Artificial intelligence-assisted colonoscopy in real-world clinical practice: A systematic review and meta-analysis

Abstract

Background: Artificial intelligence (AI) could minimize the operator-dependent variation in colonoscopy quality. Computer-aided detection (CADe) has improved adenoma detection rate (ADR) and adenomas per colonoscopy (APC) in randomized controlled trials (RCTs). There is a need to assess the impact of CADe in real-world settings. Methods: We searched MEDLINE, EMBASE, and Web of Science for non-randomized, real-world studies of CADe in colonoscopy. Random-effects meta-analyses were performed to examine the effect of CADe on ADR and APC. The study is registered under PROSPERO (CRD42023424037). There was no funding for this study. Results: Twelve of 1,314 studies met inclusion criteria. Overall, ADR was statistically significantly higher with vs. without CADe (36.3% vs. 35.8%, risk ratio [RR] 1.13, 95% CI 1.01-1.28). This difference remained significant in subgroup analyses evaluating 6 prospective (37.3% vs 35.2%, RR 1.15, 95% CI 1.01-1.32) but not 6 retrospective (35.7% vs 36.2%, RR 1.12, 95% CI 0.92-1.36) studies. Among 6 studies with APC data, APC rate ratio with vs. without CADe was 1.12 (95% CI 0.95-1.33). In 4 studies with GI Genius (Medtronic), there was no difference in ADR with vs. without CADe (RR 0.96, 95% CI 0.85-1.07). Conclusions: ADR, but not APC, was slightly higher with vs. without CADe among all available real-world studies. This difference was attributed to the results of prospective but not retrospective studies. The discrepancies between these findings vs. those of RCTs call for future research on the true impact of current AI technology on colonoscopy quality, and the subtleties of human-AI interactions.