Task leaderboard

Sleep apnea (OSA) screening

Published wearable, PPG and oximetry sleep apnea results per dataset and metric (AHI agreement, AHI thresholds, severity class), with SpO₂ use and caveats on every row.

Last updated Oct 20, 2026

These are research results as reported by each study, shown for education. A wearable or model result is not a diagnosis; talk to a clinician about a sleep study if you have concerns about sleep apnea.

Task leaderboard

Sleep apnea (OSA) screening

DatasetMetric

13 reported results · Leaderboard page

Sleep apnea (OSA) screening: Stanford enriched cohort, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 100% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	100%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 100% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	97.4%
—	Samsung Galaxy Watch OSA feature, default eAHI 15 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 66.7% at the same threshold Not ranked: Shipped threshold; 38.8% lacked a valid eAHI on at least one lab night (commentary). Weights: not released	94.1%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 66.7% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.1%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 65.8% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	92.3%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 42.9% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	84.6%
—	Samsung Galaxy Watch OSA feature, cohort-optimised eAHI 25.95 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.9% at the same threshold Not ranked: Post hoc cutoff fitted on the same cohort. Weights: not released	82.4%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.9% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	82.4%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.7% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	80.8%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 83.3% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	80.6%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 82.4% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	79.5%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.4% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	79.2%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.1% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	78.1%

Rows are ranked only within this task × dataset × metric and one protocol group. Protocols can still differ in detail; read each method’s evaluation details before comparing.

Recent changes

Oct 20, 2026
SOTA star awarded to PaPaGei-S, 0.70
Sleep-disordered breathing: SDB (Garde et al.) AUROC

From the hub's dated sync history: new results as they are added, and SOTA stars as they are awarded or move.

All reported results for this task (42 benchmarks)

Rankings apply within each dataset and metric. Open evaluation details for differences in protocols.

Apple validation study: sensitivity (AHI ≥ 15)

Sleep apnea (OSA) screening: Apple validation study, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Apple Watch Sleep Apnea Notification Apple Watch Sleep Apnea Notification (accelerometer only, no PPG) Evaluation details · caveat Site: wrist; SpO₂ used: no; Specificity 98.5% at the same threshold Not ranked: Home HSAT reference; 30-day multi-night decision; prevalence-weighted; vendor whitepaper. Weights: not released	66.3%

Apple validation study: sensitivity (AHI ≥ 30)

Sleep apnea (OSA) screening: Apple validation study, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Apple Watch Sleep Apnea Notification Apple Watch Sleep Apnea Notification (accelerometer only) Evaluation details · caveat Site: wrist; SpO₂ used: no; Threshold: AHI ≥ 30 Not ranked: 164/184; HSAT reference; vendor whitepaper. Weights: not released	89.1%

Apple validation study: specificity

Sleep apnea (OSA) screening: Apple validation study, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Apple Watch Sleep Apnea Notification Apple Watch Sleep Apnea Notification (accelerometer only, no PPG) Evaluation details · caveat Site: wrist; SpO₂ used: no; Sensitivity 66.3% at the same threshold Not ranked: Home HSAT reference; 30-day multi-night decision; prevalence-weighted; vendor whitepaper. Weights: not released	98.5%

APPLES: AUROC

Sleep apnea (OSA) screening: APPLES, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	LPSGM LPSGM (EEG/EOG/EMG PSG model, no respiratory channels) Evaluation details · open weights · caveat Site: PSG (no respiratory); SpO₂ used: no; Threshold: severe OSA Not ranked: Reference only, not a wearable input. Weights: open	0.874

CFS: ICC

Sleep apnea (OSA) screening: CFS, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: External; lower in Black participants. Weights: not released	0.92

CFS: macro-F1

Sleep apnea (OSA) screening: CFS, Macro-averaged F1. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-averaged F1↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: External. Weights: not released	0.78

