Statistical trial designs and clinical practice: are they compatible?

prognosis (M1c). Of course the statistics are not to blame for this, but performing a second trial in another part of the world may decrease such bias due to differences in clinical practice and therefore inclusion policies. Also, with more effective drugs becoming available in several tumor types, the effect of a single novel drug will be more difficult to evaluate when overall survival is the primary endpoint. In the case of stage IV colorectal cancer, positive results have been shown for irinotecan and oxaliplatin as firstline treatment, while the use of subsequent effective treatments was either unknown or the novel effective drug was not available for subsequent treatment in the control arm. This aspect does not invalidate the efficacy of the drug under investigation in any way, but it does invalidate the common interpretation that the novel drug has outright superiority in first line. The inclusion of more centers from different countries in a different trial may decrease this potential bias. Secondly, the authors state that it is not necessary to include all available patients into one trial, and that there are no statistical reasons against performing two competing trials within a single institution. This may, however, create the selection bias as discussed above in the case of melanoma. Although this should not invalidate the results of either trial, this again In this issue, Edler and Kopp-Schneider try to unmask some myths around statistical principles of randomized clinical studies, and they present their case in a clear and straightforward manner [1]. Comments from a statistical viewpoint are presented elsewhere [2]; in this editorial comments are presented from a clinical viewpoint. Firstly, the dogma that two independent randomized studies showing comparable favorable results are needed is criticized. Although in theory a single study may be sufficient indeed, the danger of a selection bias (which is inherent to all studies since any inclusion implies a selection) may render its results less applicable for the general population. Obviously, in case of known or unknown prognostic variables, this may be overcome by appropriate stratification or the inclusion of sufficient patient numbers, respectively. However, it is not unusual that prognostic criteria as defined within the trial are not considered in daily practice. For instance, when a phase III stage IV melanoma trial with a novel cytotoxic drug is open for all patients with M1a–c disease, but the majority of participating centers exclude M1a–b patients since they have a separate vaccine trial for this category, a negative result will usually be the end of this experimental cytotoxic in melanoma despite the fact that this was only shown for patients with the worst