标题： 半参数加速失效时间模型中分层抽样设计的变量选择与聚类失效时间 显示英文标题

标题： Variable Selection for Stratified Sampling Designs in Semiparametric Accelerated Failure Time Models with Clustered Failure Times

摘要： 在大规模流行病学研究中，由于分层抽样设计下的失效时间，统计推断常常受到高维协变量的复杂影响。 针对完整队列数据开发的变量选择方法并不能自然地扩展到分层抽样设计，因此需要对抽样方案进行适当的调整。 当失效时间存在聚类并且表现出组内相关性时，会带来进一步的挑战。 作为当比例风险假设不成立时Cox比例风险（PH）模型的替代方法，对于加速失效时间（AFT）模型的惩罚Buckley-James（BJ）估计方法可以通过结合广义估计方程（GEE）技术来处理这种情形中的组内相关性，尽管其实际实施仍受计算不稳定的阻碍。 我们提出了一种在GEE框架内的正则化估计方法，用于分层抽样设计，其理念类似于惩罚BJ方法，但具有可靠的推断过程。 我们建立了所提出估计量的一致性和渐近正态性，并证明它们实现了Oracle属性。 大量的模拟研究表明，我们的方法在忽略抽样偏差或组内相关性的现有方法中表现更优。 此外，正则化方案即使在中等样本量的情况下也能有效选择相关变量。 所提出的方法通过一个牙科研究的应用进行了说明。 显示英文摘要

摘要： In large-scale epidemiological studies, statistical inference is often complicated by high-dimensional covariates under stratified sampling designs for failure times. Variable selection methods developed for full cohort data do not extend naturally to stratified sampling designs, and appropriate adjustments for the sampling scheme are necessary. Further challenges arise when the failure times are clustered and exhibit within-cluster dependence. As an alternative of Cox proportional hazards (PH) model when the PH assumption is not valid, the penalized Buckley-James (BJ) estimating method for accelerated failure time (AFT) models can potentially handle within-cluster correlation in such setting by incorporating generalized estimating equation (GEE) techniques, though its practical implementation remains hindered by computational instability. We propose a regularized estimating method within the GEE framework for stratified sampling designs, in the spirit of the penalized BJ method but with a reliable inference procedure. We establish the consistency and asymptotic normality of the proposed estimators and show that they achieve the oracle property. Extensive simulation studies demonstrate that our method outperforms existing methods that ignore sampling bias or within-cluster dependence. Moreover, the regularization scheme effectively selects relevant variables even with moderate sample sizes. The proposed methodology is illustrated through applications to a dental study.