GLIF, an execution model for clinical guidelines, was augmented by attributes from GEM, a document model. Our preliminary evaluation, based on an Alz-.
Combining a Document Model and an Execution Model for Clinical Guidelines James Q. Yin, M.D., Ph.D. 1 , Mor Peleg, Ph.D.2 , Aziz A. Boxwala, M.B.B.S., Ph.D.1 , Robert A. Greenes, M.D., Ph.D.1 1 Decision Systems Group, Harvard Medical School, Brigham & Women’s Hospital, Boston, MA; 2 Stanford Medical Informatics, Stanford University School of Medicine, Stanford, CA GLIF, an execution model for clinical guidelines, was augmented by attributes from GEM, a document model. Our preliminary evaluation, based on an Alzheimer’s disease treatment guideline, shows that the merged model has enhanced capacity to relate decision and actions to explanations and the evidence upon which they are based. Background: The Guideline Elements Markup (GEM) model [1] was developed at Yale as a way of tagging elements of narrative guidelines, to enable them to be more precisely accessed and retrieved. Many of these elements relate to the nature of the process used to develop the guideline, kinds of evidence, and explanations and references for recommendations. The GuideLine Interchange Format (GLIF) [2] is a modeling approach developed at Harvard, Stanford, and Columbia (the “InterMed Collaboratory”) aimed at creating a shared guidelines representation that is computer interpretable, in order to integrate guidelines into clinical systems, and deliver patient-specific recommendations at the point of care. GEM uses XML tags to identify various elements in a guideline document; hence it is a document model. The GLIF representation of a guideline is based on an object-oriented model of classes of guideline steps that support execution. It was our hypothesis that GLIF would benefit from the inclusion of GEM elements that provide links to portions of the narrative document providing background information on which various decision steps and actions are based; we therefore extended the GLIF model by adding GEM elements to it. We evaluated the resulting model informally by applying it to an Alzheimer’s disease treatment guideline [3]. Methods: We mapped GEM elements to GLIF class definitions, by finding those that were equivalent, or adding new attributes to GLIF classes representing relevant GEM documentation descriptors, or in some cases adding new classes to the GLIF model. We used both the Together/J UML tool and the Protégé knowledge-modeling tool [4] to represent the merged GLIF-GEM ontology. While the UML tool was used to output a report of the ontology, including class diagrams and documentation, Protégé was used to encode the Alzheimer guideline by creating instances of the ontology classes. We also marked the same guideline with the GEM Cutter tool [5] to ensure that
we did not miss any GEM elements. We tabulated the frequencies of occurrence of the various GEM elements in the encoding of the Alzheimer guideline using the GLIF-GEM ontology. Results: In developing the merged ontology, we mapped 8 GEM classes and about 80 elements to the GLIF class definitions, by modifying 7 of the existing GLIF classes and by creating 10 new classes, where no such constructs were present in GLIF. The only GEM class that was not mapped was Knowledge Components because the GLIF model contains the same or similar attributes and corresponding functions. Of all the mapped classes, Evidence was most used (50 times) when encoding the Alzheimer’s guideline. A total of 288 new information items in the Alzheimer guideline were incorporated by inclusion of GEM elements. Discussion: The GLIF and GEM models are complementary, in that an execution model would benefit from links to the background information on which various decision steps and actions are based; and readers of narrative guidelines should be able to relate phrases in the guidelines to steps that actually implement them, for example in a browsable computer-based flow chart rendition of a guideline. Acknowledgment: We thank R.N. Shiffman for making GEM materials available and review of the work. Supported in part by Grant LM06594 with funds from the Department of the Army, Agency for Healthcare Research and Quality, and National Library of Medicine. References 1. Shiffman RN, Karras BT, Nath SD, A Preliminary Evaluation of Guideline Content Mark-up Using GEM – An XML Guideline Elements Model. Proc AMIA Symp 2000;413-417 2. Peleg M, Boxwala A, Ogunyemi O, et al. GLIF3: The Evolution of a Guideline Representation Format. Proc. AMIA Annual Symposium; 2000; p. 645-649. 3. Los Angeles Alzheimer's Association. Guidelines for Alzheimer's disease management; 1999 Jan 8 4. Grosso WE, Eriksson H, Fergerson R, et.al., Knowledge modeling at the Millennium. Banff, Canada; 1999. p. 7-4-1 to 7-4-36.
5. http://ycmi.med.yale.edu/GEM/