A critical appraisal of different survival techniques in oral cancer patients.

The Cox model is the preferred survival analysis technique. We compare parametric techniques with the Cox model. 709 consecutive patients treated at a single institution. Univariate survival analysis was performed using the Cox model and parametric models. Significant factors were used to perform the multivariate analysis. The Cox model identified T stage, N stage, tumor thickness, and lymphatic embolization as significant in multivariate analysis. Non-proportional hazards were demonstrated for post-operative radiotherapy and vascular invasion. In the exponential model, T stage, N stage, post-operative radiotherapy, and tumor thickness were significant. The Weibull model identified T stage, N stage, ASA score, post-operative radiotherapy, and vascular invasion as significant. Both lognormal and generalized gamma models identified T stage, N stage, post-operative radiotherapy, tumor thickness, and vascular invasion as significant. Martingale and Cox-Snell residuals were tested. Internal validation confirmed the failure of the Cox model to correctly identify all significant covariates. In conclusion, parametric models may perform better than Cox model at identifying prognostic factors in certain circumstances and should be tested.