基于人工神经网络的老年单肺通气患者术后谵妄预警模型的构建与评价

The construction and evaluation of a postoperative pelirium early warning model for elderly single-lung ventilation patients based on artificial neural networks

目的 构建基于人工神经网络(ANN)的老年单肺通气(OLV)患者术后谵妄(POD)的风险预警模型并评价其预警效能。方法纳入542例接受OLV的老年患者,其中POD组242例,非POD组300例。数据集按照7:3随机分为训练集(POD=169例,非POD=210例)和验证集(POD=73例,非POD=90例),利用训练集分别构建ANN和Logistic回归风险预警模型,并在验证集中评价两模型预警效能。结果两模型的Hosmer and Lemeshow检验差异均无统计学意义(P>0.05);ANN风险预警模型的曲线下面积、灵敏度、特异度、阴性预测值和净重新分类指数高于Logistic回归风险预警模型(P<0.05);ANN风险预警模型的约登指数、F1 score、阳性似然比、阴性似然比优于Logistic回归风险预警模型;两模型准确率和阳性预测值比较差异无统计学意义(P>0.05)。结论 基于ANN的风险预警模型对老年OLV患者POD的预警效能优于Logistic回归风险预警模型。

Objective To construct an artificial neural network(ANN)-based risk warning model for postoperative delirium(POD)in elderly patients undergoing one-lung ventilation(OLV)and to evaluate its warning efficacy.Methods 542 elderly patients undergoing OLV were included,including 242 in the POD group and 300 in the non-POD group.The original data set was divided into a training set(POD=169,non-POD=210)and a validation set(POD=73,non-POD=90)according to 7:3.ANN and Logistic regression risk warning models were constructed using the training set,and the warning efficacy of the two models was evaluated in the validation set.Results The Hosmer and Lemeshow tests of the two models were not statistically different(P>0.05).The area under the curve,sensitivity,specificity,negative predictive value and net reclassification index of the ANN risk warning model were higher than those of the Logistic regression risk warning model(P<0.05).The Jorden index,F1 score,positive likelihood ratio and negative likelihood ratio of the ANN risk warning model were higher than those of the Logistic regression risk warning model(P<0.05).Positive likelihood ratio and negative likelihood ratio of the ANN risk warning model were better than those of the Logistic regression risk warning model,there was no statistical difference between the accuracy and positive prediction values of the two models(P>0.05).Conclusion The ANN-based risk warning model is more effective than the Logistic regression risk warning model for POD in elderly OLV patients.