Skip to content

Kubios-Compatible Output

How to configure RRational so that the HRV results match Kubios HRV Scientific output as closely as possible (typically ±5% on LF/HF for properly matched segments). For the validation evidence, see Validation.

TL;DR

  1. Open the Analysis tab → Group Analysis (or Sequence Comparison)
  2. Set Frequency-domain pipeline to "Kubios-compatible (absolute ms², 4 Hz interpolation, 180 s Welch)"
  3. Use the same artifact correction in both tools (Kubios "Automatic correction" ↔ RRational nn_correction.method: kubios in the .rrational file)
  4. Run the analysis. LF/HF values are now in absolute ms² — directly comparable to Kubios

Step-by-step

1. Match the artifact correction

Kubios offers three beat correction modes:

Kubios setting What RRational matches
none (0.3) Leave nn_correction.method empty in your .rrational file
Automatic correction (0.3) Use the Kubios-style correction in RRational (Participants tab → "Apply NN correction")
Threshold-based Not directly supported — use Automatic instead

In the .rrational file, check the per-section nn_correction block:

nn_correction:
  method: kubios
  corrected_at: '2026-01-19T16:24:30.093241'
  intervals_corrected: 2

If method is none, RRational will compute HRV on raw RR intervals — matching the Kubios "none" beat correction mode.

2. Switch to Kubios-compatible frequency pipeline

In the Analysis tab, the new "Frequency-domain pipeline" dropdown appears under each analysis mode (Group Analysis, Sequence Comparison).

Choose: - "NeuroKit2 default" for new projects (normalized PSD, NK2 reference) - "Kubios-compatible" when you want absolute ms² and Kubios cross-comparison

The setting persists in your session and applies to all subsequent analyses until you change it.

3. Use segments of comparable length

Kubios's default Welch window is 180 seconds with 50% overlap. RRational's Kubios mode replicates this exactly. To get the most reliable PSD estimates:

  • Recommended segment length: ≥ 5 minutes (gives 2+ Welch windows)
  • Optimal segment length: ≥ 10 minutes (gives 5+ Welch windows, lower variance)

Shorter segments will work but produce noisier LF/HF estimates. This is a fundamental Welch-PSD limitation, not specific to RRational.

4. Verify with a known case

Take one participant whose RR data you've also processed with Kubios. Compute HRV in both tools with matching settings. Expected agreement (from our validation):

Metric Expected Δ
MeanNN < 1%
RMSSD < 2%
HF (ms²) < 5%
LF (ms²) < 10%
LF/HF < 10%
SDNN 30–60% higher in RRational (by design — see below)

The pNN50 difference (important!)

You may see large pNN50 discrepancies between Kubios and RRational only when beat correction is active in either tool. Example: Kubios with "Automatic correction (0.3)" reports pNN50 = 16.45% while RRational reports 49.86% on the same nominal segment.

This is not a bug. pNN50 is a threshold-based binary counter (only diffs > 50 ms count) — every single corrected beat that lands just above or below the 50 ms cutoff flips the count. Continuous metrics (RMSSD, LF, HF) remain stable because they integrate over all diffs.

Quantitative evidence (Rohr et al. 2024):

Metric Error sensitivity (% per ms noise SD)
LF 0.24%
SDNN 0.61%
HF 0.71%
RMSSD 1.57%
pNN50 2.75% (highest!)

Rohr et al. recommend pNN50 "be used with caution". Berntson & Lozano 2005 (DOI) showed mathematically that RMSSD acts as a robust high-pass filter while pNN50's discrete cutoff makes it fragile.

Recommendations:

  1. Prefer RMSSD over pNN50 for parasympathetic activity — same physiological information, ~4× more robust to artifact correction
  2. For Kubios-comparable pNN50: Use BeatCorrection: none in both tools (RRational then reports identical pNN50 to ±2%)
  3. For publication: Always state the exact beat correction pipeline if you report pNN50 (Quigley et al. 2024 reporting guidelines)
  4. Alternative: Report pNN20 (Mietus et al. 2002) — less floor effect, more robust to single corrected beats

The SDNN difference (important!)

You will see a noticeable SDNN difference between Kubios and RRational. This is not a bug — it's two different definitions:

  • RRational SDNN: Standard deviation of raw NN intervals after artifact correction. This is the Task Force 1996 definition and matches every textbook formula.
  • Kubios SDNN: Standard deviation of the detrended interpolated signal (after Smoothness Priors λ=500 detrending applied to 4 Hz resampled RR). This is a Kubios-proprietary variant.

The Kubios SDNN is typically 30–60% lower than the standard SDNN, because detrending removes low-frequency variance.

Recommendation for papers: Always state which SDNN definition you used. RRational reports the international standard (Task Force 1996). If you need the Kubios variant for direct comparison with Kubios output, mention this explicitly in your methods section.

Frequency band differences

Band RRational Kubios mode RRational NeuroKit2 default Kubios default
VLF 0.0–0.04 Hz 0.0033–0.04 Hz 0.0–0.04 Hz
LF 0.04–0.15 Hz 0.04–0.15 Hz 0.04–0.15 Hz
HF 0.15–0.40 Hz 0.15–0.40 Hz 0.15–0.40 Hz
ULF — 0–0.0033 Hz —
VHF — 0.4–0.5 Hz —

Kubios does not report ULF / VHF by default.

What if my values still don't match?

Common causes of remaining discrepancies (in order of likelihood):

  1. Different segment boundaries: Verify both tools analyze the exact same time range. A 1-second shift can change LF/HF by 5%.
  2. Different beat correction outcome: Even with both set to "Kubios-style", the corrected NN intervals can differ slightly because of implementation details. Open the .rrational file and check intervals_corrected vs Kubios's S{N}_Beats corrected (%).
  3. Short segments (< 5 min): PSD variance dominates — you'll see ±10–20% differences purely from finite-window effects.
  4. Different RR series: If you used different export options from your recording device (e.g., raw vs Kubios-format), the NN intervals themselves differ.

Caveats

  • The freq_method setting is stored in session state and is not persisted to the project YAML or .rrational files. If you close and reopen RRational, the default reverts to neurokit.
  • Existing cached group_analysis_results.yml files were computed with whichever freq_method was active at the time. Re-run the analysis after switching modes to refresh cached values.
  • The Participants tab's PSD plot uses a separate computation (see analysis_plots.py). It does not yet honor the freq_method setting — this affects only the displayed graph, not the metrics in the tables.

References