[Problems which arise when testing the equivalence of 2 survival distributions]

Clinicians' interest in equivalence trials has been increasing for many years. In cancer research we are more often concerned with survival or event free survival (EFS). The usual logrank test has to be modified to test whether the true value r of the relative risk is at least equal to a limit value rL. An equivalence trial of post-operative therapy was carried out in children with stage II nephroblastomas without lymph node involvement, and serves as an illustration of some problems, discussed here, which arise in such design: ethical questions when defining the equivalence region, the choice of a one sided or a two sided formulation, the required number of patients or the power of the comparison. A formula is given in order to allow calculation for the required number of patients to ensure a given power when the treatment effects are the same. Other problems are also encountered when analyzing the observed results as the number of events which is evidently low in a de-escalation trial, may be reduced due to a stopping rule. We also discuss about the possibility to use event free survival as a surrogate of overall survival, and about the difficulties encountered when interpreting the results in terms of significance level.