Study On Recurrence Factors And Prediction Of First-episode Ischemic Stroke Patients After Discharge For One Year By Machine Learning

Objective:To explore the influencing factors of recurrence of first-episode ischemic stroke,construct the recurrence risk prediction model of first-episode ischemic stroke patients,provide scientific reference for clinical decision-making,and reduce the recurrence risk of first-episode ischemic stroke patients.Methods:Patients who were hospitalized in the third class hospital from May 2018 to may2020 and diagnosed with first-episode ischemic stroke were the subjects of the study.The demographic data,past medical history,personal history,family history,blood pressure,heart rate,blood routine,blood biochemistry and NIHSS score were collected,and the recurrence within 1 year after discharge was recorded through follow-up.Through univariate analysis,the variables with statistical significance are selected,and the variables with statistical significance are included in the model.The recurrence risk prediction models are constructed by using logistic regression,random forest,support vector machine and extreme gradient lifting decision tree algorithm on the training set data,and then the data of the validation set are used to verify each model.By comparing the sensitivity,specificity,Jordan index AUC and other indicators to evaluate the prediction effect of each model on the recurrence risk of patients with first-episode ischemic stroke after discharge for 1 year,so as to select the best recurrence risk prediction model.Results:A total of 1927 patients with first-episode ischemic stroke were included in this study.According to the recurrence within one year after discharge,they were divided into 1791 cases in the non recurrence group and 136 cases in the recurrence group.The recurrence rate was 7.05%.Multivariate logistic regression analysis showed that age(OR=1.023,P=0.009,95%CI:1.006～1.041),history of hyperlipidemia(OR=2.160,P=0.006,95%CI:1.246～3.743),admission NIHSS score ≥ 16(OR=1.649,P =0.036,95%CI:1.032～2.635)and low-density lipoprotein cholesterol(OR=5.227,P=0.014,95%CI:1.389～19.666)were the risk factors for recurrence within 1 year after discharge.The ROC curve analysis of the recurrence risk prediction model for patients with first-episode ischemic stroke within one year after discharge according to four machine learning algorithms shows that the AUC value of XGBoost algorithm is 0.943(95%CI:0.902～0.977),which is about 32 percentage points higher than that of logisitc regression model of 0.619(95%CI:0.573～0.662),and about 2 percentage points higher than that of random forest model of 0.924(95%CI:0.883 ～ 0.952),It is about 14 percentage points higher than 0.799(95%CI:0.754 ～0.843)of SVM model.Conclusion:Age,history of hyperlipidemia,NIHSS score ≥ 16,low-density lipoprotein cholesterol,hemoglobin,creatinine,high-density lipoprotein cholesterol,blood glucose and monocyte count were the main risk factors for recurrence within 1 year after discharge;The risk prediction model of recurrence within 1 year after discharge of patients with first-episode ischemic stroke based on XGBoost algorithm has the best performance.