Using Ontologies to Identify Patients with Diabetes in Electronic Health Records

This paper describes a work in progress that explores the applicability of ontologies to solve problems in the medical domain. We investigate whether it is feasible to use ontologies and ontology-based data access (OBDA) to automate common clinical tasks faced by general practitioners (GPs), which are labor-intensive and error prone in terms of relevant information retrieved from electronic health records (EHRs). Our study aims to improve the selection of diabetes patients for clinical trials or medical research. The biggest impediment to automating such clinical tasks is the essential bridging of the semantic gaps between existing patient data in EHRs, such as reasons for visit, chronic conditions and diagnoses, pathology tests and prescriptions stored in general practice EHRs (GPEHR), and the ways which medical researchers or GPs interpret those records. Our current understanding is that automated identification of diabetes patients can be specified systematically as a solution supported by semantic retrieval. We detail the challenges to building a realistic case study, which consists of solving issues related to conceptualization of data and domain context, integration of different datasets, ontology creation based on the SNOMED CT-AU® standard, mapping between existing data and ontology, and the challenge of data fitness for research use. Our prototype is based on data which scale to thirteen years of approximately 100,000 anonymous patient records from four general practices in south western Sydney.