Bayesian workflow for bias-adjustment model in meta-analysis

Area of Science:

• Statistics
• Biostatistics
• Evidence Synthesis

Background:

• Bayesian hierarchical models are valuable for meta-analysis bias adjustment.
• Their complexity and prior sensitivity require systematic application frameworks.

Purpose of the Study:

• To demonstrate a Bayesian workflow for applying and assessing bias-adjustment models in meta-analysis.
• To compare a standard random-effects model with a bias-adjustment model.

Main Methods:

• Applied a Bayesian workflow to a real-world dataset and a simulation study.
• Compared a standard random-effects model with a bias-adjustment model.
• Evaluated model performance using the widely applicable information criterion and credible intervals.

Main Results:

• Results showed high sensitivity to the prior on bias probability.
• The random-effects model had better predictive accuracy, while the bias-adjustment model produced wider, more conservative credible intervals.
• Simulations confirmed parameter recovery with well-specified priors.

Conclusions:

• The Bayesian workflow provides a principled approach for diagnosing model sensitivities in meta-analysis.
• It ensures transparent and robust application of complex bias-adjustment models in evidence synthesis.