Evidence-based sample size estimation based upon an updated meta-regression analysis.

A traditional meta-analysis examines the overall effectiveness of an intervention by producing a pooled estimate of treatment efficacy. In contrast to this, a meta-regression model seeks to determine whether a study-level covariate (X) is a plausible source of heterogeneity in a set of treatment effects. Upon performing such an analysis, the results may suggest the presence of a meaningful amount of variation in the treatment effects because of the covariate; however, the current set of trials may not provide sufficient statistical power for such a conclusion. The proposed approach provides quantitative insight into the amount of support that a new trial may provide to the hypothesis that X is a meaningful source of variation in an updated meta-regression model, which includes both the previously completed and the proposed trial. This empirical algorithm allows examination of the potential feasibility of a planned study of various sizes to further support or refute the hypothesis that X is a statistically significant source of variation. A detailed example illustrates the sample size estimation algorithm for both a planned individually or cluster randomized trial to investigate the now commonly accepted impact of geographical latitude on the observed effectiveness of the Bacillus Calmette-Guérin vaccine in the prevention of tuberculosis.