COLOFIT and colorectal cancer

The COLOFIT models included FIT, age, sex, platelet count (PLT) and mean cell volume (MCV) as predictors to estimate risk of colorectal cancer (CRC)(1):

• COLOFIT’s performance over time varied significantly in the Nottingham dataset: COLOFIT yielded a 4.6% reduction in referrals in the model derivation data and a 20.2% reduction in the newer Nottingham internal validation data

FIT results ≥ 10 µg of haemoglobin per gram of faeces commonly trigger referral, with approximately 90% sensitivity for CRC (2):

• however, about 10 in 11 patients with a positive FIT referred for investigation for possible CRC do not have cancer

Tamm et al investigated whether COLOFIT could reduce the number of patients referred for urgent colorectal cancer investigation compared to the standard practice of referring patients with FIT ≥ 10 µg/g, without missing cancers compared to FIT:

• externally validated COLOFIT in a large real-world GP-requested FIT dataset of 51,477 patients, focussing on a clinically relevant metric: the change in the proportion of patients eligible for referral without missing colorectal cancer diagnoses compared to using FIT ≥ 10 µg/g alone
  • found that COLOFIT would have led to an 8% reduction in referrals across the study period and between 23% reduction and 2% increase depending on the time period
    • variation demonstrated the role of increasing testing rates and population characteristics on model performance
• COLOFIT performance varied over time as Oxfordshire rates of testing increased incorporating higher-risk symptomatic patients with rectal bleeding or blood in stool
  • observed that a COLOFIT referral threshold may not be directly transferable if estimated using data from another setting, or in data from an earlier time period from the same setting if the tested population is changing
  • the optimal setting for COLOFIT implementation would be a population similar to the original Nottingham derivation population, a ‘pan-risk’ population including stable rates of both ‘low-risk’ and ‘high-risk’ symptoms of colorectal cancer and using a buffered stool collection kit
  • recommendations to inform the implementation of COLOFIT in new settings:
    1. Validation using local data would be the optimal approach where it is feasible and local data are readily available. Before using COLOFIT to change patient care, the model would be validated in local data to identify the referral threshold that reduces referrals without missing cancers compared to FIT. This could be done retrospectively in systems where FIT has already been used, and the other variables in the model could be retrieved, or prospectively by running the model passively for a period of time while data accrues.
    2. COLOFIT can be implemented before local validation if data are not available to evaluate the model before implementation under the following conditions: FIT positivity is at least 17% and the colorectal cancer rate is near 1.3–1.6%. Without local validation, COLOFIT may miss a small number of cancers and might not reduce referrals relative to FIT.
    3. Monitor COLOFIT performance following implementation. COLOFIT performance may still vary over time with changes in the tested population. The optimal risk score threshold to indicate referral will correspondingly change over time. COLOFIT should be regularly revalidated in local data to understand changes in the tested population and to re-estimate a risk threshold as new data become available. This will reduce the likelihood of missed cancers whilst maintaining maximal reduction in referrals.

Reference:

1. Crooks CJ et al. COLOFIT: Development and Internal-External Validation of Models Using Age, Sex, Faecal Immunochemical and Blood Tests to Optimise Diagnosis of Colorectal Cancer in Symptomatic Patients. Aliment Pharmacol Ther. 2025 Mar;61(5):852-864.
2. Tamm A et al. 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. BMC Med. 2025 Aug 27;23(1):503.