Vaccine effectiveness against laboratory-confirmed influenza hospitalizations among young children during the 2010-11 to 2013-14 influenza seasons in Ontario, Canada

Uncertainty remains regarding the magnitude of effectiveness of influenza vaccines for preventing serious outcomes, especially among young children. We estimated vaccine effectiveness (VE) against laboratory-confirmed influenza hospitalizations among children aged 6–59 months. We used the test-negative design in hospitalized children in Ontario, Canada during the 2010–11 to 2013–14 influenza seasons. We used logistic regression models adjusted for age, season, and time within season to calculate VE estimates by vaccination status (full vs. partial), age group, and influenza season. We also assessed VE incorporating prior history of influenza vaccination. We included specimens from 9,982 patient hospitalization episodes over four seasons, with 12.8% testing positive for influenza. We observed variation in VE by vaccination status, age group, and influenza season. For the four seasons combined, VE was 60% (95%CI, 44%-72%) for full vaccination and 39% (95%CI, 17%-56%) for partial vaccination. VE for full vaccination was 67% (95%CI, 48%-79%) for children aged 24–59 months, 48% (95%CI, 12%-69%) for children aged 6–23 months, 77% (95%CI, 47%-90%) for 2010–11, 59% (95%CI, 13%-81%) for 2011–12, 33% (95%CI, –18% to 62%) for 2012–13, and 72% (95%CI, 42%-86%) for 2013–14. VE in children aged 24–59 months appeared similar between those vaccinated in both the current and previous seasons and those vaccinated in the current season only, with the exception of 2012–13, when VE was lower for those vaccinated in the current season only. Influenza vaccination is effective in preventing pediatric laboratory-confirmed influenza hospitalizations during most seasons.


Introduction
Rates of influenza-attributable hospitalizations are as high among young children as older adults [1]. Canada's National Advisory Committee on Immunization (NACI) identifies children aged 6-23 months as a priority group for vaccination, and added children aged 24-59 months to the list of recommended recipients starting in the 2012-13 influenza season [2]. Uncertainty remains regarding vaccine effectiveness (VE) against serious outcomes, including hospitalizations and deaths, especially among young children. Indeed, the latest Cochrane review reported no conclusive evidence that influenza vaccines reduce hospitalizations among children younger than two years of age [3].
In addition, the impact of repeated vaccination on VE has been a topic of recent controversy, [4] but few studies have focused on children or hospitalized patients. Ohmit et al. found no negative impact of repeated vaccination on VE in children younger than 9 years of age in 2013-14 when A/H1N1 circulated. This contrasted with their results from the prior season when A/H3N2 dominated and VE was higher for those vaccinated in the current season only [5]. Thompson et al. examined this phenomenon in children younger than 8 years of age in 2012-13 and found similar VE estimates for those vaccinated in the current season only and those vaccinated in both the prior and current seasons [6].
The objective of this study was to evaluate influenza VE against hospitalizations for young children aged 6-59 months for the 2010-11 to 2013-14 seasons in Ontario, Canada. We also assessed the impact of repeated influenza vaccination in young children.

Study population, setting, and design
We studied children aged 6-59 months who were hospitalized during the 2010-11 to 2013-14 influenza seasons in Ontario. Individual-level laboratory data were collected from a network of laboratories and linked using unique encoded identifiers to health administrative data, including hospital records and physician billing claims, at the Institute for Clinical Evaluative Sciences (ICES).
We estimated VE using the test-negative design (TND) [7,8]. We included all respiratory specimens tested for influenza; those with laboratory-confirmed influenza served as cases and those testing negative served as controls. The study was restricted to periods when influenza was in circulation, based on a threshold level of 5% positivity on tested respiratory specimens for the province (S1 Table).
Ethics approval for this study was obtained from the Sunnybrook Health Sciences Centre Research Ethics Board and the Office of Research Ethics at the University of Toronto, Toronto, Canada. Written or verbal informed consent was not required from the subjects in this study, as ICES is permitted to obtain and use personal health information without patient consent for specific purposes, as outlined in Ontario's Personal Health Information Protection Act. All data were fully anonymized before access by the researchers, and the requirement for informed consent was waived by the aforementioned ethics committees.

