Exploring Prognostic Indicators in Home-Based End-of-Life Care: A Preliminary Study Using Nursing Records and Text Mining

Accurate prognostic prediction is essential in end-of-life care, particularly in Japan, where cultural values emphasize the importance of being present at the time of death. However, tools that are both accurate and feasible for use in home settings remain limited. Home-visiting nurses are often the primary healthcare professionals most involved during the terminal phase, providing detailed observations through their nursing records. These records may offer valuable insights for prognostic prediction in home-based care. A retrospective analysis was conducted using nursing records from 71 patients who received care from the Nana-r Home-visit Nursing Station between April 2016 and December 2023. Data from the final month before death were analyzed, comprising 1,217 nursing records and 57,884 words of text data. Text mining with KH Coder was used to examine co-occurrence networks and analyze keyword patterns in relation to the number of days until death. The analysis identified themes such as excretion-related activities, distress symptoms, skin and oral health, and consciousness levels. Temporal patterns revealed that terms like "walking" and " eating" were more common earlier, while "consciousness," "breathing," and "urination" appeared closer to death, reflecting systemic decline. These findings suggest that nursing records, particularly their free-text descriptions, provide meaningful insights into the terminal phase. Integrating these observations with quantitative data may support the development of practical tools to improve care strategies for patients receiving end-of-life care in home settings.