A0761
Title: Weighted least squares estimation for semiparametric accelerated failure time model with regularization
Authors: Ying Chen - The university of Texas at Dallas (United States) [presenting]
Chuan-Fa Tang - University of Texas at Dallas (United States)
Sy Han Chiou - Southern Methodist University (United States)
Min Chen - University of Texas at Dallas (United States)
Abstract: Clustered failure time data arise when failure times are sampled in clusters. Depending on the sampling schemes, the selected sample of clusters might not be representative of the population. For example, case-cohort sampling is commonly adopted for studying large cohorts with rare events, and subjects who experienced events of interest are more likely to be sampled. The semiparametric accelerated failure time (AFT) model is an appealing method in survival analysis as it directly relates the failure times to a linear combination of covariates. However, most approaches in the AFT model framework are rank-based and rely on solving nonsmooth estimating equations. To facilitate inference procedures for AFT models, we extended a generalized estimating equation (GEE) embedded approach to account for the within-cluster correlations and the sampling bias. In a high-dimensional data setting, we propose to penalize the corresponding GEE for variable selection. We demonstrate the effectiveness of the proposed methods with high-dimensional data simulated under generalized stratified case-cohort designs. The large-scale simulation shows the proposed methods are more efficient than the estimator which ignores the sampling weights or within-cluster dependence.
