| ObjectiveThe aim of this study is to construct an early-warning model of stroke recurrence with the characteristics of traditional Chinese medicine,and make multiple evaluation between models to explore a suitable early warning model of ischemic stroke recurrence.MethodsStudy designThis is a prospective,multicenter,and registry study.SettingThis study collected the clinical data of 1741 patients with first-ever ischemic stroke between July 2016 and November 2019 in a total of 7 clinical branches of the hospitals which were Dongzhimen hospital Beijing university of Chinese medicine Guangdong provincial hospital of Chinese medicine,The affiliated hospital to Changchun University of Chinese medicine The first affiliated hospital of Henan University of Chinese medicine,Dongfang hospital Beijing university of Chinese medicine,Taiyuan hospital of traditional Chinese medicine.Ethics,Consent,and PermissionsThis study was approved by the ethical review committee of the institute of basic research in clinical medicine of the China Academy of Chinese Medical Sciences(2016 NO.08),and all of the study participants provided informed consent.EndpointsThe primary endpoints were a stroke recurrence event.It was defined as brain infarction(including transient ischemic attack)or intracranial hemorrhage that was diagnosed at the hospital.The secondary endpoints measures were death events.Statistical methodsDescriptive statistics use methods of frequency statistics,mean,standard deviation,median statistics,and quartile statistics.For measurement data,the mean±standard deviation(x±s)is used for normal distribution,the median and interquartile range(IQR)are used for non-normal distribution,and frequency and percentage are used for count data.Variable screening uses statistical correlation tests,distance correlation coefficients,and mutual information entropy for screening.Use the t.est,chisq.test,and Cov()functions of R software to perform statistical correlation test and analysis on the independent variables;use the dcor function of the R software energy toolkit to calculate the distance correlation coefficient,and use the multiinformation of the infotheo toolkit The function performs information entropy calculation.Longitudinal data modeling and analysis uses the R software Ime4 toolkit to implement GLMM modeling,and the geepack toolkit to implement GEE modeling.Use the pROC toolkit of R software to calculate the AUC value for all models and draw the corresponding ROC curve.Use the survfit function in the R software survival toolkit to obtain the estimated Kaplan-Meier cumulative recurrence rate when competition events are not considered.Use the cuminc function in the R software cmprsk toolkit to calculate the cumulative occurrence of stroke recurrence and death before recurrence when considering competition events rate estimates..Results1.Follow-up and outcome1741 patients completed follow-up(including 60 deaths),follow-up time was 1-3 years.175 patients had recurrence events,and the cumulative recurrence rate was 10.05%(95%CI 8.64-11.47%).Among them,the patients with the most types of recurrence were cerebral infarction(118 cases,67.43%).60 patients died and 11 cases died of cerebrovascular disease(6 cases of cerebral infarction and 5 cases of cerebral hemorrhage).There were 43 died before stroke recurrence and 17 deaths(9 deaths due to stroke recurrence)after stroke recurrence.2.Construction of early warning model for ischemic stroke recurrence2.1 Predictive variableThe selection of variables in this study was based on the combination of literature collation,statistical methods(including statistical correlation test,distance correlation coefficient,mutual information entropy,multi-factor Cox regression)and clinical expert opinions.Finally,age,previous history of cerebral hemorrhage,family history of coronary heart disease,NIHSS>4,smoking history,physical exercise,Chinese and Western Medicine Intervention(Chinese medicine intervention and use of non-heparin platelet aggregation inhibitors),syndrome of TCM,the constitution of TCM,Severe intracranial and external stenosis,the number of cerebral infarcts(?3),and Heterogeneous unstable plaques were determined as the predictive variables of stroke recurrence,2.2 Prediction model constructionIn the warning models of ischemic stroke recurrence with "combination of disease and syndrome",the predictive value of Western medicine risk factors,TCM syndrome,and TCM constitution in the stroke recurrence was considered.The data were used to comprehensively evaluate the early warning ability of the four modes,that was the longitudinal data model(GEE,GLMM model)and cross-section data model(Cox proportional risk reggression model,competitive risk model).2.3 Comparative analysis of longitudinal data model and cross-section data modelIn the longitudinal data,The GLMM and GEE models are consistent in the prediction results.In the western medicine model,the AUC is 0.69687 and 0.69695,in the TCM syndrome model,the AUC is 0.71807 and 0.71809,and in the TCM constitution model,the AUC is 0.72661 and 0.72668,and in the TCM syndrome&constitution model,the AUC is 0.79099 and 0.79102.The AUC value of the model with TCM syndromes and physical factors is higher than that of the modern medicine model.In cross-sectional data,the modern medical model,traditional Chinese medicine syndrome model,traditional Chinese medicine constitution model and traditional Chinese medicine syndrome constitution model perform relatively well on 100 training sets,indicating that the Cox model performs well under different data The AUC values of the three models in the training set close to the average level are 0.6929,0.58625,0.66583 and 0.58621,respectively.The model of modern medicine is superior to the traditional Chinese medicine syndrome and constitution model.The results of the competitive risk model show that death cases affect the estimated results of the Cox model and gradually increase over time,and have a greater impact on the estimated results of TCM syndromes and TCM constitution models.3.Analysis of risk factors for stroke recurrenceMulti-factor Cox analysis and longitudinal data analysis found age,previous history of cerebral hemorrhage,family history of coronary heart disease,continuous smoking history,exercise training,Chinese medicine intervention,western medicine non-heparin platelet aggregation inhibitors,severe intracranial artery stenosis or more severe carotid stenosis is a recurring factor of stroke,and the difference is statistically significant(P<0.05);the dynamic changes of the internal stroke syndrome of traditional Chinese medicine syndrome and Qi deficiency constitution are related to stroke recurrence,and the difference is statistically significant(P<0.05).Conclusion1.The modeling method of the longitudinal data model is better than the cross-sectional data model,especially in the early warning model of ischemic stroke recurrence with TCM syndromes and TCM constitutions;2.Longitudinal data GLMM and GEE models can better reflect the predictive value of TCM syndromes "dynamic space-time,multi-dimensional interface" in the early warning model of ischemic stroke recurrence.;3.In the longitudinal data GLMM and GEE models,the early warning model of ischemic stroke containing TCM syndromes and TCM constitution were stabled.Compared with modern medical models,it can predict the ischemic stroke recurrence earlier,which can provide a certain reference value for the evaluation of secondary prevention of ischemic stroke;4.Age,history of previous cerebral hemorrhage,family history of coronary heart disease,continuous smoking history,physical exercise,Chinese medicine intervention,non-heparin platelet aggregation inhibitors of Western medicine,severe intracranial vascular stenosis,severe carotid stenosis were associated with stroke recurrence.The dynamic changes of internal wind syndrome in TCM syndromes and the constitution of Qi deficiency have a significant effect on recurrence.5.Death events have an influence on the risk estimation of recurrence,which increases gradually over time.However,the number of patients who died after more than 1 year in this study was relatively small,conclusions need to be confirmed by a larger sample and a longer follow-up. |
PDF Full Text Request