Use this URL to cite or link to this record in EThOS:
Title: Optimising the FIT : risk adjusted colorectal cancer screening using routine data
Author: Cooper, Jennifer Anne
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis explores the value of risk-adjusted colorectal cancer screening using the faecal immunochemical test (FIT). Following the FIT pilot study in the English Bowel Cancer Screening Programme (BCSP), there was opportunity to investigate a risk-adjusted approach to screening. This thesis was informed by several evolving areas of research including risk prediction modelling, test accuracy, as well as the use of electronic health records (EHRs) and the statistical methods best applied to utilise these data. The emphasis of the research was on routine data and used the Bowel Cancer Screening System (BCSS) as well as anonymised GP records for model development (THIN - The Health Improvement Network). Three statistical modelling techniques were investigated to build a risk prediction model for use in screening referral based decisions. A conventional approach using logistic regression was investigated first, which showed an improvement in both model performance and test accuracy for the risk-adjusted model over the FIT alone. This model was then extended further by investigating a machine learning algorithm in the form of an artificial neural network. An advantage of this approach is the flexibility to model complex nonlinear associations. The performance of this model (discrimination), as well as the sensitivity when applied as a test, was significantly better than the logistic regression model. Next an anonymised GP record was investigated for additional predictors to add to a risk-adjusted model using survival analysis, which exploits the longitudinal nature of the data. Two models were produced; one which combined the faecal occult blood test (FOBT) with lab test results, symptoms and other predictors and another which was developed for those with negative FOBT results only, to determine whether additional predictors could be used for referral decisions. In order to utilise EHRs for research, the methods need to be reproducible and transparent. An Acceptable Electronic BCSP (AEB) date was developed for quality assurance of the primary care data as well as to help determine a screening cohort for analysis. As a collective, these studies show evidence for improved performance of risk-adjusted screening over using the FIT alone. Future research should focus on further BCSS predictors and external validation of a risk-adjusted model in the BCSP. Machine learning approaches may be better placed for more complex electronic data. Future risk prediction model studies should encompass the whole pathway from model development to external validation and model impact before being implemented in practice with the ultimate aim of improving patient outcomes.
Supervisor: Not available Sponsor: Collaboration for Leadership in Applied Health Research and Care West Midlands
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: RC0254 Neoplasms. Tumors. Oncology (including Cancer)