Systematic literature review assessing tobacco smoke exposure as a risk factor for serious respiratory syncytial virus disease among infants and young children

In this systematic literature review, we included studies of infants and children up to 5 years of age, including studies of children at high risk for serious RSV disease. We defined the high-risk population as patients with prematurity, BPD, or congenital heart disease. All other patient populations were defined as general populations, which consisted predominantly of children not defined as high risk.

Following PRISMA guidelines, we conducted a systematic review of the association of ETS exposure with serious RSV disease by searching broadly for studies assessing various risk factors, including ETS exposure, for RSV, bronchiolitis, or LRTI caused by RSV or bronchiolitis.[15] We searched the PubMed (including MEDLINE) and EMBASE databases for English-language studies published and indexed between 1990 and April 2009. No additional relevant unpublished studies were obtained. Studies that identified ETS exposure as a risk factor might be more likely to mention ETS in the title or abstract than studies that evaluated ETS exposure in a set of other risk factors, but did not find it to be significantly associated with increased risk. To minimize article selection bias across studies, we searched for articles that assessed any risk factors for RSV as well as bronchiolitis, because 50% to 80% of winter bronchiolitis is due to RSV in infants and young children [1]. For PubMed, the National Library of Medicine Medical Subject Headings (MeSH terms) used in the search included tobacco smoke pollution/adverse effects respiratory syncytial virus infections bronchiolitis viral, and respiratory tract infections in combination with the MeSH subheadings epidemiology etiology, and complications. The MeSH heading respiratory tract infections was combined with the keywords syncytial or RSV OR bronchiolitis or the MESH term tobacco smoke pollution/adverse effects. Studies that included risk assessment were identified using the following MeSH headings: analysis of variance (which includes multivariate analysis), probability (which includes proportional hazards model, odds ratio, risk, risk assessment, and risk factors); and case–control studies or cohort studies, which were combined with the terms relative risk OR hazard ratio OR odds ratio. The search identified studies with the following combination of these terms: (1) any disease-related term plus any risk assessment term; or (2) any disease-related term plus the smoke exposure term. We did not search for unpublished studies or reports.

Each relevant article was read by several authors and the bibliographies of included articles were reviewed for additional studies. We excluded studies that did not assess disease risk or did not include ETS exposure as a risk factor. Figure 1 presents a summary of exclusions and rationales for exclusion during successive rounds of review.

Among the studies meeting the inclusion criteria, we evaluated the risk of bias at the outcome and study level using the Cochrane risk of bias tool [16]. The Cochrane risk of bias tool was developed primarily for use in interventional studies; we used the tool to examine factors specific to sources of bias frequently found in observational studies. The studies were examined for risk of bias that would potentially influence the association of ETS exposure with serious RSV disease. The studies were judged regarding evidence of misclassification bias of RSV disease, selective reporting bias, confounding bias, exposure ascertainment bias, or participant selection bias. Disease misclassification bias was assessed based on the likelihood of the study population having an LRTI that was not due to RSV. We expect that disease misclassification would underestimate the association between ETS exposure and serious RSV disease. Selective reporting bias was evaluated based on whether outcomes related to risk factors were clearly reported. Confounding bias was assessed based on whether the study controlled for the effects of other variables through multivariate analysis. We expect that unadjusted results may overestimate the association between ETS exposure and serious RSV disease. Exposure ascertainment bias was assessed based on whether the ETS exposure variables were described clearly and determined by the study authors to be adequate measures of ETS exposure in the study population. Finally, participant selection bias was pointed out in case–control studies in which the control group consisted of ill patients with non-RSV respiratory disease, and therefore, the effect of ETS exposure on serious RSV disease could be biased toward the null.

The results are organized by study characteristics and include a bias assessment summary for each study. (The full bias assessment is presented in Additional file 1: Table S1) Table 1 presents 14 studies that established a diagnosis of RSV disease by confirmatory laboratory testing. These studies compared children with RSV infections with children without respiratory illness and assessed whether ETS exposure increased the risk for developing a serious RSV infection. Table 2 presents 8 studies that did not confirm suspected cases of RSV disease with laboratory testing, but relied on the clinical diagnosis of RSV or bronchiolitis, often from medical charts or insurance data. These studies also compared children with RSV or bronchiolitis infections with children without respiratory illness. Table 3 presents 8 studies that examined the impact of ETS exposure on the severity of RSV disease as measured by hospitalization or degree of hypoxia. These studies started with a sample of children infected with RSV and assessed whether ETS exposure increased the severity of the RSV infection. Within each group of studies, we considered studies of premature infants separately because this population is at higher risk for serious RSV disease than the general population of children [17, 18]. We also present the results by study design (e.g., cohort, case–control) in each table.

