Robust free-living arterial stiffness trend tracking from 25-Hz wearable photoplethysmography under motion artifacts and missingness

Consumer-grade wearable photoplethysmography (PPG) commonly operates at low sampling rates and is frequently affected by motion artifacts, making the reliable localization of fine-grained morphological fiducial points difficult. In particular, dicrotic-notch localization, which is used by several arterial-stiffness indices, becomes unstable under sampling quantization, while derivative-based metrics can amplify discretization and measurement noise. We therefore investigated Tail Entropy (TE), an integral-like distributional descriptor originally developed for cuffless blood pressure estimation, as a complementary feature for PWV-related within-person trend analysis. TE characterizes the amplitude distribution of the diastolic runoff and remains computable when a distinct dicrotic notch cannot be reliably identified. In a linear mixed-effects model with subject-specific random intercepts, a one-standard-deviation increase in TE was associated with a 0.139 m/s increase in reference PWV (95% CI: 0.038–0.241 m/s, p = 0.0072 , FDR-adjusted q = 0.0408 ). Crest time and RI also remained significant after FDR correction, indicating that TE did not uniformly outperform conventional indices. The TE coefficient remained directionally stable across the tested temporal-alignment and diastolic-boundary settings. These pilot-scale findings support TE as a complementary distributional descriptor for within-person hemodynamic trend analysis under low-rate wearable sensing, but they do not establish TE as an independent arterial-stiffness marker or a clinical-grade estimator of absolute PWV.