Bayesian Methods for Multivariate Meta-Analysis of Treatment Networks with Incomplete Data

• Activity Number: 308
• Type: Invited
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Christopher Schmid*+ and Thomas Trikalinos
• Companies: Brown University and Tufts Medical Center
• Keywords: network meta-analysis ; diagnostic tests ; comparative effectiveness ; individual participant data ; missing data ; multilevel models

Abstract

Comparative effectiveness research involves the comparison of a set of interventions or tests to assess relative benefits and harms. For interventions, they necessarily involve comparisons of multiple treatments and multiple outcomes potentially at multiple times with the goal of ranking treatments from best to worse with appropriate uncertainty. For medical tests, cross-test comparisons of test performance metrics (e.g., sensitivity and specificity) are of interest. Bayesian methods are particularly useful for attaching relevant probabilities to parameters of complex hierarchical models for data with potentially small numbers and incomplete structures. This talk presents a general formulation of Bayesian approaches to meta-analysis for comparative effectiveness studies that addresses each of these elements in the context of both aggregate and individual patient data. It also focuses on the sensitivity of conclusions to choice of prior distribution on the between-study variance.