Global burden of acute lower respiratory infections due to respiratory syncytial virus in young children: a systematic review and meta-analysis.

BACKGROUND: The global burden of disease attributable to respiratory syncytial virus (RSV) remains unknown. We aimed to estimate the global incidence of and mortality from episodes of acute lower respiratory infection (ALRI) due to RSV in children younger than 5 years in 2005.
METHODS: We estimated the incidence of RSV-associated ALRI in children younger than 5 years, stratified by age, using data from a systematic review of studies published between January, 1995, and June, 2009, and ten unpublished population-based studies. We estimated possible boundaries for RSV-associated ALRI mortality by combining case fatality ratios with incidence estimates from hospital-based reports from published and unpublished studies and identifying studies with population-based data for RSV seasonality and monthly ALRI mortality.
FINDINGS: In 2005, an estimated 33.8 (95% CI 19.3-46.2) million new episodes of RSV-associated ALRI occurred worldwide in children younger than 5 years (22% of ALRI episodes), with at least 3.4 (2.8-4.3) million episodes representing severe RSV-associated ALRI necessitating hospital admission. We estimated that 66 000-199 000 children younger than 5 years died from RSV-associated ALRI in 2005, with 99% of these deaths occurring in developing countries. Incidence and mortality can vary substantially from year to year in any one setting.
INTERPRETATION: Globally, RSV is the most common cause of childhood ALRI and a major cause of admission to hospital as a result of severe ALRI. Mortality data suggest that RSV is an important cause of death in childhood from ALRI, after pneumococcal pneumonia and Haemophilus influenzae type b. The development of novel prevention and treatment strategies should be accelerated as a priority. FUNDING: WHO; Bill & Melinda Gates Foundation. Copyright 2010 Elsevier Ltd. All rights reserved.

Introduction

Acute lower respiratory infection (ALRI) is the leading cause of global child mortality. Respiratory syncytial virus (RSV) is believed to be the most important viral pathogen causing ALRI in young children, although its contribution to ALRI deaths is uncertain. Many data for incidence of and mortality from RSV-associated ALRI in developing countries remain unpublished. Therefore, we formed an RSV study group to supplement a systematic literature review with unpublished data. No global or regional estimates have been made previously of the burden of RSV-associated ALRI in children. Previous reviews have focused either on developing countries or industrialised countries. We aimed to estimate the burden of disease worldwide due to RSV-associated ALRI in children younger than 5 years for 2005. We estimated only the ALRI component of the burden due to RSV infection because the main costs of RSV disease relate to health-service use for ALRI, and first-generation RSV vaccines are likely to protect against ALRI. This report is especially timely in view of the development of candidate vaccines that are sufficiently attenuated yet immunogenic in infants, and the increased funding for development and implementation of novel vaccines in developing countries that is available through the Global Alliance for Vaccines and Immunisation. We also aimed to emphasise important gaps in knowledge and to provide information about potential sites for vaccine trials.

Methods

Search strategy and selection criteria

We did a systematic literature review using a combination of search terms (webappendix pp 3–4), hand searching of online journals, and scanning of reference lists of identified citations. The search was limited to Medline (Ovid), Embase, CINAHL, Global Health, Web of Science, WHOLIS, LILACS, IndMed, and the grey literature (SIGLE) databases and to studies published between January, 1995, and June, 2009. Panel 1 shows eligibility criteria. No language or publication restrictions were applied. We invited the participation of researchers who had done similar studies resulting in unpublished data or supplementary data from published work.

Definitions

Most investigators used the clinical pneumonia and severe pneumonia syndromic case definitions that were established by WHO. We chose to use ALRI and severe ALRI, including bronchiolitis and pneumonia, to recognise the important contribution of bronchiolitis in RSV infections. ALRI has been defined as equivalent to clinical pneumonia, which is characterised by acute-onset cough or difficulty in breathing with fast breathing for age. Acute cough or difficulty in breathing with indrawing of the lower chest wall indrawing (with or without fast breathing for age) necessitating hospital admission has been defined as severe ALRI, and is equivalent to severe clinical pneumonia.

