[Multivariate analysis for the prediction of recurrence in colorectal cancer]

We developed a predictive scale for cancer recurrence by conducting a multivariate analysis (Hayashi's discriminant analysis of qualitative data) on clinicopathologic indicators in 207 patients with Dukes' B or C colorectal cancer who underwent curative resection at our institution. Seven indicators were found to have prognostic value. Nodal status, which was divided into five categories according to the level and number of positive nodes, had the highest coefficient as a predictor of recurrence. The correlation ratio between the group showing recurrence and that showing non-recurrence was 0.49, with 74% discrimination success rate. Liver or lung metastases were more accurately discriminated than local or peritoneal failure. We assessed the accuracy of this predictive scale in 110 patients with Dukes' B or C colorectal cancer who had been surgically treated at another institution (National Cancer Center, Tokyo), and determined it to be 68%. We think that this scale using multivariate analysis for the prediction of cancer recurrence has significant clinical application in the selection of more effective postoperative adjuvant chemotherapy. It is concluded that the present predictive scale is clinically useful, but that the inclusion of other indicators such as immunohistological ones is required to increase its predictive precision.