Elastic functional Cox regression model with shape predictors.

Area of Science:

• Statistics
• Biostatistics
• Functional Data Analysis

Background:

• Phase variability in functional data can obscure important features and reduce the accuracy of predictive models.
• Existing regression models struggle to effectively handle phase variation in functional predictors.

Purpose of the Study:

• To introduce a novel elastic functional Cox regression model (EFCRM) that addresses phase variability in functional predictors.
• To capture the association between time-to-event outcomes and the shape of functional predictors, even with phase shifts.

Main Methods:

• Developed an EFCRM incorporating supervised registration to align functional predictors based on survival outcomes.
• Employed an iterative algorithm alternating between phase variation removal and regression coefficient estimation.
• Utilized gradient-based updates to enhance convergence of the estimation procedure.

Main Results:

• The EFCRM demonstrated robust estimation accuracy and predictive performance in simulation studies.
• The model showed comparable or superior performance against existing methods.
• EFCRM proved robust to the choice of basis functions for coefficient approximation.

Conclusions:

• The proposed EFCRM effectively handles phase variation in functional data for time-to-event analysis.
• EFCRM offers improved accuracy and predictive power compared to current approaches.
• The model is a valuable tool for analyzing complex functional predictor data in survival analysis.