Improving the estimation of influenza vaccine effectiveness

Abstract

Vaccination is one of the most effective measures in influenza control. Monitoring vaccine effectiveness (VE) plays a crucial role in informing vaccination policy. VE is most often measured by the test-negative design (TND) study. Although some studies have examined factors influencing VE, there remains substantial uncertainty in VE estimation.

In my thesis, I aimed to enhance our understanding of influenza VE and suggest methodological improvements. By analyzing a TND study that enrolled children hospitalized for acute respiratory infections, I found that while influenza vaccination conferred moderate protection against hospitalization in children, VE against hospitalization and the number of hospitalizations averted by vaccination was higher in younger children as compared with older children. Estimating VE in partially vaccinated children (previously unvaccinated children who were currently vaccinated with one dose instead of recommended two doses) <9 years of age, I found that partial vaccination was only protective in children 3 to 5 years of age. Similar analyses by influenza type/subtype showed that partial vaccination was only effective against influenza A(H1N1)pdm09. Relative VE comparing full vaccination with partial vaccination indicated that full vaccination was consistently more effective in these children.

By conducting simulation studies, I concluded that significant bias in VE was unlikely unless sick persons tend to be very slow in seeking care or care-seeking behavior was differential by vaccination status. I demonstrated the potential of the Bayesian approach to correct for case misclassification bias. Adjusting for delay as a linear variable and restriction analyses could also improve VE estimates, although more stringent restriction criteria may reduce precision in VE estimates. I further identified that the direction of bias could differ depending on how prior vaccination and infection influences current vaccination, implicating the importance of understanding the immune mechanism from previous exposures.

I systematically reviewed studies employing the TND to estimate VE against influenza and non-influenza infections and identified heterogeneity in key methodological choices among these studies, some of which may threaten the validity of the TND study.

My findings support the current policy to vaccinate children and the recommendation of a two-dose regimen in children <9 years of age. My studies also identified potential biases in VE requiring appropriate control to improve VE estimation. Future studies are needed to understand indirect protection from vaccinating children. Future non-influenza studies applying the TND for VE estimation should carefully consider case definitions used to recruit patients, choices of controls and adequate control for confounders.