Hospitalizations associated with respiratory syncytial virus (RSV) and influenza in children, including children diagnosed with asthma

Abstract

Background

There is uncertainty about the burden of hospitalization associated with respiratory syncytial virus (RSV) and influenza in children, including those with underlying medical conditions.

Methods

We applied previously developed methodology to HealthCare Cost and Utilization Project (HCUP) hospitalization data and additional data related to asthma diagnosis/previous history in hospitalized children to estimate RSV and influenza-associated hospitalization rates in different subpopulations of US children between 2003–2010.

Results

The estimated average annual rates (per 100,000 children) of RSV-associated hospitalization with a respiratory cause (ICD-9 codes 460–519) present anywhere in the discharge diagnosis were 2381 (95% CI(2252,2515)) in children less than a year of age; 710.6(609.1,809.2) (age 1y); 395(327.7,462.4) (age 2y); 211.3(154.6,266.8) (age 3y); 111.1(62.4,160.1) (age 4y); 72.3(29.3,116.4) (ages 5–6y); 35.6(9.9,62.2) (ages 7–11y); and 39(17.5,60.6) (ages 12–17y). The corresponding rates of influenza-associated hospitalization were lower, ranging from 181(142.5,220.3) in age <1y to 17.9(11.7,24.2) in ages 12–17y. The relative risks for RSV-related hospitalization associated with a prior diagnosis of asthma in age groups under 5y ranged between 3.1(2.1,4.7) (age <1y) to 6.7(4.2,11.8) (age 2y); the corresponding risks for influenza-related hospitalization ranged from 2.8(2.1,4) (age <1y) to 4.9(3.8,6.4) (age 3y).

Conclusions

RSV-associated hospitalization rates in young children are high and decline rapidly with age. There are additional risks for both RSV and influenza hospitalization associated with a prior diagnosis of asthma, with the rates of RSV-related hospitalization in the youngest children diagnosed with asthma being particularly high.

Introduction

Robust estimation of disease severity is essential for planning prevention programs. However, there is uncertainty about the rates of hospitalization associated with respiratory syncytial virus (RSV) and influenza in different age groups of children, as well as in children with underlying medical conditions such as asthma. Studies of hospitalizations associated with respiratory virus infections in select communities (e.g. [1–5]) have documented high rates of RSV-associated hospitalization in full- and pre-term infants ([1,2]) and young children [1], as well as high rates of influenza-associated hospitalization in children during certain influenza seasons (e.g. [3] vs. [4]). Moreover, these studies reveal a high frequency of underlying medical conditions (particularly asthma) among children hospitalized with RSV and influenza virus infections ([1,5–7]). However, community-based studies can suffer from limitations such as underdetection of viral infections (see also Discussion), and suboptimal sample sizes.

Estimates of hospitalization rates associated with respiratory viruses can also be derived from studies using hospital discharge data. Limitations of discharge diagnosis studies may include both underascertainment and misclassification of specific respiratory viral etiologies. Strategies for correction of missing or misclassified diagnoses have been developed to produce burden estimates [8]. Statistical inference based on regression models for time series of severe outcomes such as deaths or hospitalizations is a common approach for estimating the contribution of respiratory viruses to the burden of these outcomes [9–14]. However, various assumptions underlying such statistical models may be uncertain [15,16], while the model estimates and the goodness-of fit are sensitive to those assumptions (e.g. Supporting Information for [13,14]).

Reducing uncertainty in both the assumptions and burden estimates for respiratory virus infections, and increasing the specificity of estimates by narrow age group and by underlying condition, could support the development and targeting of prevention and treatment. For example, a variety of RSV vaccine candidates for different populations are currently in different stages of development [17–19], but target groups for vaccination beyond infancy have not been well described.