Data sources and definitions
Laboratory testing. We included respiratory specimens tested at five hospitals and 11 public health laboratories distributed across the province. We included specimens tested using monoplex and multiplex polymerase chain reaction (PCR), viral culture, direct immunofluorescence assay (DFA), or enzyme immunoassay tests (EIA). In this dataset, 62.3% of children were tested using PCR, 22.0% with viral culture, 15.8% with DFA and 0.3% with EIA.
Hospitalizations. Hospitalizations were identified using the Canadian Institute of Health Information Discharge Abstract Database. We restricted the analysis to individuals with specimens collected within three days of hospital admission (96% of the sample) to minimize the inclusion of hospital-acquired influenza infections. We included one hospitalization with a specimen collected per individual per season. For individuals with multiple specimens collected during multiple hospitalizations within a season, we included the first hospitalization with a specimen positive for influenza (cases), or the first hospitalization if all specimens collected within the season were negative (controls).
Influenza vaccination. We used physician billing claims recorded in the Ontario Health Insurance Plan (OHIP) database to ascertain influenza vaccination status. Children were classified as fully vaccinated if they had two influenza vaccination billing claims in the current season with administration dates !28 days apart, or one billing claim in the current season with another billing claim in any prior season, and if the administration date of the dose(s) in the current season was !14 days before specimen collection date. Children were classified as partially vaccinated if they received two doses in the current season with the second <28 days after the first or <14 days before specimen collection date, or if they received only one of two recommended doses in the current season (!14 days before specimen collection date). Both trivalent (TIV) and quadrivalent (QIV) inactivated vaccines are recommended for those aged 6-23 months but only TIV products were available in Ontario during the study period. For the 2011-12 to 2013-14 seasons, live attenuated inactivated vaccines (LAIV) were preferentially recommended over inactivated vaccines for healthy children aged 24-59 months, but they were not part of the publicly funded immunization program and therefore had minimal uptake [2].
Covariates. We used health administrative data to identify demographic characteristics, underlying health conditions (including asthma, diabetes, and cancer), and prior healthcare use, including past hospitalizations and continuity of care (defined as the percentage of primary care visits assigned to the child's primary care provider). We used postal code of residence to assign neighbourhood income quintile (based on census-derived neighbourhood income) and rurality. We modified Feudtner et al.'s methodology [9] to suit the available data to assign complex chronic condition (CCC) status using physician billing and hospital claims (S2 Table). Birthweight and gestational age were determined from the mother's delivery hospitalization record.

Statistical analysis
We used logistic regression to estimate VE by comparing the odds of vaccination in the cases to the odds of vaccination in the controls. VE was calculated as (1-OR adjusted )x100%. We estimated VE for full and partial vaccination status separately. A priori, we controlled for age (in months), season, and time within season (month relative to peak) in the adjusted estimates [10,11]. All potential confounders were evaluated for inclusion in the adjusted model using the Hosmer and Lemeshow forward model building strategy [12], but no additional covariates were included.
We tested for a difference in VE between fully and partially vaccinated children by calculating the odds ratio with those partially vaccinated as our reference [6]. We performed subgroup analyses by season, subtype, age group, sex, presence of a CCC, and time within season. In sensitivity analyses, we restricted the analysis to individuals with specimens collected during a hospitalization with an acute respiratory illness (ARI) diagnostic code recorded in any of the diagnosis fields in the hospitalization record (S3 Table); we evaluated the addition of any comorbidity (any CCC, diabetes, asthma, and/or cancer) on VE estimates; and we accounted for the possibility of misclassification of our exposure measurement [13], since not all influenza vaccinations are recorded in the OHIP database. As there are no published values for the sensitivity and specificity of physician billing codes for influenza vaccination of children younger than 5 years of age, we extrapolated measures from available data to this age group [14,15]. To demonstrate specificity of the association between influenza vaccination and influenza hospitalization, we evaluated the association between influenza vaccination and laboratory-confirmation of one of four non-influenza respiratory viruses (respiratory syncytial virus, adenovirus, parainfluenza, or human metapneumovirus), for which no association was expected.
To assess the impact of repeated vaccination, we estimated VE for any vaccination in those aged 24-59 months by vaccination history using a categorical indicator variable for those vaccinated in both the current and prior seasons, the current season only, the prior season only, and neither season (i.e., four mutually exclusive groups).
We tested two main assumptions of the TND: 1) influenza testing is not associated with vaccination status; and 2) vaccination status is not associated with non-influenza respiratory viruses [7,8,16].
All analyses were conducted using SAS Enterprise Guide 6.1 (SAS Institute Inc., Cary, NC). All tests were two-sided and used p<0.05 as the level of statistical significance.

