Skip to main content
Download PDF
Download ePub

The reduction of birth weight by fine particulate matter and its modification by maternal and neighbourhood-level factors: a multilevel analysis in British Columbia, Canada

Environmental Health volume 15, Article number: 51 (2016) Cite this article

Abstract

Background

The purpose of this research was to determine the relationship between modeled particulate matter (PM2.5) exposure and birth weight, including the potential modification by maternal risk factors and indicators of socioeconomic status (SES).

Methods

Birth records from 2001 to 2006 (N = 231,929) were linked to modeled PM2.5 data from a national land-use regression model along with neighbourhood-level SES and socio-demographic data using 6-digit residential postal codes. Multilevel random coefficient models were used to estimate the effects of PM2.5, SES and other individual and neighbourhood-level covariates on continuous birth weight and test interactions. Gestational age was modeled with a random slope to assess potential neighbourhood-level differences of its effect on birth weight and whether any between-neighbourhood variability can be explained by cross-level interactions.

Results

Models adjusted for individual and neighbourhood-level covariates showed a significant non-linear negative association between PM2.5 and birth weight explaining 8.5 % of the between-neighbourhood differences in mean birth weight. A significant interaction between SES and PM2.5 was observed, revealing a more pronounced negative effect of PM2.5 on birth weight in lower SES neighbourhoods. Further positive and negative modification of the PM2.5 effect was observed with maternal smoking, maternal age, gestational diabetes, and suspected maternal drug or alcohol use. The random intercept variance indicating between-neighbourhood birth weight differences was reduced by 75 % in the final model, while the random slope variance for between-neighbourhood gestational age effects remained virtually unchanged.

Conclusion

We provide evidence that neighbourhood-level SES variables and PM2.5 have both independent and interacting associations with birth weight, and together account for 49 % of the between-neighbourhood differences in birth weight. Evidence of effect modification of PM2.5 on birth weight across various maternal and neighbourhood-level factors suggests that certain sub-populations may be more or less vulnerable to relatively low doses PM2.5 exposure.

Peer Review reports

Background

Studies of exposure to particulate air pollution have consistently shown an association with low birth weight, a predictor of fetal growth restriction and important determinant of infant and child-wellbeing [1–3]. The fine fraction of particulate matter (PM2.5 - less than 2.5 μm) is a complex mixture of elemental and organic carbon compounds, metals and gases that stem predominantly from vehicle exhaust, residential heating and industrial emissions. PM2.5, which includes ultrafine particles less than 0.1 μm, can penetrate deep into the pulmonary alveolar tissue where inflammatory mediators and possibly the particles themselves translocate into the bloodstream causing systemic cardiovascular and immunological alterations such as platelet activation, coagulation and endothelial dysfunction [4–6]. These physiological changes extend to the placenta, a highly vascularized organ and extension of the maternal cardiovascular system with similarly affected endothelial cellular tissues particularly susceptible to oxidative and inflammatory injury [7–9]. Excess or uncontrolled oxidative stress and inflammation early in pregnancy may disrupt placental cell growth and differentiation, potentially leading to deficient deep placentation and morphological adaptations associated with several adverse pregnancy outcomes including fetal growth restriction [10].

These mechanisms by which PM2.5 may act to adversely impact the reproductive system are not fully understood; however, evidence supports the potential for a shared mode of developmental toxicity with several other known risk factors [5, 11]. This includes factors that also promote or are associated with oxidative stress and inflammation such as smoking [12], drug use [13], advanced maternal age [14] gestational diabetes [15], and low socioeconomic status (SES) in general [16]. The causal pathways in which SES contributes to adverse pregnancy outcomes can be conceptualized in terms of ‘downstream’ or mediating exposures, stresses and behaviours acting on the individual through ‘upstream’ society-level determinants such as poverty, poor education, income inequality and social discrimination/marginalization over the lifespan [17, 18]. This impact of the social environment on health behaviours and outcomes creates hierarchical structures within which individuals are nested in neighbourhoods and communities with their own set of attributes that can promote or antagonize health and healthy behaviours [19, 20].

A particular challenge in environmental epidemiology is handling data at differing geographic scales. Birth registries and vital statistics provide data on individual births and certain risk factors, but may not have data on socially patterned risk factors. Alternatively, reliable SES data such as education, income and housing quality are often only available from national census databases using arbitrary administrative spatial units. Finally, obtaining individual-level environmental exposure data is often not possible. Therefore, the epidemiologist is often left with a mix of individual-level observations clustered within neighbourhood areas each with distinct attributes. The use of multilevel statistical models separates the individual-level effects from the context of their social and physical environments and can therefore quantify the degree of clustering of individuals within neighbourhood areas and test whether neighbourhood factors themselves have direct effects on the health outcome or act indirectly via the modification of individual-level variables [21, 22].

Through the mechanisms of oxidative stress and inflammation there is evidence that SES may not only confound but modify the PM2.5-birth outcome relationship [23–25]. Various exposures and experiences may act in a non-additive manner to influence fetal development [5]. We present a multilevel cross-sectional analysis of the association between birth weight and PM2.5 in British Columbia, Canada where levels of PM2.5 are relatively low but can vary substantially between different communities [26]. We explore the potential for between-neighbourhood variability for the slope of gestational age on birth weight and whether interactions with PM2.5, neighbourhood-level SES indicators, and/or individual-level risk factors are able to explain any neighbourhood-level variability. We had three research questions: 1) does exposure to PM2.5 and residence in low SES neighbourhoods in BC have significant independent negative associations with birth weight? 2) does the effect of gestational age on birth weight differ between neighbourhoods? 3) does PM2.5 interact with neighbourhood-level SES and/or individual-level risk factors to modify their independent effects on birth weight to help explain any neighbourhood-level differences?

