Variable Selection for High-dimensional Cox Model with Error Rate Control

Simultaneously finding active predictors and controlling the false discovery rate (FDR) for high-dimensional survival data is an important but challenging statistical problem. In this paper, the authors propose a novel variable selection procedure with error rate control for the high-dimensional Cox model. By adopting a data-splitting strategy, the authors construct a series of symmetric statistics and then utilize the symmetry property to derive a data-driven threshold to achieve error rate control. The authors establish finite-sample and asymptotic FDR control results under some mild conditions. Simulation results as well as a real data application show that the proposed approach successfully controls FDR and is often more powerful than the competing approaches.