Sleep Health Data Commons

Abstract

The Sleep Health Data Commons (SHDC) is a credentialed research commons of de-identified clinical sleep and neurophysiology data, assembled from multiple contributing health systems and harmonized to common open standards. For the same patients wherever possible, it brings together overnight polysomnography (PSG), clinical electroencephalography (EEG), brain MRI, and the linked de-identified electronic health record (EHR).

Signals and images are organized with the Brain Imaging Data Structure (BIDS), and the accompanying electronic health record is represented in the OMOP Common Data Model (CDM v5.4), so that cohorts, exposures and outcomes can be defined once and applied across sites. Access to every dataset that makes up the Commons is governed by a single credentialed data use agreement.

The Commons currently includes on the order of 219,000 polysomnography studies (about 157,000 patients) and 341,000 EEG recordings (about 127,000 patients), together with linked brain MRI and OMOP-mapped EHR for the large majority of patients, across roughly 10 contributing sites and growing. It is being developed under the ARPA-H REST program.

[To be filled in: final patient and record counts at release. Current figures are drawn from the BDSP Data Atlas and will be updated as sites are added.]

Background

Sleep and brain health are closely linked to cardiovascular, metabolic, neurological and psychiatric disease, yet the physiological recordings that capture them (polysomnography, EEG and neuroimaging) are typically siloed within individual health systems, stored in incompatible formats, and separated from the clinical context needed to interpret them. Large-scale, multi-site research linking these signals to health outcomes has therefore been difficult.

The Sleep Health Data Commons was created to remove those barriers: to aggregate clinical sleep and neurophysiology data across institutions, de-identify it, harmonize it to shared open standards, and link it to longitudinal de-identified EHR, so that it can be studied at scale by credentialed researchers under common governance.

Methods

Data are contributed by participating health systems and processed through a common pipeline:

• De-identification. Protected health information is removed from signals, images and the associated records, and identifiers are replaced with study identifiers. [To be filled in: detailed de-identification procedure, date handling, and the standard under which de-identification is certified.]
• Harmonization. Signals (PSG, EEG) and images (MRI) are organized in the Brain Imaging Data Structure (BIDS). The linked electronic health record is mapped to the OMOP Common Data Model (CDM v5.4) with standardized vocabularies, so that cohorts and phenotypes can be defined consistently across sites.
• Linkage. Recordings are linked to the de-identified EHR for each patient, and across modalities for the same patient where available.
• Quality control. [To be filled in: QC steps, validation, and known limitations.]

Data Description

The Commons is an umbrella collection spanning several credentialed BDSP datasets for the same underlying patient population:

• Polysomnography (PSG): full overnight sleep studies with multichannel signals and, where available, expert sleep staging and annotations.
• Electroencephalography (EEG): clinical EEG recordings, routine and long-term.
• Brain MRI: structural brain imaging in BIDS, with de-identified radiology reports where available.
• Electronic health record (EHR): longitudinal de-identified clinical data (visits, diagnoses, medications, labs and vitals, and clinical notes) represented in the OMOP CDM and linked to the recordings.

Data are organized by modality and contributing site; sites are identified by code rather than by name. [To be filled in: exact file organization, directory layout, and per-modality counts at release.]

Usage Notes

Access to the Sleep Health Data Commons is credentialed. To use the data, a researcher must hold current BDSP credentialing, complete the required human-subjects research training, and sign the Commons data use agreement; a single agreement covers all of the datasets in the Commons. Individuals who primarily reside in a country of concern under the U.S. Department of Justice Data Security Program may not be granted access.

Signals and images follow BIDS conventions and the linked EHR follows the OMOP CDM, so standard tooling for those standards can be used. Interactive tools offered through the Commons, including a browser-based MRI viewer, require the same credentialed access. [To be filled in: worked examples, code, and the preferred citation.]

Release Notes

Version 1.0.0 — initial release of the Sleep Health Data Commons, establishing the credentialed umbrella that spans its constituent datasets: electronic health records, brain imaging, polysomnography, and harmonized sleep metadata.

Ethics

The Sleep Health Data Commons operates under single-IRB oversight (BRANY). Data are de-identified before aggregation and shared with credentialed researchers under a data use agreement. [To be filled in: IRB protocol number and approval status, and the determination under which the de-identified data are shared.]

Acknowledgements

The Sleep Health Data Commons is developed under the ARPA-H REST program. This research was funded, in part, by ARPA-H under award 1AY2AX000237. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U.S. Government.

We thank the contributing health systems and the clinical and research teams who made their data available to the Commons. [To be filled in: full list of contributing institutions and individual acknowledgements.]

Conflicts of Interest

The authors declare no competing interests.