Identifying Four Developmental Trajectories of Metabolic Syndrome and Their Influencing Factors: A Longitudinal Cohort Study of Railway Employees' Physical Examinations.

BACKGROUND

Metabolic syndrome (MetS) is one of the most common chronic disease complications and significantly increases the prevalence of chronic diseases. This study aims to identify different patterns of MetS development using longitudinal data and explore their influencing factors.

METHOD: Based on the physical examination cohort of Shanghai railway workers, longitudinal data spanning 5 years (from January 1, 2019, to December 31, 2023) were collected to analyze the development trajectories of 1954 participants with MetS. Latent growth mixture model (LGMM) was employed to classify the development trajectories of MetS into distinct groups. Additionally, mixed-effect models were utilized to explore the influencing factors, and machine learning models were constructed for trajectory prediction.

RESULTS

The LGMM model classified patients into four groups: the progressively increasing group, the steadily increasing group, the progressively decreasing group, and the steadily decreasing group. Compared to the other three groups, the progressively increasing group exhibited the highest levels of weight, body mass index (BMI), heart rate, γ-glutamyltransferase, aspartate aminotransferase, alanine aminotransferase, uric acid, and white blood cell count. Conversely, compared to the other three groups, the group with progressive decreases showed the highest levels of systolic blood pressure, total bilirubin, direct bilirubin, urea nitrogen, fasting blood glucose, high-density lipoprotein cholesterol (HDL-C), and triglycerides (TGs). Mixed-effect models revealed that an increase in BMI and TG (OR > 1, p < 0.001) significantly increased the probability of being classified into the progressively increasing group, whereas HDL-C (OR < 1, p < 0.001) had the opposite effect. Variables selected through feature engineering were utilized to construct five machine learning prediction models, among which Random Forest (with an area under the curve, AUC = 0.852) demonstrated the best overall prediction performance and was therefore chosen to develop a MetS risk calculator based on Shiny.

CONCLUSION

BMI, TG, and HDL-C were the key to influence the developmental trajectories of MetS. Therefore, these three indicators should be closely monitored, and the progression of MetS can be controlled by adjusting dietary patterns.