We used the American Association for Respiratory Care definition of hypoxaemia and regarded oxygen saturation (at sea level in room air) lower than 90% (measured by pulse oximetry) in children older than 1 month and lower than 88% in neonates as hypoxaemic, unless indicated otherwise. We used a modification of the definition for RSV season that was provided by Mullins. Any month of a year with RSV detected in at least four (or 5%) of the submitted specimens were deemed to be within the RSV season. We regarded countries as industrialised if they fell within: high-income Asia Pacific, high-income North America, western Europe, and Australasia. Remaining countries were classified as developing. Population estimates by region for 2005 are as in The State of the World's Children 2007.

Adjustment for differences across studies

Few studies adopted the WHO acute respiratory infection case definition or reported RSV data for the full age range (0 to <5 years). Where necessary, data imputation was used on the basis of the approach adopted by Rudan and colleagues. Relative to an incidence of 1·0 for RSV-associated severe ALRI in the younger than 1 year age group, we calculated an incidence rate ratio of 0·58 for children younger than 2 years and 0·30 for those younger than 5 years by taking the median of the incidence rate ratios of studies reporting incidence for the full age range. For RSV-associated ALRI, an incidence rate ratio of 1·3 was calculated for children younger than 2 years, whereas for those younger than 5 years the ratio was 0·81. Too few studies reported neonatal RSV-associated ALRI for us to report this finding reliably as a separate group. We assumed that cases defined as hospitalised acute respiratory infection or hospitalised lower or acute lower respiratory infection represented severe ALRI (webappendix p 5). Figure 1 summarises our overall approach and associated rationale for decisions adopted.

Figure 1

Approaches for estimation of global RSV incidence and mortality in children aged 0–5 years

RSV=respiratory syncytial virus. ALRI=acute lower respiratory infection. CFR=case fatality ratio. *Approach justified by large difference in reported incidence between studies using active and passive case ascertainment; studies with passive ascertainment reported much lower estimates than did those with active ascertainment. †Approach justified by the decision that hospital-based data would be most useful for population-based projections, since all severe episodes are likely to need hospital treatment; also, we noted no difference in reported incidence of RSV-associated severe ALRI between studies with active and passive case ascertainment. †Approach based on assumptions that: i) baseline proportional mortality of RSV-associated ALRI in all ALRI would be similar to proportional incidence of severe ALRI in all severe ALRI, and ii) there is no overall effect from seasonality of other respiratory pathogens; then, if all excess ALRI mortality during RSV seasonal peaks is assigned to RSV as the only cause (in a setting with many seasonal peaks) and this mortality is added to baseline mortality estimates, this approach is likely to overestimate the contribution of RSV to mortality from all ALRI. §Approach deemed to yield a lower bound for RSV-associated ALRI mortality because patients with severe RSV-associated ALRI treated in hospital might have higher CFRs than do all severe cases of RSV (treated and untreated), but a substantial (but unknown) proportion of severe cases will not present to health services for treatment, thereby increasing overall CFR.

Statistical analysis

Using Stata (version 10.1), we did a meta-analysis of incidence data and reported pooled estimates and 95% CIs using the random effects model (DerSimonian-Laird method) since data were heterogeneous (p<0·0001). Studies of Indigenous populations in industrialised countries were excluded from this analysis. We estimated incidence for industrialised and developing countries for 2005, and summed these estimates to yield the global incidence estimate for that year.

Since data were scarce, we did not model a point estimate for RSV-associated ALRI mortality. Instead, we assessed the possible boundaries of mortality that could plausibly be attributed to RSV using three approaches. First, the median of all RSV-associated ALRI case fatality ratio data from hospital-based reports was combined with hospital-based inpatient data for RSV-associated severe ALRI incidence. Since access to hospital treatment in most developing countries is typically limited, we judged this result to represent a lower bound for mortality. The second approach was similar to the first, but severity was defined on the basis of presence of hypoxaemia (an established marker of RSV-associated severe ALRI). We estimated the proportion of children with severe disease and hypoxaemia at presentation and applied a 14% case fatality ratio on the basis of published data from developing country studies reporting outcomes in hypoxaemic children with ALRI. This approach adjusted for the fact that no RSV test was done in 4–28% of children admitted with ALRI in the studies we included (webappendix p 6).

