Diagnostic application of the ColonFlag AI tool in combination with faecal immunochemical test in patients on an urgent lower gastrointestinal cancer pathway

Abstract

Objective Colorectal cancer (CRC) is the fourth most common cancer in the UK. Patients with symptoms suggestive of CRC should be referred for urgent investigation. However, gastrointestinal symptoms are often non-specific and there is a need for suitable triage tools to enable prioritisation of investigations. In this study, the performance of the faecal immunochemical test (FIT), anaemia and the artificial intelligence algorithm ColonFlag were retrospectively examined and evaluated for their potential clinical benefits in patients who had been referred on an urgent lower gastrointestinal cancer pathway. Design All patients aged over 40 years referred in a 12-month period were included. After 6 months, clinical outcomes were determined and the performance of the triage tests was evaluated. Results A total of 3822 patients completed investigations and received a diagnosis. 143 had CRC, 126 high-risk adenomas (HRA). ColonFlag would have missed 27 CRC and 29 HRA. Faecal haemoglobin (f-Hb) at a cut-off of 10 µg/g would have missed 10 CRC and 26 HRA; f-Hb in combination with anaemia would have missed 2 CRC and 14 HRA. Using f-Hb in combination with ColonFlag would have missed only 1 CRC and 5 HRA and would have reduced the need for urgent referral by over 400 patients. Conclusion ColonFlag has potential to assist detection of CRC and HRA, alone where no faecal sample is present and in combination with FIT and to reduce the need for urgent referral. Data are available on reasonable request.