logo

1. Projects using artificial intelligence (AI) or machine learning (ML)

(1) Development of machine learning programs for delirium risk assessment (National Cancer Center Japan, 2022-2023)

Two machine learning algorithms for risk assessment of delirium were developed: the first one predicts the degree of risk of developing delirium using structured data such as vital signs as training data, and was developed based on a prototype algorithm that was created in a previous project; the second one predicts the possibility of developing symptoms that cause delirium by using structured data such as vital signs as training data, and was newly developed. Technical aspects included estimation using 15 different machine learning algorithms, hyperparameter tuning using Optuna and others, and migration of the learning environment using Docker.

(2) Implementation support for development of stratified AI applicable to ultra-high dimensional data (ational Institutes of Biomedical Innovation, Health and Nutrition, 2021)

With the aim of searching for superior targets for drug discovery (proteins and other biomolecules on which drugs act), we conducted research and development of AI algorithms, including subset binding, using data from multi-omics information (comprehensive biomolecular information) and patient medical information.

(3) Development of support system for risk assessment of child abuse and others, and data analysis of child abuse consultation and others (Edogawa City, Tokyo, 2021)

In the area of risk assessment in response to child abuse, the number of years of experience of those in charge of risk assessment has been decreasing year after year as the number of consultations increases, and it has been pointed out that there are gaps in knowledge and experience as well as in the quality of assessment among individuals. Under such circumstances, algorithms for “determining whether a case is eligible for temporary protection”, “extracting similar cases”, and “extracting sentences related to child abuse from written materials” were researched, developed, and systematized. using a natural language processing AI model (transformer-based), taking into account the health status of the child, socioeconomic status of the household, and status of the guardian such as single parenting as the main explanatory variables.

(4) Development of a predictive model of return-to-work decision for employees on mental health leave using natural language processing (Okayama University, 2021)

We have developed an algorithm that uses natural language processing (transformer-based) to determine whether or not an employee on mental health leave would be able to return to work within a certain period of time, such as one year, based on the record (text) of an interview with the employee conducted by a human resources representative or other person in charge at the company.

(5) Development of AI-based "Liveliness Score" estimation system (Chiba University, 2021)

Indicator values were created by machine learning to assess the risk of older adults being certified as needing assistance or care. The estimation was conducted by using an XGBoost-based model, using as the outcome variable whether or not a person was certified as requiring level 2 or higher of care during the most recent follow-up period, and as the predictor variables health status indicators, social participation level, lifestyle, age, and gender. Using this model, the probability of being certified as requiring 2 or higher level of care during a follow-up period was calculated, and an algorithm was researched, developed, and systematized that perform calculations using the degree to which the probability of each case deviates from the average probabilities for each age group as an indicator.

(6) Health Effects Study on Chronic Arsenic Poisoning in FY2021 (Ministry of the Environment, 2021)

We analyzed the causal factors of chronic arsenic toxicosis and its incidence by condition, and others. We determined the presence or absence of the target diseases (arsenic-related) from the medical examination data recorded in Excel, converted the data into binary data for data analysis, and analyzed the incidence by condition such as age and exposure period using survival analysis and other methods.

(7) Survey on administrative resource allocation for public security measures (National Research Institute of Police Science, 2020)

We conducted a study on the occurrence of petty crimes and other crimes from a spatial perspective. Specifically, in order to verify whether or not the occurrence of minor and other crimes is spatially concentrated, research and development of estimation algorithms based on (1) panel data multilevel analysis, (2) time series models, (3) spatial autocorrelation models, etc., were conducted using spatial data.

(8) Analysis of the effect of agriculture on reducing medical costs for the elderly aged 75 and over in FY2020 (Ministry of Agriculture, Forestry and Fisheries, 2020)

We analyzed the effect of engaging in agriculture on medical cost reduction. We matched medical cost data from the National Health Insurance database (KDB) with subjects meeting certain criteria selected from farmland ledgers, and conducted regression analysis (cross-sectional) and other analyses using sex, age, agricultural activity, duration of agricultural engagement, and other variables obtained from a questionnaire survey. In addition, we developed an evaluation method and designed a subsequent research project to be conducted in the future to assess the effect of medical cost reduction over time.

(9) Survey, analysis, and other work related to the streamlining of operations for the Japanese Direct Payment System in FY2020 (Ministry of Agriculture, Forestry and Fisheries, 2020)

In order to reduce the workload of municipal employees, we conducted a survey of 1,400 municipalities nationwide and interviewed five municipalities to understand the current status and issues, and then examined the functions that an IT system for streamlining the work of the Japanese direct payment system should perform. The questionnaire also included open-ended questions asking about the challenges of the system in general, and unsupervised clustering using BERT (a ML model for natural language processing) was conducted based on the degree of similarity between these open-ended questions.