Results
We included 9,982 hospitalization events during which a respiratory specimen was collected and tested for influenza for 9,547 unique children. A minority of children (4.1%) were included in more than one season. A total of 1,280 (12.8%) individuals had specimens that tested positive for influenza (range across seasons: 11.5%-14.5%). Across seasons, 1,151 (11.5%) children were classified as having received at least one influenza vaccination (range across seasons: 10.0%-13.8%), with 6.0% fully vaccinated and 5.6% partially vaccinated. Of the 882 individuals who tested positive for influenza A, 451 were subtyped, with 163 positive for A/H1N1 only, and 286 positive for A/H3N2 only. There were 402 individuals who tested positive for influenza B (including 5 influenza A/B coinfections); lineage information was not available. Influenza-positive children were more likely to be unvaccinated, older, reside in urban areas, and to be admitted during the peak influenza month (Table 1 and S4 Table). Children who had a CCC or asthma, a past intensive care unit (ICU) admission, or a history of low birthweight or preterm birth were more likely to be vaccinated (Table 2 and S5 Table).
The overall adjusted VE against laboratory-confirmed influenza hospitalization was 60.4% (95%CI, 44.0%-72.1%) for fully vaccinated children and 39.2% (95%CI, 16.6%-55.6%) for partially vaccinated children (Table 3; unadjusted results in S6 Table). Full vaccination offered statistically significant protection for three of the four influenza seasons studied, whereas partial vaccination offered protection for two seasons. VE for any vaccination was higher for children aged 24-59 months relative to those aged 6-23 months (p-value for interaction = 0.012). VE appeared higher for fully vaccinated children in all sub-analyses except against A/H3N2; however, the only statistically significant difference in VE by vaccination status was observed against influenza B (p = 0.03). Partial vaccination was not significantly protective against influenza for the 2011-12 or 2012-13 seasons, against A/H1N1 or influenza B, for children aged 6-23 months, or for children with asthma. VE estimates were unchanged when we restricted to children with specimens collected during an ARI-coded hospitalization and when we included any comorbidity in our model (Table 3). When we corrected for misclassification of our exposure measurement based on the most plausible estimates of sensitivity (77%) and specificity (98%), VE for any vaccination changed from 50.8% (95%CI, 37.6%-61.2%) to 60.7% (95%CI, 47.9%-70.6%). We performed these sensitivity analyses on a range of estimates to determine their potential impact on VE Vaccine effectiveness against influenza hospitalizations among young children in Ontario, Canada (S1 Fig). Influenza vaccination was not associated with hospitalization for non-influenza respiratory viruses. When examining the influence of prior vaccinations status in children aged 24-59 months, VE estimates for those vaccinated in the current season only versus two sequential seasons were similar (Fig 1a), though there was variation by year (Fig 1b). In the 2010-11 and 2011-12 seasons and for A/H3N2, VE appeared higher for those vaccinated in the current season only, though the reverse was seen in the 2012-13 season; however, confidence intervals were wide due to small numbers. Confidence intervals were also wide when examining residual protection based on prior season vaccination only.
For patients who had an ARI code associated with their hospitalization, 46% were tested for influenza. Among these ARI-coded hospitalizations, 12.6% of patients who were tested for influenza were vaccinated, compared to 12.4% of patients who were not tested for influenza (p = 0.56), confirming the absence of an association between influenza testing and vaccination Vaccine effectiveness against influenza hospitalizations among young children in Ontario, Canada status. Using the 95% of our sample that were tested for at least one other respiratory virus, 32.2% of unvaccinated children tested positive for another respiratory virus compared to 32.1% of children with any vaccination (p = 0.96), confirming the absence of an association between vaccination status and non-influenza respiratory viruses. This assumption held for fully, partially, and unvaccinated children, as well as between age groups and across seasons.
We found higher VE against A/H3N2 in partially vaccinated children relative to those fully vaccinated, which is consistent with previous work [6,37]. This may be because children are defined as fully vaccinated either through receipt of two doses in their first season or one dose if they had been vaccinated in any prior season. Differentiating the effect of partial vaccination versus repeated vaccination is challenging in this age group as we may be inherently studying the effect of repeated vaccination when examining fully vaccinated children. For A/H3N2, we estimated higher VE for those vaccinated in the current season only, which is equivalent to partial vaccination in certain circumstances. In the future, studies focusing on this age group should differentiate between the two types of full vaccination when estimating VE.
