AI prediction model for endovascular treatment of vertebrobasilar occlusion with atrial fibrillation

Abstract

Endovascular treatment (EVT) for vertebrobasilar artery occlusion (VBAO) with atrial fibrillation presents complex clinical challenges. This comprehensive multicenter study of 525 patients across 15 Chinese provinces investigated nuanced predictors beyond conventional metrics. While 45.1% achieved favorable outcomes at 90 days, our advanced machine learning approach unveiled subtle interaction effects among clinical variables not captured by traditional statistical methods. The predictive model distinguished high-risk subgroups by integrating multiple parameters, demonstrating superior prognostic precision compared to standard NIHSS-based assessments. Novel findings include nonlinear relationships between dyslipidemia, stroke severity, and functional recovery. The developed predictive algorithm (AUC 0.719 internally, 0.684 externally) offers a more sophisticated risk stratification tool, potentially guiding personalized treatment strategies in high-complexity VBAO patients with atrial fibrillation.

Fig. 1. Study flowchart outlining participant selection and analysis.

This figure outlines the patient selection process for the Chinese and European cohorts, the allocation of training and test sets, and the workflow for model development, validation, and external testing. The Chinese cohort was split into training and test sets, while the European cohort served as an external validation set, highlighting exclusion criteria and key analytical steps.

Fig. 2. Comprehensive evaluation of model performance and predictor impact.

a Comparative Analysis of Model Performance Indices. This figure presents a comparative analysis of two generalized linear models (GLMs), referred to as Model 1 and Model 2, evaluated across diverse performance indices. It aids in identifying the optimal model by striking a balance between predictive accuracy and model complexity. b Model Comparison of Predictor Odds Ratios. The bar chart compares the odds ratios (OR) of predictors in two logistic regression models: a simplified model (Model 1) and a full model (Model 2). ASITN/SIR American Society of Interventional and Therapeutic Neuroradiology score, EOT Time from estimated occlusion to groin puncture, NIHSS National Institutes of Health Stroke Scale score, PC-ASPECTS Posterior circulation Alberta Stroke Program Early CT Score, sICH Symptomatic Intracerebral Hemorrhage.

Fig. 3. Model performance evaluation: ROC curve analysis, calibration assessment, and decision curve analysis for test and external validation datasets.

a, b ROC Curve Analyses and Optimal Thresholds. a and b display the ROC curve analysis results for the test dataset (a) and the external validation dataset (b), respectively. Calculated using the DeLong method, the AUC for the test dataset is 0.719 (95% CI: 0.639–0.799), with an optimal threshold corresponding to a sensitivity of 0.581 and a specificity of 0.803. For the external validation dataset, the AUC is 0.684 (95% CI: 0.586–0.783), and the optimal threshold point has a sensitivity of 0.562 and a specificity of 0.773. c, d Calibration Curve Analyses. c and d display the calibration curves for the model on the test dataset and the external validation dataset, respectively. The curve for the test dataset deviates from the ideal line to some extent, while the curve for the external validation dataset more closely follows the ideal line, particularly in the higher probability range, indicating better calibration performance for this dataset. e, f Model Diagnostic Performance via DCA Analysis. The DCA analysis diagrams evaluate the diagnostic efficacy of a logistic regression model on a test dataset and an external dataset. The model’s predictions are deemed diagnostically relevant in 82% of cases for the test dataset and 74% for the external dataset.