Within the rising tide of studies extolling the benefits of artificial intelligence for improving adenoma detection during colonoscopy comes new research suggesting the contrary, at least among people with inflammatory bowel disease (IBD).
Researchers retrospectively studied almost 1,000 colonoscopies before and after introduction of an AI system (GI Genius, Medtronic) at a tertiary medical center in Israel in which a large volume of endoscopies was performed., and it was significantly higher for colonoscopies performed by gastroenterologists who had 5 or more years of experience, compared with the ADR for AI-assisted colonoscopies.
The lower ADR rate in AI-assisted procedures could be the result of an overreliance on the AI technology and shorter procedure times, which may have led to an underrecognition of adenomas, lead investigator Asaf Levartovsky, MD, said in an interview.
“AI is an aid to the endoscopist, not a replacement to the endoscopist,” added Dr. Levartovsky, a gastroenterologist at Sheba Medical Center, Tel Aviv.
The results were presented as a poster at the annual congress of the European Crohn’s and Colitis Organisation, held in Copenhagen and virtually.
The use of AI has recently been shown to improve colorectal cancer screening overall, the authors note. ADR is a measure of the quality of screening colonoscopies. Detection rates were at least 20% among women and 30% among men, “indicative of adequate performance.”
The ADR for people with IBD can be lower than it is for average-risk patients, however, owing to a difference in age in the two populations and the presence of dysplasia-associated lesions, as opposed to sporadic adenomas, for patients with IBD, the researchers note. There is no consensus on an acceptable ADR target for patients with IBD, and the impact of AI-assisted colonoscopy in this patient population hasn’t been explored, they add.
To learn more, Dr. Levartovsky and colleagues compared 237 screening colonoscopies conducted in the 11 months before AI was introduced at the medical center in July 2021 to 759 colonoscopies performed in the 15 months after its introduction.
The pre-AI patient group and the AI patient group were similar (mean age, 44-45 years; about 55% men in each group). Crohn’s disease was more common than ulcerative colitis (63% in the pre-AI cohort and 57% in the AI-assisted cohort).
The ADR in the pre-AI group was 6.3%, compared with 4% in the AI-assisted group (P = .15). The distinction became significant, at 7.6% versus 3.8% (P = .035), when researchers evaluated colonoscopies performed by gastroenterologists who had 5 or more years of experience.
Total procedure time was longer for the patients in the pre-AI group, at 25 minutes, compared with 21 minutes in the AI-assisted group. This difference was statistically significant (P < .0001).
“I think this poster raises questions regarding the real-world utility of AI for adenoma detection [in patients with IBD],” Dr. Levartovsky said.
Dr. Levartovsky said he was not surprised by their findings, because they are similar to those reported in a recent article from his group, although this earlier study did not focus on patients with IBD.
The research had some limitations. The study was not case-control matched, and the pre-AI group was considerably smaller than the AI group.