Bivariate random effects models for meta-analysis of comparative studies with binary outcomes: methods for the...

Area of Science:

• Biomedical Research
• Statistical Modeling
• Epidemiology

Background:

• Multivariate meta-analysis is crucial for combining clinical study data on drug efficacy and safety.
• Conventional methods struggle with rare events or small studies, often excluding data or using arbitrary continuity corrections.
• Existing approaches can lead to inaccurate conclusions and inconsistencies due to differing correction methods.

Purpose of the Study:

• To present bivariate random effects models for more accurate meta-analysis, especially with rare events or small sample sizes.
• To address limitations of conventional meta-analysis techniques that exclude studies or apply ad hoc continuity corrections.
• To provide a statistically sound framework that utilizes all available data without arbitrary adjustments.

Main Methods:

• Developed a bivariate Beta-binomial model based on the Sarmanov family of bivariate distributions.
• Introduced a bivariate generalized linear mixed-effects model for binary clustered data.
• These models inherently account for within-study correlation between treatment and control groups.

Main Results:

• The proposed bivariate models effectively utilize all study data, avoiding exclusions and arbitrary continuity corrections.
• These methods naturally incorporate the correlation between treatment and control groups within studies.
• Reanalysis of two recent meta-analysis datasets demonstrated the practical application and validity of the models.

Conclusions:

• Bivariate random effects models offer a robust alternative for meta-analysis, particularly in scenarios with rare events or small studies.
• These advanced statistical approaches enhance the accuracy and reliability of conclusions drawn from combined clinical trial data.
• The models provide a more integrated and statistically sound method for evaluating drug efficacy and safety profiles.