In this study we investigated the effect of ambient air pollution during early pregnancy on fetal ultrasonic measures during mid-pregnancy (13–26 weeks gestation). To our knowledge, this is the first study of its kind because it uses ultrasound measurements as direct estimates of growth, rather than using birth weight as an indirect (and delayed) measure of growth.
When analyzing scans from women at different distances to the monitoring sites, we found that if there was a negative relationship between a pollutant and ultrasound measurement, the effect often decreased toward the null when including scans from women who lived further away from the monitoring sites. Because of spatial variation in air pollution, we would expect the measurement error of the pollutant exposure to increase with increasing distance from a monitor. This increase in error causes a decrease in true effects and is known as regression dilution (MacMahon et al. 1990
This finding highlights the importance of accurate measures of exposure for future studies. Given our results, we recommend using only monitors within a 2-km radius of the subject. We also strongly suspect that our exposure estimates would have been even more accurate (and hence shown larger pollutant effects) if we had each woman’s actual address rather than her postcode. The strong change in the association between air pollution and ultrasound measurements by distance to monitor shown here may also explain the inconsistent results from previous research, which used a range of exposure estimates and distances (e.g., network average vs. close proximity to a monitoring site). Previous air pollution–birth outcome research in Brisbane failed to find reductions in neonate birth weight, head circumference, and length associated with increased ambient air pollution during pregnancy, and this failure may be caused by the maternal exposure estimates being derived from a network average across five monitoring sites (Hansen et al. 2007
). When using exposure estimates derived from an average across a network of air pollution monitoring sites, the pollutant effect will most likely be underestimated (Zeger et al. 2000
, and PM10
during early pregnancy were associated with reductions in fetal biometry during mid-pregnancy, with the two conspicuous pollutants being PM10
. However, in relation to the timing of the exposure and the different fetal body segments, these results were irregular, and until further information is known about the biological mechanisms, it is difficult to associate specific air pollutants with reduced growth of specific fetal body segments. If anything, PM10
during days 91–120 showed a consistent pattern across all ultrasound measurements except BPD. This may indicate a sensitive period of gestation in relation to PM10
exposure, but it may also be attributed to the strong correlation in ultrasound measurements. Despite the inconsistency in the exposure period, all pollutants except NO2
were associated with reductions in abdominal circumference. Based on the premise that the fetus accrues most body fat during the second half of pregnancy, the negative effects on abdominal circumference early in pregnancy suggest that ambient air pollution may interfere with the development of internal organs (e.g., the liver), as abdominal circumference is a proxy measure of the size of these organs (Ville and Nyberg 2003
Although several biological mechanisms have been suggested, the underlying mechanisms whereby ambient air pollution interferes with fetal growth remain to be determined. It is well recognized that inhalation of air pollutants can cause inflammatory responses and oxidative stress (Donaldson et al. 2001
; Kelly 2003
; Sorensen et al. 2003
), and both of these reactions can interfere with normal intra-uterine growth via vascular dysfunction in the placenta and damaged DNA (Myatt et al. 2000
). Also, pro-inflammatory cytokines can limit trophoblast invasion during the early stages of pregnancy, restricting fetal growth (Gitto et al. 2002
; Silver et al. 2004
). Poor placental vascularity is caused partly by dysregulation of gene expression in key angiogenic factors in early pregnancy (Torry et al. 2004
), and if ambient air pollution is associated with poor placental function it may partly be caused by perturbed DNA transcription early in pregnancy (Perera et al. 1999
In , the unadjusted associations were always smaller than the adjusted associations. This is most likely explained by seasonal factors. The pollutants that showed an effect had a seasonal pattern that peaked in late spring and summer, whereas the outcomes had weaker seasonal patterns that peaked in spring or winter. This difference in phases would dampen the association between pollutant exposure and fetal biometry. We adjusted for seasonal factors because we believe that the seasonal patterns in pollutants and outcome are exogenous to the year-round association between pollutant and fetal biometry.
We examined the effect of multiple pollutants but were somewhat hampered in this comparison because of the lack of overlap in the pollutant data for all the scans. Hence, when we added a secondary pollutant the sample size decreased (). Because of this drop in sample size, the statistical power also decreased. Importantly, however, the coefficients for the changes in fetal biometry associated with the primary pollutant remained approximately similar for most of the associations. Hence, the associations found with the single pollutants are relatively robust to other pollutants, and we can be fairly confident that the named pollutants are those attributing to the reductions in fetal biometry. However, air pollution is a complex mixture of compounds, and some of the reported effects may be attributable to other unmeasured pollutants that are correlated with the pollutants analyzed.
