Development and validation of a machine learning model to predict stroke risk based on the NHANES database

Area of Science:

• Medical informatics
• Biostatistics
• Public health

Background:

• Stroke poses a significant global health burden with high incidence and mortality rates.
• Accurate stroke risk assessment is crucial for effective prevention and clinical management.
• Machine learning offers promising tools for enhancing predictive capabilities in healthcare.

Purpose of the Study:

• To evaluate the efficacy of machine learning models for stroke risk prediction.
• To compare different variable selection techniques in developing predictive models.
• To identify key predictors of stroke using robust analytical methods.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (1999-2002).
• Applied LASSO regression with stepwise selection, random forest, and Boruta algorithm with LASSO regression for variable selection and model construction.
• Assessed model performance using receiver operating characteristic (ROC) curves, precision-recall curves, calibration, and decision curve analyses.

Main Results:

• A LASSO and stepwise regression model demonstrated strong discriminative performance with an Area Under the Curve (AUC) of 0.843.
• The Boruta algorithm followed by LASSO regression achieved a comparable AUC of 0.828.
• Random forest yielded lower predictive accuracy (AUC = 0.612) with fewer selected predictors.

Conclusions:

• Different variable selection methods significantly influence the predictive accuracy of stroke risk models.
• Machine learning models, particularly those employing LASSO and Boruta algorithms, show high accuracy and clinical value for stroke risk prediction.
• The developed models utilizing the NHANES database provide a reliable tool for predicting stroke risk, aiding in prevention and intervention strategies.