Leveraging multivariate analysis and adjusted mutual information to improve stroke prediction and interpretability

Area of Science:

• Cardiovascular Disease Epidemiology
• Machine Learning in Healthcare
• Public Health Informatics

Background:

• Stroke remains a leading cause of disability and death globally.
• Accurate prediction of stroke risk is crucial for timely intervention and prevention.
• Identifying key demographic and clinical risk factors can inform targeted public health initiatives.

Purpose of the Study:

• To develop and validate a machine learning model for predicting stroke risk.
• To identify significant demographic and clinical predictors of stroke.
• To determine the optimal machine learning algorithm for stroke risk prediction.

Main Methods:

• A cross-sectional study utilized data from 438,693 adults (2021 Behavioral Risk Factor Surveillance System).
• Demographic and clinical features were analyzed using logistic regression and adjusted mutual information for feature importance.
• Multiple machine learning models were constructed and evaluated based on accuracy, AUC ROC, and F1 score.

Main Results:

• The Random Forest model demonstrated superior performance, achieving 72.46% accuracy, 0.72 AUC ROC, and 0.74 F1 score.
• Significant stroke risk factors identified include older age, diabetes, hypertension, high cholesterol, and history of myocardial infarction or angina.
• Hypertension, myocardial infarction history, angina, age, diabetes, and cholesterol were the most influential features.

Conclusions:

• The Random Forest model provides a robust method for predicting stroke risk from demographic and clinical data.
• Feature importance analysis highlights hypertension and diabetes as critical targets for clinical monitoring and intervention.
• These findings support the development of data-driven strategies for stroke prevention.