Dataset guide · Updated Nov 2026

Sleep datasets for staging and sleep-disordered breathing

These recordings pair overnight physiology with sleep-stage or breathing-event labels. Use them to train or test staging and apnea models. Wrist-only staging sets remain on the wearable sleep staging guide; this page also covers lab and home PSG cohorts from NSRR and PhysioNet, including six NSRR cards added in November 2026 (WSC, SOF, NCHSDB, HomePAP, PATS and MNC). Counts and licences below come from the dataset cards on this site. A dataset is not a diagnosis, and a leaderboard is not a clinical ranking.

Interactive · counts from the dataset cards

How large are the sleep datasets on this page?

SHHS (NSRR release)5,804NCHSDB3,673MrOS Sleep (NSRR)2,911Cleveland Family Study2,284MESA Sleep2,237Wisconsin Sleep Cohort1,500STAGES1,500PATS555CHAT464SOF sleep visit461HomePAP373HeartBEAT318DREAMT100PSGWatch43Sleep-Accel31

SHHS (NSRR release): 5,804 subjects. Signals: PSG. Card key fact: 5,804 adults in the NSRR release (6,441 enrolled).

Counts are taken from each dataset card on ChatPPG (subjects field, key facts, or summary). They are not re-counted from the raw files for this figure. MNC is omitted because the card does not list a subject count. Sleep-EDF Expanded, HMC and Apnea-ECG are recording counts, shown when you include recordings. Click a bar for the source line.

Interactive · tags from the dataset cards

Which sleep datasets carry PPG, actigraphy or apnea labels?

9 of 19 listed datasets have ppg / pleth.

Tags follow each dataset card: tasks (sleep-staging, sleep-apnea, sleep-wake), signals (ppg, accel, actigraphy), and licence (open only when the card lists ODC-By or CC BY). Pediatric is marked only when the card names a pediatric or childhood cohort (NCHSDB, PATS, CHAT). Access still requires the host agreement; “open licence” is not the same as “no application.”

Top picks and when to use each

  1. #1Sleep Heart Health Study (SHHS, NSRR)Best for: Large adult NSRR cohort for sleep-disordered breathing research

    Sleep Heart Health Study. Card: 6,441 adults aged 40 or older enrolled in 1995–1998; NSRR release listed as 5,804. Unattended PSG, with a second study in a subset and cardiovascular outcomes followed to 2011.

    Gated (NSRR Data Access and U…Original host
  2. #2MESA SleepBest for: Multi-ethnic adult cohort with PSG, fingertip PPG and wrist actigraphy

    MESA Sleep ancillary study on NSRR. Card: 2,237 subjects; signals include PSG, PPG, ECG, EEG, respiration and accelerometer. Sleep-staging and AHI labels.

    Gated (NSRR Data Access and U…Original host
  3. #3DREAMTBest for: Raw wrist PPG with technician-scored PSG stages

    DREAMT on PhysioNet. Card: 100 participants, Empatica E4 BVP, accelerometer, EDA and temperature plus PSG. Sleep-clinic cohort, not a healthy-sleeper sample.

    Gated (PhysioNet Restricted H…Original host
  4. #4Sleep-EDF Database Expanded (PhysioNet)Best for: Open, downloadable overnight PSG with manual hypnograms

    Sleep-EDF Database Expanded on PhysioNet. Card: 197 whole-night recordings with EEG, EOG and chin EMG; some also have respiration and temperature. Hypnograms scored with the Rechtschaffen and Kales manual.

    Open (ODC-By 1.0)Original host
  5. Wisconsin Sleep Cohort on NSRR (new card, 2026-11). Card: 1,500 recruited; EEG, EOG, ECG, SpO2, respiration and PSG; tasks include sleep staging and sleep apnea.

    Gated (NSRR Data Access and U…Original host
  6. NCH Sleep DataBank on NSRR (new card, 2026-11). Card: 3,984 clinical pediatric PSGs on 3,673 patients (2017–2019). Pediatric recordings; non-commercial terms on the card's access notes for this site.

    Gated (NSRR Data Access and U…Original host

Comparison table

42 datasets. Swipe sideways on mobile. Values come from each dataset record; open a dataset for sources. Popularity is cached at build time: OpenAlex citations of the source paper (or the dataset DOI when no paper is linked), results on ChatPPG leaderboards, and GitHub stars or Zenodo / Hugging Face downloads where the host reports them. Hover for the source; a dash means unknown.

