External validation of the COLOFIT colorectal cancer risk prediction model in the Oxford-FIT dataset: the importance of population characteristics and clinically relevant evaluation metrics.

Abstract

Background: A faecal immunochemical test (FIT) result ≥ 10 µg/g is recommended in the UK to triage patients with symptoms of colorectal cancer (CRC) in primary care for urgent cancer investigation. The COLOFIT model combining FIT results with demographics and blood tests was developed to reduce the proportion of people referred without CRC. This study aims to externally validate the COLOFIT using data from Oxford University Hospitals (OUH).

Methods: FITs requested by GPs between January 2017 and February 2024 were extracted from the OUH Clinical Data warehouse. Adults with COLOFIT predictors and 180-day follow-up for CRC were included. External validation of the COLOFIT equation was conducted overall and for six independent time periods. Risk score thresholds where the model captured the same number of cancers as FIT ≥ 10 µg/g were estimated to understand the number of urgent referrals avoided.

Results: A total of 51,477 individuals (659 CRC) were included; 6194 (12%) had FIT ≥ 10 µg/g. FIT positivity and testing volume increased over time, associated with a gradual change from testing lower-risk patients to including those with higher-risk symptoms. COLOFIT was poorly calibrated overall (observed/expected [O/E] ratio 1.52 with calibration slope 1.05), but calibration improved over time (up to O/E ratio 1.09 with calibration slope 1.05). COLOFIT reduced referrals by 8% overall without missing colorectal cancers compared to FIT ≥ 10 µg/g, but this varied from 23% reduction to 2% increase depending on the period evaluated.

Conclusions: The potential benefit of COLOFIT varied depending on FIT testing rates, the proportion of FIT ≥ 10 µg/g, and the symptoms in the tested population. Adopting COLOFIT into current clinical practice demands, therefore, FIT positivity of at least 17% and CRC rates within 1.3-1.6%. Further validation in local and different populations would also be of significant value and help to maximise COLOFIT's ability to improve diagnostic pathways.