Background: Previous studies of air pollution and birth outcomes have not evaluated whether complicated pregnancies might be susceptible to the adverse effects of air pollution. It was hypothesised that trimester mean pollutant concentrations could be associated with fetal growth restriction, with larger risks among complicated pregnancies.
Methods: A multiyear linked birth certificate and maternal/newborn hospital discharge dataset of singleton, term births to mothers residing in New Jersey at the time of birth, who were white (non-Hispanic), African–American (non-Hispanic) or Hispanic was used. Very small for gestational age (VSGA) was defined as a fetal growth ratio <0.75, small for gestational age (SGA) as ⩾0.75 and <0.85, and ‘reference’ births as ⩾0.85. Using polytomous logistic regression, associations between mean pollutant concentrations during the first, second and third trimesters and the risks of SGA/VSGA were examined, as well as effect modification of these associations by several pregnancy complications.
Results: Significantly increased risk of SGA was associated with first and third trimester PM2.5 (particulate matter <2.5 μm in aerodynamic diameter), and increased risk of VSGA associated with first, second and third trimester nitrogen dioxide (NO2) concentrations. Pregnancies complicated by placental abruption and premature rupture of the membrane had ∼two- to fivefold greater excess risks of SGA/VSGA than pregnancies not complicated by these conditions, although these estimates were not statistically significant.
Conclusions: These findings suggest that ambient air pollution, perhaps specifically traffic emissions during early and late pregnancy and/or factors associated with residence near a roadway during pregnancy, may affect fetal growth. Further, pregnancy complications may increase susceptibility to these effects in late pregnancy.
Statistics from Altmetric.com
A body of evidence is emerging from several countries on the adverse consequences of ambient air pollution on fetal/birth outcomes, including preterm birth and fetal growth restriction.1–20 However, the biological mechanism(s) by which ambient air pollution may impact adverse birth outcomes, which may be different in complicated and uncomplicated pregnancies, is/are not clearly established.
Pathophysiological changes that have been proposed as plausible mechanisms for fetal growth restriction (ie, decreased oxygen saturation, endothelial dysfunction, increased blood viscosity, thrombosis, etc.) have also been associated with air pollution in studies of acute pollution/cardiorespiratory responses.21–24 Further, these mechanisms may also play an important role in the occurrence of pregnancy complications including pre-eclampsia, placental abruption and placenta praevia.25–29 Thus, air pollution-related fetal growth restriction, some pregnancy complications (eg, placental abruption) and cardiorespiratory disease may share common mechanisms. Therefore, we hypothesised that elevated levels of air pollution affect fetal growth in uncomplicated pregnancies, and that pregnancy complications adversely modify the pollution/fetal growth association, making the risk of impaired fetal growth more pronounced among complicated pregnancies.
Using a multiyear, New Jersey (NJ) statewide, linked birth certificate and maternal hospital discharge dataset, and PM2.5 (particulate matter <2.5 μm in aerodynamic diameter), nitrogen dioxide (NO2), sulphur dioxide (SO2) and carbon monoxide (CO) measurements made at monitoring locations across NJ, we examined the effect(s) of ambient air pollutant concentrations during early, middle and late pregnancy on fetal growth restriction among term births. These linked data provide more complete recording of pregnancy complications than birth certificates alone, and provide an opportunity to examine whether the effect of air pollution on fetal growth differs between uncomplicated and complicated pregnancies.
Using linked birth certificate and maternal/newborn hospital discharge summaries maintained by the Division of Family Health Services, NJ Department of Health and Senior Services (NJDHSS), we selected all singleton births in NJ from 1999 to 2003 to white (non-Hispanic), African–American (non-Hispanic) or Hispanic mothers who were residents of NJ at the time of birth, with a gestational age of 37–42 completed weeks and a birthweight ⩾500 g. The study was approved by both UMDNJ and NJDHSS Institutional Review Boards.
