Bayesian heterogeneity in a meta-analysis with two studies and binary data

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Meta-analysis is crucial for synthesizing evidence, especially in rare disease research.
• Standard statistical methods face challenges with small sample sizes and heterogeneity common in such studies.
• Model uncertainty due to between-sample heterogeneity requires careful consideration in meta-inference.

Purpose of the Study:

• To propose a robust Bayesian meta-analysis method for small sample sizes and heterogeneous data.
• To address limitations of frequentist and Bayesian techniques in specific clinical research contexts.
• To incorporate model uncertainty into the meta-inference process.

Main Methods:

• Utilized Bayesian model averaging with a two-component structure.
• Employed sample clustering to measure heterogeneity.
• Determined posterior probabilities of cluster models for meta-inference.
• Applied the method to sparse binomial data from real-world examples.

Main Results:

• The proposed Bayesian model averaging method is robust to small study sizes and zero-cell counts.
• It effectively incorporates uncertainties into the estimation process.
• The approach provides a mixture of meta-inferences weighted by posterior model probabilities.

Conclusions:

• The novel Bayesian approach offers a reliable alternative for meta-analysis in challenging scenarios like rare disease research.
• It enhances statistical rigor by accounting for model uncertainty and data heterogeneity.
• This method improves the practical applicability of meta-analysis for sparse and heterogeneous datasets.