Time-to-event data are commonly encountered within the areas of biostatistics and epidemiology, with the most prominent example of such data, of course, being survival data. The defining characteristic of survival data is that right censoring occurs, i.e. for some of the individuals we only know that the timing of the specific event at study exceeds a given observed point in time (often end of study). Thus, usual statistical techniques such as multiple regressions will fail to give correct answers about relationship between explanatory variables and the timing of the event, and specialized methods are required. There exist many books on this topic, some of them being rather technical, especially if they also cover why inferential methods in this field work properly (Kalbfleisch and Prentice, 2002;Andersen et al., 1993).
The present book's goal is to present the most commonly used survival analysis techniques to readers with a minimal statistical and mathematical background. As an example of this, the illustration of maximum likelihood estimation takes a whole chapter, and also integrals and the delta method are introduced in a non-technical way. Among the topics covered are rates and life tables (Chapters 1 and 2). The Kaplan-Meier estimator is introduced in Chapter 3. The exponential and Weibull survival time distributions are covered in Chapters 5 and 6.
The key part of the book is about proportional hazards models (Chapters 7-9) either with a parametric specification of the baseline hazard function (exponential and Weibull models) or with an unspecified baseline hazard function -the Cox model. In these chapters there are many examples with good discussions on important general statistical and epidemiological concepts such as confounding and interactions.
The Cox model has become the standard choice of analysis within this field. It does, however, rely on various assumptions with the proportional hazards assumption being the most prominent one. It is important to check this assumption before obtained results can be trusted. The author describes some goodness-of-fit tools -those on checking the proportional hazards assumptions are a bit oldfashioned, however. I prefer the different versions of cumulated martingale residuals of Lin, Wei, and Ying (1993) as they are effective in revealing time-varying covariate effects, and they can furthermore be motivated with formally justified tests. These tools are also available in R via the package timereg, for examples of use see Martinussen and Scheike (2006).
To summarize, for students who know very little about statistics and mathematics this book is a welcome resource on the techniques used to analyze time-to-event data. These techniques are introduced in a non-technical way accompanied with many good examples. Also the included R-code gives a good introduction to the practical analysis.