A0571
Title: The comparison of MARMoT adjustment and template matching in a multiple treatment framework: A simulation study
Authors: Margherita Silan - Department of Statistical Sciences, University of Padova (Italy) [presenting]
Pietro Belloni - University of Padua (Italy)
Abstract: Specific statistical tools are required to estimate a causal effect when many treatments are involved. Case studies involving such a number of treatments are rare in the scientific literature and typically are based on two main methods: Matching on poset-based average rank for multiple treatments (MARMoT) and template matching. The aim is to compare the two techniques through a simulation study in various scenarios. Those artificial scenarios are built that vary in multiple aspects, such as the number of treatments (50, 250, 500), the presence of rare treatments, and the presence of rare confounders. In addition, minor technical adjustments were made to enhance the performance of both techniques in the different settings. The objective is to empirically determine which technique performs better in each scenario. These methods were also applied to real data from the Medicare database, comparing 41 medical facilities on their performance with elderly patients undergoing cardiac surgeries. Conclusions could provide valuable insights for choosing and implementing statistical methods to address self-selection bias in multi-treatment observational studies.
