Machine Learning Approaches for Stroke Risk Prediction: Findings from the Suita Study

Area of Science:

• * Medical Informatics
• * Public Health
• * Cardiovascular Research

Background:

• * Stroke is a major cause of death and disability worldwide.
• * Accurate prediction and identification of risk factors are crucial for public health interventions.
• * Machine learning offers advanced analytical capabilities for complex health data.

Purpose of the Study:

• * To evaluate machine learning algorithms for stroke prediction.
• * To identify key demographic, clinical, and novel biomarkers associated with stroke risk.
• * To explore disparities in stroke incidence across different risk clusters.

Main Methods:

• * Utilized the Suita study dataset (7389 participants, 53 variables).
• * Employed unsupervised k-prototype clustering for risk stratification.
• * Applied supervised models: Logistic Regression, Random Forest, SVM, XGBoost, LightGBM for prediction.
• * Used Shapley Additive Explanations (SHAP) for feature importance analysis.

Main Results:

• * Significant stroke incidence disparities were found among risk clusters.
• * Random Forest demonstrated superior performance in stroke outcome prediction.
• * Key predictors identified: age, systolic blood pressure, hypertension, eGFR, metabolic syndrome, blood glucose.
• * Novel potential biomarkers: elbow joint thickness, fructosamine, hemoglobin, calcium levels.

Conclusions:

• * Machine learning provides a robust framework for accurate stroke risk prediction.
• * Identified both established and novel biomarkers for enhanced stroke risk assessment.
• * Findings support data-driven approaches for early detection and prevention strategies.