Eng SWM, et al. - Using unsupervised machine learning, researchers analyzed patterns of articular involvement in juvenile idiopathic arthritis (JIA) via exploring baseline joint involvement data from 640 treatment-naïve patients from 16 Canadian centers. According to findings, multilayer non-negative matrix factorization (NMF) identified seven joint involvement patterns (pelvic girdle, fingers, wrists, toes, knees, ankles and indistinct) in children with arthritis that predicted disease trajectory. The hierarchical unsupervised approach identified a new clinical feature, the degree of localization, which predicted results in both cohorts. Investigators found that localized joint involvement patients reached zero joint involvement faster than nonlocalized patients. Detailed evaluation of each joint for children with arthritis is already part of every musculoskeletal examination. This investigation supported the continued collection of detailed joint involvement and the inclusion of patterns and localization degrees to stratify patients and inform treatment decisions. This advances pediatric rheumatology from the counting of joints to the realization of the potential use of available data from the discovery of joint involvement patterns.