Abstract
Studies of the impact of long-term exposure to outdoor air pollution on the prevalence of respiratory symptoms and lung function in children have yielded mixed results, partly related to differences in study design, exposure assessment, confounder selection and data analysis.
We assembled respiratory health and exposure data for >45,000 children from comparable cross-sectional studies in 12 countries. 11 respiratory symptoms were selected, for which comparable questions were asked. Spirometry was performed in about half of the children. Exposure to air pollution was mainly characterised by annual average concentrations of particulate matter with a 50% cut-off aerodynamic diameter of 10 μm (PM10) measured at fixed sites within the study areas.
Positive associations were found between the average PM10 concentration and the prevalence of phlegm (OR per 10 μg·m−3 1.15, 95% CI 1.02–1.30), hay fever (OR 1.20, 95% CI 0.99–1.46), bronchitis (OR 1.08, 95% CI 0.98–1.19), morning cough (OR 1.15, 95% CI 1.02–1.29) and nocturnal cough (OR 1.13, 95% CI 0.98–1.29). There were no associations with diagnosed asthma or asthma symptoms. PM10 was not associated with lung function across all studies combined.
Our study adds to the evidence that long-term exposure to outdoor air pollution, characterised by the concentration of PM10, is associated with increased respiratory symptoms.
Short-term increases in outdoor air pollution have been associated with respiratory symptoms and temporary lung function decreases [1, 2]. The impact of long-term exposure to outdoor air pollution on prevalence of respiratory symptoms and lung function in children has been investigated in studies around the world [3–15]. Results concerning symptoms have been mixed, with more evidence for significant effects of outdoor air pollution on bronchitis or symptoms such as cough and phlegm than on asthma or asthmatic symptoms such as wheeze. Results from cross-sectional studies of lung function in children were also mixed [14]. Several prospective studies documented significant effects of outdoor air pollution on lung function development, e.g. the Californian Children's Health Study [14, 15]. Some of the inconsistencies in reported associations between air pollution and lung function may be due to differences in study design [14]. Studies differ in their study area (air pollution exposure contrasts between communities versus within a community), measured pollutants (particulate matter, sulphur dioxide, nitrogen dioxide and ozone), wording of symptom questions, study population (either population based or high risk), inclusion of potential confounders and statistical methods. Particulate matter air pollution has been represented with various indices, including total suspended particulate matter (TSP), particulate matter with a 50% cut-off aerodynamic diameter of 10 μm (PM10) or 2.5 μm (PM2.5). Some studies have reported higher air pollution effect estimates on lung function for females [7, 12], but the evidence is currently not convincing [14]. In the Children's Health study, air pollution effects on symptoms were stronger in males [8], whereas the effects on lung function were stronger in females [7]. Interpretation of subgroup analysis in single studies is difficult as the power to detect significant interactions is limited.
To overcome some of these shortcomings, the Pollution and the Young (PATY) project assembled health and exposure data for 58,561 children from comparable cross-sectional studies in 12 countries on children's respiratory health. Pooling original data allowed harmonisation of data analysis and definition of confounders, pursuit of research questions not addressed originally, and inclusion of unpublished studies. We made further use of the large dataset to assess effect modification with more precision than single studies. We already published on associations between outdoor NO2 and respiratory symptoms in a subset of the five PATY studies with NO2 exposure data [16]. In the current article, we examine the association between ambient fine particulate matter (PM10) in all PATY studies and prevalence of respiratory symptoms and lung function.
METHODS
Study design
Cross-sectional studies of children were sought that assessed respiratory symptoms and individual risk factors by questionnaire, included cough and wheeze as outcomes, and allowed calculation of annual mean particulate matter measures by study area. We included all cross-sectional studies that were available at the onset of the study. From the USA and the Netherlands, we included only the most recent and largest study, excluding the Six City and Six School studies [14]. Table 1 describes the studies contributing to this paper. More details on the individual studies can be found elsewhere [4, 10, 11, 13, 16, 18–20]. The studies from Russia, Central and Eastern Europe and Italy have not previously published results of air pollution analysis in the English language. The study areas were of substantially different magnitude. The Austrian and Czech study was performed in one city, and the North American study included a very large area in the USA and Canada. We therefore anticipated some heterogeneity in effects, which was taken into account in the analysis. All studies had obtained permission from the relevant medical ethical committees.
