Improving the Diagnosis of Systemic Lupus Erythematosus with Machine Learning Algorithms Based on Real-World Data

Abstract

This study addresses the diagnostic challenges of Systemic Lupus Erythematosus (SLE), an autoimmune disease with a complex etiology and varied symptoms. The ANA (antinuclear antibody) test, currently the primary diagnostic tool for SLE, exhibits high sensitivity but low specificity, often leading to inaccurate diagnoses. To enhance diagnostic precision, we propose integrating machine learning algorithms with existing clinical classification guidelines to improve SLE diagnosis accuracy, potentially reducing diagnostic errors and healthcare costs. We analyzed real-world data from a cohort of 24,990 patients over a 10-year period at the hospitals, excluding those previously diagnosed with SLE. Patients were categorized into three groups: negative ANA, positive ANA with non-SLE, and positive ANA with SLE. Feature selection was conducted to identify key factors influencing SLE diagnosis, and machine learning algorithms were employed to develop the CDSS. Performance analysis of three machine learning algorithms—decision tree, random forest, and gradient boosting—based on feature sets of 10, 20, and all available features revealed accuracy rates of 70%, 88%, and 87%, respectively, for the 20-feature set. The proposed system, utilizing real-world medical data, demonstrated modest performance in SLE diagnosis, highlighting the potential of machine learning-based CDSS in real clinical settings.