Maternal Exposure to Particulate Air Pollution and Term Birth Weight: A Multi-Country Evaluation of Effect and Heterogeneity
Background: A growing body of evidence has associated maternal exposure to air pollution with adverse effects on fetal growth; however, the existing literature is inconsistent.
Objectives: We aimed to quantify the association between maternal exposure to particulate air pollution and term birth weight and low birth weight (LBW) across 14 centers from 9 countries, and to explore the influence of site characteristics and exposure assessment methods on between-center heterogeneity in this association.
Methods: Using a common analytical protocol, International Collaboration on Air Pollution and Pregnancy Outcomes (ICAPPO) centers generated effect estimates for term LBW and continuous birth weight associated with PM10 and PM2.5 (particulate matter ≤ 10 and 2.5 µm). We used meta-analysis to combine the estimates of effect across centers (~ 3 million births) and used meta-regression to evaluate the influence of center characteristics and exposure assessment methods on between-center heterogeneity in reported effect estimates.
Results: In random-effects meta-analyses, term LBW was positively associated with a 10-μg/m3 increase in PM10 [odds ratio (OR) = 1.03; 95% CI: 1.01, 1.05] and PM2.5 (OR = 1.10; 95% CI: 1.03, 1.18) exposure during the entire pregnancy, adjusted for maternal socioeconomic status. A 10-μg/m3 increase in PM10 exposure was also negatively associated with term birth weight as a continuous outcome in the fully adjusted random-effects meta-analyses (–8.9 g; 95% CI: –13.2, –4.6 g). Meta-regressions revealed that centers with higher median PM2.5 levels and PM2.5:PM10 ratios, and centers that used a temporal exposure assessment (compared with spatiotemporal), tended to report stronger associations.
Conclusion: Maternal exposure to particulate pollution was associated with LBW at term across study populations. We detected three site characteristics and aspects of exposure assessment methodology that appeared to contribute to the variation in associations reported by centers.
The developing fetus is known to be susceptible to environmental insults (Stillerman et al. 2008). A growing body of evidence has associated maternal exposure to ambient air pollution with a range of adverse pregnancy outcomes including low birth weight (LBW), intra-uterine growth retardation, preterm birth, stillbirth, and congenital anomalies (Glinianaia et al. 2004; Sapkota et al. 2010; Šrám et al. 2005; Vrijheid et al. 2011). However, notable inconsistencies among the findings of these studies (Parker and Woodruff 2008; Parker et al. 2011; Woodruff et al. 2009) have hindered the ability of policy makers to incorporate the research evidence into policy.
Discrepancies among previous studies may reflect genuine differences in the study settings, may be a consequence of specific biases, or may arise from differences in study designs and exposure assessments (Parker and Woodruff 2008; Woodruff et al. 2009). Study setting characteristics that may contribute to variation in reported associations include the demographic characteristics of the study population, the major sources of pollutants, the size distribution of particulate pollutants [e.g., PM2.5:PM10 ratio (particulate matter ≤ 2.5 and 10 µm)], maternal time–activity patterns, the study period, the degree of confounding by socioeconomic status (SES), and the underlying prevalence of adverse pregnancy outcomes. Relevant study design characteristics include the sources of data for feto-maternal characteristics (e.g., birth certificates, questionnaires, or hospital records), inclusion/exclusion criteria, outcome definitions (e.g., birth weight as a continuous variable, LBW, or small for gestational age), and the analysis of potential confounders and/or effect modifiers. Differences in exposure assessment include applied methods for assessing the exposure (e.g., proximity-based, monitor-based, or model-based methods), exposure time windows, exposure contrasts (e.g., spatial, temporal, or spatiotemporal), the availability of data for specific pollutants, and the analysis of associations with combinations of pollutants (Parker and Woodruff 2008).
The International Collaboration on Air Pollution and Pregnancy Outcomes (ICAPPO) was established to investigate the association between maternal exposure to ambient air pollution and pregnancy outcomes across multiple centers and to understand how differences in study settings and methods contribute to variations in findings (i.e., between-center heterogeneity) (Parker et al. 2011; Woodruff et al. 2009, 2010). To achieve this goal, we previously discussed methodological differences in the published studies (Woodruff et al. 2009), described the collaborative centers (Woodruff et al. 2010), and presented preliminary estimates of effects for each center (Parker et al. 2011).
The overarching aim of this analysis was to evaluate the association between maternal exposure to particulate air pollution and term birth weight and LBW. Toward this aim, we combined effect estimates of the individual ICAPPO centers and assessed the between-center heterogeneity in these associations. The application of a common analysis protocol across ICAPPO centers provided a unique opportunity to separate the contribution of the analytical design to between-center heterogeneity, and enabled us to assess the impact of center characteristics and exposure assessment methods on the variation in reported effect estimates by each center.
Overview. Our study was based on estimates of effects [odds ratio (OR) for LBW and regression coefficients for birth weight] that were uniformly generated and reported by each ICAPPO center according to a common protocol (Parker et al. 2011). Ambient levels of PM10 and PM2.5 were used as indicators of particulate air pollution. Our analysis was focused principally on the association between maternal exposure to PM10 during the entire pregnancy and term LBW (birth weight < 2500 g at 37–42 completed weeks of gestation) because this was reported by most ICAPPO centers. We also conducted additional analyses of data from subsets of ICAPPO centers that stratified PM10–term LBW analyses by the exposure time window (i.e., the first, second, and third trimester), analyzed birth weight as a continuous outcome variable, and estimated the association between maternal exposure to PM2.5 and term LBW. We synthesized the effect estimates across the centers by applying a meta-analysis framework. The effects of center characteristics and exposure assessment methods on between-center heterogeneity in effect estimates were explored using a meta-regression framework. The meta-analyses and meta-regressions were conducted using the R statistical package ( http://cran.r-project.org/), libraries meta, rmeta, and metafor.
