Prediction of favorable outcomes of acute basilar artery occlusion using machine learning

This study aims to develop an interpretable machine learning model using SHapley Additive exPlanations (SHAP) to predict favorable outcomes based on clinical, imaging, and angiographic data.

This study analyzed data from 184 patients with acute basilar artery occlusion (BAO) who underwent endovascular treatment (EVT) and completed a 90-day follow-up at Shanxi Provincial People's Hospital. A total of 68 medical variables were collected to develop predictive models using three machine learning algorithms: logistic regression (LR), support vector machine (SVM), and Light Gradient Boosting Machine (LightGBM). Model performance was comprehensively assessed using Accuracy, Recall, Precision, F1 score, and the area under the receiver operating characteristic curve (AUC-ROC), and the best-performing model's results were interpreted using SHAP.

The SVM model demonstrated better performance, with an AUC of 0.899±0.059 (95% confidence interval (CI) 0.840 to 0.957), accuracy of 0.859±0.057, recall of 0.858±0.068, precision of 0.872±0.084, and an F1 score of 0.857±0.059. Recursive feature elimination with random forest (RF) and SHAP analysis revealed that the absence of ventilator use, absence of tracheotomy, lower National Institutes of Health Stroke Scale (NIHSS) scores at admission, and lower preoperative serum creatinine (SCR) levels were significant predictors of favorable 90-day outcomes.

This study established a machine learning model to identify predictors of favorable outcomes in patients with acute BAO. Significant factors influencing prognosis included the use of mechanical ventilation, tracheotomy, NIHSS score, and preoperative SCR levels.