Am J Epidemiol 2002; 156:188-190.
Copyright © 2002 by the
Johns Hopkins Bloomberg School of Public Health
BOOK REVIEWS |
Epidemiology: An Introduction
Department of Epidemiology University of Michigan School of Public Health Ann Arbor, MI 48109
By Kenneth J. RothmanISBN 0–19–513553–9 (cloth), ISBN 0–19–513554–7 (paper) Oxford University Press, Inc., New York, New York (Telephone: 212–726–6000, e-mail: www.oup.com), 2002, 223 pp., $57.50 (cloth), $29.95 (paper)
As stated in the preface, the teaching objective of this textbook is to provide a "simple overview of the concepts that are the underpinnings of epidemiology," in which "the emphasis is not on statistics, formulas, or computation, but on epidemiologic principles and concepts." The conceptual scope of the book is comprehensive and sufficiently profound to provide a solid foundation for instruction in causal inference, study design, and data analysis and interpretation.
The book begins with examples of confounding by age distribution that distort comparisons of crude death rates between countries with markedly disparate socioeconomic resources or risks of death between smokers and nonsmokers. At the conclusion of this chapter, the author provides a series of questions for discussion that review concepts of age-specific rates and risks of death and confounding by age. Indeed, thoughtful questions appear after each of the 10 ensuing chapters that serve to enhance understanding of fundamental concepts, but do not require computation. There are no answers provided in an appendix or plans for an instructor’s guide. However, in the preface, there is an indication of a website that will "post contributed answers to the questions raised at the end of each chapter." The website is at http://www.oup-usa.org/epi/rothman.
The organization of the text adheres to a logical sequence of conceptual and methodological presentations on measuring disease occurrence and causal effects; rationale of cohort and case-control studies; biases in study design, random error, and hypothesis testing; controlling confounding by stratification; measuring interaction or "effect-measure modification;" using regression models; and a concluding chapter on epidemiology in clinical settings. However, prior to the above sequence of chapters, Rothman discusses causation and causal inference and introduces the concept of sufficient and component causes. This highly theoretical chapter provides a philosophical foundation for consideration of multicausality, strength of component causes, and interaction between causes and the sum of attributable fractions.
A didactic style used throughout the text is the insertion of boxed commentaries. The captions include, for example, when risk does not mean risk; whether representativeness is important; the rare disease assumption; properties of a confounding factor; and what an SMR is.
The commentaries, in general, are instructive, provide emphasis or clarification, and do not distract from the flow of the main text.
The various types of epidemiologic studies that are longitudinal, namely where the information obtained pertains to more than one point in time, are discussed at length. Cohort studies are described as closed or open (or dynamic), prospective or retrospective (or historical), and with respect to measuring incidence rates (incidence density) or risks. The discussion of case-control studies begins with an important assertion that when properly carried out, case-control studies "provide information that mirrors what could be learned from a cohort study, usually at considerably less cost and time," and that a fundamental principle is that "controls be sampled independently of exposure status." Variations in case-control design are reviewed, namely nested case-control studies and case-cohort studies, that have been described in other texts as "hybrid designs" because they combine features or advantages of cohort and case-control studies (1).
In the nested design, "density-based sampling" of cases is used commonly, and the cross-product odds ratio provides an estimate of the incidence rate ratio. In the case-cohort design, controls are selected randomly from the baseline cohort. In this method of sampling, the cross-product odds ratio is derived from the number of exposed and unexposed controls in the reference population and estimates the risk ratio rather than the rate ratio. Another variant of the case-control study discussed by Rothman is the case-crossover design. Described in detail by Maclure (2) and Maclure and Mittelman (3), the case-crossover design involves only cases and may be used when brief exposure causes a change in risk that is transient. In such studies, each subject serves as his or her own control, thus constraining potential sources of confounding. When the effect of the exposure is transient, the risk interval encompasses a period of background risk and a period of excess (or reduced) risk.
The discussion of principles of sound study design extends over several chapters with considerations of measurement accuracy, the confidence interval and precision of the point estimate, random error, and systematic error or sources of bias. The core of the discussion is about selection bias, information or misclassification bias, and confounding. The author’s perspective on the limited utility of significance testing and p values is well known. Dr. Rothman emphasizes the application of estimation and the confidence interval function (or p value function) rather than significance testing in epidemiologic analysis. The use of hypothesis testing and the reporting of a p value that is less than or more than an arbitrary value has been criticized as resulting in the loss of important information and misinterpretation of data (4). As stated by Rothman: "If a result is not statistically significant, it means that the null hypothesis cannot be rejected. It does not mean that the null hypothesis is correct (1, p. 116)." The p value function encompasses a range of p values for possible estimated values of relative risk, including the null hypothesis value of 1.0. While not embracing the determination of the confidence interval as a surrogate test of statistical significance, Rothman favors an overall strategy in data analysis of point estimation and an appropriate confidence interval for the measure of effect. The textbook provides detailed formulas and examples in deriving confidence intervals for measures of disease frequency and measures of effect.
