A-256 Prediction Model of Systemic Lupus Erythematosus Based on Machine Learning Algorithms

Abstract

This research developed disease prediction models employing six distinct machine learning algorithms. Through the analysis of variations in platelet aggregation rates between patients diagnosed with systemic lupus erythematosus(SLE) and healthy control subjects, as determined by blood coagulation assays, diagnostic markers with significant predictive value for systemic lupus erythematosus were identified. Employing a cross-sectional study design, retrospective data were gathered from 123 patients diagnosed with SLE who attended Sichuan Provincial People's Hospital between July 2022 and July 2025, alongside data from 64 healthy control subjects. Platelet aggregation rates, induced by four agonists—adenosine diphosphate (ADP), arachidonic acid (AA), epinephrine (EPI), and collagen (COL)—were quantified utilizing optical transmission turbidimetry. Subsequently, six diagnostic models for SLE were constructed, and their diagnostic performance was systematically assessed. The findings from the multivariate logistic regression analysis revealed that both AA and ADP serve as independent risk factors for patients with SLE (P < 0.05). Among the predictive models evaluated, the XGboost model exhibited the highest area under the curve (AUC = 0.984, 95% confidence interval: 0.969–1.000). Furthermore, DCA curve indicated that the XGboost model provided the greatest net benefit across the examined threshold probabilities. Feature importance assessment within the XGboost framework identified AA as the most significant predictor of SLE, followed sequentially by COL, ADP, and EPI. The diagnostic model for SLE patients, developed utilizing the XGboost algorithm, exhibits robust predictive performance. When integrated with clinical features, this model has the potential to enhance diagnostic accuracy and efficiency, thereby presenting significant value for clinical application.