While estimation of the overall burden of pediatric RSV hospitalizations in different age groups is helpful for informing mitigation efforts, particularly vaccination in the general population of children, some subgroups of children may merit extra mitigation efforts during RSV epidemics (e.g. annual vaccination, as opposed to less regular vaccination in the general population of children). One such group is children with a diagnosis of asthma [1,5–7]. Prevalence of asthma diagnosis in young children increases with age [20]. Diagnosis of asthma in infants and young children is challenging [21], with different causal mechanisms involved in the pathogenesis of asthma-like symptoms during childhood [22]. Nonetheless, several studies [1,5–7,23,24] have suggested that asthma diagnosis in children serves as a marker for the risk of RSV- and influenza-associated hospitalization. However, the age-specific magnitudes of those risks for the subpopulations of US children with an asthma diagnosis compared to the general population of children are not well established, particularly for RSV-associated hospitalizations. Quantification of those risks could therefore help inform RSV and influenza-related prevention and treatment strategies for those children.

In our earlier work [13,14], we introduced a new method for estimating the burden of severe outcomes associated with influenza and RSV, designed to address several limitations of some of the previously employed inference models. Important features of that approach include: the use of RSV and influenza (sub)type (A/H3N2, A/H1N1, and B) incidence proxies that are expected to be linearly related to the population incidence of those viruses; a flexible model for the baseline of severe outcomes not associated with influenza and RSV; and a bootstrap method for inferring the confidence bounds on the estimates of the rates of influenza and RSV-associated severe outcomes that accounts for auto-correlation in the time series for the noise.

Here, we apply the inference method in [13,14] to data on hospitalizations from US states that reported to the HealthCare Cost and Utilization Project (HCUP) for the 2003–04 through the 2009–10 seasons to better estimate the rates of influenza and RSV-associated hospitalization in US children. We estimate these rates for several categories of ICD-9-coded hospital discharge diagnoses. Our main goals are to provide detailed age-specific estimates (by year of age for the youngest children) of the incidence of influenza and RSV-associated hospitalization in the general population of children, as well as in the population of children diagnosed with of asthma.

Methods

Hospitalization Data

We used weekly data on hospitalizations with different diagnoses in different age groups of children (<1y,1y,2y,3y,4y,5–6y,7–11y,12–17y) from the State Inpatient Databases of the Healthcare Cost and Utilization Project (HCUP), maintained by the Agency for Healthcare Research and Quality (AHRQ) through an active collaboration [25]. Thirty-five states reported to HCUP for the 2003–04 through the 2009–10 seasons. For our analyses, we included 24 states for which data on hospitalizations in infants with RSV in the principal diagnosis exhibited consistency throughout the study period (eAppendix). Those 24 states, which constituted approximately 65.8% of the US population during the study period, are Arkansas, California, Colorado, Connecticut, Georgia, Hawaii, Iowa, Illinois, Indiana, Maryland, Minnesota, North Carolina, Nebraska, New Jersey, Nevada, New York, Ohio, Oregon, Tennessee, Texas, Virginia, Vermont, Washington, and Wisconsin.

Aggregate hospitalization data were used in our analyses, with no informed consent from participants sought.

Statistical Inference Framework

Our study period is between calendar week 27 of 2003, and calendar week 26 of 2010.

As the incidence rates of influenza and RSV infection are difficult to estimate directly, we use proxies for the incidence of RSV and the major influenza (sub)types (A/H3N2, A/H1N1, and influenza B) that are expected to be proportional to the population incidence of those viruses [13–15]. Those proxies are derived from the available data as described below.

For the influenza incidence proxies, we utilized the US Centers for Disease Control and Prevention (CDC) influenza surveillance data [26]. We multiplied the weekly state-specific percent of medical consultations in the CDC Outpatient Illness Surveillance Network (ILINet) that were for influenza-like illness (ILI) by the state-specific percent of respiratory specimens in the US Virologic Surveillance laboratories that tested positive for each of the major influenza (sub)types (A/H3N2, A/H1N1, and influenza B) to define state-specific weekly proxies for the incidence of each major influenza (sub)type:

$$\begin{gathered} Weeklyinfluenza\left(sub\right)typeincidenceproxy= \\ =\%ConsultationsforILI*\%Allrespiratoryspecimensthatwerepositiveforthe\left(sub\right)type \end{gathered}$$ (1)

Incidence proxy for an influenza (sub)type for the collection of 24 US states used in our analyses (Data subsection) is defined as the sum of state-specific incidence proxies for the (sub)type weighted by the weekly state populations. We estimated weekly state populations during our study period through linear interpolation (in time) for the yearly, July 1st state population estimates [27]. Finally, the relation between an influenza incidence proxy for a given influenza (sub)type and the rate of associated hospitalization in a given age group may change over time due to a variety of factors, particularly the circulation of novel influenza strains (see [13], as well as eAppendix). To accommodate those changes, we split the incidence proxies for both influenza A/H3N2 and A/H1N1 into several time periods – see eAppendix for full details.

