Predicting Inactive Disease in Juvenile Idiopathic Arthritis: A Dynamic Machine Learning Approach.

Abstract

Optimal management of juvenile idiopathic arthritis (JIA) remains challenging. The ability to predict disease progression in JIA can improve personalized treatment decisions, but few reliable predictors have been identified. Using data from the Childhood Arthritis and Rheumatology Research Alliance (CARRA) Registry (years 2015-2024), we developed machine learning models to predict attainment of inactive disease in children with non-systemic JIA. We compared three structurally distinct algorithms: Dynamic Bayesian Networks (DBN), Convolutional Neural Networks (CNN), and eXtreme Gradient Boosting (XGBoost). Using a dynamic approach, the models predicted at each timepoint whether a child would attain inactive disease approximately six months later, based on the preceding 12 months of data (demographic, clinical, and medication variables), with predictions updated as new visits accrued. A total of 8,093 participants were included. On an independent test cohort, all 3 models achieved comparable discrimination (AUC 0.76-0.79). Predicted probabilities were well-calibrated (Expected Calibration Error ≤0.03). Disease activity levels in the preceding 12 months, presence of enthesitis and uveitis were the strongest predictors. The dependency structure learned by the DBN indicated that treatment decisions were primarily reactive to disease activity and shaped by JIA category and step-therapy requirements. Machine learning approaches can predict attainment of inactive disease over time with moderate discriminative performance, offering potential to support early identification of children with JIA at risk for unfavorable outcomes. The suboptimal care patterns and systems-level barriers identified through our analyses highlight priority targets for quality improvement initiatives and policy interventions to reduce gaps in JIA care.