Association of Atmospheric Particulate Matter and Ozone With Gestational Diabetes Mellitus

Hui Hu; Sandie Ha; Barron H. Henderson; Tamara D. Warner; Jeffrey Roth; Haidong Kan; Xiaohui Xu


Environ Health Perspect. 2015;123(9):853-859. 

In This Article


We examined the association of GDM with PM2.5 and O3 during different pregnancy periods using Florida birth vital statistics records and the U.S. EPA and CDC's HBM air pollution data, which have both good spatial and temporal coverage. When assessed in single-pollutant models, GDM was significantly associated with per 5-unit increases in both PM2.5 and O3 during the first and second trimesters and the full pregnancy. The associations were also found in co-pollutant models for PM2.5 exposure during the first trimester and O3 exposure during all pregnancy periods we examined. The associations persisted with adjustment for confounding by maternal characteristics such as age, race/ethnicity, education, marital status, prenatal care, season and year of conception, urbanization, and median household income at census block group level. The results of this study add to the emerging evidence linking air pollution exposure during pregnancy to pregnancy complications such as GDM.

The causal mechanisms underlying the associations between air pollution and GDM are still unclear; however, the results observed in this study are consistent with several potential pathways suggested by previous studies. Ambient air pollutants such as PM and O3 have been reported to be associated with increased insulin resistance, dyslipidemia, and systemic metabolic dysfunction (Andersen et al. 2012; Chuang et al. 2011; Coogan et al. 2012; Kelishadi et al. 2009; Kim and Hong 2012; Krämer et al. 2010; Puett et al. 2011; Sun et al. 2013), which are all precursors associated with GDM. PM contains many toxic chemicals that are regarded as reactive oxygen species (ROS) (Lemaire and Livingstone 1997; Sun et al. 2006), which can cause oxidative damage on target tissues (Ames et al. 1993). The imbalance between the production of ROS and antioxidant defenses is acknowledged as one of the main causes of insulin signaling–pathways alterations (Lamb and Goldstein 2008), and a number of studies have linked ROS to insulin resistance (Goldstein et al. 2005; Schulz et al. 2007). In addition, a recent animal study also showed O3's ability to induce glucose intolerance and systemic metabolic effects (Bass et al. 2013). In their study on young and aged Brown Norway rats, Bass et al. (2013) observed increased α2-macroglobulin, adiponectin, and osteopontin as well as decreased phosphorylated insulin receptor substrate-1 in liver and adipose tissues following acute O3 exposure. Endoplasmic reticular stress was suggested to be the consequence of O3-induced acute metabolic impairment. Furthermore, another potential pathway induced by air pollution is inflammation, which may also lead to the development of insulin resistance (Everett et al. 2010; Hotamisligil et al. 1993).

Cigarette smoking has been widely reported to be associated with type 2 diabetes (Willi et al. 2007; Zhu et al. 2014), and we initially considered it as a potential confounder in our analyses. However, given the fact that smoking is not generally considered a risk factor for GDM as well as the consistent results we observed with or without adjusting for it in the sensitivity analyses, we finally present results without adjusting for smoking. In addition, although the underlying mechanisms remain unknown, our findings that air pollution may have an impact on risk of GDM does not conflict with the null association between smoking and GDM because their toxic components are largely different.

Our study has several strengths. First, compared with the air monitoring data that have been widely used in other studies, the daily temporal resolution and the 12-km × 12-km spatial resolution of HBM air pollution data used in this study allowed us to estimate mean air pollution concentrations during different pregnancy periods without excluding subjects not covered by air monitors, thus reducing the potential for selection bias. Second, previous studies focused only on small areas and examined limited types of air pollutants. With the HBM air pollution data, we were able to include all pregnant women in the study period throughout the entire state of Florida and investigate the association between GDM and two common air pollutants, PM2.5 and O3, which have not been reported in the extant literature. Furthermore, we used both single- and co-pollutant models to examine the association between air pollution and GDM. The robust results of O3 observed from different models suggest that it may have effects on GDM independent of PM2.5. This finding is consistent with recent experimental studies (Bass et al. 2013). It is also consistent with the positive association found between NOx (nitrogen oxides) and GDM (Malmqvist et al. 2013) because NOx is one main precursor of O3 (Sillman 1999). Finally, the robust results from the sensitivity analyses suggested that the study was not likely to be largely biased by the missing data, exposure and outcome misclassifications, and underadjustment of smoking during pregnancy or overadjustments of season of conception and urbanization.

