Abstract 4363666: A Comprehensive Study on Machine Learning Models Combining with Oversampling for One-year Persistent Coronary Artery Aneurysm in Kawasaki Disease

Background: Coronary artery aneurysm (CAA) is a major adverse cardiac complication of Kawasaki Disease (KD). Besides, persistent CAA serves as a risk factor for myocardial infarction in adulthood. Aims: This study aimed to utilize machine learning techniques for the purpose of developing a highly accurate predictive model tailored to forecast the one-year persistence of CAA in patients with KD. Methods: The data utilized in this study were collected from a tertiary-level hospital in Southwest China between June 2014 and June 2024. To address the issue of imbalanced data, the oversampling algorithm synthetic minority over-sampling technique (SMOTE) was applied to improve the model. Results: Five hundred and twenty-six clinical records were collected in our study. Following data pre-processing and feature selection, 8 of the 36 features were used to build models, including duration of fever (day), the size of CAA, time of initial IVIG (day), lymphocyte count, platelet count, plateletcrit, aspartate aminotransferase and creatine kinase MB isoenzyme to creatine kinase ratio. Nine machine learning models were applied to predictive learning. The best model (Random Forest) achieved 89.95% sensitivity, 81.75% specificity, 85.85% accuracy, and 0.9405 AUC. Conclusions: The proposed model can assist clinicians in accurately predicting the one-year prognosis of KD children with CAA.