Influenza vaccine effectiveness among outpatients in the US Influenza Vaccine Effectiveness Network by study site 2011‐2016

Abstract Background Influenza vaccination is recommended for all US residents aged ≥6 months. Vaccine effectiveness (VE) varies by age, circulating influenza strains, and the presence of high‐risk medical conditions. We examined site‐specific VE in the US Influenza VE Network, which evaluates annual influenza VE at ambulatory clinics in geographically diverse sites. Methods Analyses were conducted on 27 180 outpatients ≥6 months old presenting with an acute respiratory infection (ARI) with cough of ≤7‐day duration during the 2011‐2016 influenza seasons. A test‐negative design was used with vaccination status defined as receipt of ≥1 dose of any influenza vaccine according to medical records, registries, and/or self‐report. Influenza infection was determined by reverse‐transcription polymerase chain reaction. VE estimates were calculated using odds ratios from multivariable logistic regression models adjusted for age, sex, race/ethnicity, time from illness onset to enrollment, high‐risk conditions, calendar time, and vaccination status‐site interaction. Results For all sites combined, VE was statistically significant every season against all influenza and against the predominant circulating strains (VE = 19%‐50%) Few differences among four sites in the US Flu VE Network were evident in five seasons. However, in 2015‐16, overall VE in one site was 24% (95% CI = −4%‐44%), while VE in two other sites was significantly higher (61%, 95% CI = 49%‐71%; P = .002, and 53%, 95% CI = 33,67; P = .034). Conclusion With few exceptions, site‐specific VE estimates aligned with each other and overall VE estimates. Observed VE may reflect inherent differences in community characteristics of the sites and highlights the importance of diverse settings for studying influenza vaccine effectiveness.


| INTRODUC TI ON
Each year, infection with influenza causes an estimated 9.3 million to 45 million illnesses in the United States 1 and an associated average cost of $11.2 billion. 2 Annual vaccination is the most effective strategy for preventing influenza and reducing these burdens, as it has been shown to reduce the risk of influenza illness among the general population by 40%-60% when the vaccine is well-matched to the circulating viruses. son. 4 The current Network conducts observational studies across five sites in the United States to evaluate medically attended, laboratory-confirmed influenza and estimate vaccine effectiveness (VE) using a test-negative design. 5 Enrollees are outpatients seeking care for acute respiratory illness (ARI) who are tested for influenza using CDC's reverse-transcription polymerase chain reaction (RT-PCR) assay. VE is estimated by comparing the odds of vaccination among influenza-positive and influenza-negative outpatients who present with ARI.

Recent estimates of influenza VE have demonstrated varying VE
across influenza subtypes. 6 For example, vaccines tend to be more effective for preventing infection with influenza B and influenza A/ H1N1 viruses compared to influenza A/H3N2. 3 In a meta-analysis of VE studies during the 2004-2005 to 2014-2015 influenza seasons, pooled VE estimates were as follows: 33% (95% confidence interval [CI] = 26%-39%) for A/H3N2 viruses, 61% (95% CI = 57%-65%) for A/H1N1 viruses, and 54% (95% CI = 46%-61%) for influenza B viruses. 6 Vaccine effectiveness estimates against influenza may differ geographically due to characteristics of the study population, type of vaccine purchased by large healthcare organizations or pharmacies, vaccination coverage levels, and different circulating influenza strains. Studies from other countries indicate VE differences by strain/lineage, vaccine to circulating strain matching, age and birth cohort, and vaccination history. [7][8][9] In an agent-based cost-effectiveness analysis using synthetic populations, DePasse et al showed different epidemiologic curves and different numbers of averted cases of influenza across five US counties that differed by demographic characteristics such as size, density, and age distributions of their populations. 10 The US Flu VE Network consists of geographically and demographically distinct sites that contribute to annual estimations of influenza VE. It is unknown whether influenza VE differs across the sites that comprise the Network. In this study, we estimate site-specific influenza VE among outpatients who presented with ARI during the 2011-2016 influenza seasons at four of the five US Flu VE Network sites.

