ROC-guided survival trees and ensembles.

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Tree-based methods are widely used for time-to-event analysis.
• Existing methods have limitations in handling dynamic survivor populations and time-dependent covariates.

Purpose of the Study:

• To introduce a novel framework for survival trees and ensembles.
• To improve prediction accuracy for time-to-event data.
• To develop methods for handling time-dependent covariates.

Main Methods:

• Developed a novel framework for survival trees and ensembles.
• Introduced generalized time-dependent receiver operating characteristic (ROC) curves for performance evaluation.
• Proposed a new ensemble procedure based on averaging martingale estimating equations.

Main Results:

• The proposed tree-building algorithm targets prediction accuracy using ROC-based criteria.
• The novel ensemble method addresses the instability of single survival trees.
• Simulation studies demonstrated the effectiveness of the proposed methods.

Conclusions:

• The novel framework offers improved performance for survival tree analysis.
• The methods are applicable to complex time-to-event data with time-dependent covariates.
• The approach provides a robust alternative for survival data modeling.