Abstract
OBJECTIVE: SLE is a chronic autoimmune disease with immune complex deposition in various organs, causing inflammation. The Systemic Lupus Erythematosus Disease Activity Index 2000 assesses disease severity but is subjective. This study aimed to construct a machine learning model based on objective laboratory indicators to assess SLE disease activity. METHODS: A retrospective study was conducted on 319 patients with SLE, collecting their clinical characteristics and laboratory indicators as model-building indicators. Multiple machine learning algorithms were employed to construct models for assessing SLE disease activity. RESULTS: The patients were divided into two cohorts, cohort 1 used as the training set to build the machine learning models and cohort 2 for external validation. Six laboratory indicators, including anti-dsDNA (IFT), quantitative anti-dsDNA, neutrophils, globulin, proteinuria and NK cells, were selected to construct the SLE disease activity evaluation model. The XGBoost model demonstrated superior performance in distinguishing active SLE, with an area under the receiver operating characteristic curve of 0.934, accuracy of 0.925, sensitivity of 0.969, specificity of 0.750 and F1 score of 0.954. CONCLUSIONS: This pioneering machine learning model, using objective laboratory indicators, enhances clinical feasibility and provides a novel method for assessing SLE disease activity, that may enable timely evaluation of SLE activity, facilitating preparation for treatment and prognosis.