In our analysis of full-term Nova Scotia infants, increased risks of LBW were observed for mothers in the highest exposure quartile for PM10 and SO2 during the first trimester, unadjusted for birth year. There was a 33% and 36% increase in risk of LBW associated with maternal exposures during the first trimester to the > 75th percentile for PM10 and SO2, respectively. Adjusting for birth year attenuated the risks, especially for PM10. In models based on continuous exposure variables, SO2 exposures during the first trimester exhibited a significant association with LBW. Consistent with other studies, we did not observe an increased risk associated with maternal exposure to O3 [3, 13]. Though adjustment for birth year did attenuate the associations, a dose response relationship for the first trimester was still evident with SO2. The plot representing the natural spline function of SO2 and birth weight was consistent with a linear concentration response between SO2 and birth weight, over most of the range of exposure. This is in keeping with the apparent gradient of effect observed in the logistic model. The lack of statistical significance at the lower exposure quartiles may be a result of low power for observing relatively small increases in risk.
If a causal relationship between air pollutants and birth outcomes is confirmed, a research priority will be to identify the critical time points during pregnancy when exposure to air pollutants might be most harmful [17]. Our finding of an effect with first trimester exposure, but not exposures during the second or third trimesters, is consistent with a number of other investigations. Previous results related to both intrauterine growth retardation (IUGR) and low birth weight suggest that early pregnancy exposures may be the critical time point [1, 3, 8–10, 12, 13]. Among the studies that observed a relationship between early pregnancy exposure to air pollutants and IUGR, our exposure levels were not consistently lower or higher than those reported in other studies. For instance, the high exposure group for PM10 was ≥ 50 ug/m3 in a study conducted in the Czech Republic [9], whereas the top quartile of PM10 in our study was ≥ 19 ug/m3. In Vancouver, British Columbia, the top quartile exposure level for SO2 was ≥ 6.3 ppb [13], which was a lower level than our top quartile. Other studies that have reported an effect of air pollutants on birth weight or growth have not observed an effect limited to the first trimester [4, 6, 7, 11]. Biologic mechanisms suggested to support the hypothesis of an effect associated with early pregnancy exposures relate to the etiology of IUGR. Although likely multifactorial, one suggested mechanism for IUGR is abnormal placental development in early pregnancy [18]. Hematologic effects of air pollutants might occur from an initial inflammatory response resulting in increased blood coagulation, and subsequent decreased oxygen supply to the placenta [3, 19]. Another hypothesis suggests that the polycyclic aromatic hydrocarbons (PAH) component of PM contributes to impaired growth [20]. Newborns with elevated PAH-DNA adducts (which are used as a proxy to measure individual biologically effective dose to PAH) were found to have significantly reduced birth weight and head circumference suggesting that transplacental exposures to PAHs in ambient air may negatively impact on fetal development [21]. More work is required to fully elucidate the physiologic mechanisms by which air pollution may affect fetal growth and development and to determine if the mechanisms are pollutant specific.
Several limitations of this study are noted. With the existing database, it was not possible to identify multiple births during the observational period for a particular mother. Therefore, clustering or co-linearity could not be accounted for. In addition, there was some concern that classifying exposure based on monitoring sites up to 25 km from the mother's residence could have resulted in significant exposure misclassification. Epidemiology research on the health effects of ambient air pollution is primarily dependent on the proximity of populations to the location of monitoring stations. For the gaseous pollutant, SO2, this may be a more relevant issue, as SO2concentrations are known to decrease substantially as distance from source increases. Monitor site-pair correlations for SO2 across Canada, however, indicate a strong correlation (0.8) for SO2 within a distance of 25 km. (Environment Canada). As well, 70% of mothers in our study resided within 15 km of a monitoring station. A recent study in California found that, although neighborhood and county level exposure metrics for PM2.5 were highly correlated, the two metrics produced different estimates for the association with birth weight [22]. It is important to note that studies with exposure assignment based on geographical location may result in non-differential classification of the exposure, which can underestimate the magnitude of the association. Thus, inferences across studies that have used varying exposure metrics make comparisons difficult. A recent review of published studies on prenatal and early childhood effects concluded that future investigations should focus on using a more definitive means of characterizing exposures such as personal exposure monitoring to adequately evaluate the impact of each pollutant during different periods of pregnancy [23].
We found that year of birth appeared to be a confounder of the relationship between air pollution and low birth weight, reducing both the magnitude and statistical significance of the association. Our dataset extended over a relatively long period (13 years) during which significant time trends could be detected. Other investigators should evaluate the impact of adjusting for year of birth where extended time periods are under consideration.
As with previous investigations on this issue, there was an inability to adjust for some potentially important confounders such as occupational exposures, exposures to environmental tobacco smoke, maternal drug and alcohol use, or meteorological factors. Similarly, we were unable to account for indoor air exposures such as wood combustion or other housing characteristics. A particular strength of this analysis, however, was the availability of individual-level information pertaining to smoking during pregnancy. However, since the smoking information reflected smoking at the time of delivery, the smoking status for some women may have been underestimated for first (or second) trimester exposures (e.g., a woman who smoked in early pregnancy but quit during pregnancy would be considered a non-smoker).
In this study, it was assumed that maternal residence at the time of delivery was the same residence throughout the pregnancy. Our estimates of trimester specific exposures may be misclassified, particularly for first and second trimester exposures, since the pollutant levels for each trimester were based on residence at delivery. A recent study that evaluated change in residence among pregnant women in Nova Scotia and Eastern Ontario, Canada found that 12% of the women moved during pregnancy and among the women who had changed residence, most moved within the same municipality [24]. Given the small percentage of women who might have moved outside of the municipality between the first trimester and delivery minimizes the likelihood of exposure misclassification due to mobility.
The relationship between exposures to ambient levels of air pollution and birth weight appears to be complex given the somewhat conflicting results that have been observed in terms of pollutant and trimester effects. Whether this represents an inability to accurately characterize exposure for epidemiologic investigations, threshold effects, observed effects being markers of other air pollutants or a result of varying methodologies is not clear. The study region contains significant sources of SO2 from the coal based electricity sector and combustion of heavy fuel oils. The disparity in findings may also be partially explained by a combined multi-pollutant effect that is not being captured in the present study designs. In our study, data on other pollutants were not sufficiently available to include in the analysis.