The weekly RSV incidence proxy we used in this study was the cumulative rate of hospitalization among infants with RSV (ICD-9 codes 466.11, 480.1, or 079.6 present in the principal discharge diagnosis) during that week for the collection of states included in our analyses. We use hospitalization rates in infants because there is an apparent upward trend in annual rates of hospitalization with the principal diagnosis of RSV in all age groups above two years of age in the US, likely due to changes in testing/diagnostic practices, but no apparent trend in the rates of hospitalization with the principal diagnosis of RSV in infants. We note that the use of the above RSV incidence proxy is premised on the assumption that there is year-to-year consistency in the relation between RSV infection rates in different age groups of children. Such temporal consistency may not hold for the relation between RSV infection rate in children vs. adults, and the use of the above incidence proxy for estimating hospitalization rates in adults may be questionable. We refer to eSection S2 of the eAppendix for further details on the RSV and influenza incidence proxies, including the relevant plots..

The outcomes we evaluated are rates of hospitalization for six different diagnosis categories (eTable S1 in the eAppendix) by age group in children. Those categories are 1) respiratory cause excluding asthma (ICD-9 codes 460–519; excluding 493) in the principal discharge diagnosis; 2) respiratory cause (ICD-9 codes 460–519) present anywhere in the discharge diagnoses (principal or secondary) excluding asthma (ICD-9 code 493) in the principal diagnosis; 3) respiratory cause (ICD-9 codes 460–519) present anywhere in the discharge diagnosis; 4) pneumonia and influenza (ICD-9 codes 480–488; pneumonia and/or influenza) in the principal discharge diagnosis; 5) asthma (ICD-9 code 493) as a secondary (non-principal) discharge diagnosis; 6) asthma (ICD-9 code 493) present anywhere in the discharge diagnosis. The first three categories are used to evaluate the burden of RSV and influenza-related respiratory hospitalizations, from the more restricted (category 1) to the most inclusive category of hospitalizations (category 3). Category 4 represents an important subcategory of respiratory hospitalizations. Categories 5 and 6 are subsequently used to estimate the risks for RSV and influenza-related hospitalization associated with having a prior diagnosis of asthma.

Hospitalization Rate Estimation: For each choice of an age group and diagnosis category 1–6, we apply a linear regression model to evaluate the contribution of influenza and RSV to the hospitalization burden for the corresponding diagnosis category in the corresponding age group. The dependent variable (outcome) is the weekly hospitalization rate in that age group with that diagnosis category during the study period. The covariates are: incidence proxies for the different influenza (sub)types and RSV, as well as temporal trend (modeled by a low degree polynomial in time) and a seasonal baseline with annual periodicity (modeled by cubic splines in the calendar week with knots at every fourth week, periodic in calendar year) [13,14]. If O(t) is the weekly rate for the outcome on week t, and V_i(t) are the incidence proxies for the various viruses, namely the different influenza (sub)types (split into time periods as described in the eAppendix) and RSV, the model structure is

$$O\left(t\right)={\sum }_i{\beta }_i\left(t\right)\cdot V_i\left(t\right)+Baseline+Trend+Noise$$ (2)

To account for the autocorrelation structure in the noise, a bootstrap method devised in [13] is used to estimate the confidence bounds for the model’s estimates. Estimates derived in earlier work [14] suggest that several alternative inference methods (such as maximal likelihood estimation under the AR(1) assumption for the noise structure) yield very similar results.

