Prenatal ambient air pollution exposure and small for gestational age birth in the Puget Sound Air Basin
Several studies have identified high concentrations of air pollution as harmful to the developing fetus, but few studies of traffic-derived air pollution and birth outcomes have been conducted in areas of low to moderate air pollution. We identified singleton live births between 1997 and 2005 (N = 367,046 births) in the Puget Sound Air Basin of Washington State. We estimated nitrogen dioxide (NO2) exposure using a land use regression model of traffic, PM2.5 exposure from the nearest community monitor, and proximity to highways/roadways for the residential location of all subjects. Logistic regression estimates of odds ratios (OR) of small for gestational age (SGA) and low birth weight (<2,500 g) among term births were calculated. We observed a modest association between SGA births with increasing quartile of first trimester NO2 exposure: second (OR = 1.01, 95 % confidence interval (CI) 0.97, 1.04), third (OR = 1.06, 95 % CI 1.03, 1.10), and fourth (OR = 1.08, 95 % CI 1.04, 1.12) (p trend <0.001). We did not observe an association between PM2.5 and SGA or low birth weight among term births. Our findings suggest that prenatal exposure to traffic-derived air pollutants has a modest effect on fetal growth in a region with low overall air pollutant concentrations. Given the modest associations, future studies in similar settings that maximize the opportunity to address potential residual confounding are needed.
KeywordsAir pollutionSmall for gestational ageLow birth weightTraffic
Body mass index
Land use regression
Micrograms per cubic meter
Particulate matter ≤2.5 μm in aerodynamic diameter
Parts per billion
Small for gestational age
An accumulating evidence base links higher exposure to traffic-derived air pollutants with adverse birth outcomes, particularly increased risk of low birth weight (Bobak 2000; Glinianaia et al. 2004; Sram et al. 2005; Slama et al. 2007; Brauer et al. 2008; Aguilera et al. 2009; Woodruff et al. 2009; Wilhelm et al. 2011). Most studies have been conducted in settings with relatively high concentrations of ambient air pollutants, including major urban areas around the world and in North America (Dejmek et al. 1999; Ritz et al. 2000; Wilhelm and Ritz 2003; Bell et al. 2007; Slama et al. 2007; Aguilera et al. 2008).
In Washington State, singleton low birth weight rates have increased by 1.2 % between 1993 and 2004 with unidentified etiology (Washington State Department of Health 2007). However, Washington State and its most urbanized regions in the Puget Sound area have relatively low concentrations of ambient air pollutants. Asthma has been associated with exposure to air pollutants in the area (Sheppard et al. 1999; Mar and Koenig 2009). In addition, an increased risk of small for gestational age (SGA) birth in relation to low to moderate level air pollutant exposures has been observed in the Vancouver, Canada, air basin (Brauer et al. 2008).
Traffic exposures are ubiquitous in smaller as well as larger, dense urban areas of North America. If traffic exposures impact birth outcomes even in regions with relatively lower concentrations than major urban cores, the resulting economic and public health burden on child and maternal health could be significant. A key challenge in studies of traffic to date is addressing the combination of both the spatial and temporal components of exposure variability for studies of air pollution and pregnancy outcome (Woodruff et al. 2009). Recently, studies have employed knowledge of both the mother's residential locations and the timing of relevant periods of pregnancy in the predictions using land use regression modeling (LUR) for individual exposure assessment to begin to address some of these questions (Slama et al. 2007; Brauer et al. 2008; Aguilera et al. 2009; Wilhelm et al. 2011).
We sought to explore the role of traffic-derived air pollution in an airshed with relatively low pollution concentrations using the LUR approach to model individual-level NO2 exposures as well as straightforward measures of proximity to roadways to represent traffic exposure and proximal PM2.5 community monitoring results to assign fine particulate matter exposure.
Material and methods
The study population consisted of all singleton live births between 1997 and 2005 in the four-county central Puget Sound Region (King, Kitsap, Snohomish, Pierce) which includes the large metropolitan areas of Seattle and Tacoma in Washington State. Birth record data were obtained from the Washington State Department of Health Birth Records Database which records data from the birth as well as demographic data from the mother and father at the time of birth (N = 371,451 births). We excluded births with missing (N = 4,198) or implausible gestational age (<23 and >45 weeks, N = 183) or implausible birth weight (<450 g, N = 24). We identified multiple singleton births to the same mother (N = 281,104 clusters). The final study population consisted of 367,046 births. We defined SGA as those births below the tenth percentile by gender and week of gestation. Low birth weight was defined as <2,500 g, and term births were defined as greater than or equal to 37 weeks of gestation. We obtained data on potential confounding variables from the birth records database: maternal age, urban/rural residence (based on US Census Bureau classification of urban as areas that have a population density of at least 1,000 people/square mile), duration of residence, year of birth, season of birth, maternal race/ethnicity, parity (number of live births a mother had previous to the current birth), income (median family income per census tract of residence), alcohol use, prenatal care utilization, and smoking. This study was approved by the Washington State Department of Health Human Subjects Board.
