關鍵字：聯合模型、長期追蹤資料、線性混合效應模型、加速失敗時間風險模型 、EM演算法、拔靴法。;In this thesis, we use CD4 and viral load to predict AIDS patients’ time to death and explore whether HAART works or not when patients were treated with it. The study which contains longitudinal covariates and survival time information is usually subject to measurement errors. Thus, we use joint model to solve this problem instead of partial likelihood approach. The approach uses linear mixed effects models to fit the longitudinal part and conducts the likelihood ratio test to select the suitable longitudinal model, and utilizes the accelerated failure time model to describe the relationship between covariates and survival time information. We combine these two parts to build a joint likelihood function and estimate the maximum likelihood estimators of all parameters by EM algorithm. The estimates of standard errors are derived by bootstrap method.
Keywords: join model, longitudinal data, linear mixed effects model, accelerated failure time model, EM algorithm, bootstrap method.
