Weighted SurvClipper: Nonlinear Prognostic Biomarker Selection Incorporating Historical Information for Survival Risk With Controlled FDR

Recent advances in genome technology development have led to a high demand for statistical techniques that ensure a comprehensive selection of prognostic genes for cancer studies. In this paper, we introduce a novel high-dimensional variable selection method for right-censored outcomes, namely SurvClipper, that can control false discovery rate (FDR) and accommodate complex nonlinear associations between the covariates and the outcome. To enable full utilization of information from historical research and provide an effective strategy for variable selection, we further propose wSurvClipper, a weighted FDR control procedure that incorporates prior domain information, which enhances the integrity and interpretability of the results. Simulation studies show good control of the type-I error rate and high power of detecting important prognostic biomarkers including those with complicated nonlinear effects on the outcome in many settings. Moreover, the power gain using wSurvClipper is substantial and remarkably robust to incorrect prior information. The proposed method is illustrated via application to a skin cutaneous melanoma dataset from the public domain.