AI-derived AHI reclassified 89.7% of REI-negative home sleep apnea tests to positive, converting 15.5% of all women and 6.3% of all men from REI <5 to AHI ≥5.
Observational (n=3,094)
No
Does AI-derived AHI improve diagnostic classification compared to standard REI in adults undergoing home sleep apnea testing?
AI-derived AHI significantly reduces inconclusive home sleep apnea tests and uncovers underestimated disease burden, particularly in women.
Abstract Introduction Home sleep apnea testing (HSAT) typically reports a respiratory event index (REI, events/hour of monitoring) rather than apnea–hypopnea index (AHI, events/hour of sleep). REI underestimates disease severity, particularly for arousal-based hypopneas common in women, leaving symptomatic patients with inconclusive results (REI 5) that prompt repeat testing or loss to follow-up. We evaluated how an FDA-cleared AI system that derives AHI from HSAT signals changes HSAT classification, with emphasis on sex differences. Methods We retrospectively analyzed 3094 adult HSATs (1785 men; 1147 women) from a single sleep center. REI came from routine clinical processing. AHI was computed from the same signals by DeepRESP v2.0 (K252330), an FDA-cleared SaMD for sleep staging, arousal detection, and respiratory scoring. REI and AHI were categorized as 5, 5–14.9, 15–29.9, and ≥30 events/hour. We cross-tabulated REI versus AHI overall and by sex, focusing on reclassification of REI 5 and REI 15 relative to AHI-defined moderate–severe disease. Results By REI, 351/3094 studies (11.3%) were 5, including 122/185 men (6.8%) and 205/1147 women (17.9%); with AI-derived AHI, only 36/351 (10.3%) remained 5. Reclassification from REI 5 to AHI≥5 occurred in 113 men (6.3% of all men) and 178 women (15.5% of all women), so nearly one in six women had an REI-negative HSAT that became AHI-positive. Among patients with AHI≥15, 727/2374 (30.6%) had REI 15, affecting 27.4% of men (411/1499) and 37.4% of women (283/756). Conclusion AI-derived AHI converted most REI-negative HSATs into positive or higher-severity studies, and women with REI values just below diagnostic thresholds were more often moved into higher severity categories. These findings suggest that REI alone disproportionately underestimates disease burden in women, whereas automated AHI reduces inconclusive HSATs and may streamline diagnostic pathways; effects on repeat testing and treatment require prospective evaluation. Support (if any) Internal quality-improvement project funded by Nox Health. These operational data were not part of the device’s 510(k) submission and are investigational, not intended to represent official device performance, support promotional claims, or modify labeling.
Agustsson et al. (Fri,) conducted a observational in Sleep apnea (n=3,094). AI-derived AHI (DeepRESP v2.0) vs. Respiratory event index (REI) was evaluated on Reclassification of REI < 5 and REI < 15 relative to AHI-defined moderate-severe disease. AI-derived AHI reclassified 89.7% of REI-negative home sleep apnea tests to positive, converting 15.5% of all women and 6.3% of all men from REI <5 to AHI ≥5.