Use this URL to cite or link to this record in EThOS:
Title: An investigation of methods for predicting rheumatoid arthritis patient response in clinical trials from clinical biomarkers and patient characteristics
Author: Mahoney, Paul M.
ISNI:       0000 0004 7233 7905
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Introduction: Rheumatoid Arthritis (RA) is a chronic, destructive, autoimmune disorder of unknown cause and with no known cure. RA is a multi-factorial disease with many factors influencing onset, severity and outcome. Research Question: Of interest is to understand how selection of patients might influence patient outcome in clinical trials. Specifically, which are the patient characteristics or biomarkers measurable at enrolment that would be important in predicting response? In this PhD we will attempt to answer the question where response is the dichotomous outcome ACR20. Methods: The research for this dissertation accessed clinical trial data from over 11,700 patients enrolled into 16 late stage RA clinical trials from 1998 to 2008. Through systematic review and a review and selection process, logistic regression and CART were selected to compare head to head in simulations. In simulations each method was compared using Variable Selection (VaSe) plots developed in the PhD. In most scenarios CART outperformed logistic regression, and was selected to apply to the clinical database. Results: The CART analysis, generated a model that had an overall predictive accuracy for ACR20 of just 60%, although the explanatory variables selected in the model were plausible for predicting patient outcome: baseline tender joint count, region, joint space narrowing score, number of previous treatments and race. When one considered the 8 individual components of the ACR20, the predictive accuracy increased from 60% to between 67% - 82%. Conclusions: Through the use of a large RA dataset, we were unable to predict ACR20 response to a satisfactory level. However, we were able to predict each component of the ACR20 well. This is likely due to the complexity of the ACR20 as a tool for patient response as well as the heterogeneity of this multifactorial condition. From a patient perspective it may be of more value to be able to predict a disease symptom such as number of swollen joints rather than a composite score. This dissertation provides a framework for investigating predictive markers for patient response in clinical trial in RA and other disease areas.
Supervisor: Julious, Steven A. ; Campbell, Mike J. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available