Ozone and Preterm Birth in Women With Gestational Diabetes
Ozone and Preterm Birth in Women With Gestational Diabetes
We conducted a population-based case-control study to assess the association of air pollution with preterm birth and to see whether it varied by whether or not the woman experienced pregnancy complications. By law, all births in Taiwan must be reported to the Taiwan Birth Registry within 15 days of delivery. Almost 99% of pregnant Taiwanese women have free prenatal care, which includes at least 10 prenatal care visits during pregnancy that are covered by national health insurance. The Taiwanese birth registry is a valid source of data on preterm births, with a low rate of missing information (1.6%), high sensitivity and specificity (92.8% and 99.6%, respectively), and high reliability (Cohen's k statistic, 0.92).
In the birth registry, preterm birth is defined as a gestational age of less than 37 weeks based on routine ultrasound examination. The source population comprised 1,510,064 singleton births registered in the Taiwanese birth registry from 2001 to 2007. We excluded infants born with any birth defect (International Classification of Diseases, Ninth Revision, Clinical Modification codes 740–758; n = 9,073) and infants born to mothers with a chronic disease (chronic diabetes, chronic hypertension, or heart disease; n = 5,090). Birth defects and other health conditions were primarily diagnosed by physicians. We excluded births to women who smoked (n = 394), infants with a birth weight less than 500 g or greater than 5,000 g (n = 469), infants born at less than 20 gestational weeks or greater than 42 gestational weeks (n = 329), and those with a birthplace in an area in which air pollution monitoring data were missing (n = 15,074). The final study population used in the statistical analysis included 86,224 preterm birth cases and 344,896 randomly selected control subjects. The study was approved by the Institutional Review Board of the China Medical University Hospital, and it complied with the principles outlined in the Helsinki Declaration. Because the data were anonymized, the institutional review board specifically waived the need to obtain consent from each subject.
Data on 5 of the major air pollutants (ozone, carbon monoxide, nitrogen dioxide, particulate matter with an aerodynamic diameter of 10 μm or less (PM10), and sulfur dioxide) were obtained from 72 Taiwan Environmental Protection Agency air quality monitoring stations on Taiwan's main island. Data on air pollution concentrations are continually measured by fully automated monitoring stations on an hourly basis. Carbon monoxide is measured using nondispersive infrared absorption, nitrogen dioxide is measured using chemiluminescence, ozone is measured using ultraviolet absorption, PM10 is measured using β-gauge, and sulfur dioxide is measured using ultraviolet fluorescence.
The map coordinates of air monitoring stations and the data on air pollution were identified and managed using a geographic information system (ArcGIS, version 10; ESRI, Redlands, California). The air pollutant measurements were integrated into monthly data points and interpolated to surface-level data using an inverse distance weighting approach with suitable spatial resolution (100 m). For the inverse distance weighting approach, we used the inverse square distance method by using the 3 closest monitoring stations within 25 km of each grid cell to calculate the monthly mean concentration for each air pollutant. To obtain postcode-level pollutant concentrations, we integrated the monthly air pollution data with the postcode area for each grid cell and then assigned it to individual women using their own postcode number. The postal code was typically representative of 1 block in urban areas but was larger in rural areas.
The average concentrations were set as the daily maximum of the 8-hour period from 10:00 AM to 6:00 PM for ozone and the 24-hour average concentrations for carbon monoxide, nitrogen dioxide, PM10, and sulfur dioxide for the duration of pregnancies during 2000–2007. We averaged the air pollution data over the first trimester (1–3 months), second trimester (4–6 months), and third trimester (7 months to birth) of gestation based on the birth date and gestational age reported on the birth registry.
The covariates constructed from the routinely available birth registration data included maternal age (<20, 20–34, or >34 years), sex of the infant (male or female), season of conception (spring, summer, fall, or winter), and year of conception. We obtained salary data from the Bureau of National Health Insurance to calculate the average annual household income in each postcode area and assigned this to subjects by postcode. We stratified the socioeconomic status of each postcode area into quartiles: >75th percentile, 75th–50th percentile, 50th–25th percentile, and <25th percentile.
