A Group Structure Guided Ultra-High Dimensional Feature Screening for Survival Outcome

With the rapid advancement of high-throughput technologies, feature screening methods have attracted increasing attention for analyzing ultra-high-dimensional data. Motivated by the widespread availability of meaningful grouping structures in biomedical applications, such as brain imaging and gene expression studies, we propose a novel group-structure-guided (GSG) feature screening method for survival outcomes. The proposed approach incorporates prior grouping information among predictors and accommodates both disjoint and overlapping group structures. Furthermore, a combined GSG (C-GSG) extension is developed to integrate multiple grouping schemes. We establish the sure screening properties of the proposed procedures and demonstrate through extensive simulation studies that incorporating informative group structures can improve screening performance, particularly under challenging censoring scenarios. The proposed method is further applied to a TCGA breast cancer dataset, where pathway information is utilized to identify genes associated with patient survival outcomes.