Abstract
In the development of new vaccines, understanding the level of vaccine-induced antibody that is sufficient to protect against disease can simplify and expedite the development and licensing process. If there is an accepted threshold antibody level that is indicative of protection, then smaller trials measuring antibody concentration alone can be conducted to test new vaccines, instead of large efficacy studies powered on clinical outcomes. Commonly, threshold levels of protective antibody are determined from clinical efficacy trials in which clinical endpoints are measured on everyone and a small subset of participants have antibody concentrations measured. The proportion of participants with antibody below a threshold in each group in the immunogenicity subset can be compared to the proportions with disease in each group in the larger trial to find an appropriate threshold. Mathematically, this method seeks to compute an absolute threshold whereby antibody above the threshold provides complete, sterilizing immunity. However, in practice it is often understood that such thresholds may only be indicative of a relative degree of protection rather than an absolute one. Although this approach is common, the accuracy of such methods when the underlying mathematical assumptions do not hold true, has never been tested. We simulated data from clinical trial scenarios under varying assumptions of vaccine efficacy and calculated antibody thresholds of protection. We estimated the bias in the calculated thresholds derived from each scenario and showed that in many situations this method produces inflated estimates of thresholds, particularly if a vaccine induces high levels of antibody or when the underlying assumption of sterilizing immunity is violated.