A Stratified Model for Enhanced Cuffless Blood Pressure Estimation Using Hypertension History

This study explores how clinical characteristics impact blood pressure (BP) estimation from wearable devices using ECG and PPG signals, revealing their role in causing long-term cardiovascular changes that influence estimation models. Four clustering algorithms were applied to divide the dataset into groups, revealing hypertension history as the primary determinant for separation. These findings suggest that even individuals with normalized BP but a previous history of hypertension exhibit distinct patterns that require tailored models. To address this heterogeneity, stratified models were developed for BP estimation and compared against baseline models using various training strategies and feature sets. Experimental results demonstrate that applied the stratified model improves the estimation results significantly for 36% on SBP and 33% on DBP compared with the baseline model, despite relying on relatively simpler machine learning models. These findings highlight the potential of this efficient approach to enhance BP estimation accuracy, contributing to the development of reliable and accessible cuffless BP monitoring systems using wearable technologies.