oscar-etl

Extract clean CSVs from your OSCAR CPAP data.

---

Why this exists

The manual OSCAR export gives you three flat CSV dumps -- a daily summary, per-session stats, and a massive interleaved details file mixing pressure readings with apnea events in a narrow format. It only includes three signals (pressure, EPAP, flow limitation), omits leak, respiratory rate, tidal volume, minute ventilation, and snore entirely, and uses BiPAP column headers that are always zero for your APAP machine. Session boundaries are also only available through the OSCAR GUI -- the CSVs don't make it easy to correlate events with specific mask-on segments.

oscar-etl reads the same raw EDF files OSCAR does, but produces cleaner, more complete output. It splits sessions by each mask-on segment (matching OSCAR's internal model), gives you seven signals in a wide-format timeseries CSV (one row per 2-second timestamp instead of one row per measurement), and separates events into their own file for easy correlation with sleep-tracker data. Daily stats are recomputed from the raw signal data across all sessions -- not approximated from the longest session or averaged. The whole thing runs in about a minute with no manual clicking.

Quick Start

pip install oscar-etl

oscar-etl

Expected output:

  oscar-etl v0.1.0

  Data source
  -----------------------------------
  Location:  /Users/you/OSCAR_Data
  Profile:   Your Name
  Machine:   ResMed_23192261096
  EDF files: 2,847

  Extracting
  -----------------------------------
  Discovering sessions           ━━━━━━━━━━━━━━━━ 100%
  Parsing EDF files              ━━━━━━━━━━━━━━━━ 100%
  Writing sessions.csv           ━━━━━━━━━━━━━━━━ 100%
  Writing daily.csv              ━━━━━━━━━━━━━━━━ 100%
  Writing events.csv             ━━━━━━━━━━━━━━━━ 100%
  Writing timeseries.csv         ━━━━━━━━━━━━━━━━ 100%

  Output
  -----------------------------------
  oscar-etl-output/
    cpap_sessions.csv             42 KB
    cpap_daily.csv                18 KB
    cpap_events.csv               1.2 MB
    cpap_timeseries.csv           812.4 MB

  Done in 58s

Four CSV files land in ./oscar-etl-output/. See Output Files for what each contains.

macOS Setup

On macOS, OSCAR stores data in ~/Documents/OSCAR_Data/, which is protected by the OS. You need to grant Terminal Full Disk Access temporarily to move the data somewhere unprotected. There are two options.

Option A: Symlink (recommended)

This moves your data out of the protected Documents folder and symlinks it back so OSCAR still finds it. You only need to do this once, and afterward no special permissions are required.

1. Open System Settings > Privacy & Security > Full Disk Access and enable Terminal.

2. Move the data and create a symlink:

   mv ~/Documents/OSCAR_Data ~/OSCAR_Data
   ln -s ~/OSCAR_Data ~/Documents/OSCAR_Data

3. Open System Settings > Privacy & Security > Full Disk Access and disable Terminal.

4. Verify it works:

   ls ~/OSCAR_Data

oscar-etl now reads from ~/OSCAR_Data permanently, no special permissions needed. OSCAR itself continues to work because the symlink points it to the new location.

Option B: Leave Full Disk Access on

If you prefer not to move files around, just leave Full Disk Access enabled for Terminal. oscar-etl will read directly from ~/Documents/OSCAR_Data/.

Windows and Linux

No special setup is needed. oscar-etl auto-discovers the OSCAR data directory in the standard locations.

Output Files

All output goes to ./oscar-etl-output/ by default (override with --output-dir).

cpap_sessions.csv

One row per mask-on segment.