Health data
Original questionnaires were translated into English and critically examined for comparability of wording. 11 comparable symptom outcomes were identified: wheeze in the past 12 months, asthma ever, bronchitis in the past 12 months, phlegm, nocturnal dry cough in the past 12 months, morning cough in the past 12 months, sensitivity to inhaled allergens, hay fever ever, itchy rash ever, woken by wheeze in the past 12 months and allergy to pets. The exact wording of the symptom questions in all studies has been reported previously [21].
Several of the studies also performed spirometry (table 1). All studies used equipment that fulfilled either the 1987 criteria of the American Thoracic Society (ATS) or the 1993 European Respiratory Society (ERS) criteria. Spirometry was performed using the 1987 protocol of the ATS (North American, German and Austrian studies) or the 1993 ERS (CESAR study and the Netherlands). Further details have been published previously [17]. The main lung function measures of interest were forced vital capacity (FVC), forced expiratory volume in 1 s (FEV1), peak expiratory flow and the forced expiratory flow at 25–75% of the FVC (FEF25–75%).
Exposure assessment methods
The main exposure of interest was the annual mean concentration of PM10 in the corresponding study area. Data on the gaseous pollutants NO2and SO2 were also obtained. In all studies, air pollution concentrations were measured at fixed monitoring sites in the study area. To assess the comparability of monitoring sites and monitoring methods, a standard questionnaire was discussed with the investigators of the studies (online supplementary material pages 2 and 3 and tables S1 and S2). Briefly, PM10 was directly measured specifically for the study with Harvard impactors in Switzerland, the CESAR study, and for Russia and North America. In the Netherlands, Austria and Germany, particulate matter measurements were converted to gravimetric PM10 using co-located measurements with a standard particulate matter sampler in the study area. In the Italian study, multiple particulate matter monitoring methods were used and there was insufficient co-location with standard PM10 equipment to allow calculation of a consistent PM10 concentration. The evaluation further resulted in a few modifications of the original exposure estimates and exclusions of a few areas because of unrepresentative monitoring sites (online supplementary material pages 2 and 3).
Data analysis
A priori, we assessed that air pollution exposure contrasts between countries could not be exploited, as differences in language and (unmeasured) major risk factors were likely to be more dominant determinants for the health outcomes (especially with regards to symptoms). Hence, a two-stage analysis approach was used. In stage one, study-specific PM10 effects were estimated using logistic regression for symptoms and linear regression for lung function. This approach also has the advantage that systematic differences in PM10 sampler performance between studies do not affect the results. An area-level random intercept was included to account for clustering within the study areas. The CESAR study was conducted in five countries and, although common methods were used, the study was analysed per country. In stage two, effect estimates and standard errors were entered into a meta-analysis, obtaining a mean estimate, and a measure and Cochran Chi-squared test of heterogeneity using the STATA (StataCorp, College Station, TX, USA) metan command. Estimation of this mean and its confidence interval takes into account both between-study variation in effects and uncertainty of study-specific estimates [22]. In the first stage, we controlled for age, sex, maternal education, paternal education, household-crowding, current parental smoking, mother smoking during pregnancy, gas cooking, unvented gas/oil/kerosene heater, mould, nationality, birth order and “ever had a pet” [16]. Lung function analyses were additionally adjusted for age, height, weight, technician, instrument, season of testing and reported infection on the day of the test. The natural logarithm of lung function variables was used as the dependent variable to allow for non-normal distribution and nonlinear associations with anthropometric variables [17]. The natural logarithm of age, weight and height (sex-specific using an interaction term) were included as predictors following the North American study [13]. We further calculated predicted values for FVC, FEV1 and FEF25–75% using recently published prediction equations and used the % predicted lung function as the dependent variable in an additional analysis [23].
