Shapely additive values can effectively visualize pertinent covariates in machine learning when predicting...

Area of Science:

• Medical Informatics
• Public Health
• Cardiovascular Disease Research

Background:

• Machine learning (ML) shows promise in medical prediction, but its efficacy for long-term outcomes like blood pressure using lifestyle factors is under-explored.
• Accurate prediction of hypertension risk is crucial for public health interventions and disease prevention.

Purpose of the Study:

• To assess the accuracy of ML techniques in predicting hypertension risk.
• To identify key lifestyle and nutritional factors contributing to hypertension prediction using ML.

Main Methods:

• A cross-sectional study utilized data from the National Health and Nutrition Examination Survey (NHANES) from January 2017 to March 2020.
• The XGBoost ML model was employed for its high performance in medical applications.
• Model efficacy was evaluated using AUROC and Balanced Accuracy; covariate importance was assessed using Gain statistics and SHapely Additive exPlanations (SHAP).

Main Results:

• Age was the strongest predictor of hypertension (53.1% gain).
• Demographic factors including poverty (4.33% gain) and Black race (4.18% gain) were significant predictors.
• Nutritional factors (Sodium, Caffeine, Potassium, Alcohol intake) contributed 37% to the prediction, highlighting diet's role.

Conclusions:

• Machine learning models, particularly XGBoost, can effectively predict hypertension risk.
• Key predictors include age, socioeconomic status, race, and specific dietary components, offering targets for personalized prevention strategies.