A0851
Title: Exploiting multivariate network meta-analysis: A calibrated Bayesian composite likelihood inference
Authors: Yifei Wang - Southern Methodist University (United States)
Lifeng Lin - Florida State University (United States)
Yu-Lun Liu - University of Texas Southwestern Medical Center (United States) [presenting]
Abstract: Multivariate network meta-analysis serves as a valuable framework for synthesizing evidence across multiple treatments and outcomes, offering increased efficiency and broader clinical relevance compared to univariate approaches. Despite these advantages, practical implementation is often hindered by the lack of reported within-study correlations among treatments and outcomes, which can compromise estimation accuracy and inference validity. To address this challenge, a calibrated Bayesian composite likelihood approach is proposed, which eliminates the need for a fully specified joint likelihood while accommodating partial or missing within-study correlation structures. The estimation procedure integrates a hybrid Gibbs sampling algorithm with a post-sampling open-faced sandwich adjustment to enhance posterior inference robustness. Extensive simulation studies demonstrate that the proposed method achieves nearly unbiased estimation and maintains nominal coverage probabilities across a range of realistic scenarios. The approach is further illustrated through applications to two real-world datasets: One involving comparisons of treatment procedures for root coverage and another assessing interventions for anemia in patients with chronic kidney disease. These examples highlight the method's practical utility and potential to improve the reliability of multivariate evidence synthesis in the presence of incomplete correlation information.
