Air Pollution Exposure During Pregnancy, Ultrasound Measures of Fetal Growth, and Adverse Birth Outcomes: A Prospective Cohort Study

Background: Air pollution exposure during pregnancy might have trimester-specific effects on fetal growth. Objective: We prospectively evaluated the associations of maternal air pollution exposure with fetal growth characteristics and adverse birth outcomes in 7,772 subjects in the Netherlands. Methods: Particulate matter with an aerodynamic diameter < 10 μm (PM10) and nitrogen dioxide (NO2) levels were estimated using dispersion modeling at the home address. Fetal head circumference, length, and weight were estimated in each trimester by ultrasound. Information on birth outcomes was obtained from medical records. Results: In cross-sectional analyses, NO2 levels were inversely associated with fetal femur length in the second and third trimester, and PM10 and NO2 levels both were associated with smaller fetal head circumference in the third trimester [–0.18 mm, 95% confidence interval (CI): –0.24, –0.12 mm; and –0.12 mm, 95% CI: –0.17, –0.06 mm per 1-μg/m3 increase in PM10 and NO2, respectively]. Average PM10 and NO2 levels during pregnancy were not associated with head circumference and length at birth or neonatally, but were inversely associated with birth weight (–3.6 g, 95% CI: –6.7, –0.4 g; and –3.4 g, 95% CI: –6.2, –0.6 g, respectively). Longitudinal analyses showed similar patterns for head circumference and weight, but no associations with length. The third and fourth quartiles of PM10 exposure were associated with preterm birth [odds ratio (OR) = 1.40, 95% CI: 1.03, 1.89; and OR = 1.32; 95% CI: 0.96, 1.79, relative to the first quartile]. The third quartile of PM10 exposure, but not the fourth, was associated with small size for gestational age at birth (SGA) (OR = 1.38; 95% CI: 1.00, 1.90). No consistent associations were observed for NO2 levels and adverse birth outcomes. Conclusions: Results suggest that maternal air pollution exposure is inversely associated with fetal growth during the second and third trimester and with weight at birth. PM10 exposure was positively associated with preterm birth and SGA.


Table of Contents
Excluded due to twin pregnancy, abortion, or intrauterine death N = 98 No valid air pollution data N = 1010 Subjects with valid air pollution exposure estimates (> 80% of the days available) for at least one pregnancy period N = 7870 Supplemental Material, File S1. Air pollution exposure assessment Hourly concentrations of particulate matter (PM 10 ) and nitrogen dioxide (NO 2 ) at all addresses in the study area (Northern part of Rotterdam) were estimated for the years 2001-2006, using advanced spatiotemporal dispersion modelling techniques in combination with hourly air pollution measurements at three continuous monitoring sites located around the study area. A flow chart of the exposure assessment approach is presented in Figure S2. We took the following steps to assign exposure estimates to the addresses in this area.

Spatial pattern
First, approximately 800 000 digital calculation points, further referred to as 'receptors', were assigned to the façades of all dwellings. Annual average concentrations of PM 10 and NO 2 were assessed for every receptor using the three Dutch national standard methods for the calculation of air quality, and Geographic Information Systems (GIS). These standard methods have been established by the Dutch government, and are designated to calculate the contribution of intra-urban road traffic, traffic on highways, and industrial and other point sources (standard calculation method 1, 2 and 3, respectively) (VROM (Netherlands Ministry of Housing Spatial Planning and the Environment) 2007). For each year, the spatial distribution of annual average concentrations was assessed for eight different wind conditions (wind direction: north/east/south/west; wind velocity: light/strong). For each hour, we then derived the corresponding spatial distribution for the prevailing wind direction and wind speed at that specific hour, by means of interpolation between the eight characteristic spatial distributions.
Input data for the calculations described above were traffic characteristics (including traffic intensities, traffic composition, and traffic speed), road characteristics, vehicle emission factors, buildings and ground characteristics, and emission data from shipping, industry, and households.
Detailed digital maps with information on geographic locations and traffic characteristics for roads in the study area were obtained from the local authorities of Rotterdam. Traffic intensities and meteorological data were supplied by the DCMR Environmental Protection Agency Rijnmond (DCMR).
Emission sources and emission data were obtained from the National Institute for Public Health and the Environment (RIVM) and the DCMR.
Subsequently, the spatial distributions that corresponded to the hourly wind conditions were adjusted for fixed temporal patterns of source activities, by applying fixed scaling factors to the contributions of various air pollution sources. In this way, we accounted for temporal fluctuations in the contribution of air pollution sources during the day (e.g., morning and evening rush hour), week (e.g., working days and weekend days), and month. For example, the contribution of traffic was scaled with the hourly traffic intensity pattern. The fixed temporal patterns were derived on the basis of traffic counts (reflecting traffic fluctuation patterns) and energy usage data (reflecting residential heating patterns).

