Development and validation of a prediction model for lymph node metastasis in oral squamous cell carcinoma using serum biomarkers.

OBJECTIVE

To investigate the predictive value of serum soluble T-cell immunoglobulin and mucin domain-3 (sTIM-3), transforming growth factor-beta 1 (TGF-β1), and vasohibin-1 in the lymphatic metastasis of oral squamous cell carcinoma (OSCC).

METHODS

A total of 220 OSCC patients admitted to Shanxi Provincial Cancer Hospital between January 2022 and December 2024 were included in this retrospective study. The patients were divided into training and validation sets at a 7:3 ratio (154 and 66 patients, respectively). Baseline characteristics, blood test results, and tumor marker levels were compared between the two groups. Predictors were screened, and column-line graphical models were constructed using Least Absolute and Residual Selection Operator (LASSO) regression and multifactorial logistic regression. The performance of the model was then evaluated using ROC curves, calibration curves, and decision curve analysis.

RESULTS

LASSO regression identified the following variables as predictors: clinical stage, tumor diameter, squamous cell carcinoma antigen (SCC-Ag), and carcinoembryonic antigen, sTIM-3, TGF-β1, and vasohibin-1. Multifactorial logistic regression analysis revealed that clinical stage, SCC-Ag, sTIM-3, TGF-β1, and vasohibin-1 were independent predictors of lymphatic metastasis. The AUC of the nomogram model was 0.868 in the training set and 0.863 in the validation set, indicating strong discriminatory ability. Calibration curves showed good agreement between predicted and actual values, with p-values for goodness of fit of 0.865 (training set) and 0.872 (validation set). Decision curve analysis demonstrated significant clinical benefit, with maximum benefit rates of 39.41% in the training set and 37.80% in the validation set.

CONCLUSION

sTIM-3, TGF-β1, and vasohibin-1, along with clinical stage and SCC-Ag, are independent predictors of lymph node metastasis in OSCC patients. The risk prediction model based on these variables demonstrates strong predictive ability.