The methods used to assess ETS exposure status varied widely and included mother’s prenatal smoking status, mother’s postnatal smoking status, father’s smoking status, smoking in the home, number of smokers in the household, number of cigarettes smoked in the home, more than five cigarettes smoked per day in the home, days of smoke exposure, history of exposure to smoking, smoking by daycare provider, and cotinine levels.

Some of the studies summarized included more than one source of ETS exposure, but no study detailed missing ETS exposure status data. In the studies reporting multiple ETS exposure, we have presented all results based on the different sources (e.g., maternal smoking during pregnancy and smokers in the household) because it is difficult to determine in each setting which of several proxies for total ETS exposure results in the least misclassification error. Few of the multivariate studies provided a rationale for their selection of control variables. We therefore included positive bivariate results in our tables when bivariate and multivariate analyses produced disparate conclusions.

The studies in Table 1 seek to answer the question, does ETS exposure increase the risk that an uninfected child will develop serious RSV? In 14 studies, the diagnosis of RSV infection was confirmed by laboratory testing (Table 1). Of these, 12 studies showed a significant adverse impact of ETS exposure on serious RSV in bivariate or multivariate analysis as measured by at least one exposure variable. These studies examined the association of ETS exposure and other factors on the risk of ED visit [23] or hospital admission [5, 6, 9, 10, 12, 19‐21, 24‐26] for laboratory-confirmed RSV LRTI. Two studies examining laboratory-confirmed RSV did not find a statistically significant association between ETS exposure and serious RSV: a cohort study of outpatients in the United States assessing the risk of a child presenting to the pediatrician’s office with an RSV infection [22], and an Australian case–control study assessing the risk of RSV hospitalization [11].

Four of these 14 studies contained at least one significant association, but found mixed results with different ETS exposure variables [6, 19, 20, 23]. One prospective cohort study in premature infants found that maternal smoking and parental smoking in the home had conflicting associations with three different outcomes: RSV LRTI, all-cause hospital admission, and length of hospital stay [19]. Measures of smoke exposure in the home and maternal smoking gave conflicting results when evaluated in 2 studies [6, 20], including a study in premature infants [20]. Finally, a small case–control study in a general population in Turkey found that several exposure variables (only mother smoking, both parents smoking, any parent smoking, and serum cotinine) were associated with significant increase in risk of admittance to the ED; however, only father smoking was significantly associated with a decrease in risk [23].

Of the studies of laboratory-confirmed RSV hospitalization described in Table 1, 5 were conducted in premature infants and 9 were conducted in general populations. All 5 studies of laboratory-confirmed RSV in premature infants found increased risk related to ETS exposure in either bivariate or multivariate analyses [9, 10, 19‐21] (Table 1). The 2004 case–control study of infants with a gestational age of 33 to 35 weeks by Figueras-Aloy et al. [21] found an OR of 1.62 in the bivariate analysis that did not remain significant in the multivariate analysis. However, the same researchers published a much larger cohort study in 2008, in a premature population with a gestational age of 32 to 35 weeks, reporting a nearly identical aOR (1.61) for maternal smoking that remained significant in the multivariate analysis [20]. The authors of the 2004 study note that the prevalence of ETS exposure decreased in Spain during the years before the study (52% during the 1999–2000 RSV season compared with 30% in this study) because of anti-smoking campaigns or health education [21]. The significant association in the larger study suggests that the earlier study was underpowered for the multivariate analysis.

Of the 9 studies of laboratory-confirmed RSV illness in general populations described in Tables 1 2 were prospective cohort studies and 7 were case–control studies (Table 1). Risk of RSV LRTI in an outpatient office setting [22] and risk of RSV hospitalization from ETS exposure (smoking in the household, but not from maternal smoking during pregnancy) [6] were not significant in the cohort studies.

The 7 case–control studies of laboratory confirmed RSV illness in general populations ranged in size from 20 to 2,564 cases (Table 1), and all found a significant association between ETS exposure and risk of RSV hospitalization in either the multivariate or bivariate analyses, with the ORs from the 4 largest studies clustering tightly between 1.35 and 1.6. The 3 smallest case–control studies included in Table 1 did not provide ORs but compared ETS exposure prevalence between the cases and healthy controls [23‐25]. All 3 found significant differences in ETS exposure in cases versus controls. The Turkish study also demonstrated significantly higher cotinine levels in serum samples from cases than from healthy controls [23]. In a case–control study of Alaska native children, risk was apparent in the bivariate analysis, but not in the multivariate analyses [12] (Table 1). The authors noted that misclassification of exposure status may have been a problem. The study was conducted in Alaska, where children spend much time indoors during the winter RSV season and could be exposed to ETS when visiting other houses or community buildings. This appears likely, because there were smokers living in the households of 59% of the controls and 68% of the cases. The consistency of findings associating ETS exposure and increased risk of serious RSV, coupled with the inconsistency of the variables associated with these significant results, highlights the challenges of assessing and categorizing ETS exposure.