A third approach assumed that all excess mortality due to ALRI in children younger than 5 years in the RSV season was due to RSV, and that non-RSV mortality is equal within and between RSV epidemic periods. Since approach three represents an extreme scenario, we assumed that this method yielded an upper bound for RSV-associated ALRI mortality. We defined the duration (in months) of the RSV season for each calendar year of the study (MonRSV). For each year, we calculated the average number of total ALRI deaths (in and out of hospital) that occurred per month during (AvgRSV) and outside (AvgOTHER) the RSV season, as well as the total number of deaths (TOTAL) during the year. The proportion of yearly deaths due to RSV was then calculated as:

Application of this approach to the estimated mortality of children younger than 5 years due to ALRI in Indonesia (with 2005 population data and WHO estimates of mortality attributable to ALRI in this age group) would provide an estimate of all deaths attributable to RSV if community-based case ascertainment was used. This approach needed population-based data for RSV seasonality and monthly death records (with cause of death attribution) from the same population, and this information was only available from one developing country site (with data for a 3-year period). The proportion of excess deaths computed with this method for Indonesia was extrapolated to estimate the global mortality of children younger than 5 years attributable to RSV-associated ALRI.

Role of the funding source

The funding sources supported the organisation of a meeting of the RSV study group in Edinburgh on Jan 29 and 30, 2008. The study sponsors had no role in the writing of the report or the decision to submit for publication. HC had full access to all data in the study and had final responsibility for the decision to submit for publication.

Results

We identified 36 studies with suitable data (figure 2): 19 published population-based studies reporting incidence of RSV-associated severe or non-severe ALRI in populations under surveillance; seven published studies estimating incidence on the basis of hospital discharge records or laboratory diagnosis reports and a census-based denominator of children at risk; and unpublished data from ten population-based studies, again reporting a clear denominator of children at risk (figure 3, table 1).

Figure 2

Flow diagram for selection of studies

Figure 3

Location of the 36 studies by Global Burden of Diseases, Injuries and Risk Factors regions

Table 1

Incidence estimates of RSV-associated ALRI and severe ALRI in children younger than 5 years from published and unpublished studies

