Penalized weighted proportional hazards model for robust variable selection and outlier detection

Area of Science:

• Biostatistics
• Precision Medicine
• Cancer Research

Background:

• Identifying exceptional responders is crucial in precision medicine for understanding differential treatment effects.
• Prostate cancer clinical trials present a specific need for characterizing patient subgroups with unique responses.
• Censored survival data and outlier masking pose challenges in robust statistical modeling.

Purpose of the Study:

• To develop a robust statistical method for outlier detection and regression in sparse proportional hazards models with censored survival data.
• To simultaneously perform variable selection and identify exceptional responders in prostate cancer studies.
• To address the challenge of outlier masking in the presence of censored outcomes.

Main Methods:

• Proposed a novel method modeling observation irregularity using individual weights in the hazard function.
• Applied a LASSO-type penalty for simultaneous variable selection and outlier detection.
• Transformed the optimization problem into a penalized maximum partial likelihood problem for easy implementation.
• Extended the method to handle outlier masking issues specific to censored data.

Main Results:

• The proposed method effectively performs simultaneous variable selection and outlier detection.
• The technique is robust to potential outlier masking caused by censored outcomes.
• Demonstrated performance through extensive simulations and real-world data analyses.
• Validated the approach in both low-dimensional and high-dimensional settings.

Conclusions:

• The developed statistical framework enables simultaneous variable selection and outlier detection in sparse proportional hazards models.
• This method is valuable for characterizing exceptional responders in precision medicine, particularly in prostate cancer research.
• The approach offers a robust solution for analyzing censored survival data with potential outliers.