The use of restricted significance tests in clinical trials (Q1189004)

This is not your run of the mill book of recipes for designing and analyzing clinical trials. It is a thought-provoking book that addresses foundational issues and advocates non-standard statistical procedures specially tailored to answer biological and medical questions of interest. Such specially tailored procedures have not been used much in practice until now because of computational difficulties. However, with the advent and ready availability of high speed desktop computers, that should no longer be a deterrence. The author brings his years of experience with clinical trials to illustrate his approach with a diverse variety of examples. The centraltheme of the book is that, as a first step, careful thought must be given to how the treatment is expected to affect the patient response, to postulate a suitable model that reflects this mechanism and specify a restricted alternative hypothesis on the parameters of the model. One can then use Neyman's theory of restricted tests to construct an overall test that is powerful against the specified alternative hupothesis. A significant result with the overall thest is an indication of the efficacy of the treatment which needs to be characterized with further data exploration. Then, as a next step, treatment effects are estimated using confidence interval techniques for different subsets of patients. A broad range of statistical methodologies is surveyed including restricted chi-square tests, normal theory thests, nonparametric tests, permutation and resampling methods and Bayesian estimation. The survey of these topics is necessarily cursory. For anyone who wants to think deeply about the basic issues involved with the design and analysis of clinical trials, this book is a must reading.