Early in the SAGE Project we implemented a "synchronization cycle" concept, in which requirements and development of prototypes were driven by the encoding and deployment of a successive series of "exemplar clinical guidelines". During each such work cycle, project work in all areas was focused via the requirements for the encoding and execution of the target guideline exemplar. At the end of each cycle, the latest working prototypes of all SAGE components were used in a formal exercise of as much interoperable guideline functionality as possible at that point in time. This approach to iterative, incremental development and testing allowed us to evaluate our functionality as early as possible, and we have been pleased with the adoption of this project methodology in place of the more traditional "waterfall" methodology.

We have developed three exemplar guidelines. They were chosen to drive the research project by working on real, diverse, clinical problems. One of the objectives of using clinically real exemplar guidelines is to assess our ability to encode and execute the depth of medical knowledge required. As we extended our encoding to the reaches of the clinical problems, we discovered the needs for new features in our model and in our tools to perform the encoding and to test it as well. The first guideline, Immunizations, took an entire year to encode. In that case, we were starting without any previous model constructs or tools. The subsequent guidelines far more quickly. We were able to include much more complex features including medication models, evidence knowledge-bases, and order sets.

IMMUNIZATIONS GUIDELINE Our first exemplar, the Immunizations guideline, implements the complete breadth of current CDC recommendations for childhood and adult immunizations. It includes scenarios for The Neonate and Primary Care, and a Population Immunization Surveillance scenario that can survey a large population of patients.

Each guideline was developed in a step-wise manner. As we discovered the need to change the Guideline Model, those changes were made and the new capabilities were tested in Protégé and then in the SAGE Engine, and recommendations were surfaced in our clinical information system. We "rotated" ownership of guideline encoding (e.g., from GE clinicians to Mayo clinicians) to gain more depth, experience, and understanding of the guideline encoding process.

DIABETES MELLITUS GUIDELINE Here we provide routine monitoring recommendations for Type I and Type II diabetics, including management of both blood pressure and lipids. The guideline does not make specific recommendations for glycemic control.

The Hypertension Management section of the Diabetes guideline was chosen as the vehicle for advancing our representation and use of medication modeling in the SAGE Guideline Model. We also experimented there with alternative means for notifying clinicians about recommendations including through flowsheets. At that time we made two types of "explanations" available for the recommendations.

First, basic background information as to why a particular suggestion was made can be obtained by clicking on the information (i) icon. Here we see the JNC VI explanation for routine blood pressure measurements in Type I and Type II diabetics. In our Diabetes guideline we migrated to JNC VII in 2005.

Second, the information which relates all contributing criteria to any recommended conclusion, can be made available through a link exposed by the target CIS. For example, here we see the reasoning behind a HbA1C recommendation, and the actions being taken by the SAGE engine when evaluating the guideline for a diabetic. This rationale might be surfaced in various ways through a flowsheet, or in a suggested orders panel.

The Diabetes guideline also uses structures to represent medications: classes, individual medications, dosage forms, and other attributes. Representation of medication uses (indications) and dangers (contraindications) was also included in the form of Evidence Statements. The Apelon DTS terminology server was used as the standardized repository of medication terminology.

COMMUNITY ACQUIRED PNEUMONIA GUIDELINE The Community Acquired Pneumonia (CAP) guideline introduced the use of a computed score, the Pneumonia Severity Index to make a triage decision. The SAGE Project contributed to the development of the HL7 Order Set standard and this exemplar used that XML representation for the CAP order set to both select (and de-select) orders and annotate orders and sets of orders for a patient based on the clinical situation at the time the orders are presented to the clinician.

The CAP guideline also used the ability of the SAGE Engine to signal itself to implement the regular check for events of interest to JCAHO measures like the time post-diagnosis of administration of antibiotics. This exemplar guideline also actively gathers information to support JCAHO measures, for example, whether this patient is being transferred from an outside emergency room.

Encoding of our exemplar guidelines has led us to address the issue of external knowledge resources. The application of a guideline to the care of a patient requires that we augment the guidance specified in the guideline with general medical knowledge. It is neither practical nor desirable to assume that all applicable medical knowledge should be encoded as part of a monolithic guideline knowledge base. Thus, effective guideline encoding requires that we have well-defined methods for accessing medical knowledge external to the guideline model.