We estimated NO2 exposures based on a LUR model of NO2 ambient concentrations which we developed for the region. The LUR model followed the form of a model initially developed for Vancouver, Canada, by Henderson et al. in which regression methods were used to model pollutant concentrations measured at specific sites based on variables that characterize surrounding land use, population density, and traffic patterns (Henderson et al. 2007). An NO2 dataset collected in Seattle to refine a field sampling protocol was used to derive our LUR model for the Seattle region (Poplawski et al. 2009). The model was calibrated based on 2-week integrated field measurements of ambient NO2 using passive Ogawa samplers conducted at 26 sites in Seattle, Washington, and an agency continuous analyzer for the period of July 2004 through June 2005.
Values of the predictor variables were generated from data obtained from the Puget Sound Regional Council, county road networks, the US Environmental Protection Agency, the US Census for all of the birth residence sites, and a corresponding annual NO2 concentration estimated from the Seattle LUR model (Archive, Bureau. July 1, 2005).
The NO2 field measurements were adjusted to reflect the annual averages through comparison of annual and 2-week NO2 measured by continuous analyzer at the Washington Department of Ecology Beacon Hill monitor site where one of the passive samplers was collocated. This approach assumes the yearly variation of 2-week averages follows the same pattern at all sites and has been employed in most previous LUR models.
To derive monthly concentrations of NO2 exposure for our subjects, we derived a monthly NO2 adjustment factor for each month from 1997 through 2005. This was computed as the ratio of the monthly NO2 average at the central Beacon Hill site to the annual average for the period of July 2004 through June 2005. Monthly variation of NO2 followed a reasonably consistent pattern throughout all 9 years of the study: mean annual NO2 exposure ranged from 12.9 to 14.3 ppb. Residential addresses at the time of delivery were geocoded by the Department of Health and Human Services.
We assigned daily exposure concentrations for PM2.5 using Puget Sound Clean Air Agency regulatory data for the years 1997–2005. The mother's geocoded address was linked to the nearest monitor within 20 km. For most monitors during this study period, daily average PM2.5 measurements were conducted at least every third day. The number of ambient PM2.5 monitors in the region was 4, 9, 16, and 17 within the years 1997–2000, respectively, and 18 in subsequent study years. Annual mean PM2.5 ranged from 9.3 to 12.6 μg/m3 across all years of study.
For all births, monthly pollutant averages for potentially specific “toxicity windows” of pregnancy were calculated: first month, first 3 months, last month, last 3 months, and entire pregnancy based on delivery date and date of conception using the last menstrual period. If last menstrual period data were not available, the recorded gestational age in weeks was used. We excluded all 1997 births from the PM2.5 analysis due to a large quantity of missing PM2.5 exposure data for the pregnancies conceived in 1996 (10,531 births in 1997 were missing PM2.5 exposure data). In addition, we excluded births with any period of missing PM2.5 data (35,698 births with some missing exposure data). The final PM2.5 dataset for analysis included 323,899 births.
We also classified maternal residence according to proximity within 50 or 150 m of roads categorized as either expressways or highways such as state interstates (Rd1, Standard Metropolitan Statistical Area classification codes A10, A11, A15, A20, A21, and A23) or major roads and arterials (Rd2, Standard Metropolitan Statistical Area classification codes A30, A31, A32, and A35). The same network of roads based on 2000 data was applied for 1997–2005.
We first examined the distribution of exposures during all pregnancy windows and categorized exposures per quartile. We also examined proximity to a major roadway (within 50 and 150 m of expressways or highways and within 50 and 150 m of major roads and arterials) as binary predictor variables. In order to account for correlations arising from using more than or equal to one birth per mother, we used general estimating equations to fit logistic regression models with exchangeable correlation matrices, binomial outcome distribution, and a logit link to assess the relationship between exposures and odds of SGA births. We also examined the relationship between NO2, PM2.5 exposures, and road proximity and risk of low birth weight (<2,500 g) restricted to term births only given that toxicological impact may differ for term versus preterm infants. We chose to examine early and late pregnancy to explore the impact of exposures on different developmental periods during pregnancy. Since preterm infants may not have exposures in later gestational stages, analysis restricted to terms also allowed evaluation of the effects of late term exposures that occur at similar fetal developmental timeframes in the population. Because NO2 and PM2.5 exposures were uncorrelated, we fit logistic regression models with both NO2 and PM2.5 as variables.
