Metabolic Syndrome Prediction: Big Data–Driven Machine Learning in Japan

Metabolic syndrome (MetS) is a major risk factor for cardiovascular disease, and its prevalence continues to increase worldwide. In Japan, one of the most rapidly aging societies, approximately one in four individuals aged 40–74 years has MetS or is at high risk, making prevention an urgent public health priority. Since 2008, the Specific Health Checkups (SHC) program, implemented by the Ministry of Health, Labour and Welfare, has generated large-scale, standardized longitudinal data. However, current strategies mainly focus on high-risk individuals, and identification of pre-onset risk remains insufficient. To enable earlier intervention, we developed machine learning models to predict 5-year MetS incidence. Given known sex differences in MetS, combined models may not fully capture sex-specific profiles. This study aimed to improve predictive performance and identify sex-specific predictors using sex-stratified models. SHC data collected in Japan were analyzed. The outcome was incident MetS defined by National Health and Nutrition Survey criteria. Model performance was evaluated using recall, precision, and ROC-AUC. Analyses were conducted using LightGBM in Python 3. Among 119,012 participants, 10,633 developed MetS within 5 years. In 5-fold cross-validation, AUC, sensitivity, and specificity were 0.8222 ± 0.003, 0.8248 ± 0.021, and 0.6786 ± 0.030 in men, and 0.8864 ± 0.004, 0.8370 ± 0.034, and 0.7844 ± 0.031 in women, indicating higher performance in women. Key predictors differed by sex: waist circumference, HbA1c, and HDL cholesterol contributed most in men, whereas waist circumference, BMI, and lipid-lowering medication use were most important in women. In this research, a machine learning model predicting MetS incidence using SHC data was developed. High performance was observed in both sexes despite single-year data, and differences in contributing factors were identified. Integration of this model into the SHC framework could enhance preventive strategies.

Key messages

• Sex-stratified machine learning models accurately predicted 5-year metabolic syndrome incidence using routine health checkup data and revealed distinct sex-specific risk profiles.

• Current screening focused on high-risk individuals may miss those at risk before onset; integrating sex-specific models into national health checkup programs may support earlier, targeted prevention.