Restriction lift date: 2026-09-30
Transcriptomics-based prediction of recurrence-free survival in prostate cancer following prostatectomy
Loading...
Files
Date
2024
Authors
O'Donnell, Autumn
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
This thesis investigates the application of machine learning (ML) methods for survival analysis, to the prediction of biochemical recurrence (BCR) in prostate cancer patients following radical prostatectomy, using mRNA expression data. While conventional survival analysis methods remain commonplace in clinical applications, they are unfeasible with high-dimensional data such as mRNA assays and rely on assumptions that may not fully capture the complexity of cancer recurrence. This research explores the potential of ML techniques to overcome these limitations, offering more accurate and personalised predictions for prostate cancer outcomes. A systematic review of the current literature on the use of ML for survival analysis in cancer was carried out to identify key methodologies and their performance compared to traditional statistical approaches. This review highlights a need for improved reporting and standardisation in the field. The findings informed the selection of ML models for further study, including random survival forests (RSF), LASSO Cox, and boosted Cox models.
This research’s first experimental phase focuses on post-operative BCR prediction, where mRNA data from tumour tissue was integrated with clinical variables with the aim of improving performance. Results demonstrated that ML models incorporating mRNA expression outperformed those with clinical variables alone and could also improve on traditional Cox proportional hazards (CPH) models. The second phase extends the analysis to pre-operative settings, exploring the potential feasibility of using mRNA data to predict BCR before surgery. This pre-operative model also showed significant promise, suggesting the potential for early intervention strategies that could better inform treatment decisions.
The study contributes to the growing field of precision oncology, offering insights into how genetic information and ML can be leveraged to improve clinical decision-making in prostate cancer.
Description
Keywords
Cancer , Prostate cancer , Statistics , Survival analysis , Machine learning , Genomics , mRNA , Transcriptomics
Citation
O'Donnell, A. 2024. Transcriptomics-based prediction of recurrence-free survival in prostate cancer following prostatectomy. PhD Thesis, University College Cork.