Model-based estimation of vaccine effects from community vaccine trials

Community vaccine trials are becoming increasingly important to assess both the direct and indirect community level effects of vaccination. In this paper, we present statistical methods to analyse such trials, using a design with several matched pairs of communities. The communities are matched on similarities in infection transmission as reflected through the basic reproduction number. Two methods of analysis are presented and compared. The first is simple empirical estimation of vaccine effects. Summary measures of these effects are constructed by reciprocal variance weighted averages across the community pairs. The second is likelihood-based where we derive a mixed effects epidemic model. This model takes the intercommunity variability into account through a random effect on the basic reproduction number. With this model, we derive a distribution-free estimator for the variance of the random effect. We use simulated epidemics to explore the performance of the two estimation methods for different numbers of community pairs and different levels of inter-pair variability. Both methods provide acceptable estimates in terms of bias and precision under reasonable conditions. Although the empirical approach involves fewer assumptions than the model-based approach, the resulting vaccine effectiveness estimates are only applicable to the vaccination fraction tested in the trial. In contrast, the model-based approach can be used to predict the vaccine effectiveness at vaccination fractions other than those used in the trial. Thus, it can be used as a public health policy tool for predicting the community level effects of vaccination. We demonstrate such use by predicting total vaccine effectiveness for the whole range of vaccination fractions.