Traffic-Related Air Pollution and Asthma Onset in Children: A Prospective Cohort Study with Individual Exposure Measurement

Table 1 lists the descriptive statistics and bivariate associations with asthma for the sample and all the variables used in the analysis. Hispanic Americans and those of white race were well represented in this sample, but there were few Asian Americans, African Americans, or others. Males represented just about 40% (86) of this sample. Many of the children had potential risk factors for asthma in the home, with 95% (183) having carpet in the child’s bedroom, 83% (181) having a pet, and 79% (168) having a gas stove. About 17% (35) of the children had a parental history of asthma. The median age at baseline was 9.6 years. Except for a higher proportion of females, there were no significant differences between the frequency of these variables among the subjects of this study and the remainder of eligible subjects from which we drew the study sample. The study sample and other eligible subjects also had similar traffic exposures (results not shown).

The bivariate associations of individual and household variables as predictors of asthma were largely not statistically significant (Table 1). Associations with body mass index (BMI) and other covariates were generally consistent with associations observed in previous analyses of the entire cohort from which we drew this study sample (Gilliland et al. 2003; McConnell et al. 2002a). In this sample, Hispanics were more likely to report incident asthma during the follow-up than were whites. Because of small counts, we could not derive stable estimates for African Americans. Less than a third of the homes had mildew or a smoker, which may have led to unstable estimates.

Table 2 shows the contextual confounders tested in the modeling process. In general, lower community SEP was associated with higher risk for asthma onset. Higher winter temperature was associated with higher risk for asthma. The effects of temperature were less consistent than those for humidity, which was a risk factor in all seasons. Higher relative humidity was generally associated with higher risk for incident asthma.

In Figure 1, the bars illustrate individual residential NO₂ levels within each study community. Variation exists within communities, and a broader regional pattern is also present. Communities in the Los Angeles metropolitan area have generally higher levels than do communities to the north and south, showing a spatial pattern consistent with earlier measurements based on central monitors (Gauderman et al. 2004). We examined the incidence rate by tertile of exposure. Dividing the sample into tertiles of annual NO₂, the crude asthma incidence rates, from lowest to highest tertile of exposure, were 14.2, 19.1, and 20.7 per 1,000 person-years, respectively [with wide confidence intervals (CIs)].

Table 3 presents the descriptive statistics for the exposure variables within each community. There was a wide range of average exposure across communities but also substantial variation within communities (mean interquartile range of 6.2 ppb). Communities in the Los Angeles metropolitan area (Lake Elsinore, Long Beach, Mira Loma, Riverside, San Dimas, and Upland) generally had higher NO₂.

We tested each of the individual variables shown in Tables 1 and 2 with an association to asthma to determine whether confounding reduced the pollution coefficient by ≥ 5%. Table 4 contains the results of the final models, in which we retained Hispanic ethnicity, medical insurance, enrollment group, and relative humidity as confounding adjustment variables. All pollution metrics were positively associated with incident asthma. For average residential NO₂ the hazard ratio (HR) was 1.29 (95% CI, 1.07–1.56) across the average within-community interquartile range of 6.2 ppb. The fall–winter HR of 1.29 (95% CI, 1.11–1.49) was nearly the same as the annual average, whereas the summer HR of 1.27 (95% CI, 1.03–1.59) was slightly smaller. The sample size was 196 with 26 cases (for the annual NO₂), so it is possible that influential outliers might have affected the results. However, there were no cases with exceptionally high exposure to NO₂ and no noncases with notably low exposure to NO₂.

The σ² values are the random-effects variance estimates, which represent the residual variation in asthma incidence attributable to the community cluster after we included all other fixed predictors in the model (Table 4). Addition of the individual covariates in model 1 reduced the σ² only slightly from a null model containing only baseline age and sex strata (results not shown). Models 2 and 3 show the large decline in σ² when we included community relative humidity and pollution, respectively, indicating that both variables account for a large proportion of the residual between-community variation in asthma incidence.

We assessed exposures both as the deviation of the individual specific exposure metric from the corresponding community mean and as the community average (i.e., average of household NO₂ per community). We used these variables to compare within- and between-community pollution effects. Table 5 lists results of the tests shown in Equation 2 for difference in the between-community exposures and deviation within community exposures. The between-community summer NO₂ point estimate was larger than the within-community estimate, but these estimates had large CI values and were not significantly different based on the likelihood ratio test. For yearly average and winter NO₂, the within- and between-community effects were similar.

