心脏外科术后切口感染危险因素识别及预测研究:基于多值Logistic模型和径向基神经网络算法

Study of indentifying and forecasting risk factors of cardiac surgery postoperative wound infection: based on multinomial logistic model and radial basis function neural network algorithm

目的对心脏外科术后切口感染的危险因素进行识别并对感染的可能性进行预测。方法对2 374例成人心脏外科手术的临床资料进行回顾性分析,通过结合多值Logistic模型与径向基神经网络算法(RBF)的识别与预测方法,发掘术后切口感染的主要因素,以期降低切口感染的发生。结果多值Logistic回归分析的估计结果表明,既往吸烟、有糖尿病史且使用胰岛素、肥胖、手术用血、术后出血量和术后是否进行再次开胸止血6个变量是引起切口感染的危险因素。进而将上述6种危险因素作为预测算法的输入变量,得到预测结果的ROC曲线和未经变量筛选的RBF预测结果的ROC曲线。经比较发现,通过筛选出的变量建立的RBF有更高的准确度,预测性能也更好。结论多值Logistic模型和RBF均是心脏外科术后切口感染危险因素分析的有效定量分析方法,RBF准确性更高,预测性能更好。

Objective To explore the risk factors of postoperative wound infection,forecast the possibility of infection and reduce threats towards patients because of the infection of incisional wound. Methods A retrospective study was performed on 2 374 cases with cardiac surgery,by combining multiple valued Logistic model and RBF neural network algorithm（ RBF） identification and prediction methods,explore the main factors of postoperative incision infection,in order to reduce the infection of incision. Results Multinomial estimation of logistic regression analysis results showed that smoking,history of diabetes,used insulin,obesity,surgical blood,whether thoracotomy was reused to stanch and the amount of postoperative bleeding were the risk factors that engendered the infection of incisional wound. Further more,making the above six risk factors as the input variables of the prediction algorithm,the ROC curve of the prediction results obtained and the ROC curve of radial basis function neural network algorithm without screening factors were compared. After selecting variables,the establishment of radial basis function neural network algorithm had a higher accuracy and better performance in prediction. Conclusion Multivalued Logistic model and RBF is cardiac surgery postoperative incision infection risk factor of effective quantitative analysis method,RBF has higher accuracy and better prediction performance.