1383 Evaluating Concordance Between AI Bed Sensors, Sleep Diary, and CPAP Measures in Obstructive Sleep Apnea

Abstract Introduction Consumer sleep technologies increasingly provide passive monitoring of sleep behavior, yet their clinical validity relative to subjective and medical-grade measures remains unclear. Obstructive sleep apnea (OSA) is characterized by recurrent upper-airway obstruction leading to apneic events, which are reliably quantified by continuous positive airway pressure (CPAP) devices. In contrast, AI-enabled smart beds detect movement-based awakenings—not apneas—and the relationship between these signals remains poorly understood. Because arousals and apneas represent different physiological events, any temporal association requires cautious interpretation. This case study evaluated concordance among an AI-enabled smart bed, a subjective sleep diary, and CPAP recordings across 22 nights in a patient with OSA, focusing on sleep duration, awakenings, and the temporal proximity between awakenings and apneic events. Report of case(s) Simultaneous data were collected from (1) the AI bed (objective total sleep time and movement-based awakenings), (2) a subjective sleep diary (self-reported sleep duration and awakenings), and (3) CPAP device data (apnea–hypopnea index [AHI] and timing of apneic episodes). Correlations across systems were calculated. Temporal alignment analyses assessed whether AI-detected awakenings occurred within ±5, ±10, or ±15 minutes of CPAP-recorded apneas. Because the devices measure non-equivalent phenomena, alignment metrics reflect temporal proximity rather than apnea detection. Sensitivity, positive predictive value (PPV), and F1 scores were computed to summarize correspondence. Temporal proximity between AI-detected awakenings and CPAP-recorded apneic events varied markedly by time window. Alignment was low at ±5 minutes (sensitivity 17.8%, PPV 13.2%, F1 = 0.15) and modest at ±10 minutes (sensitivity 24.2%, PPV 16.1%, F1 = 0.19). The greatest overlap occurred at ±15 minutes, where approximately half of CPAP-recorded apneas occurred near an AI-detected awakening (sensitivity 52.2%), though precision remained low (PPV 26.5%; F1 = 0.35). These findings indicate partial temporal coupling but limited specificity, consistent with the distinct physiological bases of each signal. Conclusion The AI-enabled smart bed demonstrated minimal to moderate temporal proximity to CPAP-recorded apneic events, improving only with broader windows. While the AI bed captured general patterns of nighttime disruption, it did not reliably correspond to individual apneic episodes. Larger studies are needed to determine whether such consumer devices may complement clinical tools for monitoring sleep disruption in OSA. Support (if any)