TL;DR
This paper introduces the Learning Healthcare System Data Commons (LHSDC), a cloud-based, cost-effective, and extensible platform designed to democratize access to healthcare data for research and quality improvement. It addresses the significant challenges healthcare organizations face in utilizing their data by providing a secure, governed, and scalable solution, exemplified by a reference implementation at Rush University.
What Problem Does This Paper Address?
Healthcare organizations, especially smaller or less academically focused ones, struggle with the complex and expensive infrastructure required to effectively find, access, share, and analyze patient data for research and quality improvement. This challenge is exacerbated by the need for robust data governance, security, and interoperability, especially in light of events like the COVID-19 pandemic, which highlighted the urgent need for unified and harmonized data sets. The current ad-hoc, project-by-project approach to data preparation and sharing is inefficient, costly, and lacks centralized control, leading to potential data leakage and compliance issues.
Methods
The core of the LHSDC is built upon the Gen3 platform, an open-source data commons framework. Gen3 is cloud-native and provides a FAIR (Findable, Accessible, Interoperable, Reusable) environment for data. The authors highlight their implementation's novelty in two key areas:
- Healthcare Provider Focus: It's the first Gen3 data commons specifically designed to facilitate quality improvement and research for healthcare providers.
- Microsoft Azure Implementation: This is the first Gen3 implementation on the Microsoft Azure platform, developed in collaboration with Microsoft.
The platform utilizes a pay-as-you-go, usage-based cost model, making it economical. It employs a three-tiered governance and operations model: a data governance committee, a research informatics core, and the Gen3 LHSDC itself. Data storage is managed efficiently by linking to commodity storage for large files (images, genomics) and using a Gen3-hosted Postgres database for structured data and metadata, which is then exposed via an Elasticsearch index for interactive exploration. Deployment is orchestrated using Terraform and Azure DevOps pipelines.
Datasets
The paper describes the types of data integrated into the Rush University reference implementation, rather than specific, named datasets for training or evaluation. These include:
- Electronic Health Record (EHR) patient data
- Medical images (CT/PET/MR)
- Biospecimens data (e.g., COVID-19 sputum samples, cancer biopsy specimens)
- Diagnostic sequences
- Internet of Things (IoT) streaming data
- Social determinants of health data
- Pathology files
- Biorepository data
The paper emphasizes the multimodal nature of these datasets and the challenges in housing them in traditional databases.
Key Insights and Ideas
- Democratizing Data Access: The LHSDC aims to break down data silos and make complex healthcare data accessible to a broader range of researchers and clinicians, not just those at large academic institutions.
- Cost-Effectiveness through Cloud and Usage-Based Models: By leveraging cloud infrastructure and a pay-as-you-go model, the LHSDC significantly reduces the financial burden associated with data management and utilization.
- The Power of a Data Commons for Quality Improvement: Beyond research, the platform is designed to support operational sustainability and governance for quality improvement initiatives within a hospital setting.
- Extensibility is Key: The platform's design emphasizes extensibility, allowing it to adapt to the ever-changing landscape of healthcare data and technologies.
- Centralized Governance for Security and Auditability: Moving away from decentralized, ad-hoc data sharing to a centralized, governed approach enhances security, auditability, and compliance.
Key Results
The paper presents a reference implementation and discusses the benefits observed, rather than specific quantitative results from a large-scale deployment. The key qualitative results and benefits include:
- Improved Data Findability, Accessibility, Interoperability, and Reusability: The LHSDC architecture significantly enhances these aspects of data utilization.
- Reduced Inefficiencies: The platform streamlines operational costs and processes related to data access and research.
- Rapid, Self-Service Access: Researchers can gain access to health data sets more quickly and independently, a critical need highlighted by the COVID-19 pandemic.
- Enhanced Security and Governance: Centralized control and robust governance mechanisms improve data security and compliance.
- Facilitation of Multimodal Data Integration: The Gen3 framework is well-suited for integrating diverse data types, which is crucial for precision medicine and comprehensive health analytics.
Limitations and Future Work
- Reference Implementation: The paper describes a reference implementation at Rush University, suggesting that broader adoption and scaling to other institutions will require further effort and adaptation.
- Data Governance Evolution: The paper notes that governance is a "growth and maintenance process, with small but regular updates to be made," implying ongoing work is needed to refine governance policies and practices.
- Integration with Existing Ecosystems: The conclusion mentions integrating into "existing data ecosystems (i.e., commons of commons)," indicating a future direction for broader interoperability and data sharing across institutions.
- Incorporation of New Data Sources: The Rush pilot plans to incorporate IoT hub reporting and clinical trial management system (CTMS) integration, highlighting ongoing development and expansion of data sources.
- FHIR API for Data Sharing: The development of a data model using standard naming from HL7.org to facilitate data sharing via the FHIR API is a key feature and a direction for future interoperability.
Summary updated