Joint modelling of longitudinal and time-to-event data

Powney, Matthew
(2015) Joint modelling of longitudinal and time-to-event data. PhD thesis, University of Liverpool.

[thumbnail of PowneyMat_Oct2015_2053320 edited.pdf] Text
PowneyMat_Oct2015_2053320 edited.pdf - Unspecified

Download (16MB) | Preview


A randomised control trial (RCT) is considered to be the gold standard for investigating the efficacy of new and novel treatments. However, in RCTs with longitudinal outcomes and high percentages of dropout, a poor handing of missing data can be problematic when trying to establishing the efficacy of an intervention. Joint modelling of longitudinal and time-to-event data is a novel methodology that can be used to monitor a longitudinal outcome while simultaneously accounting for time-to-dropout. This is achieved using a mean zero latent Gaussian process, and relies on the estimation of the parameter gamma, which models the association between the longitudinal and time-to-event components. However, joint modelling is still a relatively new topic for research. The aim of this thesis is to provide and develop a greater understanding for both the design and analysis elements of joint modelling. In Chapter 2 a simulation study to test the success of various missing data handling methods is presented. This demonstrated that for RCTs with missing data, joint modelling performs as well as the common alternative methods when estimating longitudinal treatment effect. Despite these benefits, a systematic review conducted in Chapter 3 showed that Joint Modelling is rarely used in practice in RCTs with longitudinal outcome data. One contributing factor to the underuse of joint models may be the lack of understanding and research into sample size calculations for a trial using joint modelling. In Chapter 4, sample size formulae are derived for beta_2 and gamma in the Henderson et al. (2000) random slope and intercept specification of the joint model. These sample size and power calculations depend on knowledge about the value of gamma in a trial. Currently, the understanding of the interpretation of gamma is limited, and no previous investigations into the relationship between magnitude of gamma and change in longitudinal outcome for dropouts has been carried in published literature. In Chapter 5, a visualisation of this relationship is presented. In Chapter 6, software is developed in R to carry out and apply some diagnostic procedures for joint models. Chapter 7 demonstrates the applicability of the methods described in this thesis, in which joint modelling is utilised to analyse a wide range of datasets.

Item Type: Thesis (PhD)
Additional Information: Date: 2015-10-23 (completed)
Divisions: Faculty of Health and Life Sciences > Institute of Life Courses and Medical Sciences > School of Medicine
Depositing User: Symplectic Admin
Date Deposited: 14 Jul 2020 09:45
Last Modified: 18 Jan 2023 23:46
DOI: 10.17638/03093967