Associations of the correlates of protection and implication on the statistical power for demonstrating non-inferiority: Application of a re-sampling method on a large phase III influenza vaccine clinical trial

BackgroundIn the later stage of the clinical development of new vaccines it is required to demonstrate their efficacy with the immunogenicity measures established as correlates for disease protection.Evaluation of interpandemic trivalent influenza vaccines is commonly assessed by three immunogenicity measures for each strain in different age groups: Seroprotection, Seroconversion and Geometric Mean Titers. US and European guidelines with respect to this topic have been issued for the licensure of new influenza vaccines. The statistical power of comparative trials, which consider these endpoint variables, could be affected to the extent that these measures are correlated. Results from a large non-inferiority trial in the clinical development of a novel cell-derived influenza vaccine have been analyzed with the aim of evaluating how statistical dependency between the above-mentioned three immunogenicity measures might affect the power to demonstrate non-inferiority.MethodsThe statistical non-inferiority criteria, which were met in the trial, were applied to different subsets (n = 250, n = 370 and n = 500) using a re-sampling method from the original dataset (re-samples = 10,000).ResultsThe measures of immunogenicity were highly correlated, and the fulfillment or failure of any of the non-inferiority criteria for a specific measure partially predicted the same outcome for the other measures. Due to this dependency within each strain, the levels of power obtained by re-sampling methods were always higher than those obtained by theoretical calculations, which were based on the assumptions of independency between the three measures of immunogenicity. Seroconversion and Geometric Mean Ratio (GMR) showed a higher correlation. A failure in the fulfillment of the non-inferiority criteria for GMR predicted the failure for Seroconversion in >76% of cases.ConclusionsThe correlation between different measures of immunogenicity should be taken into account when evaluating statistical power for non-inferiority in influenza vaccine trials and in establishing sample sizes. Statistical approaches that include either all three measures of immunogenicity or both Seroconversion and the ratio of GMTs as co-primary non-inferiority endpoints might create redundancy and could increase the probability of not meeting at least one non-inferiority criterion by chance, due to multiplicity.