Model averaging for high-dimensional additive hazard regression with current status data

View at publisher Summarise with VIRAT
Authors: Yujing Han, Yaning Han, Jing Zhang
Cited by: 0
References: 30

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Current status data, where event times are only known or observed relative to monitoring times, are common in biomedical studies. Under the high-dimensional settings, regularization methods have been proposed for variable selection in the literature, however, they may result in misleading statistical inference and prediction if the model is missspecified. As an alternative, model averaging approach that combines the strengths of a set of candidate models can mitigate the risk of misspecification and enhance the prediction accuracy. The existing model averaging approaches are mainly developed for complete data under the low-dimensional settings, limit work has been done for survival data which may subject to censoring. This paper focuses on model averaging procedure for high-dimensional current status data within the additive hazard regression model. We first construct a class of additive hazard regression models, each with a small number of covariates, then derive non-restricted optimal weights (not constrained to sum to one) to combine them. Furthermore, we propose a computationally efficient greedy algorithm for high-dimensional scalability. Simulation studies demonstrate that our method outperforms regularization techniques in prediction accuracy. As an illustration, we apply the proposed method to analyze real data from the primary sclerosing cholangitis study.

Fields of Study & Subjects

Statistical Methods and Inference Statistical Distribution Estimation and Applications Probability and Statistical Research

Publication details

Journal / source
Communications in Statistics - Simulation and Computation
Pages
1–14
Publication date
18 March 2026
Language
EN