Evaluating the strength or persuasiveness of epidemiologic evidence is inherently challenging, both for those new to the field and for experienced researchers. There are a myriad of potential biases to consider, but little guidance about how to asses the likely impact on study results. This book offers a strategy for assessing epidemiologic research findings, explicitly describing the goals and products of epidemiologic research in order to better evaluate it successes and limitations. The focus throughout is on practical tools for making optimal use of available data to assess whether hypothesized biases are operative and to anticipate concerns at the point of study design in order to ensure that needed information is generated. Specific tools for assessing the presence and impact of selection bias in both cohort and case-control studies, bias from non-response, confounding, exposure measurement error, disease measurement error, and random error are identified and evaluated. The potential value of each approach as well as its limitations are discussed, using examples from the published literature. Such information should help those who generate and interpret epidemiologic research to apply methodological principles more effectively to substantive issues, leading to a more accurate appraisal of the current evidence and greater clarity about research needs. More detailsInterpreting Epidemiologic Evidence: Strategies for Study Design and Analysis By David A. Savitz Published by Oxford University Press US, 2003 ISBN 019510840X, 9780195108408 321 pages |
Wednesday, November 12, 2008
Free Preview Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis
Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis (Hardcover)
Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis (Hardcover)
by David A. Savitz (Author) "This book was written for both producers and consumers of epidemiologic research, though a basic understanding of epidemiologic principles will be necessary at the outset..." (more)
Key Phrases: New York, Oxford University Press, United States (more...)
Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis. David A Savitz.
Book Review
Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis. David A Savitz. Oxford: Oxford University Press, 2003, pp. 336, £34.50 (HB). ISBN: 0-19-510840-X.
Isobel Bray
My first impressions of this book were that while the topic seemed important, the text was perhaps too wordy to be easily accessible. By the end, however, I was convinced that this is a book every epidemiologist should read, or at least use as a reference text when designing and analysing studies. My initial judgement that it would be heavy-going for a new student of epidemiology remains unchanged, but the author states from the outset that this was not intended to be yet another standard text on epidemiology. Rather, the aim is to offer a comprehensive strategy for assessing the results of epidemiological studies.
The book focuses on the potential sources of bias in any study. For example, what are the most likely magnitude and direction of bias due to non-response, confounding, measurement error in the exposure variable, and misclassification of the outcome of interest? This text provides the knowledge and tools necessary to answer these questions, and many more, so that a researcher can make the most of limited resources when planning a study and learn as much as possible from existing data. The author draws heavily on the published literature (particularly in the fields of environmental and occupational epidemiology) to illustrate each point, which makes the text much easier to understand and digest. There are also plenty of methodological references for those wishing to study the issues raised in more detail.
The 12 chapters are quite long, but they are broken down into clearly labelled sections listed in the Contents. This, and the detailed index, should ensure that researchers find the sections of most interest to them quite quickly. Also, each of the central chapters (4–11) ends with a summary of the proposed strategies, referred to as an integrated assessment. To those who have the time, however, I would recommend reading the book cover to cover as the chapters form a logical sequence and cover a range of important issues barely touched on in most taught courses in epidemiology.
The Introduction provides a clear justification for the book and introduces the major themes of bias and confounding. The key message is that just because we cannot quantify these potential problems in our study, we should not fail to attempt to assess their impact. Chapter 2 (The Nature of Epidemiologic Evidence) begins by elucidating the differences between public health and epidemiology. Though this chapter is not a clear introduc-tion to epidemiology for a novice, it makes many important points (e.g. evidence of no association between two variables is equally valid as evidence for a positive association). Chapter 3 (Strategies for Drawing Inferences from Epidemiologic Evidence) sets out the goal of making evaluation of epidemiological evidence more objective. To this end, we should concentrate on the most likely sources of error rather than those that are easiest to measure. There follow substantial chapters on each of the following major themes—selection bias in both cohort and case-control studies, bias due to loss of study participants, confounding, measurement and classification of both exposure and disease, and random error. Chapter 11 (Integration of Evidence Across Studies) introduces the concepts of comparative analysis, meta-regression and narrative review. The final chapter discusses issues such as how epidemiologists should present their results to those who are interested (e.g. policy makers) but may not have the time or inclination to consider the complex assessments of bias described in this book, and how we should interpret conflicting evidence. It also revisits Hill's criteria for causality,1 setting them in a wider context which involves scrutinizing possible sources of error for all studies conducted, not just those which provide evidence of an association.
