[The method of conducting randomized trials in cancerology must be changed]

Most new treatments in oncology lead to modest improvements. To demonstrate these moderate progresses, trials ought to be much larger than what is currently done. The results of the small trials presently published do not represent reliable information but reflect the play of chance. To solve the problem, much larger trials are required. These trials need to be much simpler, for instance by giving up restrictive inclusion criteria. It is often stated that the patient population in a trial must be "homogeneous". This is a mistake : the population of trial patients must be the population of patients for whom the therapeutic question is relevant. Any a priori restriction in the population included in the trial gives up all hopes to evaluate the treatment in practical situations. In the mean time, while we are waiting for the required large trials, we can gather the data from several trials of the same question in a meta-analysis.