Methods

This was a population-based retrospective cohort of singleton births in British Columbia from 2001 to 2006 (N = 237,470). Data from the BC Perinatal Data Registry were provided by Perinatal Services British Columbia (PSBC) which included information on maternal-infant health status and outcomes, reproductive history, maternal risk factors and attributes, and residential postal codes. The Registry accounts for 99 % of births and stillbirths in BC of at least 20 weeks gestation or at least 500 g birth weight. Research data access is provided by a Partnership Accord /Memorandum of Agreement between all BC Health Authorities and PSBC through the Freedom of Information and Privacy Protection Act [27]. Research ethics board approval was granted by the University of Victoria (ethics protocol #: 11-043).

The outcome variable was continuous birth weight of singleton births. In order to avoid potential selection bias, we included all births (stillbirth and live) for all gestational ages (20–44 weeks). Excluded records included out-of-province and invalid postal codes (n = 1096), non-viable births prior to 20 weeks gestation or < 500 g (n = 14), and the list-wise deletion of births missing important data including: cigarettes smoked per day (cigs/day, n = 2501), PM2.5 (n = 1510), gestational age (n = 373), and birth weight (n = 41). All continuous variables, except cigarettes/day, were standardized and centred to ease interpretation and aid model convergence.

The spatial location of each birth record was geocoded based on the latitude-longitude coordinate of the mother’s residential postal code at the time of delivery using GeoRef [28]. Birth records were related to their corresponding census dissemination area (DA) by performing a point-in-polygon spatial join procedure in ArcGIS 10.2 [29]. DAs are the smallest geographical unit for which census data are available and represent neighbourhood blocks ranging between 200 and 800 people. While DAs do not necessarily represent existing neighbourhood communities [30], they can act as proxies for a general catchment area of personal home-life activities [31, 32]. Birth records were identified as being either rural or urban using the Statistics Canada Metropolitan Influence Zone (MIZ) codes which are based on commuting flows of small towns into larger cities and metropolitan areas [33].

Exposure to PM2.5 was estimated using a national land-use regression (LUR) model developed to estimate PM2.5 at the census street block-face level [34]. The model used a number of predictors including satellite measures, proximity to major roads and industry to account for 46 % of the variability in measured annual PM2.5 concentrations. Unlike nitrogen dioxide (NO2), PM2.5 tends to have a more homogeneous intra-urban distribution between personal, indoor and ambient exposure [35]. The LUR model estimates used for this study showed very little variability of PM2.5 exposures between individuals within a given DA. We therefore aggregated the point-level estimates of PM2.5 to their DA-level mean and related it to individual birth records as an area-level variable.

The DA-level SES and demographic data were represented by three related but independent datasets all based on the 2006 Statistics Canada national census. The first was a Canadian SES index (SESi) developed by Chan et al. [36]. The second was an education variable representing the proportion of population over 15 with any post-secondary education, including college, trades, or university. The third was the proportion of continental Asian immigrants by DA as it has been shown in BC and elsewhere that healthy babies from Asian and South Asian backgrounds are constitutionally smaller compared to their Caucasian counterparts [37, 38]. Asian and South Asian ethnicities are well-represented throughout BC but particularly in concentrated pockets throughout the major urban center of Metro Vancouver where levels of PM2.5 are also high. The correlation between immigrant density with SESi and PM2.5 was −0.62 and 0.53 respectively (p < 0.001); we therefore created a residual immigrant density variable using a sequential regression technique [39]. Here, immigrant density was regressed against SESi and PM2.5 with the saved residuals representing the uncorrelated and independent contribution of immigrant density on birth weight freed from its collinearity with SESi and PM2.5. This same method was used between SESi and education (r = 0.25) creating a residual education variable. The education and immigrant data were obtained by access to ABACUS via the Data Liberation Initiative [40].

In order to avoid data loss from rural DAs, imputation for missing SES, education and immigrant density values was performed. Taking advantage of the nested hierarchical structure of the administrative census and health boundaries, the mean value for a larger encompassing census subdivision (CSD) or local health area (LHA) was imputed for a nested DA with a missing value. There were 1441 values imputed in 52 DAs for SESi (0.6 % of final N, 0.8 % of DAs), and 3170 values imputed in 108 DAs for both education and immigrant density (1.4 % of final N, 1.7 % of DAs). Sensitivity analyses were performed using only the non-missing data.

Hierarchical (multilevel) linear regression models were used to test our research questions, thereby accounting for the clustering, or non-independence, of individuals (level-1) belonging to a given DA neighbourhood (level-2). The multilevel model allows the intercept and slope to act as random parameters having between-area (DA) variability from an overall (BC-wide) mean intercept and slope. That is, each DA has its own intercept and slope in which their variability from the overall intercept and slope can be investigated with the addition of level-1 and level-2 variables and their interactions [41]. We followed a bottom-up approach to model building to quantify the explained proportional change in variance (PCV), the multilevel model equivalent to an R 2 [22]. We started with an empty (null) random intercept model without any independent variables in which birth weight is only a function of the mother’s residential DA. The presence of significant random intercept variance indicates there are unexplained between-neighbourhood differences in mean birth weight. The proportion of the total variance in birth weight that arises due to neighbourhood differences can be quantified by computing the intra-class correlation (ICC), and hence provides the degree of clustering of individual birth weight within neighbourhoods [22].

Gestational age was added to the null model and given a random slope (i.e., the mean within-DA effect of gestational age on birth weight was allowed to differ between DAs). The presence of a significant random slope indicates that its effect is not constant (or equal) for all DAs. Subsequent models included the individual and DA-level variables along with cross-level and within-level interactions in order to assess their fixed effects on birth weight but to also determine if their inclusion addressed any unexplained intercept or slope variance. Models were tested using the Akaike Information Criterion (AIC) to evaluate model performance. All statistical analyses were conducted in Stata 13IC [42].

Finally, while multilevel models address intra-area dependence while quantifying inter-area variance, they assume spatial independence among neighbouring areas. However environmental and social processes can extend beyond arbitrary neighbourhood boundaries. Additionally as mentioned above, census DAs do not necessarily represent neighbourhood dynamics, services, infrastructure, etc.; and evidence of spatial clustering between DAs may indicate that an alternative neighbourhood areal unit should be considered. We used spatial methods to test for this by checking the level-1 model residuals and level-2 predicted random parameters (intercepts and slopes) for spatial autocorrelation using the local Moran’s I statistic [43]. The presence of significant residual spatial autocorrelation indicates the existence of unobserved spatial processes causing DAs to cluster and can be a sign of model misspecification. Prediction of the DA-level random intercept and slope errors used an Empirical Bayes method [44] available as a post-estimation command in Stata 13IC.

