Conventional statistical analysis methods encounter challenges when addressing precise medical treatment for high-dimensional survival data.Overcoming these challenges involves accounting for heterogeneity to mitigate...Conventional statistical analysis methods encounter challenges when addressing precise medical treatment for high-dimensional survival data.Overcoming these challenges involves accounting for heterogeneity to mitigate estimation bias and avoiding model overfitting for enhanced interpretability.This paper introduces a regularization technique based on the heterogeneous Cox model,enabling simultaneous variable selection,subgroup identification,and parameter estimation.The approach is entirely data-driven and capable of handling exponential increases in covariate dimension with sample size,as well as divergent numbers of significant variables.Under mild assumptions,we establish asymptotic properties for the proposed estimator,including the oracle property of variable selection,consistency in subgroup identification,and asymptotic normality.Leveraging the coordinate descent method and ADMM algorithm,we propose an efficient MCD-ADMM algorithm for optimization.Simulation studies further validate the effectiveness of our approach,complemented by an analysis of ovarian cancer data for illustrative purposes.展开更多
基金Supported by Natural Science Foundation of China(Grant Nos.12171374,12371275)。
文摘Conventional statistical analysis methods encounter challenges when addressing precise medical treatment for high-dimensional survival data.Overcoming these challenges involves accounting for heterogeneity to mitigate estimation bias and avoiding model overfitting for enhanced interpretability.This paper introduces a regularization technique based on the heterogeneous Cox model,enabling simultaneous variable selection,subgroup identification,and parameter estimation.The approach is entirely data-driven and capable of handling exponential increases in covariate dimension with sample size,as well as divergent numbers of significant variables.Under mild assumptions,we establish asymptotic properties for the proposed estimator,including the oracle property of variable selection,consistency in subgroup identification,and asymptotic normality.Leveraging the coordinate descent method and ADMM algorithm,we propose an efficient MCD-ADMM algorithm for optimization.Simulation studies further validate the effectiveness of our approach,complemented by an analysis of ovarian cancer data for illustrative purposes.
