Abstract
A total of 961 patients who had received resective surgery for gastric carcinoma were grouped according to prognosis by classification and regression trees (CART). This grouping was compared to the present UICC stage grouping. For patients resected for cure (R0) the CART approach allows a better discrimination of patients with poor prognosis (5-year survival rates 15%-30%) from patients with a 5-year survival of 50%, on the one hand, and from patients with extremely poor prognosis (5-year survival rates below 5%) on the other. In the present investigation CART grouping was not influenced by the differentiation between pT1 and pT2 or between pT3 and pT4.