| ME THODS
Data were collected using a standardized protocol, as part of the CDC's US Flu VE Network study, for which detailed methods have been described. [11][12][13][14][15][16] IRB approval for the Flu VE Network protocol for all sites was granted. The screening criteria for an ARI were intentionally broad in order to maximize the pool of potential participants and included symptoms such as sore throat, fever/feverishness, congestion, wheezing, increased nasal secretions, body aches, and cough. Eligible participants were outpatients ≥6 months old seeking care for an ARI of less than 7-day duration with a cough. Patients who had received antiviral medication in the 7 days prior to enrollment, were younger than 6 months old, had enrolled in the study during the previous 14 days, or had received influenza vaccine within 14 days of enrollment were not eligible for enrollment or data analysis. Consented participants provided data on demographics, symptoms and other measures of current health and well-being, general health status, and self-report of influenza vaccination that were collected during patient enrollment interviews. High-risk conditions identified by International Classification assigned to a medical encounter during the year prior to enrollment were used to determine the presence of underlying health conditions associated with an increased risk of severe influenza. 16,17 ICD-9/10 codes were derived from electronic medical records (EMR).

| Sites
The US Flu VE Network was designed to include geographically diverse settings. The four sites included in this analysis were as follows: a large health maintenance organization located in and around Seattle, Washington, with a significant minority population of Asian enrollees; one site in a large university-affiliated health system located in north central Texas with a significant minority population of Latino enrollees; one site that combined outpatient clinics from a university health system and an inner-city health system located in Michigan with a significant minority population of black enrollees; and one site in a single large health system located in a county in Pennsylvania with one of the highest percentages in the USA of residents >65 years.

| Influenza vaccine
The influenza vaccine strains used each year are shown in Table   S1. For all five seasons, the A/H1N1 vaccine strain was the A/ Each season, vaccination status was defined as receipt of at least one dose of any seasonal influenza vaccine (including for children < 9 years old), according to EMR, immunization registries, and/ or plausible self/parental-report (with estimated date and plausible location of vaccination). All sites used the same questionnaire to query enrollees about vaccination at enrollment. Vaccination status was confirmed using data requests from the EMR and state immunization registries (electronic records), followed by requests to health insurance plans and providers such as physicians' offices, work and community sites, and pharmacies. For the remaining unconfirmed vaccinations, plausible self-report was based on additional information provided by the enrollee that had to include a location and approximate date.

| Laboratory methods
Nasal and throat swabs were collected from participants ≥2 years old, and nasal swabs only were collected from children <2 years old.
These specimens were tested for influenza using RT-PCR with CDCprovided primers and probes and were tested for A/subtype and B/ lineage. Patients who tested positive for influenza were cases, and patients who tested negative for influenza were controls.

| Statistical analysis
The influenza circulation period was calculated for each site in the Network. It was defined as the dates between the first and last influenza-positive enrollment during each season. Participants enrolled outside the influenza circulation periods were excluded from analyses. The enrollment period details for the Network sites are shown in Table S2.
Baseline characteristics were compared across sites using chisquare tests for categorical variables, and t tests or Wilcoxon ranksum statistics for continuous variables. Further, the statistical heterogeneity of estimates across sites was assessed using I 2 statistic calculated from a random effects meta-analysis. 22 Estimates of I 2 > 50% were considered to be highly heterogeneous. Data were analyzed using SAS version 9.4 (SAS Institute). Statistical significance was set at P < .05.

| Enrollee characteristics
There were significant differences among sites (P < .001) across each demographic variable measured (Table 1) Figure 2.
In meta-analysis testing for heterogeneity across the four clinical sites, three comparisons were made. Results are shown in Table   S4. Consistent with the results reported in Table 4     Any or all of these factors might contribute to influenza epidemiology and vaccine effectiveness. However, larger sample sizes may be necessary to elucidate those relationships. Further studies of differences in vaccine effectiveness by region may provide better estimates for computational modeling studies 10 that are used to predict influenza epidemiology, burden, and anticipated resource needs.

| Strengths and limitations
Strengths of this study include using data from five influenza seasons with different circulating strains, inclusion of diverse geographical sites, and generally adequate sample size. The testnegative design is subject to potential biases such as misclassification of the exposure variable (vaccination status), confounding or effect modification by prior vaccination effects, and confounding by characteristics of study enrollees; indeed, our sites differed in demographic characteristics and vaccination rates.
However, consistent protocols and rigorous attempts by the sites F I G U R E 2 Vaccine effectiveness by site to confirm vaccination status were employed to minimize bias.

| CON CLUS IONS
With few exceptions, site-specific vaccine effectiveness estimates aligned with each other and overall VE estimates. Observed VE differences may reflect inherent differences in community characteristics of the participating sites and highlight the importance of diverse settings for studying influenza vaccine effectiveness.