When a predictive classifier has been developed but there is not compelling biological or phase II data that test negative patients do not benefit from the new treatment, it is generally best to include both classifier positive and classifier negative patients in the phase III clinical trials comparing the new treatment to the control regimen. In this case, it is essential that (i) an analysis plan be predefined in the protocol for how the predictive classifier will be used in the analysis, and (ii) the clinical trial be sized for adequate power for analysis of test-positive patients separately. It is not sufficient to just stratify, that is, balance, the randomization with regard to the classifier without specifying a complete analysis plan and sizing the trial appropriately. In fact, the main importance of stratifying the randomization is that it assures that only patients with adequate test results will enter the trial.

For phase III trials of a new treatment and prespecified binary classifier, the purpose of the trial is to evaluate the new treatment and to determine how treatment effectiveness depends on the prespecified classifier. The purpose is not to modify or optimize the classifier. If the classifier is a composite gene expression–based classifier, the purpose of the study is not to reexamine the contributions of each gene. If one does any of this, then an additional phase III trial may be needed to evaluate treatment benefit in subsets determined by the new classifier. In the following sections, I describe several analysis strategies for clinical trials that include patients positive and negative for a prespecified binary classifier. Such strategies are discussed in greater detail by Simon (2008a, 623; 2008b, 640).