Evaluating the AI App "POSOS" for Detecting Drug-Induced Iatrogenesis (A Clinical Vignettes Study) — a clinical trial (ClinicalTrials.gov)

Iatrogenesis means harm caused by medical care itself, and among its forms, drug side effects and drug-drug interactions are especially easy to overlook. When a patient takes several medications, or when symptoms resemble other illnesses, recognizing in a busy emergency setting that "this symptom might be caused by a drug" is far from straightforward. Software that can rapidly cross-check vast amounts of medication information is seen as a potential aid in reducing such oversights.

What makes this study notable is its design: the AI app is evaluated not on real patients but on "clinical vignettes" (simulated cases), and physicians are randomly split into groups that do or do not use the app for comparison. Rather than trying a new tool directly in live care, this approach seeks to isolate, in a safe simulated environment, how having the tool changes a clinician's judgment — a careful way to evaluate medical AI. Responses are recorded at two time points, after 5 and 15 minutes, alongside the time taken, the number of medical-search apps used, and the overall user experience. Note that the app's effectiveness, accuracy, and safety are not established; the study aims to test its usefulness.

The challenge of medication safety is broadly shared across countries and health systems. In the United States as well, drug interactions and adverse events are an important public-health topic, and the question of how to measure the performance of such detection-support tools carries cross-cutting lessons for how medical AI itself should be evaluated. A randomized comparison built on simulated cases is a framework that could be applied to evaluating other diagnostic-support tools too.