Network Meta-Analysis of Disconnected Networks: How Dangerous Are Random Baseline Treatment Effects?

• Activity Number: 87 - Invited ePoster Session: a Statistical Smörgåsbord
• Type: Invited
• Date/Time: Sunday, July 29, 2018 : 8:30 PM to 10:30 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Audrey Béliveau* and Sarah Goring and Robert W Platt and Paul Gustafson
• Companies: University of Waterloo and SMG Outcomes Research and McGill University and University of British Columbia
• Keywords: network meta-analysis; disconnected networks; Bayesian; exchangeability assumption; systematic review; mixed treatment comparisons

Abstract

In the context of systematic reviews, disconnected networks arise when some treatments are not compared directly (within a study) or indirectly (through studies with common treatments). The analysis of disconnected networks is usually avoided due to the lack of a gold-standard method and a fear of model misspecification. In this project, we take the suggestion of Goring et al. (2016) of using random baseline treatment effects (RBTE) in a contrast-based (CB) model for disconnected networks and investigate to what extent this approach might be dangerous. We take two publicly available datasets of connected networks, and disconnect them in multiple ways. We then compare 1. the analysis of each disconnected network using a Bayesian CB model with normally distributed RBTE to 2. the analysis of their initial connected network using a Bayesian CB model with fixed BTE (a standard approach for connected networks). For the two datasets studied, we found that the use of RBTE for treatment comparisons in disconnected networks was overall appropriate. Since those datasets were not cherry-picked, this suggests that there are other disconnected networks that could benefit from the RBTE approach.