Explainable and interpretable models for predicting early-onset hypertension in the Tlalpan 2020 cohort

Background: Early-onset hypertension results from complex interactions among demographic, lifestyle, metabolic, and psychosocial factors. While machine learning models can predict hypertension with relative accuracy, their lack of interpretability limits their clinical utility. Methods: Using a nested case-control design based on the Tlalpan 2020 prospective cohort, a 10-year study of clinically healthy adults in Mexico City, this study applies DSRegPSOP, a symbolic regression approach, to develop interpretable mathematical models. The dataset included demographic, clinical, biochemical, lifestyle, and sleep-related variables. We addressed class imbalance using oversampling and SMOTE-based strategies, and evaluated model performance with accuracy, sensitivity, specificity, F1-score, and AUC-ROC. Results: DSRegPSOP produced compact analytical expressions with predictive performance comparable to state-of-the-art machine learning algorithms while preserving interpretability. The resulting models reveal clinically meaningful predictors of early-onset hypertension. Conclusion: DSRegPSOP provides a transparent and interpretable model for hypertension risk assessment that shows promising potential to support early prevention strategies, pending external validation on independent cohorts.