2406: Machine Learning-Based Artificial Intelligence in Systemic Lupus Erythematosus: A Systematic Review of Outcome Prediction and Patient Stratification

Background/Purpose: Systematic lupus erythematosus (SLE) is an autoimmune disease with prognostic challenges due to its cyclic diversity. Artificial intelligence and machine learning (ML) has emerged as a powerful tool for analyzing multifactorial SLE data. This systematic review aims to investigate the current use of ML in predicting clinical outcomes and stratifying SLE patients.

Methods: A comprehensive search of PubMed, Scopus, Web of Science, and the Cochrane Library was conducted for studies published up to May 2025 evaluating the use of machine learning for stratifying SLE patients and determining prognosis. Extracted data included study design, country, recruitment period, sample size, ML application and algorithms, study aims, inclusion criteria, and main results.

Results: Seventeen studies applying machine learning to 16,483 SLE patients were included, with 81% of participants being female. The majority of studies (58.8%) originated from the USA. Random Forest was the most frequently used algorithm (41.1%), followed by recurrent neural networks and support vector machines. Main applications included predicting disease activity/flares, lupus nephritis, hospitalization risk, and molecular subtypes. Model performance, assessed by area under the ROC curve in nine studies, ranged from 0.595 to 0.9375. Sensitivity and specificity, reported in three studies, ranged from 0.75 to 82.1 and 0.75 to 89.6, respectively.

Conclusion: Machine learning demonstrates strong performance in analyzing complex SLE datasets, enabling accurate prediction of prognosis, disease activity, hospitalization risk, and treatment response. These models consistently outperform traditional approaches, supporting more personalized care and clinical decision-making for SLE patients across diverse clinical scenarios.