Nursing Records Annotation for Extraction of Patient Status Representation

Abstract

The nursing record is a document that tracks nursing practice series. In addition to certifying nursing practice, it is expected to be used to analyze nursing care by extracting information, to provide feedback to nursing practice, and to develop infromation systems to improve the efficiency of nursing operations. Although such applications require to extract expressions regarding nursing activieies and patient status from nursing records, existing medical named-entity recognition systems have been mainly based on clinical documents written by doctors. We propose a novel patient-status annotation scheme and created an annotated corpus of 1,000 pseudo nursing records. We experimentally applied a BERT-based named entity recognition using the corpus in order to assess the annotation scheme through the model's error analysis.