Clinical phenotypes of ANCA-associated vasculitis using ensemble clustering: prognostic and therapeutic insights from two Japanese cohorts.

Abstract

Clinical heterogeneity in anti-neutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) is not fully captured by conventional disease subtypes or ANCA serotypes. We aimed to identify data-driven, clinically meaningful disease phenotypes based on patterns of organ involvement in microscopic polyangiitis (MPA) and granulomatosis with polyangiitis (GPA), and to evaluate their prognostic and therapeutic implications. We conducted a multicenter retrospective cohort study using two nationwide Japanese registries (J-CANVAS as derivation; JPVAS as validation). Ensemble clustering was applied to high-dimensional organ involvement data derived from the Birmingham Vasculitis Activity Score and additional clinically relevant manifestations. The resulting clusters were replicated using classification and regression tree (CART) across the derivation and validation cohorts, and associations between cluster memberships and clinical outcomes (overall survival and relapse), as well as response to induction therapy, were evaluated. Among 726 patients with newly diagnosed MPA or GPA, ensemble clustering identified four reproducible clinical phenotypes: (a) renal-dominant without interstitial lung disease (ILD); (b) renal-dominant with ILD; (c) systemic multi-organ; and (d) ear, nose, and throat (ENT)-dominant. CART reproduced these clusters with good concordance. Overall survival and relapse incidence differed across clusters. The ENT-dominant cluster demonstrated favorable survival but a higher relapse risk and showed a trend toward improved relapse-free survival with rituximab compared with cyclophosphamide, even among myeloperoxidase-ANCA-positive patients. Organ involvement-based ensemble clustering identifies clinically meaningful and reproducible AAV phenotypes with distinct prognostic and therapeutic profiles. Integrating phenotypic patterns with ANCA serotype may enhance risk stratification and support more individualized management strategies in AAV.