Data-driven identification of key predictors of uncontrolled hypertension: A cross-sectional study

Abstract

Uncontrolled hypertension (HTN) increases the risk of adverse health events. This study aimed to identify key predictors of uncontrolled HTN in 1,308 Mexican adults with a prior diagnosis of HTN who were undergoing pharmacological treatment. We utilized data from the 2022 National Health and Nutrition Survey and applied data-driven algorithms within an artificial intelligence framework to enhance predictive accuracy and interpretability. Specifically, we integrated Random Forest, XGBoost, LASSO regression, and SHAP analysis. Uncontrolled HTN was defined as systolic blood pressure ≥ 130 mmHg or diastolic blood pressure ≥ 80 mmHg based on two readings. We applied LASSO regression to exclude unrelated factors and trained Random Forest and XGBoost algorithms to identify the most important predictors. The key contributors to model accuracy in Random Forest were years since HTN diagnosis (11.9), age (9.4), and source of medical care (4.6), while SHAP analysis in XGBoost further highlighted age (0.115) and source of medical care (0.065) as significant factors. When compared to a traditional logistic regression model, the data-driven approach demonstrated superior predictive performance, with Random Forest achieving an AUC of 0.75 (95% CI 0.72–0.77) versus logistic regression (AUC = 0.61, 95% CI 0.59–0.64). XGBoost exhibited lower predictive capacity (AUC = 0.54, 95% CI 0.49–0.60). These findings underscore the importance of age, duration since diagnosis, and source of medical care in predicting uncontrolled HTN. If replicated, this evidence can inform public health strategies to better target at-risk populations and optimize HTN management through data-driven interventions.