Objective:

To determine if non-component clinical variables can classify JADAS27-defined disease activity strata in juvenile idiopathic arthritis (JIA) without including the four JADAS27 components.

Approach:

• Study Design: A retrospective multi-center study involving 800 JIA patients from CARRA, PRCSG, and LHTCM, excluding JADAS27 components from model inputs.
• Machine Learning Models: Four machine learning algorithms were trained using 13 non-component variables, with external validation performed on an independent cohort.
• Performance Metrics: Model performance was evaluated using accuracy and macro AUC, with sensitivity analyses conducted to assess the impact of proxy variables.

Key Findings:

• SVM achieved the best external validation performance with an accuracy of 0.731 (95% CI 0.657–0.797) and macro AUC of 0.918 (95% CI 0.887–0.945).
• Class-wise recall was highest for inactive (82.4%) and high activity (77.4%).
• Removing proxy variables reduced accuracy to 0.619 (95% CI not provided) and macro AUC to 0.843 (95% CI not provided), indicating independent contributions from both proxy and non-proxy features.

Interpretation:

Non-component clinical features can stratify JADAS27-defined disease activity, providing an alternative for risk stratification when formal scoring is impractical.

Limitations:

• The study is retrospective and relies on existing multi-center datasets.
• No formal a priori sample size calculation was performed.

Conclusion:

This approach supports early risk stratification in JIA when formal JADAS27 scoring is unavailable.