ICAPPO. Our analysis relied on effect estimates provided by fourteen ICAPPO centers from nine countries with more than three million singleton term births (Table 1). For the ICAPPO analysis, the centers reanalyzed existing data sets that had been created to evaluate the impacts of maternal exposure to air pollution on pregnancy outcomes. The centers relied on outcome data available from routinely collected administrative records (birth certificates) or data collected for a specific study (Bell et al. 2007, 2008; Brauer et al. 2008; Darrow et al. 2011; Gehring et al. 2011; Glinianaia et al. 2008; Gouveia et al. 2004; Ha et al. 2004; Jalaludin et al. 2007; Lepeule et al. 2010; Mannes et al. 2005; Morello-Frosch et al. 2010; Pesatori et al. 2008; Rich et al. 2009; Slama et al. 2009; van den Hooven et al. 2009). More detailed description of the ICAPPO centers has been previously published (Parker et al. 2011; Woodruff et al. 2010).
|Center, location||Reference||Study period||Study area (km2)||No. of births||Measure of SES||Term LBW (%)||Median PM10 (μg/m3)||PM10 IQR (μg/m3)||Median PM2.5 (μg/m3)||PM2.5/PM10 ratioa||Exposure assessment||Exposure contrast|
|Atlanta, GA, USA||Darrow et al. 2011||1996–2004||4,538||325,221||Maternal education||2.62||23.5||3.1||15.8||0.67||Monitor||Temporal|
|California, USA||Morello-Frosch et al. 2010||1996–2006||423,970||1,714,509||Maternal education||2.43||28.9||16.1||16.5||0.57||Monitor||Spatiotemporal|
|Connecticut and Massachusetts, USA||Bell et al. 2007, 2008||1999–2002||41,692||173,042||Maternal education||2.16||22||7.4||20||0.91||Monitor||Spatiotemporal|
|EDEN, Poitiers and Nancy, France||Lepeule et al. 2010||2003–2006||480||1233||Age at completion of education||2.11||19||3||—||—||Monitor||Spatiotemporal|
|Lombardy, Italy||Pesatori et al. 2008||2004–2006||23,865||213,542||Maternal education||2.71||49||10||—||—||Monitor||Spatiotemporal|
|PAMPER, Newcastle upon Tyne, UK||Glinianaia et al. 2008||1962–1992||63||81,953||Area-level indicatorb||3.19||32.8c||87.8c||—||—||Model||Spatiotemporal|
|New Jersey, USA||Rich et al. 2009||1999–2003||22,592||87,281||Maternal education||2.75||28||6.9||13.7||0.49||Monitor||Spatiotemporal|
|PIAMA, North, West, and Center of the Netherlands||Gehring et al. 2011||1996–1997||12,000||3,471||Maternal education||1.15||40.5||6.7||20.3||0.50||Model||Spatiotemporal|
|Generation R, Rotterdam, Netherlands||van den Hooven et al. 2009||2002–2006||150||7,296||Maternal education||2.26||32.8||1.1||—||—||Model||Spatial|
|São Paulo, Brazil||Gouveia et al. 2004||2005||1,500||158,791||Maternal education||3.77||40.3||2.9||—||—||Monitor||Temporal|
|Seattle, WA, USA||Sathyanarayana S, Karr C, unpublished data||1998–2005||17,800||301,880||Maternal education||1.56||—||10.2||—||Monitor||Spatiotemporal|
|Seoul, South Korea||Ha et al. 2004||1998–2000||605||372,319||Maternal education||1.45||66.5||10.9||—||—||Monitor||Temporal|
|Sydney, Australia||Jalaludin et al. 2007||1998–2004||12,145||279,015||Area-level indicatord||1.62||16.5||8.2||—||—||Monitor||Temporal|
|Vancouver, BC, Canada||Brauer et al. 2008||1999–2002||3,300||66,467||Area-level indicatore||1.35||12.5||1.4||3.98||0.32||Monitor||Spatiotemporal|
|aRatio of PM10 and PM2.5 median levels. bThe Townsend Deprivation Score is an area-based measure of material deprivation calculated for each enumeration district (~ 200 households) based on 1971, 1981, and 1991 census data. cBlack smoke (~ PM4) was used as a measure of particulate air pollution. dThe Australian Bureau of Statistics (2003) Index of Relative Socio-economic Disadvantage uses a range of census factors and is assigned to each census collection district (~ 200 households). eThe percentage of women with postsecondary education.|
Participating centers were initially asked to provide information on their study location and period, available air pollutants, number of births, prevalence of term LBW, exposure assessment method, and available covariate data (Parker et al. 2011; Woodruff et al. 2010). Based on this information, ICAPPO participants developed a common analytical protocol detailing the inclusion criteria, outcomes and covariates of interest, statistical models, and sensitivity analyses. This protocol also specified a standardized way of reporting the results (Parker et al. 2011). Each center was asked to reanalyze its existing data set according to this protocol. The analyses were limited to live-born, singleton, term births with known birth weight, maternal education (or another measure of SES), dates of birth and conception, and ambient PM10 or PM2.5 concentrations during the entire pregnancy.
Primary meta-analysis. According to the common protocol, ICAPPO centers initially estimated the association between term LBW and maternal exposure to PM10 averaged over the entire pregnancy. Each center constructed three logistic regression models to estimate the odds of term LBW associated with each 10-μg/m3 increase in average PM10 exposure levels during the entire pregnancy: a) without any adjustment, b) with adjustment only for maternal SES, and c) with adjustment for maternal SES and center-specific covariates (e.g., maternal age, maternal ethnicity, maternal smoking, parity, and infant sex). For these center-specific covariates, there was no recommendation in the ICAPPO protocol and the centers had the flexibility to independently choose the suitable ones according to their settings. This selection of extra covariates used by each center has been reported elsewhere (Parker et al. 2011). The construction of these three models was to evaluate the effect of adjustment for the maternal SES and other covariates on the combined effect estimates.
