Reliability and inter-device agreement of multi-site heart rate variability measurements during transcutaneous auricular vagus nerve stimulation: a prospective single-arm methodological physiological study

Transcutaneous auricular vagus nerve stimulation (taVNS) has been associated with acute changes in cardiac autonomic measures; however, agreement between electrocardiography-derived heart rate variability (HRV) and photoplethysmography-derived pulse rate variability (PRV) obtained from different anatomical sites across a taVNS protocol remains unclear. This study evaluated the reliability, inter-device agreement, and pre–post measurement patterns of chest-, fingertip-, and upper-arm-derived HRV/PRV measurements. This prospective, single-center, single-arm methodological physiological study included 40 healthy volunteers. Electrocardiography-based chest measurements were obtained with Polar H10 as the reference device, while photoplethysmography-based measurements were obtained simultaneously from the fingertip using CorSense and from the upper arm using Polar Verity Sense. Bilateral taVNS was applied to the cymba conchae for 10 min. Five-minute recordings were performed before and immediately after stimulation. Heart rate (HR), RR interval, RMSSD, SDNN, low frequency (LF), high frequency (HF), and LF/HF were analyzed. Inter-device agreement was assessed using ICC(A,1), Bland-Altman analysis, and Lin’s concordance correlation coefficient. Time, device, and time-by-device effects were examined using linear mixed models. In the Polar H10 reference measurements, HR was significantly lower at post-stimulation (Hodges–Lehmann estimate = − 1.50 beats/min, 95% CI − 3.00 to − 0.50; p = 0.019), whereas RR interval (Hodges–Lehmann estimate = 15.11 ms, 95% CI 1.35 to 31.36; p = 0.033), RMSSD (Hodges–Lehmann estimate = 2.67 ms, 95% CI 0.12 to 6.62; p = 0.045), and SDNN (mean difference = 6.33 ms, 95% CI 2.00 to 10.65; p = 0.005) were significantly higher. Agreement was excellent for HR (ICC range 0.980–0.996) and RR interval (ICC range 0.952–0.998), whereas RMSSD and SDNN agreement varied by sensor pair. For RMSSD, H10-Verity Sense showed stronger agreement than H10-CorSense (ICC range 0.863–0.882 vs. 0.354–0.490; CCC range 0.861–0.881 vs. 0.351–0.486). Frequency-domain agreement was less consistent, and no time-by-device interaction was significant for any HRV parameter. Systolic blood pressure was significantly lower at post-stimulation (mean difference = − 7.0 mmHg, 95% CI − 9.7 to − 4.2; p < 0.001), and ΔRMSSD was moderately inversely correlated with Δsystolic blood pressure (Spearman ρ = −0.418; p = 0.007). Pre–post autonomic changes were broadly consistent across chest, fingertip, and upper-arm measurements, although inter-device agreement varied by HRV/PRV parameter. HR and RR interval showed excellent agreement, whereas RMSSD, SDNN, and frequency-domain indices were less consistent. Given the single-arm design, these findings should be interpreted as physiological measurement changes rather than causal effects of taVNS. ClinicalTrials.gov: NCT07548723. Registered 22 February 2026.