What question did this study set out to answer?

This research aims to evaluate the effectiveness of an FDA-cleared AI device in classifying the apnea-hypopnea index (AHI) from home sleep apnea test signals.

May 10, 2026

0564 Making Type III HSAT Recordings Conclusive: Large-Scale Validation of an FDA-Cleared AI AHI Classifier from HSAT Signals

Key Points

This research aims to evaluate the effectiveness of an FDA-cleared AI device in classifying the apnea-hypopnea index (AHI) from home sleep apnea test signals.
Retrospective multicenter validation of 5,771 polysomnograms from accredited sleep programs.
Utilized data from multiple sleep testing sites with full-montage manual scoring for reference AHI.
Analyzed HSAT channel data including nasal pressure and oximetry to compute AHI.
For AHI ≥5, positive agreement was 91.0%, negative agreement 78.0%, and overall agreement 90.6%.
For AHI ≥15, positive agreement was 78.1%, negative agreement 93.9%, and overall agreement 81.7%.
For AHI ≥30, positive agreement was 69.9%, negative agreement 97.5%, and overall agreement 85.0%.

Abstract

Abstract Introduction Type III home sleep apnea tests (HSAT) lack electroencephalography for sleep and arousal scoring. Currently, a respiratory event index based on monitoring time, rather than an apnea–hypopnea index (AHI), is commonly reported from HSATs. This underestimates disease severity and leaves HSAT studies inconclusive, necessitating in-lab polysomnography. We evaluated whether DeepRESP v2.0 (K252330), an FDA 510(k)-cleared medical device, yields conclusive AHI-based severity classification aligned with attended polysomnography from HSAT signals commonly acquired in patients’ homes. Methods We conducted a retrospective, multicenter validation using 5,771 Type I/II polysomnograms from accredited US sleep programs, manually scored by registered technologists and board-certified sleep physicians. To our knowledge, this dataset is substantially larger than the cohorts typically used in FDA 510(k) validation studies of HSAT devices. HSAT channels (nasal pressure, thoracoabdominal inductance belts, oximetry, body position) were fed to the device to score wake/NREM/REM, arousals, and respiratory events, and to compute AHI. Reference AHI came from full-montage manual scoring. Positive percentage (PPA), negative percentage (NPA), and overall percentage agreement (OPA) with manual AHI categories (≥5, ≥15, ≥30 events/hour) were estimated with 95% confidence intervals. Subgroups summarized heterogeneity in age, sex, BMI, severity, and race/ethnicity. Results The cohort was heterogeneous: adults aged 18 to 65 years; 35% were female; 58% had BMI ≥30 kg/m2; AHI ranged from normal to severe; and recordings included PSGs from both in-laboratory and home settings. For AHI classification, agreement with manual scoring was high across thresholds: for AHI ≥5, PPA 91.0%, NPA 78.0%, OPA 90.6%; for AHI ≥15, PPA 78.1%, NPA 93.9%, OPA 81.7%; and for AHI ≥30, PPA 69.9%, NPA 97.5%, OPA 85.0%. High agreement was observed across demographic subgroups (age, sex, BMI, and race/ethnicity), supporting consistent performance across a diverse adult population. Conclusion AHI derived from the device closely aligns with conventional AHI from attended polysomnography, supporting conclusive severity classification from HSAT and potentially reducing inconclusive HSAT studies and repeat in-lab testing, thereby streamlining the diagnostic pathway and earlier linkage to therapy. Support (if any) None

Bookmark

Cite This Study

Agustsson et al. (Fri,) studied this question.

synapsesocial.com/papers/6a0020aec8f74e3340f9b7bf https://doi.org/https://doi.org/10.1093/sleep/zsag091.0563

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

Bookmark