April 2019 to March 2020

[Secretariat]
Kazuko Yuma (Chief Fellow, Institute for International Socio-Economic Studies)

This study can be considered a continuation of the studies conducted on projects promoting health and wellness for the elderly (Elderly health and welfare promotion subsidies) in fiscal years 2017 and 2018, working to transform data produced in care management into data that can be used for analysis with artificial intelligence. 

User needs and goals, and service content are described with text in the care plan’s formats specified by the long-term care insurance system of Japan, so they present great difficulties when AI learns from their data. However, though difficult for training an AI, this sort of data contains important information so ways must be found to use it with AI. We applied textual analysis technology, structuring the data in terms of issues (needs), long-term objectives, short-term objectives, and service content, and then by structuring the data as a care-management process (needs→long-term objective→short-term objective→service content) attempted to perform AI analysis.

To make the AI trained using structured and organized text data even more practical, we also studied ways to reflect the knowledge and experience of specialist care managers into the care-plan AI algorithms. We also established a new working group composed of care managers with different types of qualifications, such as nurses, social workers, and occupational therapists, and worked on "Care manager oriented visualization," or ways to reduce the extremely large amount of information obtained from assessment results, to content that can be written in a care plan, and studied its potential for improving the accuracy of the AI as a knowledge base. These were summarized in the report.