Methods for estimating between-study variance and overall effect in meta-analysis of odds ratios

Area of Science:

• Biostatistics
• Statistical Methods
• Meta-Analysis

Background:

• Random-effects meta-analysis relies on between-study variance (τ²) for heterogeneity assessment and effect estimation.
• Common τ² estimation methods for odds ratios are biased, leading to inaccurate overall effects and poor confidence interval coverage.

Purpose of the Study:

• To introduce and evaluate improved point and interval estimators for τ² and overall log-odds-ratios in meta-analysis.
• To compare novel KD and SSW estimators against existing methods via extensive simulations.

Main Methods:

• Utilized an improved approximation to Cochran's Q statistic (Kulinskaya-Dollinger method) for τ² and log-odds-ratio estimation.
• Developed a simpler sample-size-weighted (SSW) estimator for the overall effect.
• Conducted extensive simulations comparing KD, SSW, Mandel-Paule, Hartung-Knapp-Sidik-Jonkman, generalized linear mixed models, and Mantel-Haenszel estimators.

Main Results:

• No single τ² point estimator is universally superior; Mandel-Paule and KD are suitable for small and large studies, respectively.
• The KD estimator demonstrates reliable coverage for τ².
• Inverse-variance-weighted, Mantel-Haenszel, and generalized linear mixed model estimators show substantial bias.
• The SSW estimator provides reliable point and interval estimation for the overall log-odds-ratio.

Conclusions:

• The KD and SSW estimators offer significant improvements for estimating heterogeneity and overall effects in random-effects meta-analysis.
• Researchers should consider the SSW estimator for reliable overall log-odds-ratio estimation and confidence intervals.
• Existing common methods for τ² and odds ratio estimation require careful re-evaluation due to observed biases.