SignalsSamplingReference / labelsLicenceAccess
Sleep Heart Health Study (SHHS, NSRR)6,441PSG, ECG, EEG, resp, SPO2——NSRR Data Access and Use AgreementGated (NSRR Data Access and Use Agreement)1,095 cited8 results
MESA Sleep2,237PSG, PPG, ECG, EEG, resp256 Hzsleep staging, AHINSRR Data Access and Use AgreementGated (NSRR Data Access and Use Agreement)1,095 cited21 results
DREAMT100PPG, accel, skin temp, EDA, PSG64Hz, 32Hz, 4HzAHI, sleep stagingPhysioNet Restricted Health Data License 1.5.0Gated (PhysioNet Restricted Health Data License 1.5.0)17 cited8 results
Sleep-EDF Database Expanded (PhysioNet)—PSG, EEG, resp, temp——Open Data Commons Attribution License v1.0 (ODC-By…Open (ODC-By 1.0)1,261 cited
Wisconsin Sleep Cohort (WSC, NSRR): longitudinal adult in-lab PSG cohort oversampled for sleep apnea (1,500 recruited)1,500EEG, eog, ECG, SPO2, resp——See hostGated (NSRR Data Access and Use Agreement; non-commercial)1 cited
NCH Sleep DataBank (NCHSDB, NSRR): 3,984 clinical pediatric PSGs on 3,673 patients with linked EHR (2017-2019)3,673PSG, EEG, ECG, SPO2, resp——See hostGated (NSRR Data Access and Use Agreement; non-commercial; pediatric)4 cited
MrOS Sleep Study (Osteoporotic Fractures in Men, NSRR)—PSG, ECG, EEG, resp, SPO2——See hostSee host0 cited2 results
Study of Osteoporotic Fractures sleep visit (SOF, NSRR): in-home PSG in 461 women aged 65+ (2002-2003)461EEG, eog, ECG, SPO2, resp——See hostSee host0 cited
HomePAP (NSRR): RCT of home limited-channel testing vs in-lab PSG for OSA diagnosis and CPAP titration (373 enrolled)373PSG, EEG, eog, ECG, SPO2——See hostSee host0 cited1 result
PATS (NSRR): Pediatric Adenotonsillectomy Trial for Snoring, PSG + 7-day wrist actigraphy in children 3-12 with mild SDB555PSG, EEG, eog, ECG, SPO2——See hostSee host0 cited
Mignot Nature Communications (MNC, NSRR): multi-cohort PSG used for neural-network staging and type-1 narcolepsy detection—PSG, EEG, eog, ECG, SPO2——See hostSee host0 cited
Cleveland Family Study (CFS)2,284PSG, PPG—sleep staging, AHISee hostSee host0 cited4 results
CHAT (Childhood Adenotonsillectomy Trial, NSRR): paediatric OSA PSG at baseline and 7-month follow-up464EEG, eog, ECG, PPG, SPO2200 Hz, 128 Hz, 32 Hzsleep_stage, apneaSee hostSee host0 cited1 result
STAGES1,500PSG, EEG, ECG, resp, SPO2—sleep staging, AHISee hostSee host0 cited
HeartBEAT318PSG, PPG, ECG, resp, SPO275 Hz, 200 Hz, 3 HzAHI, abpmSee hostSee host0 cited
ABC (Apnea, Bariatric surgery, and CPAP study, NSRR): PSG in 49 adults with class II obesity and severe OSA49EEG, eog, ECG, PPG, SPO2—sleep staging, AHISee hostSee host0 cited
CCSHS (Cleveland Children's Sleep and Health Study, NSRR): in-lab PSG with finger pleth in a pediatric/adolescent cohort (907 children)907EEG, eog, ECG, PPG, SPO2—sleep staging, AHISee hostSee host0 cited
APPLES (Apnea Positive Pressure Long-term Efficacy Study, NSRR): repeated PSG in a 6-month sham-controlled CPAP RCT (1,516 enrolled)1,516EEG, eog, ECG, SPO2, resp—sleep staging, AHISee hostSee host0 cited
Motion and heart rate from a wrist-worn wearable and labeled sleep from polysomnography31accel, HR—sleep stagingODC-By 1.0Open283 cited
Haaglanden Medisch Centrum Sleep Staging Database (HMC, PhysioNet)—ECG, PSG——See hostSee host28 cited
Apnea-ECG Database (PhysioNet)—ECG, resp, SPO2——See hostSee host19 cited
PSGWatch43accel—sleep stagingCC BY 4.0See host0 cited
Newcastle PSG + GENEActiv both wrists (sleep clinic, 2015)28accel—sleep stagingCC BY 4.0See host17 cited2 results1,187 downloads
EESM1920accel—sleep stagingCC0 1.0See host0 cited
AAUWSS13PPG, ECG, accel, skin temp, EDA64 Hz, 200 Hzsleep stagingSee hostSee host1 cited765 downloads
BIDSleep47accel, HR, EEG0.2 Hzsleep stagingOpen Data Commons Attribution License v1.0 (ODC-By…Open2 cited
Human Sleep Project (BDSP)90,000PSG, PPG200 Hzsleep stagingSee hostSee host26 cited
Wearanize+ OA130PPG, accel, skin temp, EDA, EEG64 Hz, 32 Hz, 4 Hzsleep stagingSee hostSee host6 cited28 stars
ZU-PWD '2528accel, skin temp, EDA, resp, HRminute-level aggregates (not raw wavefor…sleep staging, stress, resp rate, mwlSee hostSee host0 cited
ISRUC-Sleep: overnight PSG with two-expert sleep staging (Coimbra University Hospital, 118 subjects / 126 recordings)118EEG, eog, ECG, resp, SPO2—sleep_stageSee hostSee host—
DualSleep: overnight thigh + lower-back accelerometry with PSG sleep stages (NTNU, DataverseNO)—accel, temp——See hostSee host15 cited
Boston Children's Hospital Sleep Corpus v1.0.1—PSG, ECG, SPO2200 Hzsleep stagingSee hostSee host1 cited
HPP Sleep (Human Phenotype Project 009-sleep)7,132PPG, SPO2, accel, resp, audionot stated on the dataset pageAHI, sleep staging, SPO2See hostSee host—
Sleep Health Data Commons (SHDC v1.0.0, BDSP)~157,000 PSG patients; ~127,…PSG, EEG—sleep stagingSee hostSee host0 cited
Stanford Sleep Bench v1.012,794PSG, ECG, SPO2128 Hzsleep stagingSee hostSee host2 cited
LifeSnaps: 4-month in-the-wild Fitbit Sense + EMA + surveys (71 participants, EU H2020 RAIS)71HR, skin temp, SPO2, respiratory_rate, sleep_stages——See hostSee host4 cited2,186 downloads
PMData: 5-month sports + lifelogging set, Fitbit Versa 2 + PMSys (16 participants, Simula)16HR, activity, sleep_stages——See hostSee host72 cited
PVDOMICS: pulmonary vascular disease phenomics (home sleep apnea test) (NSRR)536hsat, SPO2, PPG, ECG, resp—sleep stagingSee hostSee host—
St Vincent's / UCD Sleep Apnea Database (UCDDB, PhysioNet)—ECG, PSG, SPO2, resp——See hostSee host26 cited
MIT-BIH Polysomnographic Database (SLPDB, PhysioNet)—ECG, PSG, BP, SPO2, resp——See hostSee host8 cited
CAP Sleep Database (PhysioNet)—ECG, PSG, SPO2, resp——See hostSee host13 cited1 result
Dreem Open Datasets (DOD-H / DOD-O)—PSG, ECG——See hostSee host2 cited4,460 downloads

