Gradient Biotech

Longitudinal wearable HRV

Turn a long ECG, Holter, or wearable recording into a time-series of HRV instead of a single number — slide overlapping windows across the recording, track how autonomic state evolves, and compare temporal variability across groups.

Research question

How does HRV change over the course of a long recording, and does its level or its temporal variability differ across subjects or groups?

Who this is for

  • Researchers working with 24-hour Holter, overnight, or multi-hour wearable recordings
  • Digital health teams building longitudinal autonomic dashboards
  • Studies where when HRV changes matters as much as its average

Data requirements

DataRequiredPurpose
ECG waveform or RR seriesYesBeat detection and per-window HRV
Long recording (minutes to hours)RecommendedRun-level HRV hides within-recording change

Workflow

Upload → Preprocess (ECG) → Windowed HRV → Trend chart → Windowed cohort comparison (optional) → Interpret

Step 1 — Preprocess

Upload an ECG dataset and run Preprocess ECG (or use an RR CSV directly).

Step 2 — Windowed HRV

Open the Windowed HRV panel and click Compute windows. The pipeline slides overlapping windows (default 5-minute window, 1-minute step) and computes per-window metrics:

OutputDescription
Per-window HRVMean HR, SDNN, RMSSD, pNN50, Poincaré SD1/SD2
Quality / missingnessBeats per window, coverage, valid-window flag
Recording summaryMedian, IQR, and across-window CV per metric, plus a first-half vs second-half trend
Feature matrixA columns × windows matrix for export and cohort modeling

Step 3 — Inspect the trend

Use the longitudinal chart to see how RMSSD, SDNN, mean HR, or pNN50 move across the recording.

Step 4 — Windowed cohort comparison (optional)

In the cohort workspace, set Feature source → Windowed features to compare groups on per-subject medians and temporal variability (CV) — variability that run-level HRV can't express.

Step 5 — Interpret (optional)

Run interpretation; windowed metrics and trends are cited so the narrative can describe longitudinal change (e.g. a decline in RMSSD over the night).

Expected outputs

  • Per-window HRV feature matrix with quality/missingness
  • Recording-level median/IQR/CV and trend per metric
  • Longitudinal trend chart
  • Group comparison on windowed features (when run)

Example insight

RMSSD declines steadily across the overnight recording (second-half median 18% below the first half), and the treatment group shows higher across-window variability than controls.

Related guides