KOTOBAKARI Study: Using Natural Language Processing of Patient Short Narratives to Detect Cancer Related Cognitive Impairment

摘要

Background: Recent reports of some studies have described that the cognitive function of cancer patients often declines by a phenomenon designated as cancer related cognitive impairment (CRCI). For patients’ decision-making, detecting CRCI is important. To do so, this study uses language-based CRCI screening to examine participants’ language ability. Objective: This study was conducted to ascertain whether a Natural Language Processing (NLP) based system can detect CRCI, or not. Materials and Methods: We obtained materials of two types from cancer patients (n = 116): (1) speech samples on three topics, and (2) cognitive function level test scores from Hasegawa’s Dementia Scale – Revised (HDS-R), a test used in Japan for dementia patients. The test is similar to the Mini-Mental State Examination. Results and Discussion: Cancer patients with lower HDS-R scores showed a significantly lower Type Token Ratio (TTR). Conclusion: This result demonstrates the feasibility of the proposed speech–language-based CRCI screening method.