Interactions between humidity and air pollution were tested, but found not to be significant, although pollution did appear to exert larger effects in more humid areas (t-values were between 0.9 and 1.3 depending on the humidity interaction term with various measures of pollution).

We conducted further sensitivity analyses to assess whether excluding early wheeze at baseline affected the results (Table 6). Excluding baseline wheeze had a small impact on the asthma effect of air pollution. We implemented another sensitivity test by removing data for all children with a history of early chest illness, including croup, bronchitis, bronchiolitis, or pneumonia (Table 6). The HR values were mildly attenuated but still elevated and similar to the HR values without these exclusions.

In this prospective cohort of children from 11 Southern California communities, we found significant associations between incident asthma and measured outdoor residential exposure to NO₂. A strength of the study was the individual assignment of exposure, based on measurements of NO₂ outdoors at the homes of the children. Previous research on incident asthma and air pollution has largely relied on modeled exposure (Brauer et al. 2002; Zmirou et al. 2004) or on exposure measured at central site locations (McConnell et al. 2002a; McDonnell et al. 1999; Shima et al. 2002). To our knowledge, this is the first study to employ individually measured pollutant exposures with incident disease assessment. Although the results must be interpreted in light of the limited sample size and small number of cases, they highlight a need for further examination of relationships between incident disease and individually measured exposures.

We tested numerous individual and contextual confounding variables. Some of the variables that had an association with asthma were not confounders because they exerted little or no effect on the pollution coefficient. Parsimonious models controlling for age and sex in the baseline hazard indicated that Hispanic ethnicity, medical insurance, and relative humidity were confounders in models with NO₂. The air pollution exposures generally accounted for larger portions of the residual variance in the community random effects (i.e., σ²) than did the individual-level variables included as confounders. The results for residential NO₂ were sensitive to inclusion of the relative humidity variable. This variable met our criteria for confounding and increased the pollution effect (Table 4). Because adjustment for humidity elevated the risks associated with air pollution exposure, we further investigated an interaction between pollution exposures and humidity. Although the interaction was not significant, we did observe generally higher risks for pollution in more humid communities.

In the absence of interaction, the influence of humidity may represent a case of variance suppression (Tabachnick and Fidell 1985), whereby confounding variables may increase the effect size of the primary risk variable in the model. In this case, relative humidity and the pollutants share some variance, and this shared component also partially overlaps with the asthma outcome. The part of the variance shared by relative humidity and pollution is the component of variance with asthma that has a relatively less precise association with pollution. With the inclusion of relative humidity, this less precise component of the variance overlap between pollution and asthma is removed, and the remaining relationship becomes more pronounced.

We further investigated the possibility of positive confounding by examining the correlation between humidity and NO₂ and the impact of the humidity risk with NO₂ excluded from the model. There was a moderately high, significant negative correlation between NO₂ and relative humidity. Evaluated at the household level for the annual estimates, the Pearson correlation coefficient was −0.57, and at the community average level of NO₂ it was −0.61. Except for Long Beach, most of the coastal communities had lower air pollution and higher humidity. Because humidity was positively associated with asthma but negatively associated with NO₂, we would expect humidity to be a negative confounder (i.e., excluding humidity from the model should result in a negative bias on the NO₂ estimate). In this case, we did observe a negative bias when humidity was not included in the model. We would also expect a negative bias in the humidity effect when we exclude NO₂ from the model. Table 7 shows the humidity risks with and without NO₂ in the model. Without NO₂, the risks were attenuated in all models, and the negative bias was strongest in the summer season. Thus, the impact of humidity was negatively biased when we excluded NO₂ from the model, which is consistent with the positive confounding effect observed in the NO₂ models. These results imply that simultaneous consideration of meteorology and air pollution may be important in models assessing the effects of chronic air pollution exposure on respiratory disease.

