Vaccine Effectiveness Against Pediatric Influenza Hospitalizations and Emergency Visits

We evaluated VE among study participants in the NVSN, a network of pediatric hospitals in 7 cities (Kansas City, MO; Rochester, NY; Cincinnati, OH; Pittsburgh, PA; Nashville, TN; Houston, TX; and Seattle, WA) that performed prospective surveillance for acute respiratory illness among children who were hospitalized or who sought care in and were discharged from the ED. One main objective of the NVSN was to evaluate protection conferred by vaccination against severe influenza-associated illnesses (hospitalization) among children. All sites enrolled hospitalized children <18 years. Among ED visits, 4 sites enrolled children <18 years (Rochester, Cincinnati, Nashville, and Houston), and 3 sites enrolled children only <5 years of age (Kansas City, Pittsburgh, and Seattle). Although the NVSN enrolled children year-round, for this analysis, we included participants enrolled during each site-specific influenza season, defined as the date of the first through the last influenza-positive case in enrolled subjects at each site.

Children were eligible for NVSN enrollment if they had at least one of the following symptoms or events with a duration <14 days: fever, cough, otalgia, nasal congestion, rhinorrhea, pharyngitis, myalgias, posttussive vomiting, wheezing, dyspnea or rapid or shallow breathing, apnea, and brief resolved unexplained event. Children were excluded from enrollment if they had fever and neutropenia (absolute neutrophil count of <500 neutrophils per μL) with malignancy, were a newborn who had never been discharged from the hospital, had been hospitalized for >48 hours or transferred from another hospital after an admission of >48 hours, were discharged from a hospital in the previous 4 days, had been previously enrolled within 14 days before their current hospitalization or ED visit, or had a known nonrespiratory condition as the reason for their hospitalization or ED visit.

Study staff obtained demographic information, illness characteristics, and medical history, including underlying medical conditions, for each enrolled child through parent or guardian interviews and standardized medical chart review. The data were considered preliminary at the time of analysis. The study protocol was approved by the Centers for Disease Control and Prevention (CDC) and participating institutional review boards. Written informed consent was obtained by the parent or legal guardian of the child, and assent from the child was obtained as required.

At enrollment, we collected a midturbinate nasal swab and throat swab to be combined in viral transport medium. An alternative to swabs collected by research staff was a comparable respiratory specimen (eg, tracheal aspirate, nasal wash, nasopharyngeal swab, or midturbinate nasal swab) collected for clinically indicated testing independent of this study.

Research testing was performed for influenza viruses by using molecular assays, which varied by site: Luminex NxTAG Respiratory Pathogen Panel¹⁸  (Kansas City and Cincinnati sites), BioFire FilmArray Respiratory Panel (Seattle site),¹⁹  Applied Biosystems TaqMan Array microfluidic card (Rochester site),^20,21  and in-house real-time reverse transcription-polymerase chain reaction assays with primers, probes, and the testing protocol developed by the CDC (Pittsburgh, Nashville, and Houston sites).^22,23  Subsequent reverse transcription-polymerase chain reaction assays were performed as needed to identify influenza A subtypes and B lineage on research and salvage specimens. For some patients with only a hospital clinical standard-of-care result, there were insufficient samples to determine influenza A subtypes or B lineage. Annual CDC proficiency panels were successfully completed by all participating laboratories.

We collected information from the medical record regarding clinical molecular test results for influenza obtained as part of standard patient care. In situations with discordant results between research and clinical testing, we classified a positive result by either research or clinical molecular influenza testing as influenza-positive. We excluded clinical testing by rapid antigen detection tests from this analysis.

For all children, we asked the parent or guardian about their child’s receipt of the 2018–2019 influenza vaccine during the enrollment interview to determine if a vaccine had been received for the current season (since July 1, 2018). For hospitalized children, current-season vaccination status was confirmed from documentation in state immunization information systems, from the electronic medical record, and/or via phone call or fax from providers.²⁴  When using documented vaccination status, we classified a child as vaccinated if they received a current-season vaccine at least 2 weeks before symptom onset. Because complete data regarding influenza vaccination status for past seasons were not yet available, we defined children as vaccinated or unvaccinated and did not evaluate full and partial vaccination status. For hospitalized children missing documentation, we supplemented vaccination status with that obtained by parent or guardian report. When using parental report, a child was considered vaccinated for the current season if they were said to have received a current-season vaccine before the hospitalization or ED visit.

