Association of Secondhand Smoke Exposure with Pediatric Invasive Bacterial Disease and Bacterial Carriage: A Systematic Review and Meta-analysis

Majid Ezzati and colleagues report the findings of a systematic review and meta-analysis that probes the association between environmental exposure to secondhand smoke and the epidemiology of pediatric invasive bacterial disease.


Introduction
Invasive bacterial disease (IBD) is an important cause of child mortality in developing and developed countries [1][2][3][4][5][6][7], accounting for at least as many child deaths as HIV/AIDS and malaria combined [6][7][8]. The organisms responsible for most pediatric IBD cases are S. pneumoniae, Haemophilus influenzae type B (Hib), and N. meningitidis [1,2,4,6,7,9]. In 2000, there were an estimated 14.5 million cases and 826,000 deaths from pneumococcal disease in children under 5 with estimated incidence ranging from 544 per 100,000 in the Americas to 1,778 per 100,000 in Africa [6]. The burden of Hib was estimated at 8.13 million cases and 371,000 deaths with estimated incidence ranging from 504 per 100,000 in Europe to 3,627 per 100,000 in Africa [7]. While there are currently no analyses of global invasive meningococcal disease burden, regional estimates of incidence range from 0.3 to 4 cases per 100,000 in North America to as high as 1,000 cases per 100,000 in the so-called ''meningitis belt'' in sub-Saharan Africa [9].
Secondhand smoke (SHS; also referred to as involuntary smoking, passive smoking, or environmental tobacco smoke [ETS]) has been shown to increase the risk of several adverse outcomes in children, including lower respiratory tract infections, middle ear infection, asthma, and sudden infant death syndrome [10,11]. Since the 1980s, epidemiologic studies have also found an association between SHS exposure and IBD or bacterial carriage, including those related to N. meningitidis, Hib, and S. pneumoniae, which suggests that SHS might be an independent risk factor for IBD. Given the persistent or growing SHS exposure in developing countries, especially in Asia, where IBD poses a major health risk, it is essential to delineate the role of exposure to SHS in the epidemiology of IBD. To our knowledge, no reviews have systematically investigated the quality and consistency of epidemiological evidence on this association, which is an important gap in our understanding of the effect of SHS exposure on the burden of infectious diseases. We carried out a systematic review and quantitative assessment of the association between SHS and the risk of IBD from N. meningitidis, Hib, and S. pneumoniae in pediatric populations, aged 1 mo to 19 y (see below). We also included the effects of SHS exposure on pharyngeal carriage of these three bacteria because asymptomatic carriage is also associated with clinical disease [12][13][14][15][16].

Search Strategy
We carried out a systematic literature search of SHS and IBD using Medline (via PubMed), and EMBASE from 1975 through December 2009. PubMed was searched by combining two separate queries composed of medical subject heading (MeSH) and text word (tw) keywords for the exposure and outcome of interest. The first (exposure) query was searched using the following exploded headings and independent terms: ''(cigarette smoke [

Inclusion and Exclusion
Two reviewers independently identified articles eligible for in-depth examination using the following inclusion and exclusion criteria. Studies were included if at least one of the following outcomes was analyzed: invasive S. pneumoniae disease, invasive Hib disease, invasive N. meningitidis disease, and naso-or oropharyngeal carriage of any of the above three bacteria. IBD was defined as bacterial meningitis, bacterial epiglottitis, bacteremia, or microbiologically documented infection at other normally sterile sites with relevant clinical syndrome. Relevant exposures were defined as SHS or ETS exposure, parental smoking, household smoking or presence of household smoker(s), and regular contact with smokers. We excluded studies in which active smoking was the only exposure, active smoking was not distinguished from passive smoking, or studies that also included prenatal exposure. Study types included were cohort, case-control, and cross-sectional surveys, whereas case reports, review articles, editorials, and clinical guidelines were excluded. We included studies on human participants aged 1 mo to 19 y, i.e. infants, children, and adolescents. We excluded the neonatal period because of its established epidemiologic and pathophysiologic distinction from the post-neonatal period [17]. We included adolescents because age-specific N. meningitidis incidence peaks in childhood as well as adolescence; while S. pneumoniae incidence peaks in childhood and infancy, this disease may also occur in adolescents [4,6,9].
Studies on immunocompromised populations were excluded. When multiple articles reported on the same study population, we included only the most detailed publication that met the inclusion criteria. Any discrepancies on articles meriting inclusion between reviewers were resolved by a consensus meeting of three authors (C-CL, NAM, and ME). Study selection is summarized in Figure 1.

