requires homogeneous subgroups in order to contribute to representativeness and that the incorporated characteristics are connected to the health status. At last, representative results are challenging to provide yet essential for policy advice. The inclusion of risk groups is of tremendous importance. Results different from representative would not detect risk groups and therefore mislead valuable resources in intervention. Summarizing Info Box Careful set-up of sampling, implementation of measures to obtain cooperation by prospective participants, and flexibility in adopting these measures at short notice do not liberate from comprehensive investigation of and correction for nonparticipation. Instead of being thoroughly skeptical about the characteristics of nonparticipants, different options are available to further investigate representativeness. Weighting can correct for the sampling scheme and bias in participation.

References 1 Kurth BM, Lange C, Kamtsiuris P, Hölling H: Gesundheitsmonitoring am Robert Koch-Institut. Sachstand und Perspektiven. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2009;52:557–570. 2 Kurth B-M, Ziese T, Tiemann F: Gesundheitsmonitoring auf Bundesebene. Ansätze und Perspektiven. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2005;48:261–272. 3 Kurth B-M, Ellert U: Perceived or true obesity: which causes more suffering in adolescents? Findings of the German Health Interview and Examination Survey for Children and Adolescents (KiGGS). Dtsch Arztebl Int 2008;105:406–412. 4 Lange M, Butschalowsky HG, Jentsch F, Kuhnert R, Schaffrath Rosario A, Schlaud M, Kamtsiuris P: Die KiGGS-Folgebefragung – KiGGS Welle 1. Feldarbeit, Stichprobendesign, Response, Gewichtung und Repräsentativität. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2014;57:747–761. 5 Groves RM, Fowler FJ Jr, Couper MP, Lepkowski JM, Singer E, Tourangeau R: Survey Methodoloy. Wiley, 2004. 6 Dillman DA, Smyth JD, Christian LM: Internet, Phone, Mail, and Mixed-Mode Surveys: The Tailored Design Method. Hoboken, NJ, John Wiley & Sons, 2014. 7 Kruskal W, Mosteller F: Representative sampling, I: non-scientific literature. Int Stat Rev 1979:13–24.

8 Latza U, Stang A, Bergmann M, Kroke A, Sauer S, Holle R, Kamtsiuris P, Terschüren C, Hoffmann W: Zum problem der response in epidemiologischen studien in Deutschland (Teil I). Gesundheitswesen 2004;67:326–336. 9 Kurth BM, Kamtsiuris P, Hölling H, Schlaud M, Dölle R, Ellert U, Kahl H, Knopf H, Lange M, Mensink GB: The challenge of comprehensively mapping children’s health in a nation-wide health survey: design of the German KiGGS-Study. BMC Public Health 2008;8:196. 10 Kamtsiuris P, Lange M, Schaffrath Rosario A: Der Kinder- und Jugendgesundheitssurvey (KiGGS): stichprobendesign, response und nonresponse-analyse. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2007;50:547–556. 11 Galea S, Tracy M: Participation rates in epidemiologic studies. Ann Epidemiol 2007;17:643–653. 12 Groves RM: Nonresponse rates and nonresponse bias in household surveys. Public Opin Q 2006; 70: 646–675. 13 Singer E: Toward a benefit-cost theory of survey participation: evidence, further tests, and implications. J Off Stat 2011;27:379. 14 Groves RM, Singer E, Corning A: Leverage-saliency theory of survey participation: description and an illustration. Public Opin Q 2000;64:299–308. 15 de Leeuw ED: To mix or not to mix data collection modes in surveys. J Off Stat 2005;21:233–255.

How to Recruit a Representative Sample and How to Look for It?

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

83

16 Hölling H, Kamtsiuris P, Lange M, Thierfelder W, Thamm M, Schlack R: Der Kinder- und Jugendgesundheitssurvey (KiGGS): Studienmanagement und Durchführung der Feldarbeit. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2007;50:557– 566. 17 Hölling H, Schlack R, Kamtsiuris P, Butschalowsky H, Schlaud M, Kurth BM: Die KiGGS-Studie. Bundesweit repräsentative Längs- und Querschnittstudie zur Gesundheit von Kindern und Jugendlichen im Rahmen des Gesundheitsmonitorings am Robert Koch-Institut. Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 2012; 55: 836– 842.

18 Hoffmann W, Terschüren C, Holle R, Kamtsiuris P, Bergmann M, Kroke A, Sauer S, Stang A, Latza U: Zum problem der response in epidemiologischen studien in Deutschland (Teil II). Gesundheitswesen 2004;66:482–491. 19 Singer E, van Hoewyk J, Gebler N, Raghunathan T, McGonagle K: The effect of incentives on response rates in interviewer-mediated surveys. J Off Stat 1999;15:217–230. 20 Lynn P: The problem of nonresponse; in Leeuw Edith de, Hox Joop, Dillman Don A (eds): International Handbook of Survey Methodology. New York, Lawrence Erlbaum Associates, 2008, pp 35–55. 21 Statistisches Bundesamt: Leben in Deutschland – Ergebnisse des Mikrozensus, Wiesbaden, 2005.

R. Hoffmann Department of Epidemiology and Health Monitoring, Robert Koch Institute General-Pape-Strasse 62-66 DE–12101 Berlin (Germany) E-Mail [email protected]

84

Hoffmann · Gösswald · Houben · Lange · Kurth

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 71–84 (DOI: 10.1159/000481324)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

The Epidemiology of Global Child Health Lars Åke Persson Department of Disease Control, Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, UK

Abstract Over the past 25 years of the Millennium Development Goal (MDG) era, the global figure of mortality in children under the age of 5 has dropped from 90 to 43 per 1,000 live births. The progress has been slower in terms of improvement in neonatal survival, reduction of stillbirths, and improvement in nutritional status of children. A wide range of conditions from global politics to immediate care in the household determines child health and survival. The successful efforts in recent decades to reduce mortality need to be reinforced by a stronger emphasis on geographic, social and gender equity in child health and survival. Adequate data is needed to set priorities, to monitor progress and to mobilise resources. The repeated Demographic Health Surveys and the Health and Demographic Surveillance Systems in low- and middle-income countries satisfy some of those needs. The MDGs have now been replaced by the Sustainable Development Goals that emphasise prevention and broad multi-sectorial efforts to alleviate poverty and improve health and welfare. Those goals may potentially provide direction and strengthen the commitment in response to the huge threats to global child health that war, social unrest, migration and climate change constitute. © 2018 S. Karger AG, Basel

The Child Survival Revolution

During the last 25 years, the Millennium Development Goals have emphasized global development, poverty alleviation and investment in maternal and child health. There have been major achievements. Extreme poverty has been halved, and global mortality in children under the age of 5 has been reduced from 90 per 1,000 live births in 1990 to 43 per 1,000 in 2015 (Fig.  1) [1, 2]. This corresponds to a drop in the

Fig. 1. Global under-5 mortality rate (deaths 0–5 years per 1,000 live births, above) and neonatal mortality rate (deaths 0–28 days per 1,000 live births, below) 1990–2015.

Deaths per 1,000 live births

100 90 80 70 60 50 40 30 20 10 0 1990

1995

2000

2005

2010

2015

Year

absolute number of under-5 deaths from 12.7 million in 1990 to below 6 million in 2015. While Sub-Saharan Africa still has the highest level of under-5 mortality rate, this region has also had the highest absolute decline in child deaths. A recent analysis of global health (1990–2015) could not demonstrate any change in total disability-adjusted life-years, a summary measure of premature mortality and non-fatal health loss [3]. However, behind these overall constant figures there was a decline in disability-adjusted life-years due to communicable, maternal, nutritional and neonatal causes and an increase in the non-communicable diseases. Changes in the levels of neonatal mortality have been less dramatic than the decrease in under-5 deaths. However, there has been a reduction in deaths during the first 28 days of life from 36 to 19 per 1,000 live births (Fig. 1) [4]. In the global advocacy for improved child survival after the millennium shift, the emphasis was mainly given to the post-neonatal and child health problems [5]. A series of cost-effective, evidence-based interventions had the potential especially to lower the number of deaths in infectious diseases. A series of vertical programs had promoted immunisations, management of pneumonia and oral rehydration therapy to children with diarrhoeal diseases [6]. With the decrease of the post-neonatal mortality (1–12 months of age) and child mortality (12–60 months), an increased proportion of the under-5 deaths occurred in the first month of life. Consequently, more emphasis has later been given to the survival of the newborns [7] and strategies to prevent these early deaths by adequate maternal and newborn care [8]. During the last few years, even the huge problem of preventable fresh stillbirths in low-income settings has been added to the global health agenda [9]. The perinatal health problems cannot be managed just by isolated vertical approaches. They require health system approaches with good coverage, quality and continuity [10]. Globally, in 1990, 40% of children under the age of 5 was stunted (height for age ≤2 SD of the World Health Organization reference). This decreased to around 25% in

86

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

300

Deaths per 1,000

250 200 150 100 50 0 1990 1995 2000 2005 2010 Year

Fig. 2. Rwanda had an under-5 mortality level around 80 per 1,000 live births in 1990 (upper line in the graph). The unrest in society pushed this upward, and at the time of the 1994 genocide, this level had been tripled [14]. Efforts to improve security and international assistance reduced the mortality in the post-genocide period. A second peak coincided with a reduction of international support and a temporary re-introduction of user fees in the health services. Later, increased coverage of essential maternal and child health services and targeted interventions to vulnerable groups combined with development investments in other sectors of society resulted in an under-5 mortality of around 50/1,000 in 2010. The neonatal mortality was less influenced by the dramatic developments in the country (lower line in graph). Data are based on 3 consecutive Demographic Health Surveys [14].

2015 [11]. High prevalence of stunting is especially found in South Asia and Sub-Saharan Africa. The timing of onset of linear growth restriction is different in these two geographical areas; in South Asia, maternal undernutrition and other unfavourable conditions cause growth restriction already in foetal life, and the child is born with a low birth weight [12]. In Sub-Saharan Africa, a low birth weight is less common, but children tend to slow down in growth when the period of exclusive breastfeeding ends. In 2011, stunted growth affected at least 165 million children under the age of 5 and at least 52 million children were wasted (weight for height ≤2 SD of the World Health Organization reference) [13].

Determinants of Child Health and Survival

The changes in child mortality in low- and middle-income settings reflect the political, economic and social developments in society. An example from Rwanda is given in Figure 2, where the social unrest, war and genocide of the 1990s, and the later impressive efforts to rebuild society are visualized in the dramatic mortality curves of children [14]. A wide range of conditions influences child health, nutrition and mortality. UNICEF developed a conceptual framework for causes of child malnutrition in the

Global Child Health

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

87

Disease, disability, death

Long-term consequences Undernutrition

Poor nutrition

Food insecurity

Disease

Inadequate care

Unhealthy environment

Household poverty

Fig. 3. Conceptual framework of causes of child undernutrition and short- and long-term consequences. Adapted from [67].

Available capital

Political context

early 1990s that later repeatedly has been modified in relation to child malnutrition, morbidity and mortality (Fig. 3). The basic causes of child health problems are found in global and national political contexts and the available financial, human and social capital. These conditions determine the underlying causes, that is, whether the families live in poverty with food insecurity, inadequate care, an unhealthy environment and inadequate health services. The immediate causes include inappropriate infant feeding, inadequate dietary intake and diseases that cause the undernutrition of the child, and may result in further disease, disability and death. It is estimated that the aggregated forms of undernutrition cause 45% of the under-5 deaths [13]. Children with stunted growth who live in poverty also do poorly at school, have low incomes later on, show high fertility, and are not able to provide appropriate care for their future children, thus contributing to the intergenerational vicious circle of poverty [15]. The long-term consequences of early life undernutrition also include an increased risk of non-communicable diseases (Fig. 2). Nutritional imbalance or insult in foetal life or infancy may alter later disease risk. The Developmental Origin of Health and Disease concept is underpinned by numerous epidemiological and biomedical studies [16–18]. The Developmental Origin of Health and Disease concept suggests that phenotypes are adapted to the environment in early life when plasticity is great to improve the reproductive capability. Some of these adaptations may be advantageous for health in the short perspective, but increase the risk for chronic diseases in adulthood especially when environmental conditions change [17]. Few nutritional and other lifestyle interventions have so far been made in order to

88

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

favourably influence the developmental trajectories at a stage when developmental plasticity is great [19]. In low-income and transitional societies, where food insecurity and malnourished populations are still common, interventions to secure adequate nutrition already in pregnancy may allow the next generation to maintain growth and metabolic health during a socioeconomic and nutritional transition [20].

Equity in Global Child Health

Investments in child health have the greatest potential to reduce global inequities as pointed out by the World Health Organization’s Commission on Social Determinants of Health [21]. The reduction in under-5 mortality during the past few decades has decreased the gap in child survival between countries, but frequently increased the gap between wealthy and poor children within countries [22]. However, this is not always the case. For example, studies from Nicaragua [23], Vietnam [24] and Rwanda [14] demonstrated that pro-poor policies in health planning and equity ambitions in other sectors of society abridge the difference in child survival between social groups. This is important for further downsizing of the under-5 mortality in the post-Millennium Development Goals era [25]. There may be geographic, socioeconomic, ethnic, gender or other unfair group differences in child health. An example of geographic and ethnic inequity in neonatal survival in a province in Vietnam is given in Figure 4. The overall neonatal mortality was moderately low, but some mountainous areas had very high mortality. An intervention with a community engagement strategy lowered mortality and increased equity in survival. Also, in other parts of the world, community engagement strategies have been successful in improving maternal and child health and survival [26–28]. In many settings, girls have customarily been discriminated with consequences for dietary intake, health [29], growth [30] and survival [31]. The selective abortions of female foetuses that especially occur in Asia have been labelled as gendercide with huge ethical, social and demographic consequences [32]. In contrast, investments in girls’ education have resulted in major positive impact on fertility, child health, nutrition and child survival [21, 23, 33]. The paradoxical rapid improvement of women’s and children’s health in Bangladesh that has taken place against a fond of persistent poverty and political difficulties is most likely related to the expansion of female education and women’s empowerment [34].

The Need for Data

Access to appropriate data drives development. The past few decades’ progress in child survival has been promoted by the “countdown” monitoring activities in each low-income country as well as globally [35–37]. Country progress has been evaluated

Global Child Health

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

89

N

0– 9 9– 14 14 –1 19 9 –2 29 9 –4 9 50 +

Neonatal mortality rate

Regional and provincial hospital District hospital Community health centre

0 10 20

40 km

Fig. 4. Spatial data on neonatal mortality in the Quang Ninh province, Vietnam, in 2005 revealed great geographic inequity in neonatal survival [68], that partly was explained by ethnic differences [69]. The map is based on Geographic Positioning System data on newborn deaths and smoothed by spline technique. This map was part of the baseline description of a cluster-randomised trial that focused neonatal survival. A wide variation in different local problems related to pregnancy, childbirth and the newborn babies was addressed by facilitated stakeholder groups, who met for PlanDo-Study-Act cycles, resulting in a reduction of neonatal mortality by half [70] and improved equity in survival [71].

in relation to goals and targets and the desired advancement per year or decade. This has not only stimulated countries to fulfil their own goals but also included a component of competitive spirit between countries to move forward [38]. The availability of valid routine demographic data on births and child deaths is limited in most low-income countries. Official figures on maternal and child deaths from ministries of health or national statistical offices are often underestimated. Some countries, for example, Rwanda, make efforts to benefit from the rapidly increasing coverage of mobile devices and services to monitor pregnancies and reduce maternal and child deaths [39]. In most low-income countries, there are especially difficulties in getting valid information on deaths during the first month of life [40, 41]. Births are often not officially counted until it is evident that the newborn child is surviving. The families may not approach the local authorities for a birth certificate until after one or a few months, and the deaths of young infants are therefore frequently missed in official data. As with the neonatal deaths, there is also an urgent need to identify and register stillbirths in order to visualise the size of this problem [42]. The Demographic Health Surveys (DHS) that are performed with 5-year intervals in more than 90 countries constitute very valuable data sources that are accessible by all (http://www.dhsprogram.com). More that 300 surveys have so far been performed

90

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

with a number of standardised questionnaires and measurement modules of high relevance for global child health. A careful sampling process implies that data is representative for the country at the time of interview. Modules and data-collection procedures are valid and reliable and enable comparisons between countries and over time. A country report is produced after each survey and accessible from the DHS program. Data are thereafter released for common use. Researchers and students in the field of epidemiology have used this resource and analysed determinants of different health outcomes, studied health outcomes or exposures across countries or looked into the development of health over time (see Figure 1 as an example). One of the limitations of the DHS data is the cross-sectional design that prevents the analysis of cause and effect. In several low- and middle-income countries in Africa, Asia, and to some extent Latin America, Health and Demographic Surveillance Systems (HDSS) have been established in order to create databases of a selected population that is maintained over time by repeated, frequent updates of demographic information (births, deaths, migration) and selected health information. Frequently, spatial information (based on Geographic Positioning System data) is included in the databases. These surveillance systems enable longitudinal analyses (households and individuals have unique identifiers) and analytical epidemiology with research questions on cause and effect, but also to provide a sampling frame for advanced studies, for example, community-based trials. The oldest HDSS is running in Matlab subdistrict, Bangladesh, and covers a 220,000 population in the delta area 50 km south of the capital Dhaka. It was established in the 1960s, and by now includes 3 generations. The Matlab surveillance system has close links to the provision of services. Thousands of research publications have been published from Matlab, for example, analytical approaches regarding infections and growth [43], intervention with door-step delivery of family planning and consequences for child health [44], results of vaccination trials [45] or prenatal nutrition interventions and effects on infant mortality [46]. The HDSS sites collaborate within the INDEPTH network that provides a platform for collaboration between the HDSS sites, offers methodological support and coordinates studies across different sites (http://www.indepth-network.org).

