Determining When One Treatment is Di erent From Another Based On A Censored Data Regression Model

Often when comparing the survival rates of individuals given either of two treatments the analysis stops with a test of the hypothesis of no treatment di erence and perhaps a plot of the two survival functions. The hypothesis test is usually a comparison of the two survival curves over the entire observational period. An alternative approach to this problem is to provide an investigator with a con dence set for the set of times at which the survival rates of the two treatments are the same. We discuss how such con dence sets can be constructed when the proportional hazards or additive regression model is used to adjust the comparison of interest for other factors which may in uence survival. These approaches are illustrated on retrospective data gathered to compare the survival rates of allogeneic and autologous bone marrow transplants for acute leukemia.