Viral etiology of hospitalized acute lower respiratory infections in children under 5 years of age -- a systematic review and meta-analysis.

AIM
To estimate the proportional contribution of influenza viruses (IV), parainfluenza viruses (PIV), adenoviruses (AV), and coronaviruses (CV) to the burden of severe acute lower respiratory infections (ALRI).


METHODS
The review of the literature followed PRISMA guidelines. We included studies of hospitalized children aged 0-4 years with confirmed ALRI published between 1995 and 2011. A total of 51 studies were included in the final review, comprising 56091 hospitalized ALRI episodes.


RESULTS
IV was detected in 3.0% (2.2%-4.0%) of all hospitalized ALRI cases, PIV in 2.7% (1.9%-3.7%), and AV in 5.8% (3.4%-9.1%). CV are technically difficult to culture, and they were detected in 4.8% of all hospitalized ALRI patients in one study. When respiratory syncytial virus (RSV) and less common viruses were included, at least one virus was detected in 50.4% (40.0%-60.7%) of all hospitalized severe ALRI episodes. Moreover, 21.9% (17.7%-26.4%) of these viral ALRI were mixed, including more than one viral pathogen. Among all severe ALRI with confirmed viral etiology, IV accounted for 7.0% (5.5%-8.7%), PIV for 5.8% (4.1%-7.7%), and AV for 8.8% (5.3%-13.0%). CV was found in 10.6% of virus-positive pneumonia patients in one study.


CONCLUSIONS
This article provides the most comprehensive analysis of the contribution of four viral causes to severe ALRI to date. Our results can be used in further cost-effectiveness analyses of vaccine development and implementation for a number of respiratory viruses.

Viral etiology of hospitalized acute lower respiratory infections in children under 5 years of age -a systematic review and meta-analysis Acute lower respiratory tract infections (ALRI) are the leading cause of global mortality in children under five years of age (1,2). Studies of pre-school children from developed and developing countries alike suggest that the majority of respiratory infections generally have viral etiology (2)(3)(4). Clinically, ALRIs can be divided into pneumonias and bronchiolitis (5,6). Differentiating those two conditions can be particularly difficult in younger children, who typically exhibit less specific clinical symptoms (3,(7)(8)(9). In high-income countries (HIC), pneumonia rarely causes deaths in children (10), although it continues to be a major cause of morbidity and poses a significant economic burden (11). Bronchiolitis is characterized by a distressing pattern of symptoms: low-grade/absent fever progressing to cough, coryza, tachypnoea, hyperinflation, chest retraction, and widespread crackles or wheezes (12). Bronchiolitis deaths are very rare in HIC (13,14), but children are at increased risk of recurrent wheezing and the data on mortality in low and middle income countries (LMIC) are scarce (15).
Etiology of severe ALRI episodes is not well understood: limited contribution of the three major pathogens (S. pneumoniae, H. influenza, and respiratory syncytial virus) is established, but the role of other viruses has not been explored. The importance of viruses as major causes of ALRI is becoming increasingly apparent because the sensitivity of detection techniques has greatly improved and new molecular tests increasingly replace conventional methods. The use of polymerase-chain reaction (PCR) now allows identification of viruses that have previously been difficult or impossible to culture. In the past decade, numerous novel respiratory viruses that can cause ALRI have been discovered, and new diagnostic methods for the use in high and low-resource settings alike are continuously evolving (3,(16)(17)(18)(19)(20). It seems that the conventional diagnostic methods have systematically underestimated the role of viruses as causal pathogens in ALRI (3), and also that viruses are capable of causing severe, lifethreatening ALRI (3). The emergence of the severe acute respiratory syndrome (SARS), caused by a novel coronavirus, and the avian influenza type A (H5N1) outbreak are good recent examples (16,20). Impressive progress has been made in the last decade in increasing the global availability of vaccines against the main bacterial causes of ALRI -S. pneumoniae and H. influenzae type B -leading to marked reductions in both hospitalizations and deaths (21,22). This will lead to increased focus on viral causes and their prevention and management. Strains of influenza type A and B viruses can be life threatening (3), although infection in the majority of young children is vaccine-preventable (23,24). Parainfluenza viruses (PIV) are the most common cause of croup in young children, with PIV1 and PIV3 also being the causes of severe bronchiolitis and pneumonia (3,4,25), but there are currently no licensed PIV vaccines. Adenoviruses (AV) have long been recognized as pathogens of the lower respiratory tract that can be associated with severe or lethal lower respiratory tract infection (3,26,27) or bronchiolitis obliterans (28)(29)(30)(31). Coronaviruses (CV) cause common cold and have been historically thought to be a very rare cause of ALRI (32), despite the fact that they sporadically caused catastrophic disease in livestock (33). The SARS-CV outbreak in 2003, which was a highly virulent zoonosis capable of human-to-human transmission, renewed the interest in CV as human pathogens (32). This led to a discovery of two previously unrecognized CVs as causes of ALRI (16,17).
This study analyzed the available information on the role of four viruses (IV, PIV, AV, and CV), all of which have been historically considered to be relatively uncommon causes of severe ALRI in hospitalized cases. Our study did not assess the role of common causes -RSV, S. pneumonia, and H. influenzae -because their roles have already been systematically characterized and well-established (34). We are not aware of any systematic analyses of the global prevalence of viruses in severe childhood ALRI. We aimed to assess the proportion of cases of severe ALRI with a viral etiology and explore the contribution of mixed viral infections and separate contributions of IV, PIV, AV, and CV to severe ALRI in children under five years of age.

