[Quantifying the additional clinical benefit of new medicines: little - considerable - significant - 6 remarks from a biometrician's point of view].

According to the German Pharmaceutical Market Reorganisation Act [Arzneimittelmarktneuordnungsgesetz (AMNOG)] of 22.12.2010, the benefit assessment of a new drug should include an evaluation of the "degree of additional benefit". A corresponding regulation of the German Ministry of Health states that the quantification of the degree of additional benefit should be made in the terms "major additional benefit", "considerable additional benefit" and "little additional benefit". In September 2011 the IQWiG undertook and explained in appendix A of the dossier evaluation of Ticagrelor an "operationalisation of the extent of additional benefit according to AM-NutzenV". Therein a distinction was made between the target categories "survival time (mortality)", "serious (or, respectively, severe) symptoms", "quality of life", and "not serious (or, respectively, not severe) symptoms". In the operationalisation of the IQWiG, the categorisation of the additional benefit with regard to mortality was addressed by definition of threshold values for the upper limit of the 95% confidence interval for the relative risk (RR). The statutory regulations and the operationalisation of the IQWiG will have direct long-term effects on the provision of medical care since they have a say as to which drugs are to be available at which prices. By introduction of terms such as "major additional benefit", "considerable additional benefit" or "desired effects" and linking them to statistical parameters and algorithms, they also open a series of further fundamental questions as to if and how we should handle these terms in the future and what consequences are inherent to the use of statistical criteria in their "definition". In the present article 6 of the questions that arise in this context are discussed: Can a "considerable additional benefit" be defined with statistical methods? Can a classification of the additional benefit on the basis of an estimated RR be reliable? What are the fundamental possibilities of a "statistical" operationalisation? Is there a unique relative risk for the endpoint mortality? Is a relative risk a reasonable measure for the additional benefit? Should studies be planned in future in such a way that a "true" considerable effect can be demonstrated? My presentation is intended to make it clear that even the use of mathematical/statistical methods for the quantification or classification of an additional benefit cannot provide a perfect solution to the underlying problem of decisions made under uncertainty. Furthermore, they point to fundamental questions arising in the process of quantification. For example, should absolute or relative changes in mortality be decisive for price fixing and what time horizon should be chosen to evaluate the mortality of a treatment? With regard to the planning of new studies the question arises as to whether or not they should (or may) be powered towards the detection of any benefit at all or of a considerable additional benefit.