The research team of Mr. Juki Osawa of the University of Tokyo Hospital, Mr. Yusuke Tsugawa of the University of California, Los Angeles, Mr. Yuji Yamamoto of Minakea Co., Ltd., and Mr. Masahiro Goto of TXP Medical Co., Ltd. It was shown that it is possible to predict patients who will need high medical expenses in the future by using a machine learning model using.
Annual medical expenses in Japan in 2019 reached 43.6 trillion yen, a record high for the third consecutive year.How to stop the increase in medical expenses is a common problem not only in Japan, where the population is aging, but also in developed countries.However, few developed countries have succeeded in curbing medical costs, and much research is still being done.Although it has been suggested that preventive medical intervention is necessary to effectively control medical costs, especially for groups in which medical costs are expected to be high in the future, these groups are accurately predicted. It wasn't easy to do.
Therefore, the research team built a machine learning prediction model that identifies groups whose medical expenses are expected to be high in the future from medical institution consultation data held by healthcare venture Minakea Co., Ltd., and verifies its accuracy. went.Machine learning is a type of artificial intelligence, and it is known that it is good at making more accurate predictions by learning a lot of data compared with conventional prediction models (logistic regression model, etc.).In this study, we used typical machine learning models such as random forests and neural networks.
Research has shown that this machine learning model can be used to predict the risk of becoming the top 5% of high-cost medical patients in the future with high accuracy (AUC value: 0.84), and a prediction model. It was also shown that the machine learning model is more useful than the conventional predictive model (logistic regression model) in constructing.
The machine learning prediction model developed by this study may be able to identify in advance the populations that are expected to have high medical costs in the future, and to intervene early in the populations in need of medical care. Therefore, it may be possible to optimize medical expenses and provide efficient medical care.The machine learning prediction model developed this time can be expected to be used in various situations such as service development to realize more efficient medical care and program design to promote health management.
Paper information:[Npj Digital Medicine] MACHINE-LEARNING-BASED PREDICTION MODELS FOR HIGH-NEED HIGH-COST PATIENTS USING NATIONWIDE CLINICAL AND CLAIMS DATA
