American Journal of Epidemiology Vol. 155, No. 5 : 485-486
Copyright © 2002 by The Johns Hopkins University School of Hygiene and Public Health
BOOK REVIEWS |
Epidemiologic Analysis: A Case-Oriented Approach
Department of Epidemiology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD 21205
Dr. Steve Selvin's latest book (1
) aims to bridge an existing gap by demonstrating how "efficient and effective" statistical methods can be when applied to epidemiologic problems. Using a novel case-series approach (perhaps unique among epidemiologic methods texts), he details in 16 autonomous chapters a wide variety of analyses that typically arise in epidemiologic research, including comparisons of groups and a variety of regression modeling techniques (using parametric and nonparametric approaches). Further-more, several sophisticated techniques that may not be typical biostatistical fare for epidemiologists (e.g., smoothing, nonparametric regression) are demonstrated in several chapters.
The book has a number of appealing aspects. Each chapter begins with a clearly stated objective, followed by a description of the analysis data set. In all but one chapter (chapter 5, on Mendel's celebrated 1866 data describing Pisum sativum seed properties), the data are directly relevant to epidemiologists. We applaud the author for providing both the data and much of the S-Plus analysis code on the Internet (see http://socrates.berkeley.edu/~biostat/Courses/Spring/Ph29632). The descriptions of the statistical methods are detailed and coherent with respect to both notation and the hypotheses under consideration. We enjoyed the presentation of several competing and nonstandard analyses in a number of chapters. The liberal use of both computationally intense statistical and graphical methods was a high point of this text.
The book's title (Epidemiologic Analysis) would suggest a critical epidemiologic examination of data in the context of study design and inference, but this book primarily concentrates on statistical evaluation of data. For example, there is only modest discussion of confounding and selection bias. Implementation of this textbook in epidemiology courses would probably require important supplementary material. Furthermore, the text exhibits a slightly constrained scope of epidemiologic analysis. For example, there are no chapters dealing with time-period-cohort, ecologic, case-crossover, or case-cohort analytical designs. A chapter describing basic issues in repeated-measures analysis, beyond the effects of regression to the mean (chapter 4), would also improve the scope of this book.
In terms of a case-series text, the data descriptions sometimes fall short, with omission or incomplete description of some variables (e.g., omission of the "strata" variable in the chapter on matched analysis). Importantly, the author fully neglects to deal with missing data, a common problem of applied data analysis. Each of the data sets appears to have been cleaned to eliminate any consideration of this ubiquitous problem. A separate chapter on this topic, or at least some discussion in individual chapters, seems warranted. In addition, when competing analyses are presented, there is little guidance as to which of the competing analyses is more typical of the epidemiologic literature. For example, conditional logistic regression is the fifth of five analyses in the chapter detailing the analysis of matched data. Further-more, the author neglects discussion of how to choose a penultimate model for presentation by avoiding the possibly conflicting decisions based on 1) goodness-of-fit, 2) change-in-estimate, and 3) causal effects as derived from potential outcomes. A specific shortcoming includes the statement that the "relative risk," a nonspecific term that should be abandoned, is more conservative than the odds ratio (p. 37). Starting from first principles, it is the risk ratio that provides an accurate measure and the odds ratio that, when the outcome is frequent, is liberal.
In total, we believe Dr. Selvin has provided a generally useful and well-conceived "case studies" text. The book falters (as any text must) in its lack of scope and would be improved with augmented discussion of common issues in applied epidemiologic analysis such as missing data and model choice. The intended audience appears to be second-year graduate students in epidemiology or biostatistics, as the book emanated from a course given to such an audience. We recommend the book for this purpose, perhaps with supplementary material as noted above. Furthermore, we also recommend the book to persons with a firm grasp of basic statistical inference (at the level of Selvin's Statistical Analysis of Epidemiologic Data (2)) who wish to expand their statistical toolbox.
NOTES
ISBN 0-19-514618-2, Oxford University Press, New York, New York (Telephone: 800-451-7556, Fax: 919-677-1303, World Wide Web: www.oup-usa.org), 2001, 344 pp., $75.00 Hardcover, $37.95 Paperback
REFERENCES
- Selvin S. Epidemiologic analysis: a case-oriented approach. New York, NY: Oxford University Press, 2001.
- Selvin S. Statistical analysis of epidemiologic data. New York, NY: Oxford University Press, 1991.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||