A stratified approach to juvenile idiopathic arthritis: leveraging machine learning to uncover subgroups of patients following treatment with etanercept.

OBJECTIVES: To facilitate a stratified approach to treatment of JIA and uncover trajectory-based patterns of key clinical outcomes following etanercept (ETN) initiation in children and young people (CYP). METHODS: ETN-naïve CYP with non-systemic JIA recruited to one of three UK prospective multicentre cohorts were selected if they started ETN between 2001 and 2019. JADAS71 (71-joint Juvenile Arthritis Disease Activity) components [active joint count, physician global assessment (PGA), parent global evaluation (PGE) and ESR] were collected in the year following ETN initiation. Clusters of CYP based on JADAS71 component trajectory patterns following ETN initiation were identified using multivariate group-based trajectory models, adjusting for year of ETN initiation. Multinomial logistic regression identified independent predictors of outcome clusters. In a subset of CYP with previously determined MTX trajectories, response clusters were compared descriptively. RESULTS: In 1003 CYP, five clusters were identified following ETN initiation. In three clusters, JADAS components changed in parallel: fast improvers (16%), slow improvers (10%) and persistent disease (37%). In two, all components except for one improved over time: persistent PGA (7%) and persistent PGE (30%). Where both MTX and ETN trajectories were available (n = 139), 69% of CYP with persistent disease following MTX were assigned to other response clusters following ETN initiation. There was high consistency (73%) of CYP having a persistent PGE pattern following both drugs. CONCLUSION: Across multiple DMARDs, JIA disease activity does not always improve to similar degrees across the different features of disease. A majority of CYP initially experiencing persistent disease with MTX fall into a response group following ETN.