Data-driven diagnosis of cardiovascular disease in patients with comorbid hypertension, dyslipidemia or diabetes using classification techniques 10.55131/jphd/2026/240216

Abstract

Cardiovascular disease (CVD) is a leading cause of death globally, especially among individuals with chronic comorbidities such as hypertension, diabetes, and dyslipidemia. Early identification of high-risk individuals is essential for timely intervention, yet conventional risk models often underperform in complex clinical populations. Using real-world clinical data, this study aimed to develop and evaluate machine learning (ML) models for predicting CVD with a focus on optimizing sensitivity and interpretability. We used hospital-based clinical data from patients with at least one of the following comorbidities: hypertension, diabetes or dyslipidemia. The applied ML algorithms included logistic regression, naive Bayes, k-nearest neighbors, decision tree (DT), random forest, support vector machine (SVM), and artificial neural networks. Repeated 10x10-fold cross-validation and downsampling were used to address class imbalance, and systematic hyperparameter tuning was performed. Model performance was evaluated using accuracy, precision, sensitivity, F1 score, and area under the curve (AUC). Shapley Additive Explanations (SHAP) was applied to interpret the final model and identify key features. The SVM model achieved the highest sensitivity, indicating suitability for use in CVD screening, while the DT also showed relatively strong sensitivity. Final SVM performances included a sensitivity of 89.29%, a precision of 57.94%, an F1-score of 69.97%, and an AUC of 75.65%. SHAP analysis revealed creatinine, systolic blood pressure, low-density lipoprotein, fasting blood sugar, and triglycerides as the top predictors of CVD, aligning with established cardiovascular risk factors. We present an interpretable ML model with high sensitivity for CVD prediction in patients with metabolic comorbidities. By integrating SHAP with robust validation techniques, the model supports personalized risk assessment and is suitable for screening applications aimed at early detection and preventive care in clinical settings.