Abstract
This study aimed to analyze S100A9 and CCL5 levels in patients with nasopharyngeal carcinoma (NPC) and evaluate their predictive value as blood-based indicators for NPC diagnosis. Serum S100A9 and CCL5 levels were measured in 123 patients newly diagnosed with NPC and 107 patients without NPC. Additionally, 38 patients (19 with NPC and 19 without) were recruited from Xiangya Hospital as an external validation cohort. Logistic regression was used to identify risk factors for NPC. Variable selection was conducted using least absolute shrinkage and selection operator (LASSO) regression. NPC prediction models were developed using four machine-learning algorithms, and their performance was evaluated with ROC curves. Calibration curves, decision curve analysis (DCA), and Shapley additive explanation plots were employed for further evaluation and interpretation. Serum S100A9 and CCL5 levels were significantly elevated in patients with NPC compared with patients without NPC. Multivariate logistic regression identified S100A9, CCL5, TP, and ALB as independent predictors of NPC. ROC analysis demonstrated that S100A9 had superior diagnostic performance compared to CCL5 and other blood indicators, effectively differentiating NPC from non-NPC cases. A machine-learning-based logistic regression model incorporating S100A9, CCL5, ALB, GLB, and PLR demonstrated a reliable diagnostic value for NPC, achieving an Area under the curve (AUC) of 0.877 in the training cohort. The calibration curve showed excellent agreement between predicted and actual probabilities; in contrast, the DCA curve highlighted strong clinical utility. The model also performed well in the external validation cohort, with an AUC of 0.817. Serum levels of S100A9, CCL5, and other indicators such as GLB, ALB, and PLR have diagnostic values for NPC. The logistic regression model based on these biomarkers demonstrated robust predictive performance and clinical utility for NPC diagnosis.