ICAPPO participants chose maternal education as a common indicator of maternal SES (Parker et al. 2011). If maternal education data were not available, area-level measures of SES were used [for the PAMPER (Particulate Matter and Perinatal Events Research) study: Townsend Deprivation Score; for the Sydney study: Index of Relative Socioeconomic Disadvantage; and for the Vancouver study: percentage of women with postsecondary education]. The PAMPER study (Newcastle upon Tyne, UK) provided exposure data only for black smoke, which approximates particulate matter with an aerodynamic diameter ≤ 4 µm (PM4) and has been shown to be a reasonable surrogate for PM10 (Muir and Laxen 1995). The estimates of effect for the PAMPER study were therefore analyzed alongside the studies with PM10 measures.
We used meta-analysis to estimate combined ORs across the centers. Between-center heterogeneity was quantified using the between-center variance of effect estimates, τ2. The statistical significance of between-center heterogeneity was tested by Cochran’s Q test. The I2 statistic with 95% confidence intervals (CI) was used to estimate the proportion of total variation in effect estimates across centers that was attributable to the between-center heterogeneity (τ2) rather than within-center error (Higgins 2008). If there was significant between-center heterogeneity (i.e., Cochran’s Q test p-value < 0.05), DerSimonian–Laird random-effects models were used for meta-analysis (DerSimonian and Laird 1986); otherwise, fixed-effects models were conducted using the Mantel–Haenszel method (Mantel and Haenszel 1959). Associations with a Cochran’s Q test p-value between 0.05 and 0.10 were estimated using both fixed- and random-effects models (Higgins et al. 2002).
Additional meta-analyses. Exposure window period. Nine centers analyzed the association between maternal exposure to PM10 and term LBW stratified by the trimester of exposure and adjusted for maternal SES. We carried out meta-analyses using these reported stratified ORs to evaluate the impact of the exposure window period on the association between PM10 exposure and term LBW.
Birth weight as a continuous variable. A subset of 11 centers (all centers but Connecticut and Massachusetts, New Jersey, and Seattle, Washington) estimated the change in term birth weight as a continuous outcome variable (grams) associated with each 10-μg/m3 increase in PM10 exposure levels averaged over the entire pregnancy using three sets of linear regression models with predictors as described for the primary analysis. The combined effect estimates across the centers were calculated using meta-analysis, as described above.
PM2.5 exposure. Seven centers had data on PM2.5 (Table 1) and reported the odds of term LBW associated with each 10-μg/m3 increase in average PM2.5 exposure levels during the entire pregnancy, using three sets of logistic regression models as described for the primary analysis. We synthesized the effect estimates across these centers using meta-analysis as described before.
Sensitivity analyses. We checked the robustness of all meta-analyses results to the omission of influential centers that were identified using DFBETAS, which indicates the change in estimated coefficients after excluding a center from the meta-analysis (Viechtbauer and Cheung 2010). Centers with an absolute DFBETAS value > 1 (i.e., that resulted in ≥ 1-SD change in the estimated coefficient when omitted) were considered influential (Viechtbauer and Cheung 2010).
Meta-regressions. To assess between-center heterogeneity, we hypothesized that the following characteristics of the centers could have an impact on the estimated risk by each center: study area (square kilometers), length of the study (years), number of births, continent of the study location, latitude of the study location (degrees), percentage of term LBW births, median PM10 exposure levels (micrograms per cubic meter), median PM2.5 exposure levels (micrograms per cubic meter), interquartile range of PM10 exposure levels (micrograms per cubic meter), and PM2.5:PM10 ratio. In addition, we hypothesized that the use of model-based versus monitor-based exposure assessment, and temporal versus spatiotemporal exposure contrasts could influence center-specific estimates. Temporal exposure contrasts account for temporal variation in pollutant levels by assigning measurements from a single monitoring station to all study subjects, whereas spatiotemporal contrasts account for both spatial and temporal components of variation in air pollution levels when estimating exposure.
For associations between maternal PM10 exposure during the entire pregnancy and term LBW that showed between-center heterogeneity, we performed separate univariate meta-regressions using center-specific log-transformed ORs as outcome and each of the center or exposure assessment characteristics listed above as predictors. In effect, these meta-regressions quantified the potential impact of these factors on the estimates of effects (ORs) reported by each center. We excluded the Generation R study (Table 1) from the analysis comparing temporal and spatiotemporal exposure contrasts because its exposure assessment was based on a dispersion model, which is essentially a spatial approach without any temporal adjustment. Analyses of the influence of PM2.5 and PM2.5:PM10 ratios were based on the seven centers with PM2.5 data.
Characteristics of ICAPPO centers. Of the 14 ICAPPO centers included in our analyses (Table 1), 6 were North American, 5 European, 1 South American, 1 Asian, and 1 was Oceanian. Our analysis included > 3 million births (ranging from a little more than 1,000 to almost 2 million) generally occurring between late 1990s and mid-2000s.
