Synergistic Effects of Traffic-Related Air Pollution and Exposure to Violence on Urban Asthma Etiology

1The gradient of socioeconomic position (SEP) on health may be explained partly by a combination of increased contaminant exposures and greater susceptibility to their effects. Air pollution, for instance, may be higher near major roadways, power plants, and industrial sites, where property values are lower and lower-income populations reside. Increased life stress among lower-SEP populations has also been proposed as a primary pathway through which SEP affects health.

2Because of this potential spatial covariance across exposures, and because stress and pollution may influence common physiologic pathways (i.e., oxidative stress) and health outcomes (i.e., respiratory disease), stronger methods are needed to disentangle their effects and investigate synergies. The environmental justice literature has documented significant disproportionate contaminant exposures in minority and lower-SEP communities, and the resultant influence on asthma exacerbation patterns. However, fewer studies have considered disproportionate susceptibility among lower-SEP populations.

3Exposure to violence (ETV) has been conceptualized as a chronic urban stressor, potentially elevated in communities where pollution is higher. Chronic stress effects of episodic violence are grounded in trauma theory (e.g., post-traumatic stress), detailed elsewhere. Episodic violence, post-traumatic stress, and hyper-vigilance—more prevalent in lower-SEP urban communities—may negatively influence health though physiologic alterations, including immune dysregulation, and behavioral pathways. Many urban caregivers, for example, restrict children’s behavior, keeping them indoors due to fear of violence, making children more sedentary, increasing indoor exposures, and decreasing spatial autonomy that is important to development.

4Chronic stress has been linked to asthma exacerbations in cross-sectional and prospective population studies. Other evidence suggests a role for stress in the onset of asthma. Chronic stress may influence hypothalamic–pituitary–adrenal (HPA) axis and cortisol dysregulation, glucocorticoid resistance, sympathetic–adrenal–medullary (SAM) activation, catecholamine production, immune mediator function, inflammation, and cytokine production. Stress and pollution affect some common physiologic systems, facilitating synergistic effects; for example, psychological stress and ozone both affect oxidative stress pathways.

5Few studies have examined the influence of stress on pollution susceptibility, though some findings suggest differential susceptibility by SEP, possibly mediated by life stress. Time-series studies indicate effect modification of short-term pollution exposures by SEP, though others found no significant modification. Fewer studies have considered long-term exposures, though some indicate greater associations between long-term air pollution and mortality among less-educated adults.

6In urban settings, traffic-related air pollution may be elevated along with ETV, and previous studies have linked traffic-related air pollution to asthma exacerbation and respiratory outcomes. In the United States and Europe, children living or attending school near truck routes and highways show increased asthma and allergy symptoms, hospitalizations, allergic rhinitis, and reduced lung function. Traffic-related pollutants have also been associated with asthma development.

7Incorporating traffic-related air pollution into large-scale epidemiologic studies requires models linking traffic and ambient concentrations. Relationships between traffic and health have been examined using several different traffic indicators, with no consensus on which indictors best capture variability in traffic-related pollution or health outcomes in different settings. Prior studies have successfully extrapolated traffic exposures from sampling homes to larger cohorts using predictive land-use regression (LUR) models. LUR shows strong predictive power for intraurban nitrogen dioxide variability, using traffic and land use characteristics (i.e., population density, major sources).

8In this study, we explore the hypothesis that a chronic stressor (here, lifetime ETV) predicts stronger associations between traffic-related air pollution exposure and childhood asthma development. We employ data from the Maternal-Infant Smoking Study of East Boston (MISSEB), a community-based prospective pregnancy cohort examining asthma, respiratory, and cognitive development. Questionnaires were administered detailing violence exposures (both witnessing and victimization) and avoidance behaviors (staying in at night, avoiding certain areas, keeping children indoors). Because the MIS-SEB did not examine air pollution, we used geographic information systems (GIS) and LUR at the neighborhood scale to retrospectively estimate pollution exposures, using monthly NO2 data collected over 18 years in the surrounding neighborhoods.