Future Challenges

We have the knowledge of a large number of evidence-based interventions that, if implemented, drastically could reduce the number of neonatal [4] as well as postneonatal and child deaths [47]. What is lacking is adequate knowledge on how to successfully implement these interventions in different contexts [48]. The Child Health and Nutrition Research Initiative has developed a strategy to prioritise the research that is needed to further improve global child health [49]. Research may be needed on the delivery or implementation of interventions, on the development of

Global Child Health

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

91

new intervention packages, and on the discovery of new preventive or curative interventions. Currently, top priority has been given to the delivery of interventions, for example, implementation research on the health and survival of the newborns [50], on the prevention of death of preterm babies [51], on the treatment of pneumonias [52], and on the management of diarrhoeal diseases [53]. Many low- and middle-income countries are undergoing a rapid epidemiologic transition with a decrease in infectious diseases, an increase in non-communicable diseases and a shift in the demographic characteristics from the broad-based population pyramid with high child mortality to a smaller base and an increasing proportion of elderly [54]. A large proportion of these countries are also in the middle of a rapid nutrition transition, where frequently under- and over-nutrition occur at the same time, sometimes in the same families [13, 55]. As mentioned above, early life nutritional insults may increase the susceptibility to insulin resistance, type 2 diabetes and other non-communicable diseases in adult life [56]. Thus, there is a link between poverty and undernutrition in childhood and later overweight and chronic diseases in these rapidly changing societies. In that situation, it is important to employ a lifecourse perspective on child nutrition and health. The stunting syndrome is rooted in poverty and developing over time with prominent inter-generational components. There are windows of opportunity to break these cycles and prevent both short- and long-term health consequences [57]. Recent history shows with terrifying clarity how children become the victims of war and catastrophes. The political turmoil and suffering of the population before the genocide in Rwanda drastically increased the under-5 mortality (Figure 1). The rapidly increasing child mortality of a collapsing society was also observed in Somalia before the onset of the war [58]. The dominating causes of death in such situations are preventable and treatable: diarrhoea, pneumonia and other diseases that could be prevented by the use of vaccines. The Rwanda example also illustrates a scary new phenomenon that continues to be reported from several conflict countries: children become specifically targeted in the armed conflicts [59]. The rapid urbanisation in some low- and middle-income countries with expanding slum settings is a challenge for child health and the provision of services. Previously, indicators of child health were usually favourable to the urban as compared to rural areas. This pattern has changed due to an aggregation of unfavourable factors in these slums: poverty, risky environment, lack of education, food insecurity and no access to services [60]. This is a challenge for all sectors of society. The health services may need to embark on new approaches in order to meet these needs of families in the slums. A good example of such innovative interventions to improve maternal and newborn health and survival comes from the urban slums in Dhaka, Bangladesh, where community engagement and use of local data have strengthen the services and impressively reduced neonatal mortality [61]. Climate change is the biggest global health threat of the 21st century, the Lancet and University College London Commission summarised in 2009 [62]. It will increase

92

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

inequities in health between rich and poor, and South Asia and Sub-Saharan Africa will suffer the most. Children are the most vulnerable to climate change, exacerbating the occurrence of infectious diseases and malnutrition and forcing families to migrate [63]. The delta region in Bangladesh, where the Matlab surveillance system is located (see above), is already observing the consequences of climate change. The increased flooding of the area, the erosion along the river banks and villages that are taken by the rivers, push families to the slums of Dhaka when they have lost their land [34]. In the surveillance system, the researchers have observed that the frequency and magnitude of extreme weather are associated with the occurrence of childhood diarrhoea [64]. Global child health in the Anthropocene epoch (when human activities have significantly impacted the earth’s ecosystem) opens up a new paradigm, where the solutions for our present needs must safeguard the earth’s life-support system that decide current and future generations’ welfare [65, 66]. The Sustainable Development Goal agenda integrates the poverty, health and welfare perspectives with the needs to counteract climate change and achieve peace and planetary stability – maybe the greatest prerequisites for continued improvement in child survival in the years to come [65].

References 1 You D, Hug L, Ejdemyr S, Idele P, Hogan D, Mathers C, et al: Global, regional, and national levels and trends in under-5 mortality between 1990 and 2015, with scenario-based projections to 2030: a systematic analysis by the UN Inter-agency Group for Child Mortality Estimation. Lancet 2015;386:2275–2286. 2 The Millennium Development Goals Report 2015. New York, United Nations, 2015. 3 GBD 2015 DALYs and HALE Collaborators: Global, regional, and national disability-adjusted life-years (DALYs) for 315 diseases and injuries and healthy life expectancy (HALE), 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet, Elsevier, 2016;388:1603–1658. 4 Lawn JE, Blencowe H, Oza S, You D, Lee ACC, Waiswa P, et al: Every Newborn: progress, priorities, and potential beyond survival. Lancet 2014;384:189– 205. 5 Black RE, Morris SS, Bryce J: Where and why are 10 million children dying every year? Lancet 2003; 361: 2226–2234. 6 Jones G, Steketee RW, Black RE, Bhutta ZA, Morris SS; Bellagio Child Survival Study Group: How many child deaths can we prevent this year? Lancet 2003; 362:65–71.

7 Lawn JE, Cousens S, Zupan J; Lancet Neonatal Survival Steering Team: 4 million neonatal deaths: when? Where? Why? Lancet 2005;365:891–900. 8 Darmstadt GL, Walker N, Lawn JE, Bhutta ZA, Haws RA, Cousens S: Saving newborn lives in Asia and Africa: cost and impact of phased scale-up of interventions within the continuum of care. Health Policy Plan 2008;23:101–117. 9 Pattinson R, Kerber K, Buchmann E, Friberg IK, Belizan M, Lansky S, et al: Stillbirths: how can health systems deliver for mothers and babies? Lancet 2011; 377:1610–1623. 10 Kerber KJ, de Graft-Johnson JE, Bhutta ZA, Okong P, Starrs A, Lawn JE: Continuum of care for maternal, newborn, and child health: from slogan to service delivery. Lancet 2007;370:1358–1369. 11 de Onis M, Blössner M, Borghi E: Prevalence and trends of stunting among pre-school children, 1990– 2020. Public Health Nutr 2012;15:142–148. 12 Shrimpton R, Victora CG, de Onis M, Lima RC, Blössner M, Clugston G: Worldwide timing of growth faltering: implications for nutritional interventions. Pediatrics 2001;107:E75. 13 Black RE, Victora CG, Walker SP, Bhutta ZA, Christian P, de Onis M, et al: Maternal and child undernutrition and overweight in low-income and middleincome countries. Lancet 2013;382:427–451.

Global Child Health

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

93

14 Musafili A, Essén B, Baribwira C, Binagwaho A, Persson LÅ, Selling KE: Trends and social differentials in child mortality in Rwanda 1990–2010: results from three demographic and health surveys. J Epidemiol Community Health 2015:69:834–840. 15 Grantham-McGregor S, Cheung YB, Cueto S, Glewwe P, Richter L, Strupp B: Developmental potential in the first 5 years for children in developing countries. Lancet 2007;369:60–70. 16 Barker DJ: Early growth and cardiovascular disease. Arch Dis Child 1999;80:305–307. 17 Barouki R, Gluckman PD, Grandjean P, Hanson M, Heindel JJ: Developmental origins of non-communicable disease: implications for research and public health. Environ Health 2012;11:42. 18 Gluckman PD, Hanson MA, Buklijas T: A conceptual framework for the developmental origins of health and disease. J Dev Orig Health Dis 2010;1:6–18. 19 Hanson M, Godfrey KM, Lillycrop KA, Burdge GC, Gluckman PD: Developmental plasticity and developmental origins of non-communicable disease: theoretical considerations and epigenetic mechanisms. Prog Biophys Mol Biol 2011;106:272–280. 20 Gluckman PD, Hanson MA, Beedle AS: Early life events and their consequences for later disease: a life history and evolutionary perspective. Am J Hum Biol 2007;19:1–19. 21 Marmot M, Friel S, Bell R, Houweling TA, Taylor S: Closing the gap in a generation: health equity through action on the social determinants of health. Lancet 2008;372:1661–1669. 22 Victora CG, Wagstaff A, Schellenberg JA, Gwatkin D, Claeson M, Habicht JP: Applying an equity lens to child health and mortality: more of the same is not enough. Lancet 2003;362:233–241. 23 Peña R, Liljestrand J, Zelaya E, Persson LA: Fertility and infant mortality trends in Nicaragua 1964–1993. The role of women’s education. J Epidemiol Community Health 1999;53:132–137. 24 Hoa DP, Nga NT, Målqvist M, Persson LA: Persistent neonatal mortality despite improved under-five survival: a retrospective cohort study in northern Vietnam. Acta Paediatr 2008;97:166–170. 25 Målqvist M: Abolishing inequity, a necessity for poverty reduction and the realisation of child mortality targets. Arch Dis Child 2015;100(suppl 1):S5–S9. 26 Osrin D, Prost A: Perinatal interventions and survival in resource-poor settings: which work, which don’t, which have the jury out? Arch Dis Child 2010; 95:1039–1046. 27 Prost A, Colbourn T, Tripathy P, Osrin D, Costello A: Analyses confirm effect of women’s groups on maternal and newborn deaths. Lancet 2013;381:e15.

94

28 WHO Guidelines Approved by the Guidelines Review Committee: WHO recommendation on community mobilization through facilitated participatory learning and action cycles with women’s groups for maternal and newborn health. World Health Organization, 2014. 29 Belachew T, Hadley C, Lindstrom D, Gebremariam A, Michael KW, Getachew Y, et al: Gender differences in food insecurity and morbidity among adolescents in southwest Ethiopia. Pediatrics 2011; 127:e398–e405. 30 Bharati P, Bharati S, Pal M, Chakrabarty S, Som S, Gupta R: Growth and nutritional status of preschool children in India: rural-urban and gender differences. Coll Antropol 2009;33:7–21. 31 Brinda EM, Rajkumar AP, Enemark U: Association between gender inequality index and child mortality rates: a cross-national study of 138 countries. BMC Public Health 2015;15:97. 32 Grech V: Gendercide and femineglect. Early Hum Dev 2015;91:851–854. 33 Kiros GE, Hogan DP: War, famine and excess child mortality in Africa: the role of parental education. Int J Epidemiol 2001;30:447–455, discussion 456. 34 Chowdhury AM, Bhuiya A, Chowdhury ME, Rasheed S, Hussain Z, Chen LC: The Bangladesh paradox: exceptional health achievement despite economic poverty. Lancet 2013;382:1734–1745. 35 Requejo JH, Bryce J, Barros AJ, Berman P, Bhutta Z, Chopra M, et al: Countdown to 2015 and beyond: fulfilling the health agenda for women and children. Lancet 2015;385:466–476. 36 Amouzou A, Habi O, Bensaïd K; Niger Countdown Case Study Working Group: Reduction in child mortality in Niger: a Countdown to 2015 country case study. Lancet 2012;380:1169–1178. 37 Afnan-Holmes H, Magoma M, John T, Levira F, Msemo G, Armstrong CE, et al: Tanzania’s countdown to 2015: an analysis of two decades of progress and gaps for reproductive, maternal, newborn, and child health, to inform priorities for post-2015. Lancet Glob Health 2015;3:e396–e409. 38 Countdown 2008 Equity Analysis Group, Boerma JT, Bryce J, Kinfu Y, Axelson H, Victora CG: Mind the gap: equity and trends in coverage of maternal, newborn, and child health services in 54 Countdown countries. Lancet 2008;371:1259–1267. 39 Ngabo F, Nguimfack J, Nwaigwe F, Mugeni C, Muhoza D, Wilson DR, et al: Designing and Implementing an Innovative SMS-based alert system (RapidSMS-MCH) to monitor pregnancy and reduce maternal and child deaths in Rwanda. Pan Afr Med J 2012;13:31.

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

40 Målqvist M, Eriksson L, Nguyen TN, Fagerland LI, Dinh PH, Wallin L, et al: Unreported births and deaths, a severe obstacle for improved neonatal survival in low-income countries; a population based study. BMC Int Health Hum Rights 2008;8:4. 41 Huy TQ, Johansson A, Long NH: Reasons for not reporting deaths: a qualitative study in rural Vietnam. World Health Popul 2007;9:14–23. 42 Lawn JE, Blencowe H, Pattinson R, Cousens S, Kumar R, Ibiebele I, et al: Stillbirths: where? when? why? how to make the data count? Lancet 2011;377: 1448–1463. 43 Torres AM, Peterson KE, de Souza AC, Orav EJ, Hughes M, Chen LC: Association of diarrhoea and upper respiratory infections with weight and height gains in Bangladeshi children aged 5 to 11 years. Bull World Health Organ 2000;78:1316–1323. 44 Joshi S, Schultz TP: Family planning and women’s and children’s health: long-term consequences of an outreach program in Matlab, Bangladesh. Demography 2013;50:149–180. 45 Zaman K, Fleming JA, Victor JC, Yunus M, Bari TI, Azim T, et al: Noninterference of rotavirus vaccine with measles-rubella vaccine at 9 months of age and improvements in antirotavirus immunity: a randomized trial. J Infect Dis 2016;213:1686–1693. 46 Persson LÅ, Arifeen S, Ekstrom EC, Rasmussen KM, Frongillo EA, Yunus M, et al: Effects of prenatal micronutrient and early food supplementation on maternal hemoglobin, birth weight, and infant mortality among children in Bangladesh: the MINIMat randomized trial. JAMA 2012;307:2050–2059. 47 Liu L, Black RE: Child survival in 2015: much accomplished, but more to do. Lancet 2015;386:2234–2235. 48 Sanders D, Haines A: Implementation research is needed to achieve international health goals. PLoS Med 2006;3:e186. 49 Rudan I, Kapiriri L, Tomlinson M, Balliet M, Cohen B, Chopra M: Evidence-based priority setting for health care and research: tools to support policy in maternal, neonatal, and child health in Africa. PLoS Med 2010;7:e1000308. 50 Yoshida S, Rudan I, Lawn JE, Wall S, Souza JP, Martines J, et al: Newborn health research priorities beyond 2015. Lancet 2014;384:e27–e29. 51 Bahl R, Martines J, Bhandari N, Biloglav Z, Edmond K, Iyengar S, et al: Setting research priorities to reduce global mortality from preterm birth and low birth weight by 2015. J Glob Health 2012;2:10403. 52 Rudan I, El Arifeen S, Bhutta ZA, Black RE, Brooks A, Chan KY, et al: Setting research priorities to reduce global mortality from childhood pneumonia by 2015. PLoS Med 2011;8:e1001099.

53 Wazny K, Zipursky A, Black R, Curtis V, Duggan C, Guerrant R, et al: Setting research priorities to reduce mortality and morbidity of childhood diarrhoeal disease in the next 15 years. PLoS Med 2013; 10:e1001446. 54 McKeown RE: The epidemiologic transition: changing patterns of mortality and population dynamics. Am J Lifestyle Med 2009;3(1 suppl):19S–26S. 55 Conde WL, Monteiro CA: Nutrition transition and double burden of undernutrition and excess of weight in Brazil. Am J Clin Nutr 2014; 100: 1617S– 1622S. 56 Vorster HH, Kruger A, Margetts BM: The nutrition transition in Africa: can it be steered into a more positive direction? Nutrients 2011;3:429–441. 57 Prendergast AJ, Humphrey JH: The stunting syndrome in developing countries. Paediatr Int Child Health 2014;34:250–265. 58 Ibrahim MM, Omar HM, Persson LA, Wall S: Child mortality in a collapsing African society. Bull World Health Organ 1996;74:547–552. 59 Shenoda S, Kadir A, Goldhagen J: Children and armed conflict. Pediatr Am Acad Pediatr 2015; 136:e309–e311. 60 Garenne M: Urbanisation and child health in resource poor settings with special reference to underfive mortality in Africa. Arch Dis Child 2010;95:464– 468. 61 Marcil L, Afsana K, Perry HB: First steps in initiating an effective maternal, neonatal, and child health program in urban slums: the BRAC manoshi project’s experience with community engagement, social mapping, and census taking in Bangladesh. J Urban Health 2016;93:6–18. 62 Costello A, Abbas M, Allen A, Ball S, Bell S, Bellamy R, et al: Managing the health effects of climate change: lancet and University College London Institute for Global Health Commission. Lancet 2009; 373:1693–1733. 63 Ahdoot S, Pacheco SE; Council on Environmental Health: Global Climate Change and Children’s Health. Pediatrics 2015;136:e1468–e1484. 64 Wu J, Yunus M, Streatfield PK, Emch M: Association of climate variability and childhood diarrhoeal disease in rural Bangladesh, 2000–2006. Epidemiol Infect 2014;142:1859–1868. 65 Griggs D, Stafford-Smith M, Gaffney O, Rockström J, Ohman MC, Shyamsundar P, et al: Policy: Sustainable development goals for people and planet. Nature 2013;495:305–307. 66 Whitmee S, Haines A, Beyrer C, Boltz F, Capon AG, de Souza Dias BF, et al: Safeguarding human health in the Anthropocene epoch: report of The Rockefeller Foundation-Lancet Commission on planetary health. Lancet 2015;386:1973–2028.

Global Child Health

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

95

67 Black RE, Allen LH, Bhutta ZA, Caulfield LE, de Onis M, Ezzati M, et al: Maternal and child undernutrition: global and regional exposures and health consequences. Lancet 2008;371:243–260. 68 Nga NT, Målqvist M, Eriksson L, Hoa DP, Johansson A, Wallin L, et al: Perinatal services and outcomes in Quang Ninh province, Vietnam. Acta Paediatr 2010; 99:1478–1483. 69 Målqvist M, Nga NT, Eriksson L, Wallin L, Hoa DP, Persson LÅ: Ethnic inequity in neonatal survival: a case-referent study in northern Vietnam. Acta Paediatr 2011;100:340–346.

70 Persson LÅ, Nga NT, Målqvist M, Thi Phuong Hoa D, Eriksson L, Wallin L, et al: Effect of facilitation of local maternal-and-newborn stakeholder groups on neonatal mortality: cluster-randomized controlled trial. PLoS Med 2013;10:e1001445. 71 Målqvist M, Hoa DP, Persson LÅ, Ekholm Selling K: Effect of facilitation of local stakeholder groups on equity in neonatal survival; results from the NeoKIP Trial in northern vietnam. PLoS One 2015; 10: e0145510.

Prof. Lars Åke Persson, MD, PhD Department of Disease Control, Faculty of Infectious and Tropical Diseases London School of Hygiene and Tropical Medicine Keppel Street, London WC1E 7HT (UK) E-Mail [email protected]

96

Persson

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 85–96 (DOI: 10.1159/000481325)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

How to Deal with Proxy-Reports Jon Genuneit Institute of Epidemiology and Medical Biometry, Ulm University, Ulm, Germany

Abstract Children are not always able to fully respond to all questions posed in medical research due to, for example, limited understanding of the queried items or limitations in expressing their perception or views. This is clearly true for newborns and infants and likely true for toddlers but may also apply to children at older ages. In pediatric epidemiological research, it is thus customary to obtain proxyreports rather than self-reports, most often from parents or explicitly from the mother. This chapter describes important aspects of proxy-reports that should be considered in pediatric epidemiology, including proxy perspective and proxy selection as well as quality of proxy-reports, their agreement with self-reports, and their influence on missing data, on information bias and on confounding. All interpretation of proxy-reports is context-dependent. The proxy perspective should be clearly defined and the best available proxy to inform about the items requested should be selected. Here, characteristics of the index subject, the proxy, and their relationship have to be taken into account. Ideally, pilot studies should be conducted to determine feasibility of proxy selection and quality of © 2018 S. Karger AG, Basel their reports.

Definitions

Clearly, the choice of the best proxy or surrogate respondent or informant (hereafter called proxy) for the index subject depends on the research question. If the aim is to obtain information as it would have been given by the index subject himself (i.e., substitution of the index subject assessment called the proxy-participant or proxy-patient perspective) [1], the proxy should be the one who is best suited to inform about the

a

b

Fig. 1. Illustration of participant self-report [༬], proxy-participant perspective [༬] and proxyproxy perspective [༬] with inter-rater gap [ ] and intra-proxy gap [ ]. a Equally sized interrater and intra-proxy gaps. b Intra-proxy gap greater than inter-rater gap. c Intra-proxy gap smaller than inter-rater gap. Modified and extended from Pickard and Knight [1].

c

items that cannot be ascertained from the index subject himself. There is, however, another perspective that may be of interest: the proxy-proxy perspective about aspects of the index subject [1]. Under this perspective, a proxy can reinforce or complement the self-reported index subject assessment. The perspectives are illustrated in Figure 1. The difference between the 2 perspectives is termed the intra-proxy gap or intra-proxy difference [1]. Self- and proxy-reports under the proxy-participant perspective often differ. This is called the interrater gap or inter-rater difference (Fig. 1) [1]. Intra-proxy gap and inter-rater gap do not have to have the same size (compare Fig. 1a–c). It may well be that a proxy’s perception of the index subject (proxy-proxy perception) is closer to what the index subject would self-report than to what the proxy reports on the index subject’s behalf (proxy-participant perception; Fig. 1c).

