Machine learning analysis of home blood pressure parameters predicted abnormal nighttime blood pressure features with an accuracy of 0.882 and an AUC of 0.972.
Observational (n=1,129)
Yes
Does machine learning analysis of home blood pressure monitoring parameters predict abnormal nighttime blood pressure features in patients with prehypertension and stage 1 hypertension?
Machine learning applied to routine home blood pressure monitoring parameters can accurately predict abnormal nighttime blood pressure patterns, potentially identifying high-risk patients without requiring 24-hour ambulatory monitoring.
Abstract Objective Abnormal nighttime blood pressure (BP) feature including early morning surge (EMS), nocturnal hypertension (NH) and non-dipping systolic blood pressure (NDSBP) are predictors for cardiovascular events (CVE) and mortality. This study aimed to try to determine which of the home blood pressure monitor (HBPM) parameters can predict abnormal BP feature. Methods The study population consisted of participants with prehypertension and stage 1 hypertension, recruited from 11 medical centers within the Taiwan Hypertension-Associated Cardiac Disease Consortium (TCHC). We selected participants who underwent both HBPM and 24-hour ambulatory blood pressure monitoring (ABPM) within a six-month period. Using machine learning, we analyzed various detailed home blood pressure measurement parameters, including morning blood pressure, evening blood pressure, morning-evening blood pressure difference, morning-evening blood pressure average, and various blood pressure variability indicators. Separate analyses were conducted for systolic blood pressure, diastolic blood pressure, pulse pressure, and mean arterial pressure to identify which home blood pressure parameters could predict nocturnal hypertension. Results A total of 1,129 patients were enrolled. For those with a history of previous cardiovascular disease (CVD), the difference between morning and evening average blood pressure was the best predictor of abnormal blood pressure features. On the other hand, for patients without a history of CVD, the maximum evening pulse pressure was the most predictive of abnormal blood pressure features. Using a machine learning-assisted model, the accuracy was 0.882, precision was 0.889, F1 score was 0.889, and AUC was 0.972. Conclusion Our results highlight the potential of machine learning-based Abnormal BP feature prediction using HBPM. The accuracy of the predictions demonstrated that HBPM feature that can predict EMS, NH and NDSBP.
Lee et al. (2025) conducted an observational in Prehypertension and stage 1 hypertension (n=1,129). Home blood pressure monitor (HBPM) parameters was evaluated on Abnormal nighttime blood pressure features (early morning surge, nocturnal hypertension, and non-dipping systolic blood pressure). Machine learning analysis of home blood pressure parameters predicted abnormal nighttime blood pressure features with an accuracy of 0.882 and an AUC of 0.972.