Abstract: Objective To develop and validate a model that can predict the prognosis of patients with locally-advanced oral squamous cell carcinoma (OSCC) after radical resection. Methods A total of 243 patients with primary locally-advanced OSCC who underwent radical resection in our hospital from February 2009 to January 2016 were retrospectively analyzed. Patients were randomly assigned to the training set (n=182) and the validation set (n=61). Cox regression was used to determine the independent factors that affect the overall survival (OS) in the training set, and to develop predictive 3-year and 5-year nomogram models and random survival forest (RSF) models. The performance of the predictive model was evaluated by receiver operating characteristic (ROC) curve, calibration curve and decision curve analysis (DCA), and compared with the AJCC staging system. Results Cox regression showed that age, KFI index, pT stage, pN stage, AJCC stage, the number of positive lymph nodes, and SII were independent predictors of OS (all P<0.05). In the validation set, the AUC of RSF model, nomogram model, and AJCC staging system for predicting 3-year OS rate were 0.782, 0.756 and 0.703, respectively, for predicting 5-year OS rate were 0.780, 0.731 and 0.696, respectively. The calibration curve showed that the prediction models had a good consistency; the DCA curve showed that the clinical value of the nomogram model was higher than that of the AJCC staging system, while the clinical value of the RSF model was higher than that of the nomogram model. The log-rank test showed that the RSF model had good risk stratification ability (P<0.05). Conclusions The RFS model constructed based on age, KFI index, pT staging, pN staging, AJCC staging, number of positive lymph nodes and SII, can individually predict the prognosis of patients with locally-advanced OSCC after curative surgery.

Key words: Oral squamous cell carcinoma, Radical resection, Nomogram, Random survival forest