Association Between Serum Micronutrients and Advanced CKM Syndrome: An Interpretable Machine Learning‐Based Study

Figures

• 

  FIGURE 1. RCS analysis for association between serum micronutrients and advanced CKM syndrome. Multivariable adjusted restricted cubic splines curve for associations of advanced CKM syndrome participants and (A) β‐carotene, (B) α‐tocopherol, (C) lycopene, (D) folate, (E) vitamin C, and (F) 25(OH)D. The solid and shaded areas represent the estimated values and their corresponding 95% CI.

• 

  FIGURE 2. The association between serum micronutrients and advanced CKM syndrome. (A) The WQS model weights of serum micronutrients on the prevalence of advanced CKM syndrome in a negative direction. (B) Overall effect of serum micronutrient mixtures on the prevalence of advanced CKM syndrome in the BKMR model. (C) The Qgcomp model weights of serum micronutrients on the risk of advanced CKM syndrome.

• 

  FIGURE 3. Performance comparison of machine learning models for advanced CKM syndrome prediction. (A) ROC curves for RF, XGB, CB, LGBM, DT, KNN, and NB in the training set, with AUC values shown in the legend. The dashed line represents the chance level (AUC = 0.5). (B) ROC curves for the same models in the testing set, with AUC values shown in the legend. The dashed line represents the chance level (AUC = 0.5). AUC, area under the curve; CB, CatBoost; DT, decision tree; KNN, k‐nearest neighbors; LGBM, LightGBM; NB, naïve Bayes; RF, random forest; XGB, XGBoost.

• 

  FIGURE 4. SHAP analysis of serum nutrient contributions to the risk of advanced CKM syndrome. (A) SHAP waterfall plot for the contribution of individual sample features to advanced CKM syndrome. The purple bar indicates the contribution of the variable to the development of advanced CKM syndrome in an individual, while the yellow bar represents its inhibitory effect. The f(X) values are probabilistic predictions, and E[f(X)] (the base value) represents the predictions made without any input to the model. f(X) corresponds to the log odds ratio for each observation. (B) The SHAP bar plot and bee swarm plot. Mean SHAP value ranking of the features in LGBM model; higher mean SHAP values correspond to a higher risk of advanced CKM syndrome. Each point represents a SHAP value for a feature and an instance, with the color indicating the feature value (purple = high, yellow = low). Features are ordered by importance, and the position along the X‐axis reflects the contribution to LGBM model's output.

• 

  FIGURE 5. The independent and synergistic effects of serum micronutrients levels on the predicted risk of advanced CKM syndrome. (A) The PDPs illustrate the relationships between advanced CKM syndrome and 25(OH)D, vitamin C, and lycopene. The x‐axis represents the log‐transformed values of each serum micronutrients, while the y‐axis shows the change in the predicted advanced CKM syndrome risk. Shaded regions indicate confidence intervals, and the blue histograms below each plot visualize the distribution of data points across feature values. (B) Two‐dimensional partial dependence plots (2D PDPs) illustrating the interaction effects of serum micronutrients on the predicted risk of advanced CKM syndrome. Contour plots depict the combined influence of 25(OH)D and lycopene, 25(OH)D and vitamin C, and lycopene and vitamin C on advanced CKM syndrome risk. The x‐axis and y‐axis represent the log‐transformed values of the two serum micronutrients, with each grid corresponding to a 10% range of their levels. The contour plot shows the impact of these micronutrients on the predicted advanced CKM syndrome outcome.