ZIBGLMM: Zero-inflated bivariate generalized linear mixed model for meta-analysis with double-zero-event studies

Area of Science:

• Biostatistics
• Epidemiology
• Medical Research Methodology

Background:

• Meta-analysis (MA) faces challenges with double-zero-event studies (DZS).
• Existing methods like continuity correction or study omission can introduce bias.
• Standard bivariate generalized linear mixed models (BGLMM) do not account for systemic differences in DZS.

Purpose of the Study:

• To introduce a novel zero-inflated bivariate generalized linear mixed model (ZIBGLMM) to address DZS in MA.
• To develop both frequentist and Bayesian ZIBGLMM implementations.
• To evaluate ZIBGLMM's performance in estimating risk ratios.

Main Methods:

• Developed a two-component finite mixture model (ZIBGLMM) with a zero-inflation component.
• Implemented frequentist and Bayesian versions of the ZIBGLMM.
• Compared ZIBGLMM against BGLMM and conventional two-stage MA excluding DZS via simulations and case studies.

Main Results:

• ZIBGLMM demonstrated superior performance in estimating true effect sizes compared to BGLMM and conventional MA.
• The proposed model yielded substantially less bias in risk ratio estimation.
• ZIBGLMM achieved comparable coverage probability to existing methods.

Conclusions:

• ZIBGLMM effectively handles double-zero-event studies in meta-analysis.
• The model offers a less biased approach for estimating treatment effects, particularly risk ratios.
• ZIBGLMM provides a robust statistical framework for meta-analysis involving zero-event data.