Biostatistics in Medicine (review)

BOOK REVIEWS Biostatütics in Medicine. By Edmond A. Murphy. Baltimore: The Johns Hopkins University Press, 1982. Pp. 541. $37.50. The maturation of clinical epidemiology and the change in national research priorities from biomedical research to health care delivery and cost containment have contributed to the renewed interest of inquisitive physicians in research methodology and biostatistics in recent years. There are several new texts available to introduce these concepts to clinicians and investigators, and there is acceptance of the notion that biostatistics is not simply the application of statistical tests to biological variables but, as the author so well states, ". . . biostatistics is the craft of making inferences from finite samples of authentic biological data subject to error of observation representing in part the operation of chance factors." This new work by the author of Skepsis, Dogma and Belief, TL· Logic of Medicine, and Probability in Medicine is a valuable contribution, though not an introductory text. The objectives of the book are to provide an "elementary account of statistics viewed from inside the domain of medicine," "portray statistics from an essentially biological point of view," and "capitalize on a small but agile stock of necessary probability algebra to extend the scope of biostatistics." Dr. Murphy believes that there is a middle course between "stereotyped methods without rationale" and the use of the calculus. The first two objectives are well achieved in this text, and he conveys the flavor of methods drawn from the calculus without explicidy using those often frightening symbols, the differential and the integral, which deter biologists from reading the statistical literature. The book is organized into four parts. The initial four chapters constitute elementary statistical theory, cover theoretical aspects of hypothesis testing and estimation, and include a valuable chapter on an often neglected topic: transformations . Because biological data are rarely normally distributed, transformations are important in increasing the validity with which traditional tests can be applied to biological data. Part 2 consists of seven chapters on the statistics ofthe Gaussian distribution—applications of hypothesis testing, estimation, analysis of variance, simple regression, multiple regression, correlation, and a useful chapter on discrimination and decision. Part 3 is a series of three chapters on the Permission to reprint a book review printed in this section may be obtained only from the author. 652 Book Reviews analysis of categorical data covering the binomial variate, multinomial distribution , and the Poisson distribution. The last part consists ofmethods not involving explicit distributions and includes a chapter on nonparametric methods, survivorship studies, and a chapter entitled "Makeshift," which covers a potpourri of topics that concern "disorderly data" and include combining levels of significance and "pathologies" of the mean and variance. A strength of the book lies in the discussion of criteria biostatisticians use in selecting a statistical technique to answer the scientific questions behind a study which yielded a particular set of data. Thus, he explains the concepts of simplicity , consistency, efficiency, elasticity, unbiasedness, and asymptotic normality as factors to be considered in the choice of an estimator of a parameter. There are valuable insights sprinkled throughout the book. In a discussion on rounding data from calculations, he notes "the precision of an estimate is not indicated by the number of figures carried out but by the standard deviation of the estimate"; and "the only guaranteed property of 'the normal range' is that 5 percent of every normal population will be misdiagnosed as havingdisease." Physicians who actually carry out studies will be comforted by his description of the difficulties they experience: "Once in the study, subjects may not merely die prematurely; they may be lured to a great distance by employment, politics or love; they may conceive a distaste for the investigator or the treatment and refuse to participate further; they may meddle illegally with the dosage which they are supposed to take, try supplementary medication, or pry into a blinding procedure. (I have had all these misfortunes overtake me in clinical studies)." His comments on an investigator's prior belief in the truth of a hypothesis or the actual value of a parameter as it affects the choice of the size of the test is a natural introduction to the Bayesian approach to statistical...