Meta-analysis of rare binary events in treatment groups with unequal variability

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Statistical Modeling

Background:

• Meta-analysis synthesizes study data but faces bias in rare event studies with low or zero event counts.
• Standard methods like fixed-effect models with continuity correction can be biased for rare events.
• Existing random effects models assume equal or greater variability in treatment vs. control groups, limiting applicability.

Purpose of the Study:

• To evaluate the simple average estimator (SA_0.5) under a more general random effects framework allowing unequal group variability.
• To assess estimation efficiency by considering mean squared error, beyond just bias.
• To compare various meta-analysis methods for rare events under a new model accommodating unequal variability.

Main Methods:

• Proved that SA_0.5 remains least biased for large samples under unequal variability.
• Introduced a new random effects model to accommodate unequal group variances.
• Conducted extensive simulations to compare methods based on bias, mean squared error, type I error, and confidence interval coverage.

Main Results:

• SA_0.5 is confirmed as least biased for large samples, even with unequal variability.
• SA_0.5 does not minimize mean squared error, indicating potential trade-offs in estimation efficiency.
• Simulation results provide guidance on method selection for rare event meta-analysis across different sample sizes and variability conditions.

Conclusions:

• The simple average method (SA_0.5) offers low bias for large samples in rare event meta-analysis but may not be the most efficient.
• A new random effects model is proposed for scenarios with unequal variability between groups.
• The study provides data-driven recommendations for choosing appropriate meta-analysis techniques based on statistical performance metrics for rare event data.