9Pregnant women were recruited from East Boston Neighborhood Health Center (EBNHC), Boston, Massachusetts, between 1987 and 1993, as described elsewhere. East Boston is a working-class urban neighborhood bisected by major highways and access roads to Logan International Airport (Figure 1). Following 888 live births originally enrolled, caregivers of 417 children completed questionnaires in 1997 detailing the child’s lifetime violence exposure. Loss to follow-up was attributed largely to families moving out of the neighborhood; those who moved but continued to participate are included. Written informed consent was obtained from participants (mothers) before study initiation, in accordance with both Brigham and Women’s Hospital and Beth Israel Deaconess Medical Center Human Subjects Committees.

10NO2 has been shown to be a reliable indicator for traffic-related primary air pollution. We used a long-term spatially resolved NO2 data set, explored temporal trends in pollution concentrations, compared multiple traffic indicators as predictors of concentrations, and developed retrospective exposure indices.

11Passive NO2 samples were collected contemporaneously 1 week each month from January 1987 through December 2004, using Palmes tubes and analyzed by spectrophotometry. Samples were collected at 28 unobstructed locations, 1 m above ground, across the Logan Airport grounds and surrounding communities. We used monthly averages for 13 sites within community spaces, geocoded by hand using aerial photography. Geocoding (identifying residential locations on an active map embedded in a GIS) allowed for the analysis of spatial characteristics of each location. Missing concentrations were imputed using weighted average concentrations for surrounding months at the same site.

12To explain variability in NO2, we considered 25 traffic indicators derived from Massachusetts Highway Department (MHD) 1990 data (Table 1) and site characteristics (land use, elevation, proximity to industrial areas, population density) derived from U.S. Census 2000 data and aerial photography.

13where Yearj is a categorical indicator for each sampling year j and capturing secular pollution trends; traffici is a suite of traffic characteristics for site i (candidate variables listed in Table 1); and land usei is a suite of candidate site characteristics (listed above).

14Candidate variables were selected with p < 0.05 in univariate nonparametric associations with NO2. Backward elimination produced a parsimonious model with p < 0.1 for all terms. Because the assumption of normal distribution in NO2 could not be rejected (Shapiro-Wilk p < 0.0001), concentrations were not transformed. Models were run in PROC GLM, corroborated in PROC MIXED (SAS Institute Inc., Cary, NC) with random effects by site, to account for within-site autocorrelation.

15Using the predictive NO2 model, we created residential exposure indices. Each child’s address at enrollment (during pregnancy) and follow-up were geocoded using the ESRI StreetMap Address Locator (ESRI, Redlands, CA), and spatial variables derived in GIS, to apply Equation 1 to each address. NO2 exposure estimates were defined for each year of follow-up, for both reported residences. NO2 exposures for years following 1997 were derived using the questionnaire address; between 1990 and 1997, lacking residential information, we created time-weighted annual exposures using the two known addresses. For example, estimated exposure in 1991 for a child enrolled in 1990 is: [(6/7) × (estimated NO2 in 1991 at residence reported at enrollment) + (1/7) × (estimated NO2 in 1991 for residence reported in 1997)].

16We created this lifetime NO2 exposure trajectory for each child, and calculated exposures during seven intervals potentially influencing asthma etiology: a) life course through diagnosis (or end of follow-up); b) year of birth; c) before 5 years of age (median age of diagnosis); d) year of ETV questionnaire (known residential address); e) years between first violent event and diagnosis; f ) year of diagnosis; and g) 1 year before diagnosis.

17For intervals ending at diagnosis, median diagnosis age was used for nonasthmatics. Some exposures after diagnosis (i.e., at ETV questionnaire) are considered to be attributed to the measure’s relative stability. We used univariate logistic regression for each NO2 interval against asthma diagnosis to select an optimal exposure interval, which is applied going forward. Sensitivity analyses on the final epidemiologic models assessed whether interval selection affected results.

