Nguyen D, Chang P, Kong N, Wang S, Sharma N, Ong J, Wang D, Amini M, Bui A, Bakr O, Bruce D, Dodge J, Sahakian A, Buxbaum J. S296 AI-Assisted Colonoscopy May Increase ADR in Right Side of Colon in Gastroenterology Fellows: A Randomized Control Trial. American Journal of Gastroenterology. 2024 Oct;119(10S):S210-S210. doi: 10.14309/01.ajg.0001030552.21890.27


Abstract

Introduction:

The American College of Gastroenterology has highlighted a need to train endoscopists to detect lesions in the right colon. The role of Artificial intelligence (AI) assisted colonoscopy in the training setting has not been thoroughly defined. This study explores the impact on adenoma detection rate (ADR) overall and by location for AI assisted colonoscopy compared to conventional colonoscopy (CC) (without AI) for gastroenterology (GI) fellows.

Methods:

In a randomized control trial (NCT05423964), we consented GI fellows (postgraduate year 4-6) to AI (GI Genius, Medtronic; Minneapolis, MN) vs CC rooms daily and compared their ADR, polyp detection rate (PDR), ADR in the left (LADR) and right colon (RADR). We included all colonoscopies by trainees (postgraduate year 4-6) and excluded those with inadequate bowel preparation or primary performance by an attending physician. Baseline demographics, procedural data, polyp pathology, location, and size were collected. Mean difference of ADR was compared for AI vs CC for overall and screening colonoscopies controlled by fellow. Sub-analysis was performed by year.

Results:

A total of 1,045 overall and 130 screening colonoscopies were performed by 16 fellows (postgraduate year 4-6) between March 15th to October 31st, 2023. Patient characteristics were similar between arms. In all colonoscopies, ADR was similar for AI vs CC (AI: 40.5+/-3.9% vs CC: 35.0+/-3.6%, mean difference: 5.5(CI95%:-4.3-15.3%)), but there was a higher PDR in the AI arm (AI: 63.3+/-3.4%, vs 54.1+/-2.6%, mean difference: 9.3%(1.1-17.3%)). For ADR by anatomical location, there was a 21% higher RADR (AI: 35.1+/-7.1% vs 13.7+/-5.6%, mean difference 21% (CI95%: 7.6-48.0%), P-value = 0.02). On sub-analysis for RADR, there was a 27.8% higher RADR in first years (AI: 33.3+/-9.5% vs 5.5+/-3.2%, mean difference 27.8% (CI95%: 7.2-48.0%), P-value = 0.008), 22.6% higher RADR in second years (AI: 33.3+/-9.4% vs 10.7+/-5.7%, mean difference 22.6% (CI95%: 5-40.2%), P-value = 0.01) and no difference in third years. LADR was similar between groups. There was no difference in procedural or withdrawal time with the addition of AI.

Conclusion:

This randomized control trial shows that AI assisted colonoscopies improve RADR for gastroenterology trainees. We propose a use case via AI assisted colonoscopy for trainees guiding improvement of adenoma detection in the right colon to standardize a critically needed colonoscopy quality metric and potentially to facilitate training.