Abstract
OBJECTIVE: Primary aldosteronism (PA) is a leading cause of secondary hypertension yet is largely underdiagnosed. To develop and validate an electronic health record-based prediction model to identify individuals with hypertension who are at high risk of PA. METHODS: Using a dataset of 4564 people screened for PA at an academic health system (April 2018 to January 2025), we divided the cohort into 3650 discovery and 914 validation cohorts. We evaluated 74 candidate predictors and applied machine learning methods on the discovery cohort to construct the model. The final logistic regression model with 15 variables was validated in the validation cohort. RESULTS: Among 4564 people, 456 (10.0%) had PA. PA patients were older (61.8 vs 57.9 years, P < .0001), more likely to be Black (22.9% vs 8.1%, P < .0001), had higher systolic blood pressure (148 vs 144 mmHg, P = .001), and required more antihypertensive medications (3.8 vs 3.0, P < .0001). Our model achieved an area under the receiver operating characteristic curve of 0.7 for the validation cohort. Predictors included race (P < .0001), hypokalemia (P < .0001), ≥4 antihypertensive medications (P < .0001), obstructive sleep apnea (P = .004), higher serum bicarbonate (P < .001), and reduced glomerular filtration rate (P < .001). The model showed 29% PA in the top 10% highest-risk subgroup, achieving 2.7-fold enrichment with 27% recall and 29% precision. Expanding to the top 20% highest-risk subgroup improved recall to 41% while maintaining high specificity (82%). CONCLUSION: This electronic health record-based prediction model achieved moderate discriminative performance, identifying primary aldosteronism in 29% of patients within the highest-risk decile, suggesting potential clinical utility for prioritizing screening efforts.