Basket Trials with External Control Arms: When and How to Borrow across Subtrials?
Tuesday, Aug 5: 11:15 AM - 11:35 AM
Topic-Contributed Paper Session
Music City Center
Basket trials that consists of multiple cancer types have increasingly gained attention in recent years. As one of the master protocols, basket trial provides a flexible framework for evaluating new treatment under various patient subgroups simultaneously. A common practice in phase II basket trials is to include patients with various cancer types treated with investigational drug, which poses challenges for further development decisions without comparable control information. To handle the lack of comparator issue, we propose to use the external control arm (ECA) as a benchmark, which may include external clinical trial data or real world data (RWD). Compared with the conventional design of clinical trial, basket trial allows the potential borrowing of information across sub-trials. In this paper, our focus centered on the fundamental question of the design of basket trial: when to borrow across sub-trials and how to borrow? Three different borrowing approaches are introduced and compared, including the treatment effect borrowing (TEB), treatment response borrowing (TRB), and no borrowing. The two borrowing approaches are implemented using the Bayesian commensurate predictive prior (CPP) method. Based on the results of simulation experiments, we demonstrated the key factors that influence the performance of basket trial borrowing approaches, including 1) the "nugget pattern" (i.e., number of ineffective sub-trials), 2) the effective sample size of the basket trial and RWD, and 3) which scale (the treatment effects or the treatment responses) are more similar across sub-trials. Practical recommendations are provided using a decision-making flowchart for the design of basket trial. We finally illustrate our decision-making flowchart by reexamining the DESTINY-PanTumor02 Trial under different hypothetical scenarios.
basket trial
real-world evidence
master protocol
Bayesian commensurate predictive prior
borrowing information
You have unsaved changes.