Validation of two prognostic models for recurrence and survival after radical gastrectomy for gastric cancer.

Abstract

Prognostic models from Korea and Italy have been developed that predict overall survival and cancer recurrence respectively after radical gastrectomy for gastric cancer. The aim of this study was to validate the two models in independent patient cohorts, and to evaluate which factors may explain differences in prognosis between Korean and Italian patients with gastric cancer. Patients who underwent radical gastrectomy for gastric cancer between January 2000 and December 2004 at Seoul National University Hospital and at eight centres in Italy were included. Discrimination of the models was tested with receiver operating characteristic (ROC) curves and calculation of area under the curve (AUC). Calibration was evaluated by plotting actual survival probability against predicted survival probability for the Korean nomogram, and actual against predicted risk of recurrence for the Italian score. Some 2867 and 940 patients from Korea and Italy respectively were included. The Korean nomogram achieved good discrimination in the Italian cohort (AUC 0·80, 95 per cent c.i. 0·77 to 0·83), and the Italian model performed well in the Korean cohort (AUC 0·87, 0·85 to 0·89). The Korean nomogram also achieved good calibration, but this was not seen for the Italian model. Multivariable analyses confirmed that Italian ethnicity was an independent risk factor for cancer recurrence (odds ratio (OR) 1·72, 1·31 to 2·25; P < 0·001), but not for overall survival (OR 1·20, 0·95 to 1·53; P = 0·130). Both prognostic models performed fairly well in independent patient cohorts. Differences in recurrence rates of gastric cancer may be partially explained by ethnicity.