From the birth certificate, we extracted data on maternal characteristics (ie, age, race/ethnicity, marital status, education level, and cigarette smoking, drug use and alcohol use during pregnancy), maternal place of residence at the time of birth, trimester of first prenatal care visit, infant birthweight and gender. Also from the birth certificate, we retained data on the start day, month and year of the last menstrual period (LMP) and the clinical estimate of gestational age. If either the birth certificate or the maternal discharge data indicated specific pregnancy complications (gestational hypertension, pre-eclampsia, eclampsia, gestational diabetes, placenta praevia, placental abruption or premature rupture of the membranes), we coded that subject as having that complication. This approach provides a higher sensitivity and specificity than the use of birth certificates or maternal hospital discharge data alone.30–33
We estimated gestational age based on LMP using the algorithm proposed by the National Center for Health Statistics.34 Gestational age information reported on the basis of women’s menstrual history has been shown to be reasonably reliable.33 35 36 For each birth, we calculated a fetal growth ratio as a measure of newborn size.37 38 For each gestational age/gender/race-specific stratum (eg, white males with gestational age of 38 weeks), we calculated the median birthweight. Each newborn’s/birth’s fetal growth ratio was then calculated as the newborn’s birthweight divided by the median birthweight of the corresponding stratum. We then defined very small for gestational age (VSGA) as a fetal growth ratio <0.75, small for gestational age (SGA) as ⩾0.75 and <0.85, with all fetal growth ratios ⩾0.85 comprising the reference group. The cut-off values for defining VSGA and SGA have been validated by other investigators.37 38 This method of measuring fetal growth has been used previously by our group39 and others.37 38
All pollutant measurements by the NJ Department of Environmental Protection were retrieved from the United States Environmental Protection Agency website.40 PM2.5 measurements (over a 24-hour period) were made every third day at 20 monitoring sites in NJ from September 1999 to December 2003. NO2 was measured continuously at 11 stations, SO2 continuously at 16 stations, and CO continuously at 16 stations for the study period.
To each subject/birth, we assigned measurements from the PM2.5 monitor closest to the maternal residence at birth. However, we excluded all births whose maternal residence was >10 km from the closest monitoring station. Using the estimated date of conception, we calculated the mean first trimester (first 93 days from estimated date of conception) and mean second trimester PM2.5 concentrations (second 93 days from estimated date of conception). The mean third trimester PM2.5 concentration was calculated as the mean PM2.5 concentration during the remaining pregnancy time (third trimester ranged from 73 to 108 days). We calculated a mean concentration for only those trimesters with <30% of the scheduled PM2.5 measurements missing. If ⩾30% were missing, we set that trimester-specific mean PM2.5 concentration to missing. We then calculated trimester-specific NO2, SO2 and CO concentrations in the same manner and used these concentrations in all subsequent analyses.
Neighbourhood level socioeconomic status (SES)
To control for neighbourhood characteristics of the maternal residence that may be both associated with birth outcomes and correlated with air pollution concentrations, we abstracted the following variables from the 2000 US Census, by census tract:41 percentage of persons aged 25 years and older with less than a high school education, percentage of persons aged 25 years and older with at least 4 years of college education, and percentage of persons below the federally defined poverty line. These area-based variables have been shown to be reasonable measures of neighbourhood level SES,42 which may predict health risks associated with neighbourhood characteristics independent of individual level SES measures.43 The latitude and longitude of the maternal residence at birth were used to identify the census tract in which each mother resided, using ArcGIS v.9.2 (ESRI, Redlands, CA, USA). We then assigned each birth/mother values of these three area-based US census variables.