Swiss screening cohort: sensitivity

Sleep apnea (OSA) screening: Swiss screening cohort, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Reflectance PPG ODI, upper arm Reflectance PPG ODI, upper arm (Adam et al.) Evaluation details · caveat Site: upper arm; SpO₂ used: yes (ODI); Specificity 70% at the same threshold Not ranked: Accuracy 86%; wrist ODI reported as less reliable (no numbers); abstract only.	96%

Swiss screening cohort: specificity

Sleep apnea (OSA) screening: Swiss screening cohort, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Reflectance PPG ODI, upper arm Reflectance PPG ODI, upper arm (Adam et al.) Evaluation details · caveat Site: upper arm; SpO₂ used: yes (ODI); Sensitivity 96% at the same threshold Not ranked: Accuracy 86%; wrist ODI reported as less reliable (no numbers); abstract only.	70%

Duke Sleep Center cohort: Pearson r

Sleep apnea (OSA) screening: Duke Sleep Center cohort, Pearson correlation (labs / continuous). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Pearson correlation (labs / continuous)↑ Higher is better
—	TipTraQ fingertip Type IV device Evaluation details · caveat Site: finger; SpO₂ used: not stated Not ranked: AASM 1B (4%); 0.879 with 1A (3%); vendor-authored (PranaQ). Weights: not released	0.910

Kempenhaeghe clinical cohort: AUROC (AHI ≥ 15)

Sleep apnea (OSA) screening: Kempenhaeghe clinical cohort, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Wrist rPPG deep model Wrist rPPG deep model (Papini et al.) Evaluation details · caveat Site: wrist; SpO₂ used: no Not ranked: Independent test set.	0.86

Kempenhaeghe clinical cohort: AUROC (AHI ≥ 30)

Sleep apnea (OSA) screening: Kempenhaeghe clinical cohort, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Wrist rPPG deep model Wrist rPPG deep model (Papini et al.) Evaluation details · caveat Site: wrist; SpO₂ used: no; Threshold: AHI ≥ 30 Not ranked: Independent 188-recording test set.	0.85

Kempenhaeghe clinical cohort: AUROC (AHI ≥ 5)

Sleep apnea (OSA) screening: Kempenhaeghe clinical cohort, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Wrist rPPG deep model Wrist rPPG deep model (Papini et al.) Evaluation details · caveat Site: wrist; SpO₂ used: no; Threshold: AHI ≥ 5 Not ranked: Independent 188-recording test set.	0.84

Kempenhaeghe clinical cohort: Pearson r

Sleep apnea (OSA) screening: Kempenhaeghe clinical cohort, Pearson correlation (labs / continuous). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Pearson correlation (labs / continuous)↑ Higher is better
—	Wrist rPPG deep model Wrist rPPG deep model (Papini et al.) Evaluation details · caveat Site: wrist; SpO₂ used: no Not ranked: Independent test set; error 3±10 events/h; Philips co-author.	0.61

Mayo Clinic cohort: AUROC

Sleep apnea (OSA) screening: Mayo Clinic cohort, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Mayo AI-ECG Evaluation details · caveat Signal: 12-lead ECG Not ranked: AHI ≥ 5; sens 77.0%, spec 68.6%; women 0.82, men 0.73.	0.80

Samsung Medical Center cohort: ICC

Sleep apnea (OSA) screening: Samsung Medical Center cohort, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
—	Smartwatch eAHI algorithm Smartwatch eAHI algorithm (Kim et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated Not ranked: Vs expert PSG (0.87 vs AI-scored PSG); 41% excluded for invalid data; Samsung co-authors. Weights: not released	0.82

Samsung Medical Center cohort: sensitivity

Sleep apnea (OSA) screening: Samsung Medical Center cohort, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Smartwatch eAHI algorithm Smartwatch eAHI algorithm (Kim et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 92.6% at the same threshold Not ranked: Reference is AI-scored PSG, not expert scoring; Samsung co-authors. Weights: not released	92.3%

Samsung Medical Center cohort: specificity

Sleep apnea (OSA) screening: Samsung Medical Center cohort, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Smartwatch eAHI algorithm Smartwatch eAHI algorithm (Kim et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 92.3% at the same threshold Not ranked: Reference is AI-scored PSG, not expert scoring; Samsung co-authors. Weights: not released	92.6%

