Turn on cookies to sign in
Signing in saves your progress to your Curie account. We can only do that with cookies on — turn them on to continue.
AI cuts diagnosis time in half — but accuracy stays the same. A randomized controlled trial found that an LLM-based diagnostic tool slashed case-processing time by more than half compared to conventional resources, but failed to significantly improve diagnostic accuracy in rare rheumatologic conditions. Clinicians also showed signs of over-reliance on the AI's suggestions.
AI speeds up rheumatology diagnoses but doesn't make them more accurate
A first-of-its-kind randomized controlled trial presented at ACR Convergence 2026 tested whether a CE-certified large language model (LLM) could help physicians diagnose rare rheumatologic conditions faster and more accurately. The ALLIANCE trial enrolled 82 physicians across seven hospitals, randomly assigning them to either LLM-based diagnostic support plus conventional resources, or conventional resources alone.
The results were a mixed bag. The LLM dramatically cut case-processing time — just 94 seconds vs. 206 seconds with traditional tools — but didn't move the needle on accuracy. Top-one diagnostic accuracy after the intervention was nearly identical: 33.3% in the LLM group vs. 35% in controls. Notably, overconfidence was observed across all settings and increased following the intervention.
By the Numbers
Why it matters: Faster diagnoses could help reduce the notoriously long delays in rheumatology — but speed without accuracy gains raises real concerns, especially given the overconfidence observed among physicians using the AI tool. Researchers say further studies in routine care are needed before these time savings can be considered a true clinical benefit.
