AI Colonoscopies: More Benign Lesion Removals, Potentially Higher Costs

Use of artificial intelligence (AI)-assisted colonoscopy led to a greater removal rate of exclusively benign lesions compared with colonoscopies that did not use AI, according to data analyzed from a previous single-center prospective study.

About a third more polypectomies were performed only for benign lesions in the AI-assisted colonoscopy group than in an unassisted historical cohort (12.4% vs 8.4%, P=0.04), reported Tessa Herman, MD, of the University of Minnesota and Minneapolis VA Health Care System, at the annual meeting of the American College of Gastroenterology in Philadelphia.

Although AI-assisted colonoscopy with computer-aided detection is intended to improve polyp detection, some previous low-quality evidence has suggested that using AI increases the resection of non-neoplastic lesions.

With an estimated 15 million colonoscopies done in the U.S. every year, Herman said, extrapolating the findings of this study suggests that approximately 600,000 colonoscopies result in the removal of only benign non-adenomatous lesions. The potential implications are therefore increased risks and costs without additional benefit of colorectal cancer prevention, she said.

“AI in colonoscopy is a double-edged sword,” Brennan Spiegel, MD, MSHS, the George and Dorothy Gourrich Chair in Digital Health Ethics and director of Health Services Research at Cedars-Sinai in Los Angeles, told MedPage Today. “The technology improves our ability to spot polyps, but it also leads to more procedures that may not contribute to cancer prevention, which is an important consideration for both patient safety and healthcare costs.”

Spiegel, who was not involved in the study, said that AI-assisted colonoscopy “shows promise in improving adenoma detection,” but these findings provide “a reminder that more is not always better,” he said. “It’s critical to refine these tools to ensure that we are enhancing patient care without introducing unnecessary risks.”

Herman and her fellow researchers analyzed data from a previous prospective study at the Minneapolis VA Medical Center that evaluated the impact of AI-assisted colonoscopy on adenoma detection rate compared to a historical cohort. The study included 599 adults undergoing an AI-assisted colonoscopy from May to October 2023. Only colonoscopies for screening or surveillance — not diagnostic indications — were included. Patients with inflammatory bowel disease, an incomplete colonoscopy, or inadequate bowel preparation were also excluded.

Those AI-assisted colonoscopies were compared with 441 colonoscopies without AI assistance done from November 2022 to April 2023. There were no statistically significant differences between the two cohorts’ age, biological sex, body mass index (BMI), or race. Nearly all participants in both cohorts were male (95-96%), white (91-92%), and non-Hispanic (99%), with an average age of 66-67 years and an average BMI of 31.

Average bowel preparation scores were similar in the AI group (8.3) and non-AI-assisted group (7.94), and 4.2% of participants in the AI group had a family history of colon cancer compared with 3.9% of those in the unassisted group.

The most common benign lesions removed during polypectomies were benign colonic mucosa, lymphoid aggregates, and hyperplastic polyps. Twice as many benign colonic mucosa (12) and lymphoid aggregates (20) were removed in the AI-assisted group as in the unassisted group (6 and 10, respectively). A total of 28 hyperplastic lesions (with or without another benign lesion) were removed in the AI-assisted group compared to 17 in the unassisted group.

The benefits of AI-assisted colonoscopy are greatest when AI’s use leads exclusively to adenoma removal, or, to a lesser extent, the removal of adenomas and benign lesions. But costs outweigh the benefits when AI-assisted colonoscopy leads only to benign lesion removal.

Since this was a single-center study lasting only 6 months, its generalizability is limited, necessitating high-quality, long-term studies that monitor both adenoma detection rates and benign lesion resection rates with AI-assisted colonoscopies, Herman noted. But the findings are strengthened by the homogeneity of practices and participation of the same eight endoscopists.

“AI’s role in colonoscopy is evolving rapidly,” Spiegel said. “The next frontier is to refine these systems to not only detect more but also to discern better, minimizing unnecessary polypectomies.”