Statistical Models for Predicting the Duration of Vaccine-Induced Protection

Abstract

When new vaccine products are marketed, they are generally supported by clinical studies lasting only a few years. The duration of protection from disease, however, is often expected to be considerably longer than the time spans of the clinical studies. Thus, there are both conceptual and methodological problems in trying to estimate long-term protection based on data from short-term clinical studies. We discuss the utility of predicting the duration of protection based on extrapolating observed antibody data. Several modeling strategies that have appeared in the clinical and statistical literature are reviewed. Two alternative mixed model strategies are considered and three methods of constructing confidence intervals for the mean duration of protection are proposed. Finally, a sample data set is used to illustrate and compare the various methods.