	Case ascertainment	Study denominator	Diagnostic tests	Incidence of RSV-associated ALRI (per 1000 children per year)*			Incidence of RSV-associated severe ALRI (per 1000 children per year)*
				Children aged <1 year	Children aged <2 years	Children aged <5 years	Children aged <1 year	Children aged <2 years	Children aged <5 years
Europe, western
Gipuzoka, Spain18	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	26	15	6
Kiel, Germany19	Passive, hospital (IP)	Defined population base	RT-PCR	NA	NA	NA	16	9	(5)
Multicentric, Germany23	Passive, hospital (IP)	Defined population base	PCR using test kit	NA	NA	NA	(28)	(16)	(8)
Shropshire, UK24	Passive, hospital (IP)	Defined population base	IF	NA	NA	NA	(28)	16	(8)
Northern Stockholm, Sweden25	Passive, hospital (IP)	Defined population base	DFA	NA	NA	NA	14	(8)	(4)
Southern Austria, Austria26	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	(12)	7	(4)
UK27	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	28	(16)	(8)
Netherlands28	Passive, hospital (OP/IP)†‡	Census-derived estimate	Not available	NA	NA	NA	10	(6)	(3)
Sub-Saharan Africa, east
Kilifi birth cohort, Kenya (Nokes et al)§	Active, community based	Defined population base	IFA	104	(135)	(84)	13	(8)	(4)
Kilifi hospital study, Kenya (Nokes et al)§	Passive, hospital (IP)	Defined population base	IFA	NA	NA	NA	11	(6)	3
Manhiça district, Mozambique (Roca et al)	Passive, hospital (OP)	Defined population base	ELISA	44	(57)	(36)	16	(9)	4
Sub-Saharan Africa, west
Ibadan, Nigeria22	Active, community-based	Defined population base	ELISA	116	(151)	94	NA	NA	NA
Western region, The Gambia29	Passive, hospital (IP)	Defined population base	IF	NA	NA	NA	18¶	(10)	(5)
Sub-Saharan Africa, southern
Soweto, South Africa (Madhi et al)‖	Passive, hospital (IP)	Defined population base	DFA	NA	NA	NA	10	(6)	2
Agincourt, South Africa22	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	15	(9)	9
Asia, south
Mirzapur, Bangladesh30	Active, community-based	Defined population base	ELISA	(32)	42	(22)	NA	NA	NA
Ballabgarh, India31	Active, community-based	Defined population base	IF (DFA, IFA)	33	43	(27)	14	11	(4)
Asia, southeast
Bandung, Indonesia (Simões et al)	Active, community-based	Defined population base	PCR**	53	60	48	17	14	10
Takhli district, Thailand32	Passive, hospital (IP)	Defined population base	IF	NA	NA	NA	(30)	(17)	9
Lombok, Indonesia (Gessner et al)	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	13	8	4
Asia, east
Eastern and Northern New Territories, Hong Kong, China33	Passive, hospital (IP)	Defined population base	DFA, virus isolation	NA	NA	NA	(10)	(6)	3
Australasia
Townsville (Queensland), Australia34	Passive, hospital (IP)	Defined population base	DFA, Culture	NA	NA	NA	18	(10)	5
North America, high income
Monroe and Davidson County, USA20	Passive, hospital (IP)	Defined population base	RT-PCR and culture	NA	NA	NA	13	(8)	4
YK Delta, Alaska, USA35††‡‡	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	166	(96)	50
Milwaukee, USA36	Passive, hospital (IP)	Defined population base	MPCR, virus culture, and ELISA	NA	NA	NA	(20)	(12)	6
YK Delta, Alaska, USA (Singleton et al)††§§	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	113	65	(34)
Navajo and White Mountain Apache reservations, USA (Epi study)††	Passive, hospital (IP)	Defined population base	ELISA	NA	NA	NA	91	64	(27)
Nashville, Tennessee, USA (Wright et al)¶¶	Passive, hospital (OP)	Defined population base	Culture	46	(60)	24	10	(6)	3
Tennessee, USA37	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	63	41	(19)
USA38	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	23	(13)	(7)
American Indian and Alaska Native, USA39††	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	34	(20)	(10)
USA39	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	27	(16)	(8)
Hawaii, USA21	Passive, hospital (IP)†	Census-derived estimate	Not available	NA	NA	NA	10	(6)	3
Nashville, Rochester and Cincinnati, USA40	Passive, hospital (IP)	Defined population base	RT-PCR and tissue culture	NA	NA	NA	11	3	3
Latin America, central
San Marcos, Guatemala (Bruce et al)‖‖	Active, community-based	Defined population base	ELISA	158	(205)	(128)	60	(35)	(18)
Latin America, tropical
Rio de Janeiro, Brazil41	Passive, hospital (IP)	Defined population base	IF and culture	NA	NA	NA	(47)	(27)	14

RSV=respiratory syncytial virus. ALRI=acute lower respiratory infections. IP=inpatient. NA=not applicable. RT-PCR=reverse transcriptase PCR. IF=immunofluorescence. DFA=direct fluorescent antibody test. OP=outpatient. IFA=indirect immunofluorescent antibody test. MPCR=multiplex reverse transcription PCR.

Data in parentheses are computed incidence estimates.

Incidence estimated with hospital discharge records.

Incidence estimated with weekly lab reports from 17 virological laboratories.

RSV incidence estimates are scaled by fraction of lower respiratory infection samples tested; proportion of RSV-positive patients in the tested group is thus assumed to be the same as that in the eligible but untested group.

Incidence in rural areas; in urban areas incidence is 8·7 per 1000 children per year.

Included children aged 6 weeks up to 5 years.

ELISA was used initially as diagnostic assay for published results and 120 patients with ALRI tested positive for RSV; all samples were later reanalysed with PCR and an additional 43 cases tested positive.

Between July, 1994, and June, 2001.

Although these areas have low incomes and have factors favouring RSV transmission similar to those of developing countries, they are included in North America (high income) by virtue of being part of the USA.

Between July, 2001, and June, 2004.

Excluded children with known risk factors such as prematurity, bronchopulmonary dysplasia, and congenital heart disease.

RESPIRE trial; refer to reference 42 for methods.

We estimated incidence in children younger than 5 years for all studies within Global Burden of Disease regions (table 1). Ten studies, six published and four unpublished (Madhi et al, Wright et al, Roca et al, Simões et al), reported incidence for the full age range. All studies adopted one or a combination of three methods of case ascertainment: active community-based (ALRI cases sought by health workers by regular house-to-house visits); or passive hospital-based or clinic-based (children with ALRI who have sought treatment in hospital or clinic are enrolled), which was further stratified as outpatient or inpatient. Most studies were passive hospital-based (inpatient), with only six studies using active community-based case ascertainment and two having a passive hospital-based (outpatient) approach. In active community-based studies, investigators reported a raised incidence of RSV-associated ALRI compared with passive studies. Incidence varied widely across age groups (webappendix pp 15–16). Although incidence was usually the highest in the 0–5 month age group, some investigators reported a higher incidence in children aged 6–11 months.

