Measuring vaccine effectiveness against persistent HPV infections: a comparison of different statistical approaches

Background Persistent high-risk human papillomavirus (HPV) infection is endorsed by the World Health Organization as an intermediate endpoint for evaluating HPV vaccine effectiveness/efficacy. There are different approaches to estimate the vaccine effectiveness/efficacy against persistent HPV infections. Methods We performed a systematic literature search in Pubmed to identify statistical approaches that have been used to estimate the vaccine effectiveness/efficacy against persistent HPV infections. We applied these methods to data of a longitudinal observational study to assess their performance and compare the obtained vaccine effectiveness (VE) estimates. Results Our literature search identified four approaches: the conditional exact test for comparing two independent Poisson rates using a binomial distribution, Generalized Estimating Equations for Poisson regression, Prentice Williams and Peterson total time (PWP-TT) and Cox proportional hazards regression. These approaches differ regarding underlying assumptions and provide different effect measures. However, they provided similar effectiveness estimates against HPV16/18 and HPV31/33/45 persistent infections in a cohort of young women eligible for routine HPV vaccination (range VE 93.7–95.1% and 60.4–67.7%, respectively) and seemed robust to violations of underlying assumptions. Conclusions As the rate of subsequent infections increased in our observational cohort, we recommend PWP-TT as the optimal approach to estimate the vaccine effectiveness against persistent HPV infections in young women. Confirmation of our findings should be undertaken by applying these methods after longer follow-up in our study, as well as in different populations.