This study had several limitations. First, it is possible that GDM may be underdiagnosed in the source vital statistics records. Second, as reported by the American Diabetes Association (2013), more women of childbearing age have type 2 diabetes due to an epidemic of obesity and diabetes in recent years. This trend may result in an increase in the number of women with undiagnosed type 2 diabetes, leading to potential misclassification of GDM in this study. However, because our study period covered the years 2003–2005, our results are less likely to be biased by the effects of undiagnosed diabetes in recent years. Third, information on daily mobility and behavior patterns was not available for this study. The absence of these factors may introduce misclassifications of exposure. A high correlation between personal monitored air pollution measurement and monthly aggregated modeled air pollution measurement has been reported in a cohort of 85 pregnant women in Manchester and Blackpool, United Kingdom (Hannam et al. 2013), although we cannot assess its comparability to our study due to the lack of daily mobility data. Fourth, residential mobility during pregnancy was also not available in this study. It may be possible that some subjects in this study lived elsewhere in the early stage of their pregnancy and thus were exposed to different levels of air pollution. Fifth, although the use of HBM air pollution data can avoid selection bias, the 12-km × 12-km resolution is very crude. Although the spatial variability of O3 is low, the variability of PM2.5 may be a concern, which includes a large-scale regional component and a local source component. Isakov et al. (2012) suggested that the regional component provides most of the mass, going as far as to use PM2.5 as an example of spatially homogeneous pollutants. Therefore, exposure to PM2.5 is not likely to have extremely fine-scale variability in most places in Florida. In addition, highly variable exposure fields would also be inappropriate for use with residential address only. However, future studies with higher spatial resolution modelling data and detailed time–activity patterns are warranted. Sixth, although several important confounders have been included in this study, no information on such other risk factors for GDM as prepregnancy BMI, family history of type 2 diabetes, and physical activity was available. These unadjusted factors may influence the results. For example, if obese women are more likely to live in areas with higher air pollution, the observed effects of air pollution on GDM in this study may be overestimated without controlling for this factor. In addition, low population densities, poor street connectivity, and lack of sidewalks in rural areas have been linked to increased physical inactivity and obesity (Eberhardt and Pamuk 2004), which are also characterized by having higher O3 concentrations. Although we adjusted for urbanization in this study, residual confounding may still exist. Thus, future studies with more detailed information on these factors are warranted to confirm our findings. Another potential limitation of the study is the unavailability of traffic noise data. Traffic noise induces a stress response and disturbs sleep, which has been associated with higher levels of stress hormone and decreased insulin levels and sensitivity (Sørensen et al. 2013). Both maternal stress and/or disturbances of sleep during pregnancy increase the risk of GDM. Because road traffic is the main source for both air pollution with PM2.5 and noise in urban areas, the mutual confounding is a concern. Finally, the results observed in birth registry data may also be influenced by the fixed cohort bias (Strand et al. 2011). Fixed cohort bias is a type of selection bias that could happen in retrospective cohorts with a fixed start and end date when short pregnancies are missed at the start of the study, and longer pregnancies are missed at the end. Because GDM is linked to preterm birth, fixed cohort bias may exist if GDM cases are more likely to be excluded at the beginning and to be included at the end of the study. However, given the facts that fixed cohort bias tends to decrease when the study has longer study period and/or when it has a day and month of the start date (i.e., 1 January 2004) just before day and month of the end date (i.e., 31 December 2005), the potential for this bias was reduced in this study.