Estimation and variable selection for semiparametric transformation models under a more efficient cohort sampling design
Estimation and variable selection for semiparametric transformation models under a more efficient cohort sampling design
复制标题
更高效队列抽样设计下半参数变换模型的估计和变量选择
DOI：
10.1007/s11749-017-0562-2
复制
发表时间：
2018
期刊：
影响因子：
1.3
通讯作者：
Ruofan Wu
中科院分区：
文献类型：
--
作者：
Mingzhe Wu;Ming Zheng;Wen Yu;Ruofan Wu
两阶段队列抽样设计，有时也被称为回顾性抽样设计，常用于大型队列研究中，以节省抽样时间和成本。常用的设计包括病例 - 队列设计、病例 - 对照设计、巢式病例 - 对照设计等。人们已经努力提高这些常用设计下的估计效率。我们在半参数变换模型类别下提出了一种不同的回顾性抽样设计，称为终点设计。设计了一种逆概率加权似然方法来估计模型参数，并且所提出的设计在协变量确定规模相当的情况下比病例 - 队列和病例 - 对照设计具有更高的效率。我们还考虑了在所提出的设计下进行变量选择。提出了一种带有自适应套索惩罚的特殊设计的目标函数。推导了所提出的估计和变量选择过程的大样本性质。进行了大量的模拟研究，为所提出的方法提供了有利的证据。分析了一个真实数据集用于说明。
Two-phase cohort sampling designs, or sometimes known as retrospective sampling designs, are often used in large cohort studies for saving sampling time and cost. Commonly used designs include case-cohort design, case-control design, nested case-control design, and so on. Efforts had been taken to improve the estimation efficiency under these commonly used designs. We propose a different retrospective sampling design, called end-point design, under the class of semiparametric transformation models. An inverse probability weighting likelihood approach is designed for estimating the model parameters, and the proposed design shows higher efficiency than the case-cohort and case-control design with comparable size of covariates ascertainment. We also consider variable selection under the proposed design. A specially designed objective function with adaptive lasso penalty is proposed. The large sample properties of the proposed estimation and variable selection procedure are developed. Extensive simulation studies are carried out to show favorable evidence for the proposed approaches. A real data set is analyzed for illustration.