The Study Of Prediction Model Of Ischemic Stroke Recurrence Based On Machine Learning

Objective:Our aim was to explore the influencing factors for the recurrence of patients with ischemic stroke within 1 year after the stroke,and establish models to predict the 1 year recurrence for the first-time ischemic stroke,in order to provide a scientific evidence for clinical decision making.Methods:The present study retrospectively collected the clinical data of patients with first-episode ischemic stroke who were hospitalized in the Department of Neurology,Bethune Hospital,Shanxi Province from January 2019 to December 2020,as well as the Essen stroke risk score and SPI-Ⅱ score when out of the hospital.1168 patients who met the inclusion criteria were followed up for 1 year.To investigate the recurrence of patients one year after stroke,statistically significant variables were selected to be included into the model through single-factor analysis.Logistic regression,support vector machine,random forest,XGboost,Adaboost and other machine learning algorithms were used to learn in the training set data to build the prediction model,and then the test set data were verified.The accuracy,accuracy,sensitivity,specificity and AUC of the machine learning models were compared with Essen stroke risk score and SPI-Ⅱ score,and evaluated the prediction effect of each model on 1-year recurrence of ischemic stroke.Results:A total of 1168 first-episode ischemic stroke patients were included in this study,and they were divided into the recurrence group(155 patients)and the non-recurrence group (1013 patients)according to investigate the recurrence after 1 year.The recurrence rate of first-episode ischemic stroke patients within 1 year was 13.3%.Essen stroke risk score and SPI-Ⅱ score predicted recurrence at 1 year with AUC 0.559(95%CI: 0.510-0.607)and0.686(95%CI: 0.640-0.732),respectively.Machine learning algorithm was used to predict the recurrence of first-episode ischemic stroke patients within 1 year,and the ranking of AUC values was as follows: XGboost 0.998(95%CI: 0.996~1.000),support vector machine 0.998(95%CI: 0.993~1.000),and random forest 0.997(95%CI: 0.997).0.993~1.000),Adaboost 0.996(95%CI: 0.993~0.999),Logistic regression 0.849(95%CI:0.772~0.925).In support vector machine,random forest,XGboost,Adaboost and other machine learning algorithm models,the common factors influencing 1-year recurrence of first-stroke patients were: length of hospital stay,D-dimer,lymphocyte ratio,creatinine,high density lipoprotein,diastolic blood pressure.Conclusion:1.The recurrence rate of first-episode ischemic stroke patients within 1 year was 13.3%.2.Compared with Essen score and SPI-Ⅱ score,XGboost,Adaboost,random forest,support vector machine algorithm and Logistic regression algorithm have better ability to predict the recurrence risk of first-stroke patients one year after stroke.3.Compared with the Logistic regression,XGboost,Adaboost,random forest,support vector machine,and the identification ability of the first stroke patients and the high-risk group of recurrence one year after stroke is stronger.