A Q statistic with constant weights for assessing heterogeneity in meta-analysis

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• The conventional Q statistic in random-effects meta-analysis often suffers from problems due to inverse-variance (IV) weights.
• Previous research indicated superior performance using sample-size-based weights for certain effect measures.

Purpose of the Study:

• To investigate approximations for the distribution of the Q statistic with both IV and sample-size-based weights.
• To evaluate the accuracy of these approximations for testing heterogeneity and estimating between-study variance (τ²).

Main Methods:

• A simulation study using mean difference as the effect measure was conducted.
• Approximations for Q statistic distributions were assessed for accuracy, test level, power, and bias in τ² estimation.
• Two real-world examples demonstrated τ² and overall mean difference estimation.

Main Results:

• The Q statistic with sample-size-based weights and its exact distribution demonstrated precise test levels, even with small, unbalanced sample sizes.
• The corresponding τ² estimator was nearly unbiased with 10 or more small studies.
• Standard tests using IV weights showed liberal behavior, and their τ² estimators were largely biased.

Conclusions:

• Sample-size-based weights offer a more accurate and less biased approach for random-effects meta-analysis compared to traditional IV weights.
• The proposed methods provide reliable tools for heterogeneity testing and τ² estimation, especially in challenging meta-analytic scenarios.