Discovery and validation of novel disease subsets in 806 patients with Takayasu's arteritis across four international cohorts
Arthritis and Rheumatology
Background/Purpose: Takayasu's arteritis (TAK) is characterized by variable patterns of damage throughout the large arteries. This study aimed to develop and validate a novel disease classification system in TAK based on distribution of arterial lesions using data-driven methods. Methods: Data was used from patients with TAK from four independent cohorts: one in India and three in North America (NA). All patients underwent whole-body angiography of the aorta and branch vessels, with categorization of involvement (stenosis, occlusion, or aneurysm) in 13 arterial territories. K-means cluster analysis was performed to identify subgroups of patients based on pattern of angiographic involvement. Cluster groups were identified in the Indian cohort and independently validated in the NA cohorts. Recursive partitioning was used to develop a decision tree to predict cluster assignment. Results: 581 and 225 patients with TAK were included from the Indian and NA cohorts, respectively. Three distinct clusters were identified in the Indian cohort and validated in the NA cohorts. Patients in Cluster 1 had significantly more disease in the abdominal aorta, renal, and mesenteric arteries (p < 0.01). Patients in Cluster 2 had significantly more bilateral disease in the carotid and subclavian arteries (p < 0.01). Compared to Clusters 1 and 2, patients in Cluster 3 had asymmetric disease with fewer involved territories (p < 0.01). When serial angiography was available for review, only 1 of 109 patients changed cluster assignment over time. There were more patients from India in Cluster 1 (41% vs 24%; p < 0.01) and more patients from NA in Cluster 3 (41% vs 32%; p = 0.03). Recursive partitioning predicted cluster assignment in the Indian cohort with 92% accuracy and cross-predicted cluster assignment of the NA cohorts with 87% accuracy based on involvement of the abdominal aorta, carotid, subclavian, mesenteric, and renal arteries. In the Indian and the NA cohorts, patients in Clusters 1 and 2 compared to Cluster 3 were more likely to have arterial occlusions (58% vs 82% vs 37%; p < 0.01) and a history of tuberculosis (8% vs 10% vs 3%; p = 0.03). Disease onset in childhood (28% vs 16% vs 19%; p < 0.01) and hypertension (71% vs 42% vs 39%; p < 0.01) were more common in Cluster 1. Stroke (0% vs 22% vs 5%; p = 0.03), vision loss (0% vs 33% vs 6%; p = 0.01), carotidynia (3% vs 26% vs 9%; p = 0.01) and persistent disease activity (46% vs 59% vs 44%; p = 0.02) were significantly more prevalent in Cluster 2. Conclusion: This large study in TAK identified and validated three novel subsets of patients based on patterns of arterial disease. The same subsets were seen in patients from India and NA; however, prevalence of patients within each subset differed by country. Angiographic-based disease classification may help identify causal disease factors and enable stratified clinical decision making in this complex, clinically heterogeneous disease.