Introduction: There is a need to estimate the impact of pneumococcal conjugate vaccines against pneumonia. However, due to sparse data and variations in pneumonia trends unrelated to vaccine, such as respiratory virus epidemics or changes in reporting measures, it can be difficult to estimate true effects.
Objectives: We aim to develop and explore the use of Bayesian statistical methods that pool findings between multiple studies to obtain more precise vaccine effect estimates for each individual study and to obtain average effects across all studies.
Methods: We obtained national-level hospital administrative data on all-cause pneumonia in Brazil, Chile, Ecuador, Mexico, Denmark, and the United States. Initial rate ratio estimates of PCV impact were calculated for each dataset using the synthetic controls method which has known effectiveness in evaluating vaccine impact. Bayesian hierarchical meta-analysis was performed to combine information and to obtain "improved" estimates for each location. Individual studies were weighted based on the amount of within location variability, with more variable locations being pulled more towards the average estimated trajectory. Information on vaccine uptake, schedule, and other factors were considered in the model.
Results: We obtained estimates for each post-vaccine time point using the original synthetic control method and also the newly developed meta-analysis model. The meta-analysis approach increases the probability of detecting a true effect and decreases variation among estimates compared with those initially obtained from individual time series. Number of cases averted and relative decline were calculated for each location and time point.
Conclusion: This modeling approach allows for the generation of credible estimates of vaccine impact, even in locations with sparse and/or noisy data.