Zero-Inflated Binomial Model for Meta-Analysis and Safety-Signal Detection

Area of Science:

• Biostatistics
• Clinical Trial Analysis
• Pharmacovigilance

Background:

• Meta-analysis provides overall safety signals but struggles with study-specific variability.
• Likelihood ratio tests (LRT) identify safety concerns but are unsuitable for randomized controlled trials (RCTs) with zero events in arms.
• A novel Bayesian approach is proposed to address limitations of frequentist methods in specific meta-analysis scenarios.

Purpose of the Study:

• To present a Bayesian framework for meta-analysis of RCTs.
• To identify studies with potential safety signals, particularly when zero events are present.
• To provide an overall meta-analytic estimate of safety signal magnitude.

Main Methods:

• Utilizes a Zero-inflated binomial model with spike-and-slab parameterization for treatment effects.
• Establishes a Bayesian framework to calculate the posterior probability of a safety signal for individual studies.
• Applies the method to two published datasets of RCTs, evaluating prior choices for treatment effects.

Main Results:

• The Bayesian approach successfully identifies potential safety signals for individual adverse events.
• It provides a robust overall meta-analytic estimate of the safety signal's magnitude.
• Demonstrates utility in scenarios with zero events in treatment or control arms.

Conclusions:

• The proposed Bayesian framework offers a viable alternative to frequentist LRT methods for meta-analysis of RCTs with zero events.
• It effectively identifies single adverse event signals and estimates overall safety.
• Future extensions can accommodate multiple adverse events for comprehensive safety assessment.