Bayesian decision making in confirmatory early-stage breast cancer trial

Preoperative or neoadjuvant systemic chemotherapy, once reserved for patients with locally advanced breast cancer (BC) in whom the goal was to render breast cancer operable, has become increasingly common. In the early-stage BC neoadjuvant studies, clinical benefits such as event-free survival (EFS), disease-free survival (DFS) and overall survival (OS) usually take long time to be observed. Pathological complete response (pCR) rate obtained at surgery as an endpoint after the neoadjuvant treatment has been accepted by FDA as a surrogate predictor for long-term time-to-event endpoints to support accelerated approval. Utilizing this early endpoint helps expedite the development of novel therapies in order to fulfill the unmet medical need for certain high-risk or poor prognosis subsets of early-stage BC patients. By applying the correlation between pCR and time-to-event endpoints, an early and informative Go/NoGo decision-making structure can be built with less cost so that it improves the overall clinical development efficiency. We propose a Bayesian hierarchy model procedure that utilizes Bayesian predictive power of EFS in phase III to guide the Go/NoGo decision based on a clinical plausible threshold for the pCR treatment difference in phase II. The model implements a double bootstrap method to estimate the correlation between pCR and EFS in simulated setting. Besides simulation results, a hypothetical example based on the 2-in-1 adaptive design is provided.