Variable selection and structure estimation for ultrahigh-dimensional additive hazards models

Abstract

We develop a class of regularization methods based on the penalized sieve least squares for simultaneous model pursuit, variable selection, and estimation in high-dimensional additive hazards regression models. In the framework of sparse ultrahigh-dimensional models, the asymptotic properties of the proposed estimators include structure identification consistency and oracle variable selection. The computational process can be efficiently implemented by applying the blockwise majorization descent algorithm. Simulation studies demonstrate the performance of the proposed methodology, and the primary biliary cirrhosis data analysis is provided for illustration.