Explainable machine-learning-based cardiovascular disease prediction in patients with hypertension: Algorithm...

Area of Science:

• Cardiovascular Health
• Machine Learning in Medicine
• Hypertension Management

Background:

• Cardiovascular disease (CVD) is a major cause of illness in hypertensive patients.
• Accurate and interpretable CVD prediction models are vital for early intervention in hypertension.
• Current clinical screening methods require enhancement for hypertensive populations.

Purpose of the Study:

• To develop and validate a machine-learning (ML) model for predicting CVD risk in hypertensive individuals.
• To improve the efficacy of clinical screening for CVD in patients with hypertension.
• To enhance the interpretability of CVD risk prediction models.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (2009-2018).
• Employed feature selection techniques including Least Absolute Shrinkage and Selection Operator, Boruta, and Recursive Feature Elimination.
• Constructed prediction models using four ML algorithms and evaluated performance via 10-fold cross-validation and an independent test set.
• Applied SHapley Additive exPlanations (SHAP) for mechanistic interpretability.

Main Results:

• Included 2781 participants; identified eight key predictive variables.
• The Balanced Bagging Classifier model exhibited superior performance.
• SHAP analysis identified neutrophil-lymphocyte ratio, waist-to-height ratio, age, HDL cholesterol, LDL cholesterol, kidney disease, sleep disturbance, and diabetes as top predictors.

Conclusions:

• The developed ML model is effective and generalizable for predicting CVD in hypertensive patients.
• SHAP analysis significantly enhances model interpretability, supporting clinical decision-making.
• The model shows potential as a practical tool for early CVD screening and risk assessment in hypertensive populations.