Development and Temporal Validation of a Multinomial Prediction Model for Phenotypes of Undiagnosed Hypertension in...

Background/Objectives: Early diagnosis of hypertension (HTN) is critical, but most screening models do not simultaneously distinguish phenotypes based on systolic or diastolic patterns. We developed and temporally validated a multinomial model to predict normotension and three phenotypes of undiagnosed hypertension in Peru. Methods: We used ENDES 2017-2019 for development (final analytic n = 62,091) and ENDES 2021-2024 for temporal validation (final analytic n = 77,372), excluding 2020 due to COVID-19 disruptions. We included adults aged ≥18 years without self-reported HTN. The outcome was classified as normotension, isolated diastolic hypertension (IDH), isolated systolic hypertension (ISH), or systolic-diastolic hypertension (SDH). Eight nonlaboratory predictors were used: age, BMI, sex, residential altitude, smoking, alcohol consumption, vegetable intake, and fruit intake. Results: The model achieved an AUC of 0.789 (95% CI: 0.783-0.795) in training and 0.776 (95% CI: 0.770-0.781) in temporal validation. The prevalence of undiagnosed hypertension was 11.6% in the training set and 12.6% in the validation set. At a prespecified cutoff of 0.1004, sensitivity and specificity were 79.0% and 63.2% in training and 78.7% and 60.9% in validation, respectively (NPV 95.8% and 95.2%). Decision curve and clinical impact analyses suggested a positive net benefit and plausible referral volumes across a range of thresholds. Conclusions: This model could help prioritize confirmatory blood pressure measurements in resource-limited settings.