Skip to contents

Analysing Fitbit and Google Pixel data

Source: vignettes/fitbit.Rmd
fitbit.Rmd
library(PhysioDevices)
#> Loading required package: PhysioCore

Fitbit exports your data as an account archive (Settings → Export Your Account Archive) or as the Fitbit slice of Google Takeout — a directory of per-day JSON (and some CSV) files. Google Pixel Watch stores its health data in Fitbit, so the same reader covers Pixel Watch. The analysis then reuses the same PhysioWearable / PhysioECG functions as the Apple Watch workflow.

1. Ingest the archive 

fb <- readFitbit("Takeout/Fitbit", tz = "Asia/Tokyo")
fb   # which modalities were found, with row counts and spans

readFitbit() discovers files by name (heart_rate-*, steps-*, sleep-*, resting_heart_rate-*, SpO2), parses the nested JSON, and skips anything it cannot read (with a warning). Restrict with what = if you only need some modalities.

2. Heart rate 

hr <- fitbitSeries(fb, "heart_rate")     # columns: time, bpm (intraday)
summary(hr$bpm)
fb$resting_heart_rate                    # daily resting HR

For HRV, Fitbit exports its own daily RMSSD (a Heart Rate Variability file); for beat-to-beat HRV use a Fitbit Sense/Charge ECG recording with PhysioECG.

3. Blood oxygen (SpO2) 

PhysioWearable::spo2Metrics(fb$spo2$value, time = fb$spo2$time)

4. Sleep

Fitbit scores sleep stages itself (wake/light/deep/rem). summarizeSleepStages() turns them into per-night clinical metrics — the same engine used for Apple Watch, just with Fitbit’s stage labels:

PhysioWearable::summarizeSleepStages(
  fb$sleep,
  asleep_levels = c("light", "deep", "rem"),
  wake_levels   = "wake",
  stage_cols    = c(light = "light", deep = "deep", rem = "rem"))

fb$sleep_summary          # Fitbit's own per-log efficiency / minutes asleep / time in bed

5. Activity 

steps <- fitbitSeries(fb, "steps")
tapply(steps$value, as.Date(steps$time), sum)     # steps per day

For epoch-level free-living metrics (ENMO, intensity, fragmentation, HAR → ICF) you need raw accelerometry, which Fitbit does not export — only derived activity.

Google Fit and Health Connect

Phone-level Google data has two other paths, both reusing the same analysis:

# Google Takeout "Fit": granular heart-rate/step JSON + daily-summary CSVs
gf <- readGoogleFit("Takeout/Fit", tz = "Asia/Tokyo")
gf$heart_rate
gf$daily
# Android Health Connect: a per-record-type CSV export
hc <- readHealthConnect("health_connect_export", tz = "Asia/Tokyo")
PhysioWearable::spo2Metrics(hc$spo2$value, time = hc$spo2$time)
PhysioWearable::summarizeSleepStages(
  hc$sleep, asleep_levels = c("light", "deep", "rem"), wake_levels = "awake",
  stage_cols = c(deep = "deep", rem = "rem"))

Scope and limitations

  • Covers the Fitbit account archive / Google Takeout JSON layout; exact filenames vary slightly by export version (the reader keys off filename patterns and skips what it cannot parse).
  • The Fitbit Web API (OAuth2, live JSON) is a separate path not covered here.
  • Google Pixel Watch = Fitbit data (readFitbit()).
  • Google Fit (readGoogleFit(), Takeout) is being retired by Google in favour of Health Connect (readHealthConnect(), per-type CSV export). Health Connect has no single official export file, so column names vary by exporter; the reader detects them heuristically.
  • No raw acceleration and no continuous SpO2 (spot readings) — same limits as the Apple Watch workflow.