The Impact of Covariance Priors on Arm-Based Bayesian Network Meta-Analyzes with Binary Outcomes

• Activity Number: 259 - SPEED: Missing Data and Causal Inference Methods, Part 2
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 3:05 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Zhenxun Wang* and Lifeng Lin and JIM HODGES and Haitao Chu
• Companies: University of Minnesota, and Florida State University and UNIVERSITY OF MINNESOTA and University of Minnesota
• Keywords: Bayesian inference; network meta-analysis; covariance matrix; prior; breaking randomization

Abstract

Performing Bayesian analyses with the arm-based network meta-analysis (NMA) model requires researchers to specify the prior distribution for the covariance matrix of study-specific underlying treatment effects. The commonly-used conjugate prior for the covariance matrix, the inverse-Wishart (IW) distribution, has several limitations. For example, although the IW distribution is often described as a weakly-informative prior, it may in fact provide strong information when the variance components are close to zero, which is common in practical network meta-analyses with binary outcomes. Alternatively, one of several separation strategies can be considered for modeling the covariance matrix. To study the impact of the IW prior on network meta-analysis results and compare it with separation strategies, we perform simulation studies under different mechanisms of missing treatments and we find that a separation strategy with appropriate priors for the correlation matrix and variances separately performs better. Finally, we illustrate the importance of sensitivity analyses with different prior specifications when performing NMA by re-analysis of three real examples.