Prediction and predictor elucidation of the onset of metabolic syndrome among young workers using machine learning techniques: A nationwide study in Japan

Miyuki Suda,Tadao Ooka,Zentaro Yamagata

DOI: https://doi.org/10.1101/2021.10.21.21265259

2021-10-25

Abstract:Abstract Objectives Predictive models for the onset of metabolic syndrome (MS) among people in their 30s are scarce. This study aimed to construct a highly accurate model to predict MS onset by age 40 years and to identify the important predictors of MS onset using health checkup data of Japanese company employees aged 30 and 35. Methods The study included 6,048 Japanese employees aged 40 years who had undergone periodic health examinations over 10 years. We developed prediction models for MS onset using machine learning methods including the random forest and logistic regression, and evaluated the models using the receiver operating characteristics and precision-recall curves. For the random forest models, the variable importance of each explanatory variable was calculated to identify important predictors of MS onset. Results The random forest had higher predictive power than logistic regression in all models, although the differences were non-significant. Regarding important predictors, diastolic blood pressure was the most important predictor of MS onset for men aged 30 and 35 years, while body mass index was the most important predictor for women aged 30 and 35 years. Conclusions We created a machine learning model to predict MS onset at age 40 with high accuracy from health examination data obtained at ages 30 or 35. Sex differences in important predictors of MS onset was shown by the variable importance indices of the random forest. Applying our model in routine healthcare management should provide early and appropriate health interventions to prevent MS onset in young people.