Adaptive expansion of biomarker populations in phase 3 clinical trials

It is well documented in this genomic era that an investigational new drug may have greater treatment effect in a biomarker positive population than in the biomarker negative population. However, limited by preclinical data and early phase clinical data, a lot of Phase 3 confirmatory trials are initiated without fully understanding the biomarker effect. In this article, we will investigate the impact of adaptive population expansion on the overall Type I error in two statistical designs. The endpoint for making the adaptive decision can be different from the primary endpoint of the study. The first design allows expansion of study population from biomarker positive patients to all-comers if the treatment effect in the biomarker positive population is more impressive than expected, suggesting broader activity of the study drug. We show that, under this design, the trial outcome can be tested at the desired alpha level without inflating the Type I error when the adaptive decision is based on the primary endpoint of the study or based on an endpoint non-negatively correlated with the primary endpoint, an assumption that generally holds in practice. The second design allows addition of biomarker positive patients in an all-comer study if the treatment effect in the biomarker negative population is less impressive than expected, suggesting lower probability of success in the all-comer population. We show that, under this design, the trial outcome can always be tested at the desired alpha level without inflating the Type I error.