Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study

Budzyń, K.

doi:10.1016/S2468-1253(25)00133-5

Budzyń deskilling

Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study

Patient / Population Intervention / Exposure Comparison Outcome

In 1,443 patients undergoing colonoscopy without AI assistance, three months of prior routine AI-assisted colonoscopy reduced the adenoma detection rate of later non-AI colonoscopy compared with the three months before AI was introduced.

N

1,443patients

Design

Retrospective multicentre observational study, Poland

Endpoint

Change in ADR of standard (non-AI) colonoscopy before vs after AI exposure

Relevance

3Supporting — context rather than must-know.

ResultADR of non-AI colonoscopy 28.4% (226/795) before vs 22.4% (145/648) after AI exposure; absolute difference −6.0% (95% CI −10.5 to −1.6; P=0.0089). Multivariable OR for AI exposure 0.69 (95% CI 0.53–0.89).

Budzyń K, et al. Lancet Gastroenterol Hepatol. 2025;10(10):896-903. 10.1016/S2468-1253(25)00133-5

Discussion & critique

The deskilling signal for the whole cluster: after AI became routine, endoscopists who had used it regularly detected fewer adenomas when the system was off. Retrospective before–after comparison at four centres of the ACCEPT trial (a randomised AI-colonoscopy trial), unable to exclude secular trends; the authors themselves say AI 'might' reduce ADR. It raises the question whether the ADR gains of CADe trials (Repici, COLO-DETECT) are partly offset by eroded unaided performance.