0568 Fully Automated Central Apnea Scoring in Home Sleep Apnea Testing Using Dual Volumetric RIP and AI Analysis

Abstract Introduction HSAT is recommended for adults with suspected obstructive sleep apnea (OSA). AASM HSAT rules recommend classifying apneas as obstructive or central when respiratory effort signals are available, although reporting the central apnea-hypopnea index (CAHI) is optional. Central scoring is challenging because signals are often degraded and many automated HSAT systems require manual review and rescoring of central events. We evaluated DeepRESP v2.0 (K252330), an FDA-cleared AI software-as-a-medical-device (SaMD) that provides automated central apnea scoring from HSAT data without a central-event–specific rescoring requirement beyond standard clinical review. Methods We retrospectively validated the system in adults using multi-center PSG studies reformatted as HSAT channels: nasal pressure airflow, dual thoracoabdominal volumetric RIP, SpO₂, and body position. Respiratory events, including central apneas, were manually scored per AASM rules by technologists and physicians. The device automatically detects apneas from nasal cannula or RIP, classifies them as obstructive or central from thoracic and abdominal effort, and computes HSAT indices. The dual-belt volumetric RIP system is designed for robust attachment and an approximately linear relation to flow, improving reliability and sensitivity versus single-belt or non-calibrated RIP. We summarized performance using positive, negative, and overall percent agreement (PPA, NPA, OPA) for CAHI ≥5 events/h versus the reference. Results 5771 sleep recordings were included. CAHI ≥5 classification showed PPA 80.7%, NPA 98.0%, and OPA 97.2% relative to the manually scored reference standard. Overall apnea metrics remained robust, with OPA for all apneas (obstructive and central) ≥97% and high specificity for central events, indicating that automated scoring did not inflate central counts. Central classification remained reliable when airflow degraded or cannula failed, relying on dual-belt RIP. Conclusion An FDA-cleared AI SaMD delivered fully automated central apnea scoring from HSAT signals with high agreement to expert scoring, without any central-apnea–specific rescoring beyond standard clinical review. Dual thoracic and abdominal volumetric RIP supports AASM rules for distinguishing central from obstructive events and provides reliable quantification of incidental central events on HSAT to guide evaluation and treatment. Support (if any) None