Bayesian network meta-regression hierarchical models using heavy-tailed multivariate random effects with...

Area of Science:

• Biostatistics
• Evidence Synthesis
• Meta-analysis

Background:

• Network meta-analysis (NMA) is increasingly used for evidence synthesis.
• Incorporating covariates into NMA enhances analytical power.
• Handling sparse data and heavy-tailed random effects in NMA remains challenging.

Purpose of the Study:

• To propose a novel Bayesian network meta-regression hierarchical model.
• To address data sparsity in NMA by modeling variance components.
• To incorporate aggregate covariates into the variance modeling.

Main Methods:

• Utilized a general multivariate t distribution for random treatment effects, accommodating heavy tails.
• Developed a log-linear regression model for random effect variances to manage sparsity.
• Employed a Markov chain Monte Carlo (MCMC) algorithm with collapsed Gibbs sampling for posterior inference.
• Applied deviance information criterion (DIC) and log pseudo-marginal likelihood (LPML) for model comparison.

Main Results:

• The proposed Bayesian model effectively handles heavy-tailed random effects.
• The variance modeling approach successfully mitigates issues arising from data sparsity in NMA.
• Simulation studies and a case study demonstrate the methodology's utility and robustness.

Conclusions:

• The proposed Bayesian network meta-regression model offers a flexible and robust framework for evidence synthesis.
• The incorporation of aggregate covariates in variance modeling improves NMA in sparse data settings.
• This methodology enhances the reliability of NMA by accounting for complex error structures and data limitations.