MetHodS
This systematic review was carried out using the PRISMA and MOOSE protocols (35)(36)(37). These protocols have been developed to ensure standardized and replicable approach to systematic review of the available evidence on the burden of specific health problems, the role of risk factors, or the effectiveness of available health interventions, and the unified reporting of the findings.

Literature search and inclusion criteria
A systematic literature review was performed using the search terms detailed in Supplementary online material. This was supplemented by hand searching of key online journals and reference lists of selected papers. The search included the following databases: Medline, EMBASE, CINAHL, Global Health Library, WHOLIS, LILACS, IndMed, AIM, SciELO, IMEMR, IMSEAR, WPRIM, and SIGLE (gray literature).
All studies included in the analysis reported on inpatients aged 0-4 years with a clinical diagnosis of communityacquired ALRI, bronchiolitis, or pneumonia. Investigation of viral etiology was a requirement and the participants needed to be free of co-morbid conditions. Children admitted to emergency departments were excluded, and so were intensive care patients wherever data was not reported for all other inpatients in the hospital, to avoid potential bias. We included studies conducted between 1995 and 2011 with a continuous study period of one year (or multiples of one year), to avoid effects of seasonality. Studies that relied solely on serology for diagnosis were excluded, because this method could not reliably differentiate acute from past infections (38,39). Studies that were conducted during an epidemic or pandemic outbreak were also excluded.

Study selection and data extraction
Study selection was performed following the removal of duplicates. Authors were contacted by email in cases when study data were not published in an extractable form, to collect further details. Data were extracted for study location, period of study, sample, diagnostic assay, clinical diagnosis, age range and median age of study population, potential etiological agents investigated, proportion of patients in whom no etiological diagnosis was found, viruses and bacteria detected, and age breakdown of patients by diagnosis where available ( Figure 1).

Assessment of bias within studies
During the process of data extraction, information was drawn from each study on possible sources of bias that could affect the results, such as: • Respiratory sample used (as there is no "gold standard;" samples from lower respiratory tract are preferable, but they require invasive procedures and are difficult to obtain without contamination from the upper airway; because of this, most studies consider nasopharyngeal aspirates for viral detection as acceptable, although acknowledging limitations. Viruses detected in the upper airway of a patient with ALRI are not necessarily pathogens of the lower respiratory tract); • HIV co-infection (as this is known to increase the susceptibility to ALRI and the rate of atypical infection) (40); • Viral detection technique and timing (as there is large variability in the sensitivity of different techniques; viral culture can only reliably be used within 2 days of onset of acute rhinorrhea, when viable virus shedding is at its peak (41); PCR can be used much later, because it does not require viable viruses in the sample, offering much improved sensitivity, but also greatly increased rates of detection of benign co-infections).

