Construction and validation of a risk prediction model for metabolic syndrome: a cross-sectional study based on randomized sampling

Objective: The prevalence of metabolic syndrome is high among Chinese residents, and it is crucial to understand the current situation and intervene promptly. In this study, we investigated the current status of metabolic syndrome in some regions of China, analyzed related risk factors, and developed a risk prediction model to guide preventive measures. Methods: A multistage stratified cluster random sampling method was used to select 3541 permanent residents aged 18–79 years from a district in Beijing for face-to-face questionnaire surveys, physical examinations, and laboratory tests. All participants were randomly divided into training and validation sets. Correlation analysis and multivariate logistic regression were employed to identify risk factors for metabolic syndrome, and a column-line graph prediction model was developed. The discriminative ability and predictive accuracy of the model were assessed by receiver operating characteristic (ROC) curve and calibration curves. Results: The prevalence of metabolic syndrome in this study was 18.4%. The results of multivariate logistic regression analysis showed that increasing age, being male (OR = 1.827), being overweight (OR = 4.865), being obese (OR = 11.482), hazardous alcohol consumption (OR = 1.673), marital/cohabitation history, and specific occupations (agriculture, forestry, fisheries, and water production, and unemployed) were independent risk factors for metabolic syndrome (P < 0.05). The column-line graph prediction model, constructed accordingly, performed well, and the model indicated that BMI and age were the most significant risk factors for metabolic syndrome. The results of model validation showed that the AUCs of the training and validation sets were 0.815 (95% CI: 0.795–0.836) and 0.787 (95% CI: 0.756–0.818), respectively, indicating that the model performed well in discriminating. The calibration curve had a calibration slope of 1.000, an intercept of 0.000, and a Hosmer-Lemeshow test P-value of greater than 0.05. The MAE (0.240–0.261) and Brier score (0.120–0.131) were within reasonable ranges, suggesting that the predicted probability was highly consistent with the actual risk. Conclusions: The metabolic syndrome column-line diagram risk prediction model constructed in this study, based on multivariate logistic regression analysis, has good discriminative ability and high prediction accuracy. The model shows that a large proportion of the current risk factors for metabolic syndrome are modifiable, and that the risk of metabolic syndrome in high-risk groups, such as the elderly, men, people with marital/cohabitation histories, and people with specific occupations, can be reduced through behavioral and lifestyle interventions. This model can provide a scientific basis for the early identification of high-risk groups for metabolic syndrome, and has an important guiding value for targeted preventive interventions.