Exploiting multivariate network meta-analysis: A calibrated Bayesian composite likelihood inference.

Area of Science:

• Biostatistics
• Evidence Synthesis
• Medical Research Methodology

Background:

• Multivariate network meta-analysis synthesizes evidence from multiple treatments and outcomes.
• Unreported within-study correlations pose a significant challenge, potentially leading to biased conclusions.
• Existing methods often require complete correlation data, limiting their applicability.

Purpose of the Study:

• To propose a novel calibrated Bayesian composite likelihood approach for multivariate network meta-analysis.
• To overcome the limitation of unavailable within-study correlations among treatments and outcomes.
• To enable robust posterior inference in the absence of complete correlation information.

Main Methods:

• Developed a calibrated Bayesian composite likelihood method.
• Eliminated the need for specifying a full likelihood function.
• Employed a hybrid Gibbs sampler and Open-Faced Sandwich adjustment for robust inference.

Main Results:

• Simulation studies demonstrated unbiased estimates and nominal coverage probabilities.
• The proposed method effectively handles missing within-study correlations.
• Successful application to real-world datasets for root coverage and anemia treatments.

Conclusions:

• The calibrated Bayesian composite likelihood approach offers a robust solution for multivariate network meta-analysis with missing correlations.
• This method enhances the reliability of evidence synthesis in complex medical research.
• The approach is applicable to diverse clinical scenarios, improving treatment comparisons.