Stanford enriched cohort: AUROC

Sleep apnea (OSA) screening: Stanford enriched cohort, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated Not ranked: 95% CI 0.889–0.980; spectrum bias (29 with AHI <15 on both nights); Samsung-funded. Weights: not released	0.94

Stanford enriched cohort: Pearson r

Sleep apnea (OSA) screening: Stanford enriched cohort, Pearson correlation (labs / continuous). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Pearson correlation (labs / continuous)↑ Higher is better
—	Samsung Galaxy Watch OSA feature Evaluation details · caveat Site: wrist; SpO₂ used: not stated Not ranked: eAHI vs OAHI 4%; the comparison is plotted in Alavi et al. Fig. 2c (scatter, both lab nights), but r = 0.87 is not printed on the figure or caption and the body text is paywalled, so the value is still sourced from the Khayat & Floyd commentary; First author Arash Alavi stated in an email to ChatPPG (2026-10-30) that r = 0.87 (225 night-level pairs, OAHI 4%) appears on p. 6 of Alavi et al. JCSM 2026 (Fig. 2c). Author-reported; the paywalled text wasn't independently checked.; Samsung-funded.	0.87

Stanford enriched cohort: sensitivity

Sleep apnea (OSA) screening: Stanford enriched cohort, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 100% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	100%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 100% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	97.4%
—	Samsung Galaxy Watch OSA feature, default eAHI 15 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 66.7% at the same threshold Not ranked: Shipped threshold; 38.8% lacked a valid eAHI on at least one lab night (commentary). Weights: not released	94.1%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 66.7% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.1%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 65.8% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	92.3%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 42.9% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	84.6%
—	Samsung Galaxy Watch OSA feature, cohort-optimised eAHI 25.95 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.9% at the same threshold Not ranked: Post hoc cutoff fitted on the same cohort. Weights: not released	82.4%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.9% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	82.4%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.7% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	80.8%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 83.3% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	80.6%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 82.4% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	79.5%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.4% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	79.2%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 94.1% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	78.1%

Stanford enriched cohort: specificity

Sleep apnea (OSA) screening: Stanford enriched cohort, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 100% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	100%
—	Samsung Galaxy Watch OSA feature, cohort-optimised eAHI 25.95 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 82.4% at the same threshold Not ranked: Post hoc cutoff fitted on the same cohort. Weights: not released	94.9%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 82.4% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.9%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 80.8% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.7%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 79.2% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.4%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 78.1% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	94.1%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 80.6% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	83.3%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 79.5% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	82.4%
—	Samsung Galaxy Watch OSA feature, default eAHI 15 Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 94.1% at the same threshold Not ranked: Shipped threshold; 38.8% lacked a valid eAHI on at least one lab night (commentary). Weights: not released	66.7%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 94.1% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	66.7%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 92.3% at the same threshold Not ranked: Fig. 2d; two-night decision rule; best thresholds are post hoc on the same cohort; Samsung-funded. Weights: not released	65.8%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 82.4% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	64.9%
—	Samsung Galaxy Watch OSA feature Samsung Galaxy Watch OSA feature (Alavi et al.) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 84.6% at the same threshold Not ranked: Fig. 3b; Samsung-funded. Weights: not released	42.9%

UCI Sleep Center cohort: event-type accuracy

Sleep apnea (OSA) screening: UCI Sleep Center cohort, Event-type accuracy (4 classes). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Event-type accuracy (4 classes)↑ Higher is better
—	ViT on finger PPG + respiratory flow + effort ViT on finger PPG + respiratory flow + effort (Kazemi et al.) Evaluation details · caveat Site: finger + PSG respiratory; SpO₂ used: no Not ranked: 4-class event type; weighted F1 0.74, kappa 0.58; subject-disjoint 86/18/18; uses respiratory channels.	0.74

