Confidence intervals for random effects meta-analysis and robustness to publication bias

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• The DerSimonian-Laird (DL) confidence interval is common for average treatment effects in meta-analysis with study heterogeneity.
• However, the DL interval often exhibits suboptimal coverage probability and is susceptible to publication bias.

Purpose of the Study:

• To propose a novel confidence interval for meta-analysis that enhances coverage probability.
• To develop a method less sensitive to publication bias than existing random-effects models.

Main Methods:

• The proposed interval centers on a fixed-effects estimate, which is less prone to bias from smaller studies.
• It incorporates an assessment of additional uncertainty from the random-effects setting to account for heterogeneity.

Main Results:

• Simulations demonstrate that the new confidence interval achieves superior coverage probabilities compared to the DL method.
• The proposed interval shows reduced sensitivity to publication bias.

Conclusions:

• The novel confidence interval offers an improved alternative to the DerSimonian-Laird method for meta-analysis.
• This approach provides more reliable estimates in the presence of heterogeneity and publication bias.