多个响应变量的纵向数据联合建模方法及应用

Joint Modeling of Multivariate Longitudinal Data and Its Application

纵向观测过程中经常会遇到对同一个个体测量多个响应变量的情况,由于同一个个体的不同响应变量之间存在一定的相关性,因此独立分析每一个响应变量将会损失相关信息。本文利用联合建模的方式对多个响应变量建立混合效应模型,通过不同响应变量的随机效应之间的相关性刻画不同响应变量之间的关系,并利用极大似然方法给出模型中参数的估计,最后通过中风病的实际数据分析来说明该方法的应用。

Multiple outcomes measured repeatedly for the same subject are common in longitudinal observation. If we use the approach by analyzing each outcome separately, it may lead to wrong conclusions due to the failure of accounting for joint evolution of different outcomes. To adequately capture the interdependence among multiple outcomes, we proposed a joint modeling for multivariate longitudinal data by constructing a linear mixed-effects model for each outcome and accommodating the relationship among multiple outcomes through correlation in random effects. Maximum likelihood method was adopted to estimate parameters in this model. The application of this method was demonstrated through the analysis of stroke data.