Estimating the Risk of Getting Alzheimer’s Disease by A Two-Stage Model

Alzheimer’s disease (AD) is the most common type of degenerative dementia, and the increasing prevalence of AD in this century intensifies the need for greater research efforts. Given the irreversible nature of the disease, it is especially important to detect AD at the earlier stage to treat the disease effectively. In this thesis, I analyze the data collected from an AD study conducted by RUSH University to investigate the longitudinal cognitive changes and facilitate the diagnosis of early AD. I first model the annually assessed cognitive item responses by the generalized linear mixed effects models (GLMM) to characterize the features of longitudinal impairment, while adjusting the demographic covariates. The fitted random effects, which represent the extracted individual feature, are then incorporated in the survival models to predict the early stage of AD. Here, the onset of diagnosis of mild cognitive impairment (MCI) is defined as the event time. Finally, I present three proportional Hazards models and a time-dependent Cox model based on the demographic covariates, baseline cognitive scores, and the individual random effects from the GLMM. The analysis shows that the demographic covariates are associated with baseline cognitive scores and have a significant impact on the onset of MCI. When both demographic covariates and baseline cognitive scores are used to predict the onset of MCI, the associations are confounded and hard to interpret. On the other hand, my presented two-stage model that combines the GLMM and the survival model corrects the confounding issue, captures longitudinal cognitive changes, and identifies the significant prognostic factors for detecting the early AD