O041 Exploring the Landscape of Sleep Data Resources: A Literature Survey

Abstract Introduction The complex physiologic interdependencies of sleep make sleep medicine a data-rich discipline. Emerging big data approaches hold promise for novel discoveries, however, scholars have suggested these advances may be hindered by dataset limitations. We conducted a review of the field to assess the drawbacks of available sleep data resources. Methods A systematic literature survey on PubMed. A keyword search identified studies based on sleep datasets which combined polysomnography records and contextual patient health information. For eligible studies, we extracted information on data provenance (existing repositories versus self-collected), data accessibility for onward research, dataset size, cohort characteristics, data/metadata inclusions, and use of systematic nomenclatures. Results Only approximately 40% of identified studies used public datasets, suggesting that these repositories only partially met current research needs. Alternative, self-collected cohorts were typically small, with just 20% having populations greater than 1000 subjects. Most were closed-source or lacking data access indications, which may impact the strength and reproducibility of study results. Large-size resources focussed predominantly on sleep apnoea, limiting the reach of research efforts targeting less prevalent conditions. Datasets featuring comprehensive medical histories and long-term follow-up information were rare. Where available, this information generally lacked systematisation of nomenclatures and metadata, impairing the ability of users to extract meaning from data entities and their interrelationships. Discussion Limitations in current sleep data practices were substantiated by our review. There is a clear need to develop larger, diverse, multimodal and clinically representative sleep data collections that can be more easily shared, and analysed using emergent data-driven discovery tools.