Proxy Perspectives

The 2 proxy perspectives should not be mistaken with self-completion of a questionnaire compared to an interview. They pertain to the source of data rather than the methodology to obtain it. There are various sources of data in pediatric epidemiology; most of them are discussed in a thorough guide to undertaking a birth cohort study, a major study design in pediatric epidemiology [2]. These data sources include the index subjects themselves, proxies, physical examinations, hospital and other health records, educational records, and local environmental data. In particular, health or

98

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

Table 1. Proxy perspectives – examples of questions Item

Commonly used wording

Proxy-participant Perspective

Proxy-proxy perspective

Mode of delivery

By which mode of delivery was your child born? (spontaneous vaginal/assisted vaginal/ emergency cesarean section/ planned cesarean section)

By which mode of delivery was your child born?

By which mode of delivery was your child born?

Exposure to environmental tobacco smoke

Is your child exposed to environmental tobacco smoke? (yes/no)

Would your child say she/he is exposed to environmental tobacco smoke?

Do you perceive your child to be exposed to environmental tobacco smoke?

Eating chocolate

How often does your child eat chocolate? (frequency)

What would your child say: How often does she/he eat chocolate?

What do you think: How often does your child eat chocolate?

Physical activity

How often does your child play with a ball? (frequency)

What would your child say: How often does she/he play with a ball?

What do you think: How often does your child play with a ball?

Itchy skin as a symptom of atopic dermatitis

Has your child had an itchy skin condition that was either continuous or intermittent lasting at least 4 weeks? (yes/ no)

Did your child notice an itchy skin condition that was either continuous or intermittent lasting at least 4 weeks?

Did you notice that your child was affected by an itchy skin condition that was either continuous or intermittent lasting at least 4 weeks?

Self-rated health status

How is the health of your child? (excellent/good/fair/ poor)

How would your child rate his/her health status?

How do you perceive the health status of your child to be?

educational records may be perceived as proxy-reports in pediatric epidemiology, while the most common instance of proxy-reporting is ascertainment of data by a selfcompletion questionnaire administered to or an interview with a proxy. Both the proxy-participant and the proxy-proxy perspective may be important in pediatric epidemiology. They should be determined prior to the assessment of data as should be the appropriate wording of questions to reflect the perspective. Table 1 gives some examples. More often than not the wording commonly used in questionnaires may be interpreted by the proxy in both ways: to respond as if the proxy was the index subject or to respond as if the proxy completes the questionnaire thinking about the index subject. It is crucial to use explicit wording in order to avoid confusion of the proxy perspectives both for the proxy but also for the data analyst. The examples in Table 1 highlight that the proxy perspective becomes more of an issue when the item in question is influenced by subjective views. There is little room for interpretation when it comes to the mode of delivery. Thus, the proxy perspective does not matter much. However, the knowledge about the information requested

Proxy-Reports

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

99

may matter (see Proxy selection). For exposure to environmental tobacco smoke, perception of exposure may matter. This is probably more of a problem when it comes to dosage rather than a simple yes/no dichotomy. Also, disclosure of information about exposure to tobacco smoke may differ between the participant and the proxy; this is described below in Proxy-reports and information bias. The same applies to eating chocolate. Assessment of disease symptoms is also often influenced by the proxy perspective. While itching in the index subject can be observed by the proxy (although not all the time), psychiatric traits like mood disorders may require more insight and reflection under the proxy-participant and proxy-proxy perspective respectively. For self-rated health, the proxy perspective becomes even more important. Here, the response to items in question is influenced by expectations and beliefs that may well influence the inter-rater gap but also the intra-proxy gap [1]. In particular, it has been documented that proxies tend to be imprecise and underestimate quality of life [3].

Proxy Selection

Proxy selection is depending on the required information and on the desired proxy perspective. While parents seem like a natural choice in pediatric epidemiology, there are other proxies that may be of value, for example, day care staff members, teachers, caring physicians, but also other relatives including siblings or peers from the participant’s social network. Proxy-reports from siblings or peers are rarely pursued in pediatric epidemiology. Certainly, the age of siblings or peers may come with the same limitation requiring a proxy-report in the first place. Also, informed consent may be an issue. Naturally, siblings are not available in families with single children, which may induce selection bias. Still, siblings and/or peers may provide interesting insight, particularly under the proxy-proxy perspective. Whichever proxy is selected, informed consent should be sought from the proxy (who is disclosing information) and potentially also from the index subject (on whom information is disclosed). Discussion of informed consent in pediatric epidemiology is beyond the scope of this chapter. Some information on this topic is included in the chapter by Rotzoll and Willer [this vol., pp. 1–15].

Quality of the Proxy-Report

The quality of the proxy-report can be influenced by the duration of the relationship between index subject and proxy as well as the frequency and intensity of contact. Good agreement between self- and proxy-report under the proxy-participant perspective (i.e., a small inter-rater gap) can be achieved for well-documented or incisive medical facts, for example, mode of delivery or chronic medical conditions.

100

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

Typically lower agreement is found for lifestyle factors, for example, the number of cigarettes smoked, and even less agreement for disease symptoms, minor illness, or overall health ratings. Indeed, a proxy may be able to give a more accurate response than the index subject himself, for example, the child’s (index subject’s) exposure to environmental tobacco smoke due to maternal (proxy) smoking. Agreement may also depend on the time in life at which we are interested in a specific factor. For example, if the food the child eats is solely prepared by the parents as it mostly occurs at an early age, parental proxy-reports of child food intake are likely to be the most accurate reports that can be obtained. At an older age of the child, however, the child may make his or her own decisions about what to eat and when and parental proxy reports are likely to be incomplete or wrong. Think about chocolate consumption as an example. Assessing agreement either in a pilot study or in a part of the main study is important. This is foremost agreement between self- and proxy-reports and there are many examples in adult studies [4]. However, if self-reports are not feasible as it often occurs in pediatric epidemiology, agreement between multiple proxies (e.g., mothers and fathers, parents and pediatricians, or parents and teachers) may help to understand the influence of proxy perspective and proxy selection. An example from quality-of-life research, albeit among adults with intellectual disabilities, effectively documents the differences between proxy-reports by family members and direct support staff [5]. Furthermore, a review highlights the necessity to understand and interpret differences between parental and child report of health-related quality of life and disease symptoms rather than discarding them as methodological error [6]. The accuracy of proxy-reports is discussed in more detail in the following sections.

Proxy-Reports and Missing Data

On the one hand, compared to self-reports, the use of proxy-reports may result in a higher portion of missing data because proxies may not know the requested information, thus causing item non-response. On the other hand, proxies may be in the position to respond completely to the items in question, while the index subjects would not be able to disclose all information required. This would lead to a decrease of missing data. Similarly, willingness to participate may be higher among proxies and increase participation rates if proxies are explicitly included in the study, thus reducing unit non-response. Still, being a proxy may also be an additional burden impacting on participation thus increasing unit non-response. A full discussion of participation and selection bias is beyond the scope of this chapter. Some information on these topics is included in the chapter by Hoffmann et al. [this vol., pp. 71–84]. Still, the influence of proxy selection on participation and potentially selection bias has to be acknowledged. Moreover, children may have par-

Proxy-Reports

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

101

ticular views on medical research [7, 8], which potentially have to be respected by the proxy and may influence item or unit non-response. The influences of proxy-reports on item and unit non-response depend on the requested information. If the non-response patterns are differential with regard to items under study, they are likely to invoke information bias.

Proxy-Reports and Information Bias

Proxies may not have accurate information about the item in question. Moreover, proxies may be biased in their perception of the item in question either due to their own characteristics, for example, education or lifestyle, or due to their relation with the index subject, for example, a beloved child. Also, the quality of the information obtained from a proxy can depend on characteristics of the index subject, for example, the reason for the need of a proxy in the first place and of course on the nature of the information requested. Moreover, the quality of the proxy-report can be influenced by the duration of the relationship between index subject and proxy as well as the frequency and intensity of contact. There are many examples on how a proxy report can be biased based on proxy characteristics, mostly from adult studies. For example, family members with a history of depression are more likely to proxy-report depression in their related index subject [4]. Again, the lack of child self-reports in pediatric epidemiology potentially leads to underappreciation of information bias invoked by (mostly parental) proxyreports. However, there is ample literature on how chronic disease in the child affects the parents and there is emerging literature on how it affects siblings as well [9, 10]. It is thus conceivable that information bias plays a substantial role when proxy-reports are used in pediatric epidemiology.

Proxy Reports and Confounding

Whereas general discussions of selection bias, information bias, and confounding are most often included, standard epidemiology textbooks typically do not explicitly discuss issues attached to proxy-reports with one notable exception known to the author [11]. Here, a full chapter is devoted to the discussion of proxy-reports. However, proxy-reports are only assumed for exposure assessment and not for outcome assessment. Most of the discussed issues apply to the latter as well. An extension would be that if proxy-reports are biased by the same proxy characteristic for both exposure and outcome, this may induce confounding. Options to deal with confounding are discussed in another chapter in this volume [12]. The aforementioned characteristics of the proxy-report (proxy perspective, proxy selection, potential for missing data and information bias) should be evaluated to determine the potential for confounding.

102

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

Self-Reports

The previous sections have highlighted that some of the issues are difficult to determine in pediatric epidemiology if child self-reports in contrast to proxy-reports cannot be obtained. Attempts to overcome barriers in language, understanding, or conceptualization include video or graphic questionnaires in which questions or response options are visualized [13–18]. Also, personal interviews can be attempted; but if so, the setting may be particularly important [19].

Conclusions

All interpretation of proxy-reports is context-dependent. The proxy perspective should be clearly defined and the best available proxy to inform about the items requested should be selected. Here, characteristics of the index subject, the proxy, and their relationship have to be taken into account. In general, intra-proxy gap and interrater gap are not well studied entities. Ideally, pilot studies should be conducted to determine feasibility of proxy selection and quality of their reports.

References 1 Pickard AS, Knight SJ: Proxy evaluation of healthrelated quality of life: a conceptual framework for understanding multiple proxy perspectives. Med Care 2005;43:493–499. 2 Golding J, Jones R: Sources of data for a longitudinal birth cohort. Paediatr Perinat Epidemiol 2009; 23(suppl 1):51–62. 3 Crocker TF, Smith JK, Skevington SM: Family and professionals underestimate quality of life across diverse cultures and health conditions: systematic review. J Clin Epidemiol 2015;68:584–595. 4 Vandeleur CL, Rothen S, Lustenberger Y, Glaus J, Castelao E, Preisig M: Inter-informant agreement and prevalence estimates for mood syndromes: direct interview vs. family history method. J Affect Disord 2015;171:120–127. 5 Claes C, Vandevelde S, Van Hove G, van Loon J, Verschelden G, Schalock R: Relationship between self-report and proxy ratings on assessed personal quality of life-related outcomes. J Policy Pract Intellect Disabil 2012;9:159–165. 6 Eiser C, Varni JW: Health-related quality of life and symptom reporting: similarities and differences between children and their parents. Eur J Pediatr 2013; 172:1299–1304.

7 van der Pal S, Sozanska B, Madden D, Kosmeda A, Debinska A, Danielewicz H, et al: Opinions of children about participation in medical genetic research. Public Health Genomics 2011;14:271–278. 8 Swartling U, Hansson MG, Ludvigsson J, Nordgren A: “My parents decide if I can. I decide if I want to.” Children’s views on participation in medical research. J Empir Res Hum Res Ethics 2011;6:68–75. 9 Nielsen KM, Mandleco B, Roper SO, Cox A, Dyches T, Marshall ES: Parental perceptions of sibling relationships in families rearing a child with a chronic condition. J Pediatr Nurs 2012;27:34–43. 10 Knecht C, Hellmers C, Metzing S: The perspective of siblings of children with chronic illness: a literature review. J Pediatr Nurs 2015;30:102–116. 11 White E, Armstrong BK, Saracci R: Principles of Exposure Measurement in Epidemiology: Collecting, Evaluating and Improving Measures of Disease Risk Factors. Oxford, Oxford University Press, 2008. 12 Genuneit J: How to deal with confounding. Pediatr Adolesc Med 2017, DOI 10.1159/000481329. 13 Beasley R, Lai CK, Crane J, Pearce N: The video questionnaire: one approach to the identification of the asthmatic phenotype. Clin Exp Allergy 1998; 28(suppl 1):8–12; discussion 32–36.

Proxy-Reports

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

103

14 Collins ME: Body figure perceptions and preferences among preadolescent children. Int J Eat Disord 1991; 10:199–208. 15 Carraway-Stage V, Spangler H, Borges M, Goodell LS: Evaluation of a pictorial method to assess liking of familiar fruits and vegetables among preschool children. Appetite 2014;75:11–20. 16 Pianosi PT, Huebner M, Zhang Z, Turchetta A, McGrath PJ: Dalhousie pictorial scales measuring dyspnea and perceived exertion during exercise for children and adolescents. Ann Am Thorac Soc 2015;12: 718–726.

17 Schwerdtle B, Kanis J, Kahl L, Kübler A, Schlarb AA: Children’s sleep comic: development of a new diagnostic tool for children with sleep disorders. Nat Sci Sleep 2012;4:97–102. 18 Maćkiewicz M, Cieciuch J: Pictorial personality traits questionnaire for children (PPTQ-C)-A new measure of children’s personality traits. Front Psychol 2016;7:498. 19 Coad J, Gibson F, Horstman M, Milnes L, Randall D, Carter B: Be my guest! Challenges and practical solutions of undertaking interviews with children in the home setting. J Child Health Care 2015;19:432–443.

PD Dr. med. Jon Genuneit, MSc Institute of Epidemiology and Medical Biometry, Ulm University Helmholtzstrasse 22 DE–89081 Ulm (Germany) E-Mail [email protected]

104

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 97–104 (DOI: 10.1159/000481326)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

Biology at a Young Age Differs from Biology at Later Ages: Developmental Aspects of Growth and Body Functions in Children and Young Adults Olle Söder Department of Women’s and Children’s Health, Karolinska Institutet and University Hospital, Stockholm, Sweden

Abstract Children are not just small adults. A number of anatomical and physiological features are different in children than in adults and biology at a young age differs from that of later ages. This has important implications not only in health care for the clinical management of disorders in adults and children of different developmental ages. It is also of great importance for the choice of research approach and for interpretation and comparison of research data obtained from epidemiological, physiological, clinical and other studies involving children and adults. This chapter focuses on sev© 2018 S. Karger AG, Basel eral important aspects of this subject.

Introduction

A number of aspects of body anatomy and physiology of children differ from those of adults. These include the changes of the relative sizes of tissues and organs in development, water and fat content, as well as cognitive functions and emotional skills. Such differences may have major consequences for various health and disease measures, and include, for example, variations of tissue distribution, biological effects and metabolism of pharmaceuticals at the different developmental ages. The common saying that children are not just small adults has implications not only for clinical management but also for interpretation of research data from physiological, clinical and epidemiological studies in children when compared with adults. This chapter deals with some important aspects of this topic.

Age Determination in Children

In children, one distinguishes between chronological and developmental age, which in theory should be identical in healthy subjects. Chronological age is defined as the number of years a child has lived and biological age as the developmental stage that is reached based on biomarkers, which are recordable measures. Milestone achievements such as motor function and communications skills should be continuously followed in all young children to detect developmental delays. Psychological age is the psychosocial developmental stage assessed by appropriate psychological instruments, whereas functional age is defined by combining the chronological, biological and psychological ages. It is not uncommon that children vary in the progress rate of different developmental parameters although the final outcome in late adolescence is the same. One such example is skeletal maturation recorded from radiological images as “bone age”, which is affected by the stage of pubertal maturation. Therefore, determination of the chronological age of a child using developmental measures cannot be performed with great precision. If the correct or proposed chronological age of a child is unknown or considered unlikely, age determination may be requested by authorities and others. For instance, this may occur when there is doubt on the chronological age in an adopted infant or when decisions have to be made on child or adult legal status in adolescents. An erroneous decision on chronological age in such cases may cause major consequences for the individual, for example, deportation or not at the border control of an immigrant adolescent. From the above it is clear that clinical management and research studies encompassing children must consider differences in developmental age in addition to the chronological age and gender.

Chronological Age Periods in Children

Childhood (age 0–18 years) is divided into several chronological periods associated with a number of maturational events, both biological and psychosocial. Most of these events are continuous and not step-wise and start already at prenatal age representing premature babies. Extreme prematures born at a very early gestational age may survive today and are thus appropriate to include in a potential first postnatal period. The commonly used chronological time periods of childhood are described below. Prenatal (Prematures) Today survival is possible from an extreme preterm gestational age of 24 weeks with help of modern neonatal intensive care. In such extreme prematures, all organ systems are immature. Most vulnerable are the respiratory system, brain, eyes and physical borders (skin). Body water content is higher and the relative size of the head is high (Fig. 1).

106

Söder

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

2 months

5 months

Newborn

2 years

6 years

12 years

22 years

Intrauterine

Fig. 1. Human body proportions in development from fetal to adult age [11].

Term Neonate (Newborn) Term neonatal age starts from gestational week 37 and includes the first month (0–28 days) of life. In this period, most severe malformations and metabolic and endocrine disorders including those detected by newborn screening are revealed. Infancy Infant baby age hosts the first year (1–12 months) of postnatal life. In this period, the most rapid postnatal growth velocity is found (Fig. 2), which is a declining remnant of the even more rapid foetal growth. Toddler Toddler age holds the first 1–3 years of life when milestones such as motor function (gait) and language communication develop. Play or Preschool Age The preschool period includes the ages 3–5 years when early cognitive and social functions develop. School Age Primary or elementary school age (also called prepubescence) includes the middle period of childhood taking place between 6 and 11 years of age.

Biology at a Young Age Differs from Biology at Later Ages

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

107

cm 200

Sex steroids Growth hormone

180

2 +3 1+ 2 1+

Foetal growth

160

ed

bin

140

m Co

120 100 80

Infancy 1

60

Childhood 2

40 Puberty 3

20 0

Fig. 2. ICP model of postnatal human longitudinal growth by Karlberg [4].

–1

3

7 11 Age, years

15

19

Adolescence The adolescent period starts at age 12 years, at the onset of puberty, and continues to the end of childhood at age 18. This period is characterized by cognitive, emotional and psychosocial development of critical importance for future societal functioning. The full biological basis of reproductive function is also attained during this period. Adulthood Young adulthood is defined as the age period 19–30 years.

