基于RBF神经网络模型外科手术患者的手术部位感染因素预测及临床应用价值

Prediction of surgical site infection factors in surgical patients based on RBF neural network model and its clinical application value

目的应用神经网络建立外科手术患者手术部位感染(SSI)径向基函数神经网络(RBF)神经网络预测模型, 并探讨其临床应用价值。方法收集郑州大学第二附属医院2019年1月至2022年6月间39 321例外科手术患者为研究对象, 选取其中268例感染患者为试验组, 268例未感染患者为对照组分析相关影响因素和术后感染情况, 使用SPSS 25.0, 应用Logistic回归和神经网络分别建立Logistic回归模型以及RBF神经网络模型, 感染患者按4∶1划分为训练集和测试集, 比较2种模型的预测性能。用RBF神经网络模型对2022年7月至2022年11月5 600例外科手术患者进行预测, 根据此预测结果, 对每一例患者施行临床干预, 记录发生SSI的感染例数, 计算感染率, 并进行干预前后比较。结果共39 321例手术患者, 其中有268例患者发生手术部位感染, 感染发生率为0.68%;多元Logistic回归分析结果显示, 年龄、切口类型、预防性使用抗生素、是否患有基础性疾病、手术次数、手术时间、美国麻醉医师协会(ASA)病情分级和手术切口数是外科术后患者发生手术部位感染的独立影响因素;RBF神经网络模型训练集和测试集的准确度、精确度、召回率、F1分数以及AUC均高于Logistic回归模型, 性能指标方面的表现均优于Logistic回归模型, 各指标之间差异有统计学意义(P均<0.05);RBF神经网络模型对外科手术患者的预测及临床干预后手术部位感染率为0.32%, 与干预前的感染率比较, 临床干预效果比较显著, 差异有统计学意义(P<0.05)。结论 RBF神经网络模型在预测外科手术患者手术部位感染的预测精准度及效能高于Logistic回归模型, 为医疗诊断、治疗、护理和管理决策提供客观标准。

Objective The radial basis function neural network prediction model of surgical site infection was established by using neural network,and its clinical application value was discussed.Methods A total of 39321 surgical patients in a hospital from January 2019 to June 2022 were collected as research objects:268 infected patients were selected as the test group,and 268 uninfected patients were selected as the control group to analyze relevant influencing factors and postoperative infection.SPSS 25.0 was used to establish logistic regression model and radial basis function neural network model respectively by using logistic regression and neural network.The infected patients were divided into training set and test set according to 4∶1,and the prediction performance of the two models was compared.The radial basis function neural network model was used to predict 5600 surgical patients from July 2022 to November 2022.According to the prediction results,clinical intervention was carried out for each patient,the number of infections with surgical site infection was recorded,the infection rate was calculated,and the comparison before and after the intervention was made.Results A total of 39321 patients underwent surgery,268 of whom had surgical site infection(0.68%).Multivariate logistic regression analysis showed that age,incision type,prophylactic use of antibiotics,presence of basic diseases,number of operations,operation time,ASA disease grade and number of surgical incisions were independent influencing factors for postoperative patients to develop surgical site infection.The accuracy,precision,recall,F1 score and AUC of the training set and test set of radial basis function neural network model were higher than those of the Logistic regression model,and the performance indicators were better than those of the Logistic regression model,with statistically significant differences between the indicators(P<0.05).The prediction of radial basis function neural network model for surgical patients and the surgical site infection rate after clinical intervention were 0.32%.Compared with the infection rate before intervention,the clinical intervention effect was significant(P<0.05).Conclusion The prediction accuracy and efficiency of radial basis function neural network model in predicting surgical site infection of surgical patients are higher than those of Logistic regression model.It provides objective criteria for medical diagnosis,treatment,nursing and management decisions,and has good application prospects.