High predictive performance models via semi-parametric survival regression

This project will develop novel estimation methods for high prediction performance in survival analysis. The overall aim of this project is to build high predictive performance survival regression models that can accommodate: 1) time-varying predictors with the possibility of measurement errors; 2) random effects; 3) model selection when the number of predictors is large; and 4) partly interval-censored event times with the possibility of truncation. The specific aim is then to study how to make inferences on parameters (thus a statistical computation and methodology problem) for a range of models including (i) the Cox model; (ii) the additive hazard (AH); and the (iii) accelerate-failure-time (AFT). Applications to melanoma data are the driving force behind these methodological developments