Smartwatch-based detection of moderate-to-severe and high-risk obstructive sleep apnea

Prospective Stanford study of the Samsung Galaxy Watch sleep apnea feature against two in-lab polysomnography nights in 147 adults at high pre-test risk. AUROC for moderate-to-severe OSA (AHI ≥ 15) was 0.94; at the default eAHI cutoff of 15, sensitivity was 94.1% and specificity 66.7%, against 87.7% specificity in the FDA De Novo study. The cohort was enriched and the study was Samsung-funded.

• Form factor: wrist

PurposeObstructive sleep apnea (OSA) is a prevalent condition associated with long-term consequences, such as cardiovascular and neurocognitive diseases, yet many cases remain undiagnosed and therefore untreated due to the cost and limited access to traditional in-lab overnight polysomnography (PSG). The ubiquity of consumer wearable technology offers a practical way to mitigate these limitations, enabling more accessible, large-scale screening. However, the clinical validity of these devices requires further testing.MethodsIn this novel prospective study, we rigorously evaluated the diagnostic performance of the Samsung Galaxy Watch for detecting moderate-to-severe OSA (apnea-hypopnea index, AHI ≥ 15 events/h) compared to in-lab PSG using the AHI, and "high-risk" OSA subjects using the PSG-derived hypoxic burden (HB) criteria. We enrolled 152 adults aged 22 years or older with a prior diagnosis of moderate-to-severe OSA or a high pre-test likelihood of having it (STOP-Bang score ≥ 3), who wore the smartwatch during two in-lab PSG nights (147 completed both nights; total 1,850 h of in-lab PSG sleep): an initial in-lab PSG night, followed by ≥ 3 watch-only at-home nights, and then a second in-lab PSG night. In addition to AHI, we evaluated Galaxy Watch performance across PSG-derived HB-defined risk strata (low- vs. high-risk OSA) as HB integrates the depth, duration, and frequency of oxygen desaturations and has been shown to better reflect cardiometabolic risk and disease severity than AHI.ResultsIn our cohort, the Galaxy Watch achieved an area under the receiver operating characteristic (AUROC) curve of 0.94 (95% CI 0.889-0.980) for detecting moderate-to-severe OSA. With the default estimated AHI (eAHI) threshold of 15, the device yielded a sensitivity of 94.1% (95% CI 84.1-98%) and a specificity of 66.7% (95% CI 51-79.4%), while with the cohort-optimized eAHI threshold of 25.95, sensitivity was 82.4% (95% CI 69.7-90.4%) and specificity was 94.9% (95% CI 83.1-98.6%). Notably, within the HB defined high-risk OSA group, the Galaxy Watch achieved 100% sensitivity and 100% specificity using the default threshold, indicating robust discrimination between HB-defined low- and high-risk OSA categories.ConclusionOverall, these findings highlight the substantial potential of consumer-grade wearables to accurately detect moderate-to-severe sleep apnea using both AHI- and HB-defined risk strata across controlled and real-world settings.RegistrationClinicalTrials.gov ID: NCT06603441; First posted: 2024-09-19.Current knowledge/study rationaleObstructive sleep apnea (OSA) remains substantially underdiagnosed because definitive testing with polysomnography (PSG) is resource-intensive and not readily accessible to all patients. Although consumer smartwatches offer a promising approach for large-scale OSA screening, prospective validation against PSG and evaluation using physiologically meaningful measures such as hypoxic burden have been limited.Study impactIn this prospective study of 152 adults undergoing repeated PSG and smartwatch assessment, the Galaxy Watch demonstrated high accuracy for detecting moderate-to-severe OSA and showed excellent performance for identifying individuals with high hypoxic burden, a subgroup at elevated cardiometabolic risk. These findings support the use of consumer-grade wearable technology as a scalable, accessible screening tool to improve OSA risk identification, triage, and referral for definitive diagnostic evaluation.

Verbatim from Europe PMC (opens in a new tab) · Abstract via Europe PMC; © the authors/publisher

27 rows are credited to this paper. Ranks apply within each leaderboard only, as reported by the source.