A Combined-Mode Machine Learning Model for Predicting Stroke Recurrence During Hospitalization in Patients with Acute Minor Ischemic Stroke

Abstract

Acute minor ischemic stroke patients often experience recurrence shortly after symptom onset, highlighting the importance of predicting stroke recurrence for guiding treatment decisions. This study evaluated the effectiveness of machine learning models in predicting in-hospital recurrence. The study cohort comprised 322,135 patients with acute minor ischemic stroke from 1439 centers, as established by Chinese Stroke Center Alliance. Patients were randomly allocated into training and test sets by different centers. Models including extreme gradient boosting (XGB), light gradient boosting (LGB), and adaptive boosting (ADA) were developed using fivefold cross-validation on the training set. Optimization was performed for all models based on the most important variable, history of ischemic stroke. Compared with the traditional generalized linear model (GLM), the XGB, LGB, ADA models yielded area under the curve (AUC) values ranging from 0.788 to 0.803 after optimization. All models showed significant improvements in AUC compared with GLM, with LGB exhibiting the most substantial enhancement after optimization. For the first time, this study developed models specifically designed to predict in-hospital stroke recurrence in acute minor ischemic stroke patients. This finding aids in identifying high-risk patients and prompts physicians to provide targeted treatment. However, further external validation is warranted to confirm the model's generalizability.