抄録
Showing the non-inferiority of a new drug to an existing treatment is one of the important ways of demonstrating efficacy of a new drug. However, it has been recognized that statistically rejecting a null hypothesis of inferiority does not directly imply true non-inferiority. There are several caveats in concluding non-inferiority based on the test of the inferiority hypothesis: the type I error rate may be infiated by biases which arise from inappropriate design and/or conduct of a non-inferiority trial, the active control may not be as effective as expected, or the non-inferiority margin may be too large to exclude ineffective treatment. We discuss 1) source of biases and means of eliminating them, 2) conditions of the non-inferiority margin, and 3) requirements for the design of a non-inferiority trial. We argue that the credibility of the conclusion from the trial results depends on not only the trial design but also on how the trial is conducted, protocol violations handled and results analyzed. We also present some means to ensure credibility. Finally, check lists for evaluating quality and credibility of the non-inferiority trial results are proposed.
