The Impact of Performing a Network Meta-Analysis with Imperfect Evidence

Network meta-analysis (NMA) is an important aspect of evidence synthesis in a clinical setting, as it allows us to compare treatments which may not have been analysed in the same trial. In an ideal scenario we would have a fully connected network of randomised controlled trials (RCTs) when undertaking an NMA. Ideally, these RCTs would contain the full patient population for a particular disease, and individual patient data (IPD) would be available for all trials. However, in reality we are never going to have all this information. Therefore, this thesis investigates methods for dealing with imperfect evidence. We consider two techniques for adjusting for confounding variables due to differing patient populations in a connected network. Firstly, we assess the benefit of the extra effort involved in obtaining and including IPD in an NMA. Secondly, we evaluate the impact of using IPD to adjust for differing trial populations through the increasingly popular method of matching adjusted indirect comparison. We also propose a method for including single-arm evidence in a disconnected network through aggregate level matching, and analyse the impact of this method. Although our work mainly focuses on the methodological aspects, all methods are illustrated using real world datasets, namely Hepatitis C virus (HCV) infection, melanoma and multiple myeloma