Network meta-analysis with Bayesian approach using web-based application

Vollständiger Abstract

Worum geht es in dieser Arbeit?

This research focuses on the underrepresentation of Bayesian network meta-analyses using R (NMAs-BR) in academic literature, aiming to spotlight the advantages of this Bayesian approach in complex, multi-treatment comparisons prevalent in medical and health research. The study underscores the need to understand the differences and potential benefits of NMAs-BR compared to traditional frequentist methods, particularly in terms of flexibility, integrating prior knowledge, and interpreting probabilistic results. The application of NMAs-BR against results derived from the An R package interfacing with JAGS (Just Another Gibbs Sampler) (rjags) package in R was examined. The study leverages data from a previous study to develop a custom NMAs-BR, utilizing the Metropolis-Hastings algorithm for log odds in Markov chain Monte Carlo simulations. The research design encompasses 3 major components: setting ten different prior beliefs for data distribution to be compared using Markov Chain Monte Carlo, including a normal distribution for log odds; comparing results between Bayesian and frequentist NMAs using the An R package for network meta-analysis, rjags, and An R package for network meta-analysis packages; and detailing the NMAs-BR process on a bespoke R-platform, which includes steps for data input, model fitting, and output interpretation. The results reveal that among the ten prior beliefs compared using Markov Chain Monte Carlo, the effect sizes with a normal distribution for log odds align most closely with those obtained from frequentist methods. Further comparisons between frequentist and Bayesian NMAs show that Bayesian approaches yield results very similar to frequentist ones, particularly when a prior normal distribution is set for log odds. The outcomes from NMAs-BR are comparable to those from the rjags package in R, albeit with smaller standard errors. However, the study notes that using rjags requires the separate installation of the Just Another Gibbs Sampler, software for Bayesian analysis (JAGS) software, a standalone program that complicates its setup, in contrast to the more user-friendly interface offered by NMAs-BR on an R Shiny application of MetaBNMA. The study highlights the merits of Bayesian NMAs in R over traditional frequentist methods. It underscores the comparative effectiveness and interpretative advantages of the Bayesian approach, and demonstrates that NMAs-BR yields similar results to those from the rjags package. The study provides an R platform equipped with NMAs-BR to assist researchers interested in conducting Bayesian NMAs in the future.