AI Exposure May Reduce Polyp Detection Skills in Endoscopists

A multicenter observational study conducted in Poland has found that continuous exposure to artificial intelligence (AI) tools in colonoscopy may inadvertently impair endoscopists’ performance when AI is not in use.

The findings, published in the Lancet Gastroenterol & Hepatology, highlight the unintended consequences of overreliance on AI in routine clinical practice—particularly the risk of deskilling among experienced clinicians.

“It is not known if continuous exposure to artificial intelligence (AI) changes endoscopists’ behavior when conducting colonoscopy,” the authors wrote. “We assessed how endoscopists who regularly used AI performed colonoscopy when AI was not in use.”

The retrospective analysis was conducted across four endoscopy centers participating in the ACCEPT (Artificial Intelligence in Colonoscopy for Cancer Prevention) trial. AI for polyp detection was introduced at the end of 2021. Colonoscopies were randomly performed with or without AI assistance depending on the examination date. Researchers compared the performance of non-AI assisted colonoscopies in two timeframes: 3 months before and 3 months after AI implementation.

A total of 1,443 diagnostic colonoscopies were included: 795 prior to AI adoption and 648 following its integration. The study population had a median age of 61 years, with 58.7% identifying as female. Notably, the adenoma detection rate (ADR) in standard, non-AI assisted colonoscopies declined from 28.4% before AI exposure to 22.4% after, marking a statistically significant absolute decrease of –6.0% (95% CI –10.5 to –1.6; p=0.0089).

Multivariable logistic regression identified exposure to AI (odds ratio [OR] 0.69; 95% CI 0.53–0.89), male versus female sex (OR 1.78; 95% CI 1.38–2.30), and age ≥60 years versus <60 years (OR 3.60; 95% CI 2.74–4.72) as independent predictors of ADR.

“Continuous exposure to AI might reduce the ADR of standard non-AI assisted colonoscopy, suggesting a negative effect on endoscopist behaviour,” the authors concluded.

Reference
Budzyń K, Romańczyk M, Kitala D, Kołodziej P, et al. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. Lancet Gastroenterol Hepatol. 2025 Aug 12:S2468-1253(25)00133-5. doi: 10.1016/S2468-1253(25)00133-5. Epub ahead of print. PMID: 40816301.