Predictive Value of Constructing a Risk Warning Model Based on the SMOTE Algorithm for Predicting the Prognosis of Intravenous Thrombolysis in Patients with Acute Cerebral Infarction

FENG Huifang; ZHANG Jiangang; XIE Weizheng; XIAO Caixia; YANG XinLi

doi:10.3969/j.issn.1004-437X.2026.05.020

Abstract：Objective To analyze the predictive value of a risk warning model constructed based on the synthetic minority over-sampling technique(SMOTE)algorithm for the prognosis of patients with acute cerebral infarction(ACI).Methods A prospective study was conducted on 210 ACI patients who received recombinant tissue plasminogen activator(rt-PA)intravenous thrombolytic therapy at Anyang People's Hospital from January 2023 to January 2025.These patients were divided into a poor prognosis group(61 cases)and a good prognosis group(149 cases)based on their prognosis at 3 months after thrombolysis.The clinical data of patients with different prognoses were compared,and factors influencing poor prognosis were analyzed.Based on these influencing factors,a logistic regression(LR)model 1 was constructed.Additionally,the dataset was improved using the SMOTE algorithm,LR model 2 was constructed based on SMOTE algorithm.The predictive efficiency of the two models was then compared and analyzed.Results The poor prognosis rate of ACI patients after thrombolysis at 3 months was 29.05％.The proportion of diabetes,admission National Institutes of Health stroke scale(NIHSS)score,time from onset to thrombolysis,serum chemokine CXC ligand 12(CXCL12),and monocyte chemoattractant protein-1(MCP-1)levels in the poor prognosis group were higher than those in the good prognosis group,the serum uncoupling protein 2(UCP2)level was lower than that in the good prognosis group(P＜0.05).Multivariate logistic analysis revealed that the time from onset to thrombolysis,diabetes,admission NIHSS score,serum CXCL12,and MCP-1 were all risk factors for poor prognosis in ACI patients,while serum UCP2 was an independent protective factor(P＜0.05).The constructed LR model 1 was Y=-1.482+0.657X,+0.407X2+0.326X3+0.233X4-0.074X5+0.046X6,the Y was the logit(P),X1 was diabetes mellitus,X2 was NIHSS score at admission,X3 was time from onset to thrombolysis,X4 was CXCL12,X5 was UCP2,and X6 was MCP-1.The LR model 2 based on the SMOTE algorithm was Y=-1.039+0.576X1+0.293X2+0.198X3+0.112X4-0.062X5+0.029X6.Receiver operating characteristic(ROC)curve analysis showed that the area under the curve(AUC)for LR model 2 in predicting poor prognosis in ACI patients was 0.938,significantly higher than the AUC of 0.866 predicted by LR model 1(P＜0.05).Furthermore,when compared with LR model 1,LR model 2 had a net reclassification improvement(NRI)and integrated discrimination improvement(IDI)both greater than 0(P＜0.05).Conclusion Compared with the conventional LR model,the LR model based on the SMOTE algorithm significantly improves the predictive value for predicting poor prognosis in ACI patients,providing a more reliable clinical basis for identifying high-risk patients with poor prognosis in clinical practice.