Prediction of the Onset of Lifestyle-related Diseases Using Regular Health Checkup Data

Abstract

This study proposes a method for predicting the onset of lifestyle-related diseases using periodical health checkup data. We carefully examined insurance claims data to identify the onsets of the diseases and used them as correct answers for supervised learning. We adopted the undersampling and bagging approach to address the class imbalance problem. We aimed to predict whether lifestyle-related diseases, other than cancer, will develop within one year. The precision and recall of the proposed method were 0.32 and 0.89, respectively. Compared with a baseline that sets thresholds for each examination item and considers their logical sum, it was found that much higher precision could be obtained while maintaining recall, which is meaningful as it allows for the suppression of the number of targets for health guidance, without increasing the negligence of those that are likely to become severely ill.