Detecting rapid elevations in blood pressure using a novel pulse arrival time and machine learning technique

The purpose of this study was to test the initial feasibility of an acute hypertension detection platform (AHDP) for wearable devices that may be useful for the rapid detection of malignant hypertensive crises. The overall hypothesis was that the AHDP could detect laboratory-simulated elevations in blood pressure (BP). 42 healthy young participants (21.3 ± 2.1 yr, 16 males and 26 females) rested and then performed isometric handgrip exercise with blood flow restriction to elevate BP. Electrocardiogram (ECG), BP, and photoplethysmography (PPG) were assessed continuously. Pulse arrival time (PAT) was calculated with ECG and PPG using a proprietary algorithm and was standardized to z-scores. The PAT z-scores were used to classify normotensive and elevated BPs with logistic regression, and the optimal PAT z-score that classified BP was calculated. Elevated BPs were those that exceeded the baseline central tendency by +3 SD. The precision-recall area under the curve (PR-AUC) was used to quantify model performance. K-fold cross-validation was used to assess performance variability. Compared with baseline, BP increased during exercise (Δ18.8 ± 15.1 mmHg, P < 0.001) and PAT decreased (Δ-5.6 ± 5.7 ms, P < 0.001). The PR-AUC of the model was 0.68 with 95% confidence intervals (95% CIs) of 0.53 to 0.81, which was greater than random chance (P < 0.001). No sex differences were observed in performance (P = 0.075). The optimal PAT z-score threshold was -3.87 with 95% CIs of -5.04 to -2.15. The data support that the AHDP successfully detected laboratory-simulated elevated BP. The further development of the AHDP is expected to lead to effective hypertensive crisis detection systems for wearable devices.NEW & NOTEWORTHY The initial feasibility of an acute malignant hypertension detection platform using pulse arrival time and machine learning was assessed for cuffless blood pressure devices. The acute hypertension detection platform was sensitive to relatively small laboratory-simulated elevations in blood pressure and was generalizable among individuals with no sex differences in performance. The novel acute hypertension detection platform may be useful for detecting malignant hypertensive crises in wearable devices such as smartwatches.