Cluster analysis reveals the clinical spectrum of Erdheim-Chester disease.

Erdheim-Chester disease (ECD) is a clonal-inflammatory neoplasm driven by mutations in MAPK pathway proto-oncogenes, such as BRAF. Clinical manifestations are protean, affecting virtually every system. This cohort study analyzed 661 patients with ECD to classify them based on clinical features and mutational profiles using unsupervised clustering. Nineteen clinical and mutational variables were subjected to hierarchical clustering combined with k-means. A three-cluster model emerged as the most stable. Most patients were classified according to key features, namely BRAF mutation, and large-vessel, heart, and perirenal involvement. The "Widespread Disease" (WID) cluster (320 patients, 49%) was associated with the presence of the key features and the "Limited Disease" (LIM) cluster (282 patients, 42%) was associated with their absence. The "MAP2K1-RDD" cluster (MAP) was assigned 59 patients (9%), based on MAP2K1 mutation and/or overlapping Rosai-Dorfman-Destombes disease (RDD). Survival analysis revealed worse outcomes for WID compared to LIM (hazard ratio 1.54, 95% CI 1.09-2.17), while no significant survival difference was found for MAP. The identification of these clusters, based on mutational profiles, organ involvement and overlapping conditions, offers a data-driven validation of established clinical observations. These findings substantiate the role of the somatic mutation type in shaping the ECD phenotype.