Estimation of a Matrix of Heterogeneity Parameters in Multivariate Meta-Analysis of Random-Effects Models

Area of Science:

• Biostatistics
• Epidemiology
• Medical Research Methodology

Background:

• Multivariate meta-analysis offers advantages over univariate methods.
• Estimating heterogeneity parameters in non-negative domains under random-effects models presents a common challenge.

Purpose of the Study:

• To introduce and evaluate two novel multivariate meta-analysis estimation methods.
• To compare these new methods against extended DerSimonian and Laird methods.
• To assess the precision of estimating effect size vectors and covariance matrices.

Main Methods:

• Extending Sidik and Jonkman (2005) univariate estimates to a multivariate setting.
• Implementing an iterative "Hybrid" method based on Sidik and Jonkman (2005).
• Comparing proposed methods with extended DerSimonian and Laird methods using examples and simulations.

Main Results:

• The proposed multivariate estimation methods demonstrate potential benefits in precision.
• Simulations evaluate the accuracy in estimating vectors of effect sizes and associated covariance matrices.
• Discussion of limitations, including negative definite matrices, and potential remedies.

Conclusions:

• The novel multivariate meta-analysis methods show promise for more precise estimation of heterogeneity.
• These methods enhance the utility of multivariate meta-analysis in biostatistical and epidemiological research.
• Addressing challenges with negative definite matrices is crucial for robust heterogeneity estimation.