HRV Analysis Guidelines¶
Practical guidelines for using RRational in line with current scientific standards.
Which guidelines does RRational follow?¶
RRational implements recommendations from:
- Quigley et al. (2024) — Current consensus on HRV measurement and reporting in psychophysiology
- Task Force (1996) — Foundational standards for HRV measurement
- Quintana et al. (2016) — GRAPH checklist for reporting HRV studies
- Lipponen & Tarvainen (2019) — State-of-the-art artifact detection algorithm
Data Quality¶
Artifact Detection¶
RRational uses the Lipponen-Tarvainen algorithm (via NeuroKit2's Kubios implementation) which classifies beats into six categories: ectopic, long, short, missed, extra, and normal.
flowchart TD
A[Run artifact detection] --> B{Artifact rate?}
B -->|≤ 2%| C[Excellent\nUse as-is]
B -->|2-5%| D[Good\nCorrect + analyze]
B -->|5-10%| E[Moderate\nCorrect, time-domain only]
B -->|> 10%| F[Poor\nExclude segment]
style C fill:#28a745,color:#fff
style D fill:#2E86AB,color:#fff
style E fill:#ffc107,color:#000
style F fill:#dc3545,color:#fffRecommended Workflow
- Run artifact detection on all validated sections (same segments)
- Review per-segment quality grades
- Exclude segments with > 10% artifacts
- Correct segments with 2–10% artifacts
- Report artifact rates in your publication
Quality Grades¶
RRational assigns quality grades following Quigley et al. (2024):
| Grade | Artifact Rate | Valid Metrics | Action |
|---|---|---|---|
| Excellent | ≤ 2% | All (time, frequency, nonlinear) | Use as-is |
| Good | 2–5% | All | Correct artifacts, then analyze |
| Moderate | 5–10% | Time-domain only | Correct, but avoid frequency metrics |
| Poor | > 10% | None reliably | Exclude from analysis |
Segments with fewer than 50 beats are excluded automatically regardless of artifact rate (too short for any reliable HRV metric).
Minimum Data Requirements¶
| Analysis | Minimum | Recommended | Why |
|---|---|---|---|
| Time-domain (RMSSD, SDNN) | 100 beats | 300+ beats | Statistical reliability |
| Frequency-domain (LF, HF) | 300 beats | 500+ beats | Spectral estimation needs sufficient data |
| Recording duration for frequency | 2 minutes | 5 minutes | 5 min is the Task Force (1996) short-term window; the 2 min minimum is RRational's practical floor |
Metric Interpretation¶
Recommended Metrics¶
| Metric | What it reflects | Recommendation |
|---|---|---|
| RMSSD | Parasympathetic (vagal) activity | Primary metric for short-term studies |
| SDNN | Total autonomic variability | Use with consistent segment durations |
| HF Power | Parasympathetic activity (0.15–0.4 Hz) | Corroborates RMSSD |
| SD1 | Short-term variability (Poincare) | Mathematically related to RMSSD |
Metrics to Use with Caution¶
LF/HF Ratio
The LF/HF ratio should not be interpreted as "sympathovagal balance." This interpretation is rejected by current consensus (Quigley et al., 2024; Billman, 2013). The LF band reflects both sympathetic and parasympathetic activity plus baroreflex function. RRational includes LF/HF for legacy compatibility, but we recommend against using it as your primary outcome.
SDNN Across Different Durations
SDNN scales with recording duration. Never compare SDNN values from segments of different lengths. Use consistent window sizes (e.g., always 5 minutes) within a study.
Statistical Testing (Group & Sequence Comparison)¶
When you compare groups or conditions, RRational selects an appropriate test automatically and reports an effect size alongside each p-value.
Test selection is driven by the number of groups and a normality check (Shapiro-Wilk) on each metric:
| Comparison | Distribution | Test | Effect size |
|---|---|---|---|
| 2 groups | normal | Welch's t-test (unequal variances) | Cohen's d |
| 2 groups | non-normal | Mann-Whitney U | — |
| 3+ groups | normal | one-way ANOVA | η² (eta-squared) |
| 3+ groups | non-normal | Kruskal-Wallis | η² |
Why effect sizes? A p-value indicates whether a difference is unlikely under the null hypothesis; it does not say how large the difference is. Always report an effect size (Cohen's d or η²) so readers can judge practical significance — current reporting guidelines (Quigley et al., 2024) require this.
Multiple comparisons. Testing many metrics at once inflates the false-positive rate. RRational adjusts p-values across the metrics in a comparison using Holm (default), Bonferroni, or Benjamini-Hochberg FDR. State which correction you used.
Log-normal metrics. Frequency-domain powers (LF, HF, VLF, Total Power) are right-skewed and approximately log-normal. RRational log-transforms them before parametric testing so the normality assumption holds, in line with common HRV practice.
Normality on small samples
The Shapiro-Wilk test has low power with very small groups, so a "normal" verdict on a handful of participants is weak evidence. With small or clearly skewed samples, prefer the non-parametric option and interpret parametric results cautiously.
Reporting Checklist¶
Following the GRAPH guidelines (Quintana et al., 2016), your publication should report:
- Recording device and sampling rate
- Recording duration per condition
- Artifact detection method and parameters
- Mean artifact rate per condition (with SD)
- Artifact correction algorithm
- Number of excluded segments and criteria
- HRV metrics computed, with window duration
- Beat count per analyzed segment
- Software and version used
RRational makes this easy
The Analysis Documentation panel (in Single Participant analysis) auto-generates a report with all these details. Export as HTML or Markdown and include in your supplementary materials.
Common Pitfalls¶
| Pitfall | Why it's a problem | What to do instead |
|---|---|---|
| Comparing SDNN across different durations | SDNN scales with recording length | Use identical segment lengths |
| Interpreting LF as "sympathetic" | Not supported by evidence | Report LF but don't over-interpret |
| Using LF/HF as sympathovagal balance | Rejected by current consensus | Focus on RMSSD and HF |
| Over-correcting artifacts (> 10%) | Distorts the signal | Exclude the segment instead |
| Frequency metrics on < 2 min segments | Insufficient spectral resolution | Use time-domain only |
| Ignoring respiratory confounds | Breathing rate affects HF | Note if breathing was uncontrolled |
| Not reporting artifact rates | Reviewers can't assess data quality | Always report per condition |
Key References¶
- Quigley, K.S., et al. (2024). Publication guidelines for human heart rate and heart rate variability studies in psychophysiology. Psychophysiology, 61(9), e14604. doi:10.1111/psyp.14604
- Task Force of ESC and NASPE (1996). Heart rate variability: Standards of measurement. Circulation, 93(5), 1043–1065. doi:10.1161/01.CIR.93.5.1043
- Quintana, D.S., et al. (2016). Guidelines for Reporting Articles on Psychiatry and Heart rate variability (GRAPH). Translational Psychiatry, 6(5), e803. doi:10.1038/tp.2016.73
- Lipponen, J.A., & Tarvainen, M.P. (2019). A robust algorithm for heart rate variability time series artefact correction. Journal of Medical Engineering & Technology, 43(3), 173–181. doi:10.1080/03091902.2019.1640306
- Makowski, D., et al. (2021). NeuroKit2: A Python toolbox for neurophysiological signal processing. Behavior Research Methods, 53, 1689–1696. doi:10.3758/s13428-020-01516-y
- Billman, G.E. (2013). The LF/HF ratio does not accurately measure cardiac sympatho-vagal balance. Frontiers in Physiology, 4, 26. doi:10.3389/fphys.2013.00026