Background: Network meta-analysis is an extension of the classical pairwise meta-analysis and allows to compare multiple interventions based on both head-to-head comparisons within trials and indirect comparisons across trials. Bayesian or frequentist models are applied to obtain effect estimates with credible or confidence intervals. Furthermore, p-values or similar measures may be helpful for the comparison of the included arms but related methods are not yet addressed in the literature. In this article, we discuss how hypothesis testing can be done in a Bayesian network meta-analysis.
Methods: An index is presented and discussed in a Bayesian modeling framework. Simulation studies were performed to evaluate the characteristics of this index. The approach is illustrated by a real data example.
Results: The simulation studies revealed that the type I error rate is controlled. The approach can be applied in a superiority as well as in a non-inferiority setting.
Conclusions: Test decisions can be based on the proposed index. The index may be a valuable complement to the commonly reported results of network meta-analyses. The method is easy to apply and of no (noticeable) additional computational cost.
Keywords: Hypothesis testing; Network meta-analysis; Non-inferiority; Superiority; Treatment comparison.