Results

After exclusions there were 231,929 singleton (live and stillborn) births located in 6338 neighbourhood DAs (min. = 1, max. = 781, avg. = 37). Table 1 summarizes the untransformed individual and neighbourhood covariates (non-centered, non-standardized) along with their relationship to PM2.5. Table 2 reports the adjusted coefficients for the individual and DA-level covariate fixed effects on continuous birth weight (Model 1 to 3). Gestational age was modeled using a quadratic term to account for the rapid fetal growth in mid-gestations and its slower growth post-term (>36 weeks). Maternal smoking (cigarettes/day) and PM2.5 were also modeled using a quadratic terms, both indicating a subdued dose–response with increasing exposure.

Table 1 Descriptive statisticsa for individual (Level-1) and DA (Level-2) covariates
Full size table
Table 2 Adjusted individual and DA-level fixed effects on continuous birth weight
Full size table

Model 2 added the DA-level variables of SESi, education, immigrant density and rural residence. Their fixed effects show that lower SES and higher Asian immigrant density were significantly associated with lower birth weights (Table 2). Rural DAs and DAs with higher proportion of post-secondary education were not significantly associated with birth weight in this model. However, both became significant after the addition of PM2.5 and season of birth (cold vs. warm) in Model 3. Higher education had a positive association with birth weight, while rural areas had a significant negative association with birth weight. PM2.5 was found to have a significant non-linear negative association on birth weight whereby the negative effect tapers off at higher concentrations of PM2.5 (Fig. 1). Being born in a cold (fall or winter) month also had a significant negative association with birth weight (Table 2).

Fig. 1
figure 1

Adjusted predicted effects of PM2.5 on birth weight. Predicted effects of PM2.5 on birth weight with 95 % confidence intervals are conditional on model covariates included in Model 4. Black vertical lines represent the frequency distribution of PM2.5

Full size image

Model 4 tested interactions with PM2.5 including cross-level (level-1 by level2) and level-2 by level-2 interactions to explain the between-DA random intercept variability. The model results are presented in Tables 3 and 4 including the main effects as well as the interaction effects with PM2.5. Four maternal variables showed effect modification with PM2.5 on birth weight. Maternal smoking and suspected drug or alcohol use both had positive interactions with PM2.5 on birth weight revealing a subdued association with increased PM2.5 exposure (Fig. 2a and b respectively). Maternal age was also modified by differences in PM2.5 exposure with younger maternal ages showing a larger reduction in birth weight with increased PM2.5 exposure (Fig. 2c). Alternatively, gestational diabetes was associated with a much greater reduction in birth weight with increasing PM2.5 compared to normal births, essentially nullifying the higher birth weights produced by the condition (Fig. 2d). Three DA-level variables showed significant effect modification with PM2.5 on birth weight. First, the interaction between SESi and PM2.5 revealed a more pronounced effect of PM2.5 in lower SES neighbourhoods (Fig. 3a). Higher Asian immigrant density buffered the PM2.5 effect (Fig. 3b), while rural DAs showed an additional reduction in birth weight with increasing PM2.5 levels compared to urban DAs (Fig. 3c).

Table 3 Adjusted individual and DA-level fixed effects on continuous and term birth weight and their modification by PM2.5 (Model-4)
Full size table
Table 4 Sensitivity analysis using only term birth and excluding stillbirths and congenital anomalies (N = 207,405)
Full size table
Fig. 2
figure 2

Adjusted predicted effects of maternal risk factors on birth weight across levels of PM2.5. a Maternal Smoking b Suspected Drug or Alcohol Use c Maternal Age d Gestational Diabetes. Predicted effects on birth weight with 95 % confidence intervals are conditional on model covariates included in Model 4. Black vertical lines represent the frequency distribution of PM2.5

Full size image
Fig. 3
figure 3

Adjusted predicted effects of DA-level factors on birth weight across levels of PM2.5. a Socioeconomic Status Index (SESi) b Asian Immigrant Density c Rural Residence. Predicted effects on birth weight with 95 % confidence intervals are conditional on model covariates included in Model 4. Black vertical lines represent the frequency distribution of PM2.5

Full size image

The random effects, the explained proportional change in variance (PVC), and model diagnostics are presented in Table 5. The unadjusted ICC for the Null random-intercept model was 0.019, indicating that 1.9 % of the total residual differences in birth weight are attributable to DA-level contextual factors. The inclusion of the level-1 covariates along with the random slope for gestational age, the ICCadj (now conditional on mean-centred gestational age, 38.8 weeks), increased to 2.3 %. This was due to the large reduction in the level-1 residual variance (560.5 to 435.2) relative to the reduction in the level-2 random intercept variance (78.8 to 67.0). The addition of DA-level variables in Model 2 and Model 3 removed a lot of the DA-level variance reducing the ICCadj to 1.1 and 0.8 % respectively.

Table 5 Random effects and model diagnostics from hierarchical linear models for continuous birth weight in BC, Canada
Full size table

The level-2 random intercept standard deviation indicates that the mean birth weight for every DA has a degree of variability from the overall (BC-wide) mean birth weight. For the Null model, the overall birth weight intercept is was 3434.2 g with a standard deviation of 78.8. This equals a 8.6 % difference in range between 95 % of the DAs (3434.2 ± (1.96 × 78.8) = 3280.0 and 3588.8 g). Similarly, we calculate the between-DA 95 % distributional range of slopes for gestational age to fall between 255.8 and 361.6 g (308.7 ± (1.96 × 26.6)), a 29.3 % difference in how one week gestation increases birth weight between DAs.