The interpretation and control of confounding is achieved through the presentation on stratified analysis. As described in the text, stratification is the cross-tabulation of data on exposure and disease by categories of one or more potentially confounding variables. When the effect measure is determined to be relatively uniform across the strata, Mantel-Haenszel pooling methods are widely used. Formulas and examples are detailed for cohort and case-control studies. The discussion of matching as an effective method for controlling confounding, particularly in cohort studies, is limited to a brief parenthetical statement. The analysis of matched data may be viewed as an application of stratified analysis in which one stratifies by the matching factor. The presentation on standardization, as commonly used to facilitate comparisons of vital statistics in different populations, illustrates how category-specific rates are collapsed into a single summary value. Standardization does not require a uniform effect over strata to obtain an unconfounded summary estimate.
In the chapter on measuring interactions, Rothman asserts that there is "substantial confusion surrounding the evaluation of interaction, much of which stems from the fact that the term is used differently in statistics and epidemiology" (1, p. 168). As used in statistics, interaction refers to departure from the structure of the statistical model, namely whether it assumes additivity or a multiplicative effect, when measuring the effect of an exposure variable in combination with another variable. The presentation on statistical interaction, in the context of stratification analysis, is essentially the demonstration of heterogeneity across strata in relation to measures of effect as risk or rate differences (assumption of additivity) or as ratio measures (assumption of multiplicativity). This is further pursued in the subsequent chapter on regression models, in which departures from additivity of effects are examined in the use of logistic models and other models that involve logarithmic transformation and the assumption of multiplicative effects.
The term "effect modification," or as preferred by Rothman, "effect-measure modification," suggests the concept of biologic interaction (positive or negative synergy) among component causes. Biologic interaction between two causes occurs when the effect of one exposure is dependent on the presence of the other. Biologic interactions have been demonstrated in the evaluation of gene-environment interactions, age- or gender-exposure risk factor interactions, or in joint exposures to environmental risk factors.
In summary, Epidemiology: An Introduction is a superb addition to other publications that have appeared in the past decade (1, 5–10). The author has achieved the stated goal of providing a coherent overview of epidemiologic principles and concepts. In my view, no single textbook provides in sufficient detail the conceptual and technical requirements of an introductory course in epidemiology. The Rothman textbook, for example, does not address principles of surveillance; age-cohort-period effects on rates of vital events; investigation of an epidemic outbreak (although John Snow and cholera are discussed in considerable detail); or principles, pitfalls, and analytic implications of matching. An optimal curriculum must include laboratory exercises in the critical review of published articles, methods of study design, methods of data analysis, and the interpretation of measured effects. Ultimately, at the completion of an introductory course, we would expect the student to be aware of core concepts and of the strengths, limitations, and trade-offs in selecting various epidemiologic methods of research.
REFERENCES
- Szklo M, Nieto FJ. Epidemiology: beyond the basics. Gaithersburg, MD: Aspen Publishers, Inc, 2000.
- Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol 1991;133:144–53.
[Abstract/Free Full Text] - Maclure M, Mittelman MA. Should we use a case-crossover design? Ann Rev Public Health 2000;21:193–221.[ISI][Medline]
- Shakespeare TP, Gebski VJ, Veness MJ, et al. Improving interpretation of clinical studies by use of confidence levels, clinical significance curves, and risk-benefit contours. Lancet 2001;357:1349–53.[ISI][Medline]
- Rothman KJ, Greenland S, eds. Modern epidemiology. 2nd ed. Philadelphia, PA: Lippincott-Raven Publishers, 1998.
- Gordis L. Epidemiology. 2nd ed. Philadelphia, PA: W. B. Saunders Co, 2000.
- MacMahon B, Trichopoulos D. Epidemiology: principles and methods. 2nd ed. Boston MA: Little Brown and Co, 1996.
- Kelsey JL, Whittemore AS, Evans AS, et al. Methods in observational epidemiology, 2nd ed. New York, NY: Oxford University Press, 1996.
- Timmreck TC. An introduction to epidemiology. 3rd ed. Sudbury, MA: Jones and Bartlett Publishers, 2002. 10. Lilienfeld DE, Stolley PD. Foundations of epidemiology. 3rd ed. New York, NY: Oxford University Press, 1994.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  W. Winkelstein Jr. From the Editor: The First Epidemiology Textbook? Am. J. Epidemiol., October 1, 2002; 156(7): 684 - 684. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||