Systemic Lupus Erythematosus Detection with Machine Learning Models

Abstract

Systemic Lupus Erythematosus (SLE) is a chronic autoimmune disease characterized by immune dysregulation and multisystem inflammation, making diagnosis challenging because of its diverse clinical manifestations and overlap with other autoimmune disorders. Delayed diagnosis can lead to irreversible organ damage, highlighting the need for tools that improve diagnostic accuracy. This study evaluated the use of a supervised machine learning model to distinguish SLE from other autoimmune diseases using routinely collected clinical and laboratory data. A publicly available synthetic dataset containing over 10,000 patients with autoimmune diseases was used to train and evaluate the model. Features were selected based on the 2019 European Alliance of Associations for Rheumatology/American College of Rheumatology (EULAR/ACR) classification criteria and included demographic characteristics, laboratory biomarkers, and clinical symptoms. Model performance was assessed using a 70:30 train-test split and 10-fold cross-validation. Evaluation metrics included accuracy, precision, recall (sensitivity), specificity, F1 score, and the area under the precision-recall curve (AUPRC). The train-test split model achieved an accuracy of 99.73%, precision of 88.59%, recall of 85.15%, specificity of 99.90%, and an F1 score of 85.35%. The 10-fold cross-validation model demonstrated comparable performance, with an average accuracy of 99.74%, precision of 83.40%, recall of 91.53%, specificity of 99.83%, and an F1 score of 86.25%. The model also achieved an AUPRC of 0.9166, indicating strong discrimination despite the highly imbalanced dataset. These findings suggest that machine learning models trained on clinically relevant laboratory and symptom data can accurately identify SLE and may serve as valuable clinical decision-support tools. Additional validation using independent real-world patient datasets is required before clinical implementation.