The level-1 and level-2 explained PCV (L1-PCV & L2-PCV) summarizes the relative degree of explained variance at the different levels between the different models. Using the Null model as the reference, the L1-model resulted in an L1-PCV & L2-PCV of 39.7 and 27.7 % respectively (Table 5). These are fairly large PCVs, indicating the within and between-DA variance in birth weight shown in the Null model was moderately attributable to these individual (level-1) compositional factors, largely gestational age. The addition of the DA-level variables in Model 2 explained an additional 38.3 % of the DA-level variance (cumulative L2-PVC = 66 %). The additions of PM2.5 and season of birth in Model 3 further explained an additional 8.5 % of the level-2 intercept variance beyond that of Model 2. Model 4 accounted for an additional 2.3 % L2-PCV.

Spatial analyses were used as a model diagnostic to test for significant spatial autocorrelation of the model residuals. The local Moran’s I statistics reported in Table 5 indicate the degree of local spatial autocorrelation (i.e., spatial clustering) of the level-1 (L1) residuals as well as the level-2 predicted random intercepts (L2ri) and slopes (L2rs) for all five models. Interpreted in a similar manner as a regular correlation coefficient, the Moran’s I statistic reveals the existence of significant localized clustering of residuals at both level-1 and level-2 in the Null and level-1 model. The addition of level-2 variables reduced the Moran’s I substantially; however small but significant clustering remained.

Sensitivity analyses using only the non-imputed DAs (N1 = 228,765 in 6230 DAs) showed very minor differences in magnitude of significant variables in the birth weight models. In a second sensitivity analysis, we restricted the sample to only term births excluding stillbirths and congenital anomalies. As expected, there was a large reduction in the random-slope variance for gestational age due to dropping preterm births but no other differences in the observed relationships (Table 4).

Finally, a check for potential collider bias was performed by omitting gestational age as a covariate from the models [45]. Random-intercept models equivalent to those presented in Tables 2, 3 and 4 were run and assessed for differences. Many of the individual-level covariates returned to resemble their unadjusted estimates listed in Table 1. The interaction between PM2.5 and cigarettes/day remained unchanged, whereas the PM2.5 interaction with drug or alcohol flag was no longer significant (p = 0.066). The interaction between gestational diabetes and PM2.5 was reduced by half, but was still significant. The effects of the DA-level variables SESi and immigrant density increased moderately beyond their 95 % CIs listed for Model 4. The effect for PM2.5 decreased but was still significant (-18.0 (95 % CI -21.3 – -14.7)). The DA-level interaction between PM2.5 and immigrant density was reduced by half and marginally not significant (p = 0.051), while the interaction between SESi and PM2.5 was also reduced to non-significance (p = 0.33).

Discussion

This study employed multilevel random coefficient models to assess the effect of PM2.5 on birth weight and test its interaction with individual and neighbourhood-level risk factors. Our results show that individual and neighbourhood-level factors are capable of modifying the association between PM2.5 exposure and fetal growth. Furthermore, through the use of random-slopes models we show that the effect of gestational age on birth weight can vary considerably between neighbourhood DAs which was only moderately addressed in our models. After adjusting for individual-level covariates and DA-level socio-economic and socio-demographic variables, we found a significant non-linear effect between PM2.5 and birth weight in 231,929 births in British Columbia, Canada. This association was robust to the exclusion of stillbirths and congenital anomalies as well as the use of only term births and models dropping gestational age as a covariate, demonstrating that selection bias does not affect the observed PM2.5 main effects.

Our results corroborate the growing literature supporting a negative association between PM2.5 and birth weight [3, 46, 47]. Even in settings of relatively low air pollution exposure similar to our study, significant reductions in birth weight have been observed [48]. This strengthens the evidence of the low-dose effects of PM2.5 and is exemplified by Fig. 1 which shows the largest potential effects on birth weight are seen at the low to mid concentrations of PM2.5, a not uncommon dose–response phenomenon also observed in other exposure-disease contexts [49]. Other studies testing for non-linear effects of traffic-related air pollutants on fetal growth have been mixed [23, 50, 51]. Interestingly, we found a similar non-linear dose–response between cigarettes/day and birth weight, an effect also shown by England et al. using both self-reported cigarettes/day as well as urine-cotinine levels to assess exposure [52].

Our results show a negative interaction between PM2.5 and SES such that a more pronounced effect of PM2.5 was seen in lower SES neighbourhoods (Fig. 3a); however this result could be sensitive to collider bias. We also observed significant interactions between PM2.5 and Asian immigration density as well as with PM2.5 and living in a rural location (Fig. 3b and c respectively). This suggests that that neighbourhood characteristics can not only influence fetal growth but can also modify exposures either positively or negatively. The biological mechanisms supporting such interactions have been recently reviewed [5], and have been indirectly supported in epidemiological studies that found stronger effects of PM2.5 across race, age and SES groups [23, 25, 53]. For example, the observed lower birth weights associated with neighbourhoods with higher densities of continental Asian immigrants is likely due to constitutional birth size differences [37, 38], but the positive interaction with PM2.5 may reflect the buffering effect of strong community cohesiveness and beneficial cultural practices [23, 32]. A similar interaction was found by Basu et al. in which births to Asian mothers exhibited smaller birth weight reductions for PM2.5 constituents compared to Caucasian births [23]. Currie et al. however did not find any significant interactions between traffic-related carbon monoxide exposure and risk factors such as race, education, or low income [54].

The significant negative interaction between rural address and PM2.5 may reflect the underestimation of PM2.5 in rural areas by the LUR model [34]. The composition of PM2.5, and thus its relative toxicity, is shown to vary spatially depending on its source (e.g., wood smoke vs. traffic-related emissions) and may partially explain the observed rural–urban differences [4, 6]. The significant negative association between season of birth and birth weight could also reflect the increased presence of wood heating and vehicle exhaust in combination with winter stagnation events, but could also reflect a change in diet or increased infection rates [4, 55]. An interaction between season of birth and PM2.5 was not statistically significant.