Data Extraction and Synthesis
Data were extracted on study location, setting (e.g., community, school, hospital, etc.), population characteristics including age range and sex ratio, number of participants, definition of exposure and diagnosis of outcome, crude and adjusted effect sizes as available, and confidence intervals (CIs). We also recorded quality indicators of study design including presence of appropriate controls and covariates used for adjustment in multivariate analysis. We conducted separate analyses on IBD and bacterial carriage. When studies were identified as containing pertinent data not included in the published article (e.g., when they did not differentiate between pediatric and adult participants), we contacted the authors to obtain the missing data. When a response was not provided and raw data were provided in the article, we manually calculated the unadjusted odds ratio (OR) for inclusion in the meta-analysis. Otherwise, such articles were excluded.

Statistical Analysis
We followed the PRISMA guidelines for meta-analysis of observational studies in our data extraction, analysis, and reporting (Text S1) [18]. Heterogeneity was tested using the Cochran Q statistic (p,0.05) and quantified with the I 2 statistic, which describes the variation of effect size that is attributable to heterogeneity across studies [19,20]. The value of the I 2 statistic was used to select the appropriate pooling method: fixed-effects models were used for I 2 ,50% and random-effects models for I 2 $50% [19,20]. CIs of I 2 were calculated by the methods suggested by Higgins et al. [21]. Pooled ORs were summarized with Mantel-Haenszel method for fixed-effect models and DerSimonian and Laird method for random effect models [20]. Galbraith plots were used to visualize the impact of individual studies on the overall homogeneity test statistic [22]. Meta-regression was used to evaluate whether effect size estimates were significantly different by specific study characteristics and quality factors, particularly those of adjustment for covariates and whether IBD diagnosis was only the more rigorous laboratory-confirmed or a mix of clinicalonly and laboratory-confirmed diagnosis. We defined a study as having laboratory-confirmed diagnosis as the primary outcome if the study had more than 80% of cases confirmed by a positive culture, rapid antigen test, or PCR-based identification. In addition to meta-regression, we reestimated effects size stratified on the same study characteristics and quality factors, so that they are available as separate estimates. Even when the meta-regression result was not statistically significant, we conducted a subgroup analysis when a study characteristic was clinically or epidemiologically relevant, e.g., the age range of study participants.
The presence and the effect of publication bias was examined using a combination of the Begg and Egger tests and the ''trim and fill'' procedure [19,23,24]. This procedure considers the hypothetical possibility of studies that were missed, imputes their ORs, and recalculates a pooled OR that incorporates these hypothetical missing studies [23]. Trim-and-fill ORs are reported when the tests for publication bias were significant.

Search Results and Study Characteristics
Our search identified a total of 198 studies, of which 95 were from PubMed and 103 from EMBASE. Screening based on title and abstract identified 62 citations for full-text review ( Figure 1). An additional 27 studies were identified from reference lists of the identified articles and from other databases. Of the 89 potentially relevant articles, 47 were excluded for reasons in Figure 1, leaving a total of 42 studies that met the inclusion criteria.
There were 30 studies on invasive disease and 12 on pharyngeal carriage. All invasive disease studies utilized a case-control design and all carriage studies were cross-sectional studies. The studies varied in their SHS exposure metric, age range, and case definition ( Table 1). Household smoking or the presence of household smokers was the most frequent measure of SHS exposure (26 of 42 studies), while most others used maternal, paternal, or caregiver smoking. Nineteen of 30 invasive disease studies used laboratory-confirmed diagnosis as the case definition for all participants, while others had individuals with a combination of clinical-only and laboratory-confirmed diagnosis. Where clinical diagnoses were included, they were specific to the organism, including hemorrhagic rash for invasive meningococcal disease and epiglottitis for invasive Hib disease. The inclusion of carefully clinically defined cases has the potential to reduce bias in settings where pre-referral antibiotic treatment is common, either by policy or because of wide availability of antibiotics in the community. All 12 bacterial carriage studies used a culture definition, eight from nasopharyngeal swabs and the remainder from oropharyngeal swabs. The studies also varied in covariates adjusted for as shown in Table 1.
Invasive pneumococcal disease. The four case-control studies on SHS exposure and invasive pneumococcal disease included a total of 412 cases and 842 controls (Table 1) [35,36,42,43]. Combined results from all studies yielded a nonsignificant association (pooled OR 1.21, 95% CI 0.69-2.14; test of heterogeneity p = 0.098, I 2 = 52.3%) ( Figure 2B). Once again, meta-regression did not show any significant effect size modification by specific study characteristics considered ( Table 2).    (Figure 2C). Excluding a study in Finland that was a potential source of heterogeneity led to a lower pooled effect size, which remained nonsignificant (OR 1.18, 95% CI 0.99-1.41) [46]. Of study characteristics assessed in meta-regression, only being among preschool children was statistically significant (  [57]. However, meta-regression analysis indicated that the difference between adjusted and crude effect sizes had borderline statistical significance (p = 0.061).
S. pneumoniae carriage. There were five cross-sectional surveys on SHS exposure and pharyngeal carriage of S. pneumoniae  Figure 3B). Adjustment or study characteristics did not significantly modify the effect size in meta-regression analysis ( Table 2). Subgroup analysis on the three studies with multivariate   [64,65]. The pooled association was nonsignificant (OR 0.96, 95% CI 0.48-1.95; test of heterogeneity p = 0.56, I 2 = 0%) ( Figure 3C). The study population did not include the most vulnerable age group, i.e., preschool children.

