Joint Modelling for Longitudinal and Time-to-Event Data: Application to Liver Transplantation Data

Abstract

The joint modelling approaches are often used when an association exists between time-to-event and longitudinal processes. They are recognized for their efficiency involving the association structure between these two processes. Recently, [17] and [14] suggested alternative joint modelling approaches. In this paper, we will focus our attention on the Rizopoulos’ approach. This methodology was applied to Orthotopic Liver Transplantation data (OLT) with a flexible environment for both longitudinal and survival sub-models. Different regression models were fitted to the OLT data and their predictive performances were compared by using time-dependent ROC curves, also, dynamic predictions were obtained for the survival process. Computational aspects (including software) related to the use of the joint modelling approach in practice, were also discussed. The application of joint modelling revealed a hitherto unreported effect: for non-diabetic patients, the longitudinal Glucose levels have a significant effect on survival. In addition the discrimination ability improves over time. However for diabetic patients the association between these two processes is not significant.