All studies (apart from one) from developing countries estimating incidence for RSV-associated ALRI used active case ascertainment. Investigators using passive case ascertainment reported substantially lower incidence than did those using an active approach, as would be expected in developing country settings in which access to health services is limited (table 2; webappendix pp 7–11, 18–29). Hence, we based our incidence estimate on a summary of data from developing country studies with active case ascertainment only (table 2), as the most accurate reflection of the true value. We estimated that 33·8 (95% CI 19·3–46·2) million new cases of non-severe RSV-associated infection occurred globally in children younger than 5 years (including neonates) in 2005 (table 2), of which 96% of episodes were in developing countries (where 90% of the world's population aged younger than 5 years reside).

Table 2

Estimates of incidence and number of new cases of RSV-associated ALRI and severe ALRI in children younger than 5 years from studies§ with active and passive case ascertainment, by GBD region

		RSV-associated ALRI		RSV-associated severe ALRI
		Incidence in children aged <1 year (per 1000 per year)	Number of new cases in children aged <5 years in 2005 (×103)	Incidence in children aged <1 year (per 1000 per year)	Number of new cases in children aged <5 years in 2005 (×103)
Developing countries
Active
	Number of studies	6 (1)	6 (4)	4 (0)	4 (3)
	Median estimate	78·5 (33–116)	66 (27–94)	15·5 (13·5–38·6)	7 (4–14)
	Meta-estimate	74·2 (50·2–109·7)	59·1 (40–87·5)	22·3 (9·4–52·9)	8·3 (4·4–15·6)
Passive
	Number of studies	1	1	10 (4)	10 (3)
	Median estimate	NA	NA	15·4 (11–19)	4·7 (3–9)
	Meta-estimate	NA	NA	16·4 (13·2–20·5)	5 (3·7–6·7)
Active and passive
	Number of studies	7 (1)	7 (5)	14 (4)	14 (6)
	Median estimate	53 (33–116)	48 (27–94)	15·4 (13–19)	4·7 (4–10)
	Meta-estimate	68·4 (46·5–100·5)	54·6 (37·1–80·4)	17·9 (14·5–22·2)	5·6 (4·3–7·4)
Industrialised countries
Active
	Number of studies	0	0	0	0
Passive
	Number of studies	1	1	15 (4)	15 (10)
	Median estimate	NA	NA	18 (12–28)	4·7 (3–9)
	Meta-estimate	NA	NA	19 (14·6–24·7)	5·5 (4·2–7·2)
Global
	Developing*	59·1 (40–87·5)	32512	5·6 (4·3–7·4)	3080·7
	Industrialised†	24 (19·8–30)	1301·7	5·5 (4·2–7·2)	298·3
	Total‡	48·5 (31·4–74·9)	33813·7	5·6 (4·5–7)	3379

Data are number of studies (imputed number), median estimate (IQR), or meta-estimate (95% CI). References 35 and 36, Singleton et al, and the Epi study were excluded because these studies were of Indigenous populations in industrialised countries; references 19 and 38 excluded because of insufficient data. RSV=respiratory syncytial virus. ALRI=acute lower respiratory infection. GBD=global burden of disease.

Incidence estimates for RSV-associated ALRI based on the meta-estimate for active studies only and for severe ALRI on the meta-estimate for both active and passive studies.

Incidence for RSV-associated ALRI based on the estimate from one passive study only and incidence for severe ALRI on the meta-estimate from passive studies.

Total incidence for RSV-associated ALRI was calculated on the basis of the meta-estimate from active studies in developing countries plus the estimate from passive study in industrialised countries; total incidence for RSV-associated severe ALRI was on the basis of the meta-estimate from active and passive studies in developing countries plus the meta-estimate from passive studies in industrialised countries.