(10) Decision tree analysis, clustering, and other analyses, using machine learning on risk of pregnant women (Tokyo Medical and Dental University, 2019)

A clustering analysis was conducted on the data of pregnant women at the time of their medical examinations and the subsequent occurrence of child abuse-like behaviors with the aim of understanding the characteristics of pregnant women at high risk of abusing their children. Random forest analysis, a method used to understand the level of importance of each factor, revealed that the factors of "having previously had a baby" and "being single" were ranked at the top of the list. Decision tree analyses of several patterns were also conducted, which confirmed that the same factors were highly influential in the model with the highest predictive power. Ultimately, two high-risk groups were identified.

(11) Development of a scantron-based daily report system as part of the Regional Model Projects for Promoting Forestry as a Growing Industry (Yaita City, Tochigi Prefecture. Recommissioned by FOREST MEDIA WORKS Inc., 2019)

This work is designed to save labor and reduce workload. At forestry work sites, daily reports are written on a whiteboard for information sharing and progress management by each work group of about 3 to 5 people. We developed a prototype of a system in which handwritten numerical information on the daily report is transcribed into a spreadsheet in the cloud.
When a handwritten daily report is photographed with a smartphone and the photo is sent to the system, the system corrects the skew of the photo, cuts out the handwritten numbers (using Python's open CV), and then uses deep learning (machine learning) to read the image of the cut-out numbers and digitize them, and writes the numbers into a spreadsheet in the cloud.

(12) Comprehensive research project on Comprehensive Countermeasures for Declining Birthrate (Ministry of Health, Labour and Welfare, 2019)

A questionnaire survey was conducted to examine the association between the number of children of respondents and support programs offered by the employers of respondents and of their spouses, as well as by the local government. Iterative random forests algorithm using machine learning was used to analyze the association by considering a combination of the programs offered by employers and local governments, rather than individually. The results showed that cash benefits such as child allowances provided by local governments were more important than programs provided by companies.

2. Projects using relatively advanced statistical analysis

(1) Subsidy for Employment Preparation Support Project for People in Need in FY 2008 (Social Welfare Promotion Project) "Survey and Research Project on Medical Expenses and Medical Service Usage by Low-income Households in Other Countries" (Ministry of Health, Labour and Welfare, 2018)

A literature review was conducted on the number of medical visits and changes in health status among low-income households in Japan and abroad when the form and/or status of medical costs changes. In particular, we delved into the situation in each country to understand the situation in terms of medical in-kind services, reimbursement payments, and co-payments. In addition, we conducted an online survey in Japan on the relationship between the burden of medical expenses (or reimbursement) and the decision to visit a doctor, with questions for conjoint analysis, to estimate the extent to which the amount of co-payment, the availability/non-availability of reimbursement, and other factors affect the decision to visit a doctor.

(2) Project to organize and analyze data from a questionnaire survey on the multifunctional role of agriculture in fiscal 2018 (Japan Institute of Irrigation and Drainage, Recommissioning of a Ministry of Agriculture, Forestry and Fisheries project, 2018)

We conducted compilation and analysis of data from a questionnaire survey of prefectures and municipalities regarding the "Law for the Promotion of the Multifunctional Role of Agriculture" and analysis on "improving the stability of payments for multifunctional improvement, direct payments for mountainous areas, and direct payments for environmentally friendly agriculture by legalization" and "strengthening coordination through integrated operation of the three payments" by propensity score matching.

(3) FY 2016 Data organization and other work related to the multi-purpose use of agricultural water at the time of disaster (Japan Institute of Irrigation and Drainage, Recommissioning of a Ministry of Agriculture, Forestry and Fisheries project, 2016)

We assisted in the overall process of a nationwide online questionnaire survey on the willingness to pay for the availability of water supply in the event of a major earthquake. The survey allocated the number of respondents according to the regional classification by specific earthquake risk, and used the multiple-bounded dichotomous choice CVM method to calculate the amount of willingness to pay and estimate the determinants of willingness to pay.

(4) Survey and analysis work on the proportion of project benefits related to agricultural and agricultural community development projects in FY 2013 (Ministry of Agriculture, Forestry and Fisheries, 2013)

Under the grand goal of studying the ratio of project cost burdens shared between local governments and farmers, we examined methods using conjoint analysis to estimate the ratio of willingness to bear the burden of agricultural and extra-agricultural effects and provided support for conducting the study. In addition, in order to set up options for the ratio between the two effects, we estimated the ratio of agricultural and extra-agricultural effects for each type of project based on the results of ex-post evaluation of past agricultural and agricultural community development projects and defined a range of reasonable ratios.

(5) Effectiveness verification and cost-effectiveness analysis of Taketoyo Town Community Salon Project (Seijoh University, 2011)

In order to examine the effect of salon (a place one regularly visits) participation on the maintenance and improvement of health status of the elderly, a reverse causality analysis was conducted using panel data for the entire elderly population living in the community at two time points, using instrumental variable method. (For the final results, refer to: Y Ichida et al. Social Science & Medicine 94, 83-90)