Natural Language Processing based Medical Needs Extraction for Breast Cancer Patients from Question and Answer Services (Preprint)

BACKGROUND Currently, a large number of patient narratives are available on various web services. On web question and answer (QA) services, patient questions often relate to medical needs. Therefore, we expect these questions to provide clues to understanding patients’ medical needs. OBJECTIVE This study aims to extract patient needs and classify them into thematic categories. To clarify the patient's needs would be the first step to solve social issues for cancer patients. METHODS The material of this study is patient question texts containing the keyword “breast cancer" in the Yahoo! Japan QA service, Yahoo! Chiebukuro, which contains over 60,000 questions on cancer. First, we convert the question text into a vector representation; then, the relevance between patient needs and existing cancer needs categories are calculated based on cosine similarity. RESULTS The proportion of correct classifications in our proposed method is approximately 70%. We reveal the variation and the number of needs from the results of classifying questions. CONCLUSIONS There are various clinical applications to applying the proposed method such as identifying the side effect signaling of drugs and the unmet needs of cancer patients. Revealing these needs is important to satisfy the medical needs of cancer patients.