Abstract
BACKGROUND: Metabolic syndrome (MetS) is characterized by chronic inflammation and can be worsened by circadian disruption, which is common among shift work. Machine learning can predict the risk of MetS in shift workers using inflammatory biomarkers. Most investigations have focused on the general population rather than shift workers, a distinct group that requires continuous health monitoring; therefore, we aimed to examine the relationship between inflammatory indicators and MetS using blood cell counts in this high-risk group of shift workers who require long-term health monitoring and to enhance the biological understanding of MetS by applying machine learning methods. METHODS: In this cross-sectional study, we analyzed data from shift workers included in the National Health and Nutrition Examination Survey between 2005-2010 and 2017-2018. Prediction models, including random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and linear regression (LR), were developed and validated. We compared the model performance and conducted stratified analysis, smooth curve fitting, and threshold effect analysis to further explore the relationship between inflammation and MetS risk in shift workers. RESULTS: The analysis included 3,079 participants in total. Each machine learning model demonstrated good predictive performance in assessing MetS risk among shift workers. LightGBM achieved the area under the curve (AUC) of 0.944 in training dataset and 0.722 in testing dataset; XGBoost had an AUC of 0.818 in training dataset and 0.747 in testing dataset; and LR had an AUC of 0.763 in training dataset and 0.699 in testing dataset, RF had an AUC of 0.741 in training dataset and 0.729 in testing dataset. Furthermore, the analysis revealed that body mass index, age, neutrophil, lymphocyte, monocyte, and platelet counts, along with their derived inflammatory indices, were significant predictors. Multivariate logistic regression adjusted for lifestyle and health factors showed that lymphocytes remained consistently associated with MetS in shift workers. Generalized additive model analysis revealed complex non-linear relationships between lymphocytes and platelets. Inflammatory factors strongly predicted MetS risk in shift workers, with their effects varying by concentration threshold, particularly for lymphocytes (k = 2.2, right side p < 0.001). CONCLUSION: Lymphocyte counts and related composite indices are significant predictors of MetS risk in shift workers. Consistent monitoring of these biomarkers may be useful for early odds-based stratification of MetS in this high-risk population, whereas any preventive implications would require confirmation in longitudinal and interventional studies.