Meta-regressions assessed associations between study-specific estimates and study characteristics. These a priori-defined potential sources of heterogeneity between estimates were study period, design type (between-city, within-city or mixed design), location of monitoring station, proportion of younger children (6–8 yrs), questionnaire-date variability across study areas, high response rate (>80%) and response-rate variability across study areas.
Subgroup analyses were conducted in the first stage to assess individual subject characteristics as a source of heterogeneity. We stratified by sex, age and reporting of wheeze, and sensitisation (the latter for lung function only).
All analyses were performed using STATA version 8 (StatCorp).
RESULTS
Associations between air pollution and respiratory symptoms
Data were available from 11 studies on PM10 and symptoms (table 1). The highest PM10 concentrations were measured in the Central and Eastern European areas. The range in concentration within a study was largest in the North American and Swiss studies (ratio of maximum to minimum larger than two) (fig. 1).
Annual average concentrations of particulate matter with a 50% cut-off aerodynamic diameter of 10 μm (PM10) per study area, within each country.
Complete information on health and covariates was available for 45,788 children. About one-third of the children were from the North American study (table 2). Several-fold differences in symptom prevalence were found between countries, probably partly related to cultural differences and subtle differences in wording of the question.
After adjustment for confounders, PM10 was significantly associated with phlegm and morning cough (table 3 and fig. 2). Associations between PM10 and a doctor diagnosis of bronchitis, nocturnal cough and hay fever were borderline significant. Significant heterogeneity between study-specific estimates was found for most outcomes. The most consistent pattern was found for phlegm and hay fever, for which most study-specific effect estimates were either positive or slightly negative (fig. 2). For bronchitis and the two cough variables, both positive and negative effect estimates were found. A diagnosis of asthma and the symptoms wheeze and being woken by wheeze were not associated with PM10. Meta-regression could not identify significant factors at the study level explaining the observed heterogeneity in effect estimates. Table S3 presents the data for the symptom phlegm.
Forest plots of study-specific and mean odds ratios for effects of particulate matter with a 50% cut-off aerodynamic diameter of 10 μm (PM10) on a) wheeze, b) asthma, c) bronchitis, d) phlegm, e) nocturnal cough, f) morning cough, g) sensitivity to inhaled allergens, h) hay fever, i) itchy rash, j) being woken by wheeze and k) allergy to pets. Odds ratios per 10 μg·m−3 increase in PM10 are from single pollutant models, adjusted for individual risk factors. The vertical line indicates odds ratio of 1 and horizontal lines represent 95% confidence intervals of estimates. The diamond shape at the bottom of each graph indicates position, and confidence interval, of the mean of the estimates. Symptoms are for the past 12 months, except asthma, hay fever and itchy rash, which refer to lifetime (ever).
Two-pollutant models were employed for symptoms showing (borderline) significant associations with PM10 in the single-pollutant model (table 4). Adjustment for SO2 made little difference to the PM10 effect estimates with the exception of bronchitis, which showed a substantial decrease (single versus two pollutant PM10 estimates table 4). Adjustment for NO2 made little difference to the PM10 effect estimates for bronchitis, phlegm, morning cough and hay fever. For the symptoms of nocturnal cough, sensitivity to inhaled allergens, itchy rash and allergy to pets, after inclusion of NO2, the PM10 effect estimates were reduced. Correlations (r) between the annual average concentration of NO2 and PM10 in the studies with ≥10 study areas ranged from 0.48 to 0.84. Correlations between SO2 and PM10 ranged from 0.36 to 0.82.
PM10 effect estimates did not differ significantly between males and females, or between the younger and older children in the study population (table S4). Associations for hay fever and allergy to pets were stronger in the older children, but the difference with the younger children was not significant. Effect of PM10 on nocturnal cough and allergy to pets were stronger in boys than in girls but the differences did not reach statistical significance.
