A new study suggests that doctors who regularly rely on AI for colonoscopies become significantly less effective at detecting precancerous lesions when the technology is not available. Experts say this points to a worrying decline in diagnostic skills.
Researchers in Poland tracked 1,443 colonoscopies performed without AI support at four medical centers, all by 19 experienced endoscopists. After AI became part of routine practice, the adenoma detection rate (ADR) - a key quality measure for spotting early cancer signs - dropped from 28.4% to 22.4%. The study, published in The Lancet Gastroenterology & Hepatology, links this decline to the routine use of AI during screenings.
The authors suggest that constant reliance on AI may lower cognitive engagement, motivation, attention, and a sense of responsibility when making diagnostic decisions - a phenomenon sometimes compared to losing basic navigation skills when always using GPS.
Calls for safeguards
In a commentary, Omer Ahmad of University College London describes this as an unintended loss of expertise. He recommends creating guidelines for performance monitoring, educational programs, and regular practice sessions without AI to help clinicians maintain their diagnostic skills.
Ahmad also notes there is no simple solution, emphasizing the need for high-quality crossover studies that compare both behavior and outcomes in AI-assisted and non-AI-assisted colonoscopies.
The study has its limitations: it was observational and not randomized, meaning selection bias cannot be ruled out. Only one AI system was used, so the results may not apply to other technologies. Additionally, since all participating doctors had performed at least 2,000 colonoscopies, less experienced clinicians might be even more vulnerable to skill loss.
