老年慢性心力衰竭患者院内发生急性心力衰竭风险预测模型构建与比较

Construction and comparison of models for predicting in-hospital risk of acute exacerbation in elderly patients with chronic heart failure

目的构建老年慢性心力衰竭(chronic heart failure,CHF)患者院内发生急性心力衰竭(acute heart failure,AHF)的风险预测的最佳模型。方法选取2018年1月至2020年12月北京大学第三医院收治的所有CHF患者,根据院内是否发生AHF,分为发生组(55例)和未发生组(564例)。采用支持向量机、随机森林、XGBoost法以及Logistic 4种方法构建预测模型。结果本研究使用MV值与卡方检验筛选出的老年CHF患者院内发生AHF的危险因素包括体重、入院心率、收缩压、性别、职业和静脉利尿剂;使用支持向量机、随机森林、XGBoost和Logistic 4种方法构建了院内发生AHF预测模型,预测准确率分别达到了91.92%、94.16%、94.27%和90.73%,曲线下面积(area under curve,AUC)值分别为0.6280、0.7349、0.9350和0.5505,以XGBoost的准确率最高,而Logistic模型的AUC值最低,判别能力较差。结论本研究构建的4种预测模型,以XGBoost构建的模型准确率最高,同时具有较好的敏感性和模型判别能力。

Objective To construct an optimal model for predicting the risk of in-hospital acute exacerbation in elderly patients with chronic heart failure(CHF).Method All patients with CHF admitted to a tertiary hospital in Beijing from January,2018 to December,2020 were selected,55 cases with in-hospital acute exacerbation and 564 cases without acute exacerbation.Support vector machine,random forest,XGBoost method and Logistic were used to construct the prediction models.Result Risk factors for in-hospital acute exacerbation in elderly patients with CHF screened in this study using MV with chi-square test included weight,heart rate at admission,systolic blood pressure,gender,occupation and intravenous diuretics.Prediction models for in-hospital acute exacerbation were constructed using the four methods of support vector machine,random forest,XGBoost,and Logistic,and the prediction accuracy reached 91.92%,94.16%,94.27%and 90.73%,with area under curve(AUC)values of 0.6280,0.7349,0.9350 and 0.5505,respectively.XGBoost had the highest accuracy,while the logistic model had the lowest AUC value and poor discriminative power.Conclusion Among the four prediction models constructed in this study,the model constructed by XGBoost has the highest accuracy,as well as better sensitivity and model discrimination.