Letters Section Editor: Jody W. Zylke, MD, Senior Editor.
Author Affiliations: Medical Statistics Department, London School of Hygiene and Tropical Medicine, London, England (Drs Pocock and Piaggio) (email@example.com); and Centre for Statistics in Medicine, University of Oxford, Oxford, England (Dr Altman).
In Reply: We appreciate the concerns expressed by Dr Aberegg but disagree with his argument. We do not think that noninferiority trials require symmetry in their interpretation. A noninferiority trial is conducted to see if a new treatment is as good as a reference treatment in the sense that a specific potential magnitude of inferiority (delta) is contradicted by the trial's data.
If the upper limit of a 2-sided 95% confidence interval for the treatment difference is below delta, then the authors can claim evidence of noninferiority. The strength of evidence can be expressed by the P value for noninferiority, which tests for departure from the null hypothesis that the true treatment difference is at least delta. The appropriate choice of delta during trial design is key, as was discussed in our article.
Pocock SJ, Piaggio G, Altman DG, CONSORT Group FT. Reporting Noninferiority Trials—Reply. JAMA. 2013;309(15):1584–1585. doi:10.1001/jama.2013.3095