基于连续生理指标的脓毒症休克早期预警机器学习模型研究

Machine learning model for early warning of septic shock based on continuous physiological indexes

目的:提出一种急危重症患者中脓毒症休克的预警模型。方法:在重症监护医学信息数据库第4版(MIMIC-IV)中筛选出符合条件的患者,提取患者的心率(HR)、呼吸频率(RR)、血氧饱和度(SpO_(2))和平均动脉压(MAP)4项连续生理指标的时间序列数据,然后将数据划分到时间窗口中。患者的一般特征(年龄、性别、体重)及生存指标(心率、呼吸频率、血氧饱和度和平均动脉压4个连续指标)分别按照线性和非线性参数进行统计分析。采用多种机器学习方法建立预测模型,预测患者在所选的时间窗口内是否会发作脓毒症休克。结果:随机森林模型预测分类准确率最高为85.16%,敏感度和特异度达到了56.00%和99.05%,接受者操作特征曲线下面积(AUROC)最高为0.85。AdaBoost模型预测的敏感度最高为58.00%。结论:随机森林模型具有高准确率、敏感度和特异度;此脓毒症休克预警模型比以往的模型预测准确度更高,预测时间更加提前。危重患者监护中可实时获得上述指标,因此通过脓毒症休克的实时动态预警,起到预测脓毒症休克的早期预警的作用,为临床决策支持提供参考。

Objective To propose an early warning model for septic shock in acute and critically ill patients.Methods The time series data of heart rate(HR),respiratory rate(RR),oxyhemoglobin saturation(SpO2)and mean arterial pressure(MAP)of four continuous physiological indexes were extracted from the Medical Information Mart for Intensive Care-IV(MIMIC-IV)database of critical care medicine,and then divided into time windows.The general characteristics(age,sex and weight)and survival indexes(continuous indexes of HR,RR,SpO_(2) and MAP)of patients were statistically analyzed according to linear and nonlinear parameters.Several machine learning methods were used to establish the prediction model to predict whether septic shock would occur in the selected time window.Results The maximum accuracy rate of predication and classification of the random forest model was 85.16%,the sensitivity and specificity were 56.00%and 99.05%,and the maximum area under the receiver operating characteristic curve(AUROC)was 0.85.The maximum sensitivity of AdaBoost model predication is 58.00%.Conclusion The random forest model has high accuracy,sensitivity and specificity.This septic shock early warning model has higher prediction accuracy and earlier prediction time than the previous models.The above indexes can be obtained in real time during the monitoring of critically ill patients.Therefore,the real-time dynamic early warning of septic shock can play the role of early warning of septic shock and provide reference for clinical decision support.