Total: 1
We consider basket trials in which a biomarker-targeting drug may be efficacious for patients across different disease indications. Patients are enrolled if their cells exhibit some levels of biomarker expression. The threshold level is allowed to vary by indication. The proposed SIMBA method uses a decision framework to identify optimal biomarker subgroups (OBS) defined by an optimal biomarker threshold for each indication. The optimality is achieved through minimizing a posterior expected loss that balances estimation accuracy and investigator preference for broadly effective therapeutics. A Bayesian hierarchical model is proposed to adaptively borrow information across indications and enhance the accuracy in the estimation of the OBS. The operating characteristics of SIMBA are assessed via simulations and compared against a simplified version and an existing alternative method, both of which do not borrow information. SIMBA is expected to improve the identification of patient sub-populations that may benefit from a biomarker-driven therapeutics.