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

Disclosures

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

In This Article

Results

Of the 410,267 women included in this study, 14,032 (3.4%) had GDM, including 406,334 with complete data for all covariates (n = 13,943 with GDM). Table 1 shows the distribution of exposures to PM2.5 and O3 for each pregnancy period analyzed in this study. Women with GDM had slightly higher levels of PM2.5 and O3 exposure compared with those without GDM during all pregnancy periods (all p < 0.001). Weak correlations were observed between PM2.5 and O3 in all gestational periods (Pearson's correlation coefficients range from 0.21 to 0.39).

Table 2 shows the demographic characteristics of women by GDM status. Women with GDM were older and less likely to belong to non-Hispanic black racial/ethnic categories. Higher proportions of women with GDM were married and had higher education and income levels. GDM cases were more likely among women who started prenatal care early and whose conception began in the warm season or recent years.

Table 3 provides the unadjusted and adjusted ORs of single-pollutant logistic regression models predicting GDM from exposure to PM2.5 and O3 during different pregnancy periods. After controlling for all nine covariates, increased odds of GDM for a 5-μg/m3 increase in PM2.5 were observed during both the first and second trimesters (ORTrimester1 = 1.16; 95% CI: 1.11, 1.21; ORTrimester2 = 1.15; 95% CI: 1.10, 1.20); and the full pregnancy (OR = 1.20; 95% CI: 1.13, 1.26). Associations were also found between GDM and O3. The odds of GDM were higher for a 5-ppb increase in exposure to O3 during the first and second trimesters (ORTrimester1 = 1.09; 95% CI: 1.07, 1.11; ORTrimester2 = 1.12; 95% CI: 1.10, 1.14), and over the course of the entire pregnancy (OR = 1.18; 95% CI: 1.15, 1.21).

The results from the sensitivity analyses are presented in the Supplemental Material http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf. Specifically, multiple imputation was conducted in the first set of sensitivity analyses to assess the potential effects of missing data on the results, and we observed ORs almost identical to the original results (see Supplemental Material, Table S1 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf). Second, the Monte Carlo method was used to generate two sets of simulated data sets assuming the underreported rate of GDM was 0.5% and 1.0%. Compared with the original results, the ORs from the simulated data sets slightly attenuated, but the conclusions remain consistent (see Supplemental Material, Table S2 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf). Third, we examined the effects of potential misclassifications of exposure on the results separately using capture-area analyses and the interpolated 1-km × 1-km HBM data. Compared with the original results, we observed comparable ORs for O3 during the second trimester and PM2.5 during the second trimester and full pregnancy period in the capture-area analyses. However, attenuated ORs were observed for O3 during the first trimester and the full pregnancy period, and no significant association was found for PM2.5 in the first trimester. On the other hand, the results from the interpolated HBM in the 1-km × 1-km resolution showed consistent ORs with the original results (see Supplemental Material, Table S3 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf). Fourth, we assessed whether adjusting for smoking during pregnancy may bias the findings, and we observed consistent ORs with the original results. We also analyzed the data without adjusting for season of conception, and consistent results were observed except for the slightly attenuated OR for O3 in the first trimester (see Supplemental Material, Table S4 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf). Last, a stratified analyses by urbanization was performed to examine the potential overadjustment of it, and no statistically significant difference was observed between the nonstratified results and the stratified results (see Supplemental Material, Table S5 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf).

The results of the co-pollutant models are provided in Supplemental Material, Table S6 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408456.s001.acco.pdf. Figure 1 compares the results obtained from single- and co-pollutant continuous models. The ORs for O3 after adjusting for PM2.5 were almost identical to the ORs from the single-pollutant model. However, the ORs for PM2.5 during the first trimester and the full pregnancy attenuated after adjusting for O3, and no association was observed for PM2.5 during the second trimester in the co-pollutant model (OR = 1.02; 95% CI: 0.98, 1.07 compared with OR = 1.15; 95% CI: 1.10, 1.20 from the single-pollutant model).

Figure 1.

Adjusted log(OR) for risk of GDM with per 5 units increase in gestational exposure to pollutant for single- and co-pollutant models among women who gave birth in 2004–2005 in Florida, USA. Diamonds reflect the central estimate; whiskers represent the 95% CIs.

processing....