Summary measures
Proportional contributions of IV, PIV, AV, and CV to severe ALRI and associated confidence intervals were derived through meta-analysis using StatsDirect software package (StatsDirect Ltd, Academic version 2.7.9., Cheshire, UK). Due to large variation in methodology and patient demographics between studies, random effects models were used in all analyses, as proposed by DerSimonian and Laird (42). Heterogeneity and bias analyses were also performed for all meta-analyses. All presented results were shown to be free of publication bias, as demonstrated using funnel plots and analysis methods proposed by Begg (43), Egger (44), and Harbold (45). This triple approach represents robust protection from the sources of bias.

ReSuLtS
Fifty one studies meeting the inclusion criteria were included in this review, including 56 091 episodes of severe hospitalized ALRI. Figure 2 presents geographical distribution of the retained studies, Figure 3 shows proportion of studies investigating different viruses (any virus, RSV, IV, PIV, AV, CV), while Table 1 presents their basic characteristics in terms of case definition, sample size, period of study, and diagnostic methods used (15,40,. Only four studies investigated children hospitalized with ALRI for both bacterial and viral etiology (59,64,69,74) and only six studies reported HIV co-infection as their exclusion criteria (15,58,59,63,64,86). A total of 19 studies were from high-income countries (95), investigating on average 6.5 viruses per study, while studies in LMIC investigated 2.7 viruses (unpaired t test: P = 0.002). It seems likely that this difference reflects the fact that more tests are typically used in establishing diagnosis in high-income settings, without certainty over the causal role of all identified viral pathogens, and this may introduce systematic bias and heterogeneity between studies in HIC and LMIC. Figure 4 presents the results of meta-analysis of the proportion of children with severe ALRI aged 0-4 years in whom at least one virus was detected (including RSV). Only studies that investigated three or more viruses were included in this analysis -7 studies in total. This is an arbitrary cut off: these studies were deemed sufficiently active in their approach to detect viral infection, although the final result is likely to under-estimate the true burden. Pooled proportion was 50.4% (95% confidence interval [CI], 40.0% to 60.7%), with I˛ (inconsistency) parameter estimate of 97.0% (95% CI, 96.0% to 97.7%) ( Table 2).

Pneumonia and bronchiolitis
Bronchiolitis as a clinical diagnosis is useful in identifying children who can be presumed unlikely to benefit from antibiotics. Pneumonia (especially focal) is known to have a different spectrum of etiological agents and antibiotics are usually warranted. Six studies that differentiate between these conditions were analyzed separately. The proportion of viruses detected in the bronchiolitis analysis was 66.3% (95% CI, 56.2% to 75.6%) and in the pneumonia analysis 48.7% (95% CI, 38.0% to 59.4%) (Table 2).

Mixed viral ALRI
Seven studies investigated three or more viruses using PCR method and reported the proportion of hospitalized childhood ALRI where mixed viral infections were detected (ie, more than one viral pathogen was confirmed). A meta-analysis of those studies showed that pooled proportion was 15.3% (95% CI, 10.6%-20.5%), with I˛ parameter estimate of 91.3% (95% CI, 85.7%-94.0%). Further analysis, which only included virus-positive ALRI cases, estimated that at least 21.9% (95% CI, 17.7%-26.4%) of the viral ALRI were mixed ( Figure 5).

Proportion of hospitalized ALRI due to coronaviruses
The number of available studies on human CV was much smaller than on the other 3 viruses: 8 studies were retained after the initial review, but they did not provide sufficient epidemiological information on the role of CV to perform the meta-analysis and develop reliable estimates. CV are technically difficult to culture, and they were the sole virus detected in 4.8% of patients in one ALRI study (94) and were detected in 10.6% of virus-positive pneumonia patients in another study (59) (supplementary Figure).

