Of patients who have a positive antinuclear antibody (ANA) test, only 10%-20% will develop a systemic autoimmune disease. The ANA test searches for antibodies generated by the immune system that can mistakenly attack normal proteins in cell nuclei.
While the test can help confirm or rule out any of several diseases — lupus, systemic sclerosis, mixed connective tissue disease, among others — positive results are common in people without autoimmune disease, including healthy individuals. In U.S. studies, rates of positive ANA results in the general population without autoimmune disease range from 14%-27%. And there has been no evidence-based guidance for deciding which ANA-positive patients need prompt rheumatology referral.
In a Vanderbilt Health pragmatic randomized clinical trial of adult patients with positive ANA test results, a systemic autoimmune disease risk model using routine clinical data, paired with fast-track rheumatology referral for patients it flagged as high risk, was associated with significantly shortened time to autoimmune disease diagnosis.
As reported Nov. 6 in NEJM AI, patients in the intervention group were diagnosed with a systemic autoimmune disease at nearly twice the rate of patients in the control group. Among patients who received a diagnosis, the median time from positive test to diagnosis was 43 days in the intervention group and 138 days in the control group.
“It’s widely recognized that the ANA test is over-ordered by clinicians and that positive results trigger many rheumatology referrals that don’t lead to an autoimmune diagnosis. At the same time, patients with lupus often face a yearslong diagnostic odyssey, with treatment delays that carry serious clinical consequences. Our pragmatic randomized trial showed that pairing a real-time risk score with a fast-track referral got high-risk patients diagnosed much sooner,” said the study’s lead and corresponding author, April Barnado, MD, MSCI, formerly Assistant Professor of Medicine in the Division of Rheumatology and Immunology, now working in the pharmaceutical industry.
As described in their earlier 2024 report in Frontiers in Immunology, the team selected for their risk model several variables readily available in the electronic health record from around the time of a first positive ANA result, including age and sex, ANA titer, platelet count, whether another autoantibody was present, and billing codes for nine symptom categories associated with autoimmune diseases, such as joint pain, fatigue, rash and Raynaud’s syndrome.
The new trial randomly assigned 2,385 adult outpatients whose positive ANA test was their first at Vanderbilt Health and who had not previously been seen by a VH rheumatologist; 1,960 were included in the analysis — 954 in the intervention group and 1,006 in the control group. All patients received standard care, and their clinicians could refer them to rheumatology at any time. In the control group, risk scores were calculated but not acted on. In the intervention group, when the model calculated that a patient’s probability of developing a systemic autoimmune disease within 10 years was 20% or greater, the study team notified the clinician who had ordered the test and offered an expedited rheumatology evaluation, which clinicians and patients could decline. Of the 270 patients flagged, 203 (75%) were seen through the fast-track pathway.
The risk model is built on logistic regression, a long-established statistical method. Unlike so-called black box models developed through unsupervised machine learning, it lets clinicians see which factors drive a patient’s risk score, making it a transparent form of artificial intelligence, according to the report’s senior author, Daniel Byrne, MS, former Senior Associate in Biostatistics, now at Johns Hopkins University.
“Thousands of clinical prediction AI models have been published, but very few are built into hospital workflows and tested in randomized trials to determine whether they actually improve patient care,” Byrne said. “This study shows it can be done, and that a transparent AI model paired with an expedited referral can shorten the time it takes for high-risk patients to be evaluated and diagnosed.”
Others on the study include Ryan Moore, MS; Henry Domenico, MS; Emily Grace, PA-C; Sarah Green; Ashley Suh; Nikol Nikolova; Bryan Han; and Allison McCoy, PhD.
The study was supported in part by the National Institutes of Health (awards R01AR080629, UL1RR024975 and UL1TR000445).