Interpreting the household NO₂ findings is complicated because it represents a mixture of local and regionally transported pollutants in Southern California. In an earlier analysis, Gauderman et al. (2005) reported moderate to high correlations between the measured residential NO₂ and various measures of traffic proximity or modeled concentrations. Although some toxicologic evidence supports the potential for NO₂ as a direct causative agent for respiratory health effects at high concentrations (Kelly and Sandstrom 2004), the more likely interpretation is that NO₂ was a proxy for other components of the local air pollution mixture with high toxic potential. Because NO₂ pollutants may be transported from neighboring areas, some of the effects observed could represent local and regional contributions. If NO₂ represents other toxic traffic-related pollutants, the transported component would have different toxicologic properties than purely local contributions.

Rather than the specific causal agent, NO₂ may represent a mixture of local traffic-related pollution and regional pollutants, which in Southern California include transported primary traffic-related pollutants and regional products of photochemistry. The design of the study allowed us to examine the independent effects of each of these contributions to the pollutant mixture by modeling the effects of the within- and between-community variation in NO₂, as shown in Table 5. The within-community effects indicative of long-term local traffic sources were similar in magnitude to effects of community average NO₂ across communities, suggesting that both regional and local pollution contributed to the associations with asthma. However, the range of variation of NO₂ within communities was smaller than between communities, and the HR values were therefore smaller when constrained to the range within communities. For example, using the average interquartile range across all measurements of 28.9 ppb for annual NO₂ increased the HR to 3.25. For the average within-community range of 16.4 ppb, the HR was 1.95.

Compared with earlier studies on prevalent asthma, with a similar population and the same exposure measurements for NO₂, the odds ratio per interquartile range increment of 5.7 ppb was 1.83 for prevalent asthma (Gauderman et al. 2005). Thus, the incident effects were smaller than those observed for prevalent asthma over a similar exposure contrast. The overall effects we observed here nonetheless support results from our earlier prevalence study. In both studies, we observed effects on asthma from NO₂.

Notable strengths of this study are the prospective design, individually measured pollutant exposures, incident disease assessment, and the large effects. Limitations include an inability to identify the specific constituents of pollution that were responsible for the observed health effects, the relatively small sample size, and questionnaire case ascertainment. Although a potential limitation, self-reported physician diagnosis is widely used to define asthma in epidemiologic studies (Burr 1992), is reproducible (Ehrlich et al. 1995; Peat et al. 1992), and reflects what physicians actually reported to patients (Burney et al. 1989; Greer et al. 1993). We conducted sensitivity analyses with alternative case definitions, excluding children with baseline wheeze and early childhood chest illness. The summer estimates were attenuated mildly (HR values range from 1.21 to 1.26), and after excluding baseline wheeze became insignificant (Table 6). Many of the communities had lower NO₂ levels in the summer (Table 3), and the effects during the summer season were smaller than in the other seasons. In contrast, the HR values for NO₂ were only slightly smaller for the fall–winter and annual estimates and remained significant (HR values range from 1.2 to 1.9). The general conclusions, therefore, are similar even after excluding children with wheeze at baseline or with a history of early childhood illness.

The overall asthma incidence rate per thousand person-years of 18.8 cases resulting from parent- or child-reported physician diagnosis was high but is similar to rates reported previously in the CHS cohort (McConnell et al. 2002b) and in other studies of asthma or wheeze in school children (Lombardi et al. 1997; Ronmark et al. 2001). Lower rates have been found in other studies (Broder et al. 1974; Dodge and Burrows 1980; Yunginger et al. 1992). Comparison across studies, however, is complicated by varying ascertainment methods and temporal trends in incidence rates. The rate in this sample from the larger CHS cohort was similar to those in comparable studies.

Further field studies measuring plausibly causal ultrafine particles, metals, and poly-cyclic aromatic hydrocarbons simultaneously (Nel 2005) might identify the constituents responsible for the observed association with NO₂. Although the specific pollutant or related mixture of constituents responsible for the effects remains the focus of future research, the large risks observed suggest that traffic-related air pollution exposure contributes to new-onset asthma.

This work was funded by the National Institute of Environmental Health Sciences (grants 5P30ES007048, 5P01ES011627, 5P01ES009581, and 1R03ES014046), the U.S. Environmental Protection Agency (grants RD831861, R826708, and R831845), California Air Resources Board contract 94-331, and the Hastings Foundation.