Gross Anatomy: Organ and Tissue Growth in Children

The general type of growth in children is reflected by the longitudinal growth pattern. Body proportions change largely from foetal through perinatal age and during further transition into young adult age (Fig. 1). Among the most obvious are the changes of the relative size of the head reflecting growth of the neural tissue (brain). From this example it is obvious that a larger relative size not necessarily reflects a more advanced or complex function. Other examples of tissues with a developmental growth pattern that differ from the general type are the lymphoid tissue and the genital tissue (Fig. 3).

108

Söder

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

Size attained as percentageof total postnatal growth

200

Lymphoid

180 160 140 120 100 Neural 80 60 40

General

20

Fig. 3. Childhood growth curves of different tissues of the body with 4 main types: general, lymphoid, neural and genital [12].

Genital

0 0

2

4

6

8 10 12 Age, years

14

16

18

20

Body Size Measures

Body size is most often measured as body weight and/or height [1]. Body mass index (BMI) defined as BMI = weight kg/height meter2, is often used as a proxy for fat content in obesity [2]. Body surface area (BSA) may serve as a better reference for dosing of drugs, fluid supply, and so on and for determining body functions and structures in children from 1 month of age. The BSA of children spans from 0.2 m.sq. in newborns to approximately 1.9 m.sq. in young adult males and 1.7 m.sq. in females. BSA is typically determined from a nomogram with weight and height data [3] or calculated by a formula such as: BSA = square root of (height cm × weight kg/3,600).

Longitudinal Growth Patterns

Postnatal human longitudinal growth occurs in 3 distinct phases, each with a different pattern and regulation (Fig. 2). This growth model is referred to as the infant (I)-childhood (C)- puberty (P) model [4]. The initial I phase, starting from birth as a prolongation of the fast foetal growth pattern, shows the most rapid postnatal

Biology at a Young Age Differs from Biology at Later Ages

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

109

growth velocity that may increase the infant’s length by 50% (approx. 25 cm) during the first year of life. This phase ceases after one year of age and is followed by childhood (C) growth that proceeds until the start of the pubertal (P) growth spurt. This pubertal phase begins typically 2 years earlier (often at age 9–10) in girls than in boys and is a reflection of the earlier maturation and start of puberty in girls. The P phase continues until final height is reached. The different components of the ICP model and their main endocrine regulation are depicted in Figure 2. Worth to note is that adequate nutrition is required for all growth phases and that growth hormone has a minor role for longitudinal growth in the foetus and during the infant phase. Thus, it is obvious that any study involving children – clinical, epidemiological or other – must take into consideration what gender and which growth phase of the enrolled children that apply. Growth and development at all phases are susceptible to general disturbances caused by infections, malnutrition, hypothyroidism, toxicants and other factors.

Pubertal Development

Puberty is defined as the transition period from childhood into adult life accompanied by attainment of reproductive function. Start of puberty is dependent on the awakening of the part of the brain named hypothalamus that secretes gonadotropin (Gn) releasing hormone in a pulsatile fashion stimulating production and secretion of Gn by the pituitary. Gn in turn activates the gonads to produce sex steroids, mainly estrogen in girls and testosterone in boys, exerting gender-specific biological effects on hormone-dependent target organs. Simultaneously, the gonads initiate gametogenesis to produce functional gametes with fertilizing capacity. On average, girls mature 2 years earlier than boys. Not only the timing of the start of puberty but also its pace (tempo) should be considered when relevant in clinical and epidemiological studies. A note of caution is needed for the use of pubertal measures in such studies. Clinical evidence of start of puberty is more difficult to assess in boys than in girls. The most common first sign of puberty in boys is testicular growth, which needs palpation by trained hands to assess. Such investigation is not always easy to perform due to cultural, ethical and psychological reasons. Other signs of puberty in boys are less accurate when it comes to determining pubertal onset. In girls, breast enlargement is usually the first sign of puberty. In addition to inspection of breast development, palpation is also needed to judge whether or not glandular tissue is present. In obese girls, inspection alone will not discriminate from lipomastia. The first menstrual bleeding (menarche) is a strong but late and self-reported marker of ongoing female puberty. The precise time account and the abundance of historical data make it useful, but menarcheal age should not be used to determine pubertal onset.

110

Söder

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

Table 1. Examples of normative mean haemoglobin concentration in children of various chronological ages of both genders Age

Gender

Hb conc.

Term newborn 1 month 3–6 months 2 years 12 years 16 years 16 years

Both Both Both Both Both Female Male

165 140 115 120 135 140 145

Adult Adult

Female Male

138 157

Haemoglobin (Hb) concentration (g/L) in children at various chronological ages. From Vaughan [10].

Age-Related Biochemical and Physiological Normative Reference Ranges in Children

Water Content Water is the major component of the human body. On average, it represents 60% of the body weight in adult males. This figure is 50% in adult females aged 20–50, which is due to their higher fat content compared with males. In children, the water content varies considerably at different developmental and chronological ages [5, 6]. The relative water content is highest in premature and newborn babies ranging from 70 to 80% of the body weight. During the first 6 months of life there is a fast decrease of body water as percentage of body weight. This decrease continues at a slower rate between 6 months and 2 years and then decreases further up to 11 years of age when puberty starts. It then varies between 53 and 63% with no correlation to age or gender [7]. These marked changes of body water during childhood may affect the distribution in the body of lipophilic and hydrophilic substances, such as pharmaceuticals, at various chronological and developmental ages and in males and females. Fat Mass Average fat content of young adult males is 17–21% and of females is 25–30%. These values differ in prepubertal children [8]. Newborns of both genders carry 15% body fat, which increases to 25% at 2 months and 30% at 4 months. In later childhood, fat content decreases to 15–17% in boys and to 20–22% in girls. The final changes establishing adult values appear during pubertal development. As discussed above, BMI is often used as a proxy for fat content although it has limitations in not being able to discriminate the tissue that adds to the weight of the individual. A more accurate measure of fat content frequently used by clinicians is obtained from body densitometry by dual-energy X-ray scan [9].

Biology at a Young Age Differs from Biology at Later Ages

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

111

Common Laboratory Data Laboratory tests such as those for hematological parameters often have age- and gender-related reference values. Haemoglobin concentration, one of the most commonly employed blood tests, is a good example of such a situation. Table 1 shows haemoglobin reference values for children at various ages as employed by a hospital laboratory [10]. Obviously, if such parameters are evaluated in clinical or epidemiological studies it may be difficult to discriminate between age-, developmental- or study effects-mediated changes.

References 1 Tanner JM, Whitehouse RH: Clinical longitudinal standards for height, weight, height velocity, weight velocity, and stages of puberty. Arch Dis Child 1976; 51:170–179. 2 Lazarus R, Baur L, Webb K, Blyth F: Body mass index in screening for adiposity in children and adolescents: systematic evaluation using receiver operating characteristic curves. Am J Clin Nutr 1996; 63: 500– 506. 3 Mosteller RD: Simplified calculation of body surface area. N Engl J Med 1987;317:1098. 4 Karlberg J: A biologically-oriented mathematical model (ICP) for human growth. Acta Paediatr Scand Suppl 1989;350:70–94. 5 Friis-Hansen BJ, Holiday M, Stapleton T, Wallace WM: Total body water in children. Pediatrics 1951; 7:321–327. 6 Altman PL, Katz DD: Blood and other body fluids/ analysis and compilation; in: Altman PL, Dittmer DS (ed); Federation of American Societies for Experimental Biology. Biological Handbooks, Washington, 1961 p 540.

7 Péronnet F, Mignault D, du Souich P, Vergne S, Le Bellego L, Jimenez L, Rabasa-Lhoret R: Pharmacokinetic analysis of absorption, distribution and disappearance of ingested water labeled with D2O in humans. Eur J Appl Physiol. 2012;112:2213–2222. 8 McCarthy HD, Cole TJ, Fry T, Jebb SA, Prentice AM: Body fat reference curves for children. Int J Obes (Lond) 2006;30:598–602. 9 Dietz WH, Bellizzi MC: Introduction: the use of body mass index to assess obesity in children. Am J Clin Nutr 1999;70(suppl):123S–125S. 10 Vaughan VC: Reference ranges for laboratory tests; in: Behrman RE, Vaughan VC, (eds): Nelson Textbook of Pediatrics. Saunders, Philadelphia, 12th ed, 1983, p 1839. 11 Robbins WJ, Brody S, Hogan AG, Jackson CM, Greene CW: Growth. Yale University Press, New Haven, 1928. 12 Scammon RE: The measurement of the body in childhood; in: Harris JA, Jackson CM, Paterson DG, Scammon RE (eds); The Measurement of Man. University of Minnesota Press, Minneapolis, USA 1930 pp 173–215.

Olle Söder, MD, PhD Department of Women’s and Children’s Health Karolinska Institutet and University Hospital SE–17176 Stockholm (Sweden) E-Mail [email protected]

112

Söder

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 105–112 (DOI: 10.1159/000481327)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods Mandy Vogel · Tanja Poulain · Anne Jurkutat · Ulrike Spielau · Wieland Kiess Hospital for Children and Adolescents, Center for Pediatric Research, and LIFE Research Center for Civilization Diseases, LIFE Child, University of Leipzig, Leipzig, Germany

Abstract The present chapter aims to assist young scientists in preparing the different steps necessary to conduct a study. The description of different study designs and data sources as well as the presentation of different ways of data analysis represent key points of this chapter. Furthermore, we give an overview of available statistical software and provide information on data protection and standardiza© 2018 S. Karger AG, Basel tion.

Whenever you start a scientific project, there is a need to plan ahead and a number of basic considerations have to be thought about. Not everything presented as science in publications is compliant to good scientific practice. This is particularly the case for publications on the Internet. In this chapter, we offer basic information on how to get started. The following pages also provide an outline of the most important aspects that need to be considered when planning a cohort study, an epidemiological project, or the statistical analysis of a clinical trial.

Hypotheses

Studies can be exploratory or hypothesis-driven. In exploratory studies, many variables are correlated with the effect of interest. Some of the variables may show a more or less pronounced effect that could be of relevance and meaning. It is a bit like fishing. In contrast, if you plan a study, you should have clearly expressed hypotheses and

research questions, for one obvious reason: without a hypothesis, you cannot plan a study because you do not know whom to ask or what to ask for. A hypothesis must be stated in a testable way. Exposure and outcome variables have to be exactly specified, and the expected results have to be clearly expressed. An example: you are interested to know more about the relationship that exists between daily exercises and fat mass. Then your hypothesis would be something like this: Higher daily physical activity decreases fat mass.

But wait, it is not possible to measure daily exercises as well as fat mass directly. So you have to decide on how to measure exposure and outcome. You may ask the study participants about their daily exercise patterns, or you can measure their activity using an accelerometer. What would be the more precise way? What is feasible for you and your institution? The same is true for body fat mass: how to measure it? You can use height and weight measurements and calculate the body mass index (BMI) as a proxy for body fat mass; you may use skinfold measurements or bioimpedance analysis. Again, the question is: what is the best measure that is feasible for you? The decision depends not only on the available resources but also on the study subjects of interest. Some methods are more suitable for application in adults and cannot be used in small children, and questionnaires may have to be modified and validated for different age groups. Therefore, in most cases, it is necessary to focus on a confined age group: Higher daily physical activity (measured by accelerometry) decreases BMI and skinfold thickness (as a measure of body fat mass) in young adults (aged 16–18).

The hypothesis states that the activity has a direct influence on BMI. To show such an influence, an experimental study has to be conducted and this needs a lot of time, a number of staff, and, consequently, much money. So if you do not have the time and the money, you may have to change the hypothesis again: Higher daily physical activity (measured by accelerometry) relates to decreased BMI and skinfold thickness (as measures of body fat mass) in young adults (aged 16–18).

The next question to answer is whether or not there are any known or suspected influencing factors. These so-called confounders have to be taken into account and, therefore, the related data have to be collected. Common confounding variables are age and gender, race, medical history, socioeconomic background, history of smoking and alcohol consumption, and so on. In the case of children and adolescents, the puberty status might also be taken into account. Review of Literature As you can see, the generation of hypotheses is not a trivial task. Before you can specify the measures of exposure and outcome, the confounders and the expected results, an extensive and systematic literature review has to be done: explore the current literature with regard to the state of knowledge, contradictory reports, used methods

114

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

and approaches, and the final results; do not confine your literature to articles that support your hypotheses. To collect, organize, and annotate your literature you should use a reference manager from the very beginning. It helps you to store all your knowledge sources (research articles, websites, books, etc.) in a sustainable and well-arranged manner; plugins for MS Office, LibreOffice, similar word processors, or LaTeX provide easy ways of referring to any literature during the writing process. You will be able to find a list of reference managers and a comparison of their capabilities on the following web page: https://en.wikipedia.org/wiki/Comparison_of_reference_management_software. We recommend the use of a free software tool with network functionality like Zotero (https://www.zotero.org/) because its use is not limited to the time you are working at a particular university or research institution and it is independent of the usage of a particular device. In addition, it supports collaboration and the sharing of literature. To search (and find) scientific literature, academic search engines are indispensable. One of the most important search engines within the biomedical sciences is PubMed (www.ncbi.nlm.nih.gov/pubmed/); it is a free search engine, provided by U.S. National Library of Medicine. Other examples are Web of Science or Google Scholar. A more extensive list can be found on the following web page: https:// en.wikipedia.org/wiki/List_of_academic_databases_and_search_engines.

Study Design

The design of a study aims to describe how data is collected. It comprises the methods and procedures of collecting and analyzing data, the definition of exposure, outcome, and confounding variables, data collection strategies, the time schedule, the definition of the target population, the method of blinding (if applicable), the specification of intervention (if applicable), and so on. Design Types Medical research can be classified into basic (research without a focus on application) and applied research (problem-oriented). We will focus on the basic types of study design in applied research. Figure 1 outlines the most important design types. The main subtypes are descriptive and analytical studies. Examples of descriptive studies are sample surveys or the census that aim to assess the status of a population with regard to certain properties; they can provide baseline data and generate hypotheses for further investigations. On the other hand, analytical studies try to investigate causeeffect relationships or associations and follow one of 2 strategies: observation (e.g., cohort study, case-control study) or intervention (e.g., clinical trial, field trial). The main difference between these 2 strategies is that observational studies observe naturally occurring events, whereas interventional or experimental studies involve actions

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

115

Study design

Descriptive

Case series

Analytical

Experimental Census Observational

Sample survey Random

Non-random

Prospective

Cross sectional

Laboratory experiment

Clinical trial

Field trial

Retrospective

Longitudinal

Case-control

Cohort

Nested case-control

Other

No control

Fig. 1. Various types of study designs. Adapted from Indrayan [1], 2012.

to change the natural course of events. We will focus on designs of observational studies and only briefly address interventional studies. One of the most basic forms of observational studies are aggregate or ecologic studies. They collect and analyze data at an aggregated level. Outcome, exposure, and confounders are observed in groups of individuals as aggregated value. Therefore, no individual data will be collected. The units of observation can be institutions like kindergartens, schools, classes, or whole populations. Data are typically provided by the unit itself (e.g., annual statistics). The units have to be relatively homogeneous considering the exposure of interest; inappropriate grouping may lead to considerable bias. In addition, necessary data of scientific interest are often not available. Cross-Sectional Study In a cross-sectional study, both exposure and outcome, and all confounding variables are observed at the same time point. For example, you want to compare the prevalence of a specific disease in different population groups (e.g., the prevalence of asthmatic diseases in 3 to 18-year-old German boys stratified by the smoking status of the parents). This design allows comparisons between groups within the study subjects or investigations of relationships of characteristics. However, this strategy does not allow any conclusion on cause and effect. Therefore, the design might be more suitable for descriptive studies but can also generate hypotheses concerning etiology and, in this sense, does not remain exclusively descriptive. In addition, it might be subject to recall bias and is not as reliable as information collected in prospective studies, like cohort or longitudinal studies.

116

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Prospective Studies In the context of prospective studies, the term prospective describes the direction of investigation, which is from exposure to outcome. Thus, they are also called follow-up studies. Effects are usually measured in risk ratios or attributable risk. Cohort Study Cohort studies follow a longitudinal design. The study subjects share a defining characteristic like the year of birth, an occurrence of a certain disease, or a common profession. In a prospective cohort, the study subjects are followed up over time in the future. Therefore, the dropout rates of study subjects can lead to biased results because the dropping–out of study subjects usually does not occur at random, but drop-out subjects share similar qualities or circumstances. Another type of bias related to cohort (and other longitudinal) studies is the Hawthorne effect: People may tend to change their behavior when they are being observed. On the contrary, in a retrospective cohort, data are collected from past, individual records. Despite the name, this is not a retrospective study because the direction of investigation is still from exposure to outcome. A retrospective cohort may be more reliable and less biased than the memory power of individuals – however, “the devil is in the detail”: there was no predefined standard procedure for data capturing. Additionally, coding systems and procedures might have changed and records might be incomplete or not well and consistently structured. But, because all of the events (especially exposure and outcome) of interest have already taken place in the past, one does not have to deal with dropouts or subjects lost to follow-up. There are also other longitudinal studies. The most distinctive characteristic is that observations are made at different time points on the same subjects. The time intervals can vary from minutes (e.g., hormone levels 5, 10, 30, 120, 240, and 300 min after swimming in icy water) to many years (where the latter is typical for cohort studies, for example, the effect of birth weight on glucose tolerance in adulthood). For each follow-up visit, a tolerance for the follow-up time also has to be defined: if a participant cannot come to the follow-up visit exactly one year after the baseline visit, what difference are you willing or able to tolerate? 5 weeks earlier, 4 weeks later, or even –/+ 2 months? The appropriate time intervals between the study visits as well as the tolerances depend on the subject, the developmental stage of the subjects, the available resources, and last but not the least upon the questions that are asked. The central aim of longitudinal studies is to describe the trend of values over time. Retrospective Studies The first variable to be assessed in a retrospective study is the outcome. The direction is reverse – from outcome to exposure. Relevant retrospective study designs are casecontrol and nested case-control studies.

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

117

Case-Control Study In a case-control study, subjects with a given outcome are identified (e.g., children who have allergic asthma). In a second step, one has to find one or more controls for each of the cases that do not show the outcome (e.g., do not have allergic asthma). Importantly, each of the case – control(s) pair has to be similar in terms of predefined control variables like gender, age, weight status, or other possibly confounding characteristics. In the next step, the exposure status (e.g., exclusive breastfeeding between the age of 0.5 and one year) of cases and controls will be evaluated. Respective differences (measured in odds ratios) allow the detection of potential risk factors for the development of the specific outcome. If cases as well as controls are selected from an existing cohort, the design is called nested case-control study (the case-control setup is nested within a cohort study). Interventional Studies All the studies above represent observational studies. If you plan an intervention which involves the study participants (send them to a boot camp or treat them with a new drug), you conduct an interventional or experimental study. In an interventional study, a specific condition is willingly induced to verify or falsify a specific hypothesized causal relationship. When it comes to interventional studies, important concepts concern the control group and randomization. Individuals undergoing an intervention tend to behave and react differently (placebo effect, Hawthorne effect). To control for such effects, a control group is desirable. As described for case-control studies, the control group should be similar concerning confounding variables. Therefore, the assignment of participants to the control or the treatment group should be randomized and this supports (but not guarantees) similar distributions of characteristics. Besides simple randomization, there are also stratified and block randomization, which additionally support a similar distribution of important characteristics and similar group sizes during the recruitment. For intervention involving human individuals, ideally, neither the participant nor the assessor knows to which group the participant belongs (double-blind study). If the information is only withheld from the participant, the study is called a single-blind study. Clinical trials are a prominent example of intervention studies. In clinical trials, new treatments are evaluated in terms of the safety and efficacy of the new treatment compared to an established standard or no treatment at all. In most cases, specific outcomes of interest (e.g., healing success) occurring in the experimental group (e.g., patients treated with a new medication) are compared to those of a control group (e.g., patients treated with the conventional treatment or not treated). Clinical trials are subject to legal regulations and ethical considerations. Standards To ensure to fit the requirements and needs of reporting the results of the different study types and to define uniform standards, guidelines are developed and published. You will find a list of existing standards in Table 1.