Primary meta-analysis. Thirteen centers provided estimates for the association between maternal exposure to PM10 and term LBW [Figure 1A; see also Supplemental Material, Figure S1, for forest plots ( http://dx.doi.org/10.1289/ehp.1205575)]. There was statistically significant (Cochran’s Q test p-value < 0.05) between-center heterogeneity in effect estimates (ORs) reported by these centers, with τ2 ranging between 0.0003 and 0.0004 (Table 2). Therefore we used random-effects models to estimate combined ORs across the centers, which indicated a positive association between term LBW and average maternal exposure to PM10 during the entire pregnancy before adjustment (OR = 1.04; 95% CI: 1.01, 1.06) and after adjustment for SES (OR = 1.03; 95% CI: 1.01, 1.05) and SES plus center-specific covariates (OR = 1.02; 95% CI: 1.01, 1.04) (Table 2).
|OR (95% CI)||p-Value||τ2||p-Valuea||I2 (95% CI)|
|PM10 (13 centers)|
|Unadjusted||1.04 (1.01, 1.06)||< 0.01||0.0004||< 0.01||76.5% (59.9%, 86.2%)|
|Adjusted for maternal SES||1.03 (1.01, 1.05)||< 0.01||0.0003||< 0.01||79.4% (65.4%, 87.7%)|
|Adjusted for maternal SES and center-specific covariates||1.02 (1.01, 1.04)||0.01||0.0003||0.01||54.3% (14.5%, 75.6%)|
|PM2.5 (7 centers)b|
|Unadjusted||1.17 (1.08, 1.26)||< 0.01||0.0055||< 0.01||92.3% (86.7%, 95.6%)|
|Adjusted for maternal SES||1.10 (1.03, 1.18)||< 0.01||0.0039||< 0.01||89.7% (81.3%, 94.3%)|
|Adjusted for maternal SES and center-specific covariates||1.04 (0.99, 1.09)||0.09||0.0013||< 0.01||68.5% (30.4%, 85.7%)|
|ap-Value for Cochran’s Q test for heterogeneity. bIncluding Atlanta, California, Connecticut and Massachusetts, New Jersey, PIAMA, Seattle, and Vancouver.|
Additional meta-analyses. Exposure window period. Nine centers reported estimates for the association between PM10 and term LBW stratified by the trimester of exposure and adjusted for maternal SES [see Supplemental Material, Figure S2, for forest plots ( http://dx.doi.org/10.1289/ehp.1205575)]. All combined trimester-specific ORs for term LBW in association with a 10-μg/m3 increase in PM10 were positive and comparable in magnitude based on fixed-effects models adjusted for maternal SES, with ORs of 1.01 (95% CI: 1.01, 1.01), 1.01 (95% CI: 1.01, 1.02), and 1.02 (95% CI: 1.01, 1.02) for the first, second, and third trimesters, respectively (Table 3). Corresponding random-effects ORs were smaller, with a combined OR for the first trimester of 1.0 (95% CI: 1.00, 1.01) (Cochran’s Q test p = 0.06; I2 = 45.8%; 95% CI: 0, 74.9%) (Table 3).
|Trimester of exposure||Combined risk||Heterogeneity|
|OR (95% CI)||p-Value||I2 (95% CI)||p-Valuec|
|1st trimester (fixed-effects model)||1.01 (1.01, 1.01)||< 0.001||45.8% (0.0%, 74.9%)||0.064|
|1st trimester (random-effects model)||1.00 (1.00, 1.01)||0.325|
|2nd trimester (fixed-effect smodel)||1.01 (1.01, 1.02)||< 0.001||25.7% (0.0%, 65.2%)||0.213|
|2nd trimester (random-effects model)||1.01 (1.01, 1.02)||< 0.001|
|3rd trimester (fixed-effects model)||1.02 (1.01, 1.02)||< 0.001||42.7% (0.0%, 73.6%)||0.075|
|3rd trimester (random-effects model)||1.01 (1.00, 1.02)||0.001|
|aAdjusted for maternal SES. bORs from nine centers were included in the meta-analysis. cp‑Value for Cochran’s Q test for heterogeneity.|
Birth weight as a continuous variable. Eleven centers estimated the change in term birth weight as a continuous outcome variable (grams) associated with each 10-μg/m3 increase in PM10 exposure levels [Figure 1B; see also Supplemental Material, Figure S3, for forest plots ( http://dx.doi.org/10.1289/ehp.1205575)]. The Cochran’s Q test of heterogeneity was significant (p < 0.05) for all meta-analyses of PM10 and birth weight (data not shown). Random-effects meta-analyses indicated a negative association between term birth weight and a 10-μg/m3 increase in PM10, with unadjusted, SES-adjusted, and SES- plus center-specific covariate–adjusted estimated decreases of –2.7 g (95% CI: –7.2, 1.7 g, p = 0.23), –3.0 g (95% CI: –6.9, 0.9 g, p = 0.13), and –8.9 g (95% CI: –13.2, –4.6 g, p < 0.01), respectively.
PM2.5 exposure. Seven centers reported the odds of term LBW associated with each 10-μg/m3 increase in PM2.5 exposure levels [Figure 1C; see also Supplemental Material, Figure S4, for forest plots ( http://dx.doi.org/10.1289/ehp.1205575)]. All meta-analyses showed statistically significant between-study heterogeneity (Table 2). The random-effects meta-analyses demonstrated positive associations with term LBW, with statistically significant ORs based on unadjusted and SES-adjusted models, but not the model adjusted for maternal SES and center-specific covariates (Table 2).
