Predicting Therapeutic Outcomes in Rheumatoid Arthritis Using Explainable Machine Learning on Clinical Data

Rheumatoid arthritis (RA) is a chronic autoimmune disorder marked by persistent inflammation, progressive joint damage, and diminished quality of life. While recent AI research has emphasized image-based RA diagnosis, there remains a critical need to utilize structured clinical data for predicting treatment outcomes. This study introduces an explainable machine learning framework to identify key clinical predictors of therapeutic success in RA. Using a real-world dataset of 154 patients undergoing biologic therapy, multiple models including XGBoost and LightGBM were trained on selected clinical features. The best-performing model, XGBoost, achieved an AUC of 0.86 and accuracy of 82% using non-imaging clinical variables. SHAP-based explainability revealed that disease activity (DAS28), CRP levels, and methotrexate use were among the most influential factors in predicting outcomes. These findings demonstrate that interpretable, data-driven models using readily available clinical data can effectively support personalized treatment strategies and informed clinical decision-making in RA management.