UC San Diego

In this dissertation, we utilize the novel statistical methods for obtaining causal effect under competing risks and semi-competing risks data in survival analysis. This dissertation is comprised of three main settings. In the first setting, we aim to assess the causal effect of mid-life alcohol exposure to the late life cognitive score which is related to Alzheimer’s disease (AD) using a large scale longitudinal data. We applied the marginal structural model (MSM) with inverse probability weighted (IPW) to adjust for time-varying confounding. We found that there is a significant decline in cognitive scores among heavy drinkers compared always light drinker. However, since the cognitive scores also changes over time, learning the relationship of alcohol exposure and time to cognitive impairment is also worth to explore.

In the second setting, we are interested in mid-life alcohol exposure to late life time to cognitive impairment which is also related to AD. Under this setting, as people are in their late-life stage, death prevents us from observing cognitive impairment. In survival analysis, death is considering as competing event. To estimate the causal effect of point treatment to time to event with the existence of competing event, we applied the MSM Cox proportional hazards model with IPW. Since hazard ratio is hard to interpret in medical research, we proposed predicted risk contrasts formula under the MSM Cox model.

Observing the trend that people die quickly after experiencing cognitive impairment, in the third settings, we proposed a MSM illness-death to assess the causal effect for alcohol exposure to time to cognitive impairment, death and death after cognitive impairment. We considered two specific such models, the usual Markov illness-death structural model and the general Markov illness-death structural model which incorporates a frailty term. For interpretation purposes, risk contrasts under the structural models are defined. To accommodate the possibility of misspecification of propensity score model, we also derived the augmented IPW estimator under MSM illness-death usual Markov model.