XGBoost模型分析院内心脏骤停患者恢复自主循环的特征

Feature analysis for return of spontaneous circulation in IHCA patients based on XGBoost model

目的:应用XGBoost模型分析院内心脏骤停患者的临床数据,探索影响恢复自主循环(return of spontaneous circulation,ROSC)的关键特征。方法:回顾性分析2021年1月-2024年4月在南华大学附属第一医院住院的心脏骤停患者504例。根据心肺复苏结局分为ROSC组和未恢复组。收集患者一般资料、生化指标、生命体征、心电图特征、酸碱平衡参数及临床病史等数据。采用XGBoost模型分析各特征对ROSC的预测贡献。结果:295例院内心脏骤停患者被纳入分析,ROSC成功率为44.7%。XGBoost模型特征重要性分析显示,舒张压是预测ROSC的最重要特征。其他重要特征依次为白蛋白、凝血酶时间、心率和中性粒细胞与淋巴细胞比值。XGBoost模型预测院内心脏骤停患者ROSC的AUC为0.79,预测ROSC的精准率为83%,召回率为71%,F1得分为77%,假阳率为50%,假阴率为29%。结论:XGBoost模型在预测院内心脏骤停患者ROSC方面具有良好的诊断一致性,对临床决策支持具有潜在价值。舒张压、心率等心脏功能指标及特定生化指标对预测院内心脏骤停患者ROSC具有重要意义。

Objective:To analyze clinical data of in-hospital cardiac arrest patients using the XGBoost model and explore key features influencing the return of spontaneous circulation(ROSC).Methods:A retrospective analysis was conducted on 504 hospitalized patients who experienced cardiac arrest at the First Affiliated Hospital of University of South China from January 2021 to April 2024.Patients were divided into ROSC and non-ROSC groups based on cardiopulmonary resuscitation outcomes.Data collected included general information,biochemical indicators,vital signs,ECG characteristics,acid-base balance parameters,and clinical history.The XGBoost model was employed to analyze the predictive contribution of various features to ROSC.Results:Two hundred and ninty-five patients with IHCA were included in the analysis,with a ROSC success rate of 44.7%.The feature importance analysis of the XGBoost model shows that diastolic blood pressure is the most important predictor of ROSC.Other important features in order are albumin,thrombin time,heart rate,and the neutrophil-to-lymphocyte ratio.The AUC of the XGBoost model for predicting ROSC in in-hospital cardiac arrest patients is 0.79,with a precision of 83%,recall of 71%,F1 score of 77%,false positive rate(FPR) of 50%,and false negative rate(FNR) of 29%.Conclusion:The XGBoost model demonstrates good diagnostic consistency in predicting ROSC in in-hospital cardiac arrest patients,with potential value for clinical decision support.Indicators of cardiac function,such as diastolic blood pressure and heart rate,as well as specific biochemical markers,play an important role in predicting ROSC in in-hospital cardiac arrest patients.