基于贝叶斯推断的HIV非线性混合效应联合模型研究  

Bayesian Inference for Nonlinear Mixed-effects Joint Model in HIV Study

作　　者：李莹[1] 王仲君[2] 赵华玲[1] 

机构地区：[1]武汉理工大学 [2]武汉理工大学统计学系

出　　处：《统计研究》2012年第7期86-90,共5页Statistical Research

基　　金：国家自然科学重点项目(20634030);中央高校基本科研业务费专项资金资助(2011-IV-116)

摘　　要：在纵向数据研究中,混合效应模型的随机误差通常采用正态性假设。而诸如病毒载量和CD4细胞数目等病毒性数据通常呈现偏斜性,因此正态性假设可能影响模型结果甚至导致错误的结论。在HIV动力学研究中,病毒响应值往往与协变量相关,且协变量的测量值通常存在误差,为此论文中联立协变量过程建立具有偏正态分布的非线性混合效应联合模型,并用贝叶斯推断方法估计模型的参数。由于协变量能够解释个体内的部分变化,因此协变量过程的模型选择对病毒载量的拟合效果有重要的影响。本文提出了一次移动平均模型作为协变量过程的改进模型,比较后发现当协变量采用移动平均模型时,病毒载量模型的拟合效果更好。该结果对协变量模型的研究具有重要的指导意义。In longitudinal data study,normality of the model random error is a routine assumption for mixed-effects models.Viral data,such as viral load and CD4 count,is generally skewed,therefore the assumption of normality may affect model results and even lead to inappropriate conclusion.In HIV dynamic research,viral response is always associated with covariates which are often measured with substantial errors.Considering the situations mentioned above,we propose a skew-normal nonlinear mixed-effects（SN-NLME） joint model in this paper incorporating CD4 covariate process and apply the Bayesian inference method in model estimation.Since the choice of covariate process model is crucial to fitting-effect of response model,in which within-subjects variation is partially affected by covariates,we propose moving average model as improved model to fit CD4 covariate.Comparing the MA model with LME model for covariate process,we find that fitting-effect of viral load is better when MA model is applied,which may guide for the study of covariate model.

关 键 词：纵向 艾滋病 联合模型 偏正态 协变量 

分 类 号：F222.3[经济管理—国民经济]