Assessing Machine Learning for Diagnostic Classification of Hypertension Types Identified by Ambulatory Blood Pressure Monitoring

Background Inaccurate blood pressure classification results in inappropriate treatment. We tested if machine learning (ML), using routine clinical data, can serve as a reliable alternative to Ambulatory Blood Pressure Monitoring (ABPM) in classifying blood pressure status. Methods This study employed a multi-centre approach involving three derivation cohorts from Glasgow, Gdańsk, and Birmingham, and a fourth independent evaluation cohort. ML models were trained using office BP, ABPM, and clinical, laboratory, and demographic data, collected from patients referred for hypertension assessment. Seven ML algorithms were trained to classify patients into five groups: Normal/Target, Hypertension-Masked, Normal/Target-White-Coat, Hypertension-White-Coat, and Hypertension. The 10-year cardiovascular outcomes and 27-year all-cause mortality risks were calculated for the ML-derived groups using the Cox proportional hazards model. Results Overall XGBoost showed the highest AUROC of 0.85-0.88 across derivation cohorts, Glasgow (n=923; 43% females; age 50.7±16.3 years), Gdańsk (n=709; 46% females; age 54.4±13 years), and Birmingham (n=1,222; 56% females; age 55.7±14 years). But accuracy (0·57-0·72) and F1 scores (0·57-0·69) were low across the three patient cohorts. The evaluation cohort (n=6213, 51% females; age 51.2±10.8 years) indicated elevated 10-year risks of composite cardiovascular events in the Normal/Target-White-Coat and Hypertension-White-Coat groups, with heightened 27-year all-cause mortality observed in all groups except Hypertension-Masked, compared to the Normal/Target group. Conclusions Machine learning has limited potential in accurate blood pressure classification when ABPM is unavailable. Larger studies including diverse patient groups and different resource settings are warranted.