As with the studies in Table 1, the studies in Table 2 seek to answer the question, does ETS exposure increase the risk that an uninfected child will develop serious RSV? In these studies, however, the RSV diagnosis was based on clinical findings rather than a laboratory test. Table 2 presents 8 studies, 6 of which reported a significant association between ETS exposure and bronchiolitis hospitalization or outpatient treatment. Because other organisms may cause bronchiolitis, there may be possible misclassification of RSV disease status [33].

The large general population cohort studies including bronchiolitis hospitalization as an outcome by Boyce et al. [4], Carroll et al. [28], and Koehoorn et al. [29] are in agreement with the 12 studies in Table 1 that identified an increased risk of RSV hospitalization associated with ETS exposure. Carroll and colleagues also demonstrated increased risk of ED visit or clinic visit for bronchiolitis and bronchiolitis diagnosis related to ETS exposure [28].

In other studies assessing risk of bronchiolitis LRTI in children presenting in the office setting, 2 studies found an increased risk for at least one ETS variable [28, 31], whereas 2 other studies did not [29, 30]. Of these, the positive study by Carroll et al. [28] had the greatest statistical power.

Only the Holman study assessed risk of death from bronchiolitis; the risk was increased by maternal smoking (aOR, 1.6) [3].

The 8 studies in Table 3 address the question, given that a child has contracted RSV, does ETS exposure increase the severity of illness? Four studies compared ETS exposure in children hospitalized for RSV and children with RSV who were not hospitalized. Three of these studies, including a small study of premature infants with BPD, found no effect of ETS exposure on the likelihood of hospitalization for RSV [2, 32, 34]. One small study found an effect of ETS exposure, but only 40% of the hospitalized children were found to have RSV infections, with the remaining 60% representing bronchiolitis caused by other viruses [33].

Four studies examined clinical severity (degree of hypoxemia) of RSV LRTIs among children hospitalized for treatment. All 4 found that ETS exposure was associated with more severe illness (Table 3). Bradley et al. found that hypoxemia was associated with postnatal but not prenatal smoke exposure [36]. Chatzimichael et al. found that breastfeeding was protective against the effect of ETS exposure on disease severity [37]. Based on a small number of studies of RSV severity, ETS exposure does not appear to increase the risk of hospitalization versus outpatient treatment among children infected with RSV, but hospitalized RSV patients exposed to ETS have more severe disease.

As the studies reviewed were observational, we evaluated the primary sources of bias that could potentially affect the estimates of association with ETS exposure (Additional file 2). We found that 21 of the 30 studies confirmed the RSV diagnosis through laboratory testing for RSV or multiple respiratory illnesses including RSV in all of the study populations [2, 5, 6, 9‐12, 19‐26, 32‐36, 38]. The remaining studies had a risk of disease misclassification bias of RSV because diagnostic codes (primarily for bronchiolitis) or diagnoses from medical records were used [3, 4, 27‐31, 37, 39]. Misclassification of RSV disease may attenuate the association of ETS exposure and severe RSV disease, and the studies of clinically diagnosed RSV patients may provide conservative estimates of ETS exposure risk.

Overall, we identified few studies that had potential selective reporting bias, confounding bias, exposure ascertainment bias for ETS, or participant selection bias. Selective outcome reporting was a potential problem in 2 studies [38, 39], meaning that outcomes were not described sufficiently to determine which factors were included and potentially controlled for in the analysis. Although confounding is a threat across observational studies (because it is difficult to control for both measureable and unmeasurable factors), multivariate analyses were not conducted in a number of studies [23‐25, 32, 34, 38], which could lead to an overestimate of the effect of ETS exposure. One study had a high risk of participant selection bias, where both cases and controls were diagnosed with bronchiolitis [33]. Only 2 studies had an unclear risk of ETS exposure [12, 35]. Al-Sonboli and colleagues [35] did not describe how the ETS exposure data or other demographic data were obtained. In the Bulkow et al study, ETS exposure was highly prevalent in the community (in both cases and controls) and indoor visiting was frequent during the winter RSV season [12]; therefore, the ETS exposure variable “smoker in household” was not an adequate proxy for exposure in this population.

No statistical analysis of the risk of bias across studies was performed for this review.