Borrowing Phase 2 Data in Phase 3: The PUNCH CD3 Trial

The PUNCH CD3 study was a randomized, double-blind, placebo controlled, phase III trial to study the effect of RBX 2660 in patients with recurrent Clostridioides difficile infection. A distinctive aspect of the study design is the primary analysis, which used a unique Bayesian borrowing approach that incorporated efficacy data from a previous phase II study of RBX2600. While the borrowing of external data is relatively common in some settings, it's use in a registrational study is unusual due to statistical concerns such as type I error inflation or bias.
This talk discusses the underlying rationale for borrowing in this trial, providing insight into the scientific and clinical considerations that addressed these statistical concerns. We then discuss the design process used to develop the primary analysis, addressing key aspects of borrowing methodology such as the assumption of exchangeability and the specification of Bayesian priors. The primary analysis used a dynamic borrowing strategy, and we show how this results in an improvement in the operating characteristics relative to non-dynamic borrowing. The results of the study demonstrate the value of dynamic borrowing in action and provide an example where the joint analysis of Phase II and Phase III data led to a positive conclusion and ultimately FDA approval.

borrowing

Bayesian

hierarchical model