| Abstract: | We discuss in this paper the thesis that the underlying questions in breast cancer clinical trials are complex in nature and that multivariate procedures are required for statistical analysis. The variety of statistics resulting from multiple significance testing may be insufficient and misleading. Analysis using multivariate methods centres around interactions between treatments and prognostic factors, time dependence of treatment effects, multi-state models of the disease process, and problems of quality of life. Interpretation of results is not an easy task, as can be shown by some recent examples in the field. In the Federal Republic of Germany full informed consent will have to be taken into account. Emphatically we reject the often held view that legal problems can be bypassed by using historical controls instead of controlled clinical trials. As an alternative we favour the concept of a comprehensive cohort study with a randomised trial being just a part of it. |
