April 2021 to March 2022

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

   The preparation of care plans is one of the most burdensome tasks in care management, and it has been pointed out that these plans can also vary depending on the care manager that prepares them. As a result, there are high expectations that the use of AI can help make improvements in this area. We designated the three years from FY2017 as Phase 1, and we clarified that approaches favoring both quantity and quality of data are important for the development of AI that supports the creation of care plans. In this research, which is designated as Phase 2 for the three years from FY2020, we worked on five studiesin collaboration with home care support providers and NEC, which has cutting-edge technology in white-box AI. 

   In our efforts to visualize the thought processes of care managers, we created a multilinear model of cerebrovascular disease, following on from the research of cardiac disease in the previous fiscal year. The AI engine used in this research can automatically find highly accurate rules from diverse types of data, sort the results into groups, and make optimal predictions based on these rules according to the circumstances. By allowing the AI to learn items obtained by the multilinear model in the form of rules, we have made it possible to create new groupings based on criteria similar to judgments made by a highly specialized care manager. 

   We also worked on reflecting age-related physical and mental decline in the form of adjustment variables. In the elderly, the deterioration of physical and mental faculties is an inevitable consequence of disease and aging. By using a technique called propensity scores, we extracted a group of similar users and compared them with the average of that group to determine whether the deterioration was suppressed. This enabled us to more accurately reflect changes in the condition of users in AI analysis, thereby helping to improve the accuracy of the AI algorithm. 

   In Phase 2, we have been working to improve the accuracy of the AI algorithm by conducting analyses that are closer to real circumstances using not only the accumulated data but also the knowledge of skilled care managers and related professionals. In the next fiscal year, which will be the final year of Phase 2, we will improve the accuracy of the algorithms by constructing AI models for eight categorized clusters. To further clarify the image of social implementation of these AI models in supporting the creation of care plans, we plan to verify how to present the evidence that is a characteristic by-product of white-box AI, how to promote awareness when creating care plans, and how care managers can use the results derived by AI. We are also considering how the knowledge accumulated through the implementation of this research project can be organized and widely shared with society.