Benchmark results

Reported results from papers in the ChatPPG hub. Protocols differ, so only compare rows on the same dataset and metric.

DatasetMetricMethodResultProtocol
Cleveland Family Study (CFS)Intraclass correlation (agreement of a continuous estimate, e.g. AHI) ↑OxiNet0.92Site: PSG finger; SpO₂ used: yes (SpO₂ only)
Cleveland Family Study (CFS)Macro-averaged F1 ↑OxiNet0.78Site: PSG finger; SpO₂ used: yes (SpO₂ only)
MESA SleepAHI estimation R² (fraction of variance) ↑Dense temporal FCN + metamodel0.83Site: PSG finger; SpO₂ used: yes (auxiliary)
MESA SleepAHI estimation R² (fraction of variance) ↑Dense temporal FCN + metamodel0.76Site: PSG finger; SpO₂ used: no
MESA SleepEvent-level AUROC (apnea / hypopnea events) ↑Dense temporal FCN0.92Site: PSG finger; SpO₂ used: no
MESA SleepIntraclass correlation (agreement of a continuous estimate, e.g. AHI) ↑OxiNet0.94Site: PSG finger; SpO₂ used: yes (SpO₂ only)
MESA SleepMacro-averaged F1 ↑OxiNet0.75Site: PSG finger; SpO₂ used: yes (SpO₂ only)
MESA SleepMacro-averaged F1 ↑Dense temporal FCN + metamodel0.71Site: PSG finger; SpO₂ used: yes (auxiliary)
MESA SleepMacro-averaged F1 ↑2CNN-2LSTM on PPG features + SpO₂ + apnea burden0.617Site: PSG finger; SpO₂ used: yes
MrOS Sleep Study (Osteoporotic Fractures in Men, NSRR)Intraclass correlation (agreement of a continuous estimate, e.g. AHI) ↑OxiNet0.94Site: PSG finger; SpO₂ used: yes (SpO₂ only)
MrOS Sleep Study (Osteoporotic Fractures in Men, NSRR)Macro-averaged F1 ↑OxiNet0.80Site: PSG finger; SpO₂ used: yes (SpO₂ only)
Sleep Heart Health Study (SHHS, NSRR)F1, sleep-disordered-breathing event class ↑SleepMaMi0.485Site: PSG multimodal; SpO₂ used: yes (PSG)
Sleep Heart Health Study (SHHS, NSRR)Intraclass correlation (agreement of a continuous estimate, e.g. AHI) ↑OxiNet0.96Site: PSG finger; SpO₂ used: yes (SpO₂ only)
Sleep Heart Health Study (SHHS, NSRR)Macro-averaged F1 ↑OxiNet0.84Site: PSG finger; SpO₂ used: yes (SpO₂ only)
Sleep Heart Health Study (SHHS, NSRR)AHI mean absolute error (events/h) ↓SleepMaMi8.40Site: PSG multimodal; SpO₂ used: yes (PSG)
MESA SleepAHI estimation R² (fraction of variance) ↑Dense temporal sleep apnea profiling from…0.76 [0.66, 0.82]PPG-only; 2029 MESA PSG studies; abstract states AHI estimation 'explained 76% of variance'; note: 0.83 when an auxiliar…
MESA SleepMacro-averaged F1 ↑Dense temporal sleep apnea profiling from…0.71 [0.66, 0.76]SpO2-assisted SA severity classification from single-channel PPG pipeline; PPG+auxiliary input; macro-F1 95% CI [0.66, 0…
CAP Sleep Database (PhysioNet)Cohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net20.57—
CHAT (Childhood Adenotonsillectomy Trial, NSRR): paediatric OSA PSG at baseline and 7-month follow-upCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓wav2sleep0.793—
Cleveland Family Study (CFS)Area under ROC ↑ECG-SDI0.977external cohort, never used for training or model selection (726 recordings); single-lead ECG resampled to 100 Hz; train…
Cleveland Family Study (CFS)Macro-averaged F1 ↑SleepMaMi71.2Linear probe, CFS held out from pretraining; acc 80.7, κ 72.9; arXiv v3 Table 1
DREAMTArea under ROC ↑Multimodal Self-Supervised Learning for…0.72DREAMT 4-class staging; SSL PPG+ACC pretrain (23,000 h unlabeled) + causal Mamba head; vs supervised 0.59; abstract-only…
DREAMTCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓WatchSleepNet0.684 ± 0.051—
DREAMTCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓Multimodal SSL PPG+ACC + causal Mamba0.30 (supervised 0.06)—
DREAMTCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↑Multimodal Self-Supervised Learning for…0.3DREAMT 4-class; vs supervised 0.06; abstract-only
HomePAP (NSRR): RCT of home limited-channel testing vs in-lab PSG for OSA diagnosis and CPAP titration (373 enrolled)Cohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net20.69—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net0.724—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net20.74 (κc)—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓wav2sleep0.742—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓BM-DTS0.65—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net20.75 (IQR 0.67–0.81)—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓SleepPPG-Net0.75—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓DCA-Sleep0.652 (F1 0.731)—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓wav2sleep0.78—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓Bakker et al. 20210.68—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓Random Forest, PPG morphology + IBI0.65—
MESA SleepCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓Walch NN trained on Apple Watch → MESA subcohort0.40 (acc 0.69)—
Sleep Heart Health Study (SHHS, NSRR)Area under ROC ↑ECG-SDI0.908internal held-out SHHS test participants (participant-level 6:2:2 split of SHHS+MESA+MrOS, 10,652 recordings); 95% CI 0.…
Sleep Heart Health Study (SHHS, NSRR)Cohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓sleep2vec0.75 (acc 83.0, MF1 65.9); APPLES ext. 0.61—
Sleep Heart Health Study (SHHS, NSRR)Cohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↓wav2sleep ECG-only0.739—
Sleep Heart Health Study (SHHS, NSRR)Macro-averaged F1 ↑SleepMaMi74.1Linear probe on frozen SleepMaMi embeddings, SHHS1 test split (Phan et al. 2021 partition); acc 81.9, κ 70.0; arXiv v3 T…
DREAMTAccuracy ↑Addressing Wearable Sleep Tracking Inequity0.857Same model/protocol; 0.857 ± 0.016
DREAMTArea under ROC ↑Addressing Wearable Sleep Tracking Inequity0.926Same model/protocol; 0.926 ± 0.016
DREAMTCohen's kappa (sleep staging agreement). Alias: `kappa` (kept in Rank-axis / axis_id for slug stability; exporter canonicalizes to `cohen_kappa`) ↑Addressing Wearable Sleep Tracking Inequity0.694Same model/protocol; Table 2 0.694 ± 0.025 (abstract rounds to 0.695)
DREAMTClass F1 for one stated class (e.g. SDB events), not macro ↑Addressing Wearable Sleep Tracking Inequity0.823GPBoost + apnea-severity random effect + LSTM post-processing; participant-level 5-fold CV on 80 participants (16 test/f…
MESA SleepAccuracy ↑Cole–Kripke sleep–wake scoring0.882Task Night&Day test set n=363; LSTM seq_len=100 accuracy 88.2 (paper Table 2); Device Algorithm baseline ~76.6; Manual A…
Newcastle PSG + GENEActiv both wrists (sleep clinic, 2015)Class F1 for one stated class (e.g. SDB events), not macro ↑Sadeh sleep–wake scoring0.7393Held-out test n=24 participants; RF sleep-wake F1 73.93%; trained on combined clinic PSG studies (abstract n=134); 30 s …
Newcastle PSG + GENEActiv both wrists (sleep clinic, 2015)Specificity at a stated threshold / operating point ↑40 years of actigraphy in sleep medicine and…0.475van Hees 2015 (forearm-angle heuristic); all 28 Newcastle PSG participants, per-subject metrics averaged (Table 1): acc …

