基于深度神经网络的院内感染风险预测探讨

Exploration of Nosocomial Infection Risk Prediction Based on Deep Neural Networks

目的建立院内感染风险预测模型,并分析其可行性,助力医院院内感染管理。方法选取76564例住院患者作为研究对象,按照7∶3的比例划分为训练集和测试集。根据住院期间是否发生感染分为感染组和非感染组,分析训练集两组临床相关资料,采用二元logistic回归模型和深度神经网络(DNN)模型分析院内感染发生风险的相关影响因素并构建预测模型。结果训练集中626例发生感染,感染率为1.17%,测试集中289例发生感染,感染率为1.26%。受试者特征曲线(ROC)显示,DNN模型预测发生院内感染风险的曲线下面积(AUC)显著高于二元logistic回归模型(0.978 vs 0.686)。结论基于DNN的院内感染风险预测模型具有较高的准确性,有助于早期识别院内感染高危患者。

Objective To explore the construction of nosocomial infection risk prediction model and analyze its feasibility.Methods 76,564 hospitalized patients were selected as the research objects,and the training set and test set were divided according to the ratio of 7∶3.According to whether there was infection during hospitalization,it was divided into infection group and non-infection group.The clinical related data of the two groups were analyzed.The related influencing factors of the risk of nosocomial infection were analyzed by binary logistic regression model and deep neural network model,and the prediction model was constructed.Results 626 cases of infection occurred in the training set,and the infection rate was 1.17%.289 cases of infection occurred in the test set,and the infection rate was 1.26%.ROC curve showed that the AUC of DNN model for predicting the risk of nosocomial infection was significantly higher than that of binary logistic regression model(0.978 vs 0.686).Conclusion The nosocomial infection risk prediction model constructed based on DNN has high accuracy and is helpful for early identification of high-risk nosocomial infection patients.