AI Knows When To Skip Medical Tests, Except Where Doctors Disagree, Israeli Study Finds
By The Media Line Staff
Artificial intelligence models rarely recommended unnecessary medical care in a new Israeli study, but their accuracy dropped sharply on tests that doctors themselves disagree over, researchers at Ben-Gurion University of the Negev and Clalit Health Services found.
The study, published in the Annals of Internal Medicine, tested six large language models developed by Google, Anthropic and OpenAI on clinical decisions involving whether to test or treat a patient — or leave well enough alone.
Overall, the models recommended unnecessary intervention in just 5.4% of cases where restraint was called for. They made the opposite mistake — withholding care when it was medically indicated — only 0.6% of the time. In eight of the 15 Choosing Wisely recommendations examined, the models never recommended an unnecessary procedure.
Their biggest problems emerged in two medical gray areas.
For small thyroid nodules discovered incidentally, the models recommended unnecessary ultrasounds in 42.8% of responses. Researchers noted that the American Thyroid Association and American College of Radiology themselves disagree over whether such nodules should be imaged. The models also recommended unnecessary routine EKGs for low-risk patients in 28.3% of responses, particularly when patients reported palpitations.
“One of the hardest parts of practicing medicine is knowing when to hold back,” said Dr. Dean Ariel, a family physician and researcher at Clalit Health Services (Tel Aviv District) and medical AI researcher. “Patients find comfort in an extra test, but an unnecessary MRI often catches an asymptomatic disk bulge, setting off unnecessary worry and treatment. Language models learned medicine from human clinical literature—including our less disciplined habits. We wanted to see if they also absorbed the ability to exercise restraint. For the most part, they know when to stop, stumbling primarily where we doctors ourselves hesitate.”
Researchers also found wide differences between individual AI models when medicine offered no clear consensus. On identical cases, one model recommended an unnecessary test less than 10% of the time while another did so 80% of the time, according to Dr. Guy Laban of Ben-Gurion University of the Negev.
Laban said the next question is whether AI will maintain that restraint when confronted by anxious patients who insist on additional testing.
Brought to you by www.srnnews.com
SRN News — View the original story at SRNNews.com.
