It is interesting to read this book on Bayesian biostatistics and diagnostic medicine. Even though statistical methods for diagnostic medicine are not new topics in the literatures, this book has several unique features. Based on his research and consulting experiences at the University of Texas MD Anderson Cancer Centre (MDACC) in Houston, the author introduced the widely used statistical methods for diagnostics from solely Bayesian perspective. The concept was described with clear language supported with motivating real life examples. The author also provided readers with Minitab and WinBUGS codes, so that one can apply these methods to solve practical problems. The exercises at the end of each chapter were carefully and prudently constructed to complement and enhance the understanding of the main concept. The other valuable component of this book is the concise description of clinical trials for imaging, from core contents of protocols, the process of protocols review to aims of various phases of clinical trials.
A brief overview of the book is as follows. Chapter 1 gives a short introduction and preview of the book. Chapters 2 and 3 succinctly explain the techniques commonly used in disease diagnoses, particularly those on the diagnostic imaging modalities for cancer patients. The usefulness, advantages and disadvantages of each procedure are discussed. Chapter 4 is the only theoretical section of this book. It starts with a general overview of the development of Bayesian methods. The basic Bayesian statistics concept, terminology, and MCMC technique are introduced and illustrated with simple applications. Chapters 5–7 describe how to design diagnostic studies for testing accuracy and agreement among measurements and how Bayesian methods can be utilised to analyse the data from such studies and thereafter make inferences. Chapter 8 provides more detailed description of the three phases of clinical trials in cancer research where imaging techniques are used, that described in Chapter 2. Also introduced in this chapter are the Bayesian sequential methods that are increasingly applied in the cancer trials and the software development for the design and analysis of such trials at MDACC. Finally, in Chapter 9, some advanced statistical techniques in the analysis of clinical data from studies in diagnostic medicine where imperfect procedures or no gold standard situations are encountered.
One of the noticeable shortcomings in this book is the lack of introduction and interpretation of informative priors and its impact on the inferences. It is appreciated that in practice, particularly in the clinical trials, prior information is known, rational and adequate usage of such information must be emphasized. The other is that there is little facilitated explanation to the results of the examples, although this weak point can be complemented by numerous exercises given in the relevant chapters. Also, a few sections in Chapter 2 are replicated in Chapter 8. In some places, the entire paragraph is reprinted, for example, the protocol sections in the two chapters are almost identical word by word. Some misspellings are found in the text and in the WinBUGS codes, most of these typos, however, are minor and can be identified without causing readability problems.
This book is, as indicated in the preface, written for biostatistics students. It is in my opinion an excellent introductory textbook on Bayesian methods and their application in diagnostic medicine. Non-experienced statisticians may also find that the systematic overview of the classification and purposes of the three phases in clinical trials and the basic Bayesian theory are useful references and would benefit from the program codes, particularly WinBUGS codes.
Throughout the book, the author ensures that the minimum theoretical material is kept and effortless language is used. However, the students are expected to have obtained good knowledge of statistical methods commonly used in biomedical area. And the readers should have access to several reference books on statistical methods for diagnostics, for example, the ones by Pepe 1 and Zhou et al. 2 to fully comprehend the primary content of the book, not only because many examples are taken from these references, but also very brief introduction to the examples and little explanation to relate the maximum likelihood and Bayesian results have been given. Finally, it is worth notifying that the protocol review process described in the book and that adopted in pharmaceutical companies may not be exactly the same, but the importance in application of statistical methods in clinical trials and the input and influence of statisticians in writing and reviewing the protocols should be affirmed.