The primary analysis was evaluation of VE against influenza-associated hospitalizations for any influenza virus and for influenza A subtypes. As a secondary analysis, we evaluated VE against influenza-associated ED visits. We excluded participants <6 months of age who were not eligible to receive the influenza vaccine, those with inconclusive laboratory test results or those who were influenza-positive only by a rapid antigen detection test, those enrolled >10 days after symptom onset, children with documented vaccination <2 weeks before symptom onset, and children missing data for relevant covariates to be considered in the multivariable model (Figs 1 and 2). We compared characteristics among influenza-positive case patients and influenza-negative control patients using descriptive statistics (χ² or Wilcoxon rank tests).

We estimated influenza VE using a test-negative study design, comparing the odds of vaccination among case patients who tested positive for influenza compared with control patients who tested negative for influenza, using multivariable logistic regression models with Firth penalization.²⁵  The test-negative design is widely used nationally and internationally for observational studies of influenza VE, including in previous NVSN and other US VE publications, and among ambulatory and hospitalized adults and children.^{8,10,12,13,24,26–34}  This methodology is considered to reduce bias associated with confounding by health care–seeking behavior, to lower rates of misclassification of influenza status, and to be better at identifying control patients who arise from the same source population as case patients, which is particularly important for studies of severe influenza outcomes, such as hospitalization.^30–34  VE was calculated as (1 − the adjusted odds ratio) × 100%.

A priori, we included study site, age, and calendar time (month of enrollment) in logistic regression models.¹³  The Akaike information criterion values were similar when age was included as a continuous or categorical variable; therefore, we added age as a continuous variable to conserve model parameters. Additional potential covariates to be evaluated included sex, race and/or ethnicity, insurance status, reported number of hospitalizations in the past year, days from illness onset to enrollment, and underlying medical conditions. Models were built by using a manual forward stepwise procedure. Covariates were retained if they were considered to be a priori confounders, if they remained significant at a P value <.05, or if they influenced the VE estimate by >5%. For all analyses, we conducted a subgroup analysis to estimate VE by age (<5 years and 5–17 years) and evaluated P values from an interaction term to determine if estimates between age groups were statistically significantly different. All analyses were completed in SAS version 9.4 (SAS Institute, Inc, Cary, NC).

During the 7 site-specific dates for the influenza season (encompassing November 7, 2018, to June 21, 2019), we enrolled 2748 hospitalized children and 2676 children with ED visits that did not result in hospitalization (Figs 1 and 2).

After exclusions (mostly for children <6 months), 1792 (65%) hospitalized children were eligible for the primary VE analysis. Among the hospitalized group, 226 (13%) children were positive for influenza virus infection; 211 (93%) had influenza A viruses, and 15 (7%) had influenza B viruses. Of the influenza A viruses, 107 (51%) were A(H3N2), 82 (39%) were A(H1N1)pdm09, 1 was a dual infection with both viruses, and 21 (10%) were not subtyped (Fig 1). Among the 1792 hospitalized children in the VE analysis, 1611 (90%) had confirmed documentation of vaccination status available and 181 (10%) had vaccination status by parental report alone. Only 1 participant was documented to have received a live-attenuated influenza vaccine. Of those who received an inactivated influenza vaccine, >98% received a quadrivalent vaccine. Overall, in the hospitalized VE analysis group, 88 (5%) participants received mechanical ventilation and 7 (<1%) died.

Approximately half of influenza-positive case patients (109 of 226) and 1146 (73%) of 1566 influenza-negative control patients were 6 months to <5 years old (Table 1). Influenza-positive case patients differed from control patients by race and/or ethnicity and study site but were similar with respect to other characteristics. Influenza-vaccinated children differed from unvaccinated children by race and/or ethnicity, number of hospitalizations in the past year, and insurance status.

Among ED patients, 1944 (73%) of 2676 were eligible after exclusions. Of 1944 ED patients, 420 (22%) were positive for influenza virus infection, with 398 (95%) positive for influenza A viruses and 22 (5%) positive for influenza B viruses. Among influenza A virus infections, 203 (51%) were A(H3N2), 146 (35%) were A(H1N1)pdm09, 1 was a dual infection, and 48 (12%) were not subtyped (Fig 2). Both A(H1N1)pdm09 and A(H3N2) detections overlapped throughout the season (Fig 3).

