Robust Estimation and Variable Selection for the Accelerated Failure Time Model

TL;DR

This paper introduces a robust, unified EM-based method with L1 regularization for variable selection and parameter estimation in accelerated failure time models, effectively handling censored survival data and outliers.

Contribution

It develops a flexible framework that combines robust loss functions with variable selection, extending existing methods like Buckley-James to better handle outliers and group structures.

Findings

The proposed method improves prediction accuracy in simulations.

It effectively identifies relevant variables with regularization.

Application to ovarian cancer data demonstrates practical utility.

Abstract

This paper considers robust modeling of the survival time for cancer patients. Accurate prediction can be helpful for developing therapeutic and care strategies. We propose a unified Expectation-Maximization approach combined with the L1-norm penalty to perform variable selection and obtain parameter estimation simultaneously for the accelerated failure time model with right-censored survival data. Our approach can be used with general loss functions, and reduces to the well-known Buckley-James method when the squared-error loss is used without regularization. To mitigate the effects of outliers and heavy-tailed noise in the real application, we advocate the use of robust loss functions under our proposed framework. Simulation studies are conducted to evaluate the performance of the proposed approach with different loss functions, and an application to an ovarian carcinoma study is provided. Meanwhile, we extend our approach by incorporating the group structure of covariates.