A comparison of Frequentist and Bayesian network meta analysis with a case study from dental health data

Abstract

Network meta analysis is an extension of pair wise meta analysis where both direct as well as indirect treatment effects relative to a chosen reference treatment arm can be obtained from combined pool of studies as long as all the treatment arms are connected directly or indirectly via a network based on individual trials or studies. The aim of this network meta-analysis is to summarize and compare the results from Frequentist and Bayesian methods of the direct and indirect clinical evidence on the effectiveness of professionally applied topical fluorides in preventing dental root caries. Both the statistical approaches use different frameworks and each provides unique insights into the network of treatments approved for dental root caries. While the Frequentist approach provides the relative treatment effects along with corresponding confidence intervals compared to the usual care group, the Bayesian approach provides the relative treatment effects along with corresponding credible intervals. The full posterior distribution of treatment effects can be obtained using the Bayesian framework. Both approaches show similar direction of outcomes with subtle differences. A comparative analysis is presented and discussed using a case study using few topical fluoride-based treatment for dental caries. Various aspects of differences in approach as well as diagnostic checks and treatment ranking methods in both the frameworks are described. The netmeta package of R is used for the Frequentist approach while the geMTC package of R is used for Bayesian Analysis.