Zero burden multi night monitoring with AI enabled technology reduces obstructive sleep apnea misdiagnosis

Abstract Multi-night measurement of obstructive sleep apnea (OSA) using AI-enabled technologies could reduce misdiagnosis rates from night-to-night variation. However, prospective feasibility and effectiveness of multi-night assessment in clinical populations remains uninvestigated. In this study, 100 people with suspected OSA were recruited to receive an under-the-mattress Withings Sleep Analyzer (WSA) to quantify in-home sleep for 3 months alongside single-night polysomnography. The primary outcome was feasibility of novel technology to estimate the apnea-hypopnea-index (AHI), defined as ≥14 nights of AHI recordings within the first month. Secondary outcomes included diagnostic classification of moderate-to-severe OSA (AHI ≥ 15 events/h). 92 participants (53±15yo; BMI = 31 ± 7 kg/m 2 ; 48 females) were included in the final analyses. Eighty-five (92%) had ≥14 nights of valid AHI measurements in the first monitoring month. WSA had specificity of 85% and sensitivity of 77%, and overall F1-score of 80% to detect OSA versus polysomnography. Participants with OSA on WSA but not polysomnography had higher night-to-night variability in OSA severity, and many had minimal supine sleep and short sleep duration during polysomnography (-97 [-183, -11]min, p value = 0.027). Multi-night zero-burden AI monitoring of OSA is feasible and could identify patients at risk of single-night polysomnography misdiagnosis, including those with high night-to-night variability, short sleep, and minimal supine sleep during polysomnography.