Among 1944 enrolled children with ED visits, 290 (69%) of 420 influenza-positive case patients and 1268 (83%) of 1524 influenza-negative control patients were 6 months to <5 years old (Table 2). Influenza-positive case patients differed from control patients by race and/or ethnicity, number of hospitalizations in the past year, insurance status, and time since symptom onset. Influenza-vaccinated ED children differed from unvaccinated children by race and/or ethnicity, presence of underlying medical conditions, number of hospitalizations in the past year, and insurance status.

Influenza-positive case patients were less likely to be vaccinated than influenza-negative control patients among both hospitalized (47% vs 62%) and ED patients (40% vs 58%). Overall, adjusted VE against any influenza-associated hospitalization in children was 41% (95% confidence interval [CI], 20% to 56%) (Table 3). When stratified by virus subtype, VE was 41% (95% CI, 11% to 61%) against A(H3N2) viruses and 47% (95% CI, 16% to 67%) against A(H1N1)pdm09 viruses. VE for influenza B viruses was not estimated because of insufficient influenza cases (n = 15). By age group for hospitalized patients for all influenza viruses, VE was 46% (95% CI, 19% to 64%) among children 6 months to <5 years and 23% (95% CI, −23% to 51%) among children 5 to 17 years (P value for differences between age groups = .37). When estimating VE for influenza-associated hospitalizations using vaccination status by parental report alone for hospitalized children in the data set, VE estimates were similar: 41% (95% CI, 20% to 56%) for all influenza viruses, 35% (95% CI, 2% to 57%) against A(H3N2) viruses, and 50% (95% CI, 20% to 69%) against A(H1N1)pdm09 viruses.

Among ED patients, adjusted VE was 51% (95% CI, 38% to 62%) against any influenza virus infection (Table 4). When stratified for virus subtype, VE was 39% (95% CI, 15% to 56%) against A(H3N2) viruses and 61% (95% CI, 44% to 73%) for A(H1N1)pdm09 viruses. VE for influenza B viruses was not estimated because of insufficient influenza cases (n = 22). VE for all viruses was 55% (95% CI, 39% to 67%) among children 6 months to <5 years and 35% (95% CI, −6% to 60%) among those 5 to 17 years (P value for differences between age groups = .19).

Circulating influenza viruses are constantly evolving,³⁵  and during the 2018–2019 US influenza season, influenza A viruses predominated, with cocirculating A(H1N1)pdm09 and A(H3N2). This study reveals that influenza vaccination significantly reduced laboratory-confirmed influenza hospitalization among children enrolled in the NVSN by 41%. We estimated a significant reduction in hospitalizations associated with both A(H3N2) and A(H1N1)pdm09 viruses (point estimates of 41% and 47%, respectively), despite circulating A(H3N2) viruses that were antigenically different from the A(H3N2) influenza vaccine component.¹⁷  We also found that influenza vaccination reduced laboratory-confirmed influenza ED visits by half. This study reveals that in this season, influenza vaccines prevented moderate to severe illness in children even when one of the vaccine components was not well matched. VE estimates for both outcomes were similar against all influenza viruses including against both A(H3N2) and A(H1N1)pdm09 viruses, specifically.

Studies from other countries for the 2018–2019 influenza season have demonstrated significant protection against influenza-associated ambulatory care visits and hospitalizations among children infected with A(H1N1)pdm09 viruses. However, protection against A(H3N2) viruses in children has varied: 24% in the United States (outpatients 0–8 years), 17% and 31% in England (outpatients and hospitalized patients 2–17 years, respectively), 46% in Europe (outpatients 0–14 years), and 48% in Canada (outpatients 1–19 years).^36–40  In the United States, genetic characterization of US Flu VE Network influenza A(H3N2)-positive specimens revealed that the majority of specimens belonged to clade 3C.3a viruses, which differed from the A(H3N2) subclade 3C.2a1 component of the vaccine.^17,36  Thus, it could be expected that vaccination with a 3C.2a1 virus may not result in significant protection against illnesses caused by the predominantly circulating clade 3C.3a viruses. In fact, some research groups have explored VE against A(H3N2) viruses by clade and age and have demonstrated that a pronounced lower VE was estimated for clade 3C.3a viruses among certain adult age groups compared with children.^37,40  Authors of these studies have hypothesized that the observed differences in VE suggests a differential immune effect of vaccination by age related to childhood imprinting.^37,40  Our VE data are consistent with the protective effect of vaccination against influenza illness associated with H3N2 viruses in children during the 2018–2019 season.

