The use of prognostic indexes, based on multivariate analyses, has been described in thyroid cancer (Byar et al., 1979), acute lymphoblastic leukaemia (Palmer et al., 1980) and breast cancer (Sigurdsson et al., 1990). The analysis of prognostic factors accompanying the development of prognostic indices in gastrointestinal tumours is important, as it would improve our knowledge of these diseases and might therefore have consequences regarding the choice of therapy. The overall 5-year survival rate for patients with gastric cancer is poor, usually less than 20% (Wanebo et al., 1993). The recommended surgical treatment is total gastrectomy with more or less extensive lymph-node dissection (D1-D3). However, carcinomas restricted to the mucosa (early gastric cancer) have been resected locally by endoscopy with good results (Tada et al., 1993). Recently, Ichikura et al. (1995) proposed a discriminant score based on depth and horizontal spread of cancer in the submucosa to select those patients who might be treated by endoscopic rnucosal resection, as only 15 to 25% of these patients have lymph-node involvement (Iriyama et al., 1989;Lawrence and Shiu, 1991;Maehara et al., 1992;Moreaux and Bougaran, 1993;Sano et al., 1992). It would be of great importance to ascertain the differences in prognosis of patients within each stage (UICC, 1987) in order to select the optimal treatment modalities for the patients.The value of the immunohistochemical expression of p53, Sialyl-Tn antigen (STn) and Ki-67 antigen and of flowcytometric data together with other, more traditional variables in predicting outcome of gastric cancer patients has been described earlier (Victorzon et al., 1996a,b,c). We have shown that Sialyl Tn antigen (STn) expression and aneuploidy are significant independent prognostic factors in addition to stage (Victorzon el al., 1996b,c). In this study we further investigated the combination of these variables in order to define a risk score that could divide the patients more precisely into groups with different prognoses. days of operation were excluded. Due to background debris or insufficient number of intact cells, the DNA index could not be calculated in 16 tumours, and the S-phase fraction could not be calculated in 43 tumours. Thus 196 patients were suitable for multivariate analysis.
Survival analysisTwelve variables, Ki-67, p53, STn, ploidy, S-phase fraction, age, gender, tumour location, stage, LaurCn (1965) type, lymph-node metastases, and distant metastases were studied individually by univariate analysis and the effect of each variable on prognosis was analysed individually by calculating Kaplan-Meier survival-duration curves. The differences in survival were calculated using the log-rank test (Peto et al., 1977). When the variable analysed consisted of 3 or more ordered categories, log-rank for trend was used.A proportional hazards model (Cox, 1972) was used for the multivariate analysis of all variables which were significant in the log-rank test. Variables were selected in a stepwise fashion with...