Inclusion of hospitalizations with a principal diagnosis of asthma into the inference framework described by eq. 2 can be problematic due to the substantial annual aperiodicity in the weekly rates of those hospitalizations that could not be explained by the aperiodicity in RSV and influenza circulation patterns. Rates of hospitalization for categories 3 and 6 associated with RSV and influenza infections are estimated by scaling the rates of hospitalization in categories 2 and 5 (that differ from categories 3 and 6 by the absence of corresponding hospitalizations with a principal diagnosis of asthma) up by factors related to the prevalence of the principal diagnosis of asthma in a subsample of corresponding hospitalizations, as described in eSection S4 of the eAppendix.

Risks for RSV and influenza hospitalization associated with a prior diagnosis of asthma

To estimate the relative risk for influenza hospitalization in a given age group associated with a prior diagnosis of asthma we estimate

1. Age-specific proportion H(age) of cases of influenza-related hospitalizations in the Influenza Hospitalization Surveillance Network FluSurv-NET surveillance database [28,7] between 2003–04 through 2009–10 seasons that have a previous history of asthma, including reactive airway disease.

2. Age-specific population prevalence D(age) of asthma diagnosis between 2003–2009 (provided to us by Dr. L. Akinbami of the US CDC; [20]).

The relative risk (risk ratio) RR_flu(age) for influenza hospitalization in a given age group associated with a prior diagnosis of asthma is estimated as:

$${RR}_{flu}(age)=\frac{H\left(age\right)}{D\left(age\right)}/\frac{1-H\left(age\right)}{1-D\left(age\right)}$$ (3)

Estimation of the (age-specific) relative risk for RSV hospitalization associated with having a prior diagnosis of asthma is similar, except that the estimation of the proportion of RSV-associated hospitalizations that involve children with a prior diagnosis of asthma uses the estimates for the rates of RSV hospitalization with a secondary diagnosis of asthma, as well as data on the relation between the secondary diagnosis of asthma and prior history of asthma (eAppendix, eSection S5).

Results

Figures 1,2 shows the model fits for the category of hospitalizations with a respiratory cause present anywhere in the discharge diagnosis in children aged <2y, as well as the contribution of both RSV and influenza to the rates of these hospitalizations. Figures 3,4 present the corresponding information for the categories of hospitalizations with pneumonia and influenza in the principal diagnosis. Further model fits are presented in the eAppendix. All those figures show good, temporally consistent model fits (R-squared statistic above 0.992 for all models in Figures 1–4; eAppendix), as well as the major contribution of RSV to the corresponding hospitalization burden.

Figure 1:

Weekly hospitalization rates (per 100,000) with a respiratory cause present anywhere in the discharge diagnosis for children aged under one year, 2003–04 through the 2009–10 seasons (black), model fits (red), and contributions of the respiratory syncytial virus (RSV; red curve minus green curve) and influenza (green curve minus blue curve).

Figure 2:

Weekly hospitalization rates (per 100,000) with a respiratory cause present anywhere in the discharge diagnosis for children aged 1 year, 2003–04 through the 2009–10 seasons (black), model fits (red), and contributions of the respiratory syncytial virus (RSV; red curve minus green curve) and influenza (green curve minus blue curve).

Figure 3:

Weekly hospitalization rates (per 100,000) with pneumonia and influenza in the principal discharge diagnosis for children aged under one year, 2003–04 through the 2009–10 seasons (black), model fits (red), and contributions of the respiratory syncytial virus (RSV; red curve minus green curve) and influenza (green curve minus blue curve).

Figure 4:

Weekly hospitalization rates (per 100,000) with pneumonia and influenza in the principal discharge diagnosis for children aged 1 year, 2003–04 through the 2009–10 seasons (black), model fits (red), and contributions of the respiratory syncytial virus (RSV; red curve minus green curve) and influenza (green curve minus blue curve).

