Center for Clinical Computing, Beth Israel Deaconess Medical Center, Harvard Medical. School, Boston, MA. Several efforts ... a problem-list application, it would be desirable that an underlying ... of the two as an episode of secondary syphilis.
Capturing Clinical Data, Encoding Clinical Knowledge-Implementing a Dual Perspective in Formal Clinical Ontologies Howard S. Goldberg MD, Charles Hsu, Charles Safran MD, MS Center for Clinical Computing, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA Several candidate methods exist to exploit a dual perspective within a single ontology. By directly manipulating the frame ontology language, one may access knowledge encoded in ontology definitions. By "reasoning about individuals," one can determine what concepts recognize a hypothetical instantiation. Finally, one may create parallel ontologies where instances in the "clinical knowledge" ontology are mapped to concepts in the "documentation" ontology. Tradeoffs between these methods exist in query complexity, expressivity, and ontology maintenance.
Several efforts have been undertaken to create very large, formal ontologies for health care applications."2 Such ontologies are envisioned as the cornerstone of future generations of the electronic medical record in order to 1) guide the representation of clinical events for clinical documentation; 2) represent clinical knowledge necessary for decision support and "intelligent" applications. Because ontologies developed using formalisms such as description logic create concepts by describing definitions, relationships, and constraints, it appears that such ontologies are well-suited to serve as clinical terminologies that can capture clinical data while encoding Whether this dual clinical knowledge. perspective can be adequately incorporated into a single very large clinical ontology remains to be demonstrated.
There is tension between ontology development guided by the bias of a particular application suite and the desire to reuse very large ontologies over an entire domain. As researchers publish clinical for developing methodologies ontologies, it will be important to describe standard idioms employed, expressivity constraints of the modeling language, and target applications. Further research is necessary to demonstrate modeling idioms and expressivity features that will allow a single clinical ontology to support both capture of clinical data and queries of clinical knowledge.
Formal properties of description logics support the creation of detailed, coherent clinical descriptions, automated detection of incoherent descriptions, and automated elaboration of descriptions through inference.3 For example, in a problem-list application, it would be desirable that an underlying ontology support the documentation of a presenting maculopapular rash and the finding of a positive syphilis serology, and subsequently guide the integration of the two as an episode of secondary syphilis. The documentation process also creates the necessary data artifacts to support queries for clinical data warehouse applications, such as How many cases of syphilis presented with a maculopapular rash?
ACKNOWLEDGEMENTS Supported in part by a cooperative agreement award from AHCPR and NLM (HS08749) and a contract from NIST (1995-10-0067N) 1
Rector A, Bechhofer S, Goble CA, Horrocks I, Nowlan WA, Solomon WD. The GRAIL concept modeling language for medical terminology. Artificial Intelligence in Medicine, 9:139-171, 1997. 2 Spackman KA, Campbell KE, Cote RA. SNOMED RT: A reference terminology for health care. JAMIA Suppl. 1997;4:640-644. 3MacGregor RM. Using a description classifier to enhance deductive inference. Proc 7th IEEE Conference on Al Applications, 1991:141-147.
On the other hand, ontologies designed solely to answer questions of general clinical knowledge, e.g., What rashes are associated with syphilis, will differ in organization from ontologies designed to represent clinical events. Instances of the former will describe only prototypical clinical objects and relationships. This is unsatisfactory for clinical documentation purposes because there is no opportunity to specify for any particular episode of disease what specific manifestations were present.
1091-8280/99/$5.00 O 1999 AMIA. Inc.
1067