In conclusion, this book raises many important points that, currently, are largely overlooked in the interpretation of epidemiological data. In this age of data overload, Savitz encoura-ges us to scrutinize the quality of these data in as objective way as possible. I will certainly be referring to this text in the future, and would recommend it as highly relevant to any practising epidemiologist.
Reference
1 Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965;58:295–300.[ISI][Medline]
Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis. David A Savitz. Oxford: Oxford University Press, 2003, pp. 336, £34.50 (HB). ISBN: 0-19-510840-X.
Isobel Bray
My first impressions of this book were that while the topic seemed important, the text was perhaps too wordy to be easily accessible. By the end, however, I was convinced that this is a book every epidemiologist should read, or at least use as a reference text when designing and analysing studies. My initial judgement that it would be heavy-going for a new student of epidemiology remains unchanged, but the author states from the outset that this was not intended to be yet another standard text on epidemiology. Rather, the aim is to offer a comprehensive strategy for assessing the results of epidemiological studies.
The book focuses on the potential sources of bias in any study. For example, what are the most likely magnitude and direction of bias due to non-response, confounding, measurement error in the exposure variable, and misclassification of the outcome of interest? This text provides the knowledge and tools necessary to answer these questions, and many more, so that a researcher can make the most of limited resources when planning a study and learn as much as possible from existing data. The author draws heavily on the published literature (particularly in the fields of environmental and occupational epidemiology) to illustrate each point, which makes the text much easier to understand and digest. There are also plenty of methodological references for those wishing to study the issues raised in more detail.
The 12 chapters are quite long, but they are broken down into clearly labelled sections listed in the Contents. This, and the detailed index, should ensure that researchers find the sections of most interest to them quite quickly. Also, each of the central chapters (4–11) ends with a summary of the proposed strategies, referred to as an integrated assessment. To those who have the time, however, I would recommend reading the book cover to cover as the chapters form a logical sequence and cover a range of important issues barely touched on in most taught courses in epidemiology.
The Introduction provides a clear justification for the book and introduces the major themes of bias and confounding. The key message is that just because we cannot quantify these potential problems in our study, we should not fail to attempt to assess their impact. Chapter 2 (The Nature of Epidemiologic Evidence) begins by elucidating the differences between public health and epidemiology. Though this chapter is not a clear introduc-tion to epidemiology for a novice, it makes many important points (e.g. evidence of no association between two variables is equally valid as evidence for a positive association). Chapter 3 (Strategies for Drawing Inferences from Epidemiologic Evidence) sets out the goal of making evaluation of epidemiological evidence more objective. To this end, we should concentrate on the most likely sources of error rather than those that are easiest to measure. There follow substantial chapters on each of the following major themes—selection bias in both cohort and case-control studies, bias due to loss of study participants, confounding, measurement and classification of both exposure and disease, and random error. Chapter 11 (Integration of Evidence Across Studies) introduces the concepts of comparative analysis, meta-regression and narrative review. The final chapter discusses issues such as how epidemiologists should present their results to those who are interested (e.g. policy makers) but may not have the time or inclination to consider the complex assessments of bias described in this book, and how we should interpret conflicting evidence. It also revisits Hill's criteria for causality,1 setting them in a wider context which involves scrutinizing possible sources of error for all studies conducted, not just those which provide evidence of an association.
In conclusion, this book raises many important points that, currently, are largely overlooked in the interpretation of epidemiological data. In this age of data overload, Savitz encoura-ges us to scrutinize the quality of these data in as objective way as possible. I will certainly be referring to this text in the future, and would recommend it as highly relevant to any practising epidemiologist.
Reference
1 Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965;58:295–300.[ISI][Medline]
Subscribe to:
Posts (Atom)