David A. Savitz. New York: Oxford University Press; 2003. Hardback, Price: $49.95; 321 pp. ISBN: 019510840X
Savitz has provided us with an elegant perspective on the major concerns in interpreting epidemiologic literature. I have always argued that the science of epidemiology is the science and logic of planned common sense. The issues that need to be considered are not particularly profound, but unless they are appropriately considered, common sense can become nonsense. Savitz clearly sees the discipline in a similar light.
Without ever making the reader look at or disentangle a single formula, he takes us through all of the major issues in a most readable and logical format, making this book required reading for the novice and the professional. In addition, the book is a primer for policymakers who need to use epidemiologic results in making public health policy decisions.
Topics covered include bias in selection, bias in loss to follow up, bias in outcome, confounding, measurement error in exposure and disease, misclassification, random error, pooling, metaanalysis, and interpretation related to combining evidence. Each chapter is replete with briefly described examples from the literature, with occasional tables, to illustrate particular points discussed.
This book came to my attention while I was engaged in a review of a large body of work. Although the studies in question all purported to be investigating the same exposures and the same outcomes, the results, at first blush, appeared to be inconsistent. I might have concluded that no conclusion could be made and that more research would be needed. However, after reading Savitz’ book, I came away with a much greater understanding of my task. My job was to sort out issues of bias in sample selection, detection bias, measurement error, and random error to assess how the apparent inconsistencies could or could not be explained. Although more research is often needed, Savitz points out how using planned common sense can bring order to complex datasets and clarity to disparate results and interpretations. Sound principles used in evaluating studies can lead to research that is not more of the same, but research that will reduce the uncertainty and, thus, advance interpretation and understanding.
The chapter titled “Integration of Evidence Across Studies” should not only be read by all epidemiologists, but should be studied by science writers and others who are called on to summarize the literature for the lay public or policymakers. Often these writers deal with only the latest published study. Savitz rightly points out that the findings of observation epidemiologic studies need to be put into a context that includes not only consistency or lack thereof compared with other epidemiologic studies, but also a wide variety of other elements about the study that can modify interpretation. These include the basic biologic underpinnings of the hypothesis being tested, as well as all the issues mentioned here that can influence the outcome of any particular study.
This book will not make a trainee into an epidemiologist without study of more standard analytic texts. However, it will make anyone committed to the field a better epidemiologist. Perhaps of even greater importance, it will make those who need to interpret epidemiologic findings in consideration of public health policy better able to do their jobs with understanding, and to use planned common sense that will be appropriate to the data they are asked to interpret.