Several confounding variables were considered in the SGA analysis. A priori, we decided to use maternal age, smoking during pregnancy, race/ethnicity, season of birth, and income as covariates in all analyses given their well established relationship to the outcomes of interest. We examined the other possible confounding variables to determine their influence on point estimates or improved model precision.
Characteristics of the study population (N = 367,046) within the Puget Sound Air Basin
Maternal age (years)
Census tract median income (US $)
Maternal education (years)
Smoking during pregnancy
Alcohol use during pregnancy
Duration lived at current residence (years)
Urban or rural
Season of birth
Small for gestational age
Low birth weight among term births
Distribution of NO2 exposures (by month of gestation) in parts per billion (N = 367,046)
Distribution of PM2.5 exposures (by month of gestation) in micrograms per cubic meter (N = 323,899)
Correlation matrix of NO2 exposures (N = 367,046)
Correlation matrix of PM2.5 exposures (N = 323,899)
Correlation matrix of NO2 and PM2.5 exposures (N = 323,878)
Associations (OR, 95 % CI) between ambient nitrogen dioxide (NO2) exposures during the first and last 3 months of pregnancy and SGA birth within the Puget Sound Air Basin
Unadjusted (N = 367,046)
Adjustedb (N = 337,372)
95 % CI
95 % CI
Average NO2 during first 3 months of pregnancy (ppb)
Average NO2 during last 3 months of pregnancy (ppb)
Associations (OR, 95 % CI) between ambient fine particulate matter (PM2.5) exposures during the first and last 3 months of pregnancy and SGA birth within the Puget Sound Air Basin
No. of casesa
Unadjusted (N = 323,899)
Adjustedb (N = 298,835)
95 % CI
95 % CI
Average PM2.5 during the first 3 months of pregnancy (μg/m3)
Average PM2.5 during the last 3 months of pregnancy (μg/m3)
Associations (OR, 95 % CI) between proximity to major roadways and SGA birth within the Puget Sound Air Basin
No. of casesa
Unadjusted (N = 364,757)
Adjustedb (N = 335,275)
95 % CI
95 % CI
Residence within 50 m of a freeway or highway
Residence within 150 m of a freeway or highway
Residence within 50 m of a major road or arterial
Residence within 150 m of a major road or arterial
We explored the relationship between several air pollutant measures and SGA in a very large cohort of births in the Puget Sound Air Basin, an area of low air pollution exposures. Our results suggest a relationship between two measures of traffic: prenatal NO2 exposure and proximity to freeway/highway, and SGA birth in the Puget Sound Air Basin.
These results are somewhat consistent with those observed by Brauer et al. in the Vancouver Air Basin (Brauer et al. 2008). In their analysis, pregnancy average NO2 exposure assessed based on the monitoring network yielded the greatest magnitude of effect estimate for SGA among the pollutants examined. However, their analysis of NO2 based on a similarly derived LUR also did not identify an association nor were their assessments of pregnancy averaged PM2.5 exposure (LUR and monitor based) significantly associated with SGA. Brauer et al. found moderate concordance between LUR and ambient network data for NO2 (r = 0.37). This lack of strong correlation may suggest that LUR represents a different measure of air pollution that reflects spatial heterogeneity more precisely (Marshall et al. 2008; Wilhelm et al. 2011). The most recent emission inventory from the Puget Sound Clean Air Agency reported that gasoline and diesel motor vehicles were the largest contributors to NO2 emissions in the region (PSCA 2008). Our model takes into account roadways, land use types, population density, and elevation. Because results of our LUR NO2 analysis and road proximity analyses were not in complete concordance, we believe that our LUR NO2 may also reflect the atmospheric ozone/NOx equilibrium that adds to spatial heterogeneity as suggested by Keuken et al. (2009). We were unable to validate or contrast the results of the Puget Sound NO2 LUR exposure models with monitor data due to lack of extensive NO2 monitor data in the Puget Sound region. Ambient monitors provide numerous direct exposure measurements and include more precise temporal information that the LUR model is not designed to capture. Therefore, there is a considerable amount of uncertainty in our LUR exposure estimates.
Our mean NO2 exposure estimates (13.7 ppb) were lower than annual averages derived from monitor data in the Puget Sound region for this period (approximate mean, 20 ppb from 1997 to 2005) (Puget Sound Clean Air Agency 2006) and slightly lower than those in the neighboring Vancouver Air Basin analysis (16.8 ppb) (Brauer et al. 2008). The former result is not surprising given that the single Environmental Protection Agency NO2 monitor in Seattle was initially sited in a relatively high area of NO2 (Norris and Larson 1999). In contrast, our annual mean PM2.5 exposure estimates (10.3 μg/m3) by monitor data were higher than those in the Vancouver Air Basin study (mean, 5.3 μg/m3) but still lower than the current Environmental Protection Agency annual standard of 15 μg/m3 (Puget Sound Clean Air Agency 2006).