Temporal pattern and calibration
The modelled concentrations were adjusted based on hourly continuous monitoring data at three stations in the area. This served two main purposes. First, the temporal fluctuations in background concentrations were taken into account. Second, the modelled concentrations were calibrated against measured concentrations.
The hourly calibration procedure was performed in the following way. Concentrations were modelled for each hour at the locations of three monitoring stations and compared with the actual measurements. Subsequently, the differences between modelled and measured concentrations at the three monitoring stations were averaged into a representative difference for the whole area. This difference was added to or subtracted from the modelled concentrations. Monitoring Network (LML), emission data, and modelling. The developed nationwide concentration maps are updated annually and provide a best estimate of large-scale air quality currently available (Velders et al. 2010).

Modelling performance
Two recent studies examined the performance of a dispersion modelling approach that incorporates the It can be expected that the hourly PM 10 and NO 2 concentrations estimated for the present study will be more precise than annual average concentrations, since the substantial temporal fluctuations in air pollution concentrations were taken into account by the hourly calibration procedure with hourly monitoring data, which further enhances the agreement between predicted and measured concentrations.

Exposure assignment
For each dwelling, the receptor at the most exposed façade was selected, and the corresponding air pollution values were assigned to the address. We obtained full residential history of the participants by combining the address data collected by questionnaires with data from the local authorities of Rotterdam. It was ascertained that the residential history covered the total pregnancy period. Of the women in our study, 87% did not move during pregnancy, 12% changed residence once, 1% moved twice, and 0.1% moved three times during pregnancy.
We calculated exposure estimates for the participants using the following approach. Derived from the hourly concentrations of PM 10 and NO 2 , we constructed a database containing daily averages (24h) for every address, for the years 2001-2006. Allowing for residential mobility, air pollution exposure estimates were linked to the different home addresses of the participants during pregnancy. We derived average exposure estimates for different periods in pregnancy: 1) conception until first trimester ultrasound (median 13.2 weeks of gestation, 95% range 10.5 to 17.5); 2) conception until second trimester ultrasound (median 20.5 weeks of gestation, 95% range 18.6 to 23.4); 3) conception until early third trimester ultrasound (median 30.3 weeks of gestation, 95% range 28.3 to 33.0); and 4) conception until delivery (median 40.1 weeks of gestation, 95% range 35.5 to 42.4). Average exposures were only calculated for periods with <20% of the daily averages missing. For the other periods, air pollution exposures were set to missing. There was substantial spatial and temporal variation in pregnancyspecific exposure levels. Distributions of PM 10 and NO 2 exposure levels in our study population are presented in Table 2.

Supplemental Material, File S2. Traffic-related noise exposure assessment
We adjusted for noise exposure when examining the associations for air pollution exposure with fetal growth and neonatal complications. Recent studies expressed the need to include information on noise exposure in studies on traffic-related air pollution exposure and health, since traffic is a major shared source for both air pollution and noise (Allen et al. 2009;Davies et al. 2009;de Kluizenaar et al. 2007;de Kluizenaar et al. 2009). Noise is hypothesized to induce stress responses, which may result in altered function of the sympathetic autonomic nervous system, endocrine system, and immune system (Babisch 2000;Passchier-Vermeer and Passchier 2000), possibly contributing to an increased risk for an unfavorable pregnancy outcome (Magann et al. 2005;Nurminen 1995;Passchier-Vermeer and Passchier 2000).
Road traffic noise exposure was assessed in accordance with requirements of the European Environmental Noise Directive. The method has been described previously in more detail (de Kluizenaar et al. 2007;de Kluizenaar et al. 2009). Input data for the noise calculations was a detailed digital map describing the geographic location of roads and the traffic characteristics for each road segment, provided by the local authorities of Rotterdam for the current situation at time of the study (base year 2004). This data can be reasonably applied to adjacent years, as the road network is assumed to be rather stable, with only small (if any) but equal changes in noise exposure across the population.
Noise exposure levels are expressed in the EU standard noise metric L den (day, evening, night), a measure of annual average sound levels. We assessed the road traffic noise (L den ) level at the most