Clinical heterogeneity in random-effect meta-analysis: Between-study boundary estimate problem.

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Random-effect meta-analysis is crucial for synthesizing evidence but faces challenges with unexplained heterogeneity.
• Standard methods like maximum likelihood (ML/REML) frequently produce zero estimates for between-study variance, defaulting to fixed-effect models and ignoring clinical heterogeneity.

Purpose of the Study:

• To address the boundary estimate problem in random-effect meta-analysis.
• To propose an adjusted maximum likelihood method for between-study variance estimation.
• To develop a sensitivity analysis framework for detecting boundary estimates.

Main Methods:

• An adjusted maximum likelihood method was developed, maximizing a modified likelihood function incorporating a Gaussian adjustment factor.
• A novel criterion for sensitivity analysis was introduced to identify boundary estimate occurrences.
• The proposed method was evaluated using a meta-analysis of human albumin trials.

Main Results:

• The adjusted method ensures strictly positive between-study variance estimates, especially for a small number of studies (K).
• Bias in overall effect estimates introduced by the adjustment asymptotically approaches zero for large K.
• The adjusted estimator is consistent for large K and shows comparable mean squared error to REML.
• Numerical evaluations demonstrated the absence of boundary estimates and produced results similar to standard methods.

Conclusions:

• The proposed adjusted maximum likelihood method effectively resolves the boundary estimate problem in random-effect meta-analysis.
• This approach improves the handling of clinical heterogeneity, particularly when study numbers are small.
• The method provides reliable and more accurate synthesized results compared to conventional techniques.