Use this URL to cite or link to this record in EThOS:
Title: Incorporating time-dimension in ROC curve methodology for event-time outcomes
Author: Kamarudin, A.
ISNI:       0000 0004 7970 4288
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
ROC (receiver operating characteristic) curve analysis is well established for assessing how well a continuous biomarker is capable to distinguish between healthy and diseased (event) individuals. The classical ROC curve approach is based on binary (case/control) disease outcome. However, many disease outcomes are time dependent as disease status is changing over time. Thus, estimating an ROC curve as a function of event-time is more appropriate and is a more effective tool in measuring the diagnostic accuracy of a biomarker. This thesis develops and applies novel time-dependent ROC curve analysis approaches for evaluating the diagnostic efficacy of a biomarker at the baseline level. Two major findings of the comprehensive review undertaken on the time-dependent ROC curve has motivated the methodological developments of this thesis. Firstly, lack of parametric approaches to estimate the biomarker efficacy, and secondly, although biomarkers are often measured with an error due to contamination and variable storage conditions, the current estimation methods ignores this error. The thesis develops a parametric time-dependent ROC curve exploring a range of combination of distributions for event-time and biomarker. The closed form formulae of ROC curve summaries are derived from the joint distribution of event-time and biomarker whenever possible, while numerical solutions are implemented otherwise. A joint modelling approach is proposed to adjust for measurement error of the biomarker. Individual-level deviations of the baseline biomarker measurement from the population mean is linked with the event-time within the proposed joint model. A measurement error adjusted estimator is derived from estimated random effects and association between baseline biomarker and event-time. The proposed methods are evaluated through a range of simulation studies, and illustrated using Mayo Clinic primary biliary cirrhosis (PBC) data. Software is developed in the R language to implement the methodologies. The results show that although the closed form formulae for parametric timedependent ROC curve cannot be established for many distributional combinations due to complexity of the joint density, numerical solutions can be readily derived with the current advances in computing and software. The proposed parametric method provides equally precise estimates even when the sample size is small. The observed biomarker measurement could underestimate the true diagnostic effectiveness due to measurement error and hence useful biomarkers might go unnoticed. The proposed methodology effectively adjusts for measurement error when evaluating the diagnostic effectiveness of a biomarker.
Supervisor: Kolamunnage-Dona, Ruwanthi ; Cox, Trevor Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral