Introduction

Influenza virus infections are a substantial contributor to respiratory morbidity and mortality, with the highest burden of severe disease experienced by those aged <5 y and ≥65 y [1–3]. Until recently, however, estimates of influenza burden have been largely obtained from resource-rich settings with temperate climates with clearly defined influenza seasonality, with few estimates available from lower-income countries [4–6]. This lack of information on influenza burden in resource-limited settings has hampered informed consideration of implementation of preventive measures such as vaccination.

In the last 10 y, however, the global expansion of influenza surveillance and laboratory capacity for influenza testing by reverse transcription PCR has led to dramatic increases in testing in settings with previously sparse data. Many countries worldwide now perform hospital-based influenza surveillance among patients hospitalized with severe acute respiratory illness (SARI) [7–9]. These systems have proven useful in tracking influenza activity among hospitalized patients, but only a fraction of the collected data has been used to estimate the burden of influenza-associated hospitalizations. A 2011 meta-analysis using data from 16 population-based research sites and published literature estimated that there were 1 million cases of influenza-associated SARI episodes globally in children aged <5 y old in 2008. In this study, the per capita rate of severe influenza illness in developing countries was at least double that of industrialized countries [6]. This rate-based approach was mainly derived from sites with a limited population at risk under surveillance, and therefore represented only a small fraction of hospitals conducting surveillance for severe influenza disease globally.

We aimed to update the previous influenza burden estimates in young children, making full use of updated and expanded surveillance data from the past decade (2003–2012), both before and after the 2009 pandemic influenza emergence. We further aimed to extend previous studies by providing global estimates of the prevalence of influenza among acute lower respiratory infection (ALRI) hospitalizations among both younger (<5 y) and older (5–17 y) children.

Methods

We aggregated data from all eligible published etiologic studies of influenza-associated respiratory illness among hospitalized children, which we supplemented with data from a working group of inpatient surveillance systems worldwide. We then calculated a final pooled estimate of the proportion of tested samples that were positive for influenza by reverse transcription PCR among children aged <18 y (referred to as proportion or percent positive), using age-group-specific random-effects log-binomial regression models. Finally, we applied the aggregate pooled proportion positive among children <1 and <5 y to age-specific denominators of global hospitalizations for ALRI among these two age groups [10] to obtain the number and rate (per 100,000 children per year) of pediatric influenza-associated hospitalizations, by World Health Organization (WHO) region and United Nations (UN) country development status.

Results

The systematic literature search identified 38,006 unique records from the nine scientific literature databases, of which 957 full-text articles were reviewed, and 108 included in the descriptive analysis (Fig 1) (full list of included articles provided in S1 Appendix). In addition, 37 surveillance datasets for periods ranging from 1 to 7 y, each with data from 1–49 inpatient facilities, were provided by the GRIPP Working Group. Combined, the literature search and working group resulted in a total of 145 unique data sources from 350 sites in 60 countries across all WHO regions, including southern and northern hemisphere temperate climates as well as tropical and arid regions of Asia and Africa. The two data sources combined covered a 31-y period from 1982 to 2012. More than half (55/108, 51%) of the published studies had a 3/3 quality score. Compared to published articles, surveillance datasets were more recently collected (median start year 2009 versus 2002), of longer duration (median duration 3 y versus 2 y), more likely to be PCR-based (41% versus 28%), and more likely to use the SARI case definition (84% versus 0%). Additionally, surveillance datasets were also more commonly from low-income countries compared to published articles (41% versus 6%, respectively; Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow diagram for systematic review process.

https://doi.org/10.1371/journal.pmed.1001977.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of published studies and surveillance data sources about influenza-associated respiratory illness among hospitalized children, 1982–2012.

