DW4TR: A Data Warehouse for Translational Research

Hai Hu, Mick Correll, Leonid Kvecher, Michelle Osmond, Jim Clark, A Bekhash, Gwendolyn Schwab, De Gao, Jun Gao, Vladimir Kubatin, Craig D. Shriver, Jeffrey A. Hooke, Larry G. Maxwell, Albert J. Kovatich, Jonathan G. Sheldon, Michael Liebman, Richard MuralJournal of Biomedical Informatics 2011 · DOI

TL;DR

This paper introduces DW4TR, a flexible data warehouse designed to integrate diverse clinical and molecular data for translational research. It addresses the challenge of making complex, multi-source data accessible and usable for researchers, enabling better study design, cohort selection, and hypothesis testing across different disease areas.

What Problem Does This Paper Address?

Translational research, the process of translating scientific discoveries into clinical applications, is hampered by the fragmentation and complexity of biomedical data. Integrating clinicopathologic data (from patient records, questionnaires) with molecular data (genomic, proteomic) from various platforms is a significant challenge. Existing Electronic Medical Record (EMR) systems are often not designed for the specific analytical and reporting needs of translational research, and data is frequently siloed. Researchers need a system that can unify these disparate data sources, manage temporal relationships, and provide an intuitive interface for both aggregated analysis and detailed subject investigation, all while ensuring data security and privacy.

Methods

The authors developed DW4TR using a "bottom-up" approach, focusing on three key challenges:

  • Pragmatic Data Model: They created a patient-centric, modularly-structured clinical data model and a specimen-centric molecular data model. This model incorporates temporal relationships and is designed to be extensible to new diseases. They opted for a light-weight ontology, integrating existing standards (like MeSH, SNOMED-CT) where appropriate but also developing custom classifications to meet specific project needs.
  • User-Friendly Interface: Two interfaces were developed:
  • Aggregated Biomedical-Information Browser (ABB): A ROLAP-based tool for aggregated data analysis, allowing users to dynamically create multidimensional pivot views.
  • Individual Subject Information Viewer (ISIV): For detailed viewing and analysis of individual subject data, particularly temporal information.
  • Temporal Information Management: A specific temporal data model was implemented, defining three types of temporal properties (Static, Event, Interval) to tag data attributes and enable temporal querying and analysis.

Datasets

The DW4TR was initially developed to support the Clinical Breast Care Project (CBCP) and has been extended to support the Gynecological Disease Program (GDP). These programs involve collecting clinical data through questionnaires and biospecimens for genomic and proteomic experiments. The paper mentions that as of February 2011, over 5000 subjects were enrolled in the CBCP, with more than 43,000 specimens collected. The GDP had about 500 subjects enrolled. The system also integrates data from other sources through federation.

Key Insights and Ideas

  • Ontology Pragmatism: The paper highlights that a "one-size-fits-all" approach to ontologies is often infeasible for translational research. A pragmatic approach, combining existing standards with custom definitions and planning for future mapping, is more effective.
  • Patient-Centricity is Key: Designing the clinical data model around the physician-patient interaction and standard clinical workflows proved crucial for user adoption.
  • Temporal Data is Critical: Recognizing disease as a process, not a state, necessitates robust management of temporal information, which DW4TR explicitly addresses.
  • Interface Design for Non-Experts: The success of such a system hinges on intuitive interfaces (ABB and ISIV) that empower clinicians and researchers without extensive informatics training.
  • Extensibility is Paramount: The ability to adapt and expand the system to new diseases and data types is essential for its long-term utility in translational research.

Key Results

  • DW4TR successfully integrates clinicopathologic and molecular data for the CBCP and GDP programs.
  • The system enables researchers to perform aggregated data analysis and detailed subject investigation through the ABB and ISIV interfaces, respectively.
  • The ABB interface allows for dynamic creation of multidimensional views, facilitating cohort selection and hypothesis generation.
  • The ISIV interface provides a detailed view of temporal information, allowing for cross-subject comparison and analysis of events over time.
  • Performance tests showed that even with a single-server configuration and a large dataset, key data levels could be displayed within seconds, with expectations of further improvement through hardware upgrades and tuning.
  • The system has demonstrated its extensibility by successfully supporting multiple disease programs.

Limitations and Future Work

  • Ontology Mapping: While the authors developed a pragmatic ontology, they acknowledge the need to map their custom classifications to existing ontologies at a later time to ensure transportability.
  • EMR Integration: The current DW4TR does not directly support EMR data input. The authors have a stepwise plan to integrate EMR records, focusing on de-identification, data extraction, and data model expansion.
  • Molecular Data Support: While progress has been made, the system currently supports a limited number of molecular data types. Future work includes expanding support for other platforms like SNP microarrays, array CGH, RT-PCR, and mass spectrometry.
  • Image Data: While medical images (mammograms, gene expression microarrays) can be accessed, the storage and annotation of these large files present ongoing challenges.
  • Performance Tuning: While performance is promising, further optimization through hardware upgrades and Oracle tuning is planned.

Summary updated

BibTeX
@article{hu2011dw,
  author = {Hai Hu and Mick Correll and Leonid Kvecher and Michelle Osmond and Jim Clark and A Bekhash and Gwendolyn Schwab and De Gao and Jun Gao and Vladimir Kubatin and Craig D. Shriver and Jeffrey A. Hooke and Larry G. Maxwell and Albert J. Kovatich and Jonathan G. Sheldon and Michael Liebman and Richard Mural},
  title = {{DW4TR: A Data Warehouse for Translational Research}},
  journal = {Journal of Biomedical Informatics},
  year = {2011},
  doi = {10.1016/j.jbi.2011.08.003},
}
Figure