Weighting Methods For Causal Inference With Survival Outcomes

This dissertation aims to propose a technique for estimating the causal effects of exposure on survival outcomes using the Rubin Causal Model (RCM), a framework for defining causal estimands, discussing assumptions, and developing methods for drawing inferences on causal effects. From a substantive perspective, the research was motivated by the evaluation of the effect of two different treatments (Interferon versus Azathioprine) on time to the first worsening of Multiple Sclerosis (MS) disease. The study uses data from an observational study on patients with MS collected between 1981 and 2019 in Tuscany, Italy. The causal analysis of this study raises several challenges due to the unknown treatment assignment mechanism, and the survival outcome is subject to two different covariate-dependent censoring mechanisms: administrative censoring and treatment switching. Then, using Marginal Structural Cox models, we propose a new weighting method to adjust for observed confounders and correct selection bias due to different types of censoring.