For the dichotomous outcome variable, we applied a logistic regression model with and without adjustment for the covariates to explore the association among air pollution, pregnancy complications, and preterm births in the 3 trimesters of pregnancy. We identified adjustment factors, including the covariates listed above, that were associated with preterm birth or air pollution based on χ tests and t tests. Further, we also selected adjustment factors a priori and included them in the final model.
Single-pollutant models were fitted to estimate the association of an individual pollutant with preterm birth. We also fitted multipollutant models to examine the association of ozone exposure with preterm birth while controlling for other pollutants. However, we did not include carbon monoxide and nitrogen dioxide data in the same model because of their high collinearity (r = 0.83).
The association of each pollutant with the risk of preterm birth was estimated as the odds ratio and 95% confidence interval per each 10-ppb change for ozone and nitrogen dioxide, per each 1-ppb change for sulfur dioxide, per each 100-ppb change for carbon monoxide, and per each 10-µg/m change for PM10. We used SAS, version 9.3 (SAS Institute, Inc., Cary, North Carolina) to perform all statistical analyses. Statistical significance was set at P < 0.05 based on a 2-sided test.
As a first step, we selected air pollutants that had significantly positive associations with preterm birth. We hypothesized that common pregnancy complications, such as gestational diabetes mellitus, gestational hypertension, and preeclampsia, could be worsened by exposure to air pollutants. We then performed a stratified analysis to explore the association between the pollutant and preterm birth by 2 levels of pregnancy complications (no disease and disease). Secondly, we created an interaction term (exposure × pregnancy complication) to add into the model and used Wald's method to assess the multiplicative interaction. We only assessed the interaction during the second and third trimesters because the general diagnostic periods for gestational hypertension and preeclampsia are after 20 weeks of gestation, and routine screening for gestational diabetes mellitus begins after 24–28 weeks of gestation.
Methods
Study Population
We conducted a population-based case-control study to assess the association of air pollution with preterm birth and to see whether it varied by whether or not the woman experienced pregnancy complications. By law, all births in Taiwan must be reported to the Taiwan Birth Registry within 15 days of delivery. Almost 99% of pregnant Taiwanese women have free prenatal care, which includes at least 10 prenatal care visits during pregnancy that are covered by national health insurance. The Taiwanese birth registry is a valid source of data on preterm births, with a low rate of missing information (1.6%), high sensitivity and specificity (92.8% and 99.6%, respectively), and high reliability (Cohen's k statistic, 0.92).
In the birth registry, preterm birth is defined as a gestational age of less than 37 weeks based on routine ultrasound examination. The source population comprised 1,510,064 singleton births registered in the Taiwanese birth registry from 2001 to 2007. We excluded infants born with any birth defect (International Classification of Diseases, Ninth Revision, Clinical Modification codes 740–758; n = 9,073) and infants born to mothers with a chronic disease (chronic diabetes, chronic hypertension, or heart disease; n = 5,090). Birth defects and other health conditions were primarily diagnosed by physicians. We excluded births to women who smoked (n = 394), infants with a birth weight less than 500 g or greater than 5,000 g (n = 469), infants born at less than 20 gestational weeks or greater than 42 gestational weeks (n = 329), and those with a birthplace in an area in which air pollution monitoring data were missing (n = 15,074). The final study population used in the statistical analysis included 86,224 preterm birth cases and 344,896 randomly selected control subjects. The study was approved by the Institutional Review Board of the China Medical University Hospital, and it complied with the principles outlined in the Helsinki Declaration. Because the data were anonymized, the institutional review board specifically waived the need to obtain consent from each subject.