Dose-Response Relationships
Dose-response relationships were examined in four invasive meningococcal disease studies [28,32,36,39], three invasive Hib disease studies [47,50,51], and one invasive pneumococcal disease study [36]. The studies had used different metrics to measure exposure and dose including number of cigarettes smoked per day and number of household smokers. The absence of a consistent definition of exposure meant that a pooled analysis of the doseresponse relationship was not possible. Broadly, with the exception of the pneumococcal study and one Hib study, there was a doseresponse relationship with the number of cigarettes smoked per day or the number of smokers in the household [28,32,36,39,47,50,51].

Publication Bias
The test for publication bias was significant in three of the six outcomes, namely invasive meningococcal and pneumococcal diseases and N. meningitidis carriage ( Table 3). The trim-and-fill ORs for meningococcal disease and N. meningitidis carriage were lower but the former remained statistically significant. The positive, but nonsignificant, OR of 1.21 (95% CI 0.69-2.14) for pneumococcal disease was replaced by a trim-and-fill OR of 0.83 (95% CI 0.45-1.53).

Discussion
This systematic review and meta-analysis of studies on the association between SHS exposure and IBD or pharyngeal carriage of pathogenic bacteria in pediatric populations revealed a consistent and positive association between SHS exposure and invasive meningococcal disease and pharyngeal carriage of N. meningitidis, as well as a positive association with S. pneumoniae carriage. There was also a positive but not statistically significant association with invasive pneumococcal and Hib diseases. The association with Hib carriage was based on only two studies and was null. When subanalyses could be conducted, the pooled effect sizes with and without adjustment for important risk factors were generally similar, becoming slightly smaller for meningococcal and pneumococcal carriage and for invasive Hib disease, and larger for invasive meningococcal and pneumococcal diseases. Studies with laboratory-confirmed diagnosis, the more rigorous outcome, had large and statistically significant effect sizes for meningococcal disease but not for pneumococcal and Hib diseases.
The nonsignificant associations with invasive pneumococcal disease may have been partially due to the relatively small pooled sample sizes (412 cases), whereas that of Hib disease is less likely to be due to sample size (1,228 cases). For Hib disease, studies on the most vulnerable ages (#6 y old) had larger and significant effect estimates. A factor that may have contributed to the insignificant effects may be the increasing use of Hib vaccine (since 1990) and pneumococcal conjugate vaccines (since 2000). Studies before the vaccine era had larger effect sizes for both Hib and pneumococcal diseases, but these were only statistically significant for Hib. These factors, and the strong association that has been observed between active smoking and invasive pneumococcal disease [66], should motivate additional high-quality studies with large sample sizes to clarify the role of SHS in the etiology of invasive pneumococcal disease.
There are plausible mechanisms for the effects of SHS on bacterial diseases. Both in vivo and in vitro experimental studies have found that SHS exposure may induce structural changes in the respiratory tract including peribronchiolar inflammation and fibrosis, increased mucosal permeability, and impairment of the mucociliary clearance [67,68]. It may also decrease immune defenses, e.g., a decreased level and depressed responses of circulating immunoglobulins, decreased CD4+ lymphocyte counts and increased CD8+ lymphocyte counts, depressed phagocyte activity, and decreased release of proinflammatory cytokines [68][69][70][71][72]. All these mechanisms might increase the risk of bacterial invasion and subsequent infection. The significant findings here regarding the association of SHS exposure with bacterial carriage also support a plausible etiological role for SHS in invasive bacterial disease, because asymptomatic carriage is an intermediate step towards invasive disease [12,13,15,16]. Asymptomatic carriage itself has a public health implication because it is important in population transmission of infectious bacteria [12,15].

