Penalized estimation for competing risks regression with applications to high-dimensional covariates.

Area of Science:

• Biostatistics
• Bioinformatics
• Medical Research

Background:

• High-dimensional regression is crucial in biomedical research for analyzing complex patient bio-profiles and outcomes.
• Standard survival analysis has advanced for high-dimensional data, but competing risks remain less developed.
• Understanding disease dynamics and patient subgroups requires analyzing adverse events alongside primary outcomes.

Purpose of the Study:

• To develop a penalized regression framework for high-dimensional data with competing risks.
• To adapt the direct binomial regression model for fitting sparse regression models in competing risks scenarios.
• To provide an easily implementable method for analyzing complex survival data with multiple event types.

Main Methods:

• Reformulation of the direct binomial regression model into a penalized framework.
• Application of penalized regression techniques to handle high-dimensional covariates in competing risks.
• Utilizing existing high-performance software for penalized regression implementation.

Main Results:

• Simulation studies demonstrate the effectiveness of the proposed penalized regression approach.
• The method was successfully applied to genomic data for progression-free survival analysis.
• An R function is provided within the 'timereg' package for practical application.

Conclusions:

• The developed penalized regression approach effectively addresses competing risks in high-dimensional settings.
• This method facilitates the identification of patient subgroups with distinct prognoses and treatment responses.
• The readily available R function promotes the adoption of regularized competing risks regression in biomedical research.