0449 Pilot Study: AI-Enhanced COMISA Diagnosis in Hong Kong Public Hospitals Using Objective Sleep and HRV Metrics

Abstract Introduction The underdiagnosed and high overlapping prevalence of comorbid insomnia and obstructive sleep apnea (COMISA) often suffer worst cardiovascular outcomes than patients with either OSA or Insomnia. The scarcity of multi-disciplinary clinics addressing both conditions may impede timely COMISA diagnosis and treatment efficacy. This study explored the feasibility of leveraging objective sleep and HRV metrics through Belun Ring (BR), the AI-powered wearable, to facilitate COMISA diagnosis and personalized treatment. Methods Consecutive suspected OSA or Insomnia cases were co-screened by clinical psychologist and pulmonologist. Participants were categorized into the OSA group if PSG-AHI ≥ 15 events/h, insomnia if Insomnia Severity Index (ISI) ≥ 15 or diagnosed based on the International Classification of Sleep Disorders, third edition (ICSD-3), or COMISA if meeting both conditions. All subjects completed a 3-night (≥ 100 mins/night) sleep assessment using the BR. Kruskal-Wallis tests and PCA analyses were conducted for comparative analyses among the three groups. Results Among 54 subjects, 31 had insomnia, 13 had OSA, and 10 had COMISA. Marked discrepancies between subjective and AI-generated sleep statistics (concordance rate= 0.07) COMISA exhibited the highest wake after sleep onset (median, mins) compared to insomnia and OSA (76.5 vs. 48.1 vs. 21.2, P 0.001). COMISA and insomnia had upregulated wakefulness (median, count) compared to OSA (14.9 vs. 15.2 vs. 8.3, P 0.05). Moreover, COMISA exhibited the lowest SDNN (median, ms) and RMSSD (median, ms) compared to insomnia and OSA (SDNN: 49.0 vs. 63.8 vs. 92.9, P 0.01; RMSSD: 44.9 vs. 53.8 vs. 79.8, P 0.01). COMISA had dropped PRR50 (median, %) compared to OSA (6.6 vs. 24.8, P 0.05). Normalized LF and LF/HF did not differ across the three groups. The novel workflow reduced waiting time for consultation from 104 to 45 weeks. Conclusion This is the first study to demonstrate the use of AI-assisted sleep and HRV parameters to single-out COMISA for better personalized treatment and efficient workflow. The worst cardiovascular outcomes in COMISA patients were objectively reflected by the lowest SDNN and RMSSD findings. This novel study elucidated the potential how AI can assist to identify COMISA patients even in monodisciplinary settings for better patient care. Support (if any)