Censored Rank Independence Screening for High-dimensional Survival Data

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• High-dimensional data, where the number of covariates exceeds the sample size, is common in statistical applications.
• Traditional correlation screening methods, while effective for linear models, struggle with contaminated data and censored outcomes in survival analysis.
• The sure screening property ensures that relevant variables are retained with high probability, which is crucial for accurate modeling.

Purpose of the Study:

• To develop a robust screening method for high-dimensional survival data that is resilient to outliers and data contamination.
• To extend the sure screening property to censored survival data, improving variable selection accuracy.
• To provide a flexible method applicable to a general class of survival models.

Main Methods:

• Censored rank independence screening is proposed as a novel approach for high-dimensional survival data.
• The method utilizes rank-based statistics to enhance robustness against outliers and data contamination.
• The sure screening property of the proposed method is theoretically established and empirically validated.

Main Results:

• The censored rank independence screening method demonstrates robustness to contaminated covariates and censored outcomes.
• Simulations and real-data analysis confirm the method's competitive performance on moderate-sized survival datasets with high-dimensional predictors.
• The proposed approach successfully retains relevant variables, achieving the sure screening property even under data contamination.

Conclusions:

• Censored rank independence screening offers a powerful and robust alternative for variable selection in high-dimensional survival data analysis.
• The method's resilience to outliers and contamination makes it particularly valuable for real-world datasets.
• This approach advances the field of survival analysis by providing a reliable tool for identifying significant predictors in complex, high-dimensional settings.