Jointly pooling aggregated effect sizes and their standard errors from studies with continuous clinical outcomes.

Area of Science:

• Biostatistics
• Statistical Methods
• Meta-Analysis

Background:

• The DerSimonian-Laird (DL) method is standard for meta-analysis but underestimates standard errors with heterogeneous effect sizes.
• Alternative methods like Hardy-Thompson (HT) and Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustments improve variance estimation but assume known within-study variances.

Purpose of the Study:

• To propose a novel bivariate likelihood approach for meta-analysis.
• To jointly estimate overall effect size, between-study variance, and heteroskedastic within-study variances.
• To evaluate the performance of the proposed method against existing approaches.

Main Methods:

• Developed a bivariate likelihood model treating observed standard errors as estimators of within-study variability.
• Conducted simulations to compare the proposed method with DL (with/without HKSJ), HT, higher-order likelihood methods, and REML.

Main Results:

• The proposed bivariate likelihood approach demonstrated comparable or superior coverage probabilities to existing methods.
• The new method showed reduced bias, particularly when within-study variances were heteroskedastic and correlated with effect sizes.

Conclusions:

• The bivariate likelihood approach offers a robust alternative for meta-analysis, especially in scenarios with complex variance structures.
• This method provides more accurate estimation of effect sizes and variances, addressing limitations of traditional techniques.