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

Area of Science:

• Cardiovascular Disease Research
• Biostatistics and Machine Learning
• Public Health and Epidemiology

Background:

• Early-onset hypertension arises from multifaceted demographic, lifestyle, metabolic, and psychosocial factors.
• Current machine learning models predict hypertension accurately but lack clinical interpretability.
• Understanding hypertension's complex etiology is crucial for effective prevention.

Purpose of the Study:

• To develop interpretable mathematical models for early-onset hypertension risk assessment using symbolic regression.
• To identify clinically meaningful predictors of hypertension.
• To offer a transparent alternative to complex predictive models in clinical practice.

Main Methods:

• A nested case-control design was employed within the Tlalpan 2020 prospective cohort.
• DSRegPSOP, a symbolic regression approach, was applied to a dataset of demographic, clinical, biochemical, lifestyle, and sleep variables.
• Class imbalance was managed using oversampling and SMOTE, with performance evaluated by accuracy, sensitivity, specificity, F1-score, and AUC-ROC.

Main Results:

• DSRegPSOP generated concise analytical expressions for hypertension prediction.
• The models achieved predictive performance comparable to advanced machine learning algorithms.
• Key predictors of early-onset hypertension were identified, enhancing model interpretability.

Conclusions:

• DSRegPSOP offers a transparent and interpretable approach to hypertension risk assessment.
• The method shows potential for supporting early hypertension prevention strategies.
• External validation on independent cohorts is recommended to confirm findings.