Statistical Considerations for Non-inferiority Trial Methodology in Regulatory Applications
H. J. Hung, S. J. Wang, R. O'Neill, C. Anello, Y. C. Wang, K. Mahjoob, E. Nevius, A. Chakravarty, CDER, FDA, Rockville, MD
Statistical methodology for design and analysis of non-inferiority trials without a placebo arm has been controversial in evaluation of regulatory applications, though a variety of methods are proposed in literature. Statistical considerations rest heavily on two unverifiable assumptions: 1) constancy assumption that the treatment effect of the selected positive control does not attenuate in the non-inferiority trial as compared to the historical trials, 2) assay sensitivity assumption that the non-inferiority trial can detect a treatment difference if it exists. In this poster presentation, the topics to be discussed are the pros and cons of some frequently proposed statistical methods, non-inferiority margin selection, and scenarios where these methods may have utility from a regulatory perspective. Also to be discussed are the contrast between the non-inferiority study designs with a placebo arm and without a placebo arm