Key result
Gradient boosting model predicts early-onset hypertension in younger adults with ~0.81 AUC.
Why the study?
Early onset hypertension is often overlooked despite its association with lifelong cardiovascular disease risk, motivating the development of an interpretable machine learning prediction model for US adults.
Does an interpretable machine learning model using demographic, socioeconomic, health care utilization, and comorbidity features accurately predict early onset hypertension in US adults aged 55 years or younger?
Cross-Sectional (n=9,850)
Does an interpretable machine learning model using demographic, socioeconomic, health care utilization, and comorbidity features accurately predict early onset hypertension in US adults aged 55 years or younger?
An interpretable extreme gradient boosting machine learning model incorporating medical and nonmedical features achieved strong discrimination (ROC AUC 0.807) and high recall (0.848) for identifying US adults aged 55 or younger at risk for early onset hypertension.
ML model may aid early-onset hypertension risk stratification; leaves open prospective validation before clinical use.
Introduction: Early onset hypertension is often overlooked despite its association with lifelong cardiovascular disease risk. We developed an interpretable machine learning model to predict early onset hypertension in US adults using the 2022 Medical Expenditure Panel Survey, incorporating demographic, socioeconomic, health care utilization, and comorbidity features. Methods: We analyzed data on adults aged 55 years or younger. Hypertension was defined by International Classification of Diseases, 10th Revision, Clinical Modification code number I10 or self-reported physician diagnosis at or before age 55 years. Eighty-four variables were screened; Least Absolute Shrinkage and Selection Operator reduced the set to 45 variables. Extreme gradient boosting was trained with stratified 5-fold cross-validation. Performance metrics included accuracy, precision, recall, F2 score, area under the receiver operating characteristic curve (ROC AUC), and area under the precision-recall curve (PR AUC). We assessed fairness across subgroups and used Shapley Additive Explanations to quantify feature influence. Results: Among 9,850 adults, 1,647 (weighted prevalence, 15.2%) had hypertension. Discrimination was strong, with a ROC AUC of 0.807 (95% CI, 0.787-0.827), PR AUC of 0.486 (95% CI, 0.439-0.530), recall of 0.848 (95% CI, 0.816-0.878), and F2 score of 0.606 (95% CI, 0.578-0.634). Key predictors included older age, high body mass index, and a diagnosis of high cholesterol. Fairness analyses showed slightly higher true positive rates for Hispanic (+0.11) and Asian (+0.08) people, with performance for African American people modestly lower than that for White people. Conclusion: Both medical and nonmedical predictors contributed meaningfully, with high cholesterol particularly influential among adults older than 40 years and race and ethnicity especially important among African American adults. Prospective validation and routine fairness audits are warranted to ensure equitable benefit across populations. This interpretable model supports recall-oriented screening to identify younger adults at risk of hypertension for further evaluation and early intervention.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) conducted a cross-sectional in Early onset hypertension (n=9,850). Extreme gradient boosting (XGBoost) machine learning model using 45 demographic, socioeconomic, health care utilization, and comorbidity features was evaluated on Early onset hypertension (defined by ICD-10-CM code I10 or self-reported physician diagnosis at or before age 55 years) (ROC AUC 0.807, 95% CI 0.787-0.827). Extreme gradient boosting using 45 medical and nonmedical features strongly predicted early onset hypertension in US adults ≤55 years (ROC AUC 0.807; 95% CI 0.787-0.827; recall 0.848).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: