P037 The Big Sleep ACT Project: Developing a Modern Dataset to Support Sleep Research

Abstract Background Limitations in scale, population diversity, technical quality, data curation methods and accessibility of existing data resources have been recognised as limiting factors for the advancement of sleep clinical research through big data approaches. To bridge this gap, this study introduces a new sleep dataset which seeks to capture a data-rich, longitudinal snapshot of a representative Australian clinical sleep cohort. Methods Retrospective collation of de-identified sleep clinical records from adult patients who underwent at least one in-lab Type-1 polysomnography between 2012 and 2018 at Canberra Sleep Clinic. We extracted polysomnography raw signals and annotations, as well as medical record information including basic demographics, comorbidities, medications, examination findings, diagnoses, therapy settings and follow-up observations throughout subjects’ time in the Clinic’s care. Records were organised according to a graph database structure, embedding SNOMED terminology encodings wherever possible. Results N=6,777 subjects were included. Gender split (M/F: 62%/38%) and age (51.7±15.3 years) distribution were consistent with typical clinical sleep cohorts. Polysomnography recordings included diagnostic (n=6,635) and non-invasive ventilation titration/therapy (n=2,834), as well as MSLT (n=270) and MWT (n=25) studies. Clinical subgroups featured healthy, Obstructive Sleep Apnea (OSA) and non-OSA dyssomnia patients, as well as small cohort of parasomnia cases. Follow-up duration varied among cases (<3 months to >5 years). Discussion Despite limitations associated with retrospective data extraction, the data-richness and scale of Big Sleep ACT compare favourably with world-leading sleep datasets. Careful data organisation makes this dataset well placed to support innovative data-driven research into precise diagnoses, personalised interventions, and automation in sleep medicine.