Statistical considerations of FDA and CPMP rules for the investigation of new anti-bacterial products

In this paper I investigate statistical properties of some guidance given by the FDA and by the CPMP on the planning, conduct and analysis of clinical trials with new anti-bacterial substances using an active control design. It is demonstrated that the non-inferiority margin proposed by the FDA has some undesirable features, and that the CPMP guidance may need further interpretation with respect to a statement that the non-inferiority margin may be smaller than 10 per cent for response rates >90 per cent. A new margin is proposed that combines the desirable properties from both the FDA and the CPMP guidance. It is also shown that the approximate unconditional tests that are in use in such trials are quite unreliable with respect to preserving the nominal type I error. Unconditional exact tests are presented as a remedy for this issue.