A positive antinuclear antibody (ANA) blood test can point to lupus, systemic sclerosis and related diseases, but only 10% to 20% of people who test positive go on to develop a systemic autoimmune disease. Positive results are common in healthy people too, and there has been no evidence-based guidance on who needs a rheumatologist quickly. Meanwhile, patients with lupus often wait years for a diagnosis.
The clever bit
The AURA model reads data already sitting in the health record around a first positive ANA: age, sex, ANA titer, platelet count, whether another autoantibody showed up, and billing codes for nine symptom groups such as joint pain, fatigue, rash and Raynaud’s. It estimates the chance of a systemic autoimmune disease within 10 years. At 20% or more, the study team alerted the clinician who ordered the test and offered a fast-track rheumatology visit, which clinicians and patients could decline. Because the model is transparent, clinicians can see which factors drive each score.
From June 2023 to March 2025, adults with a first positive ANA at Vanderbilt Health were randomly split into two groups. In the control group, scores were calculated but never shared. Among the 1,960 patients analyzed, the alert group was diagnosed sooner. For patients who did turn out to have an autoimmune disease, the median wait from blood test to diagnosis was 43 days with the alert and 138 days without. Of the 270 patients flagged as high risk, 203 (75%) were seen through the fast track.
Our pragmatic randomized trial showed that pairing a real-time risk score with a fast-track referral got high-risk patients diagnosed much sooner.
— April Barnado, MD, lead author, formerly assistant professor of medicine, Vanderbilt University Medical Center, via news.vumc.org
Hooray
In the alert group, the typical diagnosed patient got an answer 95 days sooner than in the control group, roughly three months less of waiting and wondering. The score ran on its own inside the everyday health record, set off by a routine blood test, and it was tested the hard way: in a randomized trial with real patients, which few clinical AI models ever face.
What comes next
The authors made the model open source so other health systems can check it on their own patients and run their own trials. The same team is now testing whether AI can flag mothers at risk of postpartum hemorrhage, in a pragmatic trial due to wrap up by the end of 2026.