Exposure Assessment
Data on 5 of the major air pollutants (ozone, carbon monoxide, nitrogen dioxide, particulate matter with an aerodynamic diameter of 10 μm or less (PM10), and sulfur dioxide) were obtained from 72 Taiwan Environmental Protection Agency air quality monitoring stations on Taiwan's main island. Data on air pollution concentrations are continually measured by fully automated monitoring stations on an hourly basis. Carbon monoxide is measured using nondispersive infrared absorption, nitrogen dioxide is measured using chemiluminescence, ozone is measured using ultraviolet absorption, PM10 is measured using β-gauge, and sulfur dioxide is measured using ultraviolet fluorescence.
The map coordinates of air monitoring stations and the data on air pollution were identified and managed using a geographic information system (ArcGIS, version 10; ESRI, Redlands, California). The air pollutant measurements were integrated into monthly data points and interpolated to surface-level data using an inverse distance weighting approach with suitable spatial resolution (100 m). For the inverse distance weighting approach, we used the inverse square distance method by using the 3 closest monitoring stations within 25 km of each grid cell to calculate the monthly mean concentration for each air pollutant. To obtain postcode-level pollutant concentrations, we integrated the monthly air pollution data with the postcode area for each grid cell and then assigned it to individual women using their own postcode number. The postal code was typically representative of 1 block in urban areas but was larger in rural areas.
The average concentrations were set as the daily maximum of the 8-hour period from 10:00 AM to 6:00 PM for ozone and the 24-hour average concentrations for carbon monoxide, nitrogen dioxide, PM10, and sulfur dioxide for the duration of pregnancies during 2000–2007. We averaged the air pollution data over the first trimester (1–3 months), second trimester (4–6 months), and third trimester (7 months to birth) of gestation based on the birth date and gestational age reported on the birth registry.
Covariates
The covariates constructed from the routinely available birth registration data included maternal age (<20, 20–34, or >34 years), sex of the infant (male or female), season of conception (spring, summer, fall, or winter), and year of conception. We obtained salary data from the Bureau of National Health Insurance to calculate the average annual household income in each postcode area and assigned this to subjects by postcode. We stratified the socioeconomic status of each postcode area into quartiles: >75th percentile, 75th–50th percentile, 50th–25th percentile, and <25th percentile.
Statistical Methods
For the dichotomous outcome variable, we applied a logistic regression model with and without adjustment for the covariates to explore the association among air pollution, pregnancy complications, and preterm births in the 3 trimesters of pregnancy. We identified adjustment factors, including the covariates listed above, that were associated with preterm birth or air pollution based on χ tests and t tests. Further, we also selected adjustment factors a priori and included them in the final model.
Single-pollutant models were fitted to estimate the association of an individual pollutant with preterm birth. We also fitted multipollutant models to examine the association of ozone exposure with preterm birth while controlling for other pollutants. However, we did not include carbon monoxide and nitrogen dioxide data in the same model because of their high collinearity (r = 0.83).
The association of each pollutant with the risk of preterm birth was estimated as the odds ratio and 95% confidence interval per each 10-ppb change for ozone and nitrogen dioxide, per each 1-ppb change for sulfur dioxide, per each 100-ppb change for carbon monoxide, and per each 10-µg/m change for PM10. We used SAS, version 9.3 (SAS Institute, Inc., Cary, North Carolina) to perform all statistical analyses. Statistical significance was set at P < 0.05 based on a 2-sided test.
As a first step, we selected air pollutants that had significantly positive associations with preterm birth. We hypothesized that common pregnancy complications, such as gestational diabetes mellitus, gestational hypertension, and preeclampsia, could be worsened by exposure to air pollutants. We then performed a stratified analysis to explore the association between the pollutant and preterm birth by 2 levels of pregnancy complications (no disease and disease). Secondly, we created an interaction term (exposure × pregnancy complication) to add into the model and used Wald's method to assess the multiplicative interaction. We only assessed the interaction during the second and third trimesters because the general diagnostic periods for gestational hypertension and preeclampsia are after 20 weeks of gestation, and routine screening for gestational diabetes mellitus begins after 24–28 weeks of gestation.