Machine learning differentiation of rheumatoid arthritis-Sjögren's syndrome overlap from Sjögren's syndrome with polyarthritis.

Objective: This study aimed to evaluate the utility of machine learning algorithms in differentiating rheumatoid arthritis-Sjögren's syndrome overlap (RA-SS) from Sjögren's syndrome with polyarthritis (SS-PA), and to identify key factors influencing diagnostic differentiation. Methods: This retrospective analysis included 106 RA-SS and 135 SS-PA patients randomized 7:3 into training and validation sets. Clinical, laboratory, and radiographic data were collected. Least Absolute Shrinkage and Selection Operator (LASSO) regression facilitated feature selection before constructing diagnostic models using four machine learning algorithms, with feature importance quantified through SHapley Additive exPlanations (SHAP). Results: The random forest algorithm demonstrated superior performance (AUC=0.854, 95% CI: 0.747-0.944) compared to other machine learning algorithms. SHAP analysis identified anti-CCP level, rheumatoid factor (RF) level, erosive joint count, anti-SSA/Ro60 antibodies, and C-reactive protein (CRP) as critical discriminating factors between RA-SS and SS-PA. Conclusion: The random forest algorithm demonstrates promising clinical potential for RA-SS and SS-PA differential diagnosis, with diagnostic efficiency surpassing traditional logistic regression (LR), offering a new approach for clinical differentiation.