118

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 1. Published standards for study designs Standard name

Acronym Website

Consolidated Standards Of Reporting Trials

CONSORT www.consort-statement.org

Strengthening the Reporting of Observational studies in Epidemiology

STROBE

Standards for Reporting Studies of Diagnostic STARD Accuracy

www.strobe-statement.org www.stard-statement.org

Quality assessment of diagnostic accuracy studies

QUADAS www.bris.ac.uk/quadas

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

PRISMA

www.prisma-statement.org

Consolidated criteria for reporting qualitative COREQ research Statistical Analyses and Methods in the Published Literature

SAMPL

Consensus-based Clinical Case Reporting Guideline Development

CARE

www.care-statement.org/

Standards for Quality Improvement Reporting Excellence

SQUIRE

www.squire-statement.org

Consolidated Health Economic Evaluation Reporting Standards

CHEERS

www.ispor.org/taskforces/EconomicPub Guidelines.asp

Enhancing transparency in reporting the synthesis of qualitative research

ENTREQ

Standard protocol items for clinical trials

SPIRIT

Meta-analyses of observation studies in epidemiology

MOOSE

http://www.spirit-statement.org

Further Reading More general information on study design can be found in [1–3]. A nice summary article written by Thiese was published in 2014 [4]. And an introduction to randomization is given in [5].

Data Sources

There exist different ways of acquiring data needed to answer your research questions. The decision for a particular instrument mainly depends upon the research question, the age range of the study participants, the time assigned to the assessment, as well as your financial resources and staff availability. Here, we will give an overview of the most basic methods of data collection (examinations, tests, questionnaires, interviews, secondary

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

119

sources). We thereby outline advantages and disadvantages as well as other potential aspects one has to take into account when deciding in favor of a particular method. Examinations Examinations are the best (and usually the only) way to assess several body functions. They therefore often represent an essential element of medical studies. Physical, instrumental, and laboratory examinations can be distinguished. Physical examinations include classical examinations that are performed using no instruments or only very basic instruments, like stethoscopes. Examples are the auscultation of heart and lungs, visual inspection of the skin, examination of reflexes, or classical body measuring. Instrumental examinations, in contrast, are examinations that are mainly performed using instruments. Examples are the analysis of lung function via spirometry, measurement of blood pressure via blood pressure gauge, examination of heart function via electrocardiography, or anthropometric measurements via a 3D-body scanner. Compared to physical examinations, instrumental examinations represent a more objective way of measurement. For example, classical body measuring by different study assistants bears the risk – even if they are well trained – of inter-observer discrepancies, whereas a body scanner might be not as prone to show such discrepancies. Disadvantages of instrumental examinations are the costs of the instruments, the maintenance, the routine calibration, and the need for specialists familiar with the correct usage. Imaging techniques (e.g., MRI) represent a particular form of instrumental examinations. They provide a detailed picture of participants that could be used for a variety of analyses. However, you should have in mind that imaging techniques are costly, not always appreciated by study participants, and that the analysis of imaging data is very complex and time-consuming. Also, the risk of incidental findings is particularly high in studies using imaging techniques. Laboratory examinations refer to the analysis of biological samples such as blood, urine, feces, hair, or breast milk. They allow, for example, inferences on the intake of minerals or vitamins and the physical or, in the case of hair cortisol, even psychological health status of your study participants. It is, therefore, also possible to draw conclusions on metabolic processes. Furthermore, laboratory measurements may detect diseases or medical abnormalities before they are clinically apparent or detectable by other physical or instrumental examinations, for example, genetic or allergic predispositions. They may also confirm a clinical suspicion. Disadvantages are that blood samplings do not rank among the preferred measurements of study participants. Additionally, storing biological samples represents an organizational and financial challenge (biobanking). Tests Tests are the best way to assess capabilities (e.g., intelligence, attention, language comprehension, motor skills) of participants in an objective way, especially if the test can be presented using a computer. Also, computer-based assessments require fewer per-

120

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

sonnel resources. As soon as the test has to be executed by study assistants, optimal training is essential in order to ensure standardized testing. This might be challenging, particularly if test instructions are vague and invite study assistants to rely on individual judgments. The day’s form, the temperament, and the motivation of the test subject can have an enormous impact on the test result and, therefore, definitely have to be taken into account. You might search for an appropriate test in a library or by using different online platforms. The use is often restricted and requires the payment of a license fee. We recommend that you do not create tests on your own, as test development is a very complex issue (far more complicated than creating a questionnaire); they have to be piloted and validated. Questionnaires Questionnaires represent a quick and inexpensive way of data collection, especially if answers can be directly entered into an electronic input system. However, you have to keep in mind that questionnaires are a very subjective way of data assessment. Subject A might understand a question differently than subject B. Another disadvantage of questionnaires is that you can hardly verify if the given answers are true or not. Even if participants may be more likely to give honest answers to questions of an anonymized questionnaire than to questions asked in a direct face-to-face interview, participants might intentionally manipulate answers. Standardized questionnaires have been shown to be reliable and valid, have been tested in a representative sample of people, and usually provide reference values allowing you to compare the results of your study sample with this representative sample. Standard questionnaires furthermore allow comparisons of your study results with results of other studies using the same questionnaire. Before creating questionnaires on your own, it is, therefore, advisable to check the availability of validated standard questionnaires fitting your research questions. If no standard questionnaires conform to your concepts, you might have to create a questionnaire on your own. Questionnaire development is a complex issue and entire books deal with this topic. Here, we only want to mention some aspects you should have in mind when creating questionnaires. Questionnaire development takes time! You have to decide which characteristics you want to assess and which format the questions should be of. Furthermore, you have to test the questionnaire to ensure that the questions are understood as you intended them to be. Finally, you may be obliged to make some changes, to retest the new version, to make further changes, and so on. Answering the questions of your questionnaire should be feasible and easy. Questions should not be too complex, as complexity might reduce comprehensibility and compliance. However, you should also ensure that the question contains all the information needed to answer it correctly, for example, (if appropriate), an adequate reference period (“during the last year”, “during the last week,” etc.) or an example answer (Tables 2, 3). You carefully have to consider who should complete the questionnaire, that is, who is the best informant (e.g., child, parents, teachers, friends). You might also be inter-

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

121

Table 2. The following question does not contain precise formulated information How much do you smoke currently?

A. I do not smoke B. Somewhat C. Regularly D. I am a strong smoker

Analyses of the resulting data would be hard to describe and interpret.

Table 3. The following question contains a precise reference period and a precise definition of the number of cigarettes How much did you smoke during the last 6 months?

A. I do not smoke B. 1–10 cigarettes per day C. 10–20 cigarettes per day D. >20 cigarettes per day E. I quit smoking in this period

ested in comparing answers given by different informants, for example, mother vs father, parent versus child (multiple informant approach). When deciding on the order of your questions, keep in mind that earlier questions might have an impact on the interpretation of the following questions. For example, if the question “How often did you eat chocolate during the last week?” precedes the question “How often did you eat sweets during the last week?”, participants may consider that sweets exclude chocolate (which you might not intend). You should avoid such ambiguities by accurate and selective wording: “How often did you eat sweets (including/excluding chocolate) during the last week?”. An important issue in the process of questionnaire creation concerns the decision of an appropriate answer format. You have to decide on the answer format and the concrete form of your questions. Most questionnaires contain closed-ended questions, with “yes/no” questions (e.g., “Are you taking any medicine at the moment?”) representing the simplest form. Closed-ended questions with categorical response options consist of a question and different, precise answer categories. To ensure that participants find the appropriate response category, it is important to provide exhaustive and mutually exclusive answer options (Table 4). However, sometimes you may allow (or even explicitly ask) participants to give multiple answers (Table 5). Response or Likert scales represent another form of closed-ended questions. They usually comprise several statements and a response scale specifying the level of agreement, belief, or attitude ranging between (strong) refusal and (strong) approval. They are often presented as a matrix of questions and are especially suitable for the assessment of attitudes and personality/temperament (Table 6). If you are deciding for closed-ended questions, you should also consider how many answer categories you provide and if you give participants the possibility to select a

122

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 4. Two examples of closed-ended questions. The participant is expected to choose one out of the given answers. Therefore, the alternatives have to be exhaustive and mutually exclusive Where have you grown up?

A. In a city or village with less than 5,000 inhabitants B. In a city with more than 5,000 inhabitants C. In a city with more than 10,000 inhabitants D. In a city with more than 10,000 inhabitants

What is your favorite ice cream flavor?

A. Chocolate B. Vanilla C. Strawberry D. Other E. I don’t like ice cream

Table 5. Two examples of closed-ended questions (multiple answers). The participant is expected to choose all of the choices that apply For what reason did you consult your general physician in the last 12 months?

A. Chronic disease B. Medical check-up C. Follow-up care D. Other reason(s)

Do you have any kind of indoor pets?

A. None B. Dog(s) C. Cat(s) D. Bird(s) E. Fish F. Reptile(s) G. Other

Table 6. Examples of questions using a Likert scale. The questions are usually designed as matrix Strongly agree

Agree

Neutral or unsure

Disagree

Stongly disagree

I am a sociable person I enjoy meeting new people I like cats I hate spiders

neutral answer category (e.g., “unsure,” “neutral,” or “neither yes nor no”) or to refuse an answer (e.g., “I don’t know” or “I can’t say”). In contrast to closed-ended questions, open-ended questions ask participants to report their response verbatim. This might be suitable if you are interested in easily quantifiable aspects (e.g., “How many siblings do you have?”, “How often have you been hospitalized in the last 12 months?”) or if you do not know how the possible answers might be summarized by appropriate answer categories (e.g., “What would you do if you were offered one holiday tomorrow?”). The coding and interpretation of

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

123

T0

T1

T2

T3

T4

Date of measurement Height, cm Weight, kg Smoking status (0/1) Disease status (0/1)

Date of measurement Height, cm Weight, kg Smoking status (0/1) Disease status (0/1)

Date of measurement Height, cm Weight, kg Smoking status (0/1) Disease status (0/1)

Date of measurement Height, cm Weight, kg Smoking status (0/1) Disease status (0/1)

Date of birth Gender (male/female) Year of birth mother

Date of measurement Height, cm Weight, kg Smoking status (0/1) Disease status (0/1)

Fig. 2. Example of a study with 1 baseline and 4 follow-up visits. At baseline, the person characteristics (a) date of birth, (b) sex, and (c) year of birth of the mother will be collected. These 3 parameters are assumed to be constant for the whole duration of the study. (1) date of measurement, (2) height, (3) weight, (4) smoking status and (5) disease status will be collected at every visit, as their values might change during the follow-up period.

open-ended questions can be difficult and time-consuming. Importantly, the choice of a specific answer format influences the scale level of the assessed variables and, therefore, has an impact on later data analysis, that is, on the statistical methods you will be able to apply (see the next section). Further information on designing questionnaires and surveys can be found in [6–9]. Interviews Interviews are the interactive form of asking questions and giving answers. They may take place face-to-face or via telephone. An advantage of interviews is that they enable data collection of participants (e.g., children) that are not able to read. Additionally, questions can be explained in more detail by the interviewer. This can be useful if questions might easily be misunderstood. Furthermore, specific questions might be switched or added throughout the interview as a function of already given answers or the individual background of the interviewee. However, a disadvantage is that interviews are less anonymous than questionnaires. Interviewees might, therefore, pretend socially desired behavior and give incorrect answers. For very personal information, it might, consequently, be more suitable to ask questions via questionnaires. Interviews are also prone to an interviewer bias (gender, age, the behavior of the interviewer). Therefore, a standard operating procedure and well-trained interviewers are necessary pre-conditions. Secondary Sources You may sometimes be interested in data that have already been collected on another occasion, for example, at previous medical visits. If the data of interest are well documented, for example, in health records, you may simply use this data instead of col-

124

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

I do not work Part-time, less than 20 h/week

0

5

10

15

Full-time, less than 45 h/week

Part-time, more than 20 h/week

20

25

30

35 40 45 Hours per week

Full-time, more than 45 h/week

50

55

60

65

70

75

80

Fig. 3. The 5 categories “I do not work”, “part-time, less than 20 h/week”, “part-time, more than 20 h/ week”, “full-time, less than 45 h/week” and “full-time, more than 45 h/week” reflect the underlying numerical time variable. Only for the answer “I do not work” we can determine the exact numerical value. Each of the remaining categories reflects a time-interval representing a variable length of time.

lecting it on your own. This strategy is especially useful if the information concerns aspects the participants could hardly reproduce (e.g., specific medical parameters, vaccination status). A disadvantage is that some health records might be incomplete or, if handwritten, illegible. Additionally, if the health records were collected by different physicians or institutions, the data might hardly be comparable. Furthermore, you should have in mind that the transfer of the data into your database may be time consuming, maybe even more time consuming than collecting data on your own. Other secondary data might be obtained from agencies, school records, openly available data (e.g., characteristics of the hometown of your participants), and so on. Data protection issues have to be acknowledged. Repeated Data Collection If you plan a longitudinal study (see Figure 2), in addition, you have to think about which items you want to collect at which visits and, also, which items are fixed (like birthday or gender) and which items are variable (like weight or disease status). Not all items are treated the same way for every cohort: if you, for example, follow older adults from 60 to 80 years of age, the last highest completed level of education should be a fixed item for almost all of the participants. On the other hand, in a cohort followed from 15 to 35 years of age, education levels will change for almost all of the participants. So, think about the items (or instruments) you want to collect and decide for each of them whether you want (or have) to collect it once (when?), at some of the visits, or at all of them repetitively. Group your items accordingly. For example, if you conduct a longitudinal study on nursing women and you ask them at all visits about weight before pregnancy it could happen that some women will retrospectively lose weight (and you need a plan

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

125

how to deal with inconsistent data). Asking as many questions as necessary but not more saves not only your time but also the time of the participants. For the same reason: do not collect data that you can calculate, for example, if you know the birthday and the day of measurement, you do not need to ask about current age. Ethical and Moral Issues Every datum you collect on your subjects contains personal or private information. Therefore, data are to be treated with discernment and data protection is always important and to be adhered to. All examinations, especially the invasive ones like blood withdrawal, should be done with care and respect. All questions you ask should also reflect respectfulness for the individual and the time he or she sacrifices. Avoid potentially disrespectful questions or unnecessary procedures at all costs!

Statistical Analyses

Scales of Measurement All collected characteristics are measured on specific scales. The scale of a measurement defines how to interpret a specific value. It is also crucial in determining the basic set of applicable statistical methods. There are different ways to categorize scales. The most used and cited classification of measurement scales is the scale developed by Stanley Smith Stevens (Stevens and others, 1946). It consists of the nominal, ordinal, interval, and ratio scale. The most practice-oriented classification is the distinction between nominal, metric, and ordinal variables. In most statistical software packages, corresponding variable types exist. In addition, there are special types of scales that do not fit well into these categories like date and time (they might be represented as a numerical value) or geospatial data. Another example of data that does not fit perfectly in a category is count data. A basic classification including important properties is shown in Table 7. Nominal scales are based on mutually exclusive categories like male and female (gender), different blood groups, or genotypes. These categories do not have any natural order and cannot be represented meaningfully by numerical values and, therefore, can not be added, subtracted, multiplied, or divided (even if they are coded as numbers). The only valid mathematical operations are equality or set membership. The central tendency is measured with mode as the most common category. If a characteristic is represented by exactly 2 categories, it is called dichotomous or binary variable. Prominent examples for categories of dichotomous variables are: yes/no, success/failure, or male/female, often coded as 1/0. Statistical procedures are much simpler for dichotomous variables than for those with more categories (polytomous). Metric scales, on the other hand, provide a measurement value regarding quantity. The precision and the resulting range of possible values depend on the measurement method but is in principle infinite. Therefore, comparisons and calculations are meaningful. There are 2 subtypes, interval (no absolute zero) and ratio scales (include

126

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 7. Overview of basic scale types and important characteristics

Type

Mathematical operations

Relations

Discrete/ continuous

Qualitative/ quantitative

Examples

Valid location parameters

Nominal

Frequency

Equality

Discrete

Qualitative

Gender, genotype, favorite ice cream flavor

Mode

Ordinal

Frequency, ranking

Equality, ranking

Discrete

Qualitative

Likert scales, age group, preferences

Mode, median

Equality, ranking, Continuous ratio of intervals

Quantitative

IQ, temperature degree Celsius, calender time

Mode, median, mean

Equality, ranking, Continuous Frequency, ranking, addition, subtraction, ratio of intervals, ratio of values multiplication, division

Quantitative

Height, weight, velocity

Mode, median, mean

Metric Interval Frequency, ranking, addition, subtraction Ratio

absolute zero). Interval scales do not allow assumptions about ratios between numbers. An example of an interval scaled variable is the quantitative variable IQ. It is incorrect to draw conclusions about the ratio of the IQ values of 2 individuals like “individual x is twice as intelligent as individual y.” In contrast to interval scaled variables, ratio scaled variables have an absolute zero. Therefore, assumptions on ratios between different values are possible (by multiplying or dividing numbers). Examples of ratio scaled variables are height, weight, or distances. An ordinal scale is something between a nominal and a metric scale. It is also described by categories. However, in contrast to a nominal variable, the categories of an ordinal variable have a precise order. Therefore, they are “quantitative” in a certain sense but lack accuracy and precision when compared to metric scales. They originate when genuinely metric characteristics like age or income are recorded in categories or when characteristics cannot be measured directly (e.g., severity of pain, attitude). Typical examples are questions answered on a Likert scale (e.g., strongly disagree [1] – disagree [2] – neutral [3] – agree [4] – strongly agree [5]). The order is obvious. However, the distances between 2 categories cannot be exactly quantified. In most cases, it is also impossible to tell if the distance between disagree (2) and strongly disagree (1) is the same as the distance between agree (4) and strongly agree (5). Therefore, an ordinal variable might summarize an underlying numerical variable (Figure 3). In certain research domains as well as in some statistical procedures, such underlying “latent” variables (or “traits”) are assumed. Some of them could be measured directly (like working hours in our example), but most of them are more abstract concepts like quality of life or attitudes. Depending on the scale of measurement, there are different ways to summarize data.

