Diagnostic classifiers for stratifying patients at risk of prostate cancer

Abstract

Over 48,500 men in the UK are diagnosed with prostate cancer (PCa) each year. The prediction of prostate cancer in primary care is typically based upon serum total prostate specific antigen (TPSA) and digital rectal examination results. However, these tests lack sensitivity and specificity, leading to over-diagnosis of disease and unnecessary, invasive biopsies. In a recent review, we discussed how hundreds of potential PCa markers have been studied in blood, urine, tissue, and seminal fluid. Yet it was clear from this research that using a multivariable approach to risk modelling offers a clear advantage for risk prediction modelling. The primary objective of this thesis was to investigate various clinical and biomarker risk markers that could be developed into a prediction model that stratified patient risk of PCa within primary care.

Patient samples from two sets of cohorts and their characteristics were curated into one large database. Patient comorbidities and current medications were grouped and also inserted into the database. From there a three-pronged approach was taken; (i) the statistical analysis of clinical markers assigned to patients from questionnaire data as well as patient measurements (ii) the analysis of patient serum using various proteomic methods such as proteome profiler arrays, ELISA analysis and Biochip Array Technology and (iii) the analysis of patient urine using the same techniques. With this data robust statistical analyses were performed using R software in an effort to display an association of markers within PCa and to identify any differentiating markers within the patients. Furthermore, computational modelling using LASSO regression was performed.

Results showed that various markers were identified within this study that demonstrated an association with PCa across all three approaches taken. Moreover, various models were constructed that achieved a respectable level of accuracy to differentiate non-PCa and PCa patients in our cohorts. We also demonstrated that these PCa models could have utility in primary care to improve and effectively stratify patients compared to TPSA alone.

Date of Award	Sept 2021
Original language	English
Sponsors	Randox Laboratories Ltd.
Supervisor	Declan Mc Kenna (Supervisor), Tara Moore (Supervisor) & Mark Ruddock (Supervisor)