This cross-sectional observational study was conducted using the Symphony Health dataset (PRA Health Sciences, Raleigh, NC, USA) from the COVID-19 Research Database. The COVID-19 Research Database was established with Institutional Review Board approval and an exemption from patient consent because it included only data considered to be de-identified by the Health Insurance Portability and Accountability Act (HIPAA), HIPAA-limited data, or non-HIPAA-covered data, along with the strong governance measures in place to control access to all data. The Symphony Health dataset was derived from pharmacy and medical claims from several sources including Medicare, which covers about 280 million patients (almost all older adults), 1.8 million prescribers, and 16,000 health plans in the United States.
We used data collected from July 1, 2019, to June 30, 2020, about older adults (age 65 years old or older). To identify incident COVID-19 cases, we excluded individuals who had received a COVID-19 diagnosis before January 15, 2020. We also excluded individuals who had received an influenza vaccination on or after January 1, 2020, for two reasons. First, it generally takes 14 days for the body to develop antibodies against influenza after vaccination, so we hypothesized that the effect of the influenza vaccination would begin 14 days after the vaccination. Second, although our design is closer to a cross-sectional study, we used the 14-day window to strengthen the temporality between influenza vaccination and COVID-19 infection. All the exposure (that is, influenza vaccination) occurred before the outcome (that is, COVID-19 infection). Figure 3 shows the timeline of covariate assessment, receipt of influenza or other vaccine, and outcomes.
We classified individuals into an influenza-vaccination group and a no-influenza-vaccination group on the basis of their influenza vaccination status between September 1, 2019, and December 31, 2019. The differences among vaccines were considered as covariates, including trivalent or quadrivalent, live attenuated or inactivated, and with or without an adjuvant. The intranasal influenza vaccine was not recommended to older adults, thus we did not find any intranasal influenza vaccine users in our study sample. The National Drug Codes for the influenza vaccines and other vaccines were from the Centers for Disease Control (CDC). Two outcomes were identified on and after January 15, 2020, in this study: incidence of COVID-19 infection and incidence of severe illness because of COVID-19 infection. These two outcomes follow the U.S. Food and Drug Administration’s definition (Supplementary Figure S1).
The covariates included in this study were as follows: age 75 or older, gender, vaccinated against a disease other than influenza between July 1 and December 31, whether the influenza vaccine contained an adjuvant, and comorbidities that may increase the risk of COVID-19. The four types of vaccines that are recommended by the CDC and commonly administered to older adults were adjusted in Analysis 1 and served as comparators in Analysis 2 (that is, herpes zoster, pneumococcal pneumonia, tetanus, and hepatitis A). The influenza vaccines included in this study were mainly FLUZONE High-Dose (Sanofi Pasteur Inc.) and FLUAD (Seqirus USA Inc.), which accounted for 56% and 29% of the study sample, respectively. These were the two vaccines that were recommended for older adults by the CDC. The use of other influenza vaccines was never more than 5%. Both vaccines were trivalent in the study period; FLUZONE High-Dose is a vaccine without an adjuvant; FLAUD contains an adjuvant. Thus, we did not control the vaccine valency; instead, we adjusted the appearance of the adjuvant in multivariate analyses because a vaccine adjuvant may reduce COVID-19 severity. Comorbidities were selected according to the CDC’s warning about at-risk populations and included asthma, chronic kidney disease with dialysis, chronic lung disease, diabetes mellitus, hemoglobin disorders, immunocompromised, liver disease, serious heart conditions, and severe obesity (Supplementary Figure S1).
Two analyses were performed in this study. We first compared the odds of contracting COVID-19 between individuals who received an influenza vaccination versus those who did not (Analysis 1). To clarify the healthy user effect, we repeated the aforementioned analyses by comparing individuals who received an influenza vaccination to those who receive a vaccination other than for influenza (Analysis 2). Vaccines that are recommended for older adults were selected as the comparator (herpes zoster, pneumococcal pneumonia, tetanus, and hepatitis A).
The risk of COVID-19 infection and severe COVID-19 illness were evaluated with univariate and multivariate logistic regressions. To compare the influenza-vaccination group with the no-influenza-vaccination group, we divided the study cohort into 2000 subcohorts because computing efficiency limited the processing of such a large amount of data at once. We calculated the odds ratio in each subcohort and used a meta-analysis approach to get the pooled results. The final result was pooled using a random-effects model. Data were managed using the Snowflake® data warehouse (Snowflake Inc., San Mateo, CA, USA), and the analyses were performed using SAS® version 9.4 (SAS Institute Inc., Cary, NC, USA).