Associations between air pollution and lung function
Valid lung function data were available for 22,809 children (table S5). About 60% of the children with a valid lung function test were from the North American study. While in the North American, German, Austrian and Dutch studies, the percentage of children with a valid lung function test was well over 80%, it ranged from between 26 and 60% for the four CESAR study countries. This was related to too early termination of the test, resulting in too low FVC values. In all studies except the Austrian study, the mean percentage predicted lung function using the Stanojevic equations was close to 100.
On average, no association was found between the average PM10 concentration of the study area and lung function (fig. 3). In individual studies, both increases (Slovakia) and decreases (North America and Poland) in lung function were found with increasing PM10. Removing the CESAR countries from the analysis did not change the associations substantially (table S6). Analysis of the data using the recent Stanojevic prediction equations also resulted in no significant associations between PM10 and lung function (table S6).
Forest plots of study-specific and mean effects of particulate matter with a 50% cut-off aerodynamic diameter of 10 μm (PM10) for a) forced vital capacity (FVC), b) forced expiratory volume in 1 s (FEV1), c) peak expiratory flow (PEF) and d) forced expiratory flow at 25–75% of the FVC (FEF25–75%). Figures are given as percentage difference in lung function per 10-μg·m−3 increase in PM10 from single pollutant models, adjusted for individual risk factors. The vertical line indicates null, i.e. 0% change (no effect), and horizontal lines represent 95% confidence intervals of estimates. “Combined” indicates the mean and its confidence interval of the individual estimates.
There were no significant differences in PM10 effect estimates between males and females or younger and older children (table 5).
DISCUSSION
Statistically (borderline) significant positive associations were found between PM10 and the prevalence of phlegm, morning cough, hay fever, bronchitis and nocturnal cough. There were no associations with diagnosed asthma and asthma symptoms. PM10 effect estimates did not differ between males and females. PM10 was not associated with lung function.
The main strength of our study is the large number of children (45,000) taken from studies in 12 countries. This reduces the risk of finding spurious associations due to, for example, unmeasured confounding at the study area level as in single studies. The large study size also allowed for analyses of subgroups with more precision. Compared with a standard meta-analysis of studies, our study offers several advantages related to having the original data of all the studies available instead of only the air pollution effect estimates. First, we selected end-points for a common analysis that were considered sufficiently similar in wording. Secondly, the same data analysis procedures were used, including a common set of confounders. Thirdly, the comparability of exposure data collection could be assessed in detail, which resulted in the removal of some study areas from the epidemiological analysis. Finally, several unpublished studies were included in the analysis.
Comparison with previous studies
Our findings of significant associations between PM10 and respiratory symptoms and no associations with lung function are in agreement with the Harvard Six City study and two German studies, which were not included in the current analysis [3, 9, 24]. In contrast, other studies not included in the PATY study did find associations between long-term average air pollution exposure and lung function [14]. For the studies included in the PATY study the large North American study found associations with both lung function and bronchitis [13], whereas the Dutch, German and CESAR studies [5, 10, 18] did not find lung function associations.
