神经网络对老年冠心病患者发生冠脉支架内再狭窄的初步研究预测

A preliminary Study of Neural Network To Predict Coronary In-Stent Restenosis in Elderly Patients with Coronary Heart Disease

目的探讨老年冠心病患者发生经皮冠状动脉介入治疗(percutaneous coronary intervention,PCI)术后支架内再狭窄(in-stent restenosis,ISR)的危险因素,并利用多层感知器神经网络分析、学习和预测冠状动脉ISR。方法随访93例有PCI手术史的老年患者(年龄≥65岁),用简单随机化分组抽取其中76例患者作为训练组,17例作为预测组。对93例患者的年龄、PCI术后年份、原有冠脉病变血管数、支架长度、支架内径、糖化血红蛋白(HbA1c)、低密度脂蛋白(LDL-C)、高密度脂蛋白(HDL-C),脂蛋白a、尿酸(UA)进行非条件Logistic回归分析,总结出有统计学意义的因素,并对预测组进行预测。再使用多层感知器神经网络学习训练组,建立与ISR之间的隐性联系模型,寻找重要的变量。最后,利用多层感知器神经网络预测预测组中17例老年冠心病患者PCI手术后是否会发生ISR,比较两种预测结果并进行评估。结果非条件Logistic回归分析结果表明,PCI术后ISR患者和无ISR患者,原植入支架的内径具有显著差异,支架内径越小,发生ISR概率越高(P=0.004);PCI术后随着时间的推移,发生ISR的概率有明显增高(P=0.012);存在多支病变的患者发生ISR比例明显高于单支病变的患者,两者具有显著差异(P=0.037);ISR患者的HbAlc水平明显增高,与无ISR的患者比较,两者具有显著差异(P=0.026)。多层感知器神经网络模型的变量重要程度表中,Hb A1c为最为重要,其次为脂蛋白a,第三为支架直径。用多层感知器神经网络预测测试组中会发生ISR的正确率为82.4%,高于逻辑回归预测的76.5%。多层感知器神经网络预测结果更令人满意。结论支架直径、多支病变、术后年份、HbA1c、脂蛋白a对冠状动脉ISR具有显著性影响。使用多层感知器神经网络模型预测冠状动脉ISR的发生正确率更高。

Objective To investigate the risk factors of in-stent restenosis(ISR)in elderly patients with coronary artery disease after percutaneous coronary intervention(PCI)and to analyze,learn and predict coronary artery ISR with multi-layer perceptron neural network.Methods 93 elderly patients(aged over 65)with a history of PCI were followed up,and 76 ones of them were randomly selected as the training group and 17 ones as the prediction group.The age of 93 patients,the years after PCI,the lesions of original coronary artery,the length of stent,the diameter of stent,level of HbA1 c,LDL-C,HDL-C,lipoprotein A and UA were analyzed by non-conditional logistic regression.The statistically significant factors were summarized and used for prediction of prediction group.The implicit connection model with ISR was established in order to look for important variables.Finally,the multi-layer perceptron neural network was used to predict whether in-stent restenosis would occur in the prediction group after PCI.The two prediction results were evaluated and compared each other.Results The results of non-conditional logistic regression analysis showed that there was significant difference in the diameter of original stent between the patients with and without in-stent restenosis after PCI.The smaller the diameter of stent,the higher the probability of in-stent restenosis(P=0.004).The probability of in-stent restenosis increased significantly with time after PCI(P=0.012).The incidence of restenosis in patients with multi vessel disease was significantly higher than that in patients with single vessel disease(P=0.037).The glycated hemoglobin level in patients with stent restenosis was significantly higher than that in patients without stent restenosis(P=0.026).HbA1 c was the most important variable in the multiplayer perceptron neural network model,followed by lipoprotein A and stent diameter.The accuracy of predicting the occurrence of ISR in the prediction group with multilayer perceptron neural network is 82.4%,which is higher than that of logistic regression prediction 76.5%.Conclusions Stent diameter,multivessel lesion,postoperative time,HbA1 c,lipoprotein A have significant effects on coronary ISR.The multilayer perceptron neural network model is more accurate in predicting the occurrence of coronary ISR.