Grant: Catalyst: Methods Post-Market DSE - Bayesian methods and statistical models

Over the past 2 decades, the alternative treatment options for the same clinical condition and the ever-increasing number of clinical trials has considerably increased the conduct of systematic reviews including network meta-analysis (NMA). NMA compares the efficacy of all available treatments assuming consistency, whereby different sources of evidence in the network should be in agreement. A potential violation of consistency may lead to misleading results. Despite the efforts of researchers to construct a consistent network of treatments, significant inconsistency may arise which should be investigated when found. One of the strategies suggested to tackle inconsistency is to encompass it in the NMA results and apply inconsistency models relaxing the consistency assumption. This is a similar approach that researchers use to conduct a pairwise meta-analysis using the random effects model. In such models, an extra variation is added, named the inconsistency variance, so as to incorporate any unexplained inconsistency. However, in networks with only a few closed loops (i.e. certain paths in the network of evidence that start and end at the same treatment node via one or more intermediate treatments) and a small number of studies in each closed loop, it is difficult to estimate the inconsistency parameters and inconsistency variance. Bayesian models allow incorporation of external evidence on inconsistency variance, and offer advantages over conventional inconsistency models where no prior information on inconsistency is considered. This research will provide empirical evidence on the extent of the inconsistency variance, which will be particularly beneficial in NMAs with few closed loops and few included studies. The comparison of the estimated magnitude of inconsistency variance in a network with the empirical distribution of inconsistency variance can provide insight into the validity of the NMA results and inform decision making.