Risk Factors Extraction from Clinical Texts based on Linked Open Data

Proceedings of Recent Advances in Natural Language Processing, pages 161–167, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_019 161 Risk Factors Extraction from Clinical Texts based on Linked Open Data Svetla Boytcheva1 and Galia Angelova1 and Zhivko Angelov2 1 Institute of Information and Communication Technologies, Bulgarian Academy of Sciences, Sofia, Bulgaria svetla.boytcheva@gmail.com , galia@lml.bas.bg 2 ADISS Lab Ltd, 4 Hristo Botev blvd., 1463 Sofia, Bulgaria angelov@adiss-bg.com Abstract This paper presents experiments in risk factors analysis based on clinical texts en- hanced with Linked Open Data (LOD). The idea is to determine whether a pa- tient has risk factors for a specific disease analyzing only his/her outpatient records. A semantic graph of "meta-knowledge" about a disease of interest is constructed, with integrated multilingual terms (labels) of symptoms, risk factors etc. com- ing from Wikidata, PubMed, Wikipedia and MESH, and linked to clinical records of individual patients via ICD–10 codes. Then a predictive model is trained to fore- tell whether patients are at risk to de- velop the disease of interest. The testing was done using outpatient records from a nation-wide repository available for the period 2011-2016. The results show im- provement of the overall performance of all tested algorithms (kNN, Naïve Bayes, Tree, Logistic regression, ANN), when the clinical texts are enriched with LOD re- sources. 1 Motivation Recently, with the improving quality of Natural Language Processing (NLP), it is increasingly rec- ognized as the most useful tool to extract clinical information from the free text of scientific medical publications and clinical records. In this way NLP becomes an instrument supporting biomedical re- search and new application scenarios are sought to reveal patterns and dependencies expressed by medical texts. Open-source NLP software ap- pears, tailored to clinical text, and this increases NLP dissemination and acceptance. The construc- tion of language resources for biomedical NLP goes in parallel to technology development. The large variety of medical terminology systems is continuously transformed and integrated into stan- dardized, structured repositories of Linked Open Data1; de-identified data sets of electronic health records (EHRs) are made available as open re- sources2. Current hype in open linked data and collective efforts for their generation allow to ben- efit from the multilingual versions of some en- cyclopedic datasets like Wikidata and Wikipedia. Still there is a lack of NLP tools and linguistic re- sources with sufficient quality for processing med- ical texts in languages other than English but the interest to process such texts increases too. Our goal is to determine whether a patient has risk factors for a specified disease, according to the information in his/her outpatient record. We suggest to enrich patient-related clinical narratives with additional information sources in order to en- able a deeper investigation of dependencies be- tween diseases and risk factors. In general it is difficult to predict the risk of a certain disease from the text of a clinical record only. Patient his- tory contains numerous facts that are documented within a series of records but most often the med- ical expert reads them …

Content truncated to 500 words