Development and Internal Validation of a Competing‐Risk Nomogram for Predicting Cancer‐Specific Death in Oral Cavity Squamous Cell Carcinoma
Kuan‐Yu ChuABSTRACT
Objective
Oral cavity squamous cell carcinoma (OCSCC) patients face competing risks of cancer and non‐cancer death. This study developed and internally validated a competing‐risk nomogram for cancer‐specific death.
Methods
A total of 84,456 SEER 17‐registry adults (age ≥ 20 years) with OCSCC (2000–2023) were identified. A Fine–Gray model was fit on a 70% training split, with backward elimination of nine candidate predictors (gender retained a priori). The nomogram, predicting 3‐ and 5‐year cancer‐specific death, was evaluated on the 30% test set against a stage‐only benchmark, with Cox sensitivity analyses.
Results
The final model retained age, race, subsite, grade, stage, surgery, radiation and gender. Distant (vs. localized) stage (sHR 3.48, 95% CI: 3.30–3.66) and surgery (sHR 0.48, 95% CI: 0.46–0.50) were the strongest predictors. Time‐dependent AUC was 0.777 (3‐year) and 0.765 (5‐year), 5–6 points above the stage‐only model ( p < 0.001), with good calibration and Brier scores better than a null model ( p < 0.001). Proportional‐hazards assumptions were violated ( p < 0.001), so estimates are time‐averaged; radiation's estimated direction differed between Fine–Gray and Cox models, likely reflecting confounding by indication.
Conclusion
This competing‐risk nomogram provides individualized, reasonably well‐calibrated 3‐ and 5‐year cancer‐specific death risk estimates for OCSCC. External validation is needed before clinical use.