Contribution of influenza viruses to medically attended acute respiratory illnesses in children in high-income countries: a meta-analysis.

Aim: The burden of disease in children attributable to influenza viruses is difficult to quantify given the similarity of symptoms caused by infection due to influenza and other viruses. This uncertainty impacts clinical decision‐making and estimates of burden. We aimed to systematically review the literature to determine the proportion of healthy children presenting for health care with an acute respiratory illness (ARI) who have laboratory‐confirmed seasonal influenza (PROSPERO ID#CRD42014013896). Method: We searched Ovid MEDLINE, EMBASE, Scopus, and references of included articles. We included studies that used polymerase chain reaction methods to test for influenza in healthy children aged ≤5 years who presented for health care in high‐income countries with an influenza‐like or ARI. A standardized form was used to collect data on positivity and other relevant study elements. Results Seventeen studies covering 12 different influenza seasons were included. The proportion of influenza positivity ranged from 11% to 56%. Subgroup analyses were performed by influenza season, continent, healthcare setting, age group, and vaccination status. Higher influenza positivity was reported among children aged 3–5 years compared with children aged ≤2 years, and for unvaccinated children. Conclusion The minority of healthy patients aged ≤5 years with medically attended influenza‐like or acute respiratory symptoms have laboratory‐confirmed influenza virus infection, although this varied by influenza season. Prevention efforts should be targeted accordingly. Statement Most influenza‐like illnesses are not laboratory‐confirmed and have similar clinical presentations. Consequently, the true contribution of influenza to acute respiratory infections in children remains uncertain. Our systematic review estimates that this proportion ranges from 11% to 56%. This finding can help both clinicians and public health professionals target prevention.


| INTRODUCTION
Annual epidemics of seasonal influenza continue to cause substantial morbidity and mortality among children. [1][2][3] Previous studies have used non-specific outcomes such as influenza-like illness (ILI) or acute respiratory illness (ARI) to estimate disease burden and vaccine effectiveness. 4 However, other respiratory viruses cause similar symptoms, making it challenging to distinguish infections due to influenza viruses from others. 5 This uncertainty impacts clinical decision-making and the estimation of the true burden of influenza, as well as vaccine effectiveness. 6,7 Laboratory testing identifies the specific virus responsible for the illness, but is not routinely performed in clinical practice; thus, the true contribution of influenza in the pediatric population remains uncertain.
While other reviews have examined the contribution of influenza viruses to burden in sub-Saharan Africa or to acute lower respiratory infections (ALRI), 8,9 this study fills a gap in knowledge of the contribution to ILI/ARI in pediatric populations. To our knowledge, no other published systematic reviews have quantified the contribution of influenza viruses to medically attended respiratory illness in children in high-income countries. The objective of this study was to systematically review the published peer-reviewed literature evidence to determine the proportion of healthy children aged ≤5 years presenting for health care with ILI/ARI who have laboratory-confirmed seasonal influenza.

| Search strategy
We developed a detailed search strategy in consultation with a scientific librarian to identify articles related to children aged ≤5 years in high-income countries (population) who had an ILI/ARI (exposure) and were tested for influenza with polymerase chain reaction (PCR) methods (outcome). This strategy was applied to Ovid MEDLINE, EMBASE, and Scopus (from inception to August 6, 2014). Search terms included "influenza," "flu," "polymerase chain reaction," "PCR," "laboratoryconfirmed," "child," "infant," and "adolescent." Similar terms were combined with an "OR" operator and distinct terms linked with an "AND" operator. The full search strategy is outlined in File S1. Reference lists of included studies were also searched. The search strategy, along with the inclusion and exclusion criteria, was outlined in a registered study protocol (PROSPERO ID #CRD42014013896). 10 The results of our systematic review and meta-analysis are reported according to the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. 11

| Study selection
We included experimental, observational, and surveillance studies that used PCR methods to test for influenza infection in healthy children aged ≤5 years in high-income countries who presented for health care with ILI/ARI, as defined by individual studies (see Table 1 for definitions). High-income categorization was based on the definition developed by the World Bank. 12 No language restrictions were placed on the initial review, although only full-text articles available in English, French, or Spanish were included. We excluded studies in which PCR methods were not used or not used exclusively, studies that tested all participants (not just those with ILI/ARI), studies in which all participants were >5 years of age or which did not report results for this age group separately, studies of children with underlying medical conditions, studies that did not report each influenza season separately, studies that reported outbreak investigations, studies reporting interim estimates for an ongoing influenza season, studies that focused on diagnostic techniques, genetic characterization, clinical characteristics of patients, or describing the use of surveillance systems, studies that only reported on the 2009 A(H1N1) pandemic season, and studies of co-infections that did not report influenza separately. Conference proceedings and abstracts, non-peer-reviewed reports, reviews, letters, editorials, case reports, and case series were excluded.