Non-public external wrist+PSG set: ICC

Sleep apnea (OSA) screening: Non-public external wrist+PSG set, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
—	Wrist PPG event detector, finger-to-wrist transfer Wrist PPG event detector, finger-to-wrist transfer (Fonseca et al.) Evaluation details · caveat Site: wrist; SpO₂ used: no Not ranked: Preprint; development test split 0.87 (N=3,111 development incl. 969 wrist). Weights: not released	0.91

Sleep apnea task, long context: AUROC

Sleep apnea (OSA) screening: Sleep apnea task, long context, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	ECG-CPC ECG-CPC (BenchECG) Evaluation details · open weights · caveat Signal: single-lead ECG Not ranked: xECG fine-tuned 0.928, linear probe 0.879. Weights: open	0.935

Multi-center enriched study: sensitivity

Sleep apnea (OSA) screening: Multi-center enriched study, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Samsung Sleep Apnea Feature Samsung Sleep Apnea Feature (FDA De Novo DEN230041) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Specificity 87.7% at the same threshold Not ranked: Vendor submission; specificity missed the prespecified criterion (post hoc 91.1%); 16.7% of nights had insufficient data. Source: FDA decision summary	82.7%

Multi-center enriched study: specificity

Sleep apnea (OSA) screening: Multi-center enriched study, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Samsung Sleep Apnea Feature Samsung Sleep Apnea Feature (FDA De Novo DEN230041) Evaluation details · caveat Site: wrist; SpO₂ used: not stated; Sensitivity 82.7% at the same threshold Not ranked: Vendor submission; specificity missed the prespecified criterion (post hoc 91.1%); 16.7% of nights had insufficient data. Source: FDA decision summary	87.7%

iSLEEPS: event AUROC

Sleep apnea (OSA) screening: iSLEEPS, Event-level AUROC (apnea / hypopnea events). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Event-level AUROC (apnea / hypopnea events)↑ Higher is better
—	MM-Net MM-Net (EEG-driven PSG, cardiorespiratory fusion) Evaluation details · caveat Site: PSG multimodal; SpO₂ used: yes (PSG) Not ranked: Ten-fold patient-independent; ISRUC-I 0.889 retrained; not a wearable input. Weights: not released	0.825

MESA: AHI R²

Sleep apnea (OSA) screening: MESA, AHI estimation R² (fraction of variance). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	AHI estimation R² (fraction of variance)↑ Higher is better
1	Dense temporal FCN + metamodel Dense temporal FCN + metamodel (+ SpO₂ desaturation feature) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (auxiliary) Caveat: Split not stated in abstract; no code. Weights: not released	0.83
2	Dense temporal FCN + metamodel Dense temporal FCN + metamodel (PPG only) Evaluation details · caveat Site: PSG finger; SpO₂ used: no Caveat: Best PPG-only MESA AHI result found. Weights: not released	0.76

MESA: event AUROC

Sleep apnea (OSA) screening: MESA, Event-level AUROC (apnea / hypopnea events). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Event-level AUROC (apnea / hypopnea events)↑ Higher is better
—	Dense temporal FCN Dense temporal FCN (apnea events) Evaluation details · caveat Site: PSG finger; SpO₂ used: no Not ranked: AUPRC 0.28 at 1.9% prevalence; central apnea remained hard. Weights: not released	0.92

MESA: ICC

Sleep apnea (OSA) screening: MESA, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: Trained on 90% SHHS1; MESA fully external; no PPG waveform. Weights: not released	0.94

MESA: macro-F1

Sleep apnea (OSA) screening: MESA, Macro-averaged F1. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-averaged F1↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: Severity from estimated AHI. Weights: not released	0.75
2	Dense temporal FCN + metamodel Dense temporal FCN + metamodel (SpO₂-assisted) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (auxiliary) Caveat: 1.18% of apnea cases called non-apneic; split not stated in abstract. Weights: not released	0.71
3	2CNN-2LSTM on PPG features + SpO₂ + apnea burden 2CNN-2LSTM on PPG features + SpO₂ + apnea burden (Rinkevičius et al.) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes Caveat: Subject-disjoint 216/30/30; small test set, so the rank is indicative only.	0.617