Sensitivity analyses. The studies that were most frequently classified as influential were the California study (the largest center with more than 1.7 million births), the PAMPER study (covering a long time period with wide variation in sources and levels of exposure to black smoke, which was used as a surrogate for PM10), and the Lombardy study [a relatively large study of a heterogeneous region including a metropolitan area (Milan), a large mainly agricultural area (Po valley), and a northern mountainous area] [see Supplemental Material, Table S1 ( http://dx.doi.org/10.1289/ehp.1205575)]. The meta-analyses were generally robust to the exclusion of influential studies with regard to the magnitude of the estimated associations (data not shown), but there were some exceptions. For example, the unadjusted OR for LBW with a 10-μg/m3 increase in PM2.5 increased from 1.17 (95% CI: 1.08, 1.26) to 1.20 (95% CI: 1.00, 1.45) after California was excluded, and decreased to 1.06 (95% CI: 0.99, 1.13) after exclusion of Connecticut and Massachusetts. In addition, the average estimated reduction in SES-adjusted term birth weight with a10-μg/m3 increase in PM10 (random-effects model) increased from –3.0 g (95% CI: –6.9, 0.9 g) based on all centers (Figure 1B) to –5.5 g (95% CI: –9.3, –1.6 g) after excluding Atlanta, and to –4.9 g (95% CI: –8.6, –1.1 g) after excluding Lombardy. Similarly, the reduction in unadjusted mean birth weight increased from –2.7 g (95% CI: –7.2, 1.7 g) based on all centers (see Supplemental Material, Figure S3A) to –6.6 g (95% CI: –11.7, –1.5 g) and –5.0 g (95% CI: –10.0, 0.0 g) after removing Atlanta and Lombardy, respectively.
Meta-regressions. Of the characteristics we evaluated, median PM2.5 exposure levels, PM2.5:PM10 ratio, and exposure contrast (temporal vs. spatiotemporal approach) influenced the between-center heterogeneity in the reported PM10-term LBW associations, with centers that had higher median PM2.5 exposure levels and PM2.5:PM10 ratios, and that used a temporal exposure contrast, reporting stronger associations in most cases (Table 4). The results of leave-one-out sensitivity analyses for these three meta-regressions were generally consistent with those of meta-regressions including all centers (data not shown). However, removing Connecticut and Massachusetts almost doubled the meta-regression coefficient for the PM2.5:PM10 ratio and nullified the association for median PM2.5 exposure levels, and the meta-regression estimate for median PM2.5 exposure levels was null after New Jersey was excluded.
|Meta-regression||Temporal vs. spatiotemporal approacha,b||Median PM2.5 levels||PM2.5:PM10 ratio|
|Regression coefficient (95% CI)||Residual heterogeneity (τ2) (95% CI)||Regression coefficient (95% CI)||Residual heterogeneity (τ2) (95% CI)||Regression coefficient (95% CI)||Residual heterogeneity (τ2) (95% CI)|
|Unadjusted||0.07 (0.03, 0.11)||0.0033 (0.0007, 0.0259)*||0.03 (0.01, 0.05)||0.0105 (0.0, 0.3901)||0.47 (0.13, 0.82)||0.0019 (0.0, 0.3808)|
|Adjusted for maternal SES||0.08 (0.05, 0.12)||0.0015 (0.0001, 0.0116)*||0.02 (0.00, 0.04)||0.0077 (0.0, 0.3405)||0.39 (0.15, 0.64)||0.0010 (0.0, 0.3776)|
|Adjusted for maternal SES and center-specific covariates||0.06 (0.02, 0.10)||0.0006 (0.0, 0.0079)||0.01 (0.00, 0.03)||0.0 (0.0, 0.0608)||0.19 (0.01, 0.37)||0.0 (0.0, 0.1421)|
|aRegression coefficients for using temporal approach compared with the spatiotemporal approach. bGeneration R cohort was excluded because its exposure assessment was based on a dispersion model which is a spatial approach. *Cochran’s Q test p < 0.05.|
Meta-regressions for the exposure contrast showed statistically significant heterogeneity in residuals (from both unadjusted and SES-adjusted models), whereas meta-regressions of the median PM2.5 exposure levels and PM2.5:PM10 ratios did not (Table 4). Heterogeneity was reduced when ORs were adjusted for center-specific selection of covariates in addition to SES. Associations between center-specific log-ORs and other center or study characteristics were not statistically significant (data not shown).
To our knowledge, this study is the largest multicenter study so far reporting on the association between air pollution and fetal growth using a common analytical protocol. We synthesized reported estimates of effects by 14 ICAPPO centers around the globe and for the first time quantified the impacts of study settings and aspects of applied exposure assessment methods on between-center heterogeneity in the reported effect estimates (ORs) for this association. We found that maternal exposures to PM10 and PM2.5 during the entire pregnancy were positively associated with term LBW. For PM10, all trimester-specific exposures were associated with slightly increased odds of term LBW. Furthermore, term birth weight was reduced in association with average PM10 exposure over the entire pregnancy. Most associations showed between-center heterogeneity in the center-specific estimates of associations. Meta-regression assessments of factors possibly affecting PM10-term LBW associations indicated that heterogeneity was influenced by median PM2.5 exposure levels, the PM2.5:PM10 ratio, and the applied exposure contrast (temporal vs. spatiotemporal).
Primary meta-analyses. Making decisions about the risks from environmental exposure in the policy or clinical setting requires synthesizing and interpreting the available epidemiologic evidence, which may include evaluating a number of relevant studies with different designs. A recent systematic review and meta-analysis of the available evidence on the association between particulate pollution and pregnancy outcomes reported by Sapkota et al. (2010) estimated a combined OR of 1.02 (95% CI: 0.99, 1.05) for LBW in association with a 10-μg/m3 increase in average maternal PM10 exposure during pregnancy. Our estimates, which were based on center-specific ORs estimated using a common protocol, were slightly more precise than reported by Sapkota et al. (e.g., OR = 1.03; 95% CI: 1.01, 1.05 for a 10-μg/m3 increase in average maternal PM10 exposure based on SES-adjusted estimates); but both point estimates support a comparable deleterious effect of PM10 on term birth weight, an indicator of fetal growth. This comparable finding supports not overrelying on statistical significance nor using it as deciding factor in the evaluation of the evidence, but rather describing the degree of confidence and precision of the estimate, which is consistent with other scientific writings on this topic. Bacchetti et al. (2005, 2008), for example, caution against inappropriate reliance on statistical significance and rather support the decision-making process evaluating the degree of confidence in the findings and the effect size. As we have shown by comparing our results to those of Sapkota et al. (2010), even though we were able to improve the signal-to-noise ratio using our approach, the interpretation of the relationship between air pollution and fetal growth remained similar.
