Keywords: Basket trial, Bayesian hierarchical model, Heterogeity
In current oncology community, researchers have put more efforts to create therapies targeting specific molecular aberrations, which often occur across different cancer types. As an adaption to this evolvement, basket trials with target therapies investigated in multiple cancer types have emerged as a novel design to improve the efficiency and streamline oncology drug development. The design aims to identify any cancer types with promising effect, usually under the limitation of a small number of patients in each cancer type. It is also reasonable to design a basket trial with interim analysis to drop the cancer types which obviously don't work. It is not efficient to implement multiple independent two-stage designs, which also leads to a much higher false-positive rate. A Bayesian Hierarchical Model, which build on exchangeable cancer type parameter, may be a good solution to borrow information across different cancer types. However, the concern may be raised when there are extreme cancer types in the basket which are very different from the rest. One more thing to consider is about the operational feasibility to fix a timepoint for interim decision making when enrollment varies among indications. Key questions for discussion: How will the heterogeneity affect the interim decision making, and what are the solutions to address it? Are there any other caveats we need to pay attention when applying Bayesian models in basket trial design?