基于机器学习算法的重症缺血性脑卒中早期死亡预测效果评价

目的 评价支持向量机(SVM)、随机森林(RF)、极限梯度提升(XGBoost)3种机器学习算法与Logistic回归在重症缺血性脑卒中30天死亡结局的预测效果。方法 使用2008年至2019年MIMIC-IV数据库中符合纳排标准的2358例重症缺血性脑卒中患者资料，分别用SVM、随机森林、XGBoost3种机器学习算法与Logistic回归结合合成少数类过采样(SMOTE)技术建立早期死亡预测模型，并使用通过受试者工作特征曲线下面积(AUC)、准确度、F1-score、布里尔分数等指标评价预测效果。结果 SVM、随机森林、XGBoost与Logistic回归模型在原始类不平衡数据死亡预测中AUC值分别为0.78、0.81、0.84、0.83。应用SMOTE合成数据集后，SVM、随机森林、XGBoost与Logistic回归模型的AUC值分别为0.72、0.84、0.83、0.83。除SVM 外，机器学习算法与Logistic回归之间有相似的预测能力，但准确率、布里尔分数等优于Logistic回归，综合分类性能更优。结论 机器学习算法在重症缺血性脑卒中早期死亡预测中性能较传统方法更优，在解决重症患者预后预测研究问题中具有优势。

Prediction of early mortality among severe ischemic stroke patients based on machine learning model

Objecyive To evaluate the predictive efficiency of three machine learning (support vector machine [SVM], random forest, and extreme gradient boosting [XGBoost] )with logistic regression in prediction of 30-day mortality among severe ischemic stroke. Methods The data on 2358 patients with severe ischemic stroke who qualified for the criteria in the MIMIC-IV database from 2008 to 2019.SVM, random forest, XGBoost and logistic regression combined with SMOTE were used respectively to build early mortality prediction models. The prediction performance of models was evaluated by the area under the subject working characteristic curve (AUC), accuracy, F1-score and Brier score and more. Results The AUC values of SVM, random forest, XGBoost and logistic regression models using original class imbalance data were 0.78, 0.81, 0.84 and 0.83 respectively. After using SMOTE-based synthetic data, the AUC values of SVM, random forest, XGBoost and logistic regression models were 0.72, 0.84, 0.83 and 0.83 respectively. Except for SVM, the machine learning algorithms had similar predictive ability to logistic regression, but the accuracy and Brier score were better than logistic regression, and the overall classification performance was better. Conclusion Machine learning algorithms have good classification performance in early death prediction among ischemic stroke and are advantageous in addressing the prognosis prediction research problem in critically ill patients.