SSO: THE SYNDROMIC SURVEILLANCE ONTOLOGY Okhmatovskaia A, Chapman WW, Collier N, Espino J, Conway M, Buckeridge DL Ontology Description The SSO was developed in OWL (web ontology language) using Protégé Concepts In SSO, three main classes describe fundamental knowledge about clinical syndromes used for public health surveillance: Syndrome, Clinical Condition (member of sensitive or specific definition of a syndrome), and Clinical Concept (used to define clinical conditions). Additional classes: Keyword, Regular Expression and Coding are used to link syndrome definitions to their potential applications (e.g. chief complaint classifier), and to major external coding systems and vocabularies (e.g. UMLS). Particular syndromes, clinical conditions, and concepts are encoded as the subclasses of the corresponding main classes, for example Respiratory Syndrome is a subclass of Syndrome, while Apnea is a subclass of Clinical Condition. Properties Relationships among the classes are encoded using OWL properties. For example, one of the properties of the Syndrome class is hasSensitiveDefinition, which can be filled in by multiple Clinical Conditions. It is the values of these properties that distinguish among different syndromes, clinical conditions and clinical concepts. For example, both ILI Syndrome and GastroIntestinal Syndrome have sensitive definition, but their sensitive definitions are not the same. One way to express these differences would be to add restrictions to class definitions. Alternatively, these classes can be created as instances of OWL meta-classes, so that a specific value can be assigned to each class property (same as for OWL individuals). We have chosen the second option that provides better human readability. Individuals Specific keywords and regular expressions used to match clinical concepts are encoded as instances of corresponding classes. Background Most syndromic surveillance systems use data from ED visits, often free-text chief complaints [1]. Classification of chief complaints into syndromes is often inconsistent, due to the lack of agreement about the concepts that define a syndrome and how individual terms or strings map to these concepts. Two threads of research have made it possible to address this problem now. First, an effort supported by the ISDS to develop a set of consensus definitions for syndromes in terms of ED chief complaints [2]. Second, experience in using ontology to encode standards and drive knowledge- based surveillance systems, most notably: to support surveillance of media [3], to enhance free-text classification [1], and to support modular aberrancy detection [4]. Present work unites these two threads to ensure that consensus syndrome definitions are disseminated broadly, maintained collaboratively, and incorporated easily into automated systems. Where the definitions come from A set of consensus syndrome definitions that constitutes a conceptual foundation for our ontology was generated by an expert group of syndromic surveillance developers and practitioners representing 10 operating surveillance systems. The group developed definitions for four syndromes: Influenza-Like Illness, Constitutional, Respiratory, and Gastrointestinal, with Respiratory and Gastrointestinal syndromes having sensitive and specific definitions [2]. Applications Automatic documentation The SSO was encoded in a standard portable (XML-based) language, which allows one to easily generate ready-to-use documentation on consensus definitions. See for example: Syndrome definition validation By formally and explicitly representing the definitions, SSO makes it possible to explicitly evaluate them against real data to calculate positive predictive value of specific concepts. This would provide a basis for improving these definitions. Maintaining consistency of definitions A number of highly optimized reasoning engines are available today to perform a variety of knowledge-base inference tasks. These tools can be used to ensure the consistency of syndrome definitions encoded as ontology. Chief complaint classifier The SSO provides an easy way to launch a chief complaint classifier that represents the experience of surveillance systems across the country. There is a demo version of such classifier. Knowledge exchange The SSO encodes the definitions in a transparent and shareable format that can be used by humans and computers. It is intended to serve as the starting point for dialog and experimentation on which concepts best indicate relevant syndromic presentations. It can also be used to compare the performance of different chief complaint classifiers and enable sharing of classification knowledge across surveillance systems. Syndrome Clinical Condition Clinical Concept Keyword Regular Expression hasSensitiveDefinition:1 or more hasSpecificDefinition:1 or more hasExactConcept: exactly 1 hasRelatedConcepts: 0 or more hasSynonymousConcepts:0 or more hasInclusionKeywords: 1 or more hasExclusionKeywords:0 or more matchesKeywords: 1 or more Coding mapsToCoding: exactly 1 hasExternalCoding: 0 or more [1] Lu H-M, Zeng D, Trujillo L, Komatsu K, Chen H. Ontology-enhanced automatic chief complaint classification for syndromic surveillance. Journal of Biomedical Informatics. 2008;41(2): [2] Chapman WW, Christensen LM, Wagner MM, et al. Classify-ing free-text triage chief complaints into syndromic categories with natural language processing. Artificial Intelligence in Medicine. 2005;33(1): [3] Collier N, Kawazoe A, Jin L, et al. A multilingual ontology for infectious disease outbreak surveillance: rationale, design and challenges". Journal of Language Resources and Evaluation. 2007;40(3-4): [4] Buckeridge DL, Okhmatovskaia A, Tu S, O'Connor M, Nyulas C, Musen MA. Understanding detection performance in public health surveillance: Modeling aberrancy-detection algorithms. Journal of the American Medical Informatics Association 2008 Nov-Dec;15(6): University of Pittsburg