Variable Selection with Broken Adaptive Ridge Regression for Interval-Censored Competing Risks Data

2025-10-19

AI Summary

This paper addresses variable selection for interval-censored competing risks data. We propose a semiparametric transformation regression method incorporating the broken adaptive ridge (BAR) penalty—the first application of BAR to this data setting. The method simultaneously performs variable selection and effect estimation under both proportional and non-proportional hazards assumptions, achieving sparsity and oracle properties. Parameter estimation is conducted via penalized likelihood maximization, implemented through an iterative reweighted algorithm that accommodates joint modeling of multiple mutually exclusive failure types. Simulation studies demonstrate high selection accuracy and robust estimation performance across diverse scenarios. Applied to a real-world HIV cohort study, the method successfully identifies clinically significant risk factors, confirming its practical utility and reliability.

Technology Category

• Reasoning under Uncertainty: Applications
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Competing risks data refer to situations where the occurrence of one event pre- cludes the possibility of other events happening, resulting in multiple mutually exclusive events. This data type is commonly encountered in medical research and clinical trials, exploring the interplay between different events and informing decision-making in fields such as healthcare and epidemiology. We develop a penal- ized variable selection procedure to handle such complex data in an interval-censored setting. We consider a broad class of semiparametric transformation regression mod- els, including popular models such as proportional and non-proportional hazards models. To promote sparsity and select variables specific to each event, we employ the broken adaptive ridge (BAR) penalty. This approach allows us to simultane- ously select important risk factors and estimate their effects for each event under investigation. We establish the oracle property of the BAR procedure and evaluate its performance through simulation studies. The proposed method is applied to a real-life HIV cohort dataset, further validating its applicability in practice.

Problem

• Selecting variables for interval-censored competing risks data
• Employing broken adaptive ridge penalty for sparsity promotion
• Estimating event-specific risk factors using semiparametric regression models

Innovation

• Broken adaptive ridge penalty for variable selection
• Handles interval-censored competing risks data
• Uses semiparametric transformation regression models