There is a large worldwide variation in the prevalence of asthma in children and there is also evidence that the prevalence has been increasing.¹ Both genetic and environmental factors play important roles in the aetiology of childhood asthma, and there is probably also genetic susceptibility to the effects of air pollution.² Short term changes in the occurrence of asthma are more likely to be influenced by changes in the environment, diet, and lifestyle than by changes in the genetic pool. From a preventive perspective, information on environmental, dietary, and behavioural factors is crucial.³ Identification of indicators of genetic susceptibility to environmental exposures could be useful from the preventive point of view. There is evidence that long term exposure to ambient air pollution increases the risk of childhood asthma, but the role of different sources and components needs further elaboration.^{4,5,6,7,8,9,10,11,12}

In a nationwide cross sectional study in Taiwan, Guo et al found an association between traffic related air pollution concentrations and the risk of asthma in school children.¹³ This study did not adjust for parental atopy or indoor exposures which are potential sources of confounding and effect modification. In 2001 we conducted a new nationwide cross sectional study in which we also collected information on these important potential determinants of allergic disease in children.

In the present study we have elaborated the relation between exposure to urban air pollution and the risk of asthma in school children, focusing on predominantly traffic related pollutants such as nitrogen oxides (NOx), ozone (O₃), and carbon monoxide (CO). We also assessed the role of air pollutants, mainly from other fossil fuel combustion sources such as sulphur dioxide (SO₂), and particles with an aerodynamic diameter of 10 μm or less (PM₁₀). Furthermore, we studied the joint effects of parental atopy and outdoor air pollution on the risk of asthma. Parental asthma, allergic rhinitis, and allergic atopic eczema were used as surrogates of the genes that are responsible for susceptibility to the effects of air pollutants on asthma. We applied a two stage hierarchical model to adjust for confounding and to elaborate effect modification at the individual level and to assess the effects of air pollution at the municipal level.^14,15

METHODS

Data collection and study population

A nationwide cross sectional study was conducted in Taiwan in 2001 where a modified Chinese version of the International Study of Asthma and Allergies in Childhood (ISAAC-C) questionnaire was used to collect information on children’s health, environmental exposures, and other relevant factors.¹⁶ The study population was recruited from elementary and middle schools in 22 municipalities within 1 km from Taiwan Environmental Protection Agency (EPA) air monitoring stations. The questionnaire was taken home by students and answered by parents. A total of 35 036 children aged 6–15 years were approached. The response rate was 93.2%. 2364 children were excluded because of an incomplete questionnaire, leaving a final study population of 32 672 schoolchildren.

The study protocol was approved by the Respiratory Health Screening Steering Committee of the Taiwan Department of Health and the Institutional Review Board of National Cheng Kung University Hospital, and it complied with the principles outlined in the Helsinki Declaration.¹⁷

Health outcome

The outcome of interest was childhood asthma, which was defined on the basis of the answer to the question: “Has a physician ever diagnosed your child as having asthma?”.

Exposure assessment

Complete monitoring data for the air pollutants including sulphur dioxide (SO₂), nitrogen oxides (NOx), ozone (O₃), carbon monoxide (CO), and particles with an aerodynamic diameter of 10 μm or less (PM₁₀), as well as daily temperature and relative humidity, are available from 1994 for 22 EPA monitoring stations on Taiwan’s main island (fig 1). Concentrations of each pollutant are measured continuously and reported hourly—CO by non-dispersive infrared absorption, NOx by chemiluminescence, O₃ by ultraviolet absorption, SO₂ by ultraviolet fluorescence, and PM₁₀ by beta-gauge. Exposure parameters in the present study were annual average concentrations, calculated from the monthly averages of the year 2000. Forty four schools in Taiwan’s 22 municipalities were investigated. Stratified sampling by grade was applied in each school.¹⁸

Covariates

Information on potential confounders was obtained from the parent administered questionnaire. The covariates in the present analyses included age, sex, parental atopy, parental education, maternal smoking history during pregnancy, environmental tobacco smoke (ETS), and visible mould (table 1). Parental atopy was a measure of genetic predisposition to asthma and it was defined as the father or mother of the index child ever having been diagnosed as having asthma, allergic rhinitis, or atopic eczema.

