Elastic meta-analytic-predictive prior for dynamically borrowing information from historical data with application to biosimilar clinical trials

Abstract

A biosimilar is a biological product that is highly similar to and has no clinically meaningful differences from an approved reference product. Focusing on two-arm randomized clinical trials that aim to establish the equivalence between a test biosimilar product and the reference product, we propose the elastic meta-analytic-predictive (EMAP) prior method to leverage rich historical data available on the reference product to improve the power of the biosimilar trials. We first extract the prior information from multiple historical studies through meta-analysis, and then we discount the resulting meta-analytic-predictive (MAP) prior adaptively according to the congruence between the historical reference data and the trial reference arm data via an elastic function. The EMAP prior method is information-borrowing consistent in that asymptotically it achieves full information borrowing when trial reference arm data are congruent to historical reference data, and no information borrowing when trial reference arm data are not congruent to historical reference data. As a result, the method asymptotically controls the type I error rate at the nominal value. Extensive simulation studies show that the EMAP prior outperforms the robust MAP prior. The EMAP prior generates comparable or higher power and provides better-controlled type I errors. We illustrate the proposed methodology using two trial examples.