Using meta-analysis of Phase II trials to predict Phase III trial results

Methods We consider a meta-analysis of nine randomised Phase II trials comparing the efficacy of two therapies for acute myocardial infarction. Results for four outcomes were collected: intracranial haemorrhage, stroke, reinfarction and total mortality. We apply univariate and multivariate random-effects meta-analysis methods, and use the obtained summary results to derive 95% prediction intervals, which give the predicted treatment effects for the four outcomes in a future trial. The multivariate approach jointly synthesizes all outcomes whilst accounting for their correlation. The methods are applicable in both frequentist and Bayesian frameworks. Predictions calculated are compared to results from subsequent Phase III trials.