Exploiting relationships between outcomes in Bayesian multivariate network meta-analysis with an application to...

Area of Science:

• Biostatistics
• Evidence Synthesis
• Pharmacoeconomics

Background:

• Network meta-analysis (NMA) often faces incomplete evidence, with missing treatment-outcome combinations.
• Existing models struggle with missing data or rely on strong, often unrealistic, assumptions.
• Many NMA implementations are limited in scope, handling only a few treatments or outcomes.

Purpose of the Study:

• To develop a flexible Bayesian multivariate NMA model capable of estimating missing treatment-outcome combinations.
• To address limitations of existing methods regarding data sparsity and model complexity.
• To provide a robust tool for synthesizing evidence from multiple, potentially inconsistently reported, outcomes.

Main Methods:

• A novel Bayesian multivariate NMA model was developed using mappings between population mean effects.
• The model allows for imperfect correlations between study-specific effects.
• A unique decomposition of treatment effect variance facilitates efficient implementation for aggregate data.

Main Results:

• The model successfully estimated missing treatment-outcome combinations in a complex dataset.
• Fingolimod and interferon beta-1b showed high efficacy but concerning liver safety profiles.
• Dimethyl fumarate and glatiramer acetate demonstrated balanced efficacy and safety outcomes.

Conclusions:

• The proposed Bayesian multivariate NMA model offers a flexible and robust approach to evidence synthesis with incomplete data.
• This method is particularly valuable for analyzing multiple sparsely reported outcomes.
• The findings provide insights into the comparative efficacy and safety of multiple sclerosis treatments.