L1 penalized estimation in the Cox proportional hazards model

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• High-dimensional data presents challenges for traditional statistical modeling.
• The Least Absolute Shrinkage and Selection Operator (LASSO) is effective for variable selection and shrinkage in such datasets.
• Existing algorithms for LASSO estimation may lack efficiency in complex models.

Purpose of the Study:

• To present a novel, efficient algorithm for computing L(1) penalized (LASSO) estimates.
• To demonstrate the algorithm's applicability across various models, including generalized linear models.
• To provide an implementation of the algorithm in an accessible R package.

Main Methods:

• The algorithm combines gradient ascent optimization with the Newton-Raphson method.
• It is designed for general likelihood functions, accommodating diverse statistical models.
• The method is specifically applied to the Cox proportional hazards model for survival analysis.

Main Results:

• The novel algorithm efficiently computes LASSO estimates in high-dimensional settings.
• Demonstrated effectiveness in predicting breast cancer patient survival using gene expression data.
• Performance comparison with existing methods shows competitive or superior results.

Conclusions:

• The new algorithm offers an efficient approach to LASSO estimation for high-dimensional data.
• Its flexibility allows application in various statistical models, including survival analysis.
• The availability of the 'penalized' R package facilitates broader adoption and use.