A novel variable selection method for a functional linear Cox model using spline-based semiparametric estimation and group minimax concave penalty effectively identifies key temporally varying distributional patterns of physical activity related to all-cause mortality in older US adults.
Key Findings
Methods
The proposed method uses a spline-based semiparametric estimation approach combined with a group minimax concave type penalty to perform variable selection in functional linear Cox models with multiple functional and scalar covariates.
The approach integrates smoothness and sparsity into the estimation of functional coefficients simultaneously.
An efficient group descent algorithm is used for optimization.
An automated procedure is provided to select optimal values of both smoothing and sparsity parameters.
The method accommodates multiple functional and scalar covariates measured at baseline alongside time-to-event outcomes.
Results
Simulation studies demonstrated the method's ability to perform accurate variable selection and estimation in functional linear Cox models.
The simulation studies evaluated the method's performance in correctly identifying relevant functional and scalar predictors.
The method was shown to effectively balance smoothness and sparsity in estimating functional coefficients.
Results indicated accurate variable selection under the simulation settings considered.
Results
Application to the 2003-2006 NHANES cohort identified key temporally varying distributional patterns of physical activity and demographic predictors associated with all-cause mortality in older US adults.
The dataset used was the 2003-2006 cohort of the National Health and Nutrition Examination Survey (NHANES).
The analysis focused on older US adults with physical activity measured via wearable devices and time-to-event outcome of all-cause mortality.
The method identified specific daily distributional patterns of physical activity as significant predictors of mortality.
Demographic scalar covariates were also identified as relevant predictors through the variable selection procedure.
Discussion
The analysis shed light on the intricate association between daily distributional patterns of physical activity and all-cause mortality among older US adults.
Physically measured activity signals were treated as functional covariates, capturing temporally varying patterns across the day.
The functional coefficient estimates revealed which time periods of physical activity were most strongly associated with mortality risk.
The study highlights the utility of functional data analysis methods for interpreting wearable sensor data in survival analysis contexts.
Background
Modern biomedical studies collecting high-dimensional physiological signals from wearables necessitate efficient variable selection methods for survival models, motivating the development of this functional linear Cox model approach.