A novel approach to bivariate meta-analysis of binary outcomes and its application in the context of surrogate endpoints

Bivariate meta-analysis provides a useful framework for combining information across related studies and has been widely utilised to combine evidence from clinical studies in order to evaluate treatment efficacy. Bivariate meta-analysis has also been used to investigate surrogacy patterns between treatment effects on the surrogate and the final outcome. Surrogate endpoints play an important role in drug development when they can be used to measure treatment effect early compared to the final clinical outcome and to predict clinical benefit or harm. The standard bivariate meta-analytic approach models the observed treatment effects on the surrogate and final outcomes jointly, at both the within-study and between-studies levels, using a bivariate normal distribution. For binomial data a normal approximation can be used on log odds ratio scale, however, this method may lead to biased results when the proportions of events are close to one or zero, affecting the validation of surrogate endpoints. In this paper, two Bayesian meta-analytic approaches are introduced which allow for modelling the within-study variability using binomial data directly. The first uses independent binomial likelihoods to model the within-study variability avoiding to approximate the observed treatment effects, however, ignores the within-study association. The second, models the summarised events in each arm jointly using a bivariate copula with binomial marginals. This allows the model to take into account the within-study association through the copula dependence parameter. We applied the methods to an illustrative example in chronic myeloid leukemia to investigate the surrogate relationship between complete cytogenetic response (CCyR) and event-free-survival (EFS).