W. Zane Billings, Savannah L. Miller, Jessica H. Knight, Murphy John, Hayley Hemme, Andreas Handel

Potential other authors: Ted M. Ross, Andrea Sant, Ye Shen, Natalie E. Dean, Benjamin J. Cowling

Explore how age and vaccine dose interact to effect the antibody response. The UGAFluVac cohort allows participants over the age of 65 to choose whether they receive FluZone standard dose or FluZone high dose, and Andrea Sant’s cohort study administered FluZone HD to individuals aged 18 – 49. By combining the two datasets, we can use causal inference and hiearchical modeling techniques to understand the effect of dose, and how this relates to previous mechanistic modeling predictions.

Introduction: The effectiveness of conventional seasonal influenza vaccines such as FluZone (Sanofi Pasteur) is limited in elderly populations. Increasing the dose of antigen in the vaccine formulation can allow vaccines to induce a robust immune response in the elderly. FluZone High Dose (HD) was licensed for public use in participants 65 and older in [YEAR] (CITE THIS). The HD vaccine contains 4x the amount of influenza hemagglutinin antigen as the standard dose (SD) formulation (CITE THIS), and achieved standard immunogenicity targets in elderly participants in randomized clinical trials (CITE THIS). Recent research has suggested that HD vaccination may be beneficial in young populations as well (MORITZSKY AND OTHERS FROM AS GROUP). In our study, we combine data from multiple longitudinal vaccine studies to analyze the effect of FluZone vaccine dose across age groups in a causal inference framework.

Methods: We used data from two vaccine studies, both of which followed a prospective, longitudinal open cohort design. Individuals were recruited into the studies across four study sites between 2013 – 2020. The UGAFluVac cohort study has been described previously (CITE) and consisted of two clinical sites in Pittsburgh, PA and Port St. Lucie, FL from 2013 to 2016, before moving to Athens, GA in 2016 (samples were collected at all three study sites in the 2016–2017 influenza season). The UoRFluVac study has also been described previously (CITE) and consisted of one study site recruiting from 2015 – 2020 in Rochester, NY. Both studies collected pre-vaccination and post-vaccination (target 21 or 28 days depending on season and study) serum draws, and ran hemagglutination inhibition assays against similar influenza strains to those contained in the current seasonal vaccine (the so-called “homologous” HAI assays). Importantly, participants in UGAFluVac aged from 18 and older with an additional subcohort of children aged 10 – 17 beginning in the 2017–2018 influenza season. UGAFluVac participants who were 65 or older were given the choice between SD and HD vaccines, while participants under the age of 65 were given SD. Participants in the UoRFluVac study aged from 18 – (MAX AGED) and were randomly assigned to one of four vaccine groups, including Fluzone SD and HD. We estimated the conditional average treatment effect (CATE) of dose on age using Bayesian regression, and accounted for study site using a random intercept. Other confounders were identified for adjustment using a directed acyclic graph (DAG). We also implemented nonparametric, double-robust targeted maximum likelihod estimation (TMLE) as a robust comparator for estimating the CATE, although TMLE is less interpretable than regression methods.

QUANTITATIVE BIAS ANALYSIS FOR UNMEASURED CONFOUNDING?

Results:

• DAG: we saw the following
• Descriptive statistics: across the two studies we saw that….
• Stratified analysis across studies: we saw that ….in the crude analysis and …. in the stratified analysis adjusting for relevant confounders
• Bayesian multilevel regression analysis: after adjusting for relevant confounders and residual differences across study sites, we observed ….
• TMLE analysis: in the TMLE analysis we saw that …. this estimate WAS / WAS NOT the same as the regression analysis.
• QBA: we saw that unmeasured confounding would have to have a magnitude of …. to qualitatively change our results.

Discussion:

• What do stratified vs multilevel analyses tell us about the comparability of the data
• Why the regression/TMLE analysis were or weren’t the same
• While there are several assumptions necessary which may limit the external validity of our study, to our knowledge no direct comparison has been made between SD/HD vaccines across multiple age groups. Our results suggest that HD vaccines do/do not noticeably increase immunogenicity in WHICH AGE GROUPS.
• It is likely that fluctuating VE due to seasonal factors and antigenic evolution affect the relative benefits of HD in different years.
• While immunogenicity may be increased, this does not directly translate into an appreciable increase in protection from disease or severe outcomes, due to the inherently nonlinear relationship between those things. True efficacy trials or challenge studies are necessary to understand protective benefits.

Text and figures: all text and images I create in this repository are licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License, CC BY-NC-SA 4.0.

Code: all code created by me is licensed under the GNU Affero General Public License, AGPL 3.0.

Please see my website for contact information: https://wzbillings.com.

Bug reports can be filed here.