Time-varying Bayesian Network Meta-Analysis

TL;DR

This paper introduces a novel time-varying Bayesian Network Meta-Analysis (tBNMA) method that models changes in treatment effects over time, applied to MRSA cSSSI treatments to reveal evolving efficacy trends.

Contribution

It develops a time-varying BNMA approach using Gaussian Processes to address inconsistencies in treatment effect over time, applied to a comprehensive MRSA dataset.

Findings

Vancomycin's effectiveness decreased relative to linezolid between 2002-2007.

Vancomycin's efficacy has since recovered to be statistically equivalent.

The method captures non-linear trends in treatment effects over time.

Abstract

The presence of methicillin-resistant \textit{Staphylococus Aureus} (MRSA) in complicated skin and soft structure infections (cSSSI) is associated with greater health risks and economic costs to patients. There is concern that MRSA is becoming resistant to other "gold standard" treatments such as vancomycin, and there is disagreement about the relative efficacy of vancocymin compared to linezolid. There are several review papers employing Bayesian Network Meta-Analyses (BNMAs) to investigate which treatments are best for MRSA related cSSSIs, but none address time-based design inconsistencies. This paper proposes a time-varying BNMA (tBNMA), which models time-varying treatment effects across studies using a Gaussian Process kernel. A dataset is compiled from nine existing MRSA cSSSI NMA review papers containing 58 studies comparing 19 treatments over 19 years. tBNMA finds evidence of a non-linear trend in the treatment effect of vancomycin - it became less effective than linezolid between 2002 and 2007, but has since recovered statistical equivalence.