Birth data. Birth certificate data for Connecticut, Delaware, Maryland, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont, Virginia, Washington, DC, and West Virginia (USA), from 1 January 2000 through 31 December 2007 were obtained from the National Center for Health Statistics (Atlanta, GA). Data that were provided include county of residence, county of birth, birth order, trimester of first prenatal care, date of last menstrual period (LMP), gestational age, infant’s sex and birth weight, as well as maternal and paternal ages and races, and maternal education, marital status, alcohol consumption, and smoking during pregnancy. Further description of these data is available elsewhere (Bell et al. 2007b (link)).
Births with unspecified county of residence or birth, plural deliveries (e.g., twins), gestational period > 44 weeks, gestational period < 37 weeks (nonterm births), birth weight < 1,000 g or > 5,500 g, different counties of residence and delivery, or impossible gestational age and birth weight combinations were excluded from analysis (Alexander et al. 1996 (link)). Births also were excluded if LMP was missing or the estimated birth based on LMP and gestational length was > 30 days from the midday of the birth month reported on the birth certificate.
Air pollution and weather data. PM2.5 chemical components data were obtained from the U.S. Environmental Protection Agency (EPA) Air Explorer (U.S. EPA 2010a). PM10 and PM2.5 total mass, CO, NO2, O3, and SO2 data were obtained from the U.S. EPA Air Quality System for 1999–2007 (U.S. EPA 2010b). We included only counties with PM2.5 chemical component data because these exposures are our primary focus. PM10, PM2.5, and PM2.5 chemical components were measured every 3–6 days. Gaseous pollutants were measured daily, although O3 was measured mainly during the warm season. Some monitors began or ceased observation during the study period. We investigated PM2.5 chemical components identified by previous research and literature review to have potential links to health and/or contribute substantially to PM2.5 total mass: aluminum, ammonium ion, arsenic, cadmium, calcium, chlorine, elemental carbon, lead, mercury, nickel, nitrate, organic carbon matter, silicon, sodium ion, sulfur, titanium, vanadium, and zinc (Bell et al. 2007a (link); Franklin et al. 2008 (link); Haynes et al. 2011 (link); Ostro et al. 2007 (link); Zanobetti et al. 2009 (link)).
We calculated apparent temperature (AT), a measure that reflects overall temperature discomfort (Kalkstein and Valimont 1986 ), based on daily temperature and dew point temperature data obtained from the National Climatic Data Center (2010) . If weather data were unavailable for a given county, we assigned the AT value for the closest county with weather data.
Exposure estimation. For each birth we calculated the average level of each pollutant during gestation and each trimester, and average AT during each trimester. Delivery date was estimated based on self-reported LMP and gestational length, assuming conception 2 weeks after LMP. We defined trimesters as 1–13 weeks, 14–26 weeks, and week 27 to delivery, as in previous studies (Bell et al. 2007b (link)).
Exposures were estimated based on county of residence. Not all counties had data for all pollutants. Measurements from multiple monitors in the same county on the same day were averaged to generate daily pollutant -levels. To avoid biases due to changes in measurement frequency, daily pollutant levels and AT values were combined to estimate weekly exposures, which were then averaged to estimate gestational or trimester exposure. Births for which exposure estimates were unavailable for > 25% of the weeks in any trimester for a given pollutant were excluded from analyses of that pollutant.
Statistical analysis. Each birth was cate-gorized as low or normal birth weight using clinically defined LBW (< 2,500 g). Logistic regression was used to estimate associations between LBW and gestational exposure to each pollutant with adjustment for maternal race (African American, Caucasian, other), marital status (married, unmarried), tobacco consumption during pregnancy (yes, no, unknown), alcohol consumption during pregnancy (yes, no, unknown), highest education (< 12 years, 12 years, 13–15 years, > 15 years, unknown), age (< 20, 20–24, 25–29, 30–34, 35–39, ≥ 40 years), infant sex (male, female), gestational length (37–38, 39–40, 41–42, 43–44 weeks), the trimester prenatal care began (1st, 2nd, 3rd, no care, unknown), first in birth order (yes, no, unknown), delivery method (vaginal, cesarean section, unknown), average AT for each trimester, season of birth, and year of birth. In addition we included regional indicators to adjust for local factors such as area-level socioeconomic conditions (Table 1 ). We conducted sensitivity analyses restricted to first births to assess the influence of multiple births by the same mother on associations (Zhu et al. 1999 (link)).
For pollutants showing statistically significant associations with LBW in single--pollutant models, we conducted two-pollutant models that included pairs of pollutants that were not highly correlated (correlation < 0.5). Similarly, for pollutants associated in single-pollutant model, we assessed effects by trimester using a model with trimesters’ exposures included simultaneously. Because trimester exposures could be correlated, we performed sensitivity analysis with trimester exposures adjusted to be orthogonal using a method we published previously (Bell et al. 2007b (link)). In brief, we predicted exposures of two trimesters using exposure level of a given trimester (reference trimester), calculated their residuals, and put them into models besides exposure of reference trimester. This approach can avoid covariance among trimester exposures. This procedure was repeated using each trimester as the reference trimester, and we have four models for trimester analysis in total (main model and three models as sensitivity -analyses). Further description of this approach is available elsewhere (Bell et al. 2007b (link)).
