基于机器学习的重症股骨颈骨折患者死亡风险预测模型构建和验证

Construction of predictive models for death risk of patients with femoral neck fracture in intensive care units based on machine learning

关键词:  机器学习  股骨颈骨折  院内死亡

Key words:  machine learning  femoral neck fracture  hospital mortality

摘要:

目的采用机器学习方法，构建重症股骨颈患者院内死亡预测模型，辅助临床医生尽早进行临床决策。方法使用公开数据库——重症监护医疗信息市场(Medical Information Mart for Intensive Care, MIMIC)Ⅲ中入住ICU的股骨颈骨折患者信息进行回顾性分析。采用SMOTE算法平衡数据集后，按7：3随机划分训练集和验证集。以患者发生院内死亡作为结局，分别构建随机森林、XGBoost和BP神经网络预测模型。模型性能通过受试者工作特征曲线下面积(area under the receiver operating characteristic curve, AUROC)、准确率、精确率、灵敏度和特异度进行评估，并与传统Logistic模型对比验证模型的预测价值。结果共纳入366例股骨颈骨折患者，其中院内死亡48例。按死亡组：生存组=1：1平衡数据集后共获得636例患者数据。3种机器学习模型具有较高的预测准确性，随机森林、XGBoost和BP神经网络的AUC分别为0.98、0,97和0.95，预测性能均高于传统Logistic回归模型。对特征变量重要性进行排序，得到对预测患者院内死亡风险有意义的前10个特征变量为: 维生素D、乳酸脱氢酶、肌酐、SAPSⅡ评分、血清钙、入住ICU时长、白细胞、年龄、BMI和肌酸激酶。结论使用机器学习构建的死亡风险评估模型对预测重症患者的院内死亡有着积极的意义，并为减少院内死亡，改善患者预后提供有效的依据。

Abstract:

ObjectiveTo construct predictive models for nosocomial death risk of patients with femoral neck fracture in intensive care unit based on machine learning. MethodsData of patients with femoral neck fracture were obtained from an open database the Medical Information Mart for Intensive Care(MIMIC) Ⅲ. After balancing the dataset using the SMOTE algorithm, patients were randomly divided into a training set and a testing set in a ratio of 7：3 for the development and validation of the prediction model. Random forest, XGBoost, and BP neural network prediction models were constructed with the nosocomial death as the outcome. Model performance was assessed using the receiver operating characteristic curve(ROC). The predictive value of the models was verified in comparison to the traditional logistic model. ResultsA total of 366 patients with femoral neck fractures were collected, including 48 cases of in-hospital death. Data from 636 patients were obtained by balancing the dataset with the in-hospital death group：survival group as 1：1. The three machine learning models exhibited high predictive accuracy, and the area under the ROC curve(AUC) of the random forest, XGBoost, and BP neural networks were 0.98, 0.97, and 0.95, respectively；all with higher predictive performance than the traditional logistic regression model. Ranking the importance of the feature variables, the top ten feature variables that were meaningful for predicting the risk of in-hospital death of patients were vitamin D, lactate, creatinine, the Simplified Acute Physiology Score(SAPS) Ⅱ score, calcium, loss in ICU, white blood cell counts, age, BMI and CK. ConclusionDeath risk assessment models constructed using machine learning are effective for predicting the in-hospital mortality of patients with femoral neck fracture in ICU, which may provide a valid basis for reducing in-hospital mortality and improving patient prognosis.