Relative efficiency of using summary versus individual data in random-effects meta-analysis.

Area of Science:

• Biostatistics
• Statistical Methodology
• Evidence Synthesis

Background:

• Meta-analysis combines diverse study data for reliable conclusions.
• Individual participant data (IPD) meta-analysis is often considered the gold standard.
• The relative efficiency of IPD versus summary statistics meta-analysis for random-effects models is not fully understood.

Purpose of the Study:

• To examine the relative efficiency of IPD and summary statistics meta-analysis.
• To investigate efficiency gains under a general likelihood inference framework for random-effects models.

Main Methods:

• Theoretical examination of relative efficiency under a general likelihood inference setting.
• Numerical simulations and analysis of a real-world alcohol interventions study.
• Comparison of methods assuming Gaussian random effects and maximum likelihood estimation for summary statistics.

Main Results:

• Summary-statistics-based meta-analysis is theoretically at most as efficient as IPD analysis under specified conditions.
• The two methods demonstrate asymptotic equivalence.
• Summary-statistics-based inference may lose efficiency with smaller sample sizes.

Conclusions:

• IPD meta-analysis does not inherently guarantee greater efficiency over summary statistics meta-analysis.
• The choice between IPD and summary statistics should consider sample size and statistical assumptions.
• Findings hold under the assumption of constant between-study heterogeneity.