AI Speeds Up Rheumatology Diagnoses but Not Accuracy
A randomized trial found that an EU-certified large language model, Prof. Valmed, reduced diagnosis time for rheumatologic conditions but did not improve accuracy compared to physicians. Doctors using the AI were more confident, possibly overconfident.
The randomized ALLIANCE trial enrolled 82 physicians from seven institutions across Germany and Norway. Participants diagnosed three specific scenarios—Cogan syndrome, dermatomyositis, and familial Mediterranean fever—drawn from published case reports.
Prof. Valmed, which earned European CE certification in March 2025, was created by an attorney and a neuroimmunologist as an "AI copilot" with guardrails to limit hallucinations. Investigators observed substantial AI over-reliance in the intervention group, indicating the tool's speed advantage may come with behavioral trade-offs.
Faster diagnostic support could alleviate rheumatology clinic bottlenecks and reduce patient waiting times. However, the observed overconfidence and AI over-reliance may lead to diagnostic errors if physicians defer too readily to the tool. This could impact patient safety and medical liability, suggesting that healthcare systems must implement robust training and calibration protocols before integrating such AI into routine practice.