High-Dimensional Variable Selection for the Additive Cox Model with Interval-Censored Data

• Activity Number: 417 - Recent advancement on life time data analysis
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: Lifetime Data Science Section
• Author(s): Tian Tian* and Jianguo Sun
• Companies: University of Missouri-Columbia and Univerisity of Missouri-Columbia
• Keywords: High-dimensional variable selection; Additive Cox model; Interval censored data; Bernstein polynomials; Group penalization; Sieve estimation

Abstract

This article studies variable selection for the additive Cox's model with interval-censored data when the number of additive components p is greater than the sample size n. Compared with the standard Cox's model, one advantage of the additive Cox's model is that it can capture both linear and nonlinear effects of the covariates. By combining Bernstein polynomials approximation and group penalization, a group penalized sieve maximum likelihood approach is proposed. To compute the proposed estimators, an efficient algorithm based on group coordinate descent is developed and is easy to implement. An extensive simulation study demonstrates that the proposed method performs well in finite sample situations. Finally, the proposed method is applied to an Alzheimer’s disease study to select important demographic, clinical and genetic factors that are significantly related to the risk of developing dementia.