See all leaderboards

Preprocessing pitfalls

  • AHI, REI and ODI are not interchangeable. A home-test REI divides by recording time; ODI counts oxygen dips only. State which index a label file uses.
  • Hypopnea rules change the event count. A 3% desaturation-or-arousal rule and a 4% desaturation rule can move the same night into a different severity band.
  • Subject-wise splits matter. Random epoch splits leak the same person into train and test, especially in large NSRR cohorts.
  • Pediatric and older-adult cohorts are not interchangeable with mid-life community samples. NCHSDB, PATS and CHAT are pediatric; SOF and MrOS are older adults.
  • NSRR datasets require the NSRR Data Access and Use Agreement. WSC, NCHSDB and PATS are treated here as non-commercial. Commercial-use terms for SOF, HomePAP and MNC were not verified for these pages; read the current NSRR agreement before any commercial use.
  • DREAMT and several PhysioNet health-data sets are restricted. An open PhysioNet page is not the same as a licence that allows unrestricted redistribution.
  • MNC has no subject count on the card; do not invent one. Sleep-EDF Expanded is counted in recordings (197), not subjects.

FAQ

What is a sleep dataset?

In this guide it is a public or application-gated collection of overnight recordings with sleep-stage or breathing-event labels, usually polysomnography or a limited-channel home study, sometimes paired with a wearable. The list is for research use. It is not a clinical database of patients.

How is this different from the wearable sleep staging datasets page?

The wearable sleep staging guide is limited to sets that pair a wearable signal (PPG, heart rate or actigraphy) with staging labels. This page also includes lab and home PSG cohorts used for apnea and staging research, including NSRR studies that have no wrist file.

Which new NSRR cards were added?

WSC (Wisconsin Sleep Cohort), SOF (Study of Osteoporotic Fractures sleep visit), NCHSDB (NCH Sleep DataBank, pediatric), HomePAP (home limited-channel testing versus in-lab PSG), PATS (Pediatric Adenotonsillectomy Trial for Snoring, with actigraphy) and MNC (Mignot Nature Communications multi-cohort PSG). All six are gated under an NSRR agreement.

Can I use NSRR data commercially?

Not by default. WSC, NCHSDB and PATS are treated here as non-commercial. Commercial-use terms for SOF, HomePAP and MNC were not verified on these pages. Read the current NSRR Data Access and Use Agreement for the cohort you want, and do not assume a DUA allows commercial training.

Which sets are open to download without an NSRR application?

On the cards used here, Sleep-EDF Expanded and Sleep-Accel are offered under ODC-By, and PSGWatch under CC BY 4.0. DREAMT uses the PhysioNet Restricted Health Data License. Always check the host page; licences can change.

Where are the benchmarks?

Sleep-staging results are on /research/leaderboards/sleep-staging. AHI estimation results are on /research/leaderboards/sleep-apnea-ahi. Those boards compare published papers. They are not diagnostic rankings and not buying advice.

Sources