Evaluation and Application of a Probability of Success (POS) Framework in Oncology Trials

Oncology drug development encounters considerable challenges due to positive Phase I trials rarely leading to regulatory approvals, extremely competitive landscape, and high financial burdens associated with research and development. To enhance decision-making in this area and address high costs and risks associated with late-stage oncology trials, we developed a Bayesian decision-making framework by enhancing existing mWethodologies for estimating the probability of success (PoS) of these trials. We proposed a hierarchical Bayesian model that incorporates mixture prior weights from estimated benchmark probability of approval using machine learning methods, observed treatment effect from earlier studies, and the design features of Phase III trials for PoS estimation. We illustrated the adaptability of our framework to scenarios where prior data is available only from a single-arm study or when the earlier study pertain to endpoints that differ from the primary endpoint intended for the pivotal trial. We discovered through case and simulation studies that multiple factors affect the PoS predictions, including treatment effect uncertainty from earlier studies; and assessed the robustness of our model by varying the assumed correlation between the early study endpoint and the pivotal studys primary endpoint is important. Our framework was designed to facilitate a consistent evaluation of proposed oncology clinical trial designs regarding their statistical success, either within a specific program or throughout a pipeline portfolio. We anticipate that this framework will help streamline decision-making process for oncology pivotal trials from trial sponsors side, improve drug development efficiency and ultimately benefit patients.