Adaptive design theory and implementation using SAS and R
By Mark Chang, Boca Raton, Chapman & Hall/CRC, 2007, xxii+418 pp., £48.99 or US$88.95 (hardback), ISBN 978-1584889625.
This book describes extensions to experimental design to allow for utility functions beyond the power function. Adaptive designs allow the scientist to also take account of timeliness of results, safety of treatment, and ethical considerations.
My impressions of the book are mixed. On the positive, the author has gone to great pains to provide the reader with useful infrastructure, such as a section devoted to a road-map in Chapter 1, each chapter having a section that is dedicated to summary and discussion, and useful signposting about the relative challenges of certain portions of the material. Chang also makes some excellent use of tables, providing an overview of conditional error functions, type I error rates, stopping boundaries etc.
Also, Chang reaches beyond the technical details and provides useful and informative background information. For example, the book describes the motivations and challenges that are peculiar to pharmaceutical trials, the new drug application process, as well as the significant players in the process, and the challenges that they face. Furthermore, Chang devotes discussion space to a candid description of the issues surrounding the interpretation of clinical trials from a frequentist point of view, including e.g. publication bias effects. He also devotes an interesting chapter to debates within adaptive designs, for example the frequentist vs. Bayesian debate. I found this contextual material very helpful.
However, the computer code is presented poorly, relying on mark-up symbols, viz. >>SAS>> and <<SAS<<, to differentiate it from the main text of the book. This is true of the programs and the demonstrations of program calls. Many other books use a distinct, fixed-width font for computer code, which works very well. Despite the balance implied by the title, SAS code predominates, and what R code there is seems to be a direct translation of the SAS.
Furthermore, the problem sections are sparse. I felt that more worked examples and more numerical problems, with solutions, would have been very useful. Some invaluable larger-scale examples are provided; more would have been welcome.
Numerous fixable problems plague the reader. For example, there are no fewer than six different references answering to Chang (2006), and four that are Chang (2007). There is no way to know from the text which is being referred to. I also found many typographical errors. And, I can only imagine that the label ‘time-to-skeleton’ must be a rather grim jest. Here, the book feels rushed.
Overall the book is a solid collection of material, with considerable authorial effort in very specific areas. What is done well is done very well, but opportunities for improvement remain.