Preoperative identification of Warthin tumor (WT) can inform surgical planning and patient counseling. This study aimed to develop and externally validate a simple clinical prediction model based on routinely available preoperative variables.

This two-center retrospective study included 230 patients in the development cohort and 245 in the external validation cohort. Nine routinely available preoperative variables were assessed using LASSO-penalized logistic regression with stratified five-fold cross-validation. Model performance was evaluated using discrimination, calibration, overall prediction error, 1000-resample bootstrap internal validation, decision curve analysis, and clinical impact curves.

Among 475 patients, 81 (17.1%) had histopathologically confirmed WT. The final model included age (OR = 1.037, 95% CI: 1.008–1.067), male sex (OR = 3.828, 95% CI: 1.146–12.783), and smoking history (OR = 3.591, 95% CI: 1.550–8.316). The AUC was 0.791 (95% CI: 0.721–0.862) in the development cohort and 0.868 (95% CI: 0.816–0.919) in external validation. After bootstrap correction, the AUC was 0.785, the calibration intercept was −0.048, the calibration slope was 0.942, the Brier score was 0.129, and Emax was 0.088. In the external cohort, the calibration intercept and slope were 0.379 and 1.503, respectively; the Brier score was 0.101 and Emax was 0.139. Decision curve analysis suggested potential net benefit across selected threshold ranges.

A model based on age, sex, and smoking history showed useful preoperative discrimination between WT and other parotid tumors. The web-based dynamic nomogram may support initial assessment and patient counseling, although further prospective and geographically diverse validation is needed.