The effect of air pollution may have been too modest to be reflected in detectable changes of lung function. The (asthma) symptoms related to the largest deficits in lung function were also not associated with outdoor air pollution. Alternatively, different biases in the analysis of symptoms and lung function may explain the inconsistencies. First, the typically small effects of air pollution on lung function may have been masked by factors such as variability in coaching and judgement by the technician, instrument, subtle shifts in instrument calibration, short-term weather factors and (past) infections [10]. In the CESAR study, good-quality tests were only obtained in <50% of the children, despite a rigorous quality assurance/quality control protocol. Exclusion of the CESAR study countries from the analysis, however, did not change our results. Secondly, reporting bias of symptoms may explain some of the positive associations in studies where parents of the children are aware of high air pollution exposures. This is an unlikely explanation, especially in the North American study where none of the included cities was highly polluted by local sources, and the Dutch study where all schools were located within 400 m from a major road. The pattern of associations, with bronchitis but not asthma, also makes reporting bias unlikely. Thirdly, lung function data were available for a smaller subset of children, but the precision of estimated effects on lung function was good because of pooling data from a large number of children. Fourthly, pollution levels were moderate. Air pollution may affect lung function at higher pollution levels, such as in many developing countries. Longitudinal studies in highly polluted Mexico City support this [14]. Fifthly, spirometry was performed with now updated guidelines. The updates in performance of testing are probably not sufficiently major to explain the lack of overall associations in our study. We further showed that the use of recent prediction equations for spirometry also resulted in no association between PM10 and lung function. Finally, we only included cross-sectional studies. Several prospective studies have found associations between air pollution and lung function growth, although a series of Austrian studies did not find consistent effects of particulate matter [14].
Heterogeneity of effects
Despite the efforts to select symptoms with similar wording and harmonise confounder data and data analysis, significant heterogeneity was present in both the symptom and lung function country-specific effect estimates. A priori-defined study-level factors could not explain this heterogeneity, possibly related to the relatively small number of studies. Differences in study design may have contributed to the observed heterogeneity. Six of the included studies assessed exposure contrasts between communities (Switzerland, Germany, North America, Poland, Hungary and Bulgaria), two studies assessed within-community contrasts (Austria and the Czech Republic), and in the remaining three studies exposure contrasts derived from a mixture of between and within community contrasts. The major common exposure variable PM10 is likely to represent different air pollution mixtures in the different studies, with long-range transported particles contributing only in studies with a between-community component and freshly emitted (ultrafine) particles contributing more to the contrast in the within-city studies. However, particles emitted by motorised traffic have contributed to the exposure contrast in all studies. Within the PATY study, there was no consistent pattern of stronger effects for those studies with a strong traffic-exposure component.
We did not find any difference in effect estimates between males and females, either for symptoms or for lung function. In previous single studies, stronger effects were reported for females in some studies [7, 12] and males in other studies [8].
Limitations
The small number of study areas is a limitation in some of our studies, although in Switzerland, Russia, the Netherlands and North America, the number of areas was fairly large (≥10).
The annual average concentration of PM10 was the main exposure variable within the PATY study. We also evaluated NO2 and SO2 as pollutants representing the urban air pollution mixture, but the number of studies with data on multiple pollutants was limited. Hence, the ability to assess effects of PM10 independent of the gaseous pollutants was limited. Two-pollutant models are difficult to interpret, especially when the same source affects both pollutants, as is the case for PM10 and NO2, for which motorised traffic is an important source. We cannot exclude that other pollutants such as the soot or elemental carbon content of particulate matter or the ultrafine particle concentration were associated more strongly with the health outcomes.
Several studies have assessed indicators of motorised-traffic emissions at a fine spatial scale (<100–300 m) [6, 10, 12, 24]. Our central monitoring data do not reflect these fine-scale air pollution variations. Because we did not have access to individual addresses, we were unable to generate individual exposure estimates at the residential address. Hence, our study does not provide information on the role of primary traffic emissions. Although this is a limitation, studies comparing the average concentration across communities remain valuable, as they allow estimation of the potential health effects of more aged pollution mixtures. Furthermore, a large fraction of the population does not live directly on major roads. Personal monitoring studies have shown that the population average personal exposure of subjects is strongly related to the outdoor air pollution level measured at a central site in the community [25]. For individual subjects, differences in exposure from the population average may occur. However, much of the measurement error is likely to be Berkson error, which generally does not lead to bias [26]. Finally, the Swiss SAPALDIA (Swiss study on Air Pollution and Lung Disease In Adults) and the Children's Health study study reported an effect of both within and between community contrasts in NO2 exposure on lung function of adults and children's respiratory symptoms [7, 27, 28]. Chronic bronchitis of adults in the European Community Respiratory Health Survey was associated with individual level variables representing traffic, but not centre level variables such as the central site PM2.5concentration [29]. In that study, the between-community contrast, however, was largely due to differences between countries, a contrast that we specifically decided not to assess because of the potential for too many differences for which we did not have data. Despite the value of assessing community-level pollution, there is a clear need for assessment of the health effects of near-traffic exposures. There is, for example, some evidence that asthma incidence in children and adults may be associated with near-traffic exposures and not with urban background pollution [30].
