Professor Kazuhiko Oe of the University of Tokyo and others, in collaboration with Nippon Telegraph and Telephone Corporation (NTT), use electronic medical record data of about 900 diabetic patients as one of the NTT Group's AI technology "corevo" to treat diabetic patients. Announced that it has built a model that predicts patient behavior "interruption of medical records," which is one of the causes of worsening symptoms.
The number of diabetic patients has been increasing in recent years, and the number of diabetic patients reached 2014 million in 316 (26 patient survey, Ministry of Health, Labor and Welfare).As diabetes progresses, it causes complications and leads to a decrease in QOL and an increase in medical expenses, so continuous treatment is required.However, there are many cases in which about 1% of outpatients with diabetes discontinue consultation and resume consultation after the onset of complications and the condition worsens.
Therefore, the University of Tokyo and NTT aimed to build an AI that predicts the interruption of consultation based on individual electronic medical record data.A model was constructed based on the features generated by referring to the medical data analysis of the University of Tokyo and the knowledge of clinical guidance to patients, and the knowledge of machine learning in NTT's "corevo".
By inputting the electronic medical record data and the feature amount generated from it, the appointment failure (non-examination of the appointment outpatient that may cause the consultation to be interrupted) and the consultation interruption risk ranking (the length of the number of days until the future consultation interruption date) of the patient (Ranking) is predicted.When the model was evaluated using electronic medical record data of about 2 patients who visited the University of Tokyo Hospital for the treatment of diabetes from 2011 to 2014, it had excellent predictive performance of predicting 900% of consultation interruptions. It was confirmed.
It was also found that items related to the patient's reservation behavior, such as the reservation registration date, the day of the week of the reservation date, and the interval between the reservation registration date and the reservation date, which the doctor did not notice until now, influence the prediction.Based on the prediction results, it is possible to narrow down the patients who should be actively supported, determine the start time, and adjust the degree of support. It is expected that it will lead to.
