The choice of analytical methodology can alter conclusions regarding herd effects of paediatric pneumococcal vaccination programmes

Jean-Yves Pirçon,Carla A Talarico,Kaatje Bollaerts,William P Hausdorff,Christopher J Clarke

doi:10.1016/j.vaccine.2018.10.003

Abstract

BackgroundEstimation of the magnitude of the herd effect on invasive pneumococcal disease (IPD) is important when evaluating health benefits and cost-effectiveness of paediatric pneumococcal conjugate vaccine (PCV) programmes and may influence policy makers’ decisions on PCV use. Several epidemiological, programmatic, and immunological factors can affect the magnitude of the PCV herd effect. We investigated to what extent the choice of analytical methodology may also influence herd effect estimates. MethodsTo estimate the magnitude of the herd effect from paediatric PCV programmes, we examined overall IPD incidence rates in ≥65-year-olds from Finland, Australia, England/Wales, and the United States under different analytical scenarios. We used two different statistical methods: before/after comparison of average IPD incidence rates and interrupted time series (ITS) analysis accounting for underlying time trends. We also investigated how varying the length of the pre- and post-PCV analysis periods influenced the outcomes. ResultsThe estimated impact of paediatric PCV programmes on IPD incidence rates in adults ≥65 years varied substantially across the different scenarios within each country. The choice of statistical method and analysis periods contributed to this variation, and their influence varied by setting. For the datasets from England/Wales and the United States, the different scenarios produced relatively minor variation in estimated impact. For the Australian and Finnish datasets, differences were more prominent. In particular, for Finland, opposite conclusions could be drawn depending on the methodology: while no estimated herd effect was seen with the before/after method, a herd effect was evident with the ITS method. ConclusionsThe choice of statistical method and analysis periods can substantially influence the magnitude of estimated herd effects from paediatric PCV programmes. It is important to consider the reliability and presence of pre-PCV patterns in the IPD surveillance data used for analysis, the methodology and associated assumptions used to estimate herd effects.

Highlights

Using the before/after method, the estimated impact of the paediatric pneumococcal non-typeable Haemophilus influenzae protein Dconjugate vaccine (PHiD-CV) programme on overall invasive pneumococcal disease (IPD) incidence rate (IR) in !65-year-olds ranged between a 6.6% increase and a 9.0% decrease for 20/21 scenarios, in each case with 95% confidence interval (CI) including zero
The two scenarios yielding estimates suggestive of an increase in IPD IRs (26.1% and 34.9%) were those based on two years of data pre-and post-vaccine introduction, including the first year of FinIP enrolment; 95% CIs for these estimates were wide and included zero
For datasets where downward trends in pre-vaccine IPD IRs were observed in individuals !65 years, the interrupted time series (ITS) method could result in the absence of a herd effect and a qualitatively different conclusion than with the before/after method, as observed for the Australian 13-valent pneumococcal conjugate vaccine (PCV13) dataset

Summary

Introduction

q All authors attest they meet the ICMJE criteria for authorship.⇑ Corresponding author.Paediatric pneumococcal conjugate vaccine (PCV) programmes were primarily introduced to provide direct protection against pneumococcal disease in young children [1]. Estimation of the magnitude of the herd effect on invasive pneumococcal disease (IPD) is important when evaluating health benefits and cost-effectiveness of paediatric pneumococcal conjugate vaccine (PCV) programmes and may influence policy makers’ decisions on PCV use. Methods: To estimate the magnitude of the herd effect from paediatric PCV programmes, we examined overall IPD incidence rates in !65-year-olds from Finland, Australia, England/Wales, and the United States under different analytical scenarios. Results: The estimated impact of paediatric PCV programmes on IPD incidence rates in adults !65 years varied substantially across the different scenarios within each country. Conclusions: The choice of statistical method and analysis periods can substantially influence the magnitude of estimated herd effects from paediatric PCV programmes.

Methods

Results

Discussion

Conclusion