Safety of artificial intelligence-assisted optical diagnosis for leaving colorectal polyps in situ during colonoscopy (PRACTICE): a non-inferiority, randomised controlled trial

Abstract

Background: Guidelines recommend leaving in situ rectosigmoid polyps diagnosed during colonoscopy that are 5 mm or smaller if the endoscopist optically predicts them to be non-neoplastic. However, no randomised controlled trial has been done to examine the efficacy and safety of this strategy.

Methods: This open-label, multicentre, non-inferiority, randomised controlled trial enrolled adults age 18 years or older undergoing colonoscopy for screening, surveillance, or clinical indications across four Italian centres. Eligible patients were randomised 1:1 (with stratification by patient sex, age, and previous adenoma removal) via a central web-based system, to either the leave-in-situ group, in which endoscopists could leave non-neoplastic lesions in place after optical diagnosis, or the resect-all group, in which all detected polyps were systematically removed, regardless of optical diagnosis. Patients and endoscopists were not masked to group allocation but pathologists and investigators assessing outcomes were masked. All procedures in both groups were done with the assistance of a computer-aided detection and diagnosis system. Endoscopists optically diagnosed lesions through a combination of white light, blue light, and computer-aided detection. The primary outcome was the adenoma detection rate (ADR), defined as the proportion of participants with at least one adenoma detected (per-patient analysis), assessed by intention-to-treat, to determine whether the leave-in-situ strategy was non-inferior to the resect-all approach, with an absolute 10% non-inferiority margin. This trial was registered with ClinicalTrials.gov (NCT05500248) and is completed.

Findings: Between Oct 1, 2022, and April 30, 2024, 1147 patients were recruited and 895 patients (507 [57%] females, 388 [43%] males, mean age 61·1 years [SD 9·8]) were randomly assigned to either the leave-in-situ group (n=441) or resect-all group (n=454). 197 adenomas or colorectal cancers were detected in the leave-in-situ group and 211 in the resect-all group; the ADR was 44·7% (95% CI 40·4 to 49·5) in the leave-in-situ group and 46·5% (41·8 to 51·2) in the resect-all group (absolute difference -1·8 percentage points, 95% CI -8·9 to 4·9; p_{non-inferiority}=0·013). No colonoscopy-related complications, including perforation and bleeding, were reported in either group.

Interpretation: The leave-in-situ strategy through optical diagnosis with computer-assisted diagnosis support does not reduce oncological safety of colonoscopy, as measured by ADR. By reducing unnecessary polypectomies, this strategy could be considered as an attractive option in colonoscopy practice.

Funding: European Society of Gastrointestinal Endoscopy.