Screening for primary aldosteronism (PA) in unselected patients with hypertension remains difficult in routine clinical practice. A derivation-validation study published in the European Journal of Preventive Cardiology evaluated a new risk stratification algorithm, PAstrat, alongside machine learning models for identifying patients requiring PA screening.

The analysis included 15,507 patients aged 18-65 years with confirmed hypertension who underwent complete diagnostic evaluation for PA screening. PAstrat was developed using clinically defined patient risk groups to estimate the likelihood of PA across subgroups. Logistic regression and XGBoost models were also evaluated, with external validation performed in an independent cohort of 768 patients.

In the derivation cohort, the area under the receiver operating characteristic curve was 0.80 for PAstrat, 0.82 for logistic regression, and 0.83 for XGBoost. Similar performance was observed in the validation cohort, with values of 0.79, 0.82, and 0.82, respectively. Only 6 of 768 patients (0.7%) in the validation cohort were incorrectly classified as low risk by PAstrat, corresponding to 4.2% of patients with confirmed PA.

Logistic regression and XGBoost demonstrated higher false-negative rates than PAstrat, with 49 (6.7%) and 84 (10.9%) false negatives, respectively. These findings suggest that PAstrat demonstrated good screening sensitivity and a high negative predictive value while maintaining clinical interpretability.