Statistical methods

The odds ratio (OR) was used as a measure of the relation between exposure to air pollution and the risk of childhood asthma. Adjusted ORs were estimated in a two stage hierarchical model using logistic and linear regression analyses. The models assume two sources of variation—the variation among subjects in the first stage, part of which could be explained by the individual characteristics, and the variation among municipalities in the second stage, part of which could be explained by variables measured at the municipal level. In the analyses we assumed that (1) the outcome variable follows Bernoulli distribution; (2) intercept terms are random at the municipal level; and (3) all the explanatory variables are fixed effects. A logistic regression model was fitted in the first stage for the risk of childhood asthma as a function of site-specific intercepts j, where αj  =  1, ..., 22, and personal covariates. The adjusted site-specific intercepts and prevalence rates are related by Pj  =  e^αj/(1+e^αj). In the second stage these intercept terms representing the logit of the site-specific prevalence rates (Pj; j  =  1, ..., 22), adjusted for personal covariates, were regressed on each site specific ambient pollutant level by using a linear “ecologic” regression—that is, logit αj  = α +Uj + βZj where Uj denotes the random departure from the general prevalence αj on the logit scale for site j and Zj denotes the ambient pollution level for site j. Thus, β can be interpreted as the log OR (per unit change) for each pollutant, adjusted for personal characteristics. The results from the models are presented as ORs, along with their 95% confidence intervals (CIs).

The goodness of fit was assessed with likelihood ratio tests (LR) to determine whether a variable contributed significantly to the model. Firstly, we fitted a full model with a complete set of covariates. To study further the sources of confounding we fitted models with different combinations of covariates and compared the effect from models with and without the covariate of interest. If the adjusted OR differed from the crude OR by more than 10%, that covariate was be included in the final model. We first fitted one pollutant models and then considered two-pollutant models by fitting one traffic related and one stationary fossil fuel combustion related pollutant. Finally, we fitted two-pollutant models with O₃ and another pollutant. The two-pollutant models provide estimates of the independent effects of CO, NOx, SO₂, PM₁₀, and O₃ on childhood asthma, controlling for the second pollutant in the model. We also considered three-pollutant models with one traffic related, one stationary fossil fuel combustion related pollutant, and O₃. The effect of each pollutant on the risk of childhood asthma was presented as ORs per 10 ppb changes for SO₂, NOx, and O₃, 100 ppb changes for CO, and 10 μg/m³ changes for PM₁₀, along with their 95% CIs. We assessed potential effect modification by parental atopy by comparing crude and adjusted effect estimates for children with and without atopic parents. The two-stage hierarchical model was used not only to derive more precise estimates of site specific parameters and site level effects, but also to adjust for multiple comparisons.¹⁹

RESULTS

Study population and occurrence of childhood asthma

The characteristics of the study population and the prevalence of childhood asthma according to the covariates are shown in table 1. The prevalence of asthma was 6.86% (95% CI 6.59 to 7.13). The prevalence of childhood asthma was related to young age, high level of parental education, male sex, parental atopy, maternal smoking during pregnancy and the presence of cockroaches and visible mould in the home. The prevalence of asthma was lower in children exposed to environmental tobacco smoke (ETS) than in those not exposed.

Air pollution

The distributions of the annual mean air pollutant concentrations, temperature, and relative humidity in the 22 monitoring stations in the year 2000 are presented in table 2 and the correlations between different pollutants are shown in table 3. The correlation between NOx and CO concentrations was high (0.88), which reflects the common source of motor vehicles. The concentrations of PM₁₀ and SO₂ were also highly correlated (0.58), indicating a common source of stationary fuel combustion, although SO₂ concentrations were also correlated with both traffic related pollutants. The concentration of O₃ was negatively correlated with the mainly traffic related pollutants but positively correlated with PM₁₀ and SO₂, and it was only weakly correlated with that of traffic related and stationary fossil fuel combustion related air pollutants.

Air pollution and childhood asthma

In the one-pollutant model, the risk of asthma was not related to NOx levels (adjusted OR 1.005 per 10 ppb change (95% CI 0.945 to 1.060)). Addition of either SO₂ (adjusted OR 1.048 (95% CI 0.983 to 1.117)) or PM₁₀ (adjusted OR 1.065 (95% CI 1.009 to 1.123)) increased the effect estimate for NOx substantially, and addition of O₃ (adjusted OR 1.029 (95% CI 0.973 to 1.089)) slightly (table 4). In the three-pollutant model the estimates for NOx were 1.113 (95% CI 1.038 to 1.194) and 1.152 (95% CI 1.082 to 1.227), respectively, when (SO₂ and O₃) or (PM₁₀ and O₃) were added and showed statistical significances (table 5). The adjusted OR for 100 ppb change in CO was 1.045 (95% CI 1.017 to 1.074) and the estimates changed little when a second or third pollutant was added. The adjusted OR for a 10 ppb change in SO₂ alone was 0.874 (95% CI 0.729 to 1.054), but inclusion of either of the traffic related pollutants reduced the effect estimate substantially whereas addition of O₃ had little influence. The risk of childhood asthma was not related to PM₁₀ concentrations in any combination of air pollutants. The risk of asthma was significantly related to O₃ in the one-pollutant model (OR 1.138 (95% CI 1.001 to 1.293)) and the addition of other pollutants increased the risk estimates further.

