Machine Learning Methods May Help Identify Subgroups of Patients With Juvenile Idiopathic Arthritis

orthopedic pediatric doctor visit
Study authors used machine learning approaches to identify unique clusters of children and youth with juvenile idiopathic arthritis, both at first presentation and during follow-up.

Machine learning techniques can be used to identify subgroups of children and youth with juvenile idiopathic arthritis (JIA), according to results an analysis published in Lancet Rheumatology. Distinct patient subgroups, as defined by disease manifestation or trajectories of disease progression, may help in the better management of patients with JIA.

Because clusters of children and youth with JIA may experience varying global patterns in the signs and symptoms of their condition, the study authors sought to identify these subgroups during a 3-year follow-up after diagnosis. They noted that recognition of patient clusters may allow for an enhanced understanding of disease progression in patients with JIA, including the association between patient- and physician-reported outcomes during disease course.

A multicenter, prospective, longitudinal study was conducted among children and youth enrolled in the Childhood Arthritis Prospective Cohort (CAPS) — a United Kingdom inception cohort — before January 1, 2015. Between January 1, 2001 and December 31, 2014, a total of 1423 children and youth with JIA were enrolled to the cohort. Among these patients, 239 were excluded from the study (238 with no record of clinical Juvenile Arthritis Disease Activity Score [cJADAS] and 1 who died), yielding a final study population of 1184 children and youth. Overall, 65% (n=773) of the participants were girls. Among participants with available International League of Associations for Rheumatism (ILAR) category data, 50% (n=594/1179) had oligoarthritis. Median age at initial presentation was 7.4 years (interquartile range, 3.4-11.7 years).

During the 3-year period, 205 children and youth were discharged from pediatric rheumatology for reasons that included being “well” (39%), repeat nonattendance (13%), and transfer to adult service (29%). Overall, cJADAS 10 (cJADAS with an active joint count up to 10) data were available for 96% of the children and youth at baseline, 77% at the 6-month follow-up, 94% at the 1-year follow-up, 87% at the 2-year follow-up, and 80% at the 3-year follow-up.

Using longitudinal follow-up data, the study authors were able to identify 5 clusters at baseline and 6 trajectory groups. Although disease course could not be well predicted from clusters at baseline, in both cross-sectional and longitudinal analyses, significant percentages of children and youth (approximately 1 in 4) had high patient or parent global scores despite having low or improving joint counts and physician global scores. Of note, the participants in these groups were older, and a higher percentage of them had enthesitis-related JIA and a lower socioeconomic status than those in other groups.

Study authors concluded, “Distinct patient subgroups defined by disease manifestation or trajectories of progression could help to better personalize health care services and treatment plans for individuals with JIA.”

Disclosure: The CLUSTER consortium was supported by Pfizer Inc. One study author declared affiliations with the pharmaceutical industry.


Shoop-Worrall SJW, Hyrich KL, Wedderburn LR, Thomson W, Geifman N; CAPS and the CLUSTER Consortium. Patient-reported wellbeing and clinical disease measures over time captured by multivariate trajectories of disease activity in individuals with juvenile idiopathic arthritis in the UK: a multicenter prospective longitudinal study. Lancet Rheumatol. Published online December 4, 2020. doi:10.1016/S2665-9913(20)30269-1