Estimation of within-study covariances in multivariate meta-analysis

TL;DR

This paper introduces a generic method to estimate within-study covariances in multivariate meta-analysis, improving the handling of unknown correlations across multiple outcomes and treatment groups.

Contribution

It proposes a novel approach to approximate within-study covariances, addressing a common practical challenge in multivariate meta-analysis.

Findings

The method effectively estimates within-study covariances in various scenarios.

Improves accuracy of multivariate meta-analysis when correlations are unknown.

Applicable to multiple outcomes and treatment groups.

Abstract

Multivariate meta-analysis can be adapted to a wide range of situations for multiple outcomes and multiple treatment groups when combining studies together. The within-study correlation between effect sizes is often assumed known in multivariate meta-analysis while it is not always known practically. In this paper, we propose a generic method to approximate the within-study covariance for effect sizes in multivariate meta-analysis and apply this method to the scenarios with multiple outcomes and one outcome with multiple treatment groups respectively.