We used a cohort study design and polytomous logistic regression (SAS Proc Catmod, SAS Inc, Cary, NC, USA) to estimate the risk of SGA and VSGA, compared with the reference group, associated with incremental increases in mean PM2.5 concentration in the first trimester. In this model, we included those covariates that were not thought to be on the causal pathway from PM2.5 to SGA/VSGA, which changed the pollutant effect estimate by 10% and/or were predictors of SGA/VSGA. These included maternal age, education and race, trimester of prenatal care initiation, maternal smoking, drug use and alcohol use during pregnancy, marital status, percentage of the maternal residence census tract’s population 25 years and older with <12 years of education, percentage with ⩾4 years of college education and the percentage of the census tract’s population living below the poverty line. We then re-ran this same model without the first trimester PM2.5 to separately examine effects associated with second and then third trimester mean PM2.5 concentrations, as well as first, second and third trimester mean NO2, SO2 and CO concentrations. From each model, we report the excess risk and its 95% confidence interval.
To evaluate our assumption of a linear concentration response, we replaced the continuous pollutant concentration (eg, first trimester PM2.5) with indicator variables based on quintiles and re-ran the same one pollutant model described above. We then used an ordinal variable to replace these quintiles to perform a test for trend. To assess the stability of our single pollutant model risk estimates (eg, first trimester PM2.5) after adjustment for other pollutant concentrations, we ran the same models including two pollutant concentrations from the same trimester (eg, first trimester PM2.5 and first trimester NO2). To determine whether our findings were sensitive to the definitions of SGA/VSGA used (ie, fetal growth ratio vs <10% tile), we re-defined VSGA as a birthweight less than the third percentile of the corresponding gestational age-, gender- and race-specific distribution of birthweights, SGA as greater than or equal to the third percentile and less than the 10th percentile, and our reference birth group as greater than or equal to the 10th percentile. We then re-ran the same model described above. To evaluate whether our findings were restricted to one racial/ethnic group, we evaluated effect modification by maternal race. To evaluate whether our findings were sensitive to control for long-term trends, season and temperature, we included indicator variables for the month and calendar year of birth, and linear and quadratic terms of first trimester mean apparent temperature.44 For each subject, we used temperature and dew point measurements made at the closest airport to the maternal residence, and from these calculated apparent temperature as a measure of the subject’s perceived air temperature given the humidity.
Effect modification by pregnancy complications
We investigated whether the association between fetal growth restriction and PM2.5 differed in those women with and without pregnancy complications. We created an indicator variable for the presence of each pregnancy complication (ie, gestational hypertension, gestational diabetes, pre-eclampsia, eclampsia, placenta praevia, placental abruption and premature rupture of the membrane), and then included an interaction term (PM2.5 * pregnancy complication) in the model. All statistical analyses were done using SAS v.9.1 (SAS Inc., Cary, NC, USA).
There were 492,678 singleton births to white (non-Hispanic), African–American (non-Hispanic) and Hispanic mothers who were residents of NJ from 1999 to 2003. After retaining only those births with gestational ages of 37 to 42 weeks, and excluding all observations with missing data on birthweight, date of birth, LMP and other covariates, 350,107 births remained (n = 27,943 SGA births (8%) and n = 7773 (2%) VSGA births). Births with a maternal residence >10 km from a monitoring station or those missing trimester-specific mean pollutant concentrations were then excluded, leaving n = 88,678 births for analyses involving PM2.5, n = 132,888 for SO2, n = 114,411 for NO2 and n = 134,798 births for analyses involving CO. There were n = 199,221 births included in at least one pollutant-specific analysis.
Mothers of SGA and VSGA infants were more likely to be less than 25 years old and less likely to have completed high school, compared with mothers of appropriate size births (table 1). They were also more likely to be single, African–American and to have smoked during pregnancy. The frequencies of gestational hypertension, pre-eclampsia, fetal distress, placental abruption and premature rupture of membranes were highest for mothers of VSGA infants, intermediate for mothers of SGA infants and lowest for mothers of appropriate size infants. Mothers of VSGA and SGA infants lived in census tracts where greater proportions of residents had less than a high school education and lived in poverty, compared with mothers of births in the referent group (table 1).
