An interpretable XGBoost model using 2022 MEPS data predicted early onset hypertension in US adults aged 55 or younger with strong discrimination (ROC AUC 0.807) and high recall (0.848), supporting recall-oriented screening, though prospective validation and routine fairness audits are warranted.
Among 9,850 US adults aged 55 years or younger in the 2022 MEPS, 1,647 had early onset hypertension, a weighted prevalence of 15.2%.
The extreme gradient boosting model showed strong discrimination for predicting early onset hypertension.
The model achieved high recall, supporting its use for recall-oriented screening.
Feature selection reduced 84 screened variables to 45 using LASSO before model training.
Key predictors of early onset hypertension were older age, high body mass index, and a diagnosis of high cholesterol.
Predictor influence varied by subgroup: high cholesterol was particularly influential among adults older than 40 years, and race and ethnicity were especially important among African American adults.
Fairness analyses showed differences in true positive rates across racial and ethnic groups.
The authors conclude the interpretable model could support recall-oriented screening of younger adults at risk of hypertension, but call for prospective validation and routine fairness audits.
This study built a computer model to identify younger adults (55 or under) who may have high blood pressure, using 2022 national survey data on nearly 10,000 people. About 15% had hypertension. The model used information such as age, weight, cholesterol diagnosis, health care use, and socioeconomic factors, and it was designed so that people can see which factors drive its predictions. The model was fairly good at telling people with and without hypertension apart, and it caught about 85% of those who had it. That makes it suited to screening, where missing cases is a bigger concern than flagging some people who turn out not to have the condition. Older age, higher body mass index, and a high cholesterol diagnosis were among the most important factors. Cholesterol mattered more for those over 40, and race and ethnicity were especially important in predictions for African American adults. The model did not perform identically across groups: it caught slightly more true cases among Hispanic and Asian people and somewhat fewer among African American people compared with White people. This research suggests such tools might help flag younger adults for earlier evaluation, but the authors note it needs testing in real-world, forward-looking studies and regular checks for fairness before it could be relied upon.
Wu L, Kim K, Washington P, Park C. (2026). Interpretable Machine Learning for Predicting Early Onset Hypertension in the US.. Preventing chronic disease. https://doi.org/10.5888/pcd23.260109