Tables 1 and 2 present the estimates of the average annual rates of hospitalization associated with influenza and RSV for the six categories of discharge diagnoses (Methods) in select age subgroups of children. For all age groups and diagnosis types studied (except pneumonia and influenza hospitalizations in children aged 7–17y), the estimates of the rates of RSV-associated hospitalization are higher than the estimates of the rates of influenza-associated hospitalization. The estimated rates of RSV-associated hospitalization with a respiratory cause present anywhere in the discharge diagnosis are highest in children aged <1y (Table 1), and decline rapidly with age. For the more restricted categories 2 and 1 of hospitalizations (Methods), the rates of RSV-associated hospitalization were progressively lower (Table 1). A sizeable fraction of both RSV and influenza-associated hospitalizations with a respiratory cause present anywhere in the diagnosis in different age groups (except RSV hospitalizations in infants aged <1y) had pneumonia and influenza in the principal diagnosis (Table 2). Additionally, for all age groups of children except 4 year-olds, the majority of RSV and influenza-associated hospitalizations with asthma present anywhere in the discharge diagnosis have asthma as a secondary (non-principal) diagnosis (Table 2).

Table 1:

Average annual rates per 100,000 children in different age groups (with 95% confidence intervals) of influenza and respiratory syncytial virus (RSV)-associated hospitalizations that have a respiratory cause (excluding asthma) in the principal discharge diagnosis (ICD-9 codes 460–519 excluding 493); a respiratory cause present anywhere in the discharge diagnosis (excluding asthma in the principal diagnosis); and respiratory cause present anywhere in the discharge diagnosis, 2003–04 through the 2009–10 seasons.

Age group	Hospitalizations with a respiratory cause (excluding asthma) in the principal diagnosis		Hospitalizations with a respiratory cause present anywhere in the diagnosis (excluding asthma in the principal diagnosis)		Hospitalizations with a respiratory cause present anywhere in the diagnosis
	Flu	RSV	Flu	RSV	Flu	RSV
<1y	140.9 (114.1,168.4)	2129 (2044,2214)	179.3 (141.6,217.6)	2347 (2224,2471)	181 (142.5,220.3)	2381 (2252,2515)
1y	66.3 (41.3,91.4)	561.2 (483,639.4)	80.7 (53.6,108.4)	631.2 (543.9,716.9)	85.6 (57.3,114.5)	710.6 (609.1,809.2)
2y	41.2 (28.1,53.9)	265 (223.2,307.3)	57.4 (41.7,73.3)	322.3 (272.8,372.2)	62 (45.3,79)	395 (327.7,462.4)
3y	32.1 (22.8,41.7)	140.9 (111.2,171)	44.5 (32.7,56.5)	168.8 (131.4,207.2)	47.7 (35.6,60.9)	211.3 (154.6,266.8)
4y	26.5 (19.8,33.3)	66.1 (44.3,88.1)	36.8 (28.2,45.5)	87.7 (59.3,116.2)	41.3 (31.7,51.3)	111.1 (62.4,160.1)
5–6y	27.6 (22.2,33)	41.5 (23.6,59.2)	36.8 (29.6,44)	62.8 (39.3,86.3)	40.2 (32.4,48.4)	72.3 (29.3,116.4)
7–11y	14.7 (12.3,17.1)	16.8 (9.2,24.5)	20.4 (16.3,24.4)	27.6 (14.1,41.1)	23 (18.5,27.7)	35.6 (9.9,62.2)
12–17y	9.9 (8.5,11.3)	14 (9.3,18.8)	15.9 (10.5,21.3)	33.8 (15.2,52.6)	17.9 (11.7,24.2)	39 (17.5,60.6)

ICD indicates international classification of diseases, RSV respiratory syncytial virus.

Table 2:

Average annual rates per 100,000 children in different age groups (with 95% confidence intervals) of influenza and respiratory syncytial virus (RSV)-associated hospitalizations that have pneumonia and influenza (ICD9 codes 480–488) in the principal discharge diagnosis; asthma as a secondary (non-principal) discharge diagnosis; and asthma present anywhere in the discharge diagnosis, 2003–04 through the 2009–10 seasons.

