Adjusting for unmeasured confounding bias in the Statin Treatment OPtions during vaccination against Flu (STOPFlu) study: an application of the Prior Event Rate Ratio Pairwise method

Background: Influenza (flu) vaccination is routinely advised for adults aged≥65 years. Many of these patients are also likely to be prescribed statins for control of cardiovascular risk. Previous studies have suggested that statin use may reduce the effectiveness of flu vaccination. However, these studies were susceptible to bias from unmeasured confounding.

Objectives: We report findings from the STOPFlu study investigating a potential modifying effect on the flu vaccine effectiveness from statins taken at the time of vaccination. Specifically, we compare results from analyses using two different approaches to confounding bias. Firstly, we adjusted for variables listed as confounders in previously published studies of the effect of statins on flu vaccine effectiveness. We then compared the results to those from a new quasi-experimental method to address unobserved confounding.

Methods: We recruited cohorts of vaccinated adults aged≥65y from primary care practices registered to the UK Clinical Practice Research Datalink for each year from 2010 to 2018 to compare new statin users to statin-free patients. The outcome was a composite of acute respiratory infection (ARI) diagnosed in primary care and hospital admissions for ARI. Using propensity scores in an inverse probability treatment weighted survival analysis, we adjusted for observed confounders informed by those published in the literature. Using the Pairwise likelihood, we then adjusted for the statin-free prior period among the vaccinated patients, as a measure of confounding bias from observed and unobserved sources.

Results: Weighted hazard ratios (HRs) varied between 1.29 (95%CI 1.21, 1.37) in 2014 and 1.09 (95%CI 1.02, 1.17) in 2017, indicating that at least 8 more ARI cases may have occurred for every 1000 new statin users. After further adjusting for unobserved confounders measured in the prior period, the HRs varied between 1.06 (95%CI 0.97, 1.16) in 2014 and 0.94 (95%CI 0.85, 1.04) in 2016, indicating no consistent or significant effect of statins on the flu vaccine.

Conclusions: Routinely collected health data are a valuable resource for studying real-world interactions between different therapies. The results from our weighted analyses were consistent with previously published literature using similar methods in showing a reduction flu vaccine effectiveness during statin use. However, with our adjustment for unobserved confounding bias, there was no consistent or significant effect on the flu vaccine against ARIs. Our study has demonstrated why caution is advised in interpreting results based on adjustment for observed confounders only, and that statins do not necessarily reduce flu vaccine effectiveness against ARIs.