Quantile forward regression for high-dimensional survival data

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Existing prediction models often focus on average outcomes, failing to capture individual variations.
• Covariate effects can differ across the entire outcome distribution, necessitating quantile-specific analysis.

Purpose of the Study:

• To develop a flexible, high-dimensional survival data model that accounts for individual characteristics.
• To propose a quantile forward regression model for personalized risk prediction.

Main Methods:

• Utilizing quantile forward regression for high-dimensional survival data.
• Employing asymmetric Laplace distribution (ALD) maximization for variable selection.
• Applying extended Bayesian Information Criterion (EBIC) for final model derivation.

Main Results:

• The proposed method demonstrates sure screening property and selection consistency.
• Application to a national health survey dataset highlights the benefits of quantile-specific prediction.

Conclusions:

• The quantile forward regression model provides a more accurate and flexible approach to risk prediction.
• This method enhances personalized medicine by considering individual-specific covariate effects.