Establishment and validation of a recurrence risk model in early-stage tongue squamous cell carcinoma patients incorporating immune-inflammatory biomarkers and clinicopathological parameters.

OBJECTIVE

To develop and validate a machine learning-based predictive model incorporating immuno-inflammatory biomarkers and clinicopathological parameters to predict recurrence risk in early-stage tongue squamous cell carcinoma (TSCC) patients.

METHODS

This retrospective study included 515 early-stage TSCC patients treatment at Xinyu People's Hospital between May 2014 and May 2019. Medical records and laboratory data were reviewed. Patients were randomly divided into a training cohort (n=339) and a validation cohort (n=176). Feature selection was performed using LASSO, Xgboost, and Support Vector Machine (SVM) algorithms to identify key features associated with recurrence. A predictive nomogram was then built based on multivariate Cox regression analysis. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, calibration plots, and decision curve analysis (DCA).

RESULTS

Recurrence was observed in 160 cases (31.07%), with 111 (32.74%) in training cohort (n=339) and 49 (27.84%) in the validation cohort (n=176). Machine learning algorithms identified several key risk factors for recurrence, including immuno-inflammatory markers (e.g., white blood cell count [WBC], platelet count [PLT], C-reactive protein [CRP], neutrophil-to-lymphocyte ratio [NLR], systemic inflammation response index [SIRI], C-reactive protein-to-albumin ratio [CAR]) and clinicopathological characteristics (e.g., pathological classification, chemotherapy status, tumor location). The nomogram achieved areas under the ROC curve (AUCs) of 0.902 (95% CI: 0.866-0.937) in the training set and 0.819 (95% CI: 0.759-0.876) in the validation set. Calibration curves demonstrated good predictive consistency (P=0.621). DCA showed a clear net clinical benefit across a wide range of thresholds probabilities (P<0.001).

CONCLUSION

This predictive model, integrating immuno-inflammatory markers and clinicopathological features, exhibits excellent predictive performance for recurrence risk in early-stage STCC and offers substantial clinical utility.