Prediction of health check-up test values using machine learning based on special health check-up data

Abstract

Public health nurses and dietitians provide health guidance according to the level of health guidance based on the results of examination values and questionnaire items in the specified health checkup, which started in 2008 for people aged 40 and over. However, the content of the guidance for lifestyle improvement differs depending on the instructor's experience and judgment, and the evidence for lifestyle improvement is weak because the guidance is based only on the health checkup result report. In order to solve this problem, we are developing a system that predicts future test values based on past test results and presents possible lifestyle-related diseases. In this study, we constructed a model for predicting test values using a multichannel deep convolutional neural network, which is a type of machine learning, with the aim of predicting test values for the next year based on past health examination results. As a result of constructing a model to predict the abdominal circumference, the error to the true value was 2.50%. The features that the model used as the basis for its prediction were analyzed by sensitivity analysis to evaluate the validity of the prediction. As a result, the gradient of the items considered to be related to the abdominal circumference was large, suggesting that the model was learning appropriate features.