JMIR Publications

ABSTRACT

Background:

Cuffless blood pressure (BP) monitoring using wearable devices enables frequent measurement in everyday settings, yet its real-world reliability remains difficult to assess because of calibration dependency and the absence of paired reference BP measurements in uncontrolled settings.

Objective:

This study aimed to characterize measurement stability in watch-based BP monitoring using calibration difference as a reference-free analytical index and to examine how calibration stability varies according to user characteristics and measurement frequency.

Methods:

Calibration difference was defined as the absolute difference between the average cuffless systolic BP (SBP) values measured during the 1-week periods before and after calibration. We analyzed large-scale real-world data from 3 nationwide campaigns conducted between 2021 and 2023, comprising 1714 participants and 5728 calibration-difference instances obtained from Samsung Galaxy Watch measurements. Observed calibration-difference distributions were compared with theoretically derived reference ranges based on day-to-day SBP variability reported across multiple independent cohorts. Statistical analyses and machine learning–based interpretation were used to examine nonlinear and interaction effects associated with age, average SBP, hypertension status, and measurement frequency.

Results:

Among 5728 calibration-difference instances, the mean calibration difference was 5.95 (SD 4.73) mm Hg. Under the primary analytical assumption of 3 measurements per averaging period, calibration differences in the overall population and in younger or lower-SBP groups were broadly comparable to the theoretically derived reference ranges, whereas older individuals and those with higher average SBP consistently exhibited larger deviations. Age, average SBP, and the minimum number of measurements were independently associated with calibration difference in multivariable regression (all P<.01). Machine learning–based interpretation showed that the contributions of age and average SBP increased more prominently beyond approximately 55 years and 125 mm Hg, respectively. Elevated calibration differences were also observed among younger individuals with hypertension and older individuals without hypertension. Increasing measurement frequency reduced calibration difference according to a significant exponential decay model in the overall population (P<.01), with diminishing returns beyond 3 measurements, whereas older and higher-SBP subgroups continued to benefit from additional measurements.

Conclusions:

Calibration stability varied substantially across user groups, suggesting that uniform usage protocols may not be sufficient for watch-based BP monitoring in real-world settings. In settings where paired reference BP measurements are unavailable, calibration difference may serve as a scalable analytical index for examining measurement stability, rather than as a substitute for regulatory accuracy assessment. These findings provide a practical framework for evaluating real-world stability and optimizing measurement strategies for cuffless BP monitoring systems.