| Review process
The titles and abstracts of retrieved articles were reviewed for general relevancy (SB). Studies meeting the predefined inclusion criteria were included for full-text review, performed independently by two reviewers (SB and TSH). The inter-rater reliability for studies selected for inclusion in the review was good (κ=0.79, 95% CI, 0.64-0.94), and any disagreements (n=7 studies) were resolved through discussion and consensus.

| Data extraction
We used a standardized data extraction form to collect data on the following study elements: study publication details (author, journal, year of publication), study design, influenza season(s) and period of circulation, age groups, country of study, study setting (hospital, emergency department, physician office), outcome with case definition, number of participants studied, and number testing positive for influenza.

| Outcome
The outcome of interest was the proportion of swabbed children testing positive for influenza after presenting for health care with ILI/ ARI.

| Assessment of risk of bias
We used a Newcastle-Ottawa scale adapted for cross-sectional studies by Herzog et al. 13 to assess the quality of included studies as they related to our outcome: the proportion tested who were influenza positive. Given the nature of our study question, we chose a tool that related to our outcome over one that assessed the quality of the study in general. This scale evaluated the quality of the study as it related to domains of selection, comparability, and outcome. We extracted a proportion from the studies that was not directly reported; therefore, the statistical test criterion in the outcome domain was deemed not applicable to this study.

| Data synthesis and analysis
A stratified meta-analysis was planned a priori due to anticipated heterogeneity of the estimates. Subgroup categories were as follows: influenza season, continent, age group (≤2 years, 3-5 years, or ≤5 years), study setting (hospital, emergency department, or physician office), and vaccination status (fully vaccinated, partially vaccinated, or unvaccinated). We analyzed strata with at least two eligible studies.
A post hoc analysis was performed to examine positivity by timing of data collection, focusing on whether testing was restricted to periods when influenza viruses were in circulation, as defined by individual studies.

Meta-analyses were performed in MetaXL (Version 2.0, EpiGear
International Ltd, Queensland, Australia) with the inverse variance heterogeneity (IVhet) method. 14 Unlike a random effects (RE) estimate, this method maintains the inverse variance weights of individual studies and provides a more conservative confidence interval. [14][15][16] Given the high levels of heterogeneity and the wide range of sample sizes in this study, we chose the IVhet model for the primary analysis. [15][16][17] We repeated the pooled analysis with the RE model using the DerSimonian and Laird method with double arcsine transformation to stabilize the variance, 15 in order to compare the results. The analysis was run in R Statistical Software using the meta package to provide a prediction interval to estimate the range of values that future studies can be expected to fall within. 18 Forest plots were created for each subgroup in order to examine clinical (e.g., age, setting, influenza season) and methodological (e.g., study design, study quality) heterogeneity. Statistical heterogeneity was assessed using the Cochran Q statistic (with P<.10 indicating statistical significance), as well as the I 2 statistic. 19 We calculated the proportions (with corresponding 95% confidence intervals) of PCRconfirmed influenza among children presenting with ILI/ARI overall and for each subgroup.

| RESULTS
A total of 3474 titles and abstracts were screened and 141 were selected for full review (Fig. 1). Of these, 17 studies satisfied the inclusion criteria and were included ( and two studies did not report the type of specimen collected. 25,28 There were four surveillance studies, 27 [20][21][22][23][24]26,29,[33][34][35] while the remaining two studies did not explicitly state this. 28,30 In evaluating risk of bias using the modified Newcastle-Ottawa scale, we found that of the 17 included studies, five were at low risk (4-5 stars), 11 at moderate risk (2-3 stars), and one at high risk in the selection domain (Table S1 and Fig. S1). Few studies reported or compared non-respondents or had adequate sample size for the included age group. In the comparability domain, four studies were at low risk (2 stars) and 13 studies at moderate risk (1 star). As this study was restricted to patients tested with PCR methods, all studies were at low risk in the outcome domain.
The pooled proportion across all studies indicated that 20% (95% CI, 15-25) of healthy children aged ≤5 years seeking health care with an ILI or ARI tested positive for influenza (Fig. 2)

| Influenza Season
The

| Continent
The proportion of influenza positives ranged from 18% to 29% across continents, although Asia was not included as it was only represented by one study. North America had the lowest proportion positive (18%, 95% CI, 12-25) (n=5 studies) while the European region had the highest (29%, 95% CI, 21-37) (n=8 studies).

