P2391 - Time and Experience Do Not Lead to Improved Adenoma Detection Rate With Artificial Intelligence-Assisted Colonoscopy: An 11-Month Implementation Trial

Tessa Herman, MD¹, Kawthar Mohamed, MD², Daniela Guerrero Vinsard, MD¹, Morgan Freeman, MD², Amy Gravely, MA³, Anders Westanmo, PharmD, MBA³, Lisa Smith, RN³, Mohammad Bilal, MD¹, Brian Hanson, MD^3
1University of Minnesota and Minneapolis VA Health Care System, Minneapolis, MN; ²University of Minnesota, Minneapolis, MN; ³Minneapolis VA Health Care System, Minneapolis, MN

Introduction: Artificial intelligence-assisted colonoscopy (AIAC) with computer-aided detection (CADe) technology is designed to improve colonoscopy quality. While randomized controlled trial data is largely positive with improved adenoma detection rate (ADR) using AIAC, real-world implementation trials demonstrate mixed results, including a six-month pre-post implementation trial we previously reported on using two AIAC technologies which showed lower ADR with AIAC. However, current implementation trials are limited by short study durations and thus the impact of AIAC on ADR over time is unknown.

Methods: We extended the duration of our single-center, quasi-experimental implementation study to 11 months using one AIAC technology with the aim to evaluate if the ADR improves over time with continued AIAC use and experience. We compared the ADR at 6-month intervals after AIAC implementation to each other, as well as to a historical cohort with unassisted colonoscopy. Colonoscopies performed for screening, surveillance, or positive fecal immunochemical test indications were included. Secondary endpoints included sessile serrated lesion (SSL), hyperplastic polyp, and benign, non-adenomatous, non-hyperplastic (BNANH) lesion (e.g. lymphoid aggregate, normal mucosa) detection rates. Statistical analysis was performed using Pearson’s chi-square or ANOVA Type III F-test, as applicable.

Results: We evaluated 295 AIAC colonoscopies in the first 6-month interval (May-October 2023) and 419 AIAC colonoscopies in the second 5-month interval (November 2023-March 2024). Groups were balanced without differences in patient demographics (age, gender, body mass index), endoscopists involved, or polyps per colonoscopy. ADR did not improve between AIAC cohorts (74.2% versus 74.7%, p=0.89). Similarly, we did not see improvement in other polyp detection rates with continued AIAC use (Table 1), although there was a near-significant trend towards decreased resection of BNANH lesions (36.3% versus 29.4%, p=0.05) and rebounding ADR over time (Figure 1).

Discussion: Our real-world implementation study failed to show improvement in ADR with continued use and experience with AIAC at nearly 1 year after its implementation. This is unexpected as we hypothesized that with time and experience ADR would improve. This finding suggests a deficit of the CADe technology in detecting adenomatous lesions in our cohort and that time and experience do not enhance the endoscopist-AI interaction.



Disclosures:


Tessa Herman, MD¹, Kawthar Mohamed, MD², Daniela Guerrero Vinsard, MD¹, Morgan Freeman, MD², Amy Gravely, MA³, Anders Westanmo, PharmD, MBA³, Lisa Smith, RN³, Mohammad Bilal, MD¹, Brian Hanson, MD³. P2391 - Time and Experience Do Not Lead to Improved Adenoma Detection Rate With Artificial Intelligence-Assisted Colonoscopy: An 11-Month Implementation Trial, ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.