Multivariate meta-analysis of mixed outcomes: a Bayesian approach

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Synthesizing multiple correlated outcomes from studies is challenging.
• Bayesian frameworks offer powerful tools for evidence integration.

Purpose of the Study:

• To propose a Bayesian model for multivariate random effects meta-analysis (MRMA) of mixed continuous and binary outcomes.
• To extend existing bivariate models to the trivariate case.
• To incorporate external evidence through informative prior distributions.

Main Methods:

• Developed a Bayesian trivariate model for MRMA.
• Constructed informative prior distributions for within-study and between-study correlations using external data.
• Utilized a double bootstrap method for mixed outcome correlations.
• Parameterized the between-study model using univariate conditional normal distributions.

Main Results:

• The proposed model successfully synthesizes mixed continuous and binary outcomes.
• Informative priors enhance the integration of external evidence.
• The model allows for explicit prior distributions on between-study correlations, considering their inter-relationships.

Conclusions:

• The Bayesian MRMA model provides a flexible approach for synthesizing mixed outcomes.
• This method can incorporate diverse evidence sources, potentially reducing uncertainty.
• The model is applicable to complex research questions, such as in rheumatoid arthritis.