Statistical design for a confirmatory trial with a continuous predictive biomarker: A case study

With targeted therapies, it is often hypothesized that their effect may be specific to the subpopulation in which the target pathway is activated. We consider the problem of designing a confirmatory trial when the biological hypothesis of the experimental therapy is strongly supported by the pre-clinical data but limited clinical data is available to pre-define a subpopulation based on a biomarker with continuous values. The study design is further complicated if interim evaluations of the biomarker-based subpopulations are also being considered. We compared several strategies, including a naïve threshold nomination approach, a modification of the "explore and confirm" strategy proposed by Friedlin et al. (2005), and a novel biomarker sequential testing approach, motivated by the "general bivariate normal method" discussed by Wang el al. (2007), and further discussions in Spiessens and Debois (2010) and Holmgren (2017), in a setting where all-comers and biomarker subpopulation evaluations can be performed at interim analyses as well as the end of study. Based on extensive simulations, we concluded that the novel biomarker sequential testing approach out-performed other strategies when there was limited prior information for biomarker threshold determination. This design was implemented in a recently completed clinical trial of simtuzumab (RAINIER study) and provides a useful case study for designing future confirmatory clinical trials of novel targeted therapies.