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

127

Further information on variable types and measurement scales can be found in [10–14]. Descriptive Statistics Every statistical analysis starts with a descriptive data analysis. It aims to understand and describe the features of a specific data set by summarizing and simplifying a more or less large amount of data into characteristic numbers. The 3 main characteristics of a metric variable are the distribution, the location parameter (or central tendency), and the scale parameter (or variability, dispersion). Distributions: the most widely known distribution is the normal or Gaussian distribution. Several statistical tests assume normality. There are formal tests of normality (e.g., Shapiro-Wilk test, Kolmogorov-Smirnoff test, Anderson-Darling test) as well as visualizations (histogram, quantile-quantile or QQ plot). We do not recommend to trust tests blindly. They often do not show any significant deviation from normality if the number of cases is low (lack of power) but show significant deviation from normality in almost all tests if the number of cases is high enough (overpowered). In practice, many tests (like the t test) are robust against deviations from normality if the number of cases is high enough. Therefore, blind testing might be misleading. Plotting the empirical distribution as histogram or as QQ plot can give a first impression. Normally distributed values should be symmetric around the mean. This can be also expressed in numbers: the mean and the median should not be very different on the scale of the data. If, for example, the distribution is right-skewed (mean >> median), a logarithmic transformation can result in normally distributed values. Other common distributions are the binomial distribution (binary data), the Poisson distribution (count data), F- and t-distribution, and many more. Location: the location parameter or central tendency describes the central position or a typical value of the variable. In case of the normally distributed values, the location parameter is the arithmetic mean. Other well-known location parameters are the median, trimmed means, mode, geometric and harmonic mean. Scale: the scale parameter is a measure of the variability or spread of the data. The most used parameters of scale are the variance, the SD, the interquartile range, and the range. Other, but less used, parameters of interest are skewness and kurtosis. The choice of used parameters should show the crucial properties of the data. For example, normally distributed data can be well summarized using the mean and the SD, the 2 parameters defining the distribution completely. On the other hand, these 2 parameters are inappropriate if the data are highly skewed. In such cases, the first and third quartiles and the median or the combination of mean and median can give a more appropriate summary. In the case of discrete variables, the distribution of the data can be shown by a frequency table or a table of proportions. The former one contains the frequencies or counts, whereas the latter one contains the relative frequencies or percentages. Do not restrict summarizing your data to numerical summaries. Summarizing

128

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 8. Statistical procedures for hypothesis tests on locations. Adapted from Indrayan [1]

Number of samples

Conditions

Sample or group size (s)

Model/test

One sample

Normal

Any n

One sample t test

Two samples

Paired, normal distributed differences

Any n

Paired t test

Paired, non-normal

Any n 5 ≤ n ≤ 19 20 ≤ n ≤ 29 n ≥ 30

Sign test Wilcoxon signed-rank Wilcoxon signed-rank with normal approximation paired t test

Unpaired, normal

Any n

Two sample t test

Unpaired, non-normal

n ≥ 30 10 ≤ n ≤ 29

Two sample t test Standardized Wilcoxon rank-sum Wilcoxon rank-sum

4≤n≤9 Three or more

One-way or two-way, normal Any n One-way, non-normal

n≤5 n≥6

ANOVA Kruskal-Wallis Kruskal-Wallis chi-squared approximation

means always a loss of information. Visualize your data to learn its key characteristics and peculiarities. “The greatest possibilities of visual display lie in vividness and inescapability of the intended message. A visual display can stop your mental flow in its tracks and make you think. A visual display can force you to notice what you never expected to see” (Tukey 1990). Group Comparisons One of the most frequent tasks in statistics is the comparison of groups. These comparisons are carried out in terms of comparisons of the above-mentioned parameters. Again, the test depends upon the measurement scale, but also on the chosen parameter and the number of cases. In the case of metric (quantitative, continuous) variables, the most common analysis is the comparison of the center or the location, which is mostly represented by the mean or the median. Table 8 helps you to choose the appropriate statistical test when comparing the means or medians of quantitative variables. Table 9 gives an overview of statistical tests applied when comparing proportions defined by qualitative (categorical) variables. As you can see, the choice of a specific test depends on the number of groups you want to compare, on whether the groups are dependent or independent, and on your sample size. For quantitative variables, it is, furthermore, important to consider the distribution of the variable. For qualitative variables, it is important to further differentiate between binomial, multinomial, or ordinal variables.

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

129

Table 9. Statistical procedures for hypothesis tests on proportions Number of samples/ distribution

Conditions

Sample sizes

One sample, binomial

Independent trials Any n Large n

One sample, multinomial Independent trials Large n Small n Two samples, binomial

Model/test

One sample proportion test When approximate normality can be assumend: Z-test Chi-quared test Exact multinomial test

Unpaired

Large n Chi-squared, G test of contingency Small n (one or more Fisher’s exact test expected frequencies smaller than 4 or 5)

Paired

Large n Small n

McNemar test Exact test on number of cases which are differently classified (Agresti, Chapt. 11.1 [15])

Three or more, binomial vs. multinomial

Unpaired

Large n

Chi-squared, G test of contingency

Three or more, binomial vs. ordinal

Unpaired

Large n

Cochran-Armitage trend test

The number of groups is defined by the number of levels of the variable(s) you use to stratify your sample (e.g., age groups, groups defined by different weight or BMI values, or treatment vs. control group). In some cases, you might consider only one sample, for example, your entire study population or one specific subsample, and compare the mean in this sample with an already known mean (e.g., the mean IQ, or a mean published in a former study). However, in most cases, you might compare 2 (e.g., 2-year-old children with 5-year-old children, or boys with girls) or more groups (e.g., the 3 possible genotypes of a 2 allele gene). The differentiation between dependent or independent samples is another distinction that has an impact on the choice of statistical test. In dependent samples, each subject of one group has an exact match in the other group. Dependent samples can (but do not have to) comprise the same study participants in different conditions or at different time points (e.g., at baseline and at the first follow-up). Twin studies, with one twin representing the match of the other twin, are another prominent example for dependent samples. In general, whenever a meaningful one-to-one matching is part of the study design, tests for dependent samples are appropriate and should, due to their higher statistical power, be preferred to tests of independent samples. Independent samples, on the other hand, comprise study participants without individual matching (e.g., a group of one-year-olds and a group of 10-year-olds, or a group receiving medication A and a group receiving medication B). For variables on a metric scale, the distribution of the variable in the target population has to be considered

130

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

(which is – due to fluctuations – not necessarily the same as the distribution of the sample). Importantly, some statistical tests provide stable results for large sample sizes even in the case of deviation from the test conditions. For example, normal distribution (or normality) is a condition for the application of t tests. However, for group sizes n >30 (not for smaller group sizes!), the t test is stable against deviations from normality and should be preferred to nonparametric tests (which, in general, require no normality). Therefore, the sample size also influences the choice of test. Some statistical tests provide stable results for large sample sizes. Other tests, in contrast, should be used in small samples but not in larger ones. Given all necessary information on the variable type, sample size, number of (dependent or independent) groups, and distribution of your variable, you are able to select the appropriate test to test your differences (of means, medians, or proportions). Here are some examples: – Imagine you want to compare the mean IQ of 5-year-old children (n = 50) with the mean IQ of 10-year-old children (n = 50). You have, thus, a normally distributed quantitative variable (IQ) and 2 independent samples (5-year-olds and 10-yearolds), each with a sample size of 50. Therefore, you have to apply a 2-sample t test (see Table 8). – If you want to compare the mean IQ of your whole study sample (one sample) with the mean IQ in the general population (IQ = 100), you have to apply a one-sample t test (see Table 8). – If you were interested in comparing the value of a specific serum parameter in patients (n = 15) before and after therapy with a specific medication, you would have a (not normally distributed) quantitative variable and 2 dependent samples. You, therefore, should apply a sign test or a Wilcoxon signed-rank test (see Table 8). – Imagine you want to compare if the marital status (4 answer categories) of the parents of your study participants differs between girls (n = 50) and boys (n = 50). You thus have a qualitative variable (marital status) with a multinomial answer format (4 answer categories), and 2 independent samples (girls, boys). In this case, you could apply the chi-square test or the G test of contingency (see Table 9). – If you were interested in comparing 3 age groups (e.g., 2-, 5-, and 8-year-old children), with n = 20 in each group, in terms of their mathematical achievement (e.g., failure/success in mathematical test) you would have a binary variable (failure/success), and 3 independent samples (2-, 5-, and 8-year-old children). Therefore, the most appropriate statistical test would be the chi-squared test (see Table 9). There are also tests of the equality of variances, for example, the F test, the Levene’s test, the Bartlett’s test, or the Brown-Forsythe test. Relations The question on further possible relations between variables (e.g., between different blood parameters and children’s wellbeing, between children’s physical activity and

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

131

Table 10. Basic statistical methods for studying relationships Outcome

Predictor(s)

Model

Key words (software menus)

Continuous

Continuous One or more continuous Mixed (discrete and continuous) Covariates

Correlation (Multiple) linear regression ANCOVA

Correlation Linear model/linear regression Linear model/linear regression

Binary

Mixed (discrete and continuous) Covariates

Logistic regression

General linear models/logistic regression

Count

Mixed (discrete and continuous) Covariates

Poisson regression

General linear models/poisson regression

Survival

Groups

Life table Kaplan-Meier Log-rank

Life table Kaplan-Meier Log-rank

their BMI, between a family’s socio-economic status and children’s leisure behavior) represents another key question in the context of child health examinations. As for the investigation of differences between groups, several statistical models exist to examine possible relations between variables (see Table 10). When examining relations between variables, predictor (or independent) and outcome (or dependent) variables have to be distinguished. Predictor variables are variables that might explain some variance (differences) of an outcome variable, that is, predictor variables represent the explanatory variables, whereas outcome variables represent the variables being explained. The most decisive factors that have to be taken into account when choosing the appropriate statistical model are the measurement scales of the predictor or outcome variables (see Table 10). Examples: If you are interested in the relation between age and the score in an intelligence test, that is, in the relation between continuous variables, you have to apply a correlation. If you are interested in how age (a continuous variable), gender (a binary variable), and socio-economic status (a qualitative variable) predict success or failure (a binary outcome variable) in a mathematical test, you have to apply a multiple logistic regression. These synopses of tests and models should not be used as a recipe for data analyses. They are more like a shopping list for the first meal you will ever prepare. They are neither exhaustive nor rigid. You will increase your knowledge about methods over time.

Graphics

Visualization is an important part of every statistical analysis. It is useful for inspection, data cleaning, exploring the data structure, detecting trends, clusters, and unexpected features, and so on. Later on, graphics can be used to evaluate the quality of

132

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 11. Basic types of data display (plots) dependent on the number and scale of variables Number of Scale variables variable 1 1

2

3

Scale variable 2

Scale variable 3

Graphics

Continuous

Histogram Density plot Stem and leaf plot Boxplot Violin plot Letter-value plots 1D scatterplot (index plot) QQ plot Beeswarm plot

Discrete

Barplot Pie chart

Continuous

Continuous

Scatterplot Scatterplot including smoothed curve 2D density plot

Discrete

Continuous

All plots for 1 continuous variable conditional on the discrete variable

Discrete

Discrete

Stacked or grouped barcharts Pyramid plots

Continuous

Continuous

Continuous

Scatterplot including a continuous color scale on the third variable Bubble chart Contour and filled contour plot 3D scatterplot

Continuous

Continuous

Discrete

Scatterplot, colour or shape mapped to the 3rd variable Scatterplot including smoothed curve conditional on the 3rd variable

models. Finally, graphics are essential in presenting and communicating the results of an analysis. Exploration graphics should be used to get to know your data. Visualization of data can transport much more information than numerical summaries; it is essential in understanding the data. Use color and shape of points and lines to project more than 2 dimensions of the data on a 2-dim plot. As for statistical models, the appropriateness of a graphical display depends on the number of variables and their scale. In Table 11 and Figure 4 we present some basic types of graphics. Of course, there are a lot more: special graphics to visualize spatial data, categorical data, results of an analysis of variances and so on. The final graphics should be simplified to the essential ones. The data presented in the graph is the reason of its existence. Make them stand out, avoid any superfluity like unnecessary 3D effects.

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

133

Histogram

Density plot

Boxplot

Letter-value plot

Viollin plot

Beeswarm plot

Index plot

QQ plot

Barplot

Pie chart

Scatterplot

Scatterplot + smoothed fit

Contour plot

2D density

Stacked barplot

Grouped barplot

Pyramid plot

Conditional boxplot

Conditional letter-value plot

Conditional scatter +smoothed fit

Fig. 4. Examples of basic types of data display. The first 2 rows of plots show plot types for visualization of the distribution of 1-dimensional continuous variables. Bar and pie chart visualize the distribution of 1-dimensional discrete variables. Scatterplots, contour plots, and the 2D density plot visualize two continuous variables resp. their relationship. Stacked and grouped barplots and the pyramid plot visualize the relationship between 2 discrete variables. The conditional boxplot and the conditional letter-value plot show the distribution of a continuous variable conditionally on a discrete variable. The last plot shows the relationship between two continuous variables conditionally on a discrete variable.

134

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Hypothesis Testing and p Values

Statistical tests are based upon some more or less mystical summary statistics like the t value in the t test, the F value in the analysis of variance or the W value in the Wilcoxon-tests. The t value in the t test for the difference between the means of 2 groups, for example, combines the difference itself, the variance of the data, and the sample size to one number – t. Under the assumption that there is no difference (the null hypothesis: H0), t follows a specific, predefined distribution (not surprisingly called t distribution). The p value indicates the probability of having the same or a more extreme test statistic given the null hypothesis is true – p(data|H0). It is not the probability of the null hypothesis being true given the data p(H0|data). These 2 probabilities are not the same, as illustrated in the following example from Carver 1978 [16]: What is the probability of obtaining a dead person (D) given that the person was hanged (H); that is, in symbol form, what is p(D|H)? Obviously, it will be very high, perhaps 0.97 or higher. Now, let us reverse the question: what is the probability that a person has been hanged (H) given that the person is dead (D); that is, what is p(H|D)? This time the probability will undoubtedly be very low, perhaps 0.01 or lower. No one would be likely to make the mistake of substituting the first estimate (0.97) for the second (0.01); that is, to accept .97 as the probability that a person has been hanged given that the person is dead. Even though this seems to be an unlikely mistake, it is exactly the kind of mistake that is made with the interpretation of statistical significance testing – by analogy, calculated estimates of p(H|D) are interpreted as if they were estimates of p(D|H), when they are clearly not the same. Think carefully what that means: if you got a p value of 0.01 (a p value that makes you reject your null hypothesis), the probability of the null hypothesis being true can be 0.97 and, vice versa, if you got a p value of 0.97, the probability of the null hypothesis being true can be 0.01. This is the reason why you should never reason on the p value alone; always consider its meaning in the context of effect sizes, model fit, your subject (what do you know about mechanisms, pathways, etc. – it is worth a thousand p values), and common sense. Further general information on statistical analyses can, for example, be found in [1, 17]. A more in-depth but also comprehensive summary of general statistical methods is given in the book of Heiberger and Holland [18]. The principles of graphics display are described in [19–24]. Critical literature on the correct usage of hypothesis testing and p values is provided, for example, by Carver [16], Thiese [25], and Wasserstein [26].

Statistical Software

There are tons of statistical software out there. To make a choice might not be easy. Some of them are easy to use but comprise only a limited range of functions, some are freely available, whereas others are expensive; some are extensible but they seem to be

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

135

intimidating at first sight. We list the most commonly used statistical software packages in Table 12. A more extensive list of free statistical software is available at http:// freestatistics.altervista.org/. Please consider the following issues when choosing your software. • The focus of your research question: is the package suitable? Are all necessary methods implemented or can they easily be implemented? Do you rather need a general purpose or a specialized one? Or, maybe, could a tool for basic statistics do the job? • Costs: How much money are you able and willing to spend? There may be packages already available at your institution. Or you can consider a free and open source package. Or, of course, you can buy one. • Available help: is the software you are considering the same software the people in your department or your friends are working with? Someone whom you can ask is most valuable (but be careful, you may risk some friendships here). There are a lot of online resources and books available for the common tools. • Your future plans: maybe you invest a lot of time into software you cannot use anymore if you change to another institute. To choose one of the “big players” is not a guarantee, but SPSS, SAS, Stata, and R are used at most sites. • Free and open source software often has a large community supporting each other. Fast-growing repositories make newly developed methods often immediately available (you may have to pay the price of steeper learning curves but most of the time it is worth the effort). • Some industrial sectors prefer specific software packages; in the pharmaceutical industry, for example, SAS is the most applied software.