Interactions between PM2.5 and maternal-level variables shown to reduce birth weight independently revealed some counter-intuitive results. This included the PM2.5 interaction with maternal smoking (cigarettes/day) and with suspected drug/alcohol use where increasing PM2.5 levels tempered the negative effect of these risk factors (Fig. 2a and b). This finding was counterintuitive to our original hypothesis and published literature [54], and gave rise to the suspicion of survival bias due to competing risks (i.e., risk behaviours leading to early miscarriage, preterm or stillbirths). Although we were not able to control for fetal loss prior to 20 weeks gestation, survival bias was mitigated by using a near full population sample that included stillbirths, congenital anomalies and preterm births. Furthermore, the positive interaction between maternal smoking and PM2.5 was unchanged after the sensitivity analyses; however, the interaction between drug and alcohol use and PM2.5 remained positive but was no longer significant (p = 0.054). The persistence of this finding leads to a hypothesis that some individual-level exposures may act as a pre-conditioning stress that activates an adaptive response of increased biological resistance to similar stressors [56].

A protective effect of older maternal age against PM2.5 exposure was also observed by Basu et al. [23], and may stem from increased nutritional awareness among older women and/or more secure income and support networks thereby reducing potential stress and anxiety [57, 58]. Currie et al. also found significant interactions between traffic-related carbon monoxide exposure and maternal age, but that both younger (< age 19) and older (> age 34) maternal age had greater reductions in birth weight [54]. Gestational diabetes has been shown to be associated with PM2.5 and other air pollutants [59, 60]; however, their interaction with respect to birth weight has not been assessed. Our study showed that pregnancies affected by gestational diabetes had significantly higher birth weights as expected but revealed a sharp reduction in birth weight with increasing PM2.5. This significant negative interaction between PM2.5 and gestational diabetes could be related to excess of systemic or placental oxidative stress and inflammation resulting in restricted fetal growth [15, 61].

While the application of multilevel models in perinatal epidemiology has become more common [62], most have been random-intercept models with very few including a random-slope parameter. Permitting the slope for individual level-1 gestational age to be random can elucidate how its effect on birth weight differs between DAs. For example, the addition of level-1 covariates reduced the random-slope variability from 30.1 to 26.6. This suggests that that these maternal risk factors act through gestational age to influence birth weight and are not distributed homogeneously across DAs. In light of these findings, significant inter-DA variance remained for both the random intercept and slope. In other words, despite explaining a substantial proportion of the between-DA variance in birth weight with both level-1 compositional and level-2 contextual factors, there remained unmeasured DA-level mechanisms acting either directly on fetal growth and/or through gestational age to produce between neighbourhood differences in birth weight.

Spatial analyses were used to examine the wider spatial context within which the DAs are situated and also served as a measure of model specification. The inclusion of the DA-level variables and interactions substantially reduced the spatial autocorrelation in the level-1 and level-2 random-intercept residuals (L1 & L2ri Moran’s I in Table 5). There was very little spatial autocorrelation in the level-2 random-slope parameters (L2rs Moran’s I in Table 5), but it was also reduced in the DA-level models. This suggests that the census DAs perform well in capturing neighbourhood-level processes and that the selected DA-level variables do well to addess the underlying spatial processes acting on birth weight at this neighbourhood-level.

A key component of this research was the use of a land-use regression (LUR) model of air pollution [34]. While the LUR model was independently validated and achieved decent overall results in its predicted estimates, the very nature of our study design ensured some degree of exposure misclassification to our study population. Our analysis was based on maternal place of residence at delivery, and therefore intra-urban commuting and potential inter-urban relocation within the pregnancy period was not accounted for which could affect the results. Time-activity patterns show that pregnant women spend more time at home in the later stages of pregnancy, but mobility patterns may differ by age, parity and SES [63, 64]. Another limitation regarding the PM2.5 exposure assessment is that the LUR model is cross-sectional based on 2006 air quality monitoring data, while the study period of our perinatal dataset spans 6 years (2001 to 2006). We therefore assume all pregnancies were exposed to the same levels of PM2.5 for their entire pregnancy, regardless of their year of birth, based on their residential DA. While this method prevents the assessment of exposure windows by trimester, spatiotemporal studies of PM2.5 have shown little to no difference between trimester-specific and entire pregnancy effects on birth weight [3, 46, 48]. Finally, the mean PM2.5 concentrations may be underestimated by the LUR model with less variability and missing several high PM2.5 outlier locations in BC compared to compiled monitored data [26]. This could potentially result in an underestimation of our observed association of reduced birth weight with increasing PM2.5 levels.

We were unable to control for maternal-level SES, and therefore the neighbourhood-level effect estimates and interactions could reflect individual-level differences. For example, the protective effect of older maternal age buffering the PM2.5 effect on birth weight could be due to individual-level SES factors not accounted for in our models such as diet, income or stress. However, studies have found that adjustment for individual-level measures of SES did not significantly change the area-level associations [20, 65]. Maternal education is a variable provided in the BC Perinatal Data Registry, but was only available for 10 % of our population. However, the adjustment for socially-patterned behavioural risk factors such as maternal smoking, suspected drug or alcohol use and low number of prenatal care visits will control for some individual-level SES differences [66].

Conclusions

This study supports the growing literature of an effect of PM2.5 on birth weight and its modification by both maternal and neighbourhood-level factors. Most notably, it shows that lower SES neighbourhoods may be more negatively affected by higher levels of PM2.5. We observed both positive and negative interactions between maternal factors and PM2.5 that require further scrutiny but may reflect a PM2.5-oxidiative stress pathway expressed via either protective pre-conditioning or harmful overload. Targeted municipal-level interventions to reduce PM2.5 and improved neighbourhood SES may help improve birth outcomes at the population-level.

Abbreviations

BC:

British Columbia

CSD:

census subdivision

DA:

dissemination area

ICC:

intra-class correlation

L1:

level-1

L2:

level-2

LHA:

local health area

LUR:

land use regression

NO2 :

nitrogen dioxide

PCV:

proportional change in variance

PM2.5 :

particulate matter less than 2.5 microns in diameter

PSBC:

Perinatal Services British Columbia

RI:

random intercept

RS:

random slope

SES:

socio-economic status

SESi:

socio-economic status index

References

  1. Kramer MS. Determinants of low birth weight: methodological assessment and meta-analysis. Bull World Health Organ. 1987;65(5):663–737.

