In 2018 colorectal cancer (CRC) was estimated to account for 1,800,977 new cancer diagnoses and 861,663 deaths (1). CRC screening is frequently employed (2) to reduce this burden and has been demonstrated to be effective (3-7). However, the incorporation of risk prediction models into these programs could provide additional benefits, such as the ability to identify individuals at increased risk and thereby tailor their screening accordingly (8). This thesis explores the potential use of CRC risk prediction models in population-based screening and, as an extension, the impact that lifestyle can have on CRC risk.
A systematic review (9) and its extension identified 16 CRC risk prediction models that did not require invasively acquired predictors. Their external validation in two large European cohorts (the European Prospective Investigation into Cancer and Nutrition (EPIC) and UK Biobank), identified several models with good levels of discrimination (C-statistics of up to 0.71) and calibration.
Using two of the better performing models (10, 11) as exemplars, it was then investigated if the addition of a genetic risk score (GRS) for CRC could enhance model performance. Although it did not meaningfully do so at a population-level, it was observed that it may have a role in the further stratification of those at higher risk (absolute risk ≥1%).
To conclude, the impact of lifestyle and genetic risk on the future incidence of CRC was explored. Demonstrating that a less healthy lifestyle is associated with an increased CRC risk (hazard ratio 1.60 [1.32, 1.95] in the UK Biobank and incidence rate ratio 1.93 [1.37, 2.72] in EPIC).
This thesis supports the potential use of CRC risk prediction models within screening programmes. However, further work is required; this includes elucidating the optimal predictors for inclusion and undertaking impact and transportability assessments to further determine model utility.