Mothers of infants excluded from the analysis (ie, no pollutant monitoring station <10 km from the maternal residence) were generally older (23% ⩾35 years), had earlier prenatal care (86% in the first trimester) and were more likely to be white (77%), married (77%) and have had some college education (63%) than the mothers of reference births included in the analysis (table 1). The frequencies of specific pregnancy complications, however, were similar (eg, pre-eclampsia 2%; gestational diabetes 4%; placental abruption 1%; premature rupture of the membrane 4%) to the reference group.
Subject-specific first trimester mean PM2.5 concentrations ranged from 2 to 29 μg/m3, NO2 from 5 to 47 ppb, SO2 from 1 to 14 ppb and CO from 0.137 to 2.195 ppm. The mean and standard deviation for subjects’ trimester-specific mean pollutant concentrations are shown in table 2. Subject-specific first, second and third trimester NO2 and CO concentrations were each highly correlated (eg, first trimester CO and second trimester CO: r = 0.88), but subject-specific first, second and third trimester SO2 and PM2.5 concentrations were not (table 3). Trimester-specific NO2 and CO concentrations were moderately correlated (eg, first trimester NO2 and first trimester CO: r = 0.51), and all other pollutant/trimester pairs were uncorrelated.
When we evaluated each trimester-specific pollutant concentration separately, each 4 μg/m3 increase in both the first and the third trimester mean PM2.5 concentration was associated with significantly increased risk of SGA (table 4). The first and third trimester VSGA excess risk estimates were also greater than 0, but not statistically significant. Each 10 ppb increase in each of the first, second and third trimester mean NO2 concentrations was associated with significantly increased risk of VSGA, but not SGA. No trimester-specific mean SO2 or CO concentration was associated with increased risk of SGA or VSGA (table 4).
When including first trimester PM2.5 and NO2 concentrations in a model simultaneously (n = 59,955 births with both PM2.5 and NO2 trimester mean concentrations), the PM2.5/SGA and NO2/VSGA risk estimates were not substantially different from the risk estimates from single pollutant models on those same n = 59,955 subjects (table 5). This was also true for the second and third trimester risk estimates. Risk of SGA or VSGA generally increased with increasing quintiles of first and third trimester PM2.5 concentration (figure 1) and first, second and third trimester NO2 concentrations, although not always (figure 2).
When we redefined SGA and VSGA as less than the 10th and third percentiles, respectively, the excess risk estimates were generally consistent with our previous PM2.5/SGA estimate (first trimester: 4.5%, 95% CI −0.5% to 8.7%; third trimester: 4.1%, 95% CI 0.3% to 8.0%), and our NO2/VSGA estimates (first trimester: 7.0%, 95% CI 1.8% to 12.4%; second trimester: 7.7%, 95% CI 2.6% to 13.0%; third trimester: 7.4%, 95% CI 2.5% to 12.5%). When we included apparent temperature, calendar month and year of birth in our models, our excess risk estimates were consistent with our previous PM2.5/SGA estimates (first trimester: 5.5%, 95% CI 0.3% to 11.0%; third trimester: 3.3%, 95% CI −1.7% to 8.6%) and our NO2/VSGA estimates (first trimester: 7.5%, 95% CI 1.9% to 13.4%; second trimester: 7.3%, 95% CI 1.8% to 13.0%; third trimester: 8.0%, 95% CI 2.7% to 13.7%).
When we evaluated effect modification by maternal race, the third trimester NO2/VSGA excess risk estimate was greatest for Hispanic mothers (9.5%; 95% CI 0.5% to 19.2%) and smaller but similar for white (non-Hispanic) (5.2%, 95% CI −2.3% to 13.3%) and African–American (non-Hispanic) mothers (5.0%, 95% CI −3.9% to 14.8%). However, the third trimester PM2.5/SGA risk estimate was greatest for African–American mothers (7.9%, 95% CI 0.1% to 16.2%), smaller for white mothers (4.2%, 95% CI −1.4% to 10.1%), but there was no apparent effect in Hispanic mothers (−0.1%, 95% CI −6.4% to 6.7%).
