|
Abstract:
|
Network Meta-Analysis (NMA) is a useful tool to simultaneously compare multiple interventions while combining information from a number of studies. One common challenge in applying NMA in clinical data is that outcomes of interest are not always reported in all the studies. This can possibly be handled by imposing a "constant relative potency" assumption, meaning any intervention effects have a constant mean across all outcomes. In reality, however, this assumption rarely holds and cannot be validated. To address this challenge, we propose modeling the correlation structure between outcome of interest that is potentially missing from part of the network and the outcome(s) available throughout the network. Then the correlation structure can be extended to studies where outcome of interest is assumed to be missing at random (MAR). Simulation studies are conducted to demonstrate advantages of the proposed model in comparison with currently available methods, followed by applications to real-life data.
|
