Development and external validation of a nomogram for overall survival in oral tongue squamous cell carcinoma using machine learning-assisted feature selection.

This study aimed to identify prognostic factors using machine learning-assisted feature selection and develop a Cox-based nomogram for predicting 2-year and 5-year overall survival in Oral Tongue Squamous Cell Carcinoma (OTSCC). Clinical data from 5746 OTSCC patients in the SEER database were randomly divided into training and internal validation cohorts. An independent external cohort of 174 patients from Bengbu Medical University was retrospectively collected for external validation. Machine learning algorithms (LASSO, XGBoost, C-SVC, and RSF) were used for feature selection, and selected variables were incorporated into a multivariable Cox nomogram. Model performance was evaluated using the concordance index (C-index), time-dependent receiver operating characteristic curves, calibration curves, and decision curve analysis (DCA). Age, sex, tumor grade, and TNM stage were incorporated into the final nomogram. The multivariable model demonstrated improved discriminative performance compared with the TNM-only model across all cohorts. The C-indexes were 0.717 (95% CI 0.707-0.728), 0.720 (95% CI 0.704-0.736), and 0.711 (95% CI 0.634-0.788) in the training, internal validation, and external validation cohorts, respectively. The corresponding 5-year AUCs were 0.757, 0.767, and 0.737. Calibration curves showed good agreement between predicted and observed survival outcomes, while DCA suggested potential net clinical benefit. This externally validated nomogram may serve as a potentially useful tool for pre-operative risk assessment and prognostic stratification in patients with OTSCC.