Unlike other studies that suggest that seasonal patterns in birth weight are related to ambient temperature (Elter et al. 2004
; Lawlor et al. 2005
; Murray et al. 2000
) and sunlight (Tustin et al. 2004
), we failed to find an association between fetal biometry and temperature and sunlight during pregnancy. The lack of effect for sunlight could be attributable to the lack of accurate exposure data, as only the city airport measured hours of sunlight. Also, sunlight exposure depends more on individual behavior than either temperature or air pollution. Whereas a person needs to go outdoors to be exposed to sunlight, this is less true for air pollution, as the air also pervades into homes. This pervasiveness is particularly true in subtropical Brisbane; there is a generally greater need to lose (rather than trap) heat, and many houses are designed to maximize air flow.
This study has a number of important limitations. First, when using residential proximity to a monitoring site as a proxy of exposure, the mother’s postcode at the time of exposure is unknown; it was recorded only at the date of the ultrasound scan. Of the 2,522 pregnancies that had more than one scan, only 176 (7%) had changed postcode from the time of the first scan. This was not a concern because the change of address occurred after the exposure period we investigated. However, for all scans there is the assumption that the postcode at the first scan is the post-code during the first 4 months of pregnancy. This is important because studies investigating maternal residential mobility during pregnancy in relation to birth defects have reported varying rates (12–32%) of women changing address during pregnancy (Canfield et al. 2006
; Fell et al. 2004
; Khoury et al. 1988
; Shaw and Malcoe 1992
A second limitation is that we had limited data on maternal lifestyle factors such as diet, smoking, and alcohol consumption. When the exposure assessment is based purely on temporal variations in air pollution derived from a network average across a number of sites, these factors do not confound the results because they are constant over time and not associated with the temporal variations in air pollution. However, to improve exposure accuracy we used air pollution data from the closest monitoring site to the women’s post-code area, and this may create residual confounding associated with spatial variation in maternal characteristics that we were unable to control for. We attempted to control for these unknown maternal factors via an index of SES linked to the women’s postcode. Surprisingly, in our preliminary analyses area-level social class had a statistically significant effect only on FL. This may be an indicator that area-level social class is a poor proxy for individual-level social class, and stronger differences would be expected if an individual measure was used. When we included an interaction between exposure and SES quartile, the effect modification of social class on the association between pollutant and fetal biometry was only minor for most associations. Interestingly, the pollutant effects were still significant in the highest social class group for all but two of the results. No pollutant effects were significant in the lower social class group. These effects are somewhat associated with sample size: The highest SES group had the largest sample size, and the lowest SES group the smallest.
A study of the effects of particulate matter on death showed a reduction in risk when adjusting for area-level deprivation (Næss et al. 2007
). It is possible that the results shown here would be ameliorated if we had individual SES or a more accurate measure of area-level SES. This would occur if exposure and SES were correlated and lower-SES groups lived in heavily built-up areas with more traffic. This disparity in air pollution exposure has been previously reported among pregnant women (Woodruff et al. 2003
). Air pollution monitors are often placed in the areas of highest pollution. These areas might be expected to have the lowest SES, although clearly this is not the case in Brisbane (). However, there is still the potential of residual confounding as a result of within-area (postcode) variation in both exposure and maternal characteristics that could not be assessed.
Another limitation of this study is that we were unable to obtain subsequent birth outcome data from the state health department for the pregnancies within our study. Therefore, it is difficult to conclude whether these reductions associated with air pollution during early pregnancy persisted until birth and whether there is any clinical relevance to these reductions.
Given the large number of comparisons performed, some of our findings may have occurred by chance (type 1 errors).
Despite these limitations, our study adds to the growing body of literature suggesting that ambient air pollution during pregnancy influences fetal growth. We strongly recommend further exploration of fetal ultrasonic measures in air pollution–birth outcome research to corroborate our findings and examine some of the confounders (e.g., SES, parity) and possible effect modifiers (e.g., air conditioning) that we were unable to examine here. The results shown here suggest that the pollutant monitor needs to be within at least 2 km of the subject, and optimally each woman’s address would be geocoded. We also recommend collecting individual-level data such as distance to major road and time spent outside, as well as accurate data on temperature and sunlight exposure.
Although future work including more individually detailed data is needed to confirm our results, we recommend that pregnant women (where possible) reduce their exposure to air pollution.