Medical Statistics at a Glance

Medical Statistics at a Glance Petrie A, Sabin C. Blackwell Science Ltd., Oxford, 2000 (reprinted 2001), 111 pages + references and appendices. Although few breath-taking discoveries have been made in statistics since Sir Cox described the survival analysis in 1972 (1,2), the use of statistics is changing and is expected to change even further (3). For this and other reasons there is an increased awareness of the need to establish a core curriculum in medical statistics (4). Simultaneously, the literature on communication and teaching skills is growing fast (5). Considering that, a short and richly illustrated statistics textbook may be good news. With this in mind I set out to read Medical Statistics at a Glance by Petrie and Sabin (henceforth referred to as Medical Statistics) (6). Medical Statistics is an extremely short and concise textbook, which constitutes the book's main strength (and limitation). It is richly illustrated. The book is divided into seven parts (handling data, sampling and estimation, study design, hypothesis testing, basic techniques for analyzing data, additional topics, and appendices). I sometimes ask myself why medical doctors and undergraduate students are often told to start off with heavyweight statistical textbooks (7,8), when both the book by Altman and that of Armitage are all-encompassing and vast (they may even scare the beginner off). Why not begin with something lighter? It was to fill the need for a short book that Petrie and Sabin wrote Medical Statistics (see the Preface). This book is intended as a reference guide and as an adjunct to statistical lectures. The authors admit the restrictions of such an approach, and both Altman's and Armitage's books are mentioned in the Preface for readers who want to gain a greater insight into particular areas of medical statistics. CONTENT Petrie and Sabin try to explain statistics using words and examples. I think this is a more appropriate way to teach the beginner than overloading him/her with statistical formulae. Emphasis is also put on interpretation of computer output. Because there are several computer packages available in the market, the authors have chosen to use more than one when preparing their examples (SAS, SPSS, and STATA; Appendix C). The flowcharts indicating appropriate statistical techniques in different circumstances are most useful. The flowcharts page was initially left out of the book but has now been enclosed in the form of an erratum. Several practical areas are covered: How does one handle missing data (?), how does one check the data set for errors (?), and the importance of displaying data graphically. I liked the explanation of arithmetic means (what we often call the mean) versus geometric means (used in skewed data that have to be transformed) as well as the checklist of advantages and disadvantages of “averages.” The difference between the t distribution and the similar normal distribution is explained. I enjoyed the chapter on when and how to transform data (logarithmic transformation, square root transformation, reciprocal transformation, and square transformation). The chapters on study design include concise overviews (including advantages and disadvantages) of clinical trials, cohort studies, and case-control studies. In the chapter about clinical trials, the CONSORT statement's format is presented with suggested patient flowcharts (9). [Since the publication of Medical Statistics, the CONSORT statement has been revised, and I strongly urge anyone planning a clinical trial to visit the revised CONSORT web site (10).] Sample size calculations, a crucial part of every clinical trial, are presented in chapter 33 with Lehr's useful sample size formula mentioned (11). The authors explain several concepts, among them the true meaning of the P value (the P value is the probability of obtaining “our results” or something more extreme, if the null hypothesis is true; it is not the probability that the null hypothesis is true). The authors also recommend that one cite the exact P value, which nowadays is easily obtained from every computer statistical output (statistical tables are no more in fashion). The problem of multiple hypothesis testing is addressed and a remedy suggested (the Bonferroni approach). Various forms of regression analyses are discussed and richly illuminated. Other more complicated topics include time series, survival analyses (the authors explain how relative hazards ratios should be interpreted), methods for repeated measures, and Bayesian statistics. Personally, I think that Bayesian statistics should be used more often in research and in the clinical setting (12). What I like about Bayesian statistics is that it comes close to the reasoning we clinicians perform every day: we consider a patient's clinical history, heredity, symptoms, and signs before we decide on the most likely diagnosis. The probability that a patient has a certain diagnosis (considering what we know about the patient) may be called the prior or pretest probability. The presence of such a chapter in Medical Statistics shows that the authors do not shy away from difficult subjects. LIMITATIONS The book is short. There is not space enough to explain why, e.g., the variance (crucial for the calculation of the standard deviation and the standard error) is not simply the mean of the squared deviations. Instead, the variance is the squared deviations divided by “n − 1.” Sometimes the concise approach may lead to less understanding instead of more. I let a PhD student read a few chapters of the book, and he found the text difficult since it was so short. Although the authors often cite “classic references” [e.g., Everitt's book on cluster analysis, now available in a later edition (13)], the number of references is small. For the next edition I hope for more references. FOR WHOM IS THIS BOOK? This is not a beginner's book. For such a beginner's book, I still consider Understanding Biostatistics the “Number 1” book (14). Nevertheless, it is a good book. It is inexpensive, concise, and richly illustrated. Medical Statistics is the ideal statistics textbook for JPGN readers who already have a basic understanding of statistics but now and then need a handy reference guide.