A comparison of one-stage vs two-stage individual patient data meta-analysis methods: A simulation study

Area of Science:

• Biostatistics
• Epidemiology
• Health Research Methodology

Background:

• Individual patient data (IPD) meta-analysis offers advantages over aggregate data meta-analysis for exploring heterogeneity and identifying treatment subgroups.
• Both one-stage and two-stage IPD meta-analysis methods exist, but a direct practical comparison is lacking.

Purpose of the Study:

• To comprehensively compare the performance of one-stage and two-stage individual patient data meta-analysis methods.
• To evaluate how different modeling choices and heterogeneity assumptions impact analysis outcomes.

Main Methods:

• Simulated 1000 datasets across various scenarios varying IPD size and between-study variance (heterogeneity).
• Assessed performance based on mean bias, mean error, coverage, and power for main and interaction effects.
• Compared fully specified one-stage models (random/fixed intercepts, random/fixed effects) with two-stage models.

Main Results:

• Fully specified one-stage models demonstrated superior performance, especially for interaction effects.
• For main effects, performance was similar across models, except when intercept heterogeneity was present, favoring one-stage and two-stage models.
• Two-stage models were consistently outperformed by one-stage models when analyzing interaction effects.

Conclusions:

• Fully specified one-stage models are recommended for IPD meta-analysis, particularly when investigating interactions, due to their robust performance.
• If convergence issues arise with random study intercepts, a fixed study-specific intercept one-stage model is a viable alternative.