Request access to the OSA-COVID sleep dataset

Access is granted immediately once you complete the form and accept the terms.

This dataset contains clinical sleep recordings. By requesting access you agree to use it for non-commercial research or education only and not to attempt to re-identify participants.

Log in or Sign Up to review the conditions and access this dataset content.

OSA-COVID: Home Sleep Sound and Physiological Recordings for Sleep Apnea Detection in Long COVID

Dataset summary

This dataset contains synchronised smartphone sleep-sound recordings and Level III home sleep apnea test (HSAT) recordings from adults with long COVID. It covers 100 nights from 59 participants (nearly 700 hours) recorded in their own homes in the Sheffield and Greater Manchester areas (UK) between March and September 2022.

The dataset supports research on:

  • acoustic detection and screening of obstructive sleep apnea (OSA) in home environments
  • multimodal (audio + cardiorespiratory) sleep apnea detection
  • night-to-night variability of sleep-disordered breathing
  • sleep and breathing in long COVID

The dataset is described in the accompanying data descriptor (see Citation).

Participants

Participants were adults with long COVID and a STOP-Bang score of 3 or higher. Of 80 recruited participants and 170 recorded nights, 59 participants and 100 nights passed quality control.

Characteristic Value Range / %
Participants 59 —
  Male 29 49%
  Female 30 51%
Age (years) 41 ± 11 22 – 68
BMI (kg/m²) 29 ± 6 15 – 50
AHI (events/hour) 7 ± 8 0 – 56
Recording duration (hours) 7.0 ± 1.1 4.9 – 10
Nights 100 —
  Normal (AHI < 5) 58 58%
  Mild (5 ≤ AHI < 15) 30 30%
  Moderate (15 ≤ AHI < 30) 10 10%
  Severe (AHI ≥ 30) 2 2%
Event type Count (%)
Obstructive apnea 2,241 (47.5%)
Hypopnea 2,021 (42.9%)
Central apnea 452 (9.6%)
Total 5,024

Dataset structure

.
├── patients.csv        # demographic information per participant
├── nights.csv          # one row per recorded night, with technician comments
├── delays.csv          # audio ↔ HSAT offset (seconds) per night
└── data/
    └── <participant_id>/            # e.g. COV01003
        └── <YYYY-MM-DD>/            # one folder per night
            ├── audio/               # consecutive 2-minute audio segments
            ├── hst_analysed/        # processed SOMNOtouch channels and scored events
            ├── hst_raw/             # unprocessed SOMNOtouch sensor data
            └── hst_synched/         # raw HSAT segments aligned with the audio

Participant IDs and night folders link the metadata tables to the recordings.

Metadata tables

The night column in nights.csv and delays.csv is a key of the form <participant_id>/<YYYY-MM-DD>, which is also the night's folder path under data/. The date is the evening the recording started.

patients.csv: one row per participant (59 rows).

Column Description
Participant Participant identifier (e.g. COV01003); matches the folder name under data/
Device Smartphone used for the audio recording, as reported by the participant (free text)
Gender Female / Male
Age Age in years at the time of recording
Weight (kg) Body weight in kg
Height (cm) Height in cm
BMI Body mass index (kg/m²)
Ethnicity Self-reported ethnic group (UK census-style categories, some free text)
STOP Bang STOP-Bang OSA screening score (0–8); all participants scored ≥ 3
Sleeping Partner Whether the participant shared the bed (Yes / No); empty = not recorded

nights.csv: one row per recorded night (100 rows), with event counts, AHI and technician comments.

Column Description
night Night key, <participant_id>/<YYYY-MM-DD>
Comments Technician comments on the scoring, e.g. sensor dropouts with clock times, or subclinical events
Type complete, or incomplete if a sensor (flow or SpO₂) was lost for part of the night; see Comments
Audio Time (h) Duration of the audio recording, in hours
Sleep Time (h) Duration of the HSAT analysis period, in hours (denominator for AHI)
Apnoea Number of scored apneas (obstructive + central)
Hypopnea Number of scored hypopneas
A+H Apnoea + Hypopnea
AHI Apnea–hypopnea index, A+H / Sleep Time (h) (events per hour)
Apnoea (report), Hypopnea (report), A+H (report), AHI (report) The same measures as given in the SOMNOmedics analysis report (TODO: confirm), which can differ from the counts above
Avg Apnoea Dur (s), Avg Hypopnea Dur (s) Mean event duration in seconds (0 = no events of that type)
Max Apnoea Dur (s), Max Hypopnea Dur (s) Longest event duration in seconds (0 = no events of that type)

delays.csv: one row per night (100 rows), giving the offset used to align the audio with the HSAT recording.

