Table 1.
Description of biomarkers and clinical features used in the study.
Table 2.
Biomarker panels used in analysis.
Table 3.
Hyperparameter grids used for optimization.
Table 4.
Baseline characteristics of the study population (n = 545).
Table 5.
Performance summary of the best model for each biomarker panel (VIF < 5).
Fig 1.
Final validation of the inflammatory panel model in the holdout test set.
(A) ROC curve (AUC = 0.711). (B) Calibration curve demonstrating acceptable agreement between predicted and observed probabilities.
Table 6.
Updated classification metrics of the final model on the holdout test set.
Fig 2.
Confusion matrix of the final LightGBM model on the holdout test set.
Fig 3.
SHAP analysis of the final LightGBM model.
(A) Global feature importance, ranking predictors by their mean absolute impact. (B) Beeswarm plot showing the impact of each predictor’s value on the model output for every individual. The analysis highlights IGF-1, IL-10, and CRP as top influential biomarkers.