Development and validation of a 10-year predictive model for cardiovascular and metabolic disease risk: insights from a large-scale health examination cohort

Abstract

Objective To develop and validate a 10-year predictive model for cardiovascular and metabolic disease (CVMD) risk using comprehensive health examination data from nearly 37 701 individuals. Methods This retrospective cohort study used health examination data, including demographic information, clinical measurements, laboratory tests and lifestyle factors. Potential predictors were selected based on a literature review and exploratory analysis. Machine learning techniques (including random forest and gradient boosting) were employed to develop the predictive model. The model’s performance was evaluated using accuracy, sensitivity, specificity and area under the receiver operating characteristic curve (AUC). Model validation was conducted on separate test and validation sets. Results A total of 37 701 electric power employees were included in this study after applying rigorous inclusion and exclusion criteria. The dataset was divided into training, validation and testing sets in a 70:15:15 ratio, with no significant differences observed in baseline characteristics, ensuring robust analysis. Feature selection using the random forest classifier identified the top predictors of CVMD. Machine learning models, particularly random forest and gradient boosting, demonstrated superior predictive performance compared with traditional Cox regression methods. These results significantly outperformed traditional Cox models, which yielded an AUC of approximately 0.60. Correlation analysis revealed strong associations between key variables, such as systolic and diastolic blood pressure, low-density lipoprotein and total cholesterol, and creatinine and blood urea nitrogen, highlighting the complex interactions among CVMD risk factors. Conclusion The developed 10-year predictive model for CVMD risk, based on health examination data, shows promising potential for early identification and targeted intervention in individuals at high risk for CVMDs. This approach could contribute to the reduction of CVMD incidence and related morbidity and mortality.