Variable selection in functional linear Cox model.

Area of Science:

• Biostatistics
• Data Science
• Biomedical Engineering

Background:

• Biomedical studies increasingly use complex, high-dimensional physiological signals from wearables and sensors.
• Time-to-event outcomes are common, necessitating efficient variable selection for survival model accuracy and interpretation.

Purpose of the Study:

• To propose a novel variable selection method for functional linear Cox models.
• To handle both functional and scalar covariates measured at baseline.
• To improve interpretation and accuracy of survival models in wearable-based health studies.

Main Methods:

• A spline-based semiparametric estimation approach for functional coefficients.
• A group minimax concave type penalty for integrating smoothness and sparsity.
• An efficient group descent algorithm for optimization and automated parameter selection.

Main Results:

• Simulation studies confirmed the method's ability for accurate variable selection and estimation.
• Application to National Health and Nutrition Examination Survey data identified key predictors of all-cause mortality.
• Revealed temporally varying distributional patterns of physical activity associated with mortality.

Conclusions:

• The proposed method effectively performs variable selection in complex functional data settings.
• It offers insights into the association between physical activity patterns and all-cause mortality in older adults.
• This approach enhances the analysis of wearable sensor data for health outcome prediction.