The course of cirrhosis after the first episode of variceal bleeding is characterized by recurrent episodes of bleeding and bleeding-free periods, with each new stage being associated with an initially high mortality rate. Analysis of the course of this disease is usually based on traditional survival models measuring observation time from admission to hospital. This approach cannot provide valid answers to the following clinically relevant questions: 1) what is the risk of death while bleeding? 2) what is the risk of continuing bleeding? 3) what is the risk of death after bleeding has stopped? 4) what is the risk of a new bleeding episode among those who remain alive and do not bleed? and 5) do the answers to the former four questions depend on the number of previous bleeding episodes or on the time since the first episode? These questions may be addressed using the multistage, competing risks model presented here. By measuring time from the start of each new bleeding or bleeding-free period, this model synchronizes the patients with respect to disease development and, hence, to level of the rate of new clinical events. This reduces the heterogeneity of risk levels in the patient group at any given time. In an analysis of 94 patients with cirrhosis admitted for their first episode of variceal bleeding, we estimated the mortality rates during bleeding and without current bleeding, the rate of cessation of bleeding and the rate of rebleeding. The influence of number of previous rebleedings and of the time since the first bleeding was examined using a Cox regression model for competing risks.(ABSTRACT TRUNCATED AT 250 WORDS)