Reported incidences of RSV-associated severe ALRI were similar in studies with active and passive case ascertainment, which is consistent with increased health-service use for children with severe disease. We thus based the estimate of incidence of severe infection on data from studies with both active and passive case ascertainment. The estimated global incidence of severe disease in children younger than 5 years was roughly 3·4 million cases, of which 91% occurred in developing countries (table 2).

We identified 25 published and six unpublished studies providing case fatality ratio data for deaths due to RSV-associated severe ALRI in children admitted to hospital (webappendix pp 7–8). We considered separately the case fatality ratios for severe disease from hospital-based studies in developing and industrialised countries. Not all studies that were included in the younger than 5 years age group provided case fatality ratios for the full age range. Age-stratified mortality data from unpublished studies (Madhi et al, Nokes et al) show that mortality due to RSV-associated severe ALRI is very low beyond age 2 years, and the number of cases of severe infection also decreases sharply. Hence, we assumed that all deaths in severe disease occurred in the first 2 years of life. Table 3 shows median case fatality ratio estimates for developing and industrialised countries. In industrialised countries, almost all deaths due to RSV-associated severe ALRI occurred in infants, whereas in developing countries deaths also occurred in the second year of life.

Table 3

Estimated case fatality ratio in children younger than 5 years admitted to hospital for RSV-associated severe ALRI in industrialised and developing countries

	Number of studies included*		RSV-associated ALRI case fatality ratio (%)
	Children aged <1 year	Children aged <5 years	Children aged <1 year	Children aged <5 years
Industrialised countries	5	7	0·7% (0·3–4·8)	0·3% (0·2–0·4)
Developing countries	3	12	2·1% (1·6–2·2)	2·1% (1·3–3·4)

Data are number, or median (IQR). RSV=respiratory syncytial virus. ALRI=acute lower respiratory infections.

Seven studies reporting zero case fatality ratios were excluded from final analysis.

Approach 1 was based on the estimated yearly number of new cases of RSV-associated severe ALRI from hospital-based studies (table 2) and the case fatality ratios from children admitted with severe disease reported in hospital-based studies (table 3) separately for developing and industrialised countries. This approach showed that roughly 66 000 children younger than 5 years died in 2005 because of RSV-associated severe ALRI (panel 2). Approach 2 considered the median proportion of children admitted to hospital with severe disease who had hypoxaemia (0·36 [IQR 0·17–0·59]; webappendix p 14) and the related case fatality ratios reported for this group of children in resource-poor settings (28% and 14% in children admitted to hospital with a mixture of bacterial and viral pneumonias who had hypoxaemia at presentation). Adoption of the most conservative estimate of 14% leads to a crude estimate of roughly 155 000 deaths due to severe disease in children younger than 5 years in 2005 (panel 2). Data were insufficient to allow calculation of precise age-specific case fatality ratio estimates in children admitted to hospital, which would have yielded an improved estimate. Since this estimate includes only children admitted to hospital (a large underascertainment of all severe cases), we judged it to represent a plausible lower bound of RSV-associated severe ALRI mortality.

Only one study, in Lombok, Indonesia, reported cause of death in children not admitted to hospital, who were assigned by verbal autopsy and RSV isolations in the same population (mid-study analysis, reference 16; unpublished data, Gessner et al). Data were reported continuously during a 3-year period (figure 4, table 4). We estimated that the number of deaths calculated with approach 3 was 2·67 times higher than was estimated with approach 1. If we assume that these data are broadly representative of Indonesia, then 16% of all ALRI deaths are associated with RSV. If extrapolated to other developing country settings, this approach yields a crude estimate (for developing countries) of roughly 199 000 deaths attributable to with RSV-associated ALRI in young children for 2005 (panel 2). We think that this method is likely to overestimate the number of deaths because it assumes that all excess ALRI mortality during the RSV season is due to RSV, an assumption that is unlikely to be true, in view of the seasonality of other respiratory pathogens and the likelihood that RSV deaths occur outside the defined RSV season. Though data from Lombok were for children younger than 2 years only, this limitation is not likely to be important since most pneumonia deaths in children occur during the first 2 years of life.

Figure 4

ALRI-associated mortality pattern in children younger than 2 years in Lombok, Indonesia

RSV=respiratory syncytial virus. ALRI=acute lower respiratory infection.

