Simulation models for aggregated data meta-analysis: Evaluation of pooling effect sizes and publication biases

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Meta-analysis methods are frequently evaluated using simulation studies.
• Current simulations often directly generate aggregated data, bypassing individual participant data (IPD) simulation.

Approach:

• This study derives and provides the distribution of aggregated data statistics from a heteroscedastic mixed effects model for continuous IPD.
• A procedure for direct simulation of aggregated data statistics is presented.
• The approach is compared with existing simulation methods, highlighting theoretical differences.

Key Points:

• Directly simulating aggregated data can influence meta-analysis method performance assessments.
• The choice of simulation model impacts conclusions regarding meta-analysis method efficacy.
• Three meta-analysis methods (DerSimonian and Laird, Trim & Fill, PET-PEESE) were evaluated.

Conclusions:

• Multiple aggregated data simulation models should be employed to assess the sensitivity of meta-analysis method performance.
• Researchers are encouraged to explicitly state their IPD models and derive aggregated statistics' distributional consequences to inform simulation model selection.