ABSTRACT
Prognostic models can link molecular measurements to clinical outcomes, that is, molecular measurements take the role of prognostic markers, which provide information on the likely course of disease of (untreated) individuals. Similarly, predictive models can incorporate molecular markers that provide information on the treatment effect with respect to the clinical outcome. If the outcome of interest is an event time, then the observation for each individual includes a time and status information. The status may either indicate that the individual experienced the event of interest, or that a competing event
CONTENTS
15.1 Introduction ................................................................................................ 271 15.2 Bladder Cancer Application ..................................................................... 273 15.3 Concepts and Models ................................................................................ 273
15.3.1 Competing Risks ............................................................................ 274 15.3.2 Binomial Regression and Pseudovalues ..................................... 275
15.4 Techniques for Fitting Risk Prediction Rules ........................................ 276 15.4.1 Overview of Approaches .............................................................. 276 15.4.2 Selecting the Tuning Parameters .................................................277 15.4.3 Boosting for the Bladder Cancer Application ............................ 278 15.4.4 Random Survival Forests .............................................................. 281
15.5 Evaluating Risk Prediction Rules ............................................................ 282 15.5.1 Assessing the Distinction between Risk Groups ...................... 282 15.5.2 Prediction Error Curves ................................................................283 15.5.3 Concordance Index ........................................................................283 15.5.4 Performance in the Bladder Cancer Application .......................284
