Health Level 7 Common Document Architecture, LOINC, and SNOMED-CT: Lessons Learned ... School of Information, University of Michigan, Ann Arbor, MI. 2. Center for ... of Michigan Health System resources: the Depression. Network's IVR ...
Representing Natural-Language Case Report Form Terminology Using Health Level 7 Common Document Architecture, LOINC, and SNOMED-CT: Lessons Learned
Dale Hunscher1, 2 BA; Andrew Boyd, MD3; Lee A. Green, MD, MPH5; Daniel J. Clauw, MD2, 4 1. School of Information, University of Michigan, Ann Arbor, MI 2. Center for the Advancement of Clinical Research, University of Michigan Medical Center, Ann Arbor, MI 3. Department of Psychiatry, University of Michigan Medical School, Ann Arbor, MI 4. Department of Internal Medicine, Division of Rheumatology, University of Michigan Medical School, Ann Arbor, MI 5. Department of Family Medicine, University of Michigan Medical School, Ann Arbor, MI
Abstract: Clinicians and biomedical research investigators ordinarily use natural language when describing biomedical concepts and constructs, even in the context of highly structured case report forms. We describe work in progress and lessons learned in translating complex natural-language concepts on case report forms into machine-readable format using the HL7 CDA, LOINC, and SNOMED-CT standards. Description of the work: Clinicians and biomedical research investigators ordinarily use natural language when describing history, observations, diagnoses, prognoses, therapies, and other biomedical concepts and constructs. Even in the context of a widely used and well-validated case report form, natural language description is often employed. While this generally improves readability by other human being, it makes translation into machine-level semantics much more complicated. This poster describes work in progress on and fully funded by an NIH Roadmap contract under the Broad Agency Announcement BAA-RM04-23, “Re-Engineering the Clinical Research Enterprise: Feasibility of Integrating and Expanding Clinical Research Networks”. The contract involves creating an automated Honest Broker system that can mediate between heterogeneous clinical care and research data management systems deployed on several University of Michigan Health System resources: the Depression Network’s IVR depression monitoring system, MDOCC; the Cardiovascular Network, consisting of secondary and tertiary care hospitals treating cardiovascular disease throughout Michigan; GRIN, a practice-based research network with a statewide membership of family practices and community clinics; and Velos, a clinical research data management system maintaining deidentified clinical data sets, deployed at the University of Michigan Medical School.
For purposes of electronic data interchange, we created encoding definitions for messages containing data from several case report forms of various types, including the following: SF-12, to assess quality-oflife; PHQ-9, for depression screening; an internally developed form for assessing medication compliance and satisfaction with the quality of care provided by attending physicians and nurses; and a cardiovascular incident report form. The first three consisted of ten to twenty Likert scale questions whose definitions were not included in any standard repository, while the cardiovascular form consisted of well over one hundred fifty questions fitted onto two densely populated pages. After investigating syntactic options, we chose the Health Level 7 Common Document Architecture (HL7 CDA) as our format for the representation of forms data transmissions. With the permission of Pfizer and the original author of the PHQ-9, we worked with the Regenstrief Institute to include the definition of the PHQ-9 data points in the LOINC database as part of its growing collection of survey instruments. We developed a LOINC-like internallymaintained encoding for the compliance form and SF-12 data points. For internally maintained encoding definitions we obtained ASN.1 standard OIDs with an eye toward the future, when publication of our encoding schemes might be possible and desirable. The cardiovascular form presented a significant translation challenge, since a great many diagnostic, procedural, prognostic, historical, and other clinical data points were included, all of which had terse natural language prompts. After much investigation and experimentation we managed to encode all data points using SNOMED-CT. Along the way we attempted to define best practices for representing common semantic constructs and handling ambiguity using SNOMED-CT and HL7 CDA, which practices are fully described in the poster.
AMIA 2006 Symposium Proceedings Page - 961