What question did this study set out to answer?

The study aims to create an interpretable machine learning model to forecast 90-day functional outcomes in patients with vertebrobasilar dolichoectasia and posterior circulation stroke.

April 7, 2026Open Access

Explainable XGBoost model and nomogram for predicting 90-day functional outcome in posterior circulation stroke with vertebrobasilar dolichoectasia: a machine learning-based retrospective cohort study

Puntos clave

The study aims to create an interpretable machine learning model to forecast 90-day functional outcomes in patients with vertebrobasilar dolichoectasia and posterior circulation stroke.
Retrospective enrollment of 380 patients with VBD and posterior circulation ischemic stroke from 2022 to 2023.
Assessment of 19 candidate variables including demographic data, vascular geometry, neuroimaging, and laboratory markers.
Feature selection using LASSO regression and multivariate logistic regression for model building.
Comparison of five machine learning algorithms including XGBoost, with evaluation metrics like ROC curves and decision curve analysis.
Development of an interpretable nomogram for clinical application based on the selected features.
XGBoost achieved the best performance with a training AUC of 0.787 and validation AUC of 0.745.
NIH Stroke Scale, Fazekas score, basilar artery tortuosity distance, and hypertension were identified as independent predictors.
The nomogram displayed strong prognostic discrimination with a significant p-value (< 0.0001).
SHAP analysis indicated NIHSS and Fazekas score as key contributors to outcomes.

Resumen

Vertebrobasilar dolichoectasia (VBD) is a rare but clinically significant vasculopathy that predisposes patients to posterior circulation ischemic stroke (PCIS) and is frequently associated with poor outcomes. Given the complex hemodynamics and multifactorial nature of VBD-related stroke, early identification of high-risk individuals remains challenging. This study aimed to develop and validate an interpretable machine learning (ML) model to predict 90-day poor functional outcome in VBD-induced PCIS patients. We retrospectively enrolled 380 adult patients with radiologically confirmed VBD and PCIS from August 2022 to December 2023 at the Second Hospital of Lanzhou University. A total of 19 candidate variables encompassing demographic data, vascular geometry, neuroimaging, laboratory markers, and comorbidities were assessed. Missing data (< 5%) were imputed using k-nearest neighbors. Feature selection was performed via least absolute shrinkage and selection operator (LASSO) regression, followed by multivariate logistic regression with backward stepwise elimination based on the Akaike information criterion. Five supervised machine learning algorithms—decision tree (DT), Light Gradient Boosting Machine (LGBM), logistic regression (LR), support vector machine (SVM), and eXtreme Gradient Boosting (XGBoost)—were constructed and compared. Model performance was evaluated using ROC curves, calibration plots, and decision curve analysis (DCA). The final model was interpreted using SHapley Additive exPlanations (SHAP), and a nomogram with an integrated nomoScore was developed for bedside application. Four independent predictors were identified: NIH Stroke Scale (NIHSS) score, Fazekas score, basilar artery tortuosity distance (BA distance), and history of hypertension. Among all models, XGBoost achieved the best performance (training AUC: 0.787; validation AUC: 0.745), superior to traditional logistic regression and other classifiers. Calibration and DCA confirmed its robustness and clinical utility. SHAP analysis revealed NIHSS and Fazekas score as dominant contributors, followed by BA distance and hypertension. A nomogram based on these features was constructed, and its derived nomoScore showed strong prognostic discrimination (p < 0.0001). Logistic regression confirmed nomoScore as an independent outcome predictor, and decision metrics (sensitivity, specificity, PPV, NPV, Youden index) demonstrated its diagnostic efficacy. We developed and validated an interpretable XGBoost-based ML model for predicting 90-day unfavorable outcomes in patients with VBD-related PCIS. The integration of key vascular, imaging, and clinical indicators into an individualized nomogram enables early risk stratification and may support timely, targeted interventions in clinical practice.

Me gusta

Guardar

Ver artículo completo