A Risk Prediction System Based on Machine Learning for Acute Ischemic Stroke in Patients with Atrial Fibrillation

Atrial fibrillation (AF) significantly increases the risk of acute ischemic stroke (AIS), which may lead to disability and death. It is a clinical challenge for the risk assessment and management of AF. In this study, 5,330 cases were obtained from the Department of Cardiology, Chinese PLA General Hospital, containing demographic information, laboratory results, medical history, and other clinical variables. After preprocessing the data and using the random forest method to fill in the missing values, 12 variables with significant differences were selected as the features of the prediction model. To address the issue of imbalanced categories in the dataset, we implemented the Synthetic Minority Over-sampling Technique (SMOTE) to achieve a balanced distribution of samples between the two categories. Five machine learning algorithms were used to establish prediction models. After comprehensive evaluation and comparison, the XGBoost algorithm, with an AUC of 0.908 (95% CI 0.872-0.938), demonstrated superior performance in AIS risk prediction among the five machine learning methods tested. The model was implemented within a user-friendly local prediction system constructed using the Flask framework. Predictions of AIS risk were provided to healthcare professionals, accompanied by a graphical representation illustrating the impact of individual patient characteristics on the predicted outcome. Our study shows the potential of machine learning techniques in enhancing stroke risk assessment among patients with AF, offering clinicians a valuable tool for early identification and intervention in high-risk individuals. With the system's predictions, doctors may be able to intervene earlier in the rehabilitation of patients, potentially speeding up the recovery process and mitigating the onset of stroke symptoms.