MrOS: ICC

Sleep apnea (OSA) screening: MrOS, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: External. Weights: not released	0.94

MrOS: macro-F1

Sleep apnea (OSA) screening: MrOS, Macro-averaged F1. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-averaged F1↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: External. Weights: not released	0.80

OSASUD: macro-F1

Sleep apnea (OSA) screening: OSASUD, Macro-averaged F1. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-averaged F1↑ Higher is better
1	2CNN-2LSTM on PPG features + SpO₂ + apnea burden 2CNN-2LSTM on PPG features + SpO₂ + apnea burden (Rinkevičius et al.) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes Caveat: Trained on MESA, tested on OSASUD unchanged; SpO₂-only baseline 0.288.	0.448

Mixed cohorts: AUROC

Sleep apnea (OSA) screening: Mixed cohorts, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	Pooled wearables Pooled wearables (meta-analysis, 59 studies) Evaluation details · caveat Site: mixed; SpO₂ used: mixed Not ranked: Pooled reference, not a model; AHI >15; sens 0.87 / spec 0.86. Weights: not released	0.93

Mixed cohorts: sensitivity

Sleep apnea (OSA) screening: Mixed cohorts, Sensitivity within a stated window. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Sensitivity within a stated window↑ Higher is better
—	Pooled wearables Pooled wearables (meta-analysis) Evaluation details · caveat Site: mixed; SpO₂ used: mixed; Specificity 86% at the same threshold Not ranked: Pooled reference, not a model. Weights: not released	87%

Mixed cohorts: specificity

Sleep apnea (OSA) screening: Mixed cohorts, Specificity at a stated threshold / operating point. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Specificity at a stated threshold / operating point↑ Higher is better
—	Pooled wearables Pooled wearables (meta-analysis) Evaluation details · caveat Site: mixed; SpO₂ used: mixed; Sensitivity 87% at the same threshold Not ranked: Pooled reference, not a model. Weights: not released	86%

Pediatric SDB set: AUROC

Sleep apnea (OSA) screening: Pediatric SDB set, Area under ROC. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Area under ROC↑ Higher is better
—	PaPaGei-S PaPaGei-S (probe, DNA-PPG benchmark) Evaluation details · caveat Signal: PPG Not ranked: Binary AHI; DNA-PPG 0.640.	0.665

SHHS1: SDB-event F1

Sleep apnea (OSA) screening: SHHS1, F1, sleep-disordered-breathing event class. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	F1, sleep-disordered-breathing event class↑ Higher is better
—	SleepMaMi SleepMaMi (PSG foundation model) Evaluation details · open weights · caveat Site: PSG multimodal; SpO₂ used: yes (PSG) Not ranked: arXiv v3 Table 2; KISS 0.662, CFS 0.523, SOF 0.327. Weights: open	0.485

SHHS1: ICC

Sleep apnea (OSA) screening: SHHS1, Intraclass correlation (agreement of a continuous estimate, e.g. AHI). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Intraclass correlation (agreement of a continuous estimate, e.g. AHI)↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: Same cohort as training; SHHS2 0.95. Weights: not released	0.96

SHHS1: macro-F1

Sleep apnea (OSA) screening: SHHS1, Macro-averaged F1. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-averaged F1↑ Higher is better
1	OxiNet OxiNet (SpO₂ CNN) Evaluation details · caveat Site: PSG finger; SpO₂ used: yes (SpO₂ only) Caveat: Same cohort as training; SHHS2 0.83. Weights: not released	0.84

SHHS1: AHI MAE (events/h)

Sleep apnea (OSA) screening: SHHS1, AHI mean absolute error (events/h). Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	AHI mean absolute error (events/h)↓ Lower is better
—	SleepMaMi SleepMaMi (PSG foundation model, Macro-Encoder) Evaluation details · open weights · caveat Site: PSG multimodal; SpO₂ used: yes (PSG) Not ranked: Ablation row (micro-only 9.69); not a wearable input. Weights: open	8.40