Data Protection

Data protection and data privacy are very serious issues in every study dealing with person-related data. How to handle person-related data is regulated by law and may, therefore, be different in each country. In most cases, studies have to be approved by an ethics committee. Be sure to be inline with the regulations prevalent in your country before starting a study. The HIPAA (Health Insurance and Accountability Act, Safe Harbor) – lists 18 items that could be used to identify a person. This list of identifiers should be taken care of. Remove or pseudomize them to ensure the data privacy of the study participants. • Names • All geographic subdivisions smaller than a state usually except for the initial 3 digits of the ZIP code • All elements of dates except years • Telephone numbers

136

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 12. Overview of common software packages used in data analyses

Platforms Comments

Website

Analyse-it Windows, MS Excel 2007+ is required

https://analyse-it.com/

SPSS

Windows, Linux, Mac

https://www.ibm.com/analytics/us/en/technology/spss/

PSPP

Windows, free alternative to SPSS Linux, Max

https://www.gnu.org/software/pspp/

Stata

Windows, comprehensive and very user Linux, Mac friendly data analyis package

https://www.stata.com/

SAS

http://www.sas.com/ Windows, comprehensive data analyis software and language/free Linux, University Edition as Virtual Machine Mac? (Virtual Box) or AWS/standard in pharmaceutical industry

Statistica

Windows

comprehensive and very user friendly data analyis package

http://software.dell.com/products/statistica/

GraphPad Windows, statistical package containing most PRISM Mac common procedures

http://www.graphpad.com/prism

JMP

Windows, powerful data analyis package/part Mac of the SAS family

http://jmp.com/

Minitab

Windows

http://www.minitab.com/

Mplus

Windows, structural equation modeling Linux, Mac

python

Windows, programming language with https://www.python.org/ Linux, Mac comprehensive data analyis libraries

Julia

Windows, High level programming language Linux, Mac for data analyis and numerical computing

http://julialang.org/

R

Windows, almost all you could ever need tool Linux, Mac

https://www.r-project.org/; http://bioconductor.org/

BUGS

Windows, Bayesian Gibbs Sampler Linux, Mac

http://www.openbugs.net

JAGS

Windows, Bayesian Gibbs Sampler Linux, Mac

http://mcmc-jags.sourceforge.net/

Stan

Windows, Bayesian Statistic Package Linux, Mac

http://mc-stan.org/

comprehensive data analyis package

https://www.statmodel.com/

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

137

• • • • • • • • • • • • • •

Fax numbers Email addresses Social security numbers Medical record numbers Health plan beneficiary numbers Account numbers Certificate/license numbers Vehicle identifiers and serial numbers including license plates Device identifiers and serial numbers Web URLs Internet protocol addresses Biometric identifies (i.e., retinal scans, fingerprints) Photos Any unique identifying number, characteristic or code

The Concept of Standardization – What SDS Values Are For

Most of the measurement values in pediatrics are subject to considerable age and gender dependencies. Therefore, it may be useful to transform the measured values (e.g., the BMI of your study participants) to so-called standardized values (SDS values or z-values). A standardized value represents a given measured value expressed as the numbers of SDs above or below the respective expected value. This sounds a bit technical, but let us look at the IQ example again: the IQ is designed in a way that the expected (mean) value is 100 and its SD is 15.1 It is expected to be normally distributed. A person whose IQ is 100 meets exactly the expected value, its deviation from it is zero. So, its standardized IQ-score is 0. A person with an IQ of 85 is exactly one SD below the expected value: the standardized IQ-score is –1. In the table, you find further examples of the calculation of standardized IQ-scores (Table 13). Because this standardized score measures the values in SDs, we also call it SD scores, in short SDS. Given the normal distribution, we expect around two thirds (68%) of the population to have values between –1 and +1 SDS and around 95% between –2 and +2 SDS. Values between –1 and +1 are considered average scores and values above +2 SDS are considered extremely high. SDS can directly be transformed into percentiles, where an SDS value of 0 corresponds to the 50th percentile (the median). In Table 14, you will find typically used SDS values and the corresponding percentiles. There are respective functions in mostly all statistical software packages. For example, Excel: NORM.S.INV(perc) and NORM.S.DIST(SDS, TRUE)2, R: pnorm() and qnorm(), SAS: PROBNORM() and PROBIT(). 1

Therefore, the IQ itself is already a standardized value. Excel functions are language dependent, please look for the correct translation, e.g., here: http://dolf.trieschnigg.nl/excel/index.php. 2

138

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Table 13. Selected IQ values and the respective SD- or Z-scores IQ – observed value

Deviation from the mean

IQ – standardized score

70 77.5 85 100 115 122.5 130

–2 × 15 –1.5 × 15 –1 × 15 0 +1 × 15 +1.5 × 15 +2 × 15

–2 –1.5 –1 0 1 1.5 2

Table 14. Standard deviation scores and the respective percentiles for selected values SDS

Percentile

–2 –1.96 –1.881 –1.28 –1 0 1 1.28 1.881 1.96 2

2.3 2.5 3 10 16 50 84 90 97 97.5 97.8

In pediatrics, SDSs play a prominent role. As mentioned above, most measures are subject to a strong age dependency. To assess the meaning of a measurement value, we need to compare it with a value that is normal in the context of gender and age (the mean expected value). This comparison is done by means of SDS. The information content of a child being of height 110 cm is approximately zero. Without the information on age and gender, we could not access its meaning. It could be an extremely tall child if he or she is 3 years old (+3.5 SDS for a boy, +3.9 SDS for a girl), the value is normal for a child of age 5 (+0.2 SDS for a boy, +0.5 SDS for a girl), and it is an extremely small value for a child aged 7 (–2.2 SDS for both genders) (Figure 5). As we can see, the transformation to SDSs facilitates the comparison of the measured value with the expected value for the respective value (negative SDS: below the expected value, positive SDS: above the expected value). In addition, you can assume that a value below –2 or above +2 is extreme in the context of age and gender and might, therefore, be a signal for illness or abnormal development. But how to know what is normal? Growth references aim to answer this question. There are different sources of growth

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

139

140

Height, cm

120

Above 97th percentile!

Between 10th and 90th percentile: normal range Below 3rd percentile!

100

80

2

3

4

5

6

7

8

9

Age, years

Fig. 5. The 3rd, 10th, 50th, 90th, and 97th percentile curves of height in cm. The points indicate 3 different children with a height of 110 cm: at the age of 3 years the child would be exceptionally tall, at the age of 5 years the child’s height would be near the expected mean, and at the age of 7 years the child would be exceptionally small.

references dependent on the target population. For example, 2013, Zong et al. [27] published references of the height of urban Chinese children or 2000, Fredriks published references of the height of Dutch children of different origins. In Figure 6 we compare the 10, 50, and 90th percentile curves of urban Chinese children with the respective percentiles of Dutch children (of Dutch origin). We find that in younger ages, the curves seem to be very similar, but the more the children grow older, the more the curves differ. At the age of 16, the 90th percentile of Chinese children is less or equal to the 50th percentile of Dutch children in both genders. And a 16-year-old Chinese child at the 50th percentile (or 0 SDS) would be at the lower limit of the normal range (10th percentile or –1.28 SDS) according to the Dutch references. Therefore, if you want to assess the meaning of a child’s height aiming to answer the question “is this a normal value or is it peculiar?,” it is very important to use references suitable for the target population. For assessing associations within the study population, the choice of the references is less important (you can even derive your own standardization). Besides anthropometric measurements, the transformation of the raw measurement values into SDSs is also recommended for laboratory values. There may also be a strong and nonlinear age-dependency with different dynamics in different phases of development (e.g., growth spurts and puberty) with additional differences between genders. Therefore, the transformation to SDSs facilitates the analyses of laboratory measurement values during childhood and adolescence. Reference data are usually generated from large population samples. There are different methods to generate (age-dependent) reference data. Dependent on these methods, different (age-dependent) parameters are published and can be applied. We

140

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Male China NL

Height, cm

160

120

80

0

2.5

5.0

7.5 Age, years

10.0

12.5

15.0

Fig. 6. The 10th, 50th, and 90th percentile curves of height (cm) of Dutch and Chinese boys. The dashed lines show the percentiles curves derived from a Dutch population, the solid lines derived from a Chinese population. The curves are very similar up to an age of 2.5 years. Afterwards, the differences increase up to an age of 11 years. From the age of 11 years on, the pubertal growth spurt seems to be more pronounced in Chinese boys but, they reach their final height earlier/at a younger age.

provide a collection of anthropometric and laboratory reference data as R package. The package also comprises functions for a convenient transformation from raw measurement values to SDSs or percentiles and the creation of percentile curves and the respective tables. It is available for download from CRAN (https://cran.r-project.org/ package = childsds) and will be updated continuously. A list of more than 150 available reference tables from more than 15 countries including further information can be found at https://github.com/mvogel78/childsds/wiki. Further information on the analysis of growth and development is given in [28– 30]. Information on the generation of reference values can be found in [31–35]. References 1 Indrayan A: Medical Biostatistics. 3rd Edition, CRC Press, 2012. 2 Rothman KJ, Greenland S, Lash TL: Modern Epidemiology. Lippincott Williams & Wilkins, 2008.

3 Friedman LM, Furberg C, DeMets DL, Reboussin DM, Granger CB: Fundamentals of Clinical Trials. Springer, 1998.

Basic Epidemiology, Statistics, and Epidemiology Tools and Methods

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

141

4 Thiese MS: Observational and interventional study design types; an overview. Biochem Med 2014; 24: 199–210. 5 Vickers AJ: How to randomize. J Soc Integr Oncol 2006;4:194–198. 6 Bradburn NM, Sudman S, Wansink B: Asking Questions: the Definitive Guide to Questionnaire Design – for Market Research, Political Polls, and Social and Health Questionnaires. John Wiley & Sons, 2004. 7 Sudman S, Bradburn NM, Schwarz N: Thinking about Answers: The Application of Cognitive Processes to Survey Methodology. Jossey-Bass, 1996. 8 Marsden PV, Wright JD: Handbook of survey research. Emerald Group Publishing, 2010. 9 Tourangeau R, Rips LJ, Rasinski K: The Psychology of Survey Response. Cambridge University Press, 2000. 10 Berka K: Scales of Measurement. Lang Log Method. Dordrecht, Springer, 1983, pp 1–73. 11 Abramson JH, Abramson ZH: Scales of Measurement. Res Methods Community Med. John Wiley & Sons, Ltd, 2008. 12 Abramson JH, Abramson ZH: The Variables. Res Methods Community Med. John Wiley & Sons, Ltd, 2008, pp 101–108. 13 Stevens SS, et al: On the theory of scales of measurement. Science 1946;103:677–680. 14 Khurshid A, Sahai H: Scales of measurements: an introduction and a selected bibliography. Qual Quant 1993;27:303–324. 15 Agresti, A: Categorical Data Analysis. John Wiley & Sons, 2013. 16 Carver R: The case against statistical significance testing. Harv Educ Rev 1978;48:378–399. 17 Aho KA: Foundational and applied statistics for biologists using R. Boca Raton, Florida, CRC Press/ Taylor & Francis Group, 2014. 18 Heiberger RM, Holland B: Statistical Analysis and Data Display: An Intermediate Course with Examples in R. Springer, 2015. 19 Cleveland WS: Visualizing data. Hobart Press, 1993. 20 Cleveland WS, Cleveland WS: The Elements of Graphing Data. Wadsworth Advanced Books and Software Monterey, CA, 1985.

21 Wickham H: ggplot2: elegant graphics for data analysis. Springer, 2016. 22 Wilkinson L: The Grammar of Graphics. Springer Science & Business Media, 2006. 23 Tukey JW: Data-based graphics: visual display in the decades to come. Stat Sci 1990;5:327–339. 24 Wainer H: Graphical visions from Willial Playfair to John Tukey. Stat Sci 1990;5:340–346. 25 Thiese MS, Arnold ZC, Waker SD: The misuse and abuse of statistics in biomedical research. Biochem Med 2015;25:5–11. 26 Wasserstein RL, Lazar NA: The ASA’s Statement on p-Values: Context, Process, and Purpose. Am Stat 2016;70:129–133. 27 Zong XN, Li H: Construction of a new growth references for China based on urban Chinese children: comparison with the WHO growth standards. PLoS One 2013;8:e59569. 28 Cheung YB: Statistical Analysis of Human Growth and Development. CRC Press, 2013. 29 Hermanussen M: Auxology: Studying Human Growth and Development: with 89 Tables. Schweizerbart Science Publ., 2013. 30 Mirman D: Growth Curve Analysis and Visualization Using R. CRC Press, 2016. 31 Cole TJ: The use and construction of anthropometric growth reference standards. Nutr Res Rev 1993; 6: 19–50. 32 Stasinopoulos MD, Rigby RA, Heller GZ, Voudouris V, De Bastiani F: Flexible Regression and Smoothing: Using GAMLSS in R. CRC Press, 2017. 33 Indrayan A: Demystifying LMS and BCPE methods of centile estimation for growth and other health parameters. Indian Pediatr 2014;51:37–43. 34 Borghi E, de Onis M, Garza C, Van den Broeck J, Frongillo EA, Grummer-Strawn L, et al: Construction of the World Health Organization child growth standards: selection of methods for attained growth curves. Stat Med 2006;25:247–265. 35 Wright EM, Royston P: A comparison of statistical methods for age-related reference intervals. J R Stat Soc Ser A Stat Soc 1997;160:47–69.

Mandy Vogel LIFE Child Study Center and Hospital for Children and Adolescents Center of Pediatric Research, University Hospitals, University of Leipzig Liebigstraße 20a, DE–04103 Leipzig (Germany) E-Mail [email protected]

142

Vogel · Poulain · Jurkutat · Spielau · Kiess

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 113–142 (DOI: 10.1159/000481328)

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

How to Deal with Confounding Jon Genuneit Institute of Epidemiology and Medical Biometry, Ulm University, Ulm, Germany

Abstract Confounding is a central problem in epidemiology. It can occur in experimental research but is probably more common and considerably more important in observational studies. Confounding influences measures of association; that is, measures that contrast occurrence of the outcome (most often disease) between groups of people or populations with different conditions, for example, being male or female, smokers or nonsmokers, or children born by different modes of delivery. This chapter gives an introduction into the problem of confounding, into methods to identify it, and into methods to control for it. Confounding is a problem of all fields within epidemiology, but there are specific examples in pediatric epidemiology that are discussed in this chapter. These include examples related to Barker’s hypothesis of fetal origins of disease, to perinatal indicators like birth weight and gestational age, and to reproductive factors like spontaneous abortion. © 2018 S. Karger AG, Basel

Confounding is a central problem in epidemiology. It can occur in experimental research but is probably more common and considerably more important in observational studies. Confounding influences measures of association; that is, measures that contrast occurrence of the outcome (most often disease) between groups of people or populations with different conditions, for example, being male or female, smokers or nonsmokers, or children born by different modes of delivery. This applies to ratio measure of association (e.g., the odds ratio, the relative risk, the hazard ratio) as well as to difference measures of association (e.g., the risk difference).

Measures of Effect versus Measures of Association

In epidemiology, we are interested in effects of conditions or exposures on health or disease as an outcome. However, these effects are not readily observable. Imagine an exemplary investigation of the effect of mode of delivery on subsequent occurrence of allergic disease. Here, we would be interested in the occurrence of allergic diseases among those born by cesarean section. To determine the effect of the mode of delivery we would have to contrast this with the occurrence of allergic diseases if the same children had been born vaginally. This counterfactual condition, that is, the situation contrary to the facts, cannot be observed. Clearly, we cannot turn back time and let these children be delivered by a different birth mode. Instead, we have to find other children, born vaginally as a substitute population in whom we measure the occurrence of allergic disease. In other words, we exchange the unobservable counterfactual situation by a prediction of what it might be from an observable situation. This exchange allows us to derive a measure of association, which we hope equals the desired measure of effect. However, if this exchange is not proper, if the occurrence of allergic disease in our substitute population is not equal to the occurrence of allergic disease in the counterfactual situation, the measure of association will be confounded.

Terminology

Indeed, confounding has been termed to be a non-exchangeability bias [1]. Another component of non-exchangeability bias is selection bias, which is discussed in the chapter by Hoffmann et al. [this vol., pp. 71–84]. Other authors use the term “comparability” rather than “exchangeability.” Most of the terminology in this chapter relates to theory of causal inference; a complete description of its terminology is beyond the scope of this chapter but thorough outlines can be found elsewhere [2, 3]. Hereafter, “exposure” or “exposed” will be used to denote the exposure or condition that is examined and “outcome” will be used to denote the response, health, or disease under examination. The idea of the counterfactual situation or potential outcomes is displayed in Table 1: there are 4 different types of persons; however, we can observe each person only in one particular state, exposed or unexposed. Thus, the outcome under the unobservable state becomes the potential outcome. Apart from measures of effect, measures of association, exchangeability, counterfactual, and confounding, another important term is causal intermediates or mediators, that is, factors that cause the outcome and those that are affected by the condition or exposure under investigation. Here, the distinction between direct and indirect causal effects becomes important, the latter being causal effects exerted through a causal intermediate.

144

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

Table 1. Description of potential outcomes, counterfactual model, or causal types Type Outcome/response if

Details

exposed unexposed 1

1

1

2

1

0

3

0

1

4

0

0

“Doomed” – exposure is irrelevant because these persons will develop the outcome regardless of exposure “Causal exposure” – the outcome occurs only if these persons are exposed “Preventive exposure” – the outcome occurs only if these persons are unexposed; the outcome does not occur if these persons are exposed, thus, the exposure is preventive “Immune” – exposure is irrelevant because these persons will not develop the outcome regardless of exposure

Criteria for a Confounder

As described above, confounding occurs if the proportion of unexposed persons with the outcome in an observable population differs from the proportion of persons with the outcome under the counterfactual idea (i.e., type 1 and 3 persons in Table 1). So, a necessary criterion for a confounder is that it has to be associated with the outcome frequency within the unexposed population. Otherwise, the confounder could not lead to a difference between the observed and the counterfactual unexposed population. A further necessary criterion for a confounder is that it has to be associated with the exposure in the source population out of which the persons with the outcome derive. These 2 criteria are, however, not sufficient to define a confounder. In addition, a confounder cannot be a causal intermediate or mediator as it would be the case if the exposure had an effect on the confounder. This is why some authors speak of necessary “baseline” associations between exposure and confounder: at the instance of its occurrence, the exposure has to be associated with a preexisting confounder that is also driving the risk for the outcome [3]. The difference between confounders and causal intermediates becomes clear when visualized with directed acyclic graphs (DAGs).

Visualization of Confounding Using a DAG

One way to graphically represent the problem of confounding is to use a DAG [2, 4]. This visualization technique is not an analytical technique but more of a graphical language to represent the known or hypothesized relations of conditions or exposures, (health) outcome status, and further important factors such as confounders. DAGs consist of nodes representing variables (e.g., exposure X and outcome Y) and arrows that depict direct causal effects. These arrows go in one direction only

Confounding

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

145

Table 2. Basic directed acyclic graph (DAG) terminology Term

Meaning

Node

Nodes represent variables (e.g., X, Y, and Z in Fig. 1b)

Arrow

Arrows depict direct causal effects; another term for arrows is “edges”

Child

A node that receives an arrow from another node (e.g., Y is a child of X in Fig. 1a)

Parent

A node that sends an arrow to another node (e.g., X is a parent of Y in Fig. 1a)

Descendants

Nodes that are children along a series of arrows (e.g., M and Y are descendants of X in Fig. 1c)

Ancestors

Nodes that are parents along a series of arrows (e.g., X and M are ancestors of Y in Fig. 1c)

Covariates

All nodes except the main exposure (typically X) and the main outcome (typically Y)

Confounder

Both an ancestor of the exposure and an ancestor of the outcome (e.g., Z in Fig. 1b)

Mediator

A descendant of the exposure and an ancestor of the outcome (e.g., M in Fig. 1c)

Proxy confounder

A descendant of a confounder and an ancestor of either the exposure or the outcome (but not both, otherwise it would be a confounder)

Competing exposure

An ancestor of the outcome that is unrelated to the exposure

Path

Any consecutive sequence of arrows, disregarding their directionalities; that is, a path along the arrows’ lines regardless in which direction the arrowhead points

Collider

A node on a path that receives arrows from both directions, that is, where two arrowheads meet (e.g., W in Fig. 1d)

Unblocked path

A path that does not contain a collider; another term is “open path”

Blocked path

A path that contains a collider; another term is “closed path”

d-separated

Two variables are directionally separated if there is no unblocked path between them; d-separated variables are unassociated by causal assumptions

d-connected

Two variables are directionally connected if there is an unblocked path between them

Conditional d-separation

1. Conditioning on a non-collider, that is, a confounder or a mediator, blocks the path at that non-collider (e.g., conditioning on confounder Z in Figure 1b blocks the path X-Z-Y, which leads to the direct effect of X on Y as the only remaining path) 2. Conditioning on a collider, or any descendant of the collider, or any combination of the collider itself or its descendants, opens the path at that collider

and they have only one arrowhead, which is where the “directed” in DAG comes from. The “acyclic” describes the fact that the graphs do not include circles or feedback loops, meaning that the arrows from, for instance, the exposure X cannot go through one or many causal intermediates back to X itself. Table 2 describes some basic terminology, but again, a full description of the rules or syntax of this language is beyond the scope of this chapter; introductions can be found in standard textbooks [2].

146

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

X

Y

X

Y

a

b

Z

X

M

Y

c

Fig. 1. Exemplary DAGs depicting a simple direct effect (a), a confounder (b), a mediator (c), and a collider (d).