    CAS  Google Scholar 

  2. Sram RJ, Binkova B, Dejmek J, Bobak M. Ambient air pollution and pregnancy outcomes: a review of the literature. Environ Health Perspect. 2005;113:4.

    Article  Google Scholar 

  3. Dadvand P, Parker J, Bell ML, Bonzini M, Brauer M, Darrow LA, et al. Maternal exposure to particulate air pollution and term birth weight: a multi-country evaluation of effect and heterogeneity. Environ Health Perspect. 2013;121(3):267–373.

    Article  Google Scholar 

  4. Bell ML. Assessment of the health impacts of particulate matter characteristics. [Internet]. Boston: HEI Health Review Committee, editor. Health Effects Institute; 2012.

    Google Scholar 

  5. Erickson AC, Arbour L. The shared pathoetiological effects of particulate air pollution and the social environment on fetal-placental development. J Environ Public Health. 2014;2014:901017.

    Article  Google Scholar 

  6. HEI Review Panel on Ultrafine Particles. Understanding the Health Effects of Ambient Ultrafine Particles HEI Perspectives 3. Boston: Health Effects Institute; 2013.

  7. van den Hooven EH, Pierik FH, de Kluizenaar Y, Hofman A, van Ratingen SW, Zandveld PYJ, et al. Air pollution exposure and markers of placental growth and function: the generation R study. Environ Health Perspect. 2012;120(12):1753–9.

    Google Scholar 

  8. Janssen BG, Munters E, Pieters N, Smeets K, Cox B, Cuypers A, et al. Placental mitochondrial DNA content and particulate air pollution during in utero life. Environ Health Perspect. 2012;120:9.

    Article  Google Scholar 

  9. Veras MM, Damaceno-Rodrigues NR, Caldini EG, Maciel Ribeiro AAC, Mayhew TM, Saldiva PHN, et al. Particulate urban air pollution affects the functional morphology of mouse placenta. Biol Reprod. 2008;79(3):578–84.

    Article  CAS  Google Scholar 

  10. Burton GJ, Jauniaux E. Oxidative stress. Best Pract Res Clin Obstet Gynaecol. 2011;25(3):287–99.

    Article  Google Scholar 

  11. Al-Gubory KH. Environmental pollutants and lifestyle factors induce oxidative stress and poor prenatal development. Reprod Biomed Online. 2014;29(1):17–31.

    Article  CAS  Google Scholar 

  12. Aycicek A, Ipek A. Maternal active or passive smoking causes oxidative stress in cord blood. Eur J Pediatr. 2008;167(1):81–5.

    Article  CAS  Google Scholar 

  13. Kovacic P, Somanathan R. Mechanism of teratogenesis: electron transfer, reactive oxygen species, and antioxidants. Birth Defects Res C Embryo Today Rev. 2006;78(4):308–25.

    Article  CAS  Google Scholar 

  14. Haque SK, Siddiqui MU, Islam N, Moin S. Erythrocyte markers of oxidative stress in higher age-group preeclamptic and normal pregnant mothers. Hypertens Pregnancy. 2010;29(1):69–81.

    Article  CAS  Google Scholar 

  15. Kamath U, Rao G, Raghothama C, Rai L, Rao P. Erythrocyte indicators of oxidative stress in gestational diabetes. Acta Paediatr. 1998;87(6):676–9.

    Article  CAS  Google Scholar 

  16. Gruenewald TL, Karlamangla AS, Hu P, Stein-Merkin S, Crandall C, Koretz B, et al. History of socioeconomic disadvantage and allostatic load in later life. Soc Sci Med. 2012;74(1):75–83.

    Article  Google Scholar 

  17. Kramer MS, Seguin L, Lydon J, Goulet L. Socio-economic disparities in pregnancy outcome: why do the poor fare so poorly? Paediatr Perinat Epidemiol. 2000;14(3):194–210.

    Article  CAS  Google Scholar 

  18. Blumenshine P, Egerter S, Barclay CJ, Cubbin C, Braveman PA. Socioeconomic disparities in adverse birth outcomes: a systematic review. Am J Prev Med. 2010;39(3):263–72.

    Article  Google Scholar 

  19. Bird CE, Seeman T, Escarce JJ, Basurto-Dávila R, Finch BK, Dubowitz T, et al. Neighbourhood socioeconomic status and biological “wear and tear” in a nationally representative sample of US adults. J Epidemiol Community Health. 2010;64(10):860–5.

    Article  Google Scholar 

  20. Cubbin C, Winkleby MA. Protective and harmful effects of neighborhood-level deprivation on individual-level health knowledge, behavior changes, and risk of coronary heart disease. Am J Epidemiol. 2005;162(6):559–68.

    Article  Google Scholar 

  21. Merlo J, Chaix B, Yang M, Lynch J, Råstam L. A brief conceptual tutorial on multilevel analysis in social epidemiology: interpreting neighbourhood differences and the effect of neighbourhood characteristics on individual health. J Epidemiol Community Health. 2005;59(12):1022–8.

    Article  Google Scholar 

  22. Merlo J, Yang M, Chaix B, Lynch J, Råstam L. A brief conceptual tutorial on multilevel analysis in social epidemiology: investigating contextual phenomena in different groups of people. J Epidemiol Community Health. 2005;59(9):729–36.

    Article  Google Scholar 

  23. Basu R, Harris M, Sie L, Malig B, Broadwin R, Green R. Effects of fine particulate matter and its constituents on low birth weight among full-term infants in California. Environ Res. 2014;128:42–51.

    Article  CAS  Google Scholar 

  24. Généreux M, Auger N, Goneau M, Daniel M. Neighbourhood socioeconomic status, maternal education and adverse birth outcomes among mothers living near highways. J Epidemiol Community Health. 2008;62(8):695–700.

