纵向数据与生存数据联合模型中多变点识别问题

Multiple change points identification in joint modeling of longitudinal and survival data

提出了共享协变量和随机效应的纵向响应中含有多个变点识别的线性混合效应(LME)模型和加速失效时间(AFT)模型的联合模型,并通过Gauss-Hermite近似解决极大似然函数中的复杂积分以得到参数的估计.通过模拟研究验证了该方法的有效性,并将其应用于原发性胆汁性肝硬化(PBC)病变过程,研究发现:PBC患者的血清胆红素只在初期治疗阶段有所降低,两个月之后迅速开始反弹,直到3.5a后增速才有所放缓,说明治疗方法仍需改进.

A joint model with multiple change points ident ifying in longitudinal response process isproposed, which combines a linear mixed-effect （LME ） model and an accelerated failure time （AFT ） model with respect to shared covariates and random effects. All the parameters are estimated by themaximum likelihood function through the Gauss-Hermite approximation to deal w ith the intractable integrals in it . The effect of the method is elucidated through simulation studies and a real dataapplication about primary biliary cirrhosis （PBC）. It is shown that serum bilirub in level declines only at the beginning of treatment and lasts two months, then quickly rebounds and doesn,t slow downuntil 3. 5 years later, which indicates that the treatment methods stillneed to be improved.