Meta-analytic-predictive use of historical variance data for the design and analysis of clinical trials

Abstract

Continuous endpoints are common in clinical trials. The design and analysis of such trials is often based on models assuming normally distributed data, possibly after an appropriate transformation. When planning a new trial, information on the variance of the endpoint is usually available from historical trials. Although the idea to use historical data for a new trial is not new, literature on how to formally summarize and use these data on variances is scarce. The meta-analytic-predictive (MAP) approach consists of a random-effects meta analysis of the historical variance data and a prediction of the variance in the new clinical trial. Two applications that rely on the MAP approach are considered: first, the selection of the sample size in the new trial, guided by the prediction of the variance; and, second, the inclusion of the predicted variance in a Bayesian analysis of the new trial. A clinical trial in patients with wet age-related macular degeneration illustrates the methodology. (C) 2016 Elsevier B.V. All rights reserved.