The consistency of our findings with those of Sapkota et al. (2010), our observed consistency of combined ORs and between-center heterogeneity (τ2 values) for meta-analyses of ORs adjusted for center-specific covariates, and our meta-analyses of unadjusted or SES-adjusted ORs give us confidence that synthesizing effect estimates for the air pollution–pregnancy outcomes associations reported by studies with different designs is informative. Further, experience from this study could be important for policy makers in incorporating research evidence into policy—for example, by including estimates of fetal growth in the future reviews of air quality standards.
ICCAPO centers conducted unadjusted and SES-adjusted analyses according to a common analysis protocol, as well as analyses that allowed center-specific selection of covariates. In combining results across centers, there were clear trade-offs to the two approaches. Strictly specifying the models and covariates to be consistent across centers was one way to reduce methodological differences between centers. However, there might be important center-specific considerations that required inclusion of certain covariates for validity. For example, the Atlanta results relied entirely on temporal contrasts of exposure over an 11-year time period, and failing to control for long-term time trends, as was done in the unadjusted and SES-adjusted analyses, may have led to temporal confounding. This might explain why ORs adjusted for center-specific selection of covariates were different from those unadjusted or adjusted for SES only in some cases.
Additional meta-analyses. In their meta-analysis, Sapkota et al. (2010) estimated the association between LBW and maternal PM10 exposure during the first trimester (five studies) and third trimester (seven studies), but did not detect associations with exposure in either trimester (OR = 1.00; 95% CI: 0.97, 1.03, and OR = 1.00; 95% CI: 0.99, 1.01, respectively). We estimated slightly increased relative risks of term LBW in association with PM10 exposure during all trimesters, consistent with other previous reports (Parker and Woodruff 2008; Parker et al. 2005).
Because Cochran’s Q test of heterogeneity p-values for the trimester-specific analyses of PM10 and LBW were between 0.05 and 0.10, we reported results of both fixed-effects and random-effects meta-analyses (Higgins et al. 2002). For the second and third trimesters, there were no notable differences in estimates between the two models, whereas the random-effects OR for the association with first-trimester PM10 was null. Given the evidence of heterogeneity, fixed-effects model estimates for this trimester should be interpreted with caution.
Combined ORs for term birth weight as a continuous variable indicated a reduction in term birth weight associated with a 10-μg/m3 increase in average PM10 exposure during the entire pregnancy (–2.7 g; 95% CI: –7.2, 1.7 g). The association was stronger when based on ORs adjusted for maternal SES and center-specific covariates (–8.9 g; 95% CI: –13.2, –4.6 g). This may be partly explained by adjustment for gestational age at delivery by nine centers. Gestational age at delivery could confound the association between maternal exposure to air pollution and fetal growth. Although the outcome was term birth weight or LBW, there could be a 6-week difference (between 37 and 42 weeks of gestation) in gestational age at delivery. Confounding by gestational age might have had a stronger effect on the continuous birth weight analyses than on the LBW analyses.
LBW was associated with maternal PM2.5 exposure (combined random-effects OR = 1.10; 95% CI: 1.03, 1.18 for a 10-μg/m3 increase in average PM2.5 based on SES-adjusted ORs). The strength and direction of this association was comparable with the meta-analysis OR reported by Sapkota et al. (2010) (OR = 1.09; 95% CI: 0.90, 1.32).
Meta-regressions. To our best knowledge, the impact of study characteristics and exposure assessment methods on estimated associations between adverse birth weight and maternal exposure to air pollution has not been evaluated previously. Median PM2.5 exposure levels, PM2.5:PM10 ratios, and temporal versus spatiotemporal exposure contrasts appeared to influence the estimates of effects reported by the study centers. Parker and Woodruff (2008) suggested that variation in the composition of particulate matter may have contributed to differences in associations between maternal exposure to particulate pollution and fetal growth among seven different regions in the United States (Parker and Woodruff 2008). However, we did not have data on the composition of particulate pollutants, and therefore were not able to investigate its impact.
In our analysis, meta-regressions of the influence of temporal versus spatiotemporal exposure contrasts on associations between PM10 and LBW generally showed strong evidence of residual heterogeneity, in contrast to meta-regressions of median PM2.5 exposure levels and PM2.5:PM10 ratios (Table 4). These findings could suggest that at least part of the effect of the exposure contrast on PM10–term LBW associations might have been secondary to the effect of PM2.5:PM10 ratios and median PM2.5 levels at each center.
Limitations. Our meta-regressions were based on effect estimates (ORs) from 13 ICAPPO centers. As a rule of thumb, it has been suggested that 10 effect estimates are required in order to include a covariate in meta-regression (Borenstein et al. 2009). We therefore had to limit our analysis to univariate rather than multivariate meta-regressions and test the effects of each covariate separately. As a result, we could not evaluate whether our univariate results were confounded by other factors. Furthermore, other center characteristics that could have affected center-specific effect estimates, such as the prevalence of maternal smoking, were not included in our meta-regression analyses.
The ICAPPO protocol did not include data that might have been used to better characterize personal exposure, such as direct measurements of personal exposure levels (e.g., personal monitor data), pollutant levels at microenvironmental levels (e.g., indoor, outdoor, commuting) or maternal time–activity patterns. Some exposure misclassification would have resulted from the use of effect estimates based on associations with ambient levels of pollutants as a surrogate for personal exposure levels.
