应用机器学习构建细菌性脓毒症的菌型预测模型

Application of Machine Learning to Construct the Bacterial Type Prediction Model for Bacterial Sepsis

目的:应用机器学习方法构建细菌性脓毒症患者的菌型预测模型,辅助医生进行病情严重程度的评估以及抗感染治疗。方法:选取MIMIC-Ⅲ数据库中的脓毒症患者199例,其血培养结果均为阳性的单一菌种,其中G^+菌117例,G^-菌82例。收集患者在血培养前检测的体重、白蛋白(ALB)、C-反应蛋白(CRP)、血小板(PLT)、中性粒细胞(NEUT),结合年龄作为主要研究变量,然后采用XGBoost算法构建菌型预测模型。结果:模型的灵敏度、特异性、准确率和AUC值分别是0.83、0.88、0.85和0.83。结论:基于XGBoost算法的G^+菌和G^-菌的预测模型可以预测细菌性脓毒症患者感染的菌型,从而辅助医生评估脓毒症患者的病情,指导抗菌用药。

Objective:To construct a predictive model of bacterial type in bacterial sepsis patients by machine learning,which assists doctors in assessing the severity of the disease and anti-infective treatment.Methods:A total of 199 patients with sepsis in the MIMIC-Ⅲdatabase were selected.The blood culture results were all positive single strains,including 117 patients with G^+bacteria and 82 patients with G^-bacteria.The body weight,albumin(ALB),C-reactive protein(CRP),platelet(PLT)and neutrophil(NEUT)detected before blood culture were collected,which combining with age were used as the main research variable.Then the XGBoost algorithm was used to construct the predictive model of bacterial type.Results:The sensitivity,specificity,accuracy and AUC values of the model were 0.83,0.88,0.85 and 0.83 respectively.Conclusion:The predictive model of G^+and G^-bacteria based on XGBoost algorithm can predict the bacterial type of infection in patients with bacterial sepsis,thus assisting doctors in assessing the condition of sepsis patients and guiding antibiotics.