Last, we evaluated whether the association between late pregnancy (ie, third trimester) mean PM2.5 concentration and the risk of SGA/VSGA was modified by several pregnancy complications. Among those pregnancies with at least one pregnancy complication, each 4 μg/m3 increase in third trimester mean PM2.5 concentration was associated with a 12.6% greater risk of VSGA. Among uncomplicated pregnancies, this excess risk estimate was ∼5 times smaller (1.5%; table 6). We did not observe a similar pattern when estimating the risk of SGA associated with the same incremental PM2.5 increase, and neither of these interaction terms was statistically significant. We then evaluated each pregnancy complication separately in the same manner. Although none of the complication-specific interaction terms was statistically significant, we did observe ∼two- to fivefold larger SGA/VSGA excess risk estimates in those pregnancies complicated by placental abruption compared with those without placental abruption, and premature rupture of the membrane compared with those without this condition. For the other pregnancy complications, we did not observe larger excess risks of both SGA and VSGA associated with incremental PM2.5 concentration increases for complicated pregnancies compared with uncomplicated pregnancies (table 6).
In this large, multiyear, statewide cohort study of ambient air pollution and risk of fetal growth restriction, we found significantly increased risk of SGA associated with each 4 μg/m3 increase in mean PM2.5 concentration in the first and third trimesters, and significantly increased risk of VSGA associated with each 10 ppb increase in first, second and third trimester mean NO2 concentrations, after controlling for known risk factors. These estimates were not attenuated when both PM2.5 and NO2 were included in the same model, and each pollutant effect was generally consistent with an increasing concentration–response relationship. However, there were differences in the magnitude of the third trimester risk estimates by race/ethnicity, but the pattern of effect modification was not the same for PM2.5 (highest for African–American mothers) and NO2 (highest for Hispanic mothers). Last, we found evidence of effect modification by several pregnancy complications including placental abruption and premature rupture of membranes, although these effects were not statistically significant, probably because of the rarity of these complications.
Our findings are consistent with previous studies reporting greater risk of fetal growth restriction or low birthweight associated with first trimester pollutant concentration1 5 8 9 12 16 20 and third trimester pollutant concentrations,1 7 9–13 16–18 although the specific pollutants responsible for those increased risks may be different. Associations with NO2 suggest local traffic pollution and/or residence near a source of traffic pollution during the pregnancy may be important risk factors. Future analyses will estimate risks associated with pregnancy exposures to specific PM2.5 components (ie, sulphates, elemental carbon, organic carbon, etc.) or other traffic-related pollutants (eg, specific polycyclic aromatic hydrocarbons) to explore these PM2.5 and traffic pollution findings further.
The biological mechanism(s) by which ambient air pollution affect(s) fetal growth is/are largely unknown and may differ between early- and late-onset fetal growth restriction, as well as between uncomplicated and complicated pregnancies. Mechanisms may include a defective trophoblast invasion,45 46 decreased vascular reactivity,47 decreased oxygen and nutrient delivery48 and increased trophoblast apoptosis,49 which may act independently or jointly. Mechanisms may also include the direct transfer of pollutants across the maternal blood–placenta barrier and direct binding to the fetal DNA regulating its transcription. Polycyclic aromatic hydrocarbons (PAH) have previously been associated with DNA adducts, which have been reported to adversely affect fetal growth and development,50 especially during the period of rapid fetal growth. PAH exposure during pregnancy has also been associated with increased risk of fetal growth restriction.51 52
We observed approximately two- to fivefold larger SGA/VSGA risk estimates in those pregnancies complicated by placental abruption and premature rupture of membranes compared with those without these complications. Although fetal growth restriction and placental abruption share a common mechanism of defective placental implantation early during embryogenesis,53 elevated levels of pollution late in pregnancy may exaggerate decidual necrosis, microinfarcts and atheromatous/fibrinoid changes in the placenta of pregnancies that are prone to abruption,29 accentuating their effect on fetal growth restriction. The reason(s) for the synergy between elevated air pollution and premature rupture of membranes is/are not clear. Premature rupture of membranes may serve as an indicator of chronic infection as it is associated with chorioamnionitis.54 Thus, mothers developing certain pregnancy complications, such as placental abruption and premature rupture of the membranes, may represent a parturient group particularly susceptible to the adverse health effects of elevated air pollution. However, our results need confirmation.
