INCLUDING VARIABLE HISTORICAL INFORMATION IN THE ANALYSIS OF CLINICAL TRIALS: AN APPLICATION OF BAYESIAN POWER PRIOR MODELLING

摘要

Bayesian inference offers a natural framework to aggregate variable existing evidence into a quantitative summary of prior knowledge. In this paper, we applied the Bayesian power prior approach to incorporate historical evidence into the final analysis of new randomised controlled trial data. Data from a Dutch equine trial were used as an illustrative example, where treatment effect from oral supplementation of Glucosamine and Chondroitin Sulphate on stiffness in veteran horses was evaluated. Four relevant historical studies were included in the power prior. The relative importance of each historical study was weighted by two experts, one for the clinical relevance and one for the methodological quality. The expert opinion was used to define the weight parameter for the power prior approach. Posterior inference of the treatment effect was successfully obtained by combining the calculated power prior with the likelihood of the equine trial data, resulting in a more precise effect estimate.