JMIR Publications

ABSTRACT

Background:

Consumer wearable devices may support low-burden cardiovascular screening, but wearable-derived pulse wave velocity (W-PWV) is obtained from peripheral photoplethysmography signals and cannot be assumed to be numerically interchangeable with carotid-femoral pulse wave velocity (cfPWV), the reference measure of aortic stiffness.

Objective:

This study aimed to evaluate the screening performance, agreement, calibration behavior, and performance boundaries of consumer W-PWV for identifying elevated arterial stiffness defined by cfPWV.

Methods:

We conducted a real-world, prospective, cross-sectional method-comparison study in hospital and community settings. Adults who completed consumer wearable PWV measurement were enrolled. The primary wearable output was derived from a wrist-worn Xiaomi device, and a Huawei device was used for supportive cross-device analysis. Same-session cfPWV was measured as the reference standard. Elevated arterial stiffness was defined as cfPWV ≥10 m/s. We evaluated discrimination using receiver operating characteristic analysis, bootstrap optimism correction, and repeated 5-fold cross-validation. Individual-level agreement and error structure were assessed using Pearson correlation, intraclass correlation coefficient (ICC), concordance correlation coefficient (CCC), mean absolute error (MAE), Deming regression, Bland-Altman analysis, and calibration analyses across predefined cfPWV domains.

Results:

Among 222 enrolled participants, 206 had cfPWV measurements and 165 had complete same-session paired W-PWV and cfPWV data. In the paired analytical cohort, the mean age was 57.9 (SD 16.4) years, 75 participants (45.5%) were male, and 57 participants (34.5%) had elevated arterial stiffness. W-PWV showed good discrimination for elevated arterial stiffness, with an apparent area under the curve (AUC) of 0.827 (95% CI 0.762-0.891), a bootstrap optimism-corrected AUC of 0.826, and a repeated 5-fold cross-validated mean AUC of 0.827. At the Youden-derived cutoff of 9.28 m/s, apparent sensitivity was 91.2% and specificity was 63.0%. Agreement with cfPWV was moderate (ICC 0.610, 95% CI 0.510-0.700; CCC 0.615), with an MAE of 1.141 m/s and wide Bland-Altman limits of agreement (-2.857 to 2.910 m/s). Deming regression showed proportional bias, with a slope of 0.576 (95% CI 0.476-0.694). Across cfPWV domains, W-PWV changed from overestimation at lower cfPWV levels to underestimation at cfPWV ≥10 m/s, where the mean bias was -1.20 m/s and the MAE was 1.40 m/s.

Conclusions:

Consumer W-PWV provided a scalable screening signal for elevated arterial stiffness but showed only moderate individual-level agreement and stiffness-dependent bias relative to cfPWV. W-PWV should not be interpreted as a numerical substitute for cfPWV; rather, it may be used as a fit-for-purpose digital vascular screening phenotype, with confirmatory cfPWV measurement considered near decision thresholds or in higher-risk clinical contexts.