| Vaccination status
One study restricted participants to those vaccinated, 20  The I 2 statistic is used to quantify the degree of heterogeneity within each subgroup and the Cochran Q statistic used to test for statistical significance of this heterogeneity. 19 | 451

| DISCUSSION
The results of this systematic review and meta-analysis suggest that seasonal influenza viruses contribute to approximately 20% of medically attended respiratory illnesses in healthy young children in highincome countries, with individual study estimates ranging from 11% to 56%. However, there was significant heterogeneity in this pooled proportion and the estimate varied by influenza season, region, study population, age group, healthcare setting, and vaccination status.
Our study demonstrates that children aged 3-5 years had a higher proportion of influenza positivity than children ≤2 years of age, and the proportion for those children aged ≤5 years (i.e., not stratified further by age) fell between these two groups. While symptom severity and risk of hospitalization are greater for children aged ≤2 years, 5,38-40 including those <6 months of age who are not eligible for vaccination, the percentage positive is higher in older children aged 3-5 years. Preschool children have the highest transmission potential, 41,42 and their interactions in school and daycare settings may increase their risk of exposure to influenza, which may help to explain the higher rate of positivity for this age group. 43 Additionally, a lower threshold of disease severity for younger children may result in seeking health care, therefore increasing the denominator in this age group, which could lead to lower overall positivity; however, this mechanism could not be evaluated.
We are not aware of any prior reviews published on this topic for children in high-income countries. One meta-analysis of respiratory infections found that globally in 2008, 13% of pediatric ALRI were attributable to influenza viruses; however, this study had no geographic limitations and considered ALRI as opposed to ILI/ARI. 9 Another systematic review examined the burden of seasonal influenza, including some PCR-confirmed outcomes, and found the percent of outpatient ARI patients that tested positive for influenza ranged between 1% and 25%, but the study included all ages and was limited to sub-Saharan African countries. 8 Finally, another review described the burden of seasonal influenza in children, but it was not performed systematically nor did it include meta-analysis. 44 Our study demonstrates consistent findings with this review in noting a higher burden of influenza in outpatient settings.
Our study provides a better understanding of pediatric influenza epidemiology by estimating the contribution of influenza viruses to medically attended respiratory illnesses in young children. 45,46 Having better data on the contribution of influenza to ILI in this population can help to estimate the risk of disease in different populations; inform immunization uptake, 47 programs, and prevention strategies; evaluate treatment plans; and plan for seasonal epidemics and potential pandemics. 45 The WHO has noted that this information can guide the allocation of health resources and the establishment of thresholds of disease severity. 46 These estimates can also be used in mathematical models to predict influenza burden, [48][49][50] cost-effectiveness studies, 51 and clinical decision-making. Nevertheless, this estimate should be interpreted with caution.
A strength of this study was the rigor of the methods applied at all stages, including both the search for articles and the meta-analysis.
We employed a broad search strategy in order to capture all relevant literature related to influenza detection in children with ILI/ARI. The study was strengthened by choosing PCR as the method of laboratory confirmation, due to its high sensitivity and specificity, 63 as well as its improved detection over viral culture by 2%-13%. 64 The study was also strengthened using the IVhet model, as opposed to the more commonly used RE model. 65 While the pooled proportion of positivity reflected in the former was lower than the latter, it was not influenced by small studies with high rates of positivity that may not reflect the true contribution of influenza to respiratory illnesses.
This study also had some limitations, the first of which related to the search and selection process of the studies. The search included studies found through three databases, but did not include any gray literature. This limits our findings as surveillance data are often included in non-peer-reviewed reports, such as through the Public Health Agency

| CONCLUSION
This study quantified and investigated the proportion of medically attended acute respiratory illnesses attributable to influenza in pediatric populations in high-income countries. While the positivity was fairly consistent across seasons and locations, the remaining variability should be investigated and the pooled estimate should be interpreted with caution. Although only a minority of acute respiratory illnesses are caused by influenza viruses, influenza is still an important contributor to morbidity given the substantial number of respiratory illnesses in the pediatric population. This burden of disease can be reduced through seasonal influenza vaccination and other prevention strategies to reduce the risk of all respiratory viruses.

FUNDING SOURCE
Sarah Buchan's doctoral training is supported by a Public Health

Agency of Canada and Canadian Institutes of Health Research
Influenza Research Network (PCIRN) Trainee Award.