A Bayesian Method for the Detection of Proof of Concept in Early Phase Oncology Studies with a Basket Design

In the clinical drug development, proof of clinical concept (PoC) refers to the evidence of treatment efficacy that is obtained from early phase clinical studies. PoC is critical, as it motivates the initiation of late stage clinical trials, and has a profound impact on the “Chemistry, Manufacturing and Controls” (CMC) process, which is preferably launched as early as possible so as to save valuable time for drug development. A new type of oncology clinical trial called basket trial has emerged recently, where the experimental treatment targets on a specific oncogenic pathway that is hypothesized to modulate tumor growth and/or metastasis, and patients with potentially multiple cancer types can be enrolled. The problem of PoC in basket trials has not been formally investigated in the statistical literature. In early phase basket trials, the commonly used independent analysis lacks statistical power of detecting PoC due to limited sample size. A more powerful approach is needed, especially when the treatment effect is not strong enough for each individual cancer type. In this paper, we propose a novel approach for PoC detection in the early phase basket trials under a Bayesian framework. We classify cancer types into a “sensitive subgroup” that responds positively to the treatment, and an “insensitive subgroup” that does not respond to the treatment. We then assess PoC using the posterior probability that at least one cancer type is sensitive. Simulation results show that our proposed approach has a promising performance, with considerable gain in power compared with the independent approach when a relatively large number of the cancer types are sensitive to the treatment.