| Objectives Data mining was used to predict the prognosis of intravenous thrombolytic therapy using recombinant tissue plasminogen activator(rt-PA)in patients with acute ischemic stroke.Logistic regression,decision tree,neural network and support vector machine were used to predict the prognosis of intravenous thrombolysis in patients with acute ischemic stroke.The model was evaluated by the indexes of authenticity,benefit,reliability and clinical application value,and the optimal prediction model was obtained.Methods 1 Patients who were clinically diagnosed as acute ischemic stroke in Linyi people's hospital of shandong province from April 2017 to September 2018,who received rt-PA intravenous thrombolytic therapy and met the inclusion criteria,and whose data were complete,were selected as the study subjects.Collect the basic information of the research object,medical history,time of onset to treatment,thrombolysis before blood pressure level,blood sugar level before thrombolysis,blood routine examination,blood biochemistry,imaging examination data,treatment,the United States national institutes of health stroke scale(NIHSS)scores and other case data,through the follow-up of 90 days after thrombolysis patients Rankin scale(mRS)score,with mRS score of 0 to 2 points is defined as the favourable prognosis,mRS score of 3 ~6 points is defined as the poor prognosis.2 The patient information from April 2017 to June 2018 was used to establish the prediction model.The input variables of the prediction model were determined according to the results of univariate analysis and multivariate analysis,combined with literature review and expert opinions.Use SPSS Modeler 14.1 software build Logistic regression model,the decision tree model,neural network model and support vector machine(SVM)model,compare the four kinds of prediction model,sensitivity,specificity,about yoden index,positive likelihood ratio and negative likelihood ratio,the area under the ROC curve(AUC),accuracy,Kappa value,positive predictive value and negative predictive value and other indicators,in order to obtain the optimal prediction model.3 To use in July 2018-September 2018 patients included in the model validation and use of the optimal models predict prognosis after thrombolysis in patients with 3 months,the predicted results were compared with the actual prognosis of patients after 90 days of follow-up,calculate the model prediction accuracy is used to verify the actual application effect of the model.Results 1A total of 630 patients were included in this study,including 522 hospitalized patients from April 2017 to June 2018,341 with favourable prognosis and 181 with poor prognosis.From July 2018 to September 2018,108 patients were hospitalized,with a good prognosis of 73 and a poor prognosis of 35.Univariate analysis of 522 patients from April 2017 to June 2018 showed that the age of the group with favourable prognosis was lower than that of the group with poor prognosis.Among the groups with favourable prognosis,hypertension,coronary heart disease,atrial fibrillation,high homocysteine,carotid stenosis ?50%,the proportion of bleeding complications,average systolic blood pressure,OTT,NIHSS score before and after thrombolysis,platelet volume,urea,lactate dehydrogenase and blood glucose were all lower than those in the groups with poor prognosis.The proportion of conscious patients in the favourable prognosis group washigher than that in the poor prognosis group.The proportion of patients in the group with favourable prognosis basically cured and improved at discharge,the average levels of erythrocyte,plasma total protein,triglyceride and serum potassium were all higher than those in the group with poor prognosis.The proportion of patients with OCSP type POCI and LACI type with favourable prognosis was higher than that of patients with poor prognosis,and the above differences were statistically significant(P<0.05).2The results of unconditioned Logistic regression analysis showed that carotid stenosis?50%,OCSP type PACI and TACI,poor discharge outcome,OTT length,high NIHSS score after thrombolysis and advanced age were risk factors for thrombolysis prognosis.3According to the results of multi-factor analysis,combined with literature search,clinical experience and guidelines for the diagnosis and treatment of acute ischemic stroke in China,blood glucose and systolic blood pressure were increased.A total of 8 factors were selected as the input variables of the prediction model.4For training set data,the accuracy of neural network model,support vector machine model,Logistic regression model and decision tree model was 91.01%?85.11%?87.64% and 86.52%;the yoden index was 0.79,0.64,0.70 and 0.66;the Kappa index was 0.799,0.661,0.719 and 0.690;AUC(95%CI)was 0.874(0.835,0.907),0.820(0.776,0.859),0.849(0.807,0.884)and0.831(0.788,0.868).The AUC of the neural network prediction model was higher than that of other models,and the difference was statistically significant(P<0.05).For the test set data,the accuracy of neural network model,support vector machine model,logistic regression model and decision tree model were 90.96%,81.93%,86.75% and 83.73%;the yoden index was 0.77,0.57,0.66 and 0.59;the Kappa index was 0.694,0.589,0.691 and0.622.AUC(95%CI)was 0.835(0.769,0.888),0.787(0.717,0.847),0.828(0.762,0.882),and 0.797(0.727,0.855).There was no statistically significant difference in AUC between the four models(P>0.05).5Using the established based on neural network of acute ischemic stroke prognosis prediction model,intravenous thrombolysis in July 2018-September 2018 in 108 to predict the prognosis of patients,and intravenous thrombolysis in patients with 90 days of the actual outcome comparison,results show that the model prediction accuracy is 87.04%,the sensitivity was 87.67%,specificity was85.71%,about an index of 0.73,Kappa index is 0.713,AUC(95% CI)was 0.895(0.829,0.961).Conclusion carotid stenosis ?50%,OCSP type PACI and TACI,poor discharge outcome,OTT long,high NIHSS score after thrombolysis,advanced age,high blood glucose and high systolic blood pressure are risk factors for the prognosis of venous thrombolysis in acute ischemic stroke.In the logistic regression,decision tree,neural network and support vector machine(SVM)four types of acute ischemic stroke of intravenous thrombolysis prognosis prediction model,the neural network model is optimal,the prediction effect of practical application is also good,based on neural network of acute ischemic stroke of intravenous thrombolysis prognosis prediction can provide the basis for treatment in clinical practice.Figure 4;Table 34;Reference 168... |
PDF Full Text Request