Predictive Biomarker Classifiers in the Design of Pivotal Clinical Trials

In this chapter we distinguish the use of predictive biomarkers from surrogate endpoint biomarkers. We also distinguish the use of predictive biomarkers for selecting patients for pivotal clinical trials of a new drug from the use of predictive biomarkers for optimizing the utilization of an existing drug. We summarize the key steps in the development of predictive biomarker classifiers for use in new drug development. We discuss the design of targeted clinical trials in which a predictive biomarker classifier is used to restrict entry, and present results comparing the efficiency of targeted trials relative to standard randomized pivotal trials. We also discuss alternative designs in which the predictive biomarker classifier is not used to restrict entry of patients but is used to prospectively define an analysis plan for evaluating the new drug in classifier negative and positive patients. The development of predictive biomarker classifiers can be subjective, but pivotal trials should test hypotheses about the effectiveness of a new drug in subsets defined in a completely prespecified manner by a predictive classifier, and should not contain any subjective components. The data used to develop the predictive classifier should be distinct from the data used to evaluate a new drug in subsets determined by the classifier. The purpose of the pivotal trial is to evaluate the new drug in patient groups defined prospectively by the predictive classifier, not to refine or reevaluate the classifier or its components. New drug development should move from a correlative science mode to a predictive medicine mode.