How to develop treatments for biologically heterogeneous "diseases"

Abstract

The standard paradigm for the design of phase III clinical trials is not suitable for evaluation of molecularly targeted treatments in biologically heterogeneous groups of patients. Here, we comment on alternative clinical trial designs and propose a prospective discovery/evaluation framework for using tumor genomics in the design of phase III trials.

Figure 1

Schematic for the Adaptive Signature Design{Freidlin, 2005 #307}. The tumors of all randomized patients are molecularly characterized with regard to a trial specific set of discovery features to be evaluated. Eligibility is not restricted by the molecular characterization. At final analysis patients are randomly partitioned into a training set and validation set. A binary classifier is trained to identify patients for whom the difference in outcomes between the treatment arms is maximum. The binary classifier is tested on the validation set. The design uses a predictive classifier discovery and evaluation framework, not a multiple hypothesis testing framework. Trial is sized for discovery and validation as illustrated in Sher et al.{Sher, 2011 #844}