TMUNSW: Identification of Disorders and Normalization to SNOMED-CT Terminology in Unstructured Clinical Notes

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 394–398, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics TMUNSW: Identification of disorders and normalization to SNOMED-CT terminology in unstructured clinical notes Jitendra Jonnagaddalaa,b,c Siaw-Teng Liaw*,a Pradeep Rayb Manish Kumarc School of Public Health and Community Medicine a, Asia-Pacific Ubiquitous Healthcare Research Centre b, Prince of Wales Clinical School c University of New South Wales Sydney 2031, Australia {Z3339253, siaw, p.ray, manish.kumar}@unsw.edu.au Hong-Jie Dai* Graduate Institute of Biomedical Informatics, College of Medical Science and Technology Taipei Medical University Taipei City 110, Taiwan hjdai@tmu.edu.tw Abstract Unstructured clinical notes are rich sources for valuable patient information. Information extraction techniques can be employed to extract this valuable information, which in turn can be used to discover new knowledge. Named entity recognition and normalization are the basic tasks involved in information extraction. In this paper, identification of disorder named entities and the mapping of identified disorder entities to SNOMED-CT terminology using UMLS Metathesaurus is presented. A supervised linear chain conditional random field model based on sets of features to predict disorder mentions is used in conjunction with MetaMap to identify and normalize disorders. Error analysis conclude that recall of the developed system can be significantly increased by adding more features during model development and also by using a frame based approach for handling disjoint entities. 1 Introduction * Electronic health record (EHR) also referred to as electronic medical record (EMR), electronic patient record (EPR), or personal health record (PHR) store or capture patients’ medical history. * Corresponding author This work is licensed under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/ EHR data typically contains demographics, medications, administrative and billing data. The contents of EHR can be either in structured, semi- structured or unstructured. Clinical notes contribute to majority of the unstructured data in EHR. Clinical notes in EHR are often plain text records and valuable resources to obtain patient information (Denny, 2012). Clinical notes are rich in content and may include information on a patient’s demographics, medical history, family history, medications prescribed and lab test results. Information extraction tools can be used to extract the aforementioned unstructured data to discover new knowledge (Jensen, Jensen, & Brunak, 2012). Named entity recognition (NER) is an important subtask of information extraction to identify the boundaries of named entities. Clinical notes often include a wide variety of entities like diseases, disorders, anatomical sites, symptoms and procedures. However, often these entities are expressed in various forms and formats. Normalization is another sub-task of information extraction where the entities identified during NER are accurately mapped to concepts of standard terminologies or ontologies. Rich tools and resources are available to access various 394 standard terminologies and ontologies. Unified Medical Language System (UMLS) Metathesaurus and National Center for Biomedical Ontology (NCBO) BioPortal are two resources that are very useful for normalization in the biomedicine domain. The UMLS Metathesaurus provides access to medical standard terminologies such …

Content truncated to 500 words