A Greedy-Type Variable Selection Procedure for Selecting High-Dimensional Cox Models

• Activity Number: 127 - SPEED: Statistical Learning and Data Science Speed Session 1, Part 1
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Chien-Tong Lin* and Yu-Jen Cheng and Ching-Kang Ing
• Companies: and National Tsing Hua University and National Tsin Hua University
• Keywords: Greedy algorithm; Cox model; Sure screening property; high-dimensional information criterion; variable selection consistency

Abstract

Model selection for sparse high-dimensional Cox models has broad applications to contemporary biostatistics, in particular, to extracting relevant biomarkers from high-dimensional survival data. In this talk, we propose using a greedy-type algorithm, Chebyshev Greedy Algorithm (CGA), to iteratively include covariates in the aforementioned models, and show that with probability tending to one, all relevant covariates can be included in a moderate number of iterations. We also devise a high-dimensional information criterion (HDIC) to remove the redundant covariates chosen by CGA, thereby leading to selection consistency. Finally, the proposed method is illustrated using simulated data and a diffuse large B-cell lymphoma (DLBCL) dataset.