Reasons for variation in vaccine protection are multifactorial and can include virus-, host-, and environment-related factors. We do not have sequence data for our A(H3N2) specimens, but A(H1N1)pdm09 and A(H3N2) viruses composed mostly of clade 3C.2a viruses (matched to vaccine viruses) were more common in the United States until February 2019; subsequently, there was varying geographic distribution.³⁶  Only 14% of A(H3N2) viruses from our hospitalized subjects and 21% of A(H3N2) viruses from our ED participants were collected from the start of the season to February 2, 2019, when we would have expected the vaccine to be well matched. Thus, other factors, such as regional variation in circulating viruses, and host factors, such as age (and thus, birth cohort), imprinting, and previous vaccination, are likely related to our finding of vaccine protection against both A(H1N1)pdm09 and A(H3N2) viruses.

We observed similar VE estimates against influenza-associated ED visits and hospitalizations. One notable difference between the two populations is that hospitalized children likely represent more medically complex patients, with 58% having underlying medical conditions and 38% reporting at least one hospitalization in the past year, compared with 28% and 14%, respectively, among ED participants.

Strengths of our study include prospective multisite enrollment across clinical settings in geographically diverse locations, with systematic testing using sensitive molecular influenza assays. This study uniquely included pediatric hospitalizations and ED care, for which previous data were sparse. Our analysis also has limitations as a single-season study with a limited sample size, especially for age group–specific estimates. We controlled for enrollment month, but this may not have been sufficient to account for potential effects of calendar time on VE. We did not evaluate full and partial vaccination status. In addition, we lacked documented vaccination data for many ED patients. Verification of vaccination status for the current season and past seasons using documentation of vaccine receipt is ongoing for hospitalized children but is not routinely performed, per the NVSN protocol, for ED enrollees. Therefore, we chose to use parent- or guardian-reported vaccination status for the ED analysis, which may be subject to recall bias and exposure misclassification. However, self- or parent-reported influenza vaccination has been reported to be generally accurate for current-season vaccination,^41–43  and the estimates we present for hospitalized patients, using mostly (90%) documented vaccination status, were similar to the estimates using parent-reported vaccination status alone. Finally, ours is an observational study; we may not have controlled for all confounders, and there are limits to test-negative designs, such as the risk of selection bias.^44,45  Although we believe the test-negative design is optimal for our study of hospitalized and ED children, our results should not be interpreted as VE against influenza-associated ambulatory care visits or infections that are not medically attended.

In this study, we provide timely evidence of the overall benefit of vaccination in reducing pediatric hospitalizations and ED visits associated with influenza infections. Unlike studies that are focused on ambulatory care office visits, our data provide important VE estimates against severe influenza in children. Strikingly, the vaccine was ∼40% effective against A(H3N2)-related hospitalizations and ED visits, even in a season when antigenically drifted clade 3C.3a influenza viruses were the predominant circulating A(H3N2) viruses. These data provide important evidence supporting the annual recommendation that all children ≥6 months should receive influenza vaccination.

We acknowledge Elizabeth Schlaudecker (Cincinnati, OH); Pedro A Piedra, Vasanthi Avadhanula, and Flor M. Munoz (Houston, TX); Samar Musa, Noreen Jeffrey, and Monika Johnson (Pittsburgh, PA); Christina Albertin, Wende Fregoe, Lynne Shelley, Miranda Marchand, Theodore Pristash, and Joshua Aldred (Rochester, NY); Kirsten Lacombe, Bonnie Strelitz, Ashley Akramoff, Jennifer Baxter, Rachel Buchmeier, Kaitlin Cappetto, Jennifer Jensen, Sarah Korkowski, Katarina Ost, Hannah Schlaack, Hannah Smith, Sarah Steele, Ariundari Tsogoo, and Emily Walter (Seattle, WA); Jessie Chung, Jill Ferdinands, Yingtao Zhou, Iddrisu Sulemana, Roberto Mejia, LaShondra Berman, and John Barnes (Atlanta, GA); and all the members of the NVSN.

We thank the children and parents who participated in this study.

CDC

Centers for Disease Control and Prevention

CI

confidence interval

ED

emergency department

NVSN

New Vaccine Surveillance Network

VE

vaccine effectiveness