Trend and predictive modeling of hypertension risk using multiyear community health surveillance data: Evidence from...

Hypertension remains a major global health burden, often undetected until complications occur. In Indonesia, evidence integrating multiyear surveillance data with machine learning (ML) to predict population-level risk remains limited. The objective of the study was to examine temporal trends in hypertension risk and develop predictive models at the population level.

This observational study utilized community health surveillance data on 196,951 adults (2019-2021). Blood pressure was classified using World Health Organization/Joint National Committee 7 criteria, with prehypertension and hypertension combined as abnormal (y = 1). Predictors included sociodemographic characteristics, anthropometric measurements, various health behaviors, and comorbidities. Missing values were imputed using K-Nearest Neighbors. Logistic Regression, Random Forest, Gradient Boosting, eXtreme Gradient Boosting (XGBoost), and Support Vector Machine models were developed using an 80:20 split, 10-fold cross-validation, and hyperparameter tuning. Model interpretability was assessed using SHapley Additive exPlanations (SHAP).

Abnormal blood pressure prevalence increased from 74.4% (2019) to 79.7% (2020) and 80.2% (2021). XGBoost demonstrated superior performance, achieving 78.3% accuracy and an area under the curve of 0.736, outperforming other models. SHAP analysis identified age, waist circumference, body mass index, and male sex as the strongest predictors of the risk of hypertension.

Hypertension risk in Indonesia rose from 2019 to 2021. Interpretable ML models, particularly XGBoost, show potential for population-level risk stratification and early detection in community and primary healthcare settings.