Customer Reviews

Bayesian Biostatistics and Diagnostic Medicine

5 star:		(1)
4 star:		(0)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

Lyle Broemeling was a Professor at the prestigious Biostatistics Unit in the M.D. Anderson Cancer Center at University of Texas in Houston. He retired this year and I had the good fortune to meet him and talk with him at the JSM in Salt Lake City in August 2007. In my recent job at UBC I had the assignment to help out an imaging company that determines the status of a subjects cancer tumor in a double blinded placebo controlled trial. We had the job of helping them write the imagong data charter abd helped to construct rules for evaluating the reader quality. I wish I had the book when I was working on that problem.

To my knowledge this is the first book written about statistical techniques in the medical diagnosis area that is written exclusively in the Bayesian framework. The author makes a very good point for the Bayesian approach. He sees many applications where data from earlier trials of similar content can be incorporated. The Bayesian approach incorporates it naturally in the prior distribution. A comparable classical approach would have to be some form of meta-analysis and would be awkward and very unnatural.

The book has a very enjoyable reading style and in the first three chapters provides an overall introduction and basic ideas about diagnostic medical and procedures. Chapter four provides a basic introduction to Bayesian method. He also introduces the Markov Chain Monte Carlo (MCMC) method for implementing the more sophisticated Bayesian procedures and he discusses the Winbugs software that is the most commonly used and well debugged freeware available to students and researchers. MCMC has applications in other areas but Bayesian statistics is one of the most commonly used area of application for statisticians and was the motivation for the developers in England.

The rest of the book deals with the applications in diagnostic medicine and includes techniques for evaluating the accuracy of the diagnostic tool under study in a clinical trial let's say. It also includes methods for evaluating the degree of agreement or disagreement among those radiologists assigned to read the images.

The author does not go into the frequentist approach to diagnostic medicine but he does reference two excellent texts that use the classical (frequentist) approach.

I definitely recommend this book to anyone with a curiosity or interest in finding applications of Bayesian methods and especially those that have a practical interest but not a strong background instatistics or Bayesian statistic methods.