18Exposure to violence was assessed using the My Child’s ETV scale. At interview, children ranged from 4 to 11.5 years of age; those > 8 years of age also answered the questionnaire themselves. The questionnaire includes items on witnessing events: a) hitting, slapping, punching, b) a shooting, c) a stabbing, or d) hearing gunshots. We added one question on witnessing domestic verbal abuse. Respondents indicate the event frequency, over their lifetime and previous year, on a scale of 1 (0–1 time) to 4 (more than 10 times). Caregivers reported child’s approximate age at first witnessing.

19Rasch modeling summarized responses into a continuous score, modeling the conditional probabilities for each “yes” response, given the presumed event severity and each child’s true-but-unobserved exposure (a latent construct). Expanding on this approach, the model was generalized to account for features theoretically influencing severity—frequency, whether events occurred at home, whether the child knew the victim(s) or perpetrator(s)—and included parent and child’s report, wherever available, as detailed elsewhere.

20As a validity check, we assessed the relationship between the Rasch ETV and Checklist of Children’s Distress Symptoms (CCDS), administered contemporaneously. CCDS is a parental report of 24 post-traumatic stress symptoms over the prior 6 months (e.g., irritability, inability to fall asleep, nightmares, fear of attending school). The lifetime Rasch ETV and 6-month CCDS were significantly correlated (Spearman r = 0.21, p < 0.0001), corroborating an association between violence and distress symptoms reported elsewhere.

21During MISSEB follow-up, parental reports of child’s asthma diagnosis were ascertained through bimonthly telephone or face-to-face interviews over the child’s first 2 years, every 6 months through 4 years of age, and annually thereafter. Parents were asked “Since we last spoke to you, have you been told by a doctor or nurse that your child has asthma?” and likewise for asthmatic bronchitis. The same questions were asked on ETV questionnaire administration. Children were considered to have asthma if the parent reported any diagnosis of asthma or asthmatic bronchitis.

22We also ascertained demographic characteristics, smoking, and medical history through standardized questionnaires administered during MISSEB follow-up visits. Maternal education was categorized as less than high school, high school or technical school graduate, or some college. At each clinic visit during pregnancy, mothers were asked about smoking status. A urine specimen was obtained for determination of a creatinine-corrected cotinine level as detail, and mothers were classified as never-smokers during pregnancy if they always reported never smoking and their urine cotinine levels were consistently < 200 ng cotinine/mg creatinine. Maternal asthma status was based on report of physician-diagnosed asthma using the standardized American Thoracic Society respiratory questionnaire.

23We examined independent associations of air pollution and ETV on asthma diagnosis using univariate odds ratios (ORs). To examine the modifying effect of violence on the NO2–asthma relationship, we constructed two interaction models, of the forms used elsewhere:

24where ILowETV = 1 if below-median ETV (reference group), 0 otherwise. IHighETV = 1 if above-median ETV, 0 otherwise. NO2 is a centered continuous variable with SD = 1.0; ORs thus refer to a 1-SD increase above the mean. Potential confounders included maternal asthma, education, smoking before and after pregnancy, child’s sex and age. Equation 2 produces the slopes by ETV group and their significance.

25A second regression model contained main effects for ETV and NO2, and their interaction:

26Equation 3 produces the statistical test of the interaction; if β3 differs significantly from zero, ETV significantly modifies the association of NO2 on asthma.

27Because NO2 data were collected only in East Boston and adjoining Winthrop, we expect NO2 estimates to be more accurate for children always living in these neighborhoods. Likewise, the Rasch ETV likely better reflects chronic stress among children always living in the same community where exposure occurred. Therefore, all statistical analyses are performed for the entire cohort, and repeated using only lifetime residents of East Boston and Winthrop.

28We were able to geocode home addresses for 409 children at follow-up, and 369 at enrollment. Demographics of those children and their caregivers are presented in Table 2. Caregivers of 100 children (24%) reported asthma diagnoses during follow-up; caregivers of an additional seven children (2%) retrospectively reported asthmatic bronchitis. Only 8% of mothers reported ever having asthma. ETV varied significantly; at least one violent event was reported for 46% of children, at least two for 18%, and Rasch scores ranged from −1.00 to 2.93. Lifetime residents did not significantly differ from the full cohort in ETV, asthma rates, or other demographic factors or confounders.

