Xuantong Guo

Xuanwu Hospital, Capital Medical University,
China

Abstract Title: Construction of Machine Learning Predictive Model for In-Hospital Mortality in Patients with Acute Coronary Syndrome

Biography: Dr. Xuantong Guo studied Clinical Medicine at the Nanchang University, China and graduated as BS. Med in 2019. During this period, she completed a bachelor's degree with a double major in Biomedical Science at the Queen Mary University of London, UK. After five years of clinical and scientific training in National Clinical Research Centre of Cardiovascular Diseases, Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, China, she received her MD degree in 2024. She is now working as a clinical doctor meanwhile carrying scientific research at Xuanwu Hospital, Capital Medical University.

Research Interest: Acute coronary syndrome (ACS) remains the leading cause of mortality among cardiovascular disease. Traditional risk prediction models for ACS are vastly restricted by the predictive performance and single serological indicator. Therefore, we utilized machine learning (ML) model based on blood routine and biochemical detection data to predict the in-hospital mortality in patients with ACS. Three ML models, including XGBoost, CatBoost and ResNet, were constructed and compared. Lastly, we investigated the predictive performance of novel indicators, such as inflammatory burden index (IBI), by incorporating them into the ML models. The model interpretability was then evaluated by SHAP. A total of 2800 patients with ACS were included (median age 61, IQR 50-70). The incidence of in-hospital mortality was 11.96% during a median follow-up of 8 days. Among the three ML models, the CatBoost exhibited the best performance (F1-Score 0.7389, Recall 0.8657, AUC-ROC 0.9560). Moreover, through stepwise incorporating clinically novel indicators into the model, we found that the IBI could improve the F1 Score by 0.88, leading to significantly enhanced predictive performance. The SHAP plot revealed that Killip classification, history of stenting, and smoking status were the three most important features for predicting in-hospital mortality in patients with ACS. This study indicates that machine learning is a promising tool for early prediction and risk stratification in ACS.