Keywords: Bayesian Design, Real World Evidence, Augmented Control
Bayesian augmented control (BAC) designs have been frequently leveraged in oncology clinical trials to formally incorporate existing data regarding patient outcome on control into statistical inferences. These designs may arrive at improved statistical operating characteristics (e.g. power) while potentially reducing the number of patients who must be prospectively randomized to control therapy. This, in turn, may reduce costs/timelines and increase study desirability from an enrollment perspective. The class of BAC designs are thus characterized by a hybrid approach between the single-arm (uncontrolled) designs with a fixed null benchmark (often established from historical control data) and the fully randomized design paradigms in which prospective control data establishes the benchmark. BAC designs have most commonly relied on study-level summary statistics published in the literature. We will present a novel randomized Bayesian trial design in which patient-level real world historical control data is combined (in a statistically rigorous manner) with prospective, randomized control data. Details of the statistical approach and practical implementation aspects will be highlighted.