Table 4

Estimated RSV-associated ALRI deaths in Indonesia based on total ALRI deaths occurring in and out of hospital in children younger than 2 years in Lombok, Indonesia

	Months of RSV season (a)	Average total ALRI deaths per month during RSV season (b)	Average total ALRI deaths per month outside RSV season (c)	Total ALRI deaths per year (d)	Proportion of ALRI deaths due to RSV (e)*	Indonesia RSV deaths†‡
2000	6	73	62	814	0·081	2887
2001	6	85	49	802	0·27	9623
2002	7	45	37	499	0·12	4277
Summary	..	..	..	..	..	5596§

RSV=respiratory syncytial virus. ALRI=acute lower respiratory infections.

e=[(b−c)a]/d.

35 640e.

Calculated with the WHO proportion estimate of 22% of all deaths in children younger than 5 years (19% of all deaths in children aged 1–59 months and a third of the 26% of all neonatal deaths) in Indonesia in 2005.

Calculated as the mean of individual year values.

In summary, insufficient data are available from which to make valid estimates of global mortality from RSV-associated ALRI. We have adopted three independent approaches with differing assumptions and limitations to obtain a rough data-derived estimate of the plausible lower and upper bounds for RSV-associated ALRI mortality. Our estimates of mortality are consistent with RSV being associated with roughly 3–9% of deaths from ALRI in young children. Data for Indonesia show substantial yearly variation in magnitude of RSV epidemic activity and associated ALRI deaths. This finding suggests that national, regional, and global RSV mortality also varies widely from year to year.

Discussion

We estimated that in 2005, at least 33·8 (95% CI 19·3–46·2) million episodes of RSV-associated ALRI occurred worldwide in children younger than 5 years, with incidence in developing countries more than twice that of industrialised countries. This estimate represents roughly 22% of all episodes of ALRI in young children. By comparison, 13·8 (10·8–17·2) million episodes of pneumococcal pneumonia and 7·9 (7·2–12·9) million episodes of Haemophilus influenzae type b pneumonia occurred in the same age group. However, this comparison should be interpreted with caution since the estimates for pneumococcal pneumonia and Haemophilus influenzae type b were calculated by extrapolation from data derived from vaccine probe studies. A substantial proportion of RSV-associated morbidity occurs in the first year of life, with incidence in infants that is twice or three times greater than is reported for children younger than 5 years overall. We also estimated that in 2005, 3·4 (2·8–4·3) million young children worldwide developed RSV-associated severe ALRI necessitating hospital admission, and 66 000–199 000 children younger than 5 years died from RSV-associated ALRI, with 99% of these deaths occurring in developing countries.

Estimates of RSV-associated ALRI incidence are highly variable within countries or regions and between regions. We cannot deduce how much of this variation is due to methodological differences and how much is due to variation in RSV epidemiology between study populations. Thus, the true uncertainty is wider than that expressed in a standard 95% CI. Three methodological factors affect estimates: method of case ascertainment; precise case definition for non-severe and severe episodes; and differences in sensitivity and specificity of diagnostic assays to identify RSV infection.

Hospital-based passive case ascertainment is likely to yield falsely low estimates of RSV-associated ALRI incidence, especially in developing countries. For example, Nokes and colleagues reported that only 20–25% of cases of RSV-associated severe infections that were referred from the community could be identified as hospital admissions, mainly showing low health-service use. Additionally, in some studies a substantial proportion of all children with ALRI were not tested for RSV for various reasons, resulting in falsely low estimates. In hospital-based passive ascertainment studies, distance from hospital is an important factor. Results of a study in the western region of The Gambia showed that the rate of hospital admission for RSV-associated severe ALRI was inversely proportional to the cost of getting to the health centre. Similarly, in the Kilifi hospital study, though overall incidence for RSV-associated ALRI was 11 per 1000 infants, incidence was 21 per 1000 infants in sublocations closest to hospitals. In one study, investigators attempted to keep this effect to a minimum by reimbursing patients for travel costs, but still showed that roughly 25% of children referred with RSV-associated severe ALRI did not attend hospital. This finding supports our decision to base the RSV-associated ALRI incidence estimate in developing countries on data from studies with active case ascertainment.