In summary, positive statistically significant associations were found for O₃ and traffic related pollutants (CO and NOx). In contrast, negative or weak associations were found for SO₂ and PM₁₀.

In order to elaborate the potential effect modification, we systematically conducted stratified analyses in categories of parental atopy. The stratified analyses did not indicate any major effect modification (table 6).

DISCUSSION

The risk of childhood asthma increased according to increased annual levels of the two traffic related pollutants NOx and CO, as well as O₃. The risk of asthma was not related to the levels of PM₁₀ and SO₂ and, in multi-pollutant models, the risk of asthma was negatively associated with the levels of SO₂. The results did not provide evidence that hereditary atopy representing a genetic predisposition modifies the effects of exposure to outdoor air pollutants.

Validity of results

Routine air pollution monitoring data were used as the basis for exposure assessment. These data represented reasonably well exposures both in the school and in the home for two reasons. The schools were chosen to be in the vicinity of the monitoring stations. Almost all the children attended schools within 1 km of their homes because the density of elementary and middle schools in Taiwan is very high. Finally, the two-stage hierarchical modelling took into account the fact that municipal level exposure information was used. Although we did not collect information about the vicinity of the busy road, the present study and previous studies from German²⁰ and Mexican cities²¹ provide consistent evidence that the outdoor NO₂ level is a better predictor of traffic exposure than exposure to NO₂ at the personal level.

From previous literature we know that a high proportion of traffic related outdoor air pollutants (NOx, CO) penetrate indoors, but the penetration proportion for particles is lower when mechanical filtration or an air conditioner is used.^22–24 Most of the Taiwanese school children spend at least 8 hours/day in school. Air conditioning is rare in Taiwanese classrooms. Mechanical filtration is practically the only type of filtration in Taiwanese homes during the summer, even if the home is air conditioned. Any known or unknown factors such as air change, penetration, deposition, as well as emission strengths for indoor pollutants could be responsible for the observed association between personal exposure and municipal level exposure. This was a common problem in all the previous studies assessing the effects of air pollution on the risk of asthma in school children.

In cross sectional studies, selection bias is a potential threat to validity. A plausible mechanism of selection is that parents of children with asthma move to residential areas with lower levels of air pollution which will lead to underestimation of the relation between exposure and outcome. Any random migration is likely to result in underestimation of the air pollution effects but would not introduce a positive bias in the associations. Information on residential history in a cross sectional study could be used to reduce the possibility of selection bias, whereas a longitudinal study would provide a stronger design for minimising this problem. We did not have sufficient information on children’s residential history to formally assess the lifetime exposure to air pollution, so current exposure was used as the proxy of previous exposure. Most of the studies to date on the relationship between exposure to air pollutants and risk of asthma in children have been cross sectional.^7,9–12

In our statistical analyses we controlled for a number of potential individual level confounders such as parental education and indoor environmental exposures. In stratified analyses on different levels of covariates, we studied the possibility of residual confounding. Parental education had a positive association with concentrations of traffic related pollutants. The prevalence of childhood asthma was also positively associated with the level of parental education, which could be explained by both the influence of air pollution and better access to health care.

Assessment of the independent effects of different pollutants is difficult because urban air pollution constitutes a complex mixture of several compounds. Although all the measured pollutants have several sources, NOx and CO are predominantly from vehicle emissions while SO₂ and PM₁₀ are mainly from stationary fossil combustion processes.^25,26 In the present study, NOx and CO concentrations were highly correlated, indicating the common source of motor vehicle traffic emissions. SO₂ and PM₁₀ concentrations were also correlated, their common sources being stationary fossil fuel combustion. In the modelling, we were able to control for one stationary fossil fuel pollutant at a time as a potential confounder when assessing the effect of one of the traffic related pollutants and vice versa. Due to collinearity problems, it was not possible to separate the impact of traffic related pollutants (NOx and CO) from each other.