Column	Description
date	Night date (see evening date convention)
session_start	ISO 8601 timestamp when mask went on
session_end	ISO 8601 timestamp when mask came off
duration_minutes	Session length in minutes
ahi	Apnea-hypopnea index for this session
ca_count	Central apnea count
oa_count	Obstructive apnea count
h_count	Hypopnea count
ua_count	Unclassified apnea count
arousal_count	Arousal/RERA count
pressure_median	Median pressure (cmH2O)
pressure_95	95th percentile pressure
pressure_995	99.5th percentile pressure
leak_median	Median leak rate (L/min)
leak_95	95th percentile leak rate
resp_rate_median	Median respiratory rate (breaths/min)
tidal_vol_median	Median tidal volume (mL)
minute_vent_median	Median minute ventilation (L/min)

cpap_daily.csv

One row per night, aggregated from all sessions for that date.

Column	Description
date	Night date
sessions	Number of mask-on segments
start	Earliest session start
end	Latest session end
total_minutes	Total therapy time across all sessions
ahi	AHI across all sessions
pressure_median, pressure_95, etc.	Same stat columns as sessions, recomputed across all sessions for the night

cpap_events.csv

One row per event annotation recorded by the machine. This includes therapy events counted in AHI (Obstructive Apnea, Central Apnea, Hypopnea, Apnea, Arousal) as well as other annotations the machine records (Large Leak, CSR, etc.). The session and daily CSVs only count the five therapy event types in their AHI and event count columns.

Column	Description
datetime	ISO 8601 timestamp of the event
date	Night date the event belongs to
session_start	The session this event occurred in
event	Event type (e.g., Obstructive Apnea, Large Leak, Hypopnea)
duration_sec	Duration of the event in seconds

cpap_timeseries.csv

One row per 2-second sample, wide format. This file is large (typically around 800 MB for a year of data). Use --skip-timeseries to skip it.

Column	Description
datetime	ISO 8601 timestamp
date	Night date
session_start	The session this sample belongs to
pressure	Therapy pressure (cmH2O)
leak	Leak rate (L/min)
resp_rate	Respiratory rate (breaths/min)
tidal_vol	Tidal volume (mL)
minute_vent	Minute ventilation (L/min)
snore	Snore index
flow_limit	Flow limitation index

Evening date convention

Sessions starting before noon are attributed to the previous calendar day. For example, a session that starts at 1:00 AM on March 16 belongs to the "night of March 15." This matches OSCAR and most sleep tracker conventions.

Shift workers or anyone with a non-standard sleep schedule can adjust the cutoff with the --day-boundary flag. For example, --day-boundary 6 attributes sessions starting before 6 AM to the previous day.

CLI Reference

oscar-etl [OPTIONS]

Flag	Description
--oscar-dir PATH	Path to OSCAR_Data directory. Auto-discovered from standard locations if omitted.
--output-dir PATH	Output directory. Default: ./oscar-etl-output/
--profile NAME	Profile name. If omitted and multiple profiles exist, an interactive picker is shown.
--machine SERIAL	Machine serial number filter.
--skip-timeseries	Skip the large timeseries CSV (~800 MB).
--day-boundary HOUR	Hour (0--23) that separates sleep nights. Default: 12.
--version	Show version and exit.

Supported Machines

ResMed only, tested with AirSense 10 AutoSet.

Other ResMed models that use the same EDF-based data format should work. The tool warns on unrecognized signal labels rather than crashing, so you can try it and see what happens.

PRs welcome from users with other machines (Philips, Fisher & Paykel, etc.) who can build and test against their own data.

Contributing

Contributions are welcome. Here is a quick orientation for the codebase:

• Signal label mapping lives in src/oscar_etl/etl.py (PLD_SIGNAL_MAP). This is the dictionary that maps ResMed EDF signal labels to the normalized column names used in the output CSVs.
• EDF parser lives in src/oscar_etl/edf.py. It is format-agnostic and reads any valid EDF file.
• To add support for a new machine, you would add signal label mappings for that machine's EDF signal names and test against real data from that machine.

Issues and PRs are welcome on GitHub. If you are adding support for a new machine, please include sample output (anonymized) so reviewers can verify correctness.

License

MIT -- see LICENSE for details.