Among children aged 24-59 months, we found no difference in VE between those vaccinated in the current season only compared to those vaccinated in both prior and current seasons. However, we were limited by our sample size from evaluating the impact of repeated vaccination by subtype. This analysis could also only consider vaccination in two consecutive seasons rather than serial vaccination, although the number of vaccines in this age group would be limited. Our results may not generalize to older individuals with longer histories of both influenza and influenza vaccine exposure.
Our study has several strengths. We were able to test some of the assumptions that are routinely mentioned in test-negative studies but never evaluated. Specifically, we confirmed that vaccination status was not associated with influenza testing and that influenza vaccination was not associated with testing positive for other respiratory viruses. Pooling data across several years allowed for the evaluation of VE in selected subgroups of interest, and the nature of the vaccination data permitted evaluation of the impact of repeated vaccination in a population with limited previous vaccine exposure. The size of our study population was enhanced by including numerous types of high-specificity laboratory tests. While values for sensitivity and specificity may vary by testing method, in TND studies with laboratory-confirmed outcomes, specificity is more important in terms of bias [38].
This study had several limitations. Symptom onset date was only available for the minority (16%) of specimens, but since children shed higher levels of virus for longer periods of time and we restricted to those tested within three days of hospital admission, absence of symptom onset date is less likely to be an issue in this age group [39]. Further, there was no strict case definition for initiating influenza testing, though this would not be expected to be different between test-positives and test-negatives. Further, 88% of individuals with specimens tested were collected during an ARI-coded hospitalization, which may serve as a proxy for testing criteria. Our VE estimates were unchanged when restricting the sample to individuals with specimens collected during ARI-coded hospitalizations. Vaccination status was determined from physician billing claims; children who received a vaccination at public health clinics may have been misclassified. Children in this age group are not eligible for vaccination through pharmacists and the majority (77%) of children in this age group receive influenza vaccination through physicians [14]. We would not expect differential misclassification by influenza case status, and our VE estimates would therefore be conservative. While we may have underestimated coverage relative to other studies, our data reflect vaccination status relative to the date of influenza testing in a hospitalized cohort, and do not reflect estimates of vaccine coverage for the influenza season [14]. Further, the vaccinated proportion is similar to that of other Canadian studies [29][30][31], though differences in setting and vaccination data sources may explain any residual differences. Differentiating TIV, QIV, and LAIV was not possible with the available data. However, neither LAIV nor QIV were publicly funded in Ontario until the 2015-16 influenza season, therefore most children in this study would have received TIV. We were limited by low influenza vaccine coverage, which impacted our power to detect significant VE in certain groups, especially when considering the importance of evaluating VE by full and partial vaccination status [40]. Coverage appeared low in our sample relative to other estimates, perhaps because some vaccinations may have not been documented or because previous survey results over-estimated coverage [14]. Compared with surveillance-based methods, this approach is insufficiently timely to produce mid-season or end-of-season results to inform program planning. Last, the subtype-specific estimates were limited because only half of individuals positive for influenza A had specimens that were subtyped.

Conclusion
Our study demonstrates VE against laboratory-confirmed influenza hospitalizations for children aged 6-59 months. Despite variation across subgroups, we observed substantial protection for vaccinated children, particularly if fully vaccinated. These results support current recommendations to promote vaccination in this high-risk group.

Disclaimers
This study was supported by the Institute for Clinical Evaluative Sciences (ICES) and Public Health Ontario (PHO), which are funded by annual grants from the Ontario Ministry of Health and Long-Term Care (MOHLTC). The opinions, results, and conclusions reported in this paper are those of the authors and are independent from the funding sources. Parts of this material are based on data and information compiled and provided by the Canadian Institute of Health Information (CIHI) and by Cancer Care Ontario (CCO). However, the analyses, conclusions, opinions, and statement expressed herein are those of the authors, and not necessarily those of CIHI or CCO. No endorsement by ICES, PHO, MOHLTC, CIHI, or CCO is intended or should be inferred.