Reproducible learning in high-dimensional Cox models

In survival analysis, the Cox proportional hazards model is one of the most popular models widely used in biomedical sciences and epidemiology. Despite significant advancements, how to ensure the reproducibility and robustness of feature selection in high-dimensional Cox models remains largely unexplored, especially when the population precision matrix of features is unknown. In this paper, we introduce a new feature selection procedure using knockoffs to facilitate false discovery rate (FDR) control in high-dimensional Cox models. We establish the robustness of FDR control for the proposed method when the population precision matrix is unknown and replaced by some consistent estimate. Through numerical studies and real-data applications, we demonstrate the effectiveness of our method in controlling FDR while ensuring statistical power.