Customer Reviews

Statistical Methods in Diagnostic Medicine (Wiley Series in Probability and Statistics)

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

 

 

There are two modern books in the field. This one by Profs Zhou, Obuchowski, McClish and the book by Prof. Pepe. All four are experts in this field. Both books present the same aspects of statistical diagnostic testing and both can be of invaluable help for researchers (both applied and more academic) and graduate students. However, Professor Pepe has done an excellent job (if I may) using a clear, concise notation and language throughout. On the other hand this book (ZOM) is not that well written, giving more weight in the presentation of the personal research of the authors. As a result there is some notation inconsistency (not too puzzling though) and the flow of the text is not that smooth. Both books have full reference lists, they present interesting applications and give a number of exercises at the end of each chapter.

Another reviewer has compared this book with the one by Pepe but is not aware of the book by Broemeling. I reviewed Broemelings book and have this one so in addition to discussing the features of the Zhou-Obuchowski-McClish book I will make some comments comparing it to Lyle Broemeling's book.

Although this book was published in 2002 it is still very contemporary and useful. Both the classical and Bayesian approaches are covered but the details of Bayesian approaches using MCMC methods is not here so if you are interested in that it is well-covered in Broemeling's book. This book is comprehensive and rigorous and show all the modern techniques including the bootstrap. A published article on a bootstrap approach to a diagnostic testing problem involving a mixed linear model is covered in detail and critiqued for depending on an independence assumption.

What I like most about the are the last two chapters 11 and 12. This is material I have not seen before with chapter 11 showing the types of bias that can occur when the gold standard is imperfect (a very common problem given very thorough answers here). Chapter 12 provides statistical methods for habdling multiple studies for evaluating 1) sensitivity and specificity for a diagnostic test and 2) ROC area estimates ofor a diagnostic test using fixed and random effects models.

Also chapter 6 provides methods for estimating sample size when determining area under ROC curves and sensitivity and specificity for single tests, comparisons of two tests, determining equivalence of two tests and more. Methods are illustrated using real examples.

At a time when biomarkers are starting to be used as diagnostics this methodology becomes extremely important.

I think this is a very good text, and when used in combination with Pepe's text, you get all the statistical background you would need to design well thought out studies, and account for bias.