Although our study had several strengths, including the large number of subjects and the use of statewide, multiyear linked data from birth certificates and maternal hospital discharges, there were some limitations that should be considered. First, we had a limited number of VSGA births and pregnancy complications, and therefore less precision in these risk estimates. Second, it is likely that smoking, illicit drug use and alcohol use are underreported on birth certificates and hospital discharge data. Nonetheless, because these data are recorded during prenatal visits, it is unlikely that this misclassification is differential with respect to normal vs restricted fetal growth. However, residual confounding cannot be ruled out. Third, there is likely to be non-differential exposure misclassification, and therefore underestimation of risk, as we assigned pollutant concentrations based on residential proximity to fixed pollutant monitoring sites. Although we still found increased risks associated with PM2.5 and NO2, this non-differential misclassification may explain the lack of association with CO, a more spatially heterogeneous pollutant.
Fourth, although we assumed the maternal residence at birth was the same throughout the pregnancy, previous studies have shown that between 25% and 33% of pregnant women move during pregnancy,55 with 62% moving within the same municipality,56 and 70% moving within the same county.57 As we matched air pollution concentrations from the monitor closest to the maternal residence at birth, we may have mismatched some pollution monitors if the mother changed residences during pregnancy. Assuming this mismatching/exposure error was non-differential with respect to fetal growth category, this misclassification may have resulted in a bias towards the null and underestimation of risk. However, the magnitude of this bias may be minimal, as movement within a municipality or to a neighbouring municipality may not have resulted in a change in the air pollution monitor.
Last, only 25% of births with complete covariate data (88,678 of 350,107), had a maternal residence ⩽10 km from a PM2.5 monitoring station and were thus retained for PM2.5 analyses. As many of these monitors were located in urban areas, there were clear differences in the sociodemographic characteristics between those included (births to mothers from mostly urban areas) and excluded from analyses (births to mothers from urban, suburban and rural areas). Although this is not an issue of internal validity, these differences in subject characteristics between those included and excluded from this analysis may limit the generalisability of these findings.
Future work to examine associations between pregnancy exposure to specific PM components/sources and adverse birth outcomes, and/or to examine more powerfully the role of pregnancy complications as effect modifiers of this association or as outcomes themselves, are needed.
What is already known on this subject
Although the relationship between ambient air pollution and adverse birth outcomes is an active area of investigation, more data are needed to establish the time(s) during pregnancy when mothers are most at risk.
Also, whether the presence of pregnancy complications late in pregnancy infers greater susceptibility to the adverse effects of ambient air pollution on birth outcomes is not known.
What this study adds
Our findings suggest that ambient air pollution, perhaps specifically traffic emissions during early and late pregnancy and/or factors associated with residence near a roadway during pregnancy, may affect fetal growth.
Using more comprehensive data encompassing birth certificates and hospital discharge abstracts at the time of delivery, pregnancies complicated by placental abruption and premature rupture of the membrane had greater excess risks of SGA/VSGA than pregnancies not complicated by these conditions.
The authors would like to thank Dr Lakota Kruse and Neetu Jain of NJDHSS for their assistance in constructing the health dataset, and Drs Junfeng Zhang and Barbara Turpin for their assistance with the interpretation of air pollution analyses.
Competing interests: None.
Funding: This work was funded by a grant from the Foundation of the University of Medicine and Dentistry of New Jersey (UMDNJ) and the NIEHS-sponsored UMDNJ Center for Environmental Exposures and Disease (CEED), grant no. NIEHS P30ES005022.
Ethics approval: The study was approved by both UMDNJ and NJ Department of Health and Senior Services (NJDHSS) Institutional Review Boards.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.