Modelling survival times using frailty models

Abstract

This dissertation explores the application of univariate and shared frailty models to patients who underwent aortic valve replacement surgery at Mater Dei hospital. Traditional survival models, including Kaplan-Meier, Nelson-Aalen and Cox regression are inappropriate in the presence of heterogeneity since they assume a homogeneous population. To overcome this limitation, frailty models were introduced. The same principles from survival theory apply, however the distribution incorporates a multiplicative term to cater for unobserved random effects. Frailty models are implemented to relate survival times for censored data to pre-operative, operative and post-operative patient related variables to identify the risks factors. The study is focused on conducting parametric and semi-parametric investigations to determine the predictors that yield the highest degree of heterogeneity.