| Objective:To explore the influencing factors for the prognosis of patients with ischemic stroke,and to build models to predict the prognosis of patients with ischemic stroke,so as to provide a scientific basis for clinical decision making and improve the outcome of patients with ischemic stroke.Methods:Inpatients who were diagnosed with ischemic stroke in the Second Affiliated Hospital of Nanchang University,the First Affiliated Hospital of Nanchang University,Ganzhou People’s Hospital of Jiangxi Province and other 11 III grade hospitals from April 2019 to July 2020 were enrolled.The information of demographical information,medical history,personal history,family history,blood pressure,blood routine index,blood biochemical index and the National Institute of Health Stroke Scale(NIHSS)score at admission,antiplatelet drug and other data were collected,and the Modified Rankin scale(m RS)score at 90 days after discharge was recorded through follow-up.The input variables of prediction models were the indictors with statistical significance in univariate analyses.Logistic regression model,random forest model and BP neural network model were trained by using the training set and compared the performance of three machine learning models by sensitivity,specificity,area under the curve and other indexes based on the test set.Results:A total of 1378 inpatients diagnosed with ischemic stroke were included which1098 of them were defined as favorable outcome according to the m RS score at 90 days after discharged and 280 of them were defined as poor outcome.In logistic regression model,age,smoking history,antiplatelet drugs,NIHSS score at admission and the level of alkaline phosphatase were risk factors influencing the prognosis within 90 days of ischemic stroke(P<0.5);in random forest model,the top five influencing factors for the prognosis were age,the level of alkaline phosphatase,NIHSS score at admission,mean corpuscular volume and myoglobin;in BP neural network model,the top five influencing factors were NIHSS score at admission,neutrophil count,age,smoking history and antiplatelet drugs;the prediction performance of three machine learning models was ranked in order of the AUC value :BP neural network model(0.90,95%CI : 0.886～0.910),random forest model(0.870,95%CI:0.867～0.874),logistic regression model(0.750,95%CI:0.710～0.785).Conclusion:Age,alkaline phosphatase,smoking history,NIHSS score at admission,antiplatelet drugs,neutrophil count,mean erythrocyte volume and myoglobin were independent risk factors affecting the prognosis of patients with ischemic stroke;The prediction performance of random forest model and BP neural network model was better than logistic regression model,whose sensitivity,specificity and Youden index were all higher than logistic regression model. |