W

X

d

Z1

Y Z2

Formal examples of DAGs highlighting the position of exposure, outcome, confounders, and mediators are displayed in Figure 1. These include a simple direct effect of exposure X on outcome Y (Fig. 1a), the influence of a confounder (Z, Fig. 1b), and the influence of a causal intermediate (M, Fig. 1c). The circles around X and Y are typically not included in DAGs and have no meaning; they are displayed here to highlight the main exposure and outcome for the readers inexperienced with DAGs. Figure 1d depicts a special pattern that you may come across in several examples, including those discussed at the end of this chapter. Here, W lies on the path from X to Y and receives arrows from both directions. Such a variable at which 2 arrowheads meet is called collider (see also Table 2 for terminology). In Figure 1d, the relation between X and Y is confounded by the path X-Z1-W-Y. If this confounding is addressed during data analysis by conditioning on W, this opens a new confounding path X-Z1W-Z2-Y (see Table 2 “conditional d-separation”). The resulting bias is often called “collider bias” or “collider stratification bias.” In Figure 1d, controlling either Z1 or W and Z2 together would block all confounding paths. To be frank, for most research questions, the preexisting knowledge is insufficient to draw the perfect DAG and the data available or feasibly obtainable are inadequate to identify the correct answer. Thus, every method will remain at risk for bias, which we cannot rule out completely. One can add signs to the arrows to inform about the direction of effects (deleterious or protective), which may help to identify directions of biases. However, DAGs do not include information on the magnitude of likely biases. Further criticism of DAGs includes that they represent one certain type of causal reasoning that is based on counterfactuals and potential outcomes as described above [5]. Indeed, DAGs and counterfactual methods should just be one of many tools employed in causal inference [5].

Confounding

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

147

There are various software tools for creating and analyzing DAGs. One well-documented and comprehensible tool is the browser-based DAGitty [6]. Its website at http://dagitty.net also offers an interactive section designed to teach DAG terminology. Another program called TETRAD can be used to analyze DAGs and to identify appropriate statistical models for a given analytical problem [7]. Actions Against Confounding during Study Design Following the description above, actions against confounding during the study design should establish or secure exchangeability of (i) the unobservable counterfactual situation of the population with a given condition or exposure with (ii) the observed population with the respective condition or exposure. How can this be achieved? An often declared gold standard is randomization of a population to receive either of the exposures under examination. One described benefit of randomization is that exchangeability will not only be established with regard to known confounders but also with regard to those unknown or which cannot be measured. There are several worthwhile discussions on why “randomization” does not necessarily equal “no confounding;” some of these can be found in referenced material [3]. Also, there are many instances in which randomization is not an option; often because of ethical constraints. This may particularly apply to pediatric epidemiology, which is naturally dealing with index subjects having limited options of assenting or consenting to research and which is often concerned with early life events and longterm consequences. Also, many conditions are not readily amenable to randomization, such as gender (can be done during in-vitro fertilization only), ancestral smoking during production of gametes (would be a very long-lasting study), or number of older siblings (likely not ethical). You will certainly find your own examples quickly. If randomization is not an option, matching for confounders can be used. In cohort studies, exposed subjects are matched to unexposed subjects, that is, matching on exposure, regardless of their outcome because the outcome has not happened at the start of the study. This is something completely different than matching in case-control studies, where diseased subjects are matched to non-diseased subjects, that is, matching based on the outcome. Worked examples highlight the fundamental difference between both methods of matching [2]. If applied and analyzed appropriately, matching can prevent or reduce confounding.

Actions Against Confounding during Data Analysis

Similar to actions against confounding during study design, those during data analysis aim at (re-)establishing exchangeability. Foremost, this is stratification or “conditioning” on levels of a confounding variable. If there is only one single confounder, exchangeability will be restored within a given stratum of the confounder. If there are

148

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

multiple confounders, strata have to be formed based on the levels of all of these. Restriction is a special case of stratification in which only specific strata are carried forward during data analysis and others are discarded. Indeed, restriction could also be applied during study design in terms of restricting the population eligible for study entry to those with a particular confounding condition (e.g., only smokers). In both instances, restriction has implications for generalizability. Standardization involves stratification and weighted aggregation of stratum-specific estimates. Finally, adjustment in multivariate regression models is another method to derive a summary estimate across strata.

Examples of Confounding in Pediatric Epidemiology

Confounding in pediatric epidemiology is not different from confounding in other branches in epidemiology. However, what may be present more often in pediatric epidemiology compared to other fields is time-dependency. Early life from fertilization to childhood includes highly dynamic phases of development and rapid changes in potentially important exposures. If the state of exposure changes over time (e.g., from exposed to unexposed or, more complicated, back and forth), the exposure is said to be time-varying. Much like time-varying exposures, time-varying confounders may exist. Time-dependent confounding can include a situation in which the confounder affects the exposure, but a later (time-varying) state of the confounder is also affected by the exposure. Although including some statistical notation, there are good DAG examples in Figure 1 of a published tutorial on this issue and on methods dealing with it [8]. Note how later states of the exposures become colliders on several paths because they now receive arrows from previous states of exposures and from confounders. Another worked example from pediatric epidemiology includes a recent publication on vaccination and child hospitalizations [9]. Apart from time-dependency, pediatric epidemiology often has to deal with proxyreports. As laid out in another chapter of this book [10], a proxy characteristic that leads to biased reports of both exposure and outcome may induce confounding. Regardless of the source of confounding, methods to deal with it stay the same. However, proxy characteristics should be carefully described and evaluated for confounding, potentially beyond typical indicators characterizing proxies such as (maternal in case of maternal reports) socioeconomic status or education. One specific example of confounding involving pediatric epidemiology is centered on the Barker’s hypothesis of fetal origins of disease, which was brought up by observations of the relation of birth weight with adult blood pressure and cardiovascular disease [11]. Here, DAGs are used to document why adjustment for current weight may induce a spurious association between birth weight and blood pressure [12]. Again, note how current weight is a collider on the path from birth weight to blood pressure via a set of confounders [12]. But also remember that a DAG per se does not

Confounding

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

149

help to identify the magnitude of bias, which arises due to its constellation. Following intense controversial scientific discussion, quantification of the bias led to the assertion that its effect is less of a problem than previously assumed [13, 14]. A further example at the core of pediatric epidemiology also involves the investigation of birth weight along with parental smoking status and infant mortality. While infants born to smokers have higher risk for both low birth weight and infant mortality, among low birth weight infants, only the infant mortality is lower among those with smoking parents compared to those without smoking parents. This so-called birth weight paradox was shown to derive from unwarranted adjustment for birth weight as a confounder [15, 16]. You will easily identify birth weight as a collider in the example DAGs in the referenced article. The reasons why this “paradox” arises have been well summarized [17]. Other examples of colliders in pediatric epidemiology include gestational age, which is similar to the birth weight example above and spontaneous abortion, which is related to time-dependent confounding discussed above [18, 19].

Conclusion

This chapter has given an introduction to the problem of confounding, highlighting important parts of the theoretical basis. Confounding is a problem of all fields within epidemiology, but specific examples in pediatric epidemiology have been noted. Techniques to identify confounding as well as to avoid confounding have been explained. But which one should be applied? Is this technique still cutting edge or has it been criticized to neglect important facts? So, how to deal with confounding in your study within the field of pediatric epidemiology? When in doubt, ask your local epidemiologist for advice.

References 1 Suzuki E, Tsuda T, Mitsuhashi T, Mansournia MA, Yamamoto E: Errors in causal inference: an organizational schema for systematic error and random error. Ann Epidemiol 2016;26:788–793.e1. 2 Rothman K, Greenland S, Lash TL: Modern Epidemiology. 3rd edition, Philadelphia, Wolters Kluwer Health/Lippincott Williams & Wilkins, 2008. 3 Greenland S, Robins JM: Identifiability, exchangeability and confounding revisited. Epidemiol Perspect Innov 2009;6:4. 4 Pearl J: Causality: Models, reasoning, and Inference. 2nd edition, Cambridge, UK, New York, Cambridge University Press, 2009.

150

5 Krieger N, Davey Smith G: The tale wagged by the DAG: broadening the scope of causal inference and explanation for epidemiology. Int J Epidemiol 2016; 45:1787–1808. 6 Textor J, Hardt J, Knüppel S: DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology 2011;22:745. 7 Scheines R, Spirtes P, Glymour C, Meek C, Richardson T: The TETRAD project: constraint based aids to causal model specification. Multivariate Behav Res 1998;33:65–117. 8 Daniel RM, Cousens SN, De Stavola BL, Kenward MG, Sterne JA: Methods for dealing with time-dependent confounding. Stat Med 2013;32:1584–1618.

Genuneit

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

9 Jensen AK, Ravn H, Sørup S, Andersen P: A marginal structural model for recurrent events in the presence of time-dependent confounding: non-specific effects of vaccines on child hospitalisations. Stat Med 2016;35:5051–5069. 10 Genuneit J: How to deal with proxy-reports. Pediatr Adolesc Med 2017, DOI: 10.1159/000481326. 11 Barker DJ: The fetal and infant origins of adult disease. BMJ 1990;301:1111. 12 Weinberg CR: Invited commentary: barker meets simpson. Am J Epidemiol 2005; 161: 33–35; discussion 36–37. 13 Chiolero A, Paradis G, Kaufman JS: Assessing the possible direct effect of birth weight on childhood blood pressure: a sensitivity analysis. Am J Epidemiol 2014;179:4–11. 14 Howards PP: Invited commentary: identifying the improbable, the value of incremental insights. Am J Epidemiol 2014;179:12–14.

15 Hernández-Díaz S, Schisterman EF, Hernán MA: The birth weight “paradox” uncovered? Am J Epidemiol 2006;164:1115–1120. 16 Wilcox AJ: Invited commentary: The perils of birth weight – a lesson from directed acyclic graphs. Am J Epidemiol 2006;164:1121–1123. 17 VanderWeele TJ: Commentary: resolutions of the birthweight paradox: competing explanations and analytical insights. Int J Epidemiol 2014; 43: 1368– 1373. 18 Wilcox AJ, Weinberg CR, Basso O: On the pitfalls of adjusting for gestational age at birth. Am J Epidemiol 2011;174:1062–1068. 19 Howards PP, Schisterman EF, Poole C, Kaufman JS, Weinberg CR: “Toward a clearer definition of confounding” revisited with directed acyclic graphs. Am J Epidemiol 2012;176:506–511.

PD Dr. med. Jon Genuneit, MSc Institute of Epidemiology and Medical Biometry, Ulm University Helmholtzstrasse 22 DE–89081 Ulm (Germany) E-Mail [email protected]

Confounding

Kiess W, Bornehag C-G, Gennings C (eds): Pediatric Epidemiology. Pediatr Adolesc Med. Basel, Karger, 2018, vol 21, pp 143–151 (DOI: 10.1159/000481329)

151

Author Index

Bhutta, Z.A. 60 Bornehag, C.-G. VII

Lange, M. 71 Lemke, J.R. 30

Gennings, C. VII Genuneit, J. 97, 143 Gösswald, A. 71

Memon, Z.A. 60 Persson, L.Å. 85 Poulain, T. 113

Hiemisch, A. 41 Hoffmann, R. 71 Houben, R. 71

Rössler, F. 30 Rotzoll, M. 1

Janson, S. 16 Jurkutat, A. 113

Söder, O. 105 Spielau, U. 113

Khan, M.I. 60 Kiess, W. VII, 41, 113 Kurth, B.-M. 71

Vogel, M. 113

152

Willer, M. 1

Subject Index

Abuse, see Child maltreatment ACE-IQ, see Adverse Childhood Experiences International Questionnaire Adolescent period 108 Adverse Childhood Experiences International Questionnaire (ACE-IQ) 22 Age, determination in children 106 BMI, see Body mass index Body mass index (BMI) 109 Body surface area (BSA) 109 BSA, see Body surface area Case-control study 118 Child maltreatment child sexual abuse 25 cultural differences between countries 25, 26 data sources agency registers 19, 20 inpatient registers 19 mortality registers 18, 19 outpatient registers 19 police reports 20 socioeconomic data 20 definition 17, 18 ethics of pediatric epidemiology studies 27 history of study 17 neglect 24, 25 population-based surveys children’s personal experiences 23 parental reports of attitudes and behavior 20–23 sampling methods and size 23, 24 validity problems 26 web-based surveys 27 Child mortality, trends 85–87

Child sexual abuse, see Child maltreatment Climate change 91, 92 Clinical Laboratory Improvement Amendments of 1988 37 Cohort study 117 Computed tomography (CT), incidental findings 41, 46, 50–52 Conflict Tactic Scale (CTS) 21, 22 Confounding criteria 145 data analysis 148, 149 directed acyclic graph for visualization 145–148 examples 149, 150 hypothesis testing 114 measure of effect versus measure of association 144 overview 143 proxy reports 102 study design 148 terminology 144 Cross-sectional study 116 CT, see Computed tomography CTS, see Conflict Tactic Scale DAG, see Directed acyclic graph Data protection 136, 138 Data sources child maltreatment agency registers 19, 20 inpatient registers 19 mortality registers 18, 19 outpatient registers 19 police reports 20 socioeconomic data 20 ethics of data collection 126 examinations 120

153

interviews 124 overview 119, 120 questionnaires 121–124 repeated data collection 125, 126 secondary sources 124, 125 tests 120, 121 Demographic Health Surveys (DHS) 90, 91 Developing countries disease burden 61 equity in child health 89 presentation of data 66, 67 preventable mortality 61, 62 prospects for child health research 67, 68 research priorities 62–66 Development longitudinal growth patterns 109, 110 organ and tissue growth 108 puberty 110 reference values fat mass 111 hemoglobin 111, 112 water content 111 Developmental Origin of Health and Disease 88 DHS, see Demographic Health Surveys Directed acyclic graph (DAG), confounding visualization 145–148 Distributions 128 Ethics, pediatric epidemiology child maltreatment studies 27 data collection 126 developing countries 65, 66 guidelines 11–13 health research versus health care 5, 6 institutionalized children, historical perspective 7–11 overview 1, 2 vulnerability of children as probands 2, 3 Fat mass, reference values 111 Freedom of Information Act 37 Genetic Information Nondiscrimination Act 37 Genetic testing, legislation European Union 31, 32 France 33, 34 Germany 32, 33 Netherlands 34, 35 overview 30, 31

154

Portugal 36 Switzerland 33 United Kingdom 35, 36 United States 37, 38 German Health Interview and Examination Survey for Children and Adolescents (KiGGS) 72–74, 78–80, 82 Good epidemiological practice 11–13 Graphs 132–134 HDSS, see Health and Demographic Survey Systems Health and Demographic Survey Systems (HDSS) 81 Health Insurance Portability and Accountability Act 37 Height, growth patterns 109, 110 Hemoglobin, reference values 111, 112 HIPAA 136 Hypothesis testing 113, 114, 135 ICAST 22 ICAST-C 22, 27 Incidental findings clinical relevance 48–54 cohort studies impact of incidental findings 43–45 management of incidental findings 54 frequency 48–54 overview 41–43 usefulness versus risks 43–48 Infancy period 107 Institutionalized children, historical perspective 7–11 Interventional study 118 JVQ 22 KiGGS, see German Health Interview and Examination Survey for Children and Adolescents Kurtosis 128 LIFE Child study 53, 55, 57 Likert scales 122 Literature review 114, 115 Location parameter 128, 129 Magnetic resonance imaging (MRI) data sources 120 incidental findings 41, 49–52, 56, 57

Subject Index

Metric scale 126, 127 Millennium Development Goals 85 MRI, see Magnetic resonance imaging National DNA Index System 38 Neglect, see Child maltreatment Neonatal mortality, trends 85, 86, 90 Nominal scale 126 Ordinal scale 127 PET, see Positron emission tomography Positron emission tomography (PET), incidental findings 50 Preschool period 107 Preterm infant 106 Prospective study 117 Proxy report confounding 102 information bias 102 missing data 101, 102 overview 97, 98 perspectives 98–100 quality assessment 100, 101 question examples 99 selection of proxy 100, 101 self-reports 103 Puberty 110 p value 135 Reference data 111, 112, 140, 141 Retrospective study 117 Sampling German Health Interview and Examination Survey for Children and Adolescents 72–74, 78–80, 82 gross sample sampling 73–77 immigrant studies 80, 81 overcoverage 77 overview of representative sample selection 72–73 physical examination benefits and burdens 80 questionnaire design 81 representativeness assessment and adjustment 81–83

Subject Index

survey cooperation incentives 80 theories 78 SDS, see Standard deviation score Selection bias, see Sampling Sex differences, child health 89 Sexual abuse, see Child maltreatment SHIP, see Study of Health in Pomerania Skewness 128 Standard deviation score (SDS) 138–141 Statistical analysis data protection 136, 138 descriptive statistics 129 graphs 132–134 group comparisons 129–131 p value 135 relations 131, 132 scales of measurement 126–128 software 135–137 standard deviation score 138–141 Study design confounding visualization 148 design types case-control study 118 cohort study 117 cross-sectional study 116 interventional study 118 overview 115, 116 prospective study 117 retrospective study 117 standards 118, 119 Study of Health in Pomerania (SHIP) 44, 47, 51, 55 Stunting causes 92 trends 86, 87 Toddler period 107 t value 135 Typhoid 62 Ultrasonography, incidental findings 41 Undernutrition, causes 87, 88 Water content, reference values 111 z value 138–141

155

Pediatric and Adolescent Medicine Editor: W. Kiess ISSN 1017–5989

18

Neonatal Pharmacology and Nutrition Update Editors: F.B. Mimouni, Tel Aviv; J.N. van den Anker, Washington, D.C./Basel/Rotterdam VIII + 128 p., 8 fig., 2 in color, 13 tab., hard cover, 2015. ISBN 978–3–318–02735–8

19

Metabolic Syndrome and Obesity in Childhood and Adolescence Editors: W. Kiess, Leipzig; M. Wabitsch, Ulm; C. Maffeis, Verona; A.M. Sharma, Edmonton, Alta. X + 202 p., 28 fig., 18 in color, 9 tab., hard cover, 2015. ISBN 978–3–318–02798–3

20

Progressive Neuroblastoma Innovation and Novel Therapeutic Strategies Editors: H. Christiansen, N.M. Christiansen, Leipzig VIII + 192 p., 16 fig., 5 in color, 6 tab., hard cover, 2015. ISBN 978–3–318–05496–5

21

Pediatric Epidemiology Editors: W. Kiess, Leipzig; C.-G. Bornehag, Karlstad; C. Gennings, New York, NY VIII + 156 p., 21 fig., 7 in color, 23 tab., hard cover, 2018. ISBN 978–3–318–06122–2

Pediatric epidemiology differs substantially from general epidemiology especially when it comes to ethical, developmental and societal aspects. This unique book addresses biological considerations and ethical and legal questions in dealing with pediatric and adolescent population. Classic topics, such as how to recruit representative samples, how to deal with confounding variables, and how to work with genetic information which are the core areas of the book are also in focus. Last but not least, this volume adds to the current understanding of global trends in occurrence, transmission, and control of epidemic pediatric diseases. This book not only serves as a textbook for epidemiologists, pediatricians, geneticists, and child and public health specialists but is also a key reference for those embarking on pediatric cohort studies and epidemiological studies involving the pediatric population.