Penalty and Shrinkage Strategies Based on Local Polynomials for Right-Censored Partially Linear Regression.

Area of Science:

• Statistics
• Biostatistics
• Medical Informatics

Background:

• Semiparametric regression models are crucial for analyzing complex data, especially with censored outcomes.
• Existing estimation methods may face challenges with data exhibiting right-censorship.
• Penalized regression techniques offer potential for improved estimation in such scenarios.

Purpose of the Study:

• To propose novel modified semiparametric estimators for right-censored regression models.
• To evaluate the performance of estimators utilizing six distinct penalty and shrinkage strategies.
• To introduce the local polynomial method as a smoothing technique within this framework.

Main Methods:

• Development of modified semiparametric estimators incorporating ridge, lasso, adaptive lasso, SCAD, MCP, and elasticnet penalties.
• Application of the local polynomial method for smoothing.
• Theoretical explanation of estimation procedures.
• Simulation studies for performance evaluation and comparison.
• Analysis of real-world data using hepatocellular carcinoma data.

Main Results:

• The adaptive lasso and SCAD-based estimators exhibited enhanced resistance to right-censorship.
• These estimators demonstrated superior performance compared to the other four penalized methods evaluated.
• The local polynomial smoothing approach proved effective in the semiparametric estimation context.

Conclusions:

• Modified semiparametric estimators, particularly those using adaptive lasso and SCAD, offer robust solutions for right-censored data.
• The integration of local polynomial smoothing enhances the utility of these penalized methods.
• The findings provide valuable tools for statistical analysis in fields with censored data, such as medical research.