Spatial Proportional Hazards Model with Differential Regularization

Abstract: This paper presents a semiparametric proportional hazards model designed to handle spatially varying covariate functions, applicable to both geostatistical and areal data observed on irregular spatial domains. The model is estimated through the maximization of a penalized partial likelihood, with a roughness penalty term based on a differential of the spatial field over the target domain. The finite element method is employed for efficient estimation, enabling a piecewise polynomial surface representation of the spatial field. We apply this method to analyze response time data from the London Fire Brigade.