Additional analyses were conducted for pollutants that showed statistically robust results in two-pollutant models. We included interaction terms between gestational pollutant exposures and sex or race to investigate whether some populations are particularly susceptible, because previous analysis found higher relative risks associated with ambient air pollution in some populations than in others (Bell et al. 2007b (link)). Statistical significance was determined at an alpha level of 0.05 for the entire analyses.
Births with unspecified county of residence or birth, plural deliveries (e.g., twins), gestational period > 44 weeks, gestational period < 37 weeks (nonterm births), birth weight < 1,000 g or > 5,500 g, different counties of residence and delivery, or impossible gestational age and birth weight combinations were excluded from analysis (Alexander et al. 1996 (link)). Births also were excluded if LMP was missing or the estimated birth based on LMP and gestational length was > 30 days from the midday of the birth month reported on the birth certificate.
Air pollution and weather data. PM2.5 chemical components data were obtained from the U.S. Environmental Protection Agency (EPA) Air Explorer (U.S. EPA 2010a). PM10 and PM2.5 total mass, CO, NO2, O3, and SO2 data were obtained from the U.S. EPA Air Quality System for 1999–2007 (U.S. EPA 2010b). We included only counties with PM2.5 chemical component data because these exposures are our primary focus. PM10, PM2.5, and PM2.5 chemical components were measured every 3–6 days. Gaseous pollutants were measured daily, although O3 was measured mainly during the warm season. Some monitors began or ceased observation during the study period. We investigated PM2.5 chemical components identified by previous research and literature review to have potential links to health and/or contribute substantially to PM2.5 total mass: aluminum, ammonium ion, arsenic, cadmium, calcium, chlorine, elemental carbon, lead, mercury, nickel, nitrate, organic carbon matter, silicon, sodium ion, sulfur, titanium, vanadium, and zinc (Bell et al. 2007a (link); Franklin et al. 2008 (link); Haynes et al. 2011 (link); Ostro et al. 2007 (link); Zanobetti et al. 2009 (link)).
We calculated apparent temperature (AT), a measure that reflects overall temperature discomfort (Kalkstein and Valimont 1986 ), based on daily temperature and dew point temperature data obtained from the National Climatic Data Center (2010) . If weather data were unavailable for a given county, we assigned the AT value for the closest county with weather data.
Exposure estimation. For each birth we calculated the average level of each pollutant during gestation and each trimester, and average AT during each trimester. Delivery date was estimated based on self-reported LMP and gestational length, assuming conception 2 weeks after LMP. We defined trimesters as 1–13 weeks, 14–26 weeks, and week 27 to delivery, as in previous studies (Bell et al. 2007b (link)).
Exposures were estimated based on county of residence. Not all counties had data for all pollutants. Measurements from multiple monitors in the same county on the same day were averaged to generate daily pollutant -levels. To avoid biases due to changes in measurement frequency, daily pollutant levels and AT values were combined to estimate weekly exposures, which were then averaged to estimate gestational or trimester exposure. Births for which exposure estimates were unavailable for > 25% of the weeks in any trimester for a given pollutant were excluded from analyses of that pollutant.
Statistical analysis. Each birth was cate-gorized as low or normal birth weight using clinically defined LBW (< 2,500 g). Logistic regression was used to estimate associations between LBW and gestational exposure to each pollutant with adjustment for maternal race (African American, Caucasian, other), marital status (married, unmarried), tobacco consumption during pregnancy (yes, no, unknown), alcohol consumption during pregnancy (yes, no, unknown), highest education (< 12 years, 12 years, 13–15 years, > 15 years, unknown), age (< 20, 20–24, 25–29, 30–34, 35–39, ≥ 40 years), infant sex (male, female), gestational length (37–38, 39–40, 41–42, 43–44 weeks), the trimester prenatal care began (1st, 2nd, 3rd, no care, unknown), first in birth order (yes, no, unknown), delivery method (vaginal, cesarean section, unknown), average AT for each trimester, season of birth, and year of birth. In addition we included regional indicators to adjust for local factors such as area-level socioeconomic conditions (
For pollutants showing statistically significant associations with LBW in single--pollutant models, we conducted two-pollutant models that included pairs of pollutants that were not highly correlated (correlation < 0.5). Similarly, for pollutants associated in single-pollutant model, we assessed effects by trimester using a model with trimesters’ exposures included simultaneously. Because trimester exposures could be correlated, we performed sensitivity analysis with trimester exposures adjusted to be orthogonal using a method we published previously (Bell et al. 2007b (link)). In brief, we predicted exposures of two trimesters using exposure level of a given trimester (reference trimester), calculated their residuals, and put them into models besides exposure of reference trimester. This approach can avoid covariance among trimester exposures. This procedure was repeated using each trimester as the reference trimester, and we have four models for trimester analysis in total (main model and three models as sensitivity -analyses). Further description of this approach is available elsewhere (Bell et al. 2007b (link)).
Additional analyses were conducted for pollutants that showed statistically robust results in two-pollutant models. We included interaction terms between gestational pollutant exposures and sex or race to investigate whether some populations are particularly susceptible, because previous analysis found higher relative risks associated with ambient air pollution in some populations than in others (Bell et al. 2007b (link)). Statistical significance was determined at an alpha level of 0.05 for the entire analyses.
Free full text: Click here
