A prediction model for stroke risk based on explainable machine learning.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Cardiovascular Disease Prediction

Background:

• Stroke is a leading cause of death and disability worldwide.
• Accurate prediction of stroke risk is crucial for timely intervention and prevention.
• Existing risk assessment tools may not fully capture the complexity of stroke etiology.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting stroke risk.
• To identify key risk factors contributing to stroke incidence.
• To assess the performance of various algorithms in stroke risk prediction.

Main Methods:

• Retrospective analysis of 134 stroke patients and 354 controls.
• Development and comparison of eight machine learning models, including Random Forest (RF).
• Feature selection using LASSO and logistic regression; model evaluation via ROC/PR curves and accuracy metrics.

Main Results:

• The RF algorithm achieved high performance: ROC AUC of 0.96, PR AUC of 0.92, specificity of 0.97, and precision of 0.92.
• Key predictors identified include weekly exercise days, calf circumference, medical history, gender, BMI, and the STRATIFY score.
• Shapley Additive Explanations confirmed the significant impact of physical activity and anthropometric measures on stroke risk.

Conclusions:

• Machine learning, particularly the RF algorithm, offers a robust approach for stroke risk prediction.
• Regular physical activity and specific physiological markers are significant determinants of stroke risk.
• These predictive models can potentially enhance clinical decision-making and personalized stroke prevention strategies.