The Dantzig Selector for Censored Linear Regression Models.

Area of Science:

• Statistics
• Biostatistics
• Genomics

Background:

• Regularized regression models, like the Dantzig selector, are effective for variable selection.
• Existing Dantzig selector methods primarily address fully-observed response variables.
• Right-censored data is common in clinical studies, necessitating specialized analytical tools.

Purpose of the Study:

• To propose novel adaptive Dantzig variable selectors for linear regression models with right-censored response variables.
• To address the challenge of identifying predictive genes for event-free survival in clinical settings, specifically multiple myeloma.
• To establish theoretical guarantees for the new selectors regarding model selection consistency and estimation efficiency.

Main Methods:

• Development of a new class of adaptive Dantzig selectors tailored for right-censored data.
• Theoretical analysis to prove consistency in model selection and optimal efficiency of estimation under mild conditions.
• Extensive simulations to validate the practical performance of the proposed methods.

Main Results:

• The proposed adaptive Dantzig selectors demonstrate theoretical properties including consistency in model selection.
• The methods achieve optimal efficiency in estimation, matching that of known true models.
• Simulations confirm the practical utility and effectiveness of the new selectors.

Conclusions:

• The developed adaptive Dantzig selectors provide a robust approach for variable selection in linear regression with right-censored data.
• These methods are particularly valuable for clinical studies aiming to identify prognostic markers, such as genes predicting survival.
• The approach was successfully applied to a multiple myeloma clinical trial, identifying significant predictive genes.