BREATH: Breakthrough Research in Electromyography for the Assessment of Sleep-disordered BreaTHing

Project Summary Abstract Objectives and Specific Aims: The overarching objective is to develop and validate a diagnostic tool for obstructive sleep apnea (OSA), Deep Learning-enhanced Transmembranous Electromyography (DL-tmEMG), to enable convenient point-of-care, and accurate detection of OSA for widespread use by any practitioner evaluating patients for OSA. In phase 1, this proposal aims to establish feasibility of tmEMG to diagnose OSA patients with good accuracy and to optimize the DL algorithm. Phase 2 aims to demonstrate generalizability and accuracy of DL-tmEMG in a diverse population, identifying differences in important subpopulations and the benefits of multimodal data integration, while progressing regulatory approval. Health-relatedness: This project addresses the urgent need for efficient and accurate diagnostics for OSA, a widespread and serious disorder impacting an estimated 1 billion people worldwide, while undiagnosed OSA costs the U.S. approximately $149.6 billion annually. Successful development of DL-tmEMG will enhance personalized OSA management and improve healthcare accessibility. Efficient screening for OSA in under- resourced areas and for patients pre-surgically and perinatally are relevant to this proposal. Design: Phase 1 Study#1 (9 months) is a case-control observational feasibility study of 60 adult patients. Phase 2 Study #2 (18 months) involves a large single center prospective observational cohort study of 200 adult patients referred for PSG testing at the UCSD Sleep Center. Methods: DL-tmEMG is a new diagnostic that incorporates two novel technologies: 1) a non-invasive transmembranous sensor of muscle activity (tmEMG), and 2) deep learning algorithms applied to EMG signal to provide an automatic interpretation (EMGNet). The combination of these two innovations obviates the need for an expert electromyographer, key to allowing for widespread use to fulfill the unmet need for improved diagnostics in OSA. In Phase 1 Study #1, 30 healthy and 30 patients with moderate to severe OSA will undergo DL-tmEMG within 30 days of polysomnography. We will compare DL-tmEMG classification results with PSG results and calculate performance metrics. Phase 2 Study #2 utilizes a similar methodology as Study #1 but differs in size as well as population pretest probabilities, being a cohort study design. Patients will undergo DL-tmEMG within 30 days of PSG, which will classify patients as OSA (AHI>5) or non-OSA (AHI<5). The performance of DL-tmEMG will again be compared with PSG results. Recruitment will be adjusted to allow performance to be evaluated in important sub-populations. Results of this prospective study will provide strong evidence for validity and reliability of DL-tmEMG as a novel diagnostic tool for OSA and will be used to support a multicenter study (a phase IIB proposal) that will further evaluate the performance of DL-tmEMG across sites and diverse patient populations to meet FDA submission requirements, as part of a 510(k) premarket notification or a premarket approval (PMA) process.