Inter-Electrode Pulse Wave Velocity: A Direct Method for Maternal Arterial Stiffness Assessment During Pregnancy Using Multi-Channel ECG

TL;DR

This study validates a direct inter-electrode PWV method using three-channel abdominal ECG to assess maternal arterial stiffness during pregnancy, avoiding LVET-based timing uncertainties. By detecting R-peaks on each channel, matching them across channel pairs, and computing PWV as PWV = $L/Δt$ with a fixed inter-electrode distance, the authors obtain physiologically plausible PWV values (~5–10 m/s) and demonstrate robust cross-channel consistency, temporal stability, and LVET-independence. Longitudinal data suggest divergent PWV trajectories between Control and prenatal Yoga groups, though significance is limited by sample size, indicating potential sensitivity of vascular stiffness to pregnancy progression and intervention. The method shows promise for continuous, non-invasive arterial stiffness monitoring using standard multi-channel ECG equipment, but requires validation against gold-standard measures, accurate electrode-distance quantification, and larger cohorts to confirm clinical utility and outcome associations.

Abstract

Objective: To validate a novel inter-electrode PWV method that directly measures pulse wave propagation between spatially separated electrodes, avoiding LVET estimation. Methods: We analyzed 43 multi-channel ECG recordings (3 channels, 1000 Hz) from the FELICITy 2 cohort (pregnant women, $\sim$19 and $\sim$35 weeks gestation). R-peaks were detected independently on each channel using an ensemble detector. Time lags ($Δt$) between matched R-peaks on electrode pairs were calculated, and PWV computed as PWV = $L$/$Δt$, where $L$ is an effective inter-electrode distance. Three channel pairs provided independent PWV estimates per recording. Temporal stability was assessed using sliding window analysis (1--15 minutes). To investigate whether $Δt$ reflects morphological distortion or vascular propagation, we performed signal origin analysis using three QRS fiducial points (R-peak maximum, QRS onset, maximum $|dV/dt|$) and two bandpass settings (0.5--40 and 0.5--100~Hz). Longitudinal changes were compared between Control (n=24) and prenatal Yoga intervention (n=20) groups. Results: Inter-electrode PWV yielded physiologically plausible values (Control First: 7.40$\pm$1.51 m/s, Control Last: 6.98$\pm$1.63 m/s; Yoga First: 7.10$\pm$2.15 m/s, Yoga Last: 8.16$\pm$0.91 m/s), consistent with literature values for aortic PWV (5--10 m/s). Temporal stability analysis demonstrated PWV stabilizes at 5 minutes (CV=12.3\%), with 2.6--5.2$\times$ better stability than heart rate and HRV metrics. Signal origin analysis showed that inter-electrode delays persisted across all QRS fiducial points (15--27~ms) and were insensitive to bandpass changes ($-$8.5\%, NS), arguing against pure morphological distortion; all conditions yielded PWV within 6.8--9.1~m/s. Preliminary group comparison suggests different trajectories (Control: --5.7\% decrease, Yoga: +14.9\% increase, p=0.07 for interaction). Conclusions: Inter-electrode PWV provides direct spatial measurement of pulse wave propagation with physiologically valid values, independent of LVET estimation. Signal origin analysis supports the robustness of this empirical surrogate across fiducial and filter conditions. Method shows promise for pregnancy arterial stiffness assessment using standard multi-channel ECG equipment. Further validation against gold-standard measures and accurate electrode distance determination are needed.

Figures (6)

