Abstract
Smooth survival models using penalized regression splines are one approach for modeling of non-proportional hazards, adjusting for potential confounders, and estimating functions of survival time. We present a broad class of link-based survival models, which can integrate and extend some existing survival models (e.g. the proportional hazards, additive hazards, proportional odds and probit models, together with the class of Royston-Parmar models) via link functions. We construct a penalized likelihood to estimate the parameters and develop related numerical algorithms to implement these models. Model performance is assessed using the symmetrised Kullback-Leibler distance. The simulation study demonstrate our proposed model and estimation procedure performs well compared with existing penalized survival models on the log-hazard scale. A real dataset is analyzed to demonstrate some features of our proposed models in multivariable survival data analysis.