基于机器学习的ACS-Stacking预测模型

ACS-Stacking Prediction Model Based on Machine Learning

摘要: 急性冠脉综合征(Acute Coronary Syndrome, ACS)是威胁人类健康的重要疾病，其中急性心肌梗死的快速鉴别诊断技术仍需进一步研究。本研究包含了山西医科大学附属心血管病医院的813名患者的临床数据，由24个与人口统计学/合并症特征和住院并发症相关的预测变量描述。以“急性心肌梗死(Acute Myocardial Infarction, AMI)、不稳定心绞痛(Unstable Angina, UA)”二分类变量为目标变量，建立一个可解释性的机器学习(Machine Learning, ML)模型，确定显著相关指标来辅助临床医师对ACS患者进行快速有效的鉴别。训练并评估了这7种ML模型的性能，将在测试集中表现较好的Xgboost, Adaboost, Randomforest融合成表现最佳的可解释的Stacking融合模型(命名为：ACS-Stacking预测模型)。ACS融合预测模型实现了在测试集的AUC值为0.96562，在10-fold Cross-Validation下的准确率为89%。该模型有助于医生在临床诊断中结合模型预测结果、模型可视化和临床经验快速甄别出ACS患者。

Abstract: Acute Coronary Syndrome (ACS) is a significant disease that threatens human health, and the rapid differential diagnosis technology for acute myocardial infarction still requires further research. This study involved clinical data from 813 patients at the Cardiovascular Hospital of Shanxi Medical University, described by 24 predictive variables related to demographic/comorbidity characteristics and in-hospital complications. Using “Acute Myocardial Infarction (AMI) and Unstable Angina (UA)” as binary classification variables as the target variables, an interpretable machine learning (ML) model was established to identify significant related indicators to assist clinicians in making rapid and effective identification of ACS patients. The performance of these seven ML models was trained and evaluated, and the Xgboost, Adaboost, and Randomforest models that performed better in the test set were fused into the best-performing interpretable Stacking ensemble model (named: ACS-Stacking prediction model). The ACS ensemble prediction model achieved an AUC value of 0.96562 in the test set and an accuracy rate of 89% under 10-fold Cross-Validation. This model helps doctors quickly identify ACS patients in clinical diagnosis by combining model prediction results, model visualization, and clinical experience.