Real-time artificial intelligence-assisted colonoscopy and the effect of endoscopist experience on polyp detection rates at a tertiary referral center

Background: In recent years, new technology has emerged with the potential to considerably influence the future landscape of gastrointestinal endoscopy. One example of this is real-time artificial intelligence-assisted colonoscopy (AIC). Reported improvements in polyp detection rate (PDR) and adenoma detection rate (ADR) over standard colonoscopy (SC) have led to its rapid adoption in several tertiary endoscopy centers worldwide. However, many clinicians have mixed attitudes concerning the use of AIC. Similar to other skill-based procedures, the quality of endoscopy is directly proportional to the technical expertise and proficiency of the endoscopist. Objective: To evaluate the use of real-time AIC versus SC at a tertiary, high-volume referral center. Design: Retrospective cohort study. Settings: Single tertiary referral center. Patient and Methods: Data was retrospectively analyzed from patients who underwent elective colonoscopy at Singapore General Hospital from August 1, 2022 to October 31, 2022. Main Outcome Measures: The primary study outcome was PDR and ADR. Subgroup analysis assessed the differences in polyp detection according to the experience level of the endoscopist. Sample Size: Eight hundred and fifty-nine patients. Results: Over the 3-month study period, 859 patients underwent complete colonoscopies performed by nine accredited endoscopists; 430 patients underwent SC and 429 underwent AIC. Both groups were statistically similar in age, gender, and indication for colonoscopy. The median ADR was 34.2% (range, 24.6%–57.3%) for the entire cohort. Overall detection rates were higher for AIC than SC for diminutive polyps (≤5 mm) (45.7% vs 38.6%, P = 0.045), sessile polyps (48.5% vs 37.4%, P = 0.009), and polyps with adenomatous (42.4% vs. 36.3%, P = 0.043) or serrated histology (1.2% vs. 0%, P = 0.025). Using SC, PDR among junior versus senior endoscopists was comparable at 47.9% vs 45.6% (P = 0.672) and ADR at 36.1% vs 36.8% (P = 0.912). With real-time AI enhancement, PDR and ADR for junior endoscopists considerably increased over their senior counterparts, with PDR at 69.5% vs 44.9% (P = 0.0001) and ADR at 50.0% vs 37.7% (P = 0.016). Among senior endoscopists, AIC did not result in any statistical improvement of PDR (P = 0.999) and ADR (P = 0.854) compared to SC. Conclusions: AIC significantly increases PDR and ADR compared to SC, particularly for detecting diminutive and sessile polyps. This benefit was only observed among less experienced endoscopists. Limitations: This study is limited by its retrospective nature, short study duration, and availability of the AI system, leading to practical constraints. Conflict of Interest: The authors have no conflict of interest to declare.