We aimed to compare the Belun Sleep Platform (BSP), an artificial intelligence-driven home sleep testing device, with polysomnography (PSG) for diagnosing obstructive sleep apnea. The BSP analyzes oxygen saturation, heart rate, and accelerometry patterns. Participants scheduled for PSG and with no significant cardiovascular or neuromuscular comorbidities were recruited. They underwent simultaneous in-laboratory, full-night PSG with the BSP. We assessed diagnostic properties, including sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. A total of 40 participants (54.3% male) with a mean age of 49.9 years were enrolled. For an apnea-hypopnea index (AHI) cutoff of ≥ 15 events/h, BSP showed an accuracy of 68.5%, sensitivity of 35.2%, and specificity of 100% under American Academy of Sleep Medicine criteria 1 A and 1B. For AHI thresholds of ≥ 5 and ≥ 30 events/h, sensitivity was 82.1% and 33.3%, respectively, while specificity was 14.2% and 100%, respectively. BSP-AHI correlated moderately with PSG-AHI (intraclass correlation coefficient ICC = 0.737). BSP's oxygen desaturation index (ODI) showed a strong correlation with PSG-ODI (ICC = 0.882). Moderate correlations were observed between BSP and PSG for non-rapid eye movement sleep duration (ICC = 0.736), rapid eye movement sleep duration (ICC = 0.664), total sleep time (ICC = 0.617), and sleep efficiency (ICC = 0.719). The BSP's high specificity but low sensitivity suggests it serves better as a confirmatory tool rather than a primary screening method. Its moderate concordance with PSG underscores its potential in settings where PSG is unavailable. However, further investigation is needed to refine its clinical applications.
Tirachaimongkol et al. (Fri,) studied this question.