https://doi.org/10.1371/journal.pmed.1001977.t001

The crude median percent of respiratory samples that were influenza positive among patients aged 5–17 y was more than double that among those <5 y (15%, interquartile range [IQR] 10%–22%, versus 6%, IQR 3%–9%, p < 0.001) and was also significantly higher in surveillance data than in published articles (9%, IQR 6%–12%, versus 5%, IQR 3%–9%, p < 0.01) (Table 2). The median percent positive was also significantly higher in 2009, when pandemic influenza A(H1N1)pdm09 virus emerged, and in following years, compared to before 2009 (13% in 2009, IQR 6%–20%, versus 9% after 2009, IQR 5%–11%, and 5% before 2009, IQR 3%–9%, p < 0.001). The majority of the datasets (57%) comprised results from PCR diagnosis. Influenza positivity was significantly lower among the datasets that used immunofluorescence alone, as compared to the datasets that used other methods such as PCR, culture, or a combination of diagnostic tests (3% versus 7%, p < 0.01). The frequency of influenza detection by any assay was not significantly associated with case definition used, income level, or WHO region. To account for the possibility that differences in influenza positivity by age group were due to confounding factors such as differences in PCR use or case definition applied, we analyzed these factors by age group and found no association. The studies that did not use PCR were similar to those that did use PCR by age group and timeframe of data reported. However, studies from the Americas and the Western Pacific (particularly South America and China) were less likely to use PCR, with immunofluorescence as a more common diagnostic methodology. Among published articles, there was also no difference in influenza positivity by quality score.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Crude proportion of respiratory samples from hospitalized children testing positive for influenza by age group, study design, and population.

https://doi.org/10.1371/journal.pmed.1001977.t002

Due to steady increases in year-round hospital-based influenza surveillance over the past decade, the number of available datasets substantially increased from 2006 (n = 2) to 2011 (n = 22). The overall median influenza percent positive increased sharply from 5% in 2008 to 13% in 2009, largely driven by pandemic H1N1, the predominant subtype that year. The median total percent positive (for any influenza type/subtype) dropped substantially in 2010 to 9% and remained at 8% in 2011, as influenza A(H1N1)pdm09 circulated at lower levels and influenza A(H3N2) and B predominated (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Boxplot of the proportion of pediatric (0–17 y of age) respiratory samples testing positive for influenza virus among GRIPP datasets by year and virus type/subtype.

Data are for years with more than one dataset providing testing results by virus subtype. Unsubtyped influenza A viruses are included in influenza A totals, but not shown separately. Boxplot excludes outside values.

https://doi.org/10.1371/journal.pmed.1001977.g002

The pooled meta-analyses with only PCR-confirmed data included 63 datasets, each with 1 to 6 y of data, from 41 countries. The pooled percent positive for influenza among children hospitalized with respiratory illness varied from 4.8% (95% CI 3.3%–6.9%) among those <6 mo to 16.4% (95% CI 13.6%–19.8%) among those 5–17 y, and was 9.5% (95% CI 8.1%–11.0%) overall among children <18 y. Among children <5 y, the pooled estimate was 7.4% (95% CI 6.2%–8.8%). For this age group, we also stratified the pooled estimate by region and country development status. We found the highest percent positive in Southeast Asia (8.5%, 95% CI 6.7%–10.8%) and the lowest in the Americas (4.6%, 95% CI 2.8%–7.4%), and the percent positive was lower in industrialized countries (5.9%, 95% CI 4.6%–7.5%) than in developing countries (7.7%, 95% CI 6.4%–9.3%), although these differences were not statistically significant. There were no significant differences by country income status. Post hoc sensitivity analysis of the effect of outliers (>95th percentile values) on the global estimates showed no difference.

Finally, by applying the age-specific pooled proportion to a denominator of global hospitalized ALRI episodes, we estimated the absolute number of influenza-associated hospitalizations among children <5 y to be 870,000 (95% CI 610,000 to 1,237,000), for a per capita rate of 135/100,000 children/year (95% CI 95–193) (Table 3). The influenza-associated hospitalization rate in developing countries was 3-fold that in industrialized countries, and ranged from 174/100,000 children/year in Africa to 71/100,000 children/year in the Americas. Among children <1 y, we estimated 374,000 (95% CI 264,000 to 539,000) influenza-associated hospitalizations per year worldwide.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Pooled estimates of global pediatric influenza-associated hospitalizations per year, by age group, development status, and WHO region, among studies using PCR diagnostic testing.

https://doi.org/10.1371/journal.pmed.1001977.t003

The incidence of influenza-associated ALRI among children 0–5 mo was approximately 1.6 times higher than that of children 6–11 mo. Applying this rate ratio to the total influenza-associated hospitalizations for children <1 y, we estimate 228,000 (95% CI 150,000 to 344,000) influenza-associated hospitalizations per year among children aged less than 6 mo.

Discussion

We used influenza surveillance data from 350 sites in 60 countries to estimate that 10% of global respiratory hospitalizations in children under 18 y worldwide were associated with influenza. This proportion increased by age, with the highest percentage found among school-aged children (5–17 y). We further estimated that influenza causes approximately 374,000 respiratory hospitalizations per year in children <1 y of age (including 270,000 among those less than 6 mo) and 870,000 respiratory hospitalizations per year in children <5 y of age, with the greatest impact in developing countries.

Although we did not identify any significant trends in the overall proportion positive by WHO region in the crude analysis, the final pooled estimates among children <5 y from the Americas differed from the global estimates (4.6%, 95% CI 2.8%–7.4%, versus 7.4%, 95% CI 6.2%–8.8%). This lower percentage positive for influenza may be related to the use of influenza vaccine in the region, which has been on the rise since 2004 [15]. However, vaccine coverage data among children in the region are limited, and these trends may also be a chance finding due to the small number of study sites in this region reporting results of year-round influenza testing.

Our findings on influenza burden among children <5 y are consistent with a previously published estimate using fewer datasets and a different methodological approach [6] (870,000 annual influenza-associated hospitalizations worldwide in our analysis versus 911,000 estimated by Nair et al., with overlapping confidence intervals) and demonstrate the value of using non-population-based sentinel surveillance data for disease burden estimation. Our analysis also found a substantially higher proportion of influenza-associated hospitalizations among school-aged children (5–17 y), who are an important group for transmission of influenza virus [7,16,17] and who were not included in previous global analyses. While the prevalence of influenza as a proportion of respiratory hospitalizations is lower among younger children (<5 y and particularly <1 y) than among older children, this is largely due to a high frequency of other serious respiratory pathogens such as respiratory syncytial virus (RSV) [18]. These young children experience the highest rate of total respiratory and influenza-associated hospitalizations [2,19]. Global pediatric mortality estimates suggest that the proportion of deaths caused by influenza is highest among those aged 1–12 mo (2.8% of all deaths in this age group, worldwide) [20]. Further, severe influenza infection with non-respiratory clinical presentation is not captured in our analysis, so our calculation of the total burden of influenza-associated hospitalization among the youngest children (<1 y) is likely an underestimate, as non-standard presentations (such as fever without classic respiratory signs) are known to occur in this age group [21].

Other pathogens, particularly Streptococcus pneumoniae and RSV, are also important causes of respiratory hospitalizations in those <5 y. Globally, it is estimated that RSV is associated with 25% of all ALRI in children <5 y, and, like influenza, RSV prevalence can vary substantially by season and across geographic locations [18]. S. pneumoniae and Haemophilus influenzae type B were associated with 18.3% and 4.1% of severe ALRI episodes globally among children <5 y in 2010, respectively [22], although their relative contributions to pediatric respiratory hospitalizations may vary depending on their inclusion in pediatric vaccine programs. However, a single ALRI episode may be caused by multiple pathogens, as well. For example, severe bacterial pneumonia can be a sequela of influenza infection [23]; pediatric hospitalizations for invasive pneumococcal disease in the US have been shown to increase during the 2-wk window following peak circulation of influenza, as well as other respiratory viruses [24].

