Bayesian analysis of multicentre trial outcomes

Bayesian methods provide flexibility in the analysis of data from multicentre trials that would be difficult to achieve by other means. This paper illustrates some useful applications of Bayesian methods to the analysis of multicentre trials, with emphasis on insights that would be difficult to obtain using conventional frequentist methods. Two trials provide data for illustration: a large multicentre trial comparing two doses of a drug with placebo with respect to an essentially continuous measurement for which the original analysis revealed a significant treatment by centre effect, and a large multicentre trial with intraclass correlation induced by a categorical outcome of up to four episodes of heartburn reported by individual patients. The data from both trials had been analysed previously using conventional frequentist methods. Both sets of data were reanalysed using Bayesian and empirical Bayesian methods; all of the analyses provided the same conclusions for the key questions regarding treatment differences. The Bayesian methods provided some insights useful for model checking and also provided a way to explore some important quantitative aspects about the magnitude of treatment effects.