    Article  Google Scholar 

  25. Zeka A, Melly SJ, Schwartz J. The effects of socioeconomic status and indices of physical environment on reduced birth weight and preterm births in Eastern Massachusetts. Environ Health Perspect. 2008;7:60.

    Article  Google Scholar 

  26. Canadian Council of Ministers of the Environment. Canada-wide standards for particulate matter and ozone: five year report: 2000-2005. Winnipeg: Health Effects Institute; 2006. Report No.: PN 1374. [Accessed 19 Oct 2015]. Available from: http://www.ccme.ca/en/resources/air/pm_ozone.html.

  27. Perinatal Services BC. British Columbia Perinatal Data Registry [Internet]. Vancouver; [Accessed 15 Dec 2015]. Available from: http://www.perinatalservicesbc.ca/health-professionals/data-surveillance/perinatal-data-registry.

  28. DMTI Spatial. GeoPinpoint. Ont: Markham; 2007.

    Google Scholar 

  29. ESRI. ArcGIS 10.2. Redlands: Environmental Systems Research Institute; 2010.

    Google Scholar 

  30. Gauvin L, Robitaille E, Riva M, McLaren L, Dassa C, Potvin L. Conceptualizing and operationalizing neighbourhoods: the conundrum of identifying territorial units. Can J Public Health. 2007;98 Suppl 1:S18–26.

    Google Scholar 

  31. Holowaty EJ, Norwood TA, Wanigaratne S, Abellan JJ, Beale L. Feasibility and utility of mapping disease risk at the neighbourhood level within a Canadian public health unit: an ecological study. Int J Health Geogr. 2010;9:21.

    Article  Google Scholar 

  32. Meng G, Thompson ME, Hall GB. Pathways of neighbourhood-level socio-economic determinants of adverse birth outcomes. Int J Health Geogr. 2013;12(1):32.

    Article  Google Scholar 

  33. du Plessis V, Beshiri R, Bollman RD, Clemenson H. Definitions of “Rural”. Ottawa: Statistics Canada; 2002. Report No.: 61.

  34. Hystad P, Setton E, Cervantes A, Poplawski K, Deschenes S, Brauer M, et al. Creating national air pollution models for population exposure assessment in Canada. Environ Health Perspect. 2011;119(8):1123–9.

    Article  CAS  Google Scholar 

  35. Jerrett M, Finkelstein M. Geographies of risk in studies linking chronic air pollution exposure to health outcomes. J Toxicol Environ Health A. 2005;68(13-14):1207–42.

    Article  CAS  Google Scholar 

  36. Chan E, Serrano J, Chen L, Stieb DM, Jerrett M, Osornio-Vargas A. Development of a Canadian socioeconomic status index for the study of health outcomes related to environmental pollution. BMC Public Health. 2015;15(1):714.

    Article  Google Scholar 

  37. Shiono PH, Klebanoff MA, Graubard BI, Berendes HW, Rhoads GG. Birth weight among women of different ethnic groups. JAMA. 1986;255(1):48–52.

    Article  CAS  Google Scholar 

  38. Janssen P, Thiessen P, Klein M, Whitfield MF, MacNab YC, Cullus-Kuhl SC. Standards for the measurement of birth weight, length and head circumfrence at term in neonates of european, Chinese and South Asian ancestry. Open Med. 2007;1(2):E74–88.

    Google Scholar 

  39. Graham MH. Confronting multicollinearity in ecological multiple regression. Ecology. 2003;84(11):2809–15.

    Article  Google Scholar 

  40. Canada S. Census of Canada. Topic-based Tabulations, 2006 [Internet]. Ottawa: Statistics Canada; 2006.

    Google Scholar 

  41. Snijders TAB, Bosker RJ. Multilevel analysis: an introduction to basic and advanced multilevel modeling. 2nd ed. London: Sage; 2012.

    Google Scholar 

  42. StataCorp LP. Stata 13. College Station: StataCorp LP.; 2014.

    Google Scholar 

  43. Anselin L. Local indicators of spatial association — LISA. Geogr Anal. 1995;27(2):93–115.

    Article  Google Scholar 

  44. Skrondal A, Rabe-hesketh S. Prediction in multilevel generalized linear models. J R Stat Soc A. 2009;172(3):659–87.

    Article  Google Scholar 

  45. Wilcox AJ, Weinberg CR, Basso O. On the pitfalls of adjusting for gestational age at birth. Am J Epidemiol. 2011;174(9):1062–8.

    Article  Google Scholar 

  46. Stieb DM, Chen L, Eshoul M, Judek S. Ambient air pollution, birth weight and preterm birth: a systematic review and meta-analysis. Environ Res. 2012;117:100–11.

    Article  CAS  Google Scholar 

  47. Pedersen M, Giorgis-Allemand L, Bernard C, Aguilera I, Andersen A-MN, Ballester F, et al. Ambient air pollution and low birthweight: a European cohort study (ESCAPE). Lancet Respir Med. 2013;1(9):695–704.

    Article  CAS  Google Scholar 

  48. Brauer M, Lencar C, Tamburic L, Koehoorn M, Demers P, Karr C. A cohort study of traffic-related air pollution impacts on birth outcomes. Environ Health Perspect. 2008;116(5):680–6.

    Article  Google Scholar 

  49. Wigle DT, Lanphear BP. Human health risks from low-level environmental exposures: no apparent safety thresholds. PLoS Med. 2005;2(12):e350.

    Article  Google Scholar 

  50. Ballester F, Estarlich M, Iñiguez C, Llop S, Ramón R, Esplugues A, et al. Air pollution exposure during pregnancy and reduced birth size: a prospective birth cohort study in Valencia, Spain. Environ Health. 2010;9:6.

    Article  Google Scholar 

  51. Barnett AG, Plonka K, Seow WK, Wilson L-A, Hansen C. Increased traffic exposure and negative birth outcomes: a prospective cohort in Australia. Environ Heal BioMed Central Ltd. 2011;10(1):26.

    Google Scholar 

  52. England LJ, Kendrick JS, Gargiullo PM, Zahniser C, Hannon WH. Measures of maternal tobacco exposure and infant birth weight at term. Am J Epidemiol. 2001;153(10):954–60.

