AI-based analysis reveals 4 distinct patterns of organ involvement in AAV

A type of artificial intelligence (AI) called ensemble clustering identified four distinct patterns of organ involvement in ANCA-associated vasculitis (AAV) that were associated with differences in survival and relapse rates, a study showed.

The pattern dominated by ear, nose, and throat (ENT) involvement was associated with the highest survival but also the highest relapse rate. In contrast, the pattern marked by kidney involvement and interstitial lung disease (ILD) was linked to the lowest survival.

Recommended Reading

Sept. 9, 2026 News by Steve Bryson, PhD

Study links kidney artery inflammation to kidney failure in AAV

Organ-based patterns reveal distinct AAV profiles

“Organ involvement-based ensemble clustering identifies clinically meaningful and reproducible AAV [patterns] with distinct prognostic and therapeutic profiles,” the researchers wrote.

Combining these patterns with the type of ANCA detected in the blood “may enhance risk stratification and support more individualized management strategies in AAV,” the researchers added. ANCAs are self-reactive antibodies that drive most AAV cases.

The study, “Clinical phenotypes of ANCA-associated vasculitis using ensemble clustering: prognostic and therapeutic insights from two Japanese cohorts,” was published in Rheumatology.

AAV is a group of autoimmune diseases marked by inflammation and damage to small blood vessels, affecting various organs and tissues. These immune attacks are driven mostly by ANCAs that target one of two enzymes, proteinase 3 (PR3) or myeloperoxidase (MPO), in neutrophils, a type of immune cell.

The two most common types of AAV are granulomatosis with polyangiitis (GPA) and microscopic polyangiitis (MPA). GPA involves immune cell clusters called granulomas and affects the upper respiratory system, lungs, and kidneys. MPA also affects the lungs and kidneys, as well as the nervous system, but does not involve granulomas.

However, substantial clinical variation exists within and across these conventional types, and ANCA types do not fully explain the diversity of symptoms and outcomes among people with AAV.

To look for other clinically meaningful disease profiles, researchers in Japan used ensemble clustering, a type of AI that combines results from multiple grouping methods to identify more stable and reproducible patterns.

Study analyzes two Japanese AAV cohorts

The team retrospectively analyzed data from two nationwide Japanese registries. J-CANVAS included 726 people with newly diagnosed MPA or GPA, while JPVAS provided a separate external validation cohort. Among the J-CANVAS patients, the median age was 75 years, and more than half were women (57.7%). Most (88.3%) tested positive for MPO-ANCAs, while 13.2% tested positive for PR3-ANCAs.

Ensemble clustering was applied to 481 J-CANVAS patients to identify disease patterns based on organ involvement. The analysis included data from the Birmingham Vasculitis Activity Score (BVAS) across nine areas: general, skin, mucous membranes and eyes, ENT, chest, cardiovascular, abdominal, kidneys, and nervous system. Data about other relevant manifestations, including interstitial lung disease (ILD), were also included.

The analysis identified four groups (A, B, C, and D) based on their organ involvement patterns. Researchers then developed a classification model to reproduce those groups in additional patients.

When the classification model was applied to the 726 J-CANVAS patients, group A (28.4%) was dominated by kidney involvement without ILD. Group B (17.4%) was also kidney involvement-dominant, but all patients had ILD and the group had higher rates of smoking, chronic kidney disease, and lung disease. Both groups had the highest rates of severely reduced kidney function and the highest BVAS scores, indicating greater disease activity.

Group C (34.8%) was characterized by broad organ involvement, including general and nervous-system manifestations, but had less severe organ involvement and better-preserved kidney function. Lastly, group D (19.4%) was ENT-dominant and showed the highest proportion of patients testing positive for PR3-ANCAs.

Survival and relapse differed across AAV patterns

Among the 726 J-CANVAS patients, over a median follow-up of 2.7 years, relapse, defined as a severe, organ-threatening or life-threatening increase in disease activity, occurred in 2.8% of patients at one year and 4.8% at five years. About one in 20 (4.8%) died at one year, and about one in 10 (10.5%) had died by five years.

The ENT-dominant group D had the highest relapse incidence, but also the highest overall survival. Group B, marked by kidney involvement plus ILD, had the lowest survival. A similar pattern was seen among patients testing positive for MPO-ANCAs. In contrast, among PR3-ANCA-positive patients, survival did not clearly differ across groups, while relapse incidence was higher in groups C, marked by broad organ involvement, and D, marked by ENT involvement.

Statistical analyses adjusted for age, sex, smoking, AAV type, ANCA type, kidney function, and co-existing conditions continued to show favorable survival but a higher incidence of relapse in group D. Differences among groups A-C were less pronounced after adjustment.

Finally, one-year relapse-free survival — time without death or relapse — did not differ meaningfully by initial (induction) treatment regimen in groups A, marked by kidney involvement without ILD, and C.

The AAV-approved therapy rituximab showed a favorable trend compared with cyclophosphamide in group B and compared with cyclophosphamide or other regimens in group D. The group D treatment-response pattern was also seen among patients with MPO-ANCAs.

Because treatment was not standardized in this retrospective study, the researchers said these treatment findings should be interpreted cautiously.

The researchers noted that the treatment-response pattern in group D aligned with previous observations in PR3-ANCA-positive AAV in Western studies. Because most patients in group D were MPO-ANCA-positive, the findings suggest that some MPO-ANCA-positive patients may have a “PR3-like [profile].”

“Through ensemble clustering of organ involvement patterns, we identified four distinct clinical [patterns] among Japanese patients with AAV that differed in [outcomes] and response to induction therapy,” the team wrote. “Integrating [pattern-based] stratification with ANCA [type] will facilitate individualized risk assessment and therapeutic decision-making.”