• Figure 1: Electrode placement and inter-electrode distances for the three-channel abdominal ECG configuration. Skin–skin distances are measured in cm between electrodes. These distances illustrate that each channel pair has a distinct geometric separation. In the present analyses, we used a single effective inter-electrode distance $L = 7.5$ cm (see Section Methods \ref{['sec:PWV']}) to scale inter-electrode delays $\Delta t$ into PWV-like units; precise, pair-specific and subject-specific path lengths will be required for absolute PWV calibration in future work.
• Figure 2: Longitudinal Inter-Electrode PWV Trajectories by Group.Panel A: Individual subject trajectories (thin lines) and group means with standard error bars (thick lines with error bars) from First visit ($\sim$19 weeks gestation) to Last visit ($\sim$35 weeks gestation). Blue represents Control group ($n=11$ paired subjects), red represents Yoga intervention group ($n=9$ paired subjects). Individual subjects show varied responses while group trajectories diverge: Control decreases --5.7% (mean 7.40$\rightarrow$6.98 m/s), Yoga increases +14.9% (mean 7.10$\rightarrow$8.16 m/s). Group $\times$ Time interaction p=0.07 (trend-level significance). Panel B: Distribution of percent change in inter-electrode PWV. Violin plots with overlaid individual data points show Control group centered near 0% change (mean +0.6%, median --5.5%) while Yoga group shows positive shift (mean +18.6%, median +17.2%). Horizontal dashed line at 0% indicates no change. Independent samples t-test p=0.075, Cohen's d=0.467 (moderate effect size).
• Figure 3: Signal Origin Analysis: Fiducial and Bandpass Sensitivity.Panel A: Mean absolute inter-electrode delay ($|\Delta t|$) by QRS fiducial point at two bandpass settings. R-peak maximum (baseline method), QRS onset, and maximum $|dV/dt|$ (steepest slope) were compared. Error bars represent standard error. Max $|dV/dt|$ reduced delays by 27% ($p$=0.005), but substantial delays (15--17 ms) persisted. QRS onset delays were paradoxically larger (+15%, $p$=0.005). Dashed line near zero indicates predicted electromagnetic time-of-flight ($\sim$0.05 ms) for inter-electrode distances used. Panel B: Bandpass sensitivity for R-peak fiducial. Each point represents one subject's mean $|\Delta t|$ at 0.5--40 Hz vs 0.5--100 Hz. Strong correlation ($r$=0.66, $p$$<$0.001) and minimal systematic shift ($-$8.5%, NS) indicate delays are not driven by frequency-dependent phase distortion. Dashed line = identity; solid line = regression fit. Blue = Control, red = Yoga.
• Figure 4: Temporal Stability Analysis of Inter-Electrode PWV and Comparison Metrics.Panel A: Within-subject coefficient of variation (CV) by window size for inter-electrode PWV (green), heart rate (gray), RR interval (gray), SDNN (gray), and RMSSD (gray). PWV demonstrates superior temporal stability across all window sizes (CV range: 7.6--12.3%). Vertical dashed line indicates 5-minute stabilization point where PWV CV change from 2-minute window is $<$10% (1.4% relative change), meeting stabilization criterion. Red circle highlights PWV at 5 minutes. Panel B: Relative stability comparison at 5-minute window. Inter-electrode PWV (12.3% CV) is 2.6$\times$ more stable than heart rate (32.3% CV), 3.2$\times$ more stable than RR interval (39.0%), 5.2$\times$ more stable than SDNN (63.4%), and 5.1$\times$ more stable than RMSSD (62.7%). Fold-differences shown in red text. Panel C: Recording duration context. Both Control (45 min, 9$\times$ minimum) and Yoga (100 min, 20$\times$ minimum) groups substantially exceed the validated 5-minute minimum requirement (red bar), confirming that recording duration differences between groups do not confound PWV measurements.
• Figure S.1: BMI Adjustment Analysis.Panel A: Scatter plot showing relationship between timepoint-matched BMI (kg/m$^2$) and inter-electrode PWV (m/s). Data points distinguished by group (Control: blue circles, Yoga: red squares) and timepoint (First: semi-transparent, Last: opaque). Regression line (red dashed) shows weak negative correlation: r=--0.134, R$^2$=0.018 (1.8% variance explained), p=0.403. BMI explains minimal PWV variation. Panel B: Effect of BMI adjustment on statistical significance. Raw PWV (green) shows moderate effect size (Cohen's d=0.467) and trend-level significance (p=0.075, orange) for group differences. BMI-adjusted PWV shows reduced effect size (d=0.429) and worsened significance (p=0.091), crossing above $\alpha$=0.05 threshold (red dashed line). Arrow indicates direction of change. BMI adjustment decreases rather than improves statistical power.