Our combined effect estimates, which were based on effect estimates generated by 14 ICAPPO centers across the globe using a common analytical protocol, support an adverse impact of maternal exposure to particulate pollution on fetal growth. The estimated combined associations, although relatively small, could be of major public health importance considering the ubiquitous nature of particulate air pollution exposure and therefore the potential for considerable population attributable risks, particularly given evidence of both perinatal and lifelong effects of LBW on health (Balci et al. 2010; Gibson 2007). After reducing analytical differences as a possible source of heterogeneity by using a common protocol, we found that some of the heterogeneity in effect estimates by centers could be explained by differences in median PM2.5 exposure levels and PM2.5:PM10 ratios, suggesting geographical variation in the association between air pollution and fetal growth. In general, the direction and strength of combined estimates of association and between-center heterogeneity based on unadjusted and SES-adjusted ORs were consistent with combined estimates based on ORs adjusted for center-specific covariates. These findings highlight the contribution of study settings to inconsistencies in the available literature and can therefore increase the confidence of policy makers when summarizing existing evidence and translating it into policy.
We are grateful to D. Martinez, J. Grellier, and P. Grillo for their contributions to the analytical parts of the paper.
This project was supported by the U.S. Environmental Protection Agency under contract number EP-W-05-022 and by the National Institute of Environmental Health Sciences (R01ES016317, R01ES019587). For the Vancouver analysis, the British Columbia (BC) Ministry of Health, the BC Vital Statistics Agency, and the BC Reproductive Care Program approved access to and use of the data facilitated by Population Data BC. The U.K. PAMPER study was supported by the Wellcome Trust (grant 072465/Z/03/Z). P.D. is funded by a Juan de la Cierva fellowship (JCI-2011-09937) awarded by the Spanish Ministry of Science and Innovation.
The authors declare they have no actual or potential competing financial interests.
- Australian Bureau of Statistics. 2003. Census of Population and Housing—Socio-Economic Indexes for Areas, Australia, 2001.Available: http://www.abs.gov.au/ausstats/[email protected]/allprimarymainfeatures/09D68973F50B8258CA2573F0000DA181?opendocument [accessed 12 January 2013]. Google Scholar
Bacchetti P, McCulloch CE, Segal MR. 2008. Simple, defensible sample sizes based on cost efficiency.Biometrics 64(2):577-585 18482055. Crossref, Medline, Google Scholar Bacchetti P, Wolf LE, Segal MR, McCulloch CE. 2005. Ethics and sample size.Am J Epidemiol 161(2):105-110 15632258. Crossref, Medline, Google Scholar Balci MM, Acikel S, Akdemir R. 2010. Low birth weight and increased cardiovascular risk: Fetal programming.Int J Cardiol 144(1):110-111 19171387. Crossref, Medline, Google Scholar Bell ML, Ebisu K, Belanger K. 2007. Ambient air pollution and low birth weight in Connecticut and Massachusetts.Environ Health Perspect 115:1118-1125 17637932. Link, Google Scholar Bell ML, Ebisu K, Belanger K. 2008. The relationship between air pollution and low birth weight: effects by mother’s age, infant sex, co-pollutants, and pre-term births.Environ Res Lett 3:044003; doi: 10.1088/1748-9326/3/4/044003[Online 22 October 2008]. Crossref, Medline, Google Scholar
- Borenstein M, Hedges LV, Higgins JP, Rothstein HR. 2009. Introduction to Meta-analysis. Chippenham, UKJohn Wiley & Sons. Google Scholar
Brauer M, Lencar C, Tamburic L, Koehoorn M, Demers P, Karr C.. 2008. A cohort study of traffic-related air pollution impacts on birth outcomes.Environ Health Perspect 116:680-686 18470315. Link, Google Scholar Darrow LA, Klein M, Strickland MJ, Mulholland JA, Tolbert PE. 2011. Ambient air pollution and birth weight in full-term infants in Atlanta, 1994–2004.Environ Health Perspect 119:731-737 21156397. Link, Google Scholar DerSimonian R, Laird N.. 1986. Meta-analysis in clinical trials.Control Clin Trials 7(3):177-188 3802833. Crossref, Medline, Google Scholar Gehring U, Wijga AH, Fischer P, de Jongste JC, Kerkhof M, Koppelman GH. 2011. Traffic-related air pollution, preterm birth and term birth weight in the PIAMA birth cohort study.Environ Res 111(1):125-135 21067713. Crossref, Medline, Google Scholar Gibson AT. 2007. Outcome following preterm birth.Best Pract Res Clin Obstet Gynaecol 21(5):869-882 17531538. Crossref, Medline, Google Scholar Glinianaia SV, Rankin J, Bell R, Pless-Mulloli T, Howel D. 2004. Particulate air pollution and fetal health: a systematic review of the epidemiologic evidence.Epidemiology 15(1):36-45 14712145. Crossref, Medline, Google Scholar Glinianaia S, Rankin J, Pless-Mulloli T, Pearce MS, Charlton M, Parker L. 2008. Temporal changes in key maternal and fetal factors affecting birth outcomes: a 32-year population-based study in an industrial city.BMC Pregnancy Childbirth 8(1):39; doi: 10.1186/1471-2393-8-39[Online 19 August 2008] 18713457. Crossref, Medline, Google Scholar Gouveia N, Bremner