Age group	Pneumonia and Influenza in the principal diagnosis		Asthma as a secondary (non-principal) diagnosis		Asthma as either the principal or secondary diagnosis
	Flu	RSV	Flu	RSV	Flu	RSV
<1y	111.9 (91.4,133.1)	360.1 (297.8,421.8)	5.5 (−0.1,11.1)	158.3 (139.4,177)	6.4 (0.1,13)	209.6 (178.2,241.9)
1y	59 (46.5,71.8)	247.3 (207.4,286.8)	10.1 (3.1,17.3)	143.8 (120.2,167.6)	14.3 (4.1,24.8)	221.4 (171.4,270.8)
2y	36.5 (28.7,44.4)	152.6 (126.2,178.3)	9.3 (4.3,14.4)	96.8 (79.4,113.8)	13.1 (6,20.7)	176.4 (137.3,215.7)
3y	28.6 (20.6,36.7)	94 (69,118.9)	8.1 (3.9,12.3)	58.7 (44.4,73.1)	10.3 (5,16)	107.7 (70.9,144.9)
4y	22.9 (17.6,28.4)	47 (29.5,64.4)	6 (2.3,9.8)	22.9 (10.1,35.5)	9.5 (3.5,15.8)	49.3 (12.6,86.7)
5–6y	21.5 (16.8,26.3)	29.1 (13.8,44.3)	7.4 (4.2,10.7)	16.2 (5.3,27.1)	9.8 (5.7,14.4)	27 (−7.2,61.6)
7–11y	11.2 (9.3,13.2)	10.4 (4.2,16.5)	5.1 (3.2,7)	10.6 (4.1,17.2)	7.3 (4.5,10.2)	20.1 (−1.1,41.5)
12–17y	8.1 (7.1,9)	8 (5,11.1)	2 (−1,5.1)	13.9 (3.5,24.2)	3.4 (−1.8,8.5)	18.6 (5,32.3)

ICD indicates international classification of diseases, RSV respiratory syncytial virus.

Table 3 presents the estimates of the prevalence of asthma by age group between 2003–2009 in young children in the US, and the relative risks for RSV and influenza-associated hospitalization associated with a prior diagnosis of asthma between the 2003–04 through the 2009–10 seasons in the US. Prior diagnosis of asthma is associated with substantial risks for both RSV and influenza hospitalization in children aged <5y, with those risks ranging between 3.1(2.1,4.7) (age <1y) to 6.7(4.2,11.8) (age 2y) for RSV hospitalizations, and between 2.8(2.1,4) (age <1y) to 4.9(3.8,6.4) (age 3y) for influenza hospitalizations. Additionally, the estimated relative risks for RSV hospitalization were somewhat higher than the corresponding relative risks for influenza hospitalization.

Table 3:

Average annual prevalence of asthma diagnosis in US children aged <5y between 2003–2009, and relative risks for respiratory syncytial virus (RSV) and influenza hospitalization associated with a prior diagnosis of asthma in different age groups between the 2003–04 through the 2009–10 seasons.

Age group	Prevalence of asthma diagnosis among US children in the age group	Relative risk for RSV hospitalization associated with a prior diagnosis of asthma	Relative risk for influenza hospitalization associated with a prior diagnosis of asthma
<1y	2% (1%,2%)	3.1 (2.1,4.7)	2.8 (2.1,4.0)
1y	5% (4%,6%)	6.0(3.9,9.5)	3.7 (2.9,4.7)
2y	6% (5%,7%)	6.7 (4.2,11.8)	4.7 (3.7,6.0)
3y	8% (7%,9%)	5.9 (3.2,13.1)	4.9 (3.8,6.4)
4y	10% (9%,11%)	5.1 (1.4,27.2)	4.4 (3.4,5.8)

Discussion

Granular evaluation of the burden of severe outcomes associated with influenza and RSV in children, including children with underlying health conditions could aid in the design of mitigation efforts. For example, while a number of RSV vaccine candidates are currently in different stages of development [17–19], target groups for RSV vaccination beyond the infant population have not been well characterized. The monoclonal antibody palivizumab is recommended for infants and young children with certain underlying heath conditions that put those children under elevated risk for RSV hospitalization [29]; however, young children with asthma are not included in those recommendations. In this paper we apply our previously developed methodology [13,14] to estimate the rates of influenza and RSV-associated hospitalization for various discharge diagnoses by age group in US children between 2003–2010. Estimates of the rates of RSV-associated hospitalization in young children are very high (substantially higher than the corresponding estimates of influenza-associated hospitalization rates) and decline rapidly by year of age, most steeply for the first vs. second year of life. Additionally, our results suggest that having a diagnosis of asthma carries substantial risks for both RSV and influenza-associated hospitalization in children. In particular, the rates of RSV-associated hospitalization in the youngest children with a prior diagnosis of asthma are exceptionally high.

