A Semantic Wiki for the Engineering of Diagnostic Guideline Knowledge Reinhard Hatko, Jochen Reutelshoefer, Joachim Baumeister and Frank Puppe Institute of Computer Science University of Würzburg, Germany {hatko,reutelshoefer,baumeister,puppe}@informatik.uni-wuerzburg.de 2. CLINICAL GUIDELINE MODELLING The work presented in this paper is conducted within the ABSTRACT project “CliWE - Clinical Wiki Environments”1 . We inves- This paper presents a wiki environment for the modelling of tigate languages, tools and methodologies to collaboratively Computer Interpretable Guidelines (CIGs) using the graph- build Computer Interpretable Guidelines (CIG) by domain ical language DiaFlux. We describe a wiki-driven develop- specialists themselves. The requirement concerning the lan- ment process using a stepwise formalization and allowing guage is the development of an explicit and executable repre- for almost self-acquisition by the domain specialists. The sentation of diagnostic knowledge for active decision-support applicability of the approach is demonstrated by a project systems. Furthermore, we create a development process for developing a guideline for sepsis diagnosis and treatment by simple and effective knowledge acquisition by domain spe- a collaboration of clinicians. cialists. Finally, the completed knowledge bases will be ex- ported into mixed-initiative systems, that cooperate with the clinician during the care process. 1. KNOWLEDGE ENGINEERING WITH SEMANTIC WIKIS Clinical guidelines have shown their benefits by providing Today, Semantic Wikis are used for collaborative ontology standardized treatment based on evidence-based medicine. development, by providing a flexible, web-based interface to Many textual guidelines are readily available and also shared build semantic applications. The main benefit of Semantic through the internet, but rely on the proper application by Wikis is their possibility to interweave different formaliza- the clinician during the actual care process. Much effort tion types of knowledge in the same context. That way, has been put into the development of formal models for ontological concept definitions are mixed with free text and computer-interpretable guidelines. In the variety of CIG images within the wiki articles. models, each has its own focus, e.g. the sharability of guide- lines between various institutions or assisting patient care In this paper, we introduce the graphical language DiaFlux through active decision-support [2]. The focus of the Dia- for modeling of clinical guidelines: The contributions are Flux guideline language lies in the executability of the de- its simple application for developing decision-support sys- veloped models. tems, only providing a limited number of intuitive language elements. Due to its simplicity, it is possible to be used For the specification of a clinical guideline, two types of by domain specialists and thus eases the application in the knowledge have to be effectively combined, namely declara- knowledge engineering process. To allow for comfortable tive and procedural knowledge. While the declarative part development of DiaFlux guidelines, we introduce a visual encompasses the facts and their relationships, the procedu- editor integrated into the Semantic Wiki KnowWE [1]. ral one reflects the knowledge about how to perform a task, i.e. deciding which action to take next. DiaFlux guidelines KnowWE provides the possibility to define and maintain are based on flowcharts. Distinct node types allow, e.g., ontologies together with strong problem-solving knowledge. the conduction of tests, the changing of diagnosis states and Thus, the wiki can be used to collaboratively build decision- calling other DiaFlux guideline as self-contained modules. support systems. These enhancements require extensions of To express the procedural aspect of the guideline, nodes are the standard Semantic Wiki architecture by a task ontology connected by edges, which are labeled with conditions. They for problem-solving and an adapted reasoning process. evaluate the declarative knowledge with respect to the ob- served findings, e.g. the outcome of a given test or the status of a diagnosis. To obtain the semantics necessary for exe- cutability, we rely on an application ontology as an exten- sion to the task ontology of diagnostic problem solving [1]. The application ontology defines the declarative knowledge consisting of findings and their ranges, diagnoses, and treat- ments. 1 Funded by Drägerwerk AG & Co. KGaA, Lübeck, Ger- many, 2009-2012. Figure 1: The main module of the sepsis diagnosis and treatment guideline, opened in the visual editor. An AJAX-based editor for DiaFlux guidelines is integrated 3. CASE STUDY: SEPSIS MODELLING into KnowWE. Ontological concepts that are already defined In the context of the project “CliWE” we used a prototype in the wiki’s knowledge base can be reused or new ones can of the clinical wiki environment for the development of a be created. The model’s source code is encoded in XML guideline covering the diagnosis and therapy of sepsis. Sepsis and integrated into the corresponding wiki article and saved is a syndrome of a systemic inflammation of the whole body and versioned together with it. This allows for further docu- with a high mortality (30 to 60%). mentation of the guideline by tacit knowledge in the article. When the article is displayed in a web browser, the model The knowledge base was developed in accordance to the offi- visualization is rendered. cial guideline by the German Sepsis Society. Our formaliza- tion of the guideline covers so far the diagnostics and parts Collaborative development requires to track the changes of of the therapy (cf. Figure 1). It contains about 50 nodes in all participants. Therefore, a frequent task is to compare eight modules with several possible pathways, depending on different versions of a wiki article. For DiaFlux guidelines a how the diagnosis can exactly be established and the course visual diff is provided, highlighting changes for easy com- of the therapy. Overall, the wiki-based approach showed parison. After creating a knowledge base in the wiki, a its applicability and usefulness, as the combination of for- test session can directly be started from the wiki. The cur- mal and informal knowledge and its gradual refinement was rent state of the guideline throughout the session can be intensely used during the acquisition of the guideline and observed, highlighting the traversed pathway through the according test cases. guideline. This immediate feedback considerably eases the interactive testing of the knowledge base, especially when 4. ACKNOWLEDGEMENTS parallelism or hierarchical structuring are involved. We thank Prof. Dr. Norbert Weiler and Dr. Dirk Schädler, University Medical Center Schleswig-Holstein, Campus Kiel, For the development of DiaFlux models we propose a step- for their efforts during the development of the sepsis guide- wise formalization process: At first, informal information line. are collected in wiki articles, e.g. about goals of a protocol. During the next step, a first semi-formal guideline is created using only nodes containing free text and connecting edges. 5. REFERENCES At this stage of formalization, the flowcharts can not be au- [1] J. Baumeister, J. Reutelshoefer, and F. Puppe. tomatically executed, but “manually”. For testing purposes KnowWE: A semantic wiki for knowledge engineering. the user can explicitly select the pathway through the guide- Applied Intelligence, 2010. line. The taken pathway is highlighted for easier tracking. [2] P. de Clercq, K. Kaiser, and A. Hasman. The last step is the full formalization into a DiaFlux model Computer-interpretable guideline formalisms. In A. ten and the creation of the application ontology, resulting in Teije, S. Miksch, and P. Lucas, editors, Computer-based an executable knowledge base. By following this process of Medical Guidelines and Protocols: A Primer and gradual refinement, the entry barrier for domain specialists Current Trends, pages 22–43. IOS Press, Amsterdam, is quite low, while knowledge acquisition can start from the The Netherlands, 2008. beginning.