Artificial intelligence–assisted colonoscopy was associated with higher adenoma detection and nearly half the rate of interval colorectal cancer compared with earlier practice, according to a large US analysis of electronic health records.
“Most of the evidence behind AI-assisted colonoscopy comes from randomized trials run at academic centers under fairly controlled conditions,” Muhammad Ali Butt, MD, a gastroenterology fellow at the University of Minnesota, Minneapolis, told GI & Hepatology News ahead of Digestive Disease Week® (DDW) 2026. “What we didn't have is a clear picture of what happens when these tools get used across dozens of health systems in routine practice — “different patient populations, different procedure volumes, different levels of endoscopist experience.”
For the retrospective cohort study, Dr. Butt and colleagues drew from the TriNetX network to compare outcomes among patients aged 45 years and older undergoing colonoscopy in two time periods: 2015–2019 and 2022–2025. The earlier period reflected practice before AI was widely used, while the later period represented the era of adoption. Patients with colorectal cancer diagnosed before or within 30 days of colonoscopy were excluded. After using 1:1 propensity score matching to balance demographics, comorbidities, and procedure indications, the researchers included more than 1.5 million patients in each cohort for short-term outcomes and nearly 1.6 million per group for longer-term cancer outcomes.
The matched populations were similar across key characteristics, with a mean age of about 60 years and slightly more than half female in both groups. The large sample and matching method were meant to limit bias and better reflect real-world care across multiple US health systems.
Within 30 days of colonoscopy, adenoma detection was higher in the AI era, identified in 3.6% of patients compared with 1.8% in the earlier cohort, representing nearly twice the detection rate. Advanced adenoma detection also increased, occurring in 0.19% vs 0.13% of patients. All differences were statistically significant (P <0.001 for all comparisons).
Longer-term outcomes showed a reduction in interval colorectal cancer, defined as cancer diagnosed 6–36 months after colonoscopy. During the AI era, interval cancer occurred in 0.11% of patients compared with 0.21% in the earlier period, a 47% relative reduction (P <0.001).
“The interval CRC reduction was more pronounced than I expected; interval colorectal cancer is a hard outcome to move,” Dr. Butt said. “These are cancers that slip through after a ‘negative’ scope. A signal that large in real-world data was not something I was anticipating. That said, this is an observational study covering a period when a lot was changing in GI practice simultaneously, [including] quality improvement initiatives, increased awareness of ADR benchmarks, and changes in surveillance guidelines, so I wouldn't attribute all of that to AI alone. What the data show is a temporal association, and not causation.”
Still, he concluded, the scale of the analysis and consistency of the findings across endpoints provide support for the clinical impact of AI–enhanced colonoscopy in routine care. “I think it gives gastroenterologists and hospital administrators something more concrete to point to when weighing adoption decisions,” Dr. Butt said. “Whether it changes practice overnight is another question; cost, workflow integration, and training all matter. But for clinicians who were waiting to see whether real-world results would hold up compared to controlled trials, this is at least a data point in that direction.”
Dr. Butt reported having no disclosures.
DDW is AGA’s annual meeting, jointly sponsored by AGA, AASLD, ASGE, and SSAT. Learn more at ddw.org.