BPED: A Bayesian basket design for pediatric trials with external data

The basket trial is a novel type of trial that evaluates one treatment in multiple indications (such as cancer types) simultaneously. One challenge of applying the basket trial design to pediatric studies is limited accrual, resulting in low statistical power. To address this issue, we propose a Bayesian basket design for pediatric trials with external data (BPED) that performs dual-information borrowing to improve the design efficiency: borrow information from the external data to the pediatric trial, and borrow information between the cancer types within the pediatric trial. BPED also accommodates potential heterogeneous treatment effects across cancer types by allowing each cancer type belonging to the sensitive or insensitive latent subgroups. The design adaptively updates the members of the subgroups based on the accumulated pediatric and external data to make go/no-go decisions for each cancer type. The simulation study shows that, compared to some existing designs, BPED yields higher power to detect the treatment effect for sensitive cancer types and maintains a desirable type I error rate for insensitive cancer types.