Column Description
night Night key, <participant_id>/<YYYY-MM-DD>
delay Offset between the audio and HSAT clocks, in seconds. TODO: sign convention, e.g. "HSAT time = audio time + delay"

Audio (audio/)

  • Smartphone (iOS or Android) placed next to the bed at head level, recorded with a purpose-built Sleep Study App
  • Mono, 16 kHz, 16-bit, in consecutive 2-minute segments
  • File naming: <ID>_<StartTime>_<EndTime>_<Index>.<ext>, where times are YYYY-MM-DD-HH-MM-SS-mmm
  • The first 20 minutes and the last 2 minutes of each night were excluded, because participants were usually awake then

Processed HSAT channels (hst_analysed/)

These are the processed outputs of the SOMNOmedics SOMNOtouch RESP system, exported as text files named <Channel> - <DD.MM.YYYY>.txt. All timestamps in the processed channels are are already adjusted using the delay.

Group Channel Source Description
Respiratory Effort 1 Thorax Thoracic respiratory effort
Effort 2 Abdomen Abdominal respiratory effort
Flow Flow Processed respiratory airflow
Cheyne Stokes Flow, Effort 1–2 Detected Cheyne–Stokes respiration
Cardiovascular Cardio complex Pleth Detected cardiac complexes
Cardio rhythm Cardio complex Continuous heart rhythm
Heart Rate Curve Pleth Instantaneous heart rate
HRV LF / HRV HF Cardio rhythm Low- / high-frequency heart rate variability
SVB Pleth Stroke volume beat signal
Activity & position Acti Profile Activity Activity profile from accelerometer
Activity Accu Overall movement level
Position Position Body position
Sleep & arousal Arousal Multi-sensor Arousal events
Autonomic arousals Cardio rhythm, HRV Autonomic arousal events
Classification Arousal Arousal Arousal classification labels
Sleep–Wake Multi-sensor Sleep vs. wake
Oximetry & snoring SpO₂ SpO₂ Peripheral oxygen saturation
SpO₂ Events SpO₂ Desaturation events
Snore Snore Snoring events
Other Increase 1 / Increase 2 Thorax / Abdomen Discrete effort-related events
Markers — Event markers and annotations

Raw HSAT channels (hst_raw/) and synchronised segments (hst_synched/)

Group Channel Sampling rate In hst_synched
Respiratory Thorax 32 Hz ✓
Abdomen 32 Hz ✓
Flow 256 Hz ✓
Sum Effort 32 Hz ✓
Cardiovascular Pleth 128 Hz ✓
Pulse 4 Hz ✓
Activity & motion Accu 1 / 30 s ✗
Activity 16 Hz ✗
Position 1 Hz ✗
Oximetry & snoring SpO₂ 4 Hz ✓
Snore 256 Hz ✓

hst_synched/ holds the raw HSAT data cut into segments that match the audio segments, using the offsets in delays.csv.

How to use

You must be logged in and have accepted the access conditions on this page. Then authenticate with huggingface-cli login or the HF_TOKEN environment variable.

Load the metadata tables:

from datasets import load_dataset

patients = load_dataset("<org>/<repo>", "patients", split="train")
nights   = load_dataset("<org>/<repo>", "nights",   split="train")
delays   = load_dataset("<org>/<repo>", "delays",   split="train")

The audio_samples, flow_events and spo2 configs are small previews from a single night (participant COV01003, 2022-03-30), for the dataset viewer. They are not the full dataset.

Download everything, or a single participant:

from huggingface_hub import snapshot_download

# full dataset
path = snapshot_download("<org>/<repo>", repo_type="dataset")

# one participant only
path = snapshot_download(
    "<org>/<repo>", repo_type="dataset",
    allow_patterns=["*.csv", "data/COV01003/**"],
)

Ethics and privacy

Written informed consent was obtained from all participants, including consent to record sleep sounds and to wear HSAT sensors overnight. Data collection, storage and management were approved by the University of Sheffield Ethics Committee (approval number 044761). Participants are identified only by study codes.

Access requests: the information you submit in the access form is shared with the dataset maintainers. It is used only to keep a record of who has access and to contact users about the dataset.

License

Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0). Commercial use is not permitted.

Citation

If you use this dataset, please cite the data descriptor and the dataset:

@article{roadabike_osa_covid,
  title   = {Home-collected sleep sound and physiological recordings for sleep apnea detection in long COVID patients},
  author  = {Roa-Dabike, Gerardo and Brown, Guy J. and Manuel, Ari and Iyer, Sriram and Ma, Ning},
  journal = {TODO},
  year    = {TODO},
  doi     = {TODO}
}
Downloads last month
46