We observed an association between SGA birth (OR = 1.11) associated with living within 50 m of a freeway/highway but not with living in close proximity (50 or 150 m) to major roadways/arterials. In the Vancouver region study, an increased risk of small for gestational age birth in relation to proximity to major roadways/arterials was observed, and Wilhelm et al. also observed increased risk for low birth weight for those living in close proximity to major roadways/arterials in Los Angeles County, an area of high air pollution (Wilhelm and Ritz 2003; Brauer et al. 2008). Taken together, these studies along with ours reported here suggest that road proximity may be an independent predictor of birth outcomes. Proximity may be a more direct measure of exposure to the harmful components of air pollutants than a single pollutant LUR model.
Physiologic mechanisms for air pollutant effects on fetal growth remain largely unknown, but hypotheses include disturbances to placental blood flow, increased risk for premature contractions/premature rupture of membranes through an inflammatory prostaglandin pathway, and increased risk for maternal infections (Wilhelm and Ritz 2005; Leem et al. 2006). Timing of air pollutant exposures and potential windows of susceptibility for fetal growth are inconsistent within the literature, and our analysis explored exposures in the first and last 3 months of pregnancy which may reflect different gestational developmental timeframes for infants born preterm compared to those at term. Known risk factors for prematurity and SGA birth include anemia, drug use, infections, low socioeconomic status, smoking, stress, and previous history of pregnancy complication
Limitations of the current study include the accuracy and availability of potentially important confounder data reported on birth certificates. Smoking and socioeconomic status are recognized as among the most important confounders to address in studies of air pollution. We adjusted for maternal smoking during pregnancy although the limits of self-reporting maternal smoking during pregnancy are appreciated. Some of the potential residual confounding may be captured by the adjustments for socioeconomic status. The majority of our study population was not at a low socioeconomic status as defined by median census tract income, but we attempted to address potential confounding with individual-level data on maternal age and education and income data which were only available at the census tract level. In addition, we were unable to determine how long each woman spent at a particular residence and therefore could not weight our exposure estimates. Lack of exposure variability decreased our ability to discern between time-specific exposures during pregnancy and outcomes. This was specifically true for NO2 exposures and, to a milder extent, PM 2.5. Because categorized NO2 and PM2.5 exposures were relatively uncorrelated (Table 6), we fit logistic regression models with both NO2 and PM2.5 as variables, and results were almost identical to those reported in Tables 7 and 8, suggesting that NO2 and PM2.5 may affect fetal growth with independent pathways and mechanisms of action.
The inability to account for occupational or other sites that constitute the mother's exposure is an exposure assessment limitation. We performed a sensitivity analysis of mothers who lived at their residence for less than 1 year compared to 1–3 years and >3 years and did not observe a significant change in estimates which may reflect that mothers stayed in close proximity to their recorded residence in the birth record or may move to settings with equivalent levels of exposure. We were not able to address personal time-activity or time spent at home versus other settings. Overall, this may have produced nondifferential misclassification of exposure and biased our findings somewhat towards the null hypothesis.
The preterm birth rate of 6.7 % in our primarily white and relatively higher educated population is considerably lower than that of the non-Hispanic white US population in 2005 (12 %). The rate of low birth weight in our entire cohort was 5 % which is similar to that of the corresponding US population (5 %) according to the Centers for Disease Control and Prevention National Vital Statistics Report (Hoyert et al. 2006). Information regarding gestational age and birth weight on birth certificates has been shown to be in high concordance with that abstracted from hospital records (95–99 %) in Washington state; thus, the potential for our outcome definition to be biased is small (DiGiuseppe et al. 2002).
Increasingly, air pollution epidemiological studies are discerning health effects in populations at relatively low concentrations, often below regulatory thresholds (Puget Sound Clean Air Agency 2006). Our findings suggest that maternal first and last 3 months of traffic-derived NO2 exposures as well as road proximity to freeways/highways are associated with SGA birth although we cannot rule out residual confounding. If these findings are causal, they have important implications for public health given that the rate of SGA birth has been increasing in the Puget Sound region without known etiology and the pollutant concentrations observed are within regulatory limits and comparable to levels experienced by the majority of women in US metropolitan areas.
We acknowledge the Ambulatory Pediatric Association/AHRQ Young Investigator Grant Program which provided funds for this analysis and the Health Canada Border Air Quality Study—Western Pilot Initiative that provided funds for creating the LUR models. British Columbia Centres for Disease Control Agreement. Grant No. GEH0404.
Conflicts of interest
Each author has no conflicts of interest to report.