A Comprehensive Representation Methodology for Computer-based Patient Records
Chimezie Ogbuji
Primary takeways
- Chasm between content ontologies and message exchange formats
- The importance of requirements in conceiving reusable software solutions
- The value of dual-representation (GRDDL)
- The Computer-based Patient Record (CPR) Requirements
-
Secure (or Semi-closed) Semantic Web Architecture
- Trust and rules might not be sufficient for security in closed CMS
- Use of foundational ontologies
Backdrop
-
The Computer-Based Patient Record: An Essential Technology for Health Care.
- Cleveland Clinic's Cardiovascular Information Registry
-
3 year project to completely reconstituted as an XML & RDF repository
- 200,000 patients
- 100+ publications per year
- 15-20 new cases collected per day
-
Details..
- Controlled data entry (XForms)
- Semantic data warehouse model within the institution
- Migrated a (very large) legacy relational model to a patient record ontology
- Security model: ACL associated with Named Documents / Graphs
- High-volume content management
- Data export for statistical analysis
- Use of open source tools and open standards (XSLT,XML,RDF,OWL,N3,SKOS,XSD)
The problem: content ontology proliferation
- SNOMED, Open Galen, UMLS, OBR, FMA, biozen ..
- High proliferation, little commonality
-
Not (neccessarily) suited for uniform & concrete serialization (data collection)
- More a fault of their representation formats than the vocabularies
- The (often understated) impedance mismatch between KR and document syntax
- Mostly oriented around biomedical reality not healthcare information
The problem: use of message formats for KR
- HL7 Reference Information Model (RIM)
- Not biologically (or clinically) comprehensive
-
Ontologically inconsistent
- The primary goal is data exchange not KR
Wide spectrum of representation requirements
- Two extremes in a wide spectrum of medical record requirements
- Uniform message exchange versus knowledge representation
- XML and RDF have become mutually exclusive solutions for representation
A software engineering analogy
- What are the symptoms of poor requirements gathering in the software development cycle?
Suggested Solutions
- Dual web representation (XML,GRDDL,RDF,OWL)
- Web architecture (standards-based)
- CPR requirements
- Foundational content modeling
Dual Representation (GRDDL)
- GRDDL: Emerging technology for automated extraction of RDF from XML
- Crucial for medical record representation and content management
- Encourages fast adoption and reuse of single-purpose vocabularies
- ESL analogy
-
Cheap technology:
- Presentation (XSLT, XHTML)
- Form-based data entry (XForms)
- Uniform, unambigous interpretation (RDF,RDFS,OWL)
GRDDL best practices for vocabularies
-
Associate an XML to RDF transform for extracting RDF from established XML vocabularies
- XSLT is the more common language for writing such transforms
-
Develop a concrete XML syntax for your stand-alone RDF vocabulary
- Consider a GRDDL transformation for extracting CPR descriptions from HL7 documents / messages
Computer-based Patient Record: what is it?
- Old, but very relevant, specific set of requirements by Institute of Medicine
- Expresses the same concerns about lack of clear definition / requirements
- 15 years and counting...
-
Emphasizes:
- The need to support Problem-oriented Patient Record (POPR) standards
- The need to facilitate decision support
- Can be met with semantic web architecture, document processing, GRDDL, and rich web backplane
Meeting CPR functional requirements
- Uniform, extensible record content (dual representation)
- Record format (open vocabulary standards)
- System performance (RDF-DBMS and Content Management Systems - CMS)
- Linkages (web architecture)
- Intelligence (logical reasoning and decision support - DL, DLP, ACPP models)
- Reporting Capabilities (Document processing - XSLT, XHTML, PDF, tabular exports - CSV)
- Security (ACL, Named Document / Graph, Content Managements Systems - CMS)
- Multi-views (seperation of content / presentation / interpretation XML, GRDDL & XSLT)
- Accessiblity (browser-based standards - CSS, XHTML, XSLT, XForms, SVG, etc ...)
Architecture, representation, and content models
Meeting CPR content modeling requirements
- Alignment with POPR standard
- Longitudinal, life-long content (time-orientation OWL time)
- Provenance and source (DC)
- Provider Identification and participation (FOAF?)
- Patient Identification (FOAF,vCard,..)
- Controlled vocabularies and thesauri classifications (SKOS)
-
Uniform (core) information and process model
- (HL7 RIM OWL? - acts, entities, roles, causation, participation)
- Use of interoperable, foundational ontologies
Diagram of content models for CPR
Going forward
- Establish a group to steward best practices?
- An interest group note?
-
POPR concepts are minimal and well established
- Develop a core ontology for POPR concepts?
-
Align POPR vocabulary with:
- HL7 RIM (OWL?)
- OpenEHR?
- ACPP
- OWL-time
-
Foundational ontologies (DOLCE, OBR, SNOMED, UMLS)
- Write namespace transformations for extracting POPR RDF from HL7 RIM documents
Acknowledgements
- http://www.nap.edu/catalog/5306.html - Online book
- http://doi.ieeecomputersociety.org/10.1109/TAI.2002.1180795 - Ontologies for Knowledge Representation in a Computer-Based Patient Record
- http://www.w3.org/TR/grddl/ - Gleaning Resource Descriptions from Dialects of Languages
- http://www.w3.org/TR/webarch/ - Architecture of the World Wide Web, Volume One
- http://www.w3.org/MarkUp/Forms/2006/backplane/ - Rich Web Application Backplane
- http://www.w3.org/DesignIssues/LinkedData.html - Linked Data
- http://ontology.buffalo.edu/bio/OBR.pdf - A Strategy for Improving and Integrating Biomedical Ontologies