Case definitions varied from the WHO definition to physician-diagnosed ALRI (with or without chest radiographic confirmation) and unspecified acute respiratory infection that was severe enough to necessitate hospital admission. We noted that investigators using the WHO definition reported the highest incidences. For example, Nokes and colleagues reported in the Kilifi birth cohort study (webappendix p 17) that 75% of WHO-defined severe ALRI cases did not warrant admission to hospital, according to study clinical officers. The studies we included also differed in method of nasal sampling and in the specific diagnostic assays used, which will also have contributed to some of the variation in reported incidence estimates.

The RSV-associated ALRI incidence estimates are more likely to underestimate than to overestimate true incidence in developing countries. However, this conclusion is uncertain because of various factors affecting the estimate. First, estimation depends on the relative sensitivity and specificity of the WHO case definition for true ALRI—both are reported to be fairly high (median specificity of 86% for infants and 93% for children aged 1–4 years). Second, almost all studies identified RSV by ELISA or immunofluorescence assays, which have a 12–50% lower sensitivity than does PCR-based diagnosis. However, the overall effect of these factors on estimates depends also on relative test specificities, which are unknown for most studies. Finally, although we based our estimate on data from community-based studies with active case ascertainment and facilitated referral of patients to hospital, they could have still missed an unknown proportion of cases.

Substantial uncertainty surrounds case fatality ratio estimates from developing countries. First, 4–28% of children admitted to hospital with ALRI were not tested for RSV for various reasons (such as the child being critically ill, sampling not done on weekends or for night-time admissions, death before sampling, or refusal to get the test done; webappendix p 6). The absence of sampling in these children introduces a bias towards falsely low reported estimates because mortality tends to be higher in these groups. Second, we have estimated case fatality ratios separately for broad developing and industrialised country categories and the degree to which included studies are representative of these broad categories is unknown. We omitted three developing country studies reporting data from paediatric intensive care units and reduced bias in combination of incidence and case fatality ratio data by using only data for the incidence and case fatality ratios from cases admitted to hospital. More generally, these studies might be from settings with above-average resources, and thus report an underestimate of the true case fatality ratio in hospital settings in developing countries.

We based our estimate of the lower bound on reported incidence of RSV-associated severe ALRI necessitating hospital admission and on reported case fatality ratios in developing-country hospitals. However, not all cases of severe disease are admitted to hospital. Furthermore, hospital-based case fatality ratios from the studies we included cannot be regarded as representative of whole population groups. In the most resource-poor settings, the case fatality ratio for RSV-associated ALRI could be higher than these reported estimates and closer to that of bacterial pneumonias, especially since RSV-associated ALRI might predispose to bacterial infection. Results of studies from developing countries have shown variable rates of bacterial co-infection in RSV-associated ALRI, ranging from 3·5% in The Gambia to 31% in a study from Pakistan. In Lombok, Indonesia, peaks in ALRI case fatality ratios occurred after peaks in RSV case fatality ratios, potentially implicating bacterial co-infections (figure 4). During strategic planning to reduce childhood pneumonia deaths, co-attribution of mortality in patients with RSV and bacterial co-infection to RSV in addition to bacteria might be worthwhile.

Our estimate of the upper bound was based on only one study and so replication in other settings is needed. Additionally, although we attributed all excess ALRI mortality during the RSV season to RSV, several other viral and bacterial pathogens causing ALRI have seasonal patterns. We have thus not considered the context of multiplicity of other respiratory viruses (eg, human metapneumovirus or influenza A) that co-circulate during the RSV season and in some areas account for as much as 10–20% of ALRI hospital admissions and probably have a similar, if not higher, case fatality ratio to RSV.

The plausibility of our global mortality boundary estimates are supported by the internal consistency of the RSV-associated severe ALRI incidence, case fatality ratios, and RSV-associated ALRI mortality estimates. However, evidence to support valid and precise estimates of global RSV-associated ALRI mortality is of low quality. Research investment to gather further data is clearly needed—eg, by gathering RSV isolation data from ALRI patients in sites where demographic surveillance records community-based pneumonia mortality. Further large-scale unselected case series reporting age-specific case fatality ratios from many well described clinical settings in developing countries and large-scale post-mortem studies of ALRI that include investigation of possible RSV causes would also substantially improve the evidence base for this estimate. Until the widespread delivery of an effective RSV vaccine, measures such as promotion of health-service use, provision of regular oxygen supplies at health centres and hospitals, and immunoprophylaxis with monoclonal antibodies (when appropriate and affordable) can be expected to substantially reduce mortality associated with this disease.