Artificial Intelligence Models Accurately Answer Clinical Questions About Urinary Stones
Artificial intelligence tools capable of understanding and generating natural language show a remarkable ability to support healthcare professionals in specialized medical fields. A recent evaluation tested four of these systems against precise clinical questions regarding the management of urinary stones, a common and complex condition.
Researchers developed one hundred and five questions covering various aspects of this disease, ranging from diagnosis to treatments, including specific situations such as pregnancy or pediatric cases. Each system answered these questions in two separate sessions. Two experienced urologists then evaluated the responses by comparing them to the recommendations of reference clinical guidelines. The results reveal that all systems achieved high scores, with an average between 86 and 93 percent, indicating strong compliance with medical guidelines. Among them, one model stood out with a particularly strong correlation to expert evaluations, achieving a 94 percent agreement level.
No significant differences were observed between the systems’ performances or between the different question categories, suggesting stability in their ability to provide reliable information. Additionally, the responses generated during the two sessions were very similar, confirming the repeatability of the results.
These tools could therefore serve as support for medical training and clinical assistance, providing answers aligned with best practices. Their use as a complementary aid in the field of urology appears promising, although further validation in real-world contexts is needed to confirm their utility in daily practice. Their integration could facilitate access to standardized and up-to-date information, while emphasizing the importance of maintaining human supervision to ensure the safety and relevance of medical decisions.
Sources and Credits
Source Study
DOI: https://doi.org/10.1007/s00240-026-02013-1
Title: Performance of large language models in urological decision support: a guideline-based comparative evaluation in urolithiasis
Journal: Urolithiasis
Publisher: Springer Science and Business Media LLC
Authors: Adem Tunçekin; Yasin Aktaş