AI Assistance in Colonoscopy linked to Potential Skill Decline in Endoscopists
New research suggests that continuous use of artificial intelligence (AI) during colonoscopies may lead to a reduction in endoscopists’ adenoma detection rates (ADR), raising concerns about potential “de-skilling.” The study, published in Lancet Gastroenterology & Hepatology in 2025, found that ADR decreased slightly after the introduction of AI assistance (0.43 vs. 0.54 before AI), though this difference wasn’t statistically significant.
Researchers analyzed data from 19 experienced endoscopists, finding no difference in the number of adenomas per colonoscopy, the detection rate of advanced adenomas, or colorectal cancer detection rates between periods with and without AI assistance. However, univariable logistic regression analysis revealed a significant decrease in ADR after AI exposure (odds ratio [OR]=0.70).
Several factors were identified as predictors of ADR. men had a substantially higher ADR than women (OR=1.51), as did patients aged 60 or older compared to younger patients (OR=3.46). ADR was also higher in patients without alarm symptoms (OR=1.67). These findings largely held true in a multivariable analysis, with adjusted odds ratios of 0.69, 1.78, and 3.60 respectively; the association with alarm symptoms was no longer statistically significant in the multivariable model.
Notably, the study found that the majority – all but four - of the 19 endoscopists experienced a drop in ADR following AI implementation. The ”de-skilling effect” appeared more pronounced in centers with higher baseline adrs (29.4-39.7%) compared to those with more modest rates (21.7-22.9%),suggesting that those already performing well had less room for improvement. Procedures performed by women endoscopists also showed a more considerable decrease in ADR (-15.1%) compared to those performed by men (-2.9%).
The researchers hypothesize that continuous reliance on AI decision support systems may lead to clinicians becoming “less motivated, less focused, and less responsible when making cognitive decisions without AI assistance.” They also suggest that this observed negative effect of continuous AI exposure may have influenced the results of previous trials comparing AI-assisted and non-AI-assisted colonoscopies.
The study acknowledges several limitations, including its retrospective design, the relatively short duration of AI exposure, and the focus on a limited number of experienced endoscopists, potentially limiting generalizability. A lack of blinding and variations in case volume per endoscopist were also noted.
In a related commentary, Omer Ahmad (University College london, UK) stated the findings “temper the current enthusiasm for rapid adoption of AI-based technologies” and “highlight the importance of carefully considering possible unintended clinical consequences.” He raised questions about the underlying mechanisms of de-skilling,its reversibility,and whether similar effects could occur with other AI applications,such as optical diagnosis.
Ahmad concluded, ”Although AI continues to offer great promise to enhance clinical outcomes, we must also safeguard against the quiet erosion of fundamental skills required for high-quality endoscopy.”
The research was originally published in Lancet Gastroenterology & Hepatology 2025; doi:10.1016/S2468-1253(25)00133-5 and 10.1016/S2468-1253(25)00164-5.