29Over 18 years, 2,291 monthly NO2 samples were collected and analyzed, averaging 24.0 ppb (0.78–69.4 ppb). Some overall NO2 decline occurred over time, with some reordering of sites (Figure 2). LUR models explained significant variability as a function of secular trends, traffic density within 500 m, distance from major roads, and block group population density (R2 = 0.83; Table 3). More variability in NO2 was explained by spatial (R2 = 0.53) than temporal terms (R2 = 0.29). Residential estimates averaged 27.5 ppb (18.7–42.6 ppb) overall, higher than measured concentrations, because several NO2 samplers were located in open spaces (i.e., parks, backyards), whereas cohort homes clustered near major roads (Figure 1).

30Univariate ORs for the seven candidate NO2 exposure periods indicated that a 1-SD increase only in year-of-diagnosis NO2 (4.3 ppb) showed near-significant associations with asthma for the full cohort [OR = 1.17; 95% confidence interval (CI), 0.94–1.46], and was used going forward as our key exposure metric.

31For the full cohort, we found no independent effect of ETV on asthma (OR = 0.98; 95% CI, 0.78–1.22). Stratified analyses, however, indicated that year-of-diagnosis NO2 was significantly associated with asthma solely among children with above-median ETV (OR = 1.65; 95% CI, 1.16–2.34). For children with below-median ETV, there was no association between NO2 and asthma (OR = 0.94; 95% CI, 0.70–1.26).

32Among lifetime resident children, with less expected exposure measurement error, we found similar effects with greater magnitude. There was a near-significant effect of NO2 on asthma (OR = 1.28; 95% CI, 0.97–1.69), with no independent effect of ETV (OR = 1.12; 95% CI, 0.84–1.48). Stratified analyses indicated that NO2 was significantly associated with asthma solely among children with above-median ETV (OR = 2.33; 95% CI, 1.47–3.71, vs. OR = 0.87; 95% CI, 0.59–1.28 with below-median ETV).

33We used multivariate logistic regression to formally test the observed interaction, adjusting for potential confounders. In stratified analysis across the full cohort, we found elevated odds of asthma with increased NO2 solely among children with above-median ETV (Table 4), and the magnitude was unaffected by addition of potential confounders (adjusted OR = 1.63; 95% CI, 1.14–2.33). The interaction between NO2 and ETV was statistically significant (p = 0.03), and robust to inclusion of both main effects.

34Among the lifetime residents, in multivariate logistic regression, we found similar associations with greater magnitude. Stratified analyses indicated elevated odds of asthma with increased NO2 solely among children with above-median ETV (adjusted OR = 2.40; 95% CI, 1.48–3.88). The NO2–ETV interaction term was statistically significant (p = 0.0009), and robust to inclusion of main effects.

35Sensitivity analyses indicated that only average NO2 for the years between first violence exposure and asthma diagnosis could be substituted for year-of-diagnosis NO2 with significance (adjusted OR = 1.52; 95% CI, 1.05–2.19) for children with above-median ETV, in the full cohort multivariate model. We repeated analyses removing 10 children whose diagnosis occurred before first violent event, with little bearing on results. Last, sensitivity analyses indicated that median-dichotomized My ETV, My Child’s ETV, and CCDS scores could not be substituted for the Rasch indicator with significance.

36We found an association between traffic-related pollution (NO2) and asthma diagnosis only among children with elevated ETV, after controlling for potential confounders. Of multiple exposure intervals, year-of-diagnosis NO2 best predicted asthma (with some evidence for NO2 after first violence exposure), supporting a theoretical model wherein individuals may become susceptible or “primed” through social pathways to some environmental triggers, including traffic-related pollutants. Despite limitations of our data sets, these results agree with evidence elsewhere that chronic stress may shape biologic response in early life and potentiate effects of air pollution through common physiologic systems.

37Our findings contribute to the existing environmental justice literature by identifying potential changes in pollution susceptibility within communities affected by violence and other stressors. We also indicate ancillary effects of violence on children in urban communities, in addition to direct injury and post-traumatic stress. identified associations among neighborhood deterioration, violence, and low collective efficacy (social control of neighborhoods through residents’ collective effort); we observed that a heightened susceptibility to pollution, associated with violence exposures or fear thereof, may lead to synergistic health effects of social and physical environmental conditions.

