2017 年 37 巻 6 号 p. 313-321
Our team has been involved in developing a clinical decision support system (CDSS), which requires information about patients’ lifestyle. However, patients’ lifestyle issues are usually encoded in clinician generated narrative texts, which poses significant barriers to their information accessibility. In this paper, we propose an approach to identifying lifestyle issues of obesity, smoking and drinking in electronic health records (EHR) using machine learning and natural language processing techniques. To evaluate our approach, we conduct experiments using clinical narratives from The University of Tokyo Hospital which were generated in 2015 and saved in SS-MIX2 extended storage. The experimental results show that the proposed approach achieves equivalent high performance compared to previous studies focusing on English discharge summaries.