Synthesis with previous knowledge

The results of the present study are in line with several previous studies from Seattle,⁴ Illinois,⁵ Los Angeles,⁶ Austria,⁷ the Netherlands,⁸ Czech Republic and Poland,⁹ Germany,¹⁰ Japan,¹¹ and China¹² indicating that exposure to outdoor air pollutants increases the risk of asthma. All available studies suggest an increased risk for traffic related air pollutants (CO and NOx)^{4,6–8,11,12} and O₃,^5,6 but the findings are inconsistent for stationary fossil fuel combustion related air pollutants (SO₂ and PM₁₀).^9,10,12

A panel study conducted in Seattle showed no association between exposure to CO and the occurrence of asthma related symptoms.⁴ In a cross sectional study carried out in Illinois, increased morning and evening asthma symptom scores were associated with the levels of O₃.⁵ A population based cross sectional study in Los Angeles found positive associations between asthma symptoms and all the US criteria air pollutants (O₃, NO₂, SO₂, PM₁₀). The risk of more severe asthma symptoms was associated with 8 hour exposure to NO₂ levels above 1.4 ppb (OR 1.27, 95% CI 1.05 to 1.54).⁶ Four recent studies from Austria, the Netherlands, Japan, and China examined the relationship between exposure to NO₂ and the risk of childhood asthma.^7,8,11,12 In the Austrian study, the risk of childhood asthma was associated with NO₂ levels (highest v lowest category, prevalence odds ratio (POR) 5.81, 95% CI 1.27 to 26.5).⁷ In a cohort study conducted in the Netherlands, the risk of physician diagnosed asthma was associated with traffic related air pollution measured as NO₂ concentrations.⁸ A survey in Japan found that a 10 ppb increase in the outdoor NO₂ concentration was associated with an increased risk of asthma with an adjusted OR of 2.10 (95% CI 1.10 to 4.75).¹¹ A study conducted in four Chinese cities suggested a positive association between the risk of childhood asthma and the levels of PM₁₀ (OR 1.33, 95% CI 0.80 to 2.19), but weak or negative associations were found for SO₂ (OR 1.03, 95% CI 0.75 to 1.41) and NOx (OR 0.95, 95% CI 0.63 to 1.43).¹² In the Czech and Polish study the lifetime prevalence of physician diagnosed asthma in schoolchildren was associated with SO₂ (OR 1.39 per 50 μg/m³ increase in SO₂, 95% CI 1.10 to 1.92).⁹ In contrast, a study in Germany indicated that high levels of SO₂ in East Germany were associated with a lower prevalence of asthma compared with West Germany.¹⁰

Nitrogen dioxide has been shown to be an acute respiratory irritant in controlled exposure studies.²⁷ There are no plausible mechanisms through which CO exposure would influence the airways and increase the risk of childhood asthma. Both NOx and CO represent the complex mixture of traffic exhaust, and NO₂ is known to be the best indicator of motor vehicle traffic emissions. In the present study it was not possible to determine to what extent NOx would have direct effects on children’s airways. CO is unlikely to have any direct effects on the airways.

A positive association between the risk of childhood asthma and exposure to O₃ was identified, compatible with toxicological studies.^28,29 The prevalence of childhood asthma was associated with O₃ levels. The effect was increased when adjusting for either a traffic related pollutant, stationary fossil fuel pollutant, or both. Ozone is a secondary pollutant in the atmosphere produced from traffic exhausts, but scavenged by direct motor vehicle emissions. It is a known respiratory irritant³⁰ and could be causally related to childhood asthma.

Our finding of a lack of association between the risk of childhood asthma and PM₁₀ levels is consistent with the results from the Harvard 24 Cities Study in North America.³¹ Although the risk of childhood asthma was not related to the levels of PM₁₀, it is likely that there is an association with fine particulate matter (PM_2.5) and ultrafine particles typically present in motor vehicle exhausts and, in particular, in diesel exhausts. Further studies should assess these relations. No association between childhood asthma and SO₂ was found in the present study, possibly because the average level of combustion related air pollutants such as SO₂ in Taiwan is substantially lower than levels reported in studies in the Czech Republic and Poland⁹ and in East Germany.¹⁰

Conclusion

The present study provides additional evidence that exposure to outdoor air pollutants increases the risk of childhood asthma in school children. The observed relations between the risk of childhood asthma and NOx and CO levels suggests that emissions from motor vehicles play an important role. In addition, the relationship with O₃ levels indicates that photochemical air pollution contributes to adverse health effects.