Concerns Arise Over AI Impact on Colonoscopy Skills in Health Professionals

The increasing integration of artificial intelligence (AI) technology in medical procedures raises pressing concerns regarding its potential effects on health professionals' skills, particularly among those conducting colonoscopies. A recent study published in The Lancet Gastroenterology & Hepatology highlights a troubling trend: endoscopists may experience a decline in their ability to detect precancerous growths, known as adenomas, when they rely on AI assistance.

Colonoscopy is a vital procedure for identifying and eliminating adenomas, thereby preventing bowel cancer. While numerous studies have demonstrated that AI can enhance adenoma detection rates, this research underscores a significant gap in understanding how ongoing use of AI affects the proficiency of endoscopists. Researchers have expressed concerns that while AI may train clinicians, it could simultaneously lead to a deterioration of their skills.

Dr. Marcin Romanczyk from the Academy of Silesia in Poland, a primary author of the study, emphasized that this research is among the first to indicate a negative impact of regular AI usage on the ability of healthcare professionals to perform critical tasks. He pointed out the urgency for further studies to explore the implications of AI's role across diverse medical fields and to identify factors contributing to potential inefficiencies in human-AI collaboration.

Conducted across four colonoscopy centers in Poland, the study spanned from September 2021 to March 2022. The centers implemented regular AI assistance at the end of 2021, leading to randomly assigned colonoscopies performed either with or without AI support. In total, 1,443 colonoscopies were performed without AI: 795 prior to the integration of AI and 648 afterward, carried out by 19 experienced endoscopists, each with over 2,000 procedures under their belts.

Findings revealed a significant drop in adenoma detection rates at non-AI-assisted colonoscopies, which decreased from 28.4% (226 out of 795) before AI exposure to 22.4% (145 out of 648) post-AI exposure. This represents a 20% relative and 6% absolute reduction in adenoma detection rates. Conversely, in colonoscopies assisted by AI, the adenoma detection rate was 25.3% (186 out of 734).

Prof. Yuichi Mori from the University of Oslo remarked that these results prompt a reevaluation of previous randomized controlled trials suggesting that AI-assisted colonoscopy yields higher adenoma detection rates. It is possible that non-AI-assisted colonoscopy practices in those trials differ from standard procedures, as the endoscopists may have been adversely impacted by continuous AI exposure.

The authors also recognized limitations inherent in their observational study design, which may have allowed other factors to influence the findings. Furthermore, because the study involved only seasoned endoscopists, the results may not reflect the experiences of less experienced practitioners. Future research focused on the long-term effects of AI use among novice health professionals is essential to fully understand its impact on their adenoma detection capabilities.

In a related commentary, Dr. Omer Ahmad from University College London, who was not part of the study, noted that the findings temper the enthusiasm surrounding rapid AI adoption in medical practices. He urged a careful examination of the unintended clinical consequences that could arise from such technologies. Although AI holds the potential to enhance clinical outcomes, it is crucial to monitor and prevent the gradual erosion of fundamental skills necessary for quality endoscopic practices.