38Although our findings are biologically plausible and highly suggestive, there are potential limitations in interpreting our models. A clear limitation is our small sample size for investigating multiplicative effects; despite the relatively high asthma prevalence (26%), a larger cohort would be required to consider interactions across numerous potential risk factors. The pollution model relies on NO2 data, commonly used as a marker of primary vehicular emissions, but not necessarily representative of all pollutants of interest. Road construction and airplane technology changes made it unclear a priori whether NO2 could be predicted by spatial indicators in this neighborhood, particularly using 1990 traffic data, though our model proved robust. We lacked complete residential history, likely creating some misclassification in a cohort with significant residential instability. However, we found no difference in asthma prevalence or exposures between movers and nonmovers; thus we expect misclassification to be nondifferential, biasing results toward the null and making our significant results more noteworthy.

39The interpretation of our interaction model is complicated by the fact that behaviors may differ in violent neighborhoods, where parents often keep children indoors due to fear of violence. Therefore, children may be more exposed to indoor NO2, which is generally higher than outdoor NO2, and is influenced both by infiltration from outdoors and by smoking, gas stoves, and other sources which may be more prevalent in lower-income communities. Future studies should examine differential susceptibility to a wider range of pollution exposures, including indoor allergens and environmental tobacco smoke; although we accounted for maternal smoking, we did not examine the effect of other smokers in the home. Likewise, ETV may proxy for other social exposures responsible for the susceptibility effect we found; children witnessing violence may have greater family instability or may be directly victimized, potentially leading to injury-related susceptibilities.

40We aimed to investigate whether a psychosocial stressor modifies pollution effects on asthma, and would prefer a long-term stress measure to corroborate the ETV scale. The CCDS elicits distress using a 6-month symptoms recall, inappropriate to our goal of capturing life-course stress. The correlation between CCDS and violence does, however, corroborate an association, supporting the plausibility of stress pathways to pollution susceptibility, and should be investigated further.

41Reporting bias—generally underreporting by survivors, perpetrators, and witnesses—hampers quantitative violence research, particularly for domestic and intimate violence. Our questionnaire focused largely on within-community events, potentially more accurately reported. A prior analysis indicated that parents underestimate older children’s exposures, but are better corroborated for near-home events, potentially due to shared experience. We were unable to examine direct victimization because very little was reported, owing either to low prevalence or underreporting, and thus we likely have some misclassification of true ETV. Most asthma cases were reported during longitudinal follow-up, limiting recall bias in diagnoses, but our retrospective violence report is subject to recall bias. To assess recall bias in violence reports, we asked the caregiver to report on asthma episodes triggered by violence; because very few parents associated violent events with asthma symptoms, recall bias in ETV by asthma status is unlikely.

42Despite these limitations, our study provides evidence of a synergistic effect between social and physical factors in asthma etiology. The study also provides a model for retrospectively estimating traffic-related exposures, accounting for intracommunity heterogeneity and temporal trends using GIS. We were able to create temporally calibrated estimates using a spatially dense and temporally rich pollution data set of NO2 measurements collected throughout our epidemiologic study period. Though few studies will have access to such data, existing models provide insight about site characteristics influencing exposure variability, and spatially resolved satellite pollution data may prove useful in some settings.

43In future studies, a wider and more frequently assessed suite of social stressors and perceived stress measures should be employed to examine stress trajectories over time and temporality in susceptibility to environmental triggers. The roles of social support and health beliefs in mediating the effects of perceived stress on health outcomes should be considered as buffers in the pathway from stressor to stress response, the latter of which may influence contaminant susceptibility and health. We observed no correlation between ETV and pollution levels, but suggest that longitudinal investigation of multiple exposures, and multiple pathways, are needed to further elucidate these effects. The spatial correlation across multiple exposures deserves greater investigation, and should be addressed in larger studies investigating spatial autocorrelations across, and potential synergies among, social and physical environmental factors.