ABSTRACT
Cancer clinical trials should be designed primarily to get precise answers to important questions about the efficacy of treatment. However, there is considerable interest in also trying to learn something about the underlying biology of the disease during the course of a trial or a series of trials. Data are collected on patient demographics, tumor characteristics, and various other host factors in an attempt to understand which variables are useful in predicting patient outcome, both for use in subsequent trials and in explaining the results of a given trial. As opposed to the definitive treatment comparison, these statistical analyses are exploratory, serving to generate and not prove hypotheses. The general kinds of questions addressed include the following: What are the important prognostic factors? How can they be used in the design of future trials? Are there identifiable subsets of patients who do so well that there is little room for improvements in treatment? Are there subsets of patients who do so poorly that much more aggressive strategies should be devised?