Our finding of an increased impact among children of pandemic influenza compared to seasonal influenza (median estimate percent positive of 13% among children of all age groups in 2009 compared to 5% before 2009 and 9% after 2009) confirms previous findings of the effect of the pandemic on ALRI hospitalizations worldwide. In the United States, a 5-fold rise was reported in the rate of laboratory-confirmed influenza hospitalizations among children 5–17 y of age during the 2009–2010 influenza pandemic compared to non-pandemic influenza seasons [25]. Similar trends have been seen in India, where hospitalization rates due to pandemic influenza infection were particularly high among those 5–29 y, compared to other age groups [26]. Global mortality data from the 2009 pandemic suggest that the actual number of deaths was 15 times greater than laboratory-confirmed estimates, with 80% occurring in those younger than 65 y and 22% occurring in children 0–17 y [27].

Several important limitations should be considered when interpreting our findings. Differences in hospitalization practices, applications of case definitions, influenza testing practices (including sampling method), and factors such as time from symptom onset to specimen collection could make detection of influenza more or less likely and therefore bias the percent positive outcome. Our estimate of the burden of severe respiratory illness due to influenza is an underestimate of total burden for several reasons. First, our approach does not allow estimation of severe respiratory illness in individuals who did not present to hospital, which is particularly a problem in resource-limited settings with poor health care access, where hospitalization rates may be driven by proximity to a health care facility, limiting the generalizability of hospital-based burden estimates. Second, individuals with influenza virus may have stopped shedding by the time they presented to hospital and were tested, resulting in an underestimation of the true percent positive. Lastly, we assumed that influenza virus detected in individuals hospitalized for respiratory disease was causal for the hospitalization episode. While influenza is rarely found in well individuals [28], co-infections of influenza with bacteria and other viruses may result in a more severe illness, which is not explored in our analysis.

Additionally, year-by-year variability of influenza percent positive and external factors, such as co-infections and vaccination coverage, may affect influenza positivity. This year-by-year variability is better captured in the GRIPP data than in the published data, which was often not analyzable by both year and age group. Data from 2010 onward also include influenza A(H1N1)pdm09, which may have had a greater impact on immunologically naïve children, compared to other influenza viruses, during this period, resulting in an overestimate of annual influenza burden. This may be particularly the case during 2010, when WHO officially transitioned from pandemic to post-pandemic phase, especially in settings such as West Africa, where influenza A(H1N1)pdm09 only began circulating in 2010 [29]. However, no significant difference was seen in the overall pooled estimate when 2010 data were excluded, indicating that influenza A(H1N1)pdm09 activity in 2010 is not likely to impact our overall findings on proportion positive for seasonal influenza.

Our findings expand knowledge of the impact of severe influenza among children <1 and <5 y, and create an evidence base for both younger (<6 mo) and older (5–17 y) children, for whom, to our knowledge, no global estimates of influenza disease burden have been published to date. Countries considering possible influenza vaccination programs for children and/or pregnant women can use our estimates as inputs for vaccine impact and cost–benefit models. Our data may also stimulate further research into the development of effective influenza vaccines for young children. Several recent changes have made influenza vaccination a more realistic goal in settings that have not previously considered influenza vaccination programs, including 2012 WHO recommendations identifying pregnant women as the most important target group for influenza vaccination [30], as well as recent expansions in vaccine manufacturing and increasing investment into public health among middle-income countries [31]. As demand and supply grow, vaccines may drop in cost, and targeting of pediatric populations through childhood immunization programs such as the WHO Expanded Programme on Immunization could further reduce the cost of implementation of an influenza vaccine program. Maternal immunization is another strategy that would help reduce influenza burden in children less than 6 mo [32]. The development of improved influenza vaccines is needed, including products that require only one dose in young children or provide longer-term protection against a broader array of influenza viruses, such as live-attenuated and adjuvanted vaccines, as well as universal influenza vaccines that protect against multiple influenza strains. Further research is also needed to better define the burden of influenza in young children, especially those presenting with non-respiratory symptoms, possibly via approaches such as vaccine probe studies [33].

Supporting Information

Acknowledgments

The authors thank the influenza surveillance teams at each of the collaborating institutions for generating the data used in this analysis. We also thank colleagues at the Stephen B. Thacker CDC Library for assistance in accessing articles for the literature searches.