gms | German Medical Science

Introduction: Recent years saw a large increase in bacterial antibiotics-resistance, leading to serious impact on hospital patient safety. The DebugIT (Detecting and Eliminating Bacteria Using Information Technology) project [1] strives to integrate clinical data from seven hospitals across Europe to enable cross-country resistance-data comparison, safety guideline development and to ultimately counteract antibiotics resistance EU-wide applying an ‘ITbiotics’ approach. In order to detect variances in antibiotics prescriptions and compare their impact on resistance development between different hospitals and countries, DebugIT needs to find a way to access heterogeneous clinical data. Ontologies seem a reasonable approach to achieve such integration.

Material and methods: Access to geographically distributed, multilingual and semantically heterogeneous data is integrated via semantic web technologies like OWL ontologies, automatic reasoner and SPARQL endpoints for querying [2]. D2R servers were deployed for real-time transformation of relational data into a semantic web accessible RDF-format, providing SPARQL access by means of a site-specific Data Definition Ontology (DDO), a wrapper layer matching the relational database schemata. DDOs are used in site-specific data set queries and mediate successive further data formalisation steps, i.e. exploiting ontology conversion rules to derive instances of a highly formal DebugIT core ontology (DCO, [3]), serving as universal cross-site integration layer, allowing cross-site data querying and intelligent result analysis. In parallel database values representing standard terminology entries are converted into DCO instances via SKOS terminology mappings.

Results: The overall DebugIT interoperability platform relies on multiple layers of ontologies, conversion rules and terminology mapping files to allow data integration, subsumptive querying and rule-based data mining approaches, enabling health care pattern recognition and decision support on antibiotics prescriptions.

As the semantic gap from the clinical data to formal ontologies, and vice-versa from clinical questions in natural language to DDO SPARQL-queries, is too large to be bridged by a one step ontology integration approach, we enable users to generate queries in a stepwise multi-layered manner, from the informal to formal semantics, relying on a multiplicity of ontology and terminological mapping layers of different ‘user-compliance’ and ontological rigidity [4].

Discussion/conclusion: Data is formalized ontologically and hence can be integratively queried across all sites and languages, rendering results universally re-useable. Although the presented multilayered data integration approach is complex and requires new user soft-skills, we believe it is nevertheless feasible, as it represents the only scalable solution, enabling seamless integration of further new Hospital sites to join the project.