Analysis of survival of patients with cancer sets particular epidemiological and statistical problems, especially when one wants to take into account metastasis or local recurrences. Cox's model does not allow modeling multiple events. Wei et al. have proposed an extension of Cox's model, by formulating the marginal distributions of multivariate failure times, which allows testing covariates effects on different events. We applied these methods to data from the Registry of Digestive Tumors of Burgundy, France. Prognostic factors of recurrence are rectal location of tumor and advanced stage at diagnosis. Prognostic factors of death are male gender, age greater than 75, rectal location and advanced stages. However, relative risk of recurrence for advanced stages is significantly greater than relative risk of death (p=4.10(-3)), while tumor location has the same influence on the two events.