当前状态数据的可加风险模型变量选择方法

A Variable Selection Method for the Additive Hazards Model with Current Status Data

当协变量数目较多时,变量选择对模型构建至关重要.近年来,以LASSO为代表的各种惩罚变量选择方法备受关注,但生存分析领域的惩罚变量选择方法研究大多基于Cox比例风险模型,且研究对象多为右删失数据.文章对当前状态数据(也称Ⅰ型区间删失数据)在可加风险模型下的变量选择方法进行研究.在失效时间服从可加风险模型及观测时间与协变量相关的假定下,从计数过程的角度来构造风险函数,并给出一种基于重复迭代加权的BAR (Broken Adaptive Ridge)惩罚似然变量选择方法,证明了Oracle性质.通过模拟实验来比较BAR与其他常用惩罚似然方法在变量选择方面的效果,最后利用文章提出的方法分析一项阿尔茨海默病的研究数据.模拟实验和实证分析都表明了BAR方法在变量选择方面表现良好.

Variable selection is vital to statistical modelling when the number of covariates is large.In recent years,penalty function-based methods represented by LASSO have attracted much attention.But most researches in survival analysis are based on the Cox proportional hazards model and right-censored data.In this paper,we consider current status data(also called type I interval-censored data) with the additive hazards model which is less studied.Under the assumption that the failure time follows the additive hazards model and the censoring time is dependent on covariates,the hazard function is constructed from the perspective of counting process,and then a simple likelihood function is derived.A BAR(Broken Adaptive Ridge)variable selection method is proposed,which is based on iteratively reweighted penalization and enjoys Oracle property.We compare BAR with some popular penalized methods through simulation and apply it to the current status data arising from the Alzheimer’s disease study.Both simulation and application show that BAR performs better compared with popular penalized methods.