Use this URL to cite or link to this record in EThOS:
Title: Machine learning stratification for oncology patient survival
Author: Lloyd, Katherine L.
ISNI:       0000 0004 7425 6068
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Personalised medicine for cancer treatment promises benefits for patient survival and effective use of medical resources. This goal requires the development of predictive models for the identification and implementation of biomarkers for the prediction of patient survival given treatment options. This thesis addresses research questions in this area. The systematic review detailed in Chapter 2 investigates the literature concerning the prediction of resistance to chemotherapy for ovarian cancer patients using statistical methods and gene expression measurements. The range of models used by studies in the systematic review highlights the popularity of traditional models, such as Cox proportional hazards, with few more complex models being utilised. In Chapters 3 and 4, new methods are presented for modelling right-censored survival data. Due to the nature of biomedical data, the methods used need to be flexible and adequately account for high dimensional, noisy data. Gaussian processes fulfil these requirements and were hence used for the development of three Gaussian process models for right-censored survival data. Chapter 3 details these models, and they are applied to synthetic and cancer data in Chapter 4. In all cases the Gaussian processes for survival were found to equal or outperform all comparison models, as measured by concordance index. Given the application to molecular cancer data, it was expected that the data would be high dimensional. Two feature selection methods are investigated in Chapter 5 for use with Gaussian processes to address this. In Chapter 6 a program is developed for the analysis of the data produced by a test for cancer mutations using qPCR. The automated program was designed to isolate the analysis from the user and produce results and reports for clinical use. It is observed that this approach of automated analysis would be suitable for application to any form of clinical test or complex predictive model without the requirement of user guidance.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA76 Electronic computers. Computer science. Computer software ; R Medicine (General)