A comprehensive RCT in screening, surveillance, and diagnostic AI-assisted colonoscopies (ACCENDO-Colo study)

Background and aims: Adenoma detection rate (ADR) serves as a primary quality metric in colonoscopy. Various computer-aided detection (CADe) tools have emerged, yielding diverse impacts on ADR across different demographic cohorts. This study aims to evaluate a new CADe system in patients undergoing colonoscopy.

Methods: This is an Italian multicenter randomized control trial (RCT) that included patients aged 40-85 scheduled for screening, surveillance or diagnostic colonoscopy randomly assigned to CADe or standard colonoscopy (SC). Patients with a Boston Bowel Preparation Scale < 2 in any segment were excluded. The primary outcome was ADR in both groups. Secondary outcomes included adenoma per colonoscopy (APC), polyp per colonoscopy (PPC) and sessile serrated lesion detection rate (SSLDR).

Results: 1228 patients were enrolled of whom 70 were excluded for inadequate bowel cleansing or missed cecal intubation. Therefore, 1158 subjects (578 CADe vs 580 SC) were included in the final analysis. ADR was significantly higher in CADe than in the control group (50.2 % vs 40.5 %, p = 0.001). CADe also significantly increased PPC and APC (1.64 ± 2.03 vs 1.23 ± 1.72, p < 0.001; 1.16 ± 1.82 vs 0.80 ± 1.46 p < 0.001; respectively). No significant differences were found in SSLDR between CADe and SC (12.1 % vs 11.0 %, p = 0.631).

Conclusions: The results of this RCT indicate that AI-assisted colonoscopy significantly improved ADR in a non-selected population undergoing colonoscopy without causing any significant delay in procedure time or increasing the detection of nonneoplastic lesions. (Ethical committee approval: NCT05862948).