Methodological Development and Advances for Joint Modelling of Longitudinal and Time-To-Event Data

Abstract

Univariate joint modelling of longitudinal and time-to-event data is a simultaneous analysis of repeated measurements taken from the same individual over time, until an event of interest occurs. This method has attracted increasing interest in the literature over the last two decades. In practice, clinical studies are increasingly likely to record more complex data structures (such as multilevel longitudinal data or multiple longitudinal profiles, along with event time data) than single longitudinal and event time data. This thesis develops a methodology and software for both multilevel and multivariate joint models accounting for complex longitudinal data, by focusing on random effects selection models, where information from the longitudinal trajectories is used to inform the event-time process. The research also assesses the power of the score test, which is a prognostic tool to investigate the association between submodels, before fitting potentially complex and computationally intensive joint models under a variety of scenarios. The methodology is tested via simulation studies, and implemented in various real datasets. The results show that the advanced joint models can provide unbiased estimators when the model is specified correctly such that it utilizes all available data, and that the score test is a powerful tool when the longitudinal profile is highly associated with the event time data. Based on preliminary findings using discrimination measures, the advanced joint models should be preferred in case of complex longitudinal data in order to improve the predictive capability of the model.

Date of Award	1 Jul 2018
Original language	English
Awarding Institution	Northumbria University
Supervisor	Nan Lin (Supervisor) & Pete Philipson (Supervisor)