Encyclopaedic companion to medical statistics Everitt BS, Palmer CR (eds) (2005) ISBN 0340809981; 372 pages; £125.00, $198.50 Hodder Arnold; http://www.hoddereducation.com/

Abstract

The importance of using statistical methods in medical research has been increasingly recognized in recent years. This means that clinicians and medical researchers need to have at least a basic knowledge of statistics. The purpose of this book is to fill the gap between textbooks that are too theoretical and shorter dictionaries that do not focus enough on the needs of medical researchers or have entries that are too brief. Another intention of the book is that it may encourage productive and timely interactions between medical researchers on one hand and statisticians on the other hand. As such, the targeted readership is people involved in contemplating, conducting or contributing to medical research. The entries are written by a total of 79 contributors, most of them from the UK in a wide range of departments, institutions, centres, and companies, each having a special expertise in the field. There are more than 350 entries on statistical topics central to modern medical statistics. They span a wide field of subjects, from ‘confidence intervals’, ‘hypothesis tests’, and ‘P-values’, through ‘adaptive designs’, ‘capture-recapture methods’, and ‘multidimensional scaling’, to ‘consulting a statistician’, ‘journals in medical statistics’, and ‘pitfalls in medical research’. The entries are well written and easy to follow, with most explanations given in plain text, using formulae only when necessary. Many of the entries are illustrated with useful figures and tables, making it easier to understand the discussed topics. Simple concepts, or those that are considered of less importance, are covered only briefly; but most topics are treated extensively, sometimes with entries covering several pages. All except the shortest entries contain useful references to further resources that treat the topics in more depth. Extensive cross-references also make it easy to find other entries treating similar subjects. The level of statistical knowledge necessary varies between entries, of course, with many topics needing almost no knowledge at all, and other more theoretical topics needing at least a basic course in statistics. There are also many entries that redirect the reader to another entry; e.g. for ‘confidence level’ one is redirected to ‘confidence intervals’. However, some redirections are missing. For example, there are entries for ‘logistic regression’ and ‘multiple linear regression’, but no entry for ‘regression’ redirecting the reader to these two entries. Also, an entry may mention 'null and alternative hypotheses, but there is no entries for these topics redirecting the reader to ‘hypothesis tests’. The result is that a reader unfamiliar with these concepts may not find the right entry. I also found a reference in the entry for ‘generalized estimating equations’ to the entry for ‘repeated measures data’, but no such entry exists. As could be expected from any encyclopaedia, there are also some topics that a reader misses. For example, there are no entries treating nonlinear regression, mean absolute deviation (MAD) regression, or quantile regression. I picked a couple of entries and studied these more closely too see how well the entries handle the areas. The first entry was ‘bootstrap’. It is a well written and instructive overview of what bootstrapping is and how to perform it, mainly concentrating on using the bootstrap to estimate means and variances. Other applications, such as confidence interval calculations and hypothesis testing, are only mentioned briefly. The second entry I looked at was ‘logistic regression’. It gives an extensive overview of this area, with motivations for using it, detailed descriptions of modelling and interpretation, as well as discussions of pitfalls and misinterpretations. The entry gives the reader a good overview of this area, and several references are given, but I miss references to a modern book on logistic regression for those readers who are interested in learning more about this subject. Overall, the stated purpose of the book is met. This is a useful reference for anyone involved in medical research, especially for those with a limited knowledge of statistics; but professional statisticians will also find useful information. Practical statisticians working in the pharma industry will find many useful entries on topics related to their work.