Optimal Two-Stage Phase II Bayesian Design

Background: Bayesian methods have been adopted in Phase II clinical trial dated back to 1990s. Particularly, hierarchical Bayesian model have been widely adopted in this area due to their advantage in borrowing information across the disease subtypes or multiple treatments to enhance the study power. It is worth noting that no optimal design was rigorously formulated in Bayesian Phase II clinical trials with multiple arms of interventional treatments. In this project, we propose an optimal two-stage Bayesian design with multiple arms of interventional treatments for a homogeneous patient population. Methods: A hierarchical Bayesian model is proposed for platform trials with interventional treatments. We develop a master study protocol to investigate many potential interventional treatments for a homogeneous patient population. The primary study endpoint is assumed binary representing whether the disease responds to a treatment. Unlike other existing Bayesian designs, the optimal two-stage Bayesian design allow the stopping boundary to be flexible, thereby controlling familywise Type I error and maximize familywise study power in some sense. Results: By borrowing the homogeneity of interventional treatments and allowing flexible stopping rules, the simulation studies showed that the optimal two-stage Bayesian design achieved higher study power and lower risk of early terminating the clinical trials than other existing Phase II clinical trial designs. Conclusion: Compared with existing designs, the advantage of the optimal two-stage Bayesian design includes: (i) building a master study protocol to investigate many potential interventional treatments for a homogeneous patient population; (ii) controlling Type I error and maximize study power; (iii) achieving higher study power and lower risk of early terminating the clinical trials.