SA, Novaes HMD. 2004. Association between ambient air pollution and birth weight in São Paulo, Brazil.J Epidemiol Community Health 58(1):11-17 14684720. Crossref, Medline, Google Scholar Ha EH, Lee BE, Park HS, Kim YS, Kim H, Kim YJ. 2004. Prenatal exposure to PM10 and preterm birth between 1998 and 2000 in Seoul, Korea.J Prev Med Public Health 37(4):300-305. Medline, Google Scholar Higgins J, Thompson S, Deeks J, Altman D.. 2002. Statistical heterogeneity in systematic reviews of clinical trials: a critical appraisal of guidelines and practice.J Health Serv Res Policy 7(1):51-61 11822262. Crossref, Medline, Google Scholar Higgins JPT. 2008. Commentary: Heterogeneity in meta-analysis should be expected and appropriately quantified.Int J Epidemiol 37(5):1158-1160 18832388. Crossref, Medline, Google Scholar Jalaludin B, Mannes T, Morgan G, Lincoln D, Sheppeard V, Corbett S.. 2007. Impact of ambient air pollution on gestational age is modified by season in Sydney, Australia.Environ Health 6:16; doi: 10.1186/1476-069X-6-16[Online 7 June 2007] 17553174. Crossref, Medline, Google Scholar Lepeule J, Caïni F, Bottagisi S, Galineau J, Hulin A, Marquis N. 2010. Maternal exposure to nitrogen dioxide during pregnancy and offspring birth weight: comparison of two exposure models.Environ Health Perspect 118:1483-1489 20472526. Link, Google Scholar Mannes T, Jalaludin B, Morgan G, Lincoln D, Sheppeard V, Corbett S.. 2005. Impact of ambient air pollution on birth weight in Sydney, Australia.Occup Environ Med 62(8):524-530 16046604. Crossref, Medline, Google Scholar Mantel N, Haenszel W.. 1959. Statistical aspects of the analysis of data from retrospective studies of disease.J Natl Cancer Inst 22(4):719-748 13655060. Medline, Google Scholar Morello-Frosch R, Jesdale BM, Sadd JL, Pastor M. 2010. Ambient air pollution exposure and full-term birth weight in California.Environ Health 9:44; doi: 10.1186/1476-069X-9-44[Online 28 July 2010] 20667084. Crossref, Medline, Google Scholar Muir D, Laxen DPH. 1995. Black smoke as a surrogate for PM10 in health studies?Atmos Environ 29(8):959-962. Crossref, Google Scholar Parker JD, Rich DQ, Glinianaia SV, Leem JH, Wartenberg D, Bell ML. 2011. The International Collaboration on Air Pollution and Pregnancy Outcomes: initial results.Environ Health Perspect 119:1023-1028 21306972. Link, Google Scholar Parker JD, Woodruff TJ. 2008. Influences of study design and location on the relationship between particulate matter air pollution and birthweight.Paediatr Perinat Epidemiol 22(3):214-227 18426516. Crossref, Medline, Google Scholar Parker JD, Woodruff TJ, Basu R, Schoendorf KC. 2005. Air pollution and birth weight among term infants in California.Pediatrics 115(1):121-128 15629991. Crossref, Medline, Google Scholar Pesatori AC, Bonzini M, Carugno M, Giovannini N, Signorelli V, Baccarelli A. 2008. Ambient air pollution affects birth and placental weight. a study from Lombardy (Italy) region.Epidemiology 19(6):S178-S179. Google Scholar Rich DQ, Demissie K, Lu SE, Kamat L, Wartenberg D, Rhoads GG. 2009. Ambient air pollutant concentrations during pregnancy and the risk of fetal growth restriction.J Epidemiol Community Health 63(6):488-496 19359274. Crossref, Medline, Google Scholar Sapkota A, Chelikowsky AP, Nachman KE, Cohen AJ, Ritz B. 2010. Exposure to particulate matter and adverse birth outcomes: a comprehensive review and meta-analysis.Air Qual Atmos Health 5(4):369-381. Crossref, Google Scholar Slama R, Thiebaugeorges O, Goua V, Aussel L, Sacco P, Bohet A. 2009. Maternal personal exposure to airborne benzene and intrauterine growth.Environ Health Perspect 117:1313-1321 19672414. Link, Google Scholar Šrám RJ, Binková B, Dejmek J, Bobak M. 2005. Ambient air pollution and pregnancy outcomes: a review of the literature.Environ Health Perspect 113:375-382 15811825. Link, Google Scholar Stillerman KP, Mattison DR, Giudice LC, Woodruff TJ. 2008. Environmental exposures and adverse pregnancy outcomes: a review of the science.Reprod Sci 15(7):631-650 18836129. Crossref, Medline, Google Scholar van den Hooven E, Jaddoe V, de Kluizenaar Y, Hofman A, Mackenbach J, Steegers E. 2009. Residential traffic exposure and pregnancy-related outcomes: a prospective birth cohort study.Environmental Health 8:59; doi: 10.1186/1476-069X-8-59[Online 22 December 2009] 20028508. Crossref, Medline, Google Scholar Viechtbauer W, Cheung MWL. 2010. Outlier and influence diagnostics for meta analysis.Res Synth Meth 1(2):112-125. Crossref, Medline, Google Scholar Vrijheid M, Martinez D, Manzanares S, Dadvand P, Schembari A, Rankin J. 2011. Ambient air pollution and risk of congenital anomalies: a systematic review and meta-analysis.Environ Health Perspect 119:598-606 21131253. Link, Google Scholar Woodruff TJ, Parker JD, Adams K, Bell ML, Gehring U, Glinianaia S. 2010. International collaboration on air pollution and pregnancy outcomes (ICAPPO).Int J Environ Res Public Health 7(6):2638-2652 20644693. Crossref, Medline, Google Scholar Woodruff TJ, Parker JD, Darrow LA, Slama R, Bell ML, Choi H. 2009. Methodological issues in studies of air pollution and reproductive health.Environ Res 109(3):311-320 19215915. Crossref, Medline, Google Scholar