| Objective:Recanalization treatment for acute ischemic stroke(including intravenous thrombolysis and endovascular therapy)can restore perfusion in the ischemic penumbra,reduce stroke-related disability and mortality,and improve long-term prognosis.This study aims to analyze the risk factors for 90-day unfavorable functional outcome in patients with acute ischemic stroke after recanalization treatment;to establish and validate a clinical prediction model,create visual nomograms,and develop a user-friendly application.Moreover,the feasibility and effectiveness of using machine learning algorithms for individualized prediction of functional outcomes are explored for comprehensive decision support in clinical diagnosis and treatment.In addition,we further identify and analyze the risk factors for symptomatic intracranial hemorrhage and risk factors for unfavorable prognosis in the large vessel occlusion subgroups;and compare the efficacy and safety of recanalization treatment in the subgroups of patients with basilar artery occlusion and those with anterior circulation intracranial artery occlusion.Method:This single-center,prospective,observational clinical cohort study included 338 acute ischemic stroke patients who received recanalization treatment at Beijing Hospital from August 2018 to January 2022.Based on the modified Rankin Scale at 90 days post-treatment,patients were divided into a favorable prognosis group(mRS≤2)and an unfavorable prognosis group(mRS>2).Based on the presence of hemorrhagic transformation on non-contrast CT and a related increase of NIHSS ≥4,patients were categorized into symptomatic intracranial hemorrhage and non-symptomatic intracranial hemorrhage groups.Risk factors were screened using intergroup analysis or LASSO regression analysis,followed by multivariate logistic regression analysis to adjust for confounding factors.Independent risk factors for 90-day unfavorable functional outcomes and symptomatic intracranial hemorrhage were identified and further analyzed in subgroups.Based on the dates of treatment,patient data were divided into a 75%training set and a 25%temporal validation set.In the training set,stepwise regression was used in dimension reduction and model establishment,and the prediction model was evaluated from multiple perspectives in the temporal validation set.In the machine learning part of the study,support vector machines and gradient boosting decision tree algorithms were employed to preprocess clinical data and follow-up outcomes,extract features,build machine learning models,optimize hyperparameters,and perform thorough internal and external validation.Results:(1)Baseline National Institutes of Health Stroke Scale(NIHSS)(OR=1.200,95%CI 1.134-1.270,P<0.001),pre-treatment systolic blood pressure(OR=1.025,95%CI 1.012-1.039,P<0.001),emergency D-dimer levels(OR=2.748,95%CI 1.344-5.619,P=0.006),and large vessel occlusion of the culprit artery(OR=2.863,95%CI 1.537-5.711,P=0.001)were identified as independent risk factors for unfavorable prognosis following recanalization treatment in acute ischemic stroke through univariate and multivariate logistic regression analyses.(2)After adjusting for patients’ baseline NIHSS,platelet count,neutrophil percentage and D-dimer levels,large vessel occlusion of the responsible artery(OR=5.366,95%CI:1.569～18.345,P=0.001),concurrent diabetes(OR=2.195,95%CI:1.083～4.446,P=0.029),and atrial fibrillation(OR=2.265,95%CI:1.073～4.779,P=0.029)were identified as independent risk factors for symptomatic intracranial hemorrhage following recanalization treatment in acute ischemic stroke.(3)In the large vessel occlusion subgroup,multivariate logistic regression analysis revealed that female gender(OR=2.911,95%CI:1.304～6.496,P=0.009),baseline NIHSS(OR=1.276,95%CI:1.176～1.385,P<0.001),systolic blood pressure(OR=1.033,95%CI:1.014～1.052,P=0.001),urea(OR=1.316,95%CI:1.073～1.616,P=0.009),and D-dimer levels(OR=2.813,95%CI:1.092～7.248,P=0.032)were independent risk factors for unfavorable prognosis following recanalization treatment.(4)Compared to the anterior circulation intracranial artery occlusion group,the proportion of patients with atrial fibrillation in the basilar artery occlusion group was lower(35.71%vs.59.21%,P=0.033).There were no statistically significant differences between the two groups in terms of successful recanalization rate(67.86%vs.85.07%,P=0.056)and the proportion of unfavorable prognosis in 90-day mRS(60.71%vs.65.57%,P=0.646).However,the rates of hemorrhagic transformation(21.43%vs.47.76%,P=0.017)and symptomatic intracranial hemorrhage(3.57%vs.14.93%,P=0.031)were significantly lower in the basilar artery occlusion group compared to the anterior circulation intracranial artery occlusion group.(5)The aforementioned variables were used to establish a risk prediction model for unfavorable prognosis after forward stepwise logistic regression.Based on the temporal validation results,the AUC for the training set was 0.856(95%CI:0.810～0.902),and the AUC for the validation set was 0.864(95%CI:0.793～0.936).Moreover,the model exhibited good calibration in both the training and validation sets and showed clinical utility in decision curve analysis.(6)Feature extraction,model establishment,and multi-model comparisons were performed on the clinical data using machine learn ing algorithms.The support vector machine(SVM)algorithm with a polynomial kernel and a regularization parameter C=0.42,which used baseline NIHSS,large vessel occlusion,D-dimer levels,and age as variables for predicting the prognosis of recanalization treatment,achieved the best discrimination(AUC=0.902)in internal cross-validation.In further external validation,the SVM algorithm demonstrated good predictive performance(AUC=0.808)and outperformed previously reported clinical prediction models.Conclusion:(1)This study found that baseline NIHSS score,pre-treatment systolic blood pressure,emergency D-dimer levels,and large vessel occlusion of the culprit artery are independent risk factors for unfavorable prognosis in acute ischemic stroke following recanalization treatment.(2)Large vessel occlusion of the responsible artery,concurrent diabetes,and atrial fibrillation were identified as independent risk factors for symptomatic intracranial hemorrhage.(3)In the large vessel occlusion subgroup,female gender,baseline NIHSS score,systolic blood pressure,urea,and D-dimer levels were independent risk factors for unfavorable prognosis following recanalization treatment.(4)Lastly,compared to patients with anterior circulation intracranial occlusion,those with basilar artery occlusion exhibited similar effectiveness in terms of successful recanalization rate and 90-day functional outcomes after recanalization treatment,and demonstrated better safety in terms of hemorrhagic transformation.(5)A risk prediction model was established,demonstrating good calibration and clinical decision-making value.A nomogram and a developed application based on this model can effectively assist clinicians in their diagnostic and treatment process.(6)Machine learning algorithms,particularly the SVM algorithm,exhibited excellent prognostic prediction performance and outperformed previously reported clinical prediction models in external validation. |
PDF Full Text Request