The EON project at Stanford Medical Informatics seeks to create an architecture made up of a set of software components and a set of interfaces that developers can use to build robust decision-support systems that reason about guideline-directed care. We implemented the EON architecture by building middleware components (reusable, embeddable software modules) such as a temporal database mediator for handling requests of time-dependent data from a patient database, domain models for multiple clinical specialties, a generic and extensible ontology for modeling clinical guidelines and protocols, an eligibility-determination server, a protocol-based therapy planner; and a mediator for explaining and visualizing the behavior of other EON components. We had tested the models and the software components by building computable models of several guidelines and protocols and by collaborating with Department of Veteran Affairs and other Stanford researchers and clinicians to develop and deploy the ATHENA decision support systems for guideline-based management.
In the EON project, a guideline modeler uses the Protégé knowledge-editing environment to create and maintain models of concepts and relations in the medical specialty and of clinical guidelines and protocols. We use a meta-class facility in Protégé to define the top-level categories (e.g., a laboratory-test result) and their attributes (e.g., the upper and lower bounds of the laboratory-test result). We use the class/subclass relationship in Protégé to define kind-of hierarchies in the clinical domain. From the ontology of guidelines and protocols, Protégé generate forms for the acquisition of individual guidelines, and the Protégé Axiom Language allows us to write constraints that maintain consistency of the guideline knowledge base. The plug-in architecture of Protégé makes possible a number of specialized visual tools for entering guideline knowledge. The diagram widget, for example, is a special tool that allows us to model the clinical algorithm of a guideline or the schema of a clinical-trial protocol as a directed graph where nodes correspond to decisions or actions and arcs temporal sequencing relationships or alternatives in decision.
The current EON software components uses the Protégé API to access the domain model and the guideline knowledge base and CORBA IDL to specify interfaces among components. By generating Java classes from Protégé classes and creating Java methods that can be invoked, we are able to add behavior to the frame-based knowledge base that Protégé provides. By using the CORBA technology, we are able to distribute EON components as clients and servers that are available from anywhere in the Internet. A web service version of the software is under development at Department of Veteran Affairs (VA) Palo Alto Healthcare System (as of 2006).
Funding for the EON project ended in 2001. Use of the technology continues at VA (as of 2006).