Can AI tell when patients don't need medical tests? Israeli study puts chatbots to the test

A study by Ben-Gurion University and Clalit Health Services found that leading AI chatbots can identify unnecessary medical tests in most cases, with errors clustering around ambiguous practices like EKGs and ultrasounds. The research was published in Annals of Internal Medicine.

A new joint study from Ben-Gurion University of the Negev and Clalit Health Services reveals that leading Large Language Models (LLMs) can identify unnecessary medical interventions in most cases. The peer-reviewed study, published in Annals of Internal Medicine, evaluated six LLMs from Google, Anthropic, and OpenAI across hundreds of clinical scenarios based on the international Choosing Wisely campaign. When an intervention was medically indicated, the models mistakenly recommended withholding care in only 0.6% of cases. When an intervention was explicitly unneeded, the models inappropriately recommended testing or treatment in just 5.4% of cases. However, errors clustered around two ambiguous practices: unnecessary EKGs (28.3% of responses) and unnecessary ultrasounds (42.8% of responses), mirroring real medical disagreement. The researchers noted that models varied widely on the same cases, with one recommending an unnecessary test less than 10% of the time and another 80% of the time. The study was tested only in English, and more work is needed before clinical validation.