AI Use Linked to Doctor Skill Decline in Colonoscopy Detection
A recent study casts a shadow on the widespread integration of artificial intelligence into medical diagnostics, suggesting that doctors who routinely rely on AI during colonoscopies may experience a significant decline in their ability to detect precancerous lesions when the technology is unavailable. This finding, published in The Lancet Gastroenterology & Hepatology, raises concerns about an unintended erosion of crucial diagnostic skills in an era of increasing technological dependence.
The observational study, conducted across four medical centers in Poland, tracked 1,443 colonoscopies performed by 19 experienced endoscopists. Crucially, these procedures were carried out without AI support, but after AI had already become a standard part of the doctors’ routine practice. Researchers observed a notable drop in the adenoma detection rate (ADR)—a key quality metric indicating how often precancerous growths are identified during a colonoscopy. Before AI was routinely adopted, the ADR stood at 28.4%; after its integration, and when doctors subsequently performed procedures without AI, this rate fell to 22.4%.
The authors of the study hypothesize that constant reliance on AI for diagnostic support may inadvertently diminish a clinician’s cognitive engagement, motivation, attention, and overall sense of responsibility. They draw a parallel to the everyday experience of over-relying on GPS, which can lead to a gradual loss of fundamental navigation skills. In a high-stakes field like medicine, such a decline could have serious implications for patient outcomes.
In a commentary accompanying the study, Omer Ahmad of University College London described this phenomenon as an “unintended loss of expertise.” He stressed the urgent need for safeguards to mitigate this risk. Ahmad recommended implementing clear guidelines for performance monitoring, establishing robust educational programs, and mandating regular practice sessions for clinicians without AI assistance. Such measures, he argued, are vital to help medical professionals maintain their innate diagnostic acumen. He also called for more rigorous, high-quality crossover studies that can directly compare both clinician behavior and patient outcomes in scenarios where AI is present versus when it is not.
While the Polish study offers compelling insights, it is important to acknowledge its limitations. As an observational study, it was not randomized, meaning that potential selection bias cannot be entirely ruled out. Furthermore, the research focused on the use of a single AI system, which means its findings may not be universally applicable to other AI technologies. Another significant point is that all participating doctors were highly experienced, having performed at least 2,000 colonoscopies each. This raises the distinct possibility that less experienced clinicians, who might be more susceptible to skill degradation, could be even more vulnerable to this concerning trend.
The findings underscore a critical challenge in the ongoing integration of AI into healthcare: ensuring that technological advancements augment, rather than inadvertently diminish, the essential human skills that remain indispensable for effective patient care. As AI becomes more pervasive, striking the right balance between automation and human expertise will be paramount.