Do nutritional variables improve cardiovascular disease prediction? A comparative machine learning analysis

Cardiovascular diseases (CVD) remain a leading cause of mortality worldwide, highlighting the need for accurate prediction models. While machine learning (ML) approaches have shown promising results, the contribution of nutritional variables to prediction performance remains unclear. This study evaluates the incremental effect of dietary intake variables by comparing baseline and nutrition-extended feature sets across multiple ML models.

A cross-sectional dataset of 1,359 adults aged 19-65 years was analyzed, including demographic, anthropometric, clinical, biochemical, and dietary variables collected via 24-h dietary recall. Two feature configurations were constructed: a baseline set excluding dietary variables and a nutrition-extended set including detailed nutritional features. Six ML algorithms-Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), k-Nearest Neighbors (KNN), Decision Tree (DT), and Artificial Neural Network (ANN)-were trained and optimized using cross-validation, with SMOTE applied to the training data to address class imbalance.

The baseline Random Forest model achieved the highest discriminative performance (ROC-AUC = 0.7823). The inclusion of nutritional variables produced limited and model-dependent improvements in selected classifiers (e.g., SVM and ANN) but did not enhance the best-performing baseline model. Across all models, detection of actual cardiovascular disease cases was lower than detection of non-diseased individuals, indicating that further calibration is required before clinical screening use.

These findings suggest that nutritional variables provide limited, model-dependent, and incremental contributions to CVD prediction, emphasizing the importance of evaluating feature groups systematically rather than assuming that a larger feature set will improve performance.