dISCuSSIoN
This review estimated that 50% of all hospitalized ALRI in pre-school children, 66% of hospitalized bronciolitis episodes, and 49% of hospitalized pneumonia episodes showed viral involvement. All these estimates were derived using very heterogeneous sets of studies (I 2 >90% in many analyses). This heterogeneity is not surprising, given the large differences in study methods used, participants' ethnicity, climate and viral endemicity, to name a few. For this reason, random effects models were used, sacrificing statistical power to ensure estimates that would be as valid as realistically possible with available information.
In one third of the episodes of bronchiolitis in hospitalized patients, no virus could be detected, although bronchiolitis is expected to be almost exclusively of viral etiology. We  could hypothesize that this lack of sensitivity could be attributed both to imperfections of the tests and the timing of obtaining the sample. It is also possible that, in some studies, the diagnostic process for bronchiolitis can include some asthmatic (non-viral) patients (12). If we assume that the etiological estimates for pneumonia are subject to the same lack of sensitivity, this would mean that our estimates are likely to present a lower bound of the true role of viruses in all ALRI, and that the likely direction of bias is toward under-estimation of the true burden of viruses.
Detection of viral etiologies in hospitalized ALRI has been markedly increased by the use of PCR. We estimated that multiple viruses were involved in at least 15.3% of all cases of ALRI and 21.9% of virus positive ALRI cases. Both of those figures are likely to be underestimates, because they were based on studies that only tested for a limited number of viruses, and not for all known viruses (96).
The analyzed studies might have been affected by differences in regional practice: there were subtle differences in defining criteria for bronchiolitis in different areas (12) and inter-observer reliability in assessing some clinical signs has been low (97). Furthermore, some regions are prone to classifying tracheobronchitis and croup as LRTIs. According to several textbooks of respiratory and pediatric medicine these should be considered upper respiratory tract infections (URTIs) (5,6), because the vocal cords are not the division between upper and lower airway. Surprisingly, no study included in this review detailed the study populations' past vaccinations; in areas where vaccination against bacterial causes has been implemented, higher proportion of viral etiology would be expected. It also seems likely that a higher proportion of viral ALRI cases that are complicated by bacterial infection would be hospitalized than of those that are pure viral infections. It was beyond the scope of this study to consider virus seasonality or age breakdown for specific viral infections. In contrast to the other main respiratory viruses, PIV has been suggested to cause ALRI more frequently in summer months, while IV is thought to affect older children than RSV (3). Further work is necessary to elucidate these issues.
The delay between sample collection and establishing a diagnosis makes identifying causative agents of little practical use in the majority of acute cases of ALRI. Thus, good etiological epidemiology is important to guide management. There is an argument, both economic and humanitarian, for prevention of viral ALRI over cure. Antiviral agents (with the possible exception of neuraminidase inhibitors for IV) have been shown largely ineffective. Following the successes of vaccination programs for bacterial ALRI, this review makes the case for the growing importance of viral agents and provides the first comprehensive estimates for the burden of viral etiology other than RSV.
Our study conveys some very broad and general messages for the further development of global health policy. First, there seems to be a viral etiological component to at least half of all severe ALRI that require hospitalization, and this number is probably an underestimate for the reasons discussed in this study. This is somewhat unexpected, because it establishes a larger role for viruses in severe ALRI than generally presumed in international health community. Second, although RSV is a dominant viral cause, the role of influenza, parainfluenza, adenoviruses, and coronaviruses should not be neglected: they seem to be jointly responsible for at least a third of all viral severe ALRI and one in six of all severe ALRI. Given that respiratory viruses are amenable to prevention through vaccination, and that their role at the community level is likely to be larger than at the hospital level, our study should allow for modeling of cost-effectiveness of developing such vaccines. With a global roll-out of the existing vaccines against S. pneumoniae and H. influenzae, viral etiology of ALRI will come under increased focus, and understanding the burden associated with particular viral pathogens should help plan global prevention and save further lives.
Funding received from Bill and Melinda Gates Foundation.
ethical approval Not required.
declaration of authorship The study was initiated by HN and HC. PKK and FS collected and coded the data and performed data analysis, under supervision by HN and HC. PKK and FS made the first draft. The manuscript was conceptualized by IL and IR, who wrote and edited the final version and prepared it for submission. All authors read and approved the final manuscript.
Competing interests All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years; no other relationships or activities that could appear to have influenced the submitted work.