Strengths and Limitations
This systematic review has strengths and limitations. To the best of our knowledge, this is the first systematic review of the epidemiologic evidence on the association between SHS exposure and pediatric IBD. We were able to include clinical invasive disease as well as the etiologically and epidemiologically important intermediate stage of asymptomatic bacterial carriage. As far as possible, we assessed sensitivity to important methodological design and quality characteristics using meta-regression and subgroup analysis. Our search covered multiple databases without language limitation.
A key limitation of our study was the relatively small number of studies, specifically from developing countries in which the IBD burden is the largest, smoking is increasing, and vaccination coverage may be lower. Notably, two African studies had a nonsignificant effect size for invasive meningococcal disease [30,34]. One of these studies was from northern Ghana, where more than 93% of participants were exposed to fuelwood smoke [30]. In addition, negative residual confounding (due to the potential negative association of household smoking with economic status) cannot be ruled out as a source of the nonsignificant negative finding. The second study, from urban South Africa, had included an interaction term between recent upper respiratory tract infection (URTI) and SHS exposure in the multivariate analysis, which had an OR of 3.6 (95% CI 1.4-7.3) [34]. If recent URTI itself was caused by SHS and can increase the risk of IBD, then this adjustment would attenuate the true effect of the SHS term. A third African study from The Gambia had an OR of 2.99 (95% CI 1.10-8.14) for pneumococcal disease [35]. The limited number of studies from developing countries makes it difficult to assess the role of factors such as background incidence rate, nutritional status, vaccination, and coexposure to wood smoke on the ORs for SHS-IBD association. A second potential limitation is that both the exposure and outcome measurements may have been subject to error, which is likely to have biased our results towards the null and reduced its significance. Third, heterogeneity of effect sizes across studies restricts our ability and confidence to generalize the results of this pooled data analysis to all populations. Fourth, the studies on association with bacterial carriage were distinct from those on IBD and no SHS-IBD studies had assessed carriage at baseline. As a result, we were not able to assess whether SHS exposure only increases the risk for colonization, or increases the risk of subsequent infection, or both. Fifth, because IBD is a complex disease with multiple causes, there is potential for residual confounding in the observational studies included in our analysis. This is especially relevant given that the currently available SHS-IBD studies were case-control or nested case-control studies and those on SHS-bacterial carriage were cross-sectional studies. Our findings on the potential causal associations should motivate new prospective studies. Sixth, our study focused on pediatric SHS exposure and did not assess studies on perinatal SHS exposure as a risk factor for IBD, with some of the effect possibly mediated through low birth weight. [73,74]. Finally, while we assessed the potential for publication bias and report trim-and-fill ORs, the latter estimates are themselves subject to methodologic limitations especially when the number of studies is small [75].
Despite the limitations of current epidemiologic studies, our meta-analysis provides some evidence of an association between SHS and IBD and pharyngeal carriage, especially in preschool children. Although there are efficacious vaccines against all three pathogens assessed in this study, many children in low-income countries are not covered in routine immunizations and have limited access to case management [76][77][78][79][80]. Vaccine pricing remains an obstacle to uptake, while waning immunity and serotype replacement may undermine long-term vaccine effectiveness [78,[81][82][83]. Thus scaling up vaccine coverage and case management must be accompanied by nonvaccine interventions, such as environmental interventions, to address the large burden of IBDs. Tobacco smoking and SHS exposure have increased in low-income and middle-income countries, making SHS exposure a global problem [84,85]. While public smoking bans have been effective in reducing adult SHS exposure and adverse health effects [86,87], children's exposure to SHS may occur at home, where bans may be difficult to enforce [11]. An estimated 700 million children worldwide are exposed to SHS at home [84]. If the observed effects are causal, our results indicate that in a population where 25% of young children are exposed to SHS (e.g., Brazil or South Africa), 5%-20% of IBD cases may be attributable to this risk factor; the attributable fraction would be 10%-34% in populations where exposure is 50% (e.g., Egypt or Indonesia). Effects of such magnitude should motivate a number of research and intervention steps specifically related to SHS and pediatric IBD: Firstly, there should be well-designed prospective studies with high-quality measurement of exposure, outcome, and potential confounders to overcome the limitations of the current studies. Secondly, interventions that specifically focus on reducing children's exposure at home, schools, and other environments should be pursued. These two directions are particularly important in developing countries where the IBD burden is high and exposure to SHS is high or increasing. Finally, the effects of other combustion pollutant sources that are common in developing countries, especially smoke from wood and animal dung fuels, on IBD should be subject to research.