In conclusion, our study adds to the evidence that long-term exposure to outdoor air pollution, characterised by the concentration of PM10, is associated with increased respiratory symptoms (phlegm and morning cough) in children. We did not find an association between PM10 and lung function, possibly due to modest pollution levels and heterogeneity across studies.
Acknowledgments
The study was coordinated by T. Fletcher (London School of Hygiene and Tropical Medicine, London, UK).
Author affiliations are as follows. G. Hoek: Institute for Risk Assessment Sciences (IRAS), Utrecht University, Utrecht, the Netherlands; S. Pattenden: London School of Hygiene and Tropical Medicine, London, UK; S. Willers: Institute for Risk Assessment Sciences (IRAS), Utrecht University, Utrecht, the Netherlands; T. Antova: Environmental Health Unit, NCPHP, Sofia, Bulgaria; E. Fabianova: Regional Authority of Public Health, Banska Bystrica, Slovakia; C. Braun-Fahrländer: Swiss Tropical and Public Health Institute and University of Basel, Basel, Switzerland; F. Forastiere: Dept of Epidemiology, ASL Rome, Rome, Italy; U. Gehring: Institute for Risk Assessment Sciences (IRAS), Utrecht University, Utrecht, the Netherlands and Helmholtz Zentrum München, Institute of Epidemiology, Neuherberg, Germany; H. Luttmann-Gibson: Dept of Environmental Health, Harvard School of Public Health, Boston, MA, USA; L. Grize: Swiss Tropical and Public Health Institute and University of Basel, Basel, Switzerland; J. Heinrich: Helmholtz Zentrum München, Institute of Epidemiology, Neuherberg, Germany; D. Houthuijs: National Institute Public Health and the Environment (RIVM), Bilthoven, the Netherlands; N. Janssen: Institute for Risk Assessment Sciences (IRAS), Utrecht University, Utrecht and the National Institute Public Health and the Environment (RIVM), Bilthoven, the Netherlands; B. Katsnelson: Ural Regional Centre for Environmental Epidemiology, Ekaterinburg, Russia; A. Kosheleva: Ural Regional Centre for Environmental Epidemiology, Ekaterinburg, Russia; H. Moshammer: Institute of Environmental Health, Medical University of Vienna, Vienna, Austria; M. Neuberger: Institute of Environmental Health, Medical University of Vienna, Vienna, Austria; L. Privalova: Ural Regional Centre for Environmental Epidemiology, Ekaterinburg, Russia; P. Rudnai: National Institute of Environmental Health, "Fodor Jozsef" National Center for Public Health, Budapest, Hungary; F. Speizer: Dept of Environmental Health, Harvard School of Public Health, Boston, MA, USA; H. Slachtova: Institute of Public Health, Center of Health Services, Ostrava, Czech Republic; H. Tomaskova: Institute of Public Health, Center of Health Services, Ostrava, Czech Republic; R. Zlotkowska: Epidemiology Dept, Institute of Occupational Medicine and Environmental Health, Sosnowiec, Poland; and T. Fletcher: London School of Hygiene and Tropical Medicine, London, UK.
Footnotes
This article has supplementary material available from www.erj.ersjournals.com
Support Statement
The PATY study was funded by the EU Fifth Framework Quality of Life Programme (contract number QLRT-2001-02544).
Statement of Interest
None declared.
- Received January 7, 2011.
- Accepted December 9, 2011.
- ©ERS 2012
REFERENCES
