Bootstrapping methods for assessing causality in survival analysis: A case study on major adverse cardiovascular events

Abstract: The combination of graphical models with Aalen's additive hazards model, resulting in a model known as dynamical path analysis, permits assessing the effects of different variables on times until an event and decomposing these total effects into direct and indirect effects. This thesis proposes novel bootstrapping methods designed for left-truncated right-censored data, conditional on covariates within the framework of Aalen's additive hazards model, in order to obtain confidence intervals for the estimates. To illustrate the practical application of the bootstrapping methods, we conduct a case study utilising data from the Malmö diet and cancer study. The data set consists of left-truncated right-censored data. Our analysis aims to examine causality and estimate the direct effects of various covariates on the incidence of major adverse cardiovascular events and indirect effects between covariates. We compute confidence intervals for these effects with the proposed bootstrapping methods.