    Article  CAS  Google Scholar 

  53. Ebisu K, Bell ML. Airborne PM2.5 chemical components and low birth weight in the northeastern and mid-Atlantic regions of the United States. Environ Health Perspect. 2012;120(12):1746–52.

    CAS  Google Scholar 

  54. Currie J, Neidell M, Schmieder JF. Air pollution and infant health: lessons from New Jersey. J Health Econ. 2009;28:688–703.

    Article  Google Scholar 

  55. Rasmussen SA, Jamieson DJ, Uyeki TM. Effects of influenza on pregnant women and infants. Am J Obstet Gynecol. 2012;207(3 Suppl):S3–8.

    Article  Google Scholar 

  56. Calabrese EJ, Bachmann KA, Bailer AJ, Bolger PM, Borak J, Cai L, et al. Biological stress response terminology: Integrating the concepts of adaptive response and preconditioning stress within a hormetic dose-response framework. Toxicol Appl Pharmacol. 2007;222(1):122–8.

    Article  CAS  Google Scholar 

  57. Newburn-Cook CV, Onyskiw JE. Is older maternal age a risk factor for preterm birth and fetal growth restriction? A systematic review. Health Care Women Int. 2005;26(9):852–75. Taylor & Francis Group.

    Article  Google Scholar 

  58. Public Health Agency of Canada. Canadian Perinatal Health Report - 2008 edition. Ottawa: Maternal and Infant Health Section; 2008. Available from: http://www.publichealth.gc.ca/cphr/.

  59. Hu H, Ha S, Henderson BH, Warner TD, Roth J, Kan H, et al. Association of atmospheric particulate matter and ozone with gestational diabetes mellitus. Environ Health Perspect. 2015;123(9):853–9.

    Google Scholar 

  60. Malmqvist E, Jakobsson K, Tinnerberg H, Rignell-Hydbom A, Rylander L. Gestational diabetes and preeclampsia in association with air pollution at levels below current air quality guidelines. Environ Health Perspect. 2013;121(4):488–93.

    Google Scholar 

  61. Qiu C, Hevner K, Enquobahrie DA, Williams MA. Maternal serum heme-oxygenase-1 (HO-1) concentrations in early pregnancy and subsequent risk of gestational diabetes mellitus. PLoS One. 2012;7(11):e48060.

    Article  CAS  Google Scholar 

  62. Metcalfe A, Lail P, Ghali WA, Sauve RS. The association between neighbourhoods and adverse birth outcomes: a systematic review and meta-analysis of multi-level studies. Paediatr Perinat Epidemiol. 2011;25(3):236–45.

    Article  Google Scholar 

  63. Nethery E, Brauer M, Janssen P. Time-activity patterns of pregnant women and changes during the course of pregnancy. J Expo Sci Environ Epidemiol. 2009;19(3):317–24.

    Article  Google Scholar 

  64. Bell ML, Belanger K. Review of research on residential mobility during pregnancy: consequences for assessment of prenatal environmental exposures. J Expo Sci Environ Epidemiol. 2012;22(5):429–38.

    Article  Google Scholar 

  65. Luo ZC, Wilkins R, Kramer MS. Effect of neighbourhood income and maternal education on birth outcomes: a population-based study. CMAJ. 2006;174(10):1415–20.

    Article  Google Scholar 

  66. Erickson AC, Arbour LT. Heavy smoking during pregnancy as a marker for other risk factors of adverse birth outcomes: a population-based study in British Columbia, Canada. BMC Public Health. 2012;12(1):102.

    Article  Google Scholar 

Download references

Acknowledgements

We acknowledge Adrian Barnett in his thoughtful review and suggestions in an early draft manuscript as well as his general statistical advice. We acknowledge Eleanor Setton for her general support. We’d also like to acknowledge the staff at the Perinatal Services BC for their support regarding data access and manuscript review. All inferences, opinions, and conclusions drawn in this publication are those of the authors and do not reflect the opinions or policies of Perinatal Services BC. This publication was funded in part by the Canadian Institute of Health Research (CIHR) Operational Grant (protocol #: 200903-202069). Funders provided no role in the design, conduct, data collection, management, analysis, interpretation, preparation, review and approval of this study.

Author information

Authors and Affiliations

  1. Division of Medical Sciences, University of Victoria, Medical Science Bld. Rm-104, PO Box 1700 STN CSC, Victoria, V8W 2Y2, BC, Canada

    Anders C. Erickson & Laura Arbour

  2. Department of Geography, University of Victoria, David Turpin Bldg. Rm-B203, PO Box 1700 STN CSC, Victoria, V8W 2Y2, BC, Canada

    Aleck Ostry

  3. Center for Advanced Research in Environmental Genomics, University of Ottawa, 20 Marie-Curie, Ottawa, K1N 6N5, ON, Canada

    Laurie H. M. Chan

  4. Department of Medical Genetics, University of British Columbia, C201 - 4500 Oak Street, Vancouver, V6H 3N1, BC, Canada

    Laura Arbour

Authors
  1. Anders C. Erickson
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Aleck Ostry
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Laurie H. M. Chan
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Laura Arbour
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Laura Arbour.

Additional information

Competing interests

The authors declare they have no competing interests.

Authors’ contributions

AE was primarily responsible for the study design, acquisition of data, analysis and interpretation of results, drafting the article and making revisions. LA critically assessed the concept, study design, analysis and interpretation of results. LA provided critical edits of the content. Final approval of submitted manuscript was provided by both authors. LC and AO provided thoughtful edits on draft manuscripts and input into the manuscript framework. All authors read and approved the final manuscript.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Erickson, A.C., Ostry, A., Chan, L.H.M. et al. The reduction of birth weight by fine particulate matter and its modification by maternal and neighbourhood-level factors: a multilevel analysis in British Columbia, Canada. Environ Health 15, 51 (2016). https://doi.org/10.1186/s12940-016-0133-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12940-016-0133-0

Keywords

Environmental Health

ISSN: 1476-069X