research

As a biostatistics PhD student I've had to endure many very poorly written textbooks (though there are many good one's too). Not only is this book a great text on applied survival analysis, it's a great piece of statistical writing and should be used as an example for all applied texts. The general approach of introducing the theory followed by examples with SAS/SPlus code makes learning the material easy and fun. I wish all statistics texts were even half this good!

This text is one of the few to make the work of Andersen et al. (Statistical Models Based on Counting Processes, Springer, 1993) accessible to the average statistician. It has three limitations:
1) fails to mention the use of permutation tests for hypotheses regarding the Nelson-Aalen estimator,
2) fails to cite Good PI, Globally almost most powerful tests for censored data,Nonpar Statist 1992, 1:253-262.
3) fails to deal with multiple dependent events (the most common case).
The text also fails to be prescriptive; one is often left feeling that all tests are equal which simply isn't the case.

Terry Therneau is a research statistician at the Mayo Clinic and Patricia Grambsch is a Professor of Biostatistics at the University of Minnesota. The Cox proportional hazards model has been one of the key methods for analyzing survival data with covariates for the last 25 years. Proportionality is a key assumption that limits its use. There has long been a need to find methods which diagnose when the hazard rates are not proportional and provide alternative methods in such situations. Using the theory of counting processes the authors are able to extend the Cox model to more general situations including multiple/correlated event data using either marginal models or random effects (frailty) models. Time dependent covariates are also covered. Some of the theory of martigales and counting processes is included to make the book self-contained. Generalized residuals are used to identify outlying and influential observations (analogous to ordinary regression) and also to assess the proportional hazards assumption.

Although the topics are advanced and the mathematical level is high the book is designed for practitioners, emphasizing applications and providing numerous examples, many from the authors' experience. Statistical analyses are done in SAS and SPlus. The authors tend to use SAS for data management and analysis and SPlus for diagnostics and other plots. Therneau is an expert programmer who has written much of the necessary software in both systems.

Therneau gave an excellent short course that I attended a couple of years ago at the Joint Statistical Meetings based on a draft of the text. The finished product is as good as I expected.

The appendices include SAS and S-Plus tutorials on survival analysis and provide SAS Macros and S functions to apply the new methodology.