While community-based studies of hospitalization in children involving virologic testing (e.g. [1–4]) offer direct evidence about the burden of hospitalization in children associated with the circulation of respiratory viruses, such studies may suffer from certain limitations, particularly potential underdetection of the presence of viral infections, as well as small sample sizes. While the rates of influenza-associated hospitalization for children aged under 2y estimated in this study (Table 1) are similar to the estimates in the community-based studies [3,4], rates of influenza-associated hospitalization for children aged 2–6y estimated in this study are generally higher than the corresponding estimates in the community-based studies [3,4]. For RSV-associated hospitalizations, our estimates of hospitalization rates (Table 1) are higher than the estimates in the community-based studies [1,2] for all age groups of children under 6y, with the relative differences between our estimates and the estimates in the community-based studies [1,2] increasing with age (e.g. a nearly five-fold difference for children aged 2–5y, compare Table 1 in our paper with Table 2 in [1]). The above differences for both influenza and RSV-associated hospitalizations may suggest underdetection of the corresponding infections in hospitalized children, particularly for RSV, as well as for older children. We also note that, compared to inference based on Poisson regression [9], and to estimates derived from data on hospital discharge diagnoses [8], our estimates of the rates of RSV-associated hospitalization (Table 1) are very similar in infants aged <1y, and are somewhat higher for children aged 1–4y, with good, temporally consistent model fits produced by our inference method (Figures 1–4; eAppendix).

We note that a sizeable fraction of the estimated RSV and influenza-associated hospitalizations in children involved complications stemming from bacterial infections, but the course of disease was initially triggered by viral infections. For example, Figures 3,4 suggests that most hospitalizations for pneumonia and influenza in young children above the seasonal baseline are explained (well matched) by the patterns of RSV circulation. Influenza circulation also explains some of the pneumonia and influenza hospitalizations (category 4; ICD-9 codes 380–388) above the seasonal baseline in Figures 3,4, with major influenza epidemics (e.g. the 2003–04 season, and the Fall of 2009) corresponding to spikes in pneumonia and influenza hospitalization rates. We call those hospitalizations RSV and influenza-associated hospitalizations, though the etiology behind some of those hospitalization outcomes involves additional factors besides viral infections, such as bacterial infections (e.g. secondary infection by S. pneumoniae), underlying health conditions, etc. Our results suggest that a large fraction of pneumonia and influenza hospitalizations are associated with RSV and influenza infections (Figures 3,4; eAppendix), which agrees with the findings in [30].

The connection between RSV infection in early childhood and the subsequent risk of developing recurrent wheezing and asthma has received considerable attention in the literature [31–34]. The contribution of RSV and influenza to hospitalization burden in children with asthma is less well established, partly due to the complexities of asthma diagnoses in very young children [21,22,35]. A number of studies have focused on the contribution of respiratory viruses, particularly the human rhinovirus, to asthma exacerbations [36,37]. At the same time, infections with respiratory viruses, particularly RSV, in children with asthma can result in hospitalizations that have a principal cause other than asthma exacerbation (e.g. Table 2). Quantification of the full hospitalization burden associated with RSV and influenza circulation in children who have a diagnosis of asthma should therefore help inform prevention and treatment strategies in those children. Several studies have suggested that young children with underlying health conditions, including asthma, are disproportionately represented among young children hospitalized with influenza and RSV infections [1,5–7,23,24,38]. Hall et al. [1] and Glezen et al. [5] demonstrated a high frequency of RSV infections among respiratory hospitalizations in young children, as well as the relatively high proportions of asthma or other pulmonary conditions in RSV-infected hospitalized children. A large Danish study demonstrated that previous asthma hospitalization carried ~5-fold risk for hospitalization with a confirmed RSV infection before the age of 2 years [23]. Our estimates of the relative risks for both RSV and influenza hospitalization associated with a prior diagnosis of asthma are in line with the findings in other studies [1,5–7,23]. While the etiology behind an asthma diagnosis in young children may be subject to uncertainty [21,22], our results support the notion that an asthma diagnosis serves as a marker for the risk of an RSV- and influenza-associated hospitalization in young children. We note that the US Centers for Disease Control (CDC) recommendations for annual influenza vaccination apply to all children, with a special emphasis on children with underlying health conditions such as asthma; and that the schedule for RSV vaccines currently under development [17–19] for different target groups is being evaluated. Furthermore, while the estimated rates of RSV-associated hospitalization in the youngest children diagnosed with asthma are exceptionally high, those children are not included in the recommendations for palivizumab prophylaxis [29].

