Development and assessment of a genetic and environmental risk model for colorectal cancer

Thesis

Colorectal cancer (CRC) is the third most common cancer globally, and is readily prevented by screening. Colonoscopy is the gold standard screening tool but is invasive, expensive, and time consuming. ‘Risk-stratified’ approaches to screening, using genetic or non-genetic risk predictors to guide screening decisions, may improve screening outcomes and utilise limited resources more effectively.

The aim of this thesis was to develop and evaluate risk prediction models for CRC, to improve our ability to predict CRC risk. An initial search for novel genetic risk loci began with a new genome-wide association study (GWAS), following which meta-analysis of 5 new GWAS with 10 existing GWAS identified 31 new risk loci for CRC. Genome-wide linkage analysis combined with whole genome sequencing in two early-onset CRC families, and three families with multiple adenomas, identified several new possible risk loci for CRC including a potentially pathogenic variant in GFI-1.

Six polygenic risk scores (PRS) for CRC were then developed and tested in UK Biobank, with SNPs and weights derived from the GWAS meta-analysis. Evaluation of PRS performance showed that a genome-wide approach using LDpred2 software performed best, with a top age- and sex-adjusted C-statistic of 0.733 (95% confidence interval 0.710-0.753) in logistic regression. The PRS performed less well in women compared to men, and in minority ethnic UK Biobank participants.

Integrated genetic and non-genetic models were then developed, combining the top-performing PRS with the non-genetic QCancer-10 (Colorectal Cancer) model, which is based on primary care data. QCancer-10+PRS models modestly improved performance compared to QCancer-10 alone, with C-statistics of 0.730 (0.720-0.741) compared to 0.693 (0.682-0.704) respectively in men, and similar improvements in women. Decision curve analysis indicated small incremental improvements in net benefit, and the absolute difference in predicted 5-year risk was small (0.15-0.3%).

This thesis improves our understanding of the genetic basis of CRC risk, and risk prediction for CRC. Given the modest improvements in risk prediction with the addition of PRS to the QCancer-10 model, and the current logistical and ethical implications of PRS testing, there is no clear justification for PRS implementation in bowel cancer screening in their current form.

Authors: Briggs, S

• DOI: 10.5287/ora-aomnxbmo4
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: risk prediction; colorectal cancer; polygenic risk score
• Subjects: Risk assessment; Cancer