Penalized Cox Regression Analysis from Parkinson Disease to Freezing of Gait Using Multiple Kernels

Abstract

Objectives: The objective of this study was to investigate complex non-linear associations of multiple genetic and brain imaging factors and the interactions of the two factors with the risk of gait freezing for patients with Parkinson’s disease (PD).Methods: We employed a penalized kernel machine Cox proportional hazard regression model. Multiple kernels were created to account for multiple genetic factors and their interactions with brain imaging factors, and we identified significant elements using the group lasso penalty. We applied the proposed method to Parkinson’s Progression Markers Initiative (PPMI) data.Results: We identified LRRK2 genes and the volume of six regions of interest in brain interacting with COMT, GBA, LRRK2, and SNCA genes that are associated with the occurrence of freezing of gait for PD patients.Conclusions: It was found that there is evidence of gene-brain interactions in freezing of gait. The proposed penalized Cox model using the kernel machine method enables us to identify the nonlinear relationship between genetic and neuroimaging factors and the occurrence of neurodegenerative diseases.