This study has some limitations. Information on prior asthma diagnosis in hospitalized children was unavailable [25]. Instead, we extrapolated the relation between frequencies of a prior diagnosis of asthma vs. the hospital discharge diagnosis of asthma for cases of RSV-associated hospitalization from analogous data on influenza hospitalizations reported to FluSurv-NET surveillance – see eAppendix. While such extrapolation is imperfect, we note that relatively high proportions of cases of influenza-related hospitalization in different age subgroups of children in FluSurv-NET that have asthma listed on the discharge diagnosis also have a prior diagnosis of asthma, and vice versa, suggesting a strong connection between those two categories of hospitalization. Our estimates of several quantities, such as the risk for influenza hospitalization associated with having a diagnosis of asthma rely on how accurately the FluSurv-NET surveillance data represent certain quantities (such as the proportion of hospitalizations that have asthma as the principal discharge diagnosis) for all influenza-related hospitalizations in children (eAppendix). Another potential issue is the accuracy of the proposed regression framework for the hospitalization time-series [13,14]. Spikes in hospitalization rates during certain influenza seasons correspond visually to major influenza epidemics (as suggested by the incidence proxies that we utilize), providing additional support for the validity of our inference method for influenza-associated hospitalization rates. While RSV circulation is more periodic than influenza circulation, there is notable aperiodicity in the RSV circulation (eFigure S1 in the eAppendix); use of a flexible model for the baseline rates of hospitalization not associated with influenza or RSV helps separate the baseline rates from the rates of hospitalization associated with influenza and RSV and results in substantial improvement in the model fits compared to the use of a trigonometric model for the baseline rates (Supporting Information for [14,13]). We used rates of hospitalizations in infants with RSV in the principal diagnosis as a proxy for RSV incidence, while presence of RSV in the principal diagnosis need not be supported by a laboratory test. We excluded certain states with unusual patterns of hospitalization rates in infants with RSV in the principal diagnosis from the analyses (see eAppendix) in an effort to minimize the impact of variable quality of the proxy. We note that our models produce good, temporally consistent fits to the data (R-squared statistic above 0.98 for all models, eAppendix), especially for the younger children (e.g. Figures 1–4), and that the confidence bounds for the contribution of RSV to many of the hospitalization categories considered are reasonably tight (Tables 1 and 2). Finally, the estimated weekly baseline rates of hospitalization not associated with RSV and influenza exhibit Fall and Spring peaks (Figures 1–4), with those peaks being suggestive of the contribution of rhinoviruses and parainfluenza [39]. Further related work is needed to examine various aspects of our inference methodology, including the effect of the circulation of other respiratory viruses.

We believe that, despite those limitations, our work provides granular estimates of the rates of hospitalization associated with influenza and RSV infections in children, suggesting very high rates of RSV-associated hospitalizations in young children. Such work, in conjunction with efforts to understand the role of different age groups in propagating RSV epidemics [40] may aid the evaluation of the impact of RSV vaccination on disease rates in different population groups, including children, and inform RSV vaccination strategies. Additionally, our results demonstrate risks for both RSV and influenza hospitalization associated with a prior diagnosis of asthma in young children, with the rates of RSV-related hospitalization in the youngest children diagnosed with asthma (particularly those aged <2y) being especially high. These results may help inform prevention efforts for those children.