Supporting Information
Text S1 PRISMA checklist. Editors' Summary Background. The deleterious health effects of smoking on smokers are well established, but smoking also seriously damages the health of nonsmokers. Secondhand smoke (SHS), which is released by burning cigarettes and exhaled by smokers, contains hundreds of toxic chemicals that increase the risk of adults developing lung cancer and heart disease. Children, however, are particularly vulnerable to the effects of SHS exposure (also known as passive smoking) because they are still developing physically. In addition, children have little control over their indoor environment and thus can be heavily exposed to SHS. Exposure to SHS increases the risk of ear infections, asthma, respiratory symptoms (coughing, sneezing, and breathlessness), and lung infections such as pneumonia and bronchitis in young children and the risk of sudden infant death syndrome during the first year of life.
Why Was This Study Done? Several studies have also shown an association between SHS exposure (which damages the lining of the mouth, throat, and lungs and decreases immune defenses) and potentially fatal invasive bacterial disease (IBD) in children. In IBD, bacteria invade the body and grow in normally sterile sites such as the blood (bacteremia) and the covering of the brain (meningitis). Three organisms are mainly responsible for IBD in children-Streptococcus pneumoniae, Haemophilus influenzae type B (Hib), and Neisseria meningitidis. In 2000, S. pneumonia (pneumococcal disease) alone killed nearly one million children. Here, the researchers undertake a systematic review and meta-analysis of the association between SHS exposure in children and two outcomes-IBD and the presence of IBD-causing organisms in the nose and throat (bacterial carriage). A systematic review uses predefined criteria to identify all the research on a given topic; metaanalysis is a statistical method that combines the results of several studies. By combining data, it is possible to get a clearer view of the causes of a disease than is possible from individual studies.
What Did the Researchers Do and Find? The researchers identified 30 case-control studies that compared the occurrence of IBD over time in children exposed to SHS with its occurrence in children not exposed to SHS. They also identified 12 cross-sectional studies that measured bacterial carriage at a single time point in children exposed and not exposed to SHS. The researchers used the data from these studies to calculate a ''summary odds ratio'' (OR) for each outcome-a measure of how SHS exposure affected the likelihood of each outcome. Compared with children unexposed to SHS, exposure to SHS doubled the likelihood of invasive meningococcal disease (a summary OR for SHS exposure of 2.02). Summary ORs for invasive pneumococcal disease and Hib diseases were 1.21 and 1.22, respectively. However, these small increases in the risk of developing these IBDs were not statistically significant unlike the increase in the risk of developing meningococcal disease. That is, they might have occurred by chance. For bacterial carriage, summary ORs for SHS exposure were 1.68 for N. meningitidis, 1.66 for S. pneumonia (both these ORs were statistically significant), and 0.96 for Hib (a nonsignificant decrease in risk).
What Do These Findings Mean? These findings indicate that SHS exposure is significantly associated with invasive meningococcal disease among children. However, the evidence that SHS exposure is associated with invasive pneumococcal and Hib disease is only suggestive. These findings also indicate that exposure to SHS is associated with an increased carriage of N. meningitidis and S. pneumoniae. The accuracy and generalizability of these findings is limited by the small number of studies identified, by the lack of studies from developing countries where SHS exposure is increasing and the burden of IBD is high, and by large variations between the studies in how SHS exposure was measured and IBD diagnosed. Nevertheless, they suggest that, by reducing children's exposure to SHS (by, for example, persuading parents not to smoke at home), the illness and death caused by IBDs among children could be greatly reduced. Such a reduction would be particularly welcome in developing countries where vaccination against IBDs is low.