We propose an iterative variable selection method for the accelerated failure time model using high-dimensional survival data. Our method pioneers the use of the recently proposed structured screen-and-select framework for survival analysis. We use the marginal utility as the measure of association to inform the structured screening process. For the selection steps, we use Bayesian model selection based on non-local priors. We compare the proposed method with a few well-known methods. Assessment in terms of true positive rate and false discovery rate shows the usefulness of our method. We have implemented the method within the R package GWASinlps.

翻译：本文针对高维生存数据，提出了一种用于加速失效时间模型的迭代变量选择方法。该方法首次将近期提出的结构化筛选-选择框架应用于生存分析领域。我们采用边际效用作为关联性度量指标来指导结构化筛选过程。在选择阶段，我们使用基于非局部先验的贝叶斯模型选择方法。通过将所提方法与若干经典方法进行比较，在真阳性率和错误发现率方面的评估结果验证了本方法的有效性。我们已在R软件包GWASinlps中实现了该方法。