A Problem-oriented Medical Record Ontology
Chimezie Ogbuji, The Cleveland Clinic, <ogbujic@ccf.org>
Semantic Web Healthcare and Life Science Interest Group
Why build a patient record ontology?
- Facilitate conversation & consensus around clinical content within the IG
- Lack of ontological commitment within existing clinical vocabularies
- Uniform record format is really the only outstanding medical record requirement
- Semantic Web vision is predicated on vocabulary consensus
- The pieces are mostly there already
Influential literature
- HL7 RIM: An Incoherent Standard
-
Highlights ontological inconsistencies within HL7 RIM between:
- Act of documenting
- The phenomena being documented
- The corresponding record
-
Suggests a seperation between:
- Reference Ontology of the Healthcare Domain
- Model of Healthcare Information (HL7 RIM)
-
Knowledge Representation for Relevance Ranking of Patient-Record Content in Primary-Care Situations
- Emphasizes the use of POMR structure as a core framework
-
Proposes an alignment of
- A process ontology
- A formal (minimal) subsumption hierarchy for clinical care acts
- A POMR-based information ontology
-
A Strategy for Improving and Integrating Biomedical Ontologies
- In depth analysis of the value of foundational ontologies for biomedicine
- Proposes a minimal, comprehensive subsumption hierarchy of biomedical reality
- Aligns with (FMA, BFO)
-
Towards a Reference Terminology for Ontology Research and Development in the Biomedical Domain
- Argues against the use of cognitive representations
- Proposes a set of terms for referring (unambigously) to ontological terms
How can this ontology be used?
- To postcoordinate vocabulary terms for a Computer-based Patient Record (CPR)
-
.. offer a set of pre-enumerated codes covering only a small fraction of all the things that might be said. But when a user can’t find what they want, they can create (postcoordinate) it on demand, defining what they need in terms of the pre-enumerated codes.
-
As a bridge between healthcare information and biomedical ontologies
- As an interlingua for developing use cases which cover clinical content
Design Choices
- The use of expressive, freely available foundational ontologies
- Avoided the use of skos:Concept
- Support for a syntactic mapping to HL7 RIM (via GRDDL)
- Use of OWL time to model temporal logic
- Modeling the act of recording as a primary class
- Formal logic for (objective and subjective) interpretations and signs
- Use of FOAF for terminology related to people
Core classes
cpr:clinical-description
inf:depiction AND rim:Act AND
(foaf:maker EXACTLY 1) AND (dol:has-quality SOME time:TemporalEntity) AND
(edns:realizes ONLY cpr:patient-record) AND (cpr:description-of min 1)
cpr:screening
inf:information-realization AND cpr:diagnostic-act AND
(edns:realizes SOME cpr:medical-problem) AND (dol:particpant SOME cpr:clinician)
cpr:medical-problem
dol:linguistic-object AND (galen:hasSeverity MIN 1) AND (dol:has-quality SOME time:TemporalEntity)
cpr:medical-sign
(edns:interpreted-by ONLY cpr:clinician) AND
(cpr:interpretant-of ONLY cpr:disease) AND (cpr:interpretant-of MIN 1)
cpr:symptom
((dol:interpreted-by ONLY cpr:patient) AND
(cpr:interpretant-of ONLY (cpr:disease OR physiological-state OR physiological-process))))
Ontologies used
- Clinical
- General Architecture for Languages, Encyclopedias, and Nomenclatures in Medicine (GALEN)
- Ontology of Biomedical Reality (OBR)
- Laboratory for Applied Ontology (Clin-Act and Biologic-Functions theories)
- Non-clinical
- Descriptive Ontology of Linguistics and Cognitive Engineering (DOLCE)
- Friend of A Friend (FOAF)
- OWL time
Best practices
- Mappings to foundational ontologies to facilitate ontological commitment
- Adoption of consient terminology for ontological constructs
- Avoidance of constructs which denote cognitive representations (skos:Concept)
- Careful use of partial and complete class axioms
- Clear seperation of temporal semantics
- Exhaustive disjointeness
Motivating usecases
Coronary Artery Bypass Graft (CABG) Indications
The POMR ontology was used to model guidelines for Coronary Artery Bypass Graft operations.
Modeling an Cardiac Catheterization
A clinical description entry in a patient record which describes a cardiac catheterization which shows an arterial stricture.
[ a cpr:clinical-description;
dol:has-quality [ a time:Instant; time:inXSDDate "2006-12-06"^^xsd:date ]
foaf:maker [];
rdfs:label 'Cardiac Catheterization @ 2006-12-06';
cpr:description-of
[ a cpr:screening;
edns:realizes
[ a inf:diagrammatic-object,foaf:Image;
edns:interpreted-by [ a cpr:clinician ];
foaf:depicts
[ a galen:Stricture, cpr:pathological-structure;
galen:hasSeverity "50";
galen:hasSpecificLocation galen:LeftAnteriorDescendingCoronaryArtery;
]
]
]
]
Mapping from HL7 CDA to POMR
The POMR ontology was used as the target vocabulary for a GRDDL transform written for HL7 CDA documents.
Namespace and resources
The cpr prefix is associated with the Namespace URI: http://purl.org/cpr/1.0/problem-oriented-medical-record.owl#
The /cpr PURL top-level domain has been registered as a home for this vocabulary
The Namespace URI resolves to a formal OWL ontology
Problems and Issues
- Expressing periodic time intervals with OWL Time
- No known URI-based naming convention (or OWL export) for SNOMED CT terms
- Lossy semantic transformation from HL7 to RDF
- No feasible means of reasoning over very large ontologies (GALEN, DOLCE, etc..)?