Robust estimation and variable selection for the accelerated failure time model.

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Informatics

Background:

• Accurate prediction of patient survival time is critical for effective cancer treatment strategies.
• Existing methods may be sensitive to outliers and heavy-tailed noise in survival data.
• Accelerated failure time (AFT) models are widely used for survival data analysis.

Purpose of the Study:

• To propose a unified Expectation-Maximization (EM) approach for robust modeling of cancer patient survival time.
• To simultaneously perform variable selection and parameter estimation in AFT models with right-censored data.
• To mitigate the impact of outliers and noise using robust loss functions.

Main Methods:

• A unified Expectation-Maximization (EM) algorithm combined with L1-norm penalty for variable selection and parameter estimation.
• Application of general loss functions, including robust loss functions to handle outliers.
• Extension of the approach to incorporate group structures among covariates.

Main Results:

• The proposed EM approach with L1-norm penalty effectively performs simultaneous variable selection and parameter estimation.
• Robust loss functions improve model performance in the presence of outliers and heavy-tailed noise.
• The method demonstrated good performance in simulation studies and application to ovarian carcinoma data.

Conclusions:

• The unified EM approach provides a robust and flexible framework for modeling censored survival data in cancer research.
• The method enhances the accuracy of survival time prediction, aiding in the development of therapeutic strategies.
• The incorporation of robust loss functions is recommended for real-world applications with noisy data.