Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Epidemiology and Biostatistics

Major Professor

Kathleen O’Rourke, Ph.D.

Committee Member

Russell S. Kirby, Ph.D.

Committee Member

Amy L. Stuart, Ph.D.

Committee Member

Alfred K. Mbah, Ph.D.

Committee Member

Jason L. Salemi, Ph.D.


air pollution, birth defects, exposure assessment, benzene, particulate matter


In the United States, approximately 3% of infants are diagnosed with a major birth defect each year. Whereas prevention efforts have led to decreases in some birth defects, the cause of approximately 70% of birth defects remains unknown. More recently, there has been increasing concern regarding exposures to environmental agents, such as air pollution during pregnancy, and the risk of birth defects. Over the past decade, there has been an increase in research studies examining the association between air pollution and birth defects. The results have been inconsistent, with some studies reporting that higher levels of exposure from the same pollutant increase the risk of birth defects, whereas others report no risk or even a decrease in risk. These inconsistencies may arise from differences in exposure assessment methods across epidemiological studies.

A comprehensive series of sensitivity analyses were conducted to evaluate the impact of different exposure estimation decisions on the adjusted measures of associations between ambient air pollution exposure and birth defects. For each of six decisions – spatial scale, spatial limit, temporal scale, temporal limit, data aggregation sequence, and weighting scheme – two alternative choices were considered, and maternal exposure to PM2.5 and benzene were estimated for each ‘sensitivity case' (unique combination of choices across all decisions). For each sensitivity case, adjusted prevalence ratios (aPRs) and 95% confidence intervals (CI) were estimated from multivariable regression models comparing each exposure quartile to the first quartile; each measure of association represented the associations between either benzene or PM2.5 and one of the birth defects under study. To then evaluate the impact of selecting alternative choices for each exposure assessment decision, three metrics were used: 1) the percent difference in the aPR point estimates between alternative choices for a specific decision (calculated for each sensitivity case pairing); 2) the percent of sensitivity cases in which the aPR 95% CIs for alternative choices for a specific decision overlapped (i.e., were not statistically significantly different); and 3) the percent of sensitivity cases in which there was agreement in the substantive conclusion of the association between exposure (pollutant) and outcome (defect), the conclusion being either statistically significant increased risk, statistically significant decreased risk, or no statistically significant difference in risk.

Second, a retrospective cohort study was conducted to investigate the association between maternal exposure to ambient benzene and PM2.5 and the risk of musculoskeletal birth defects in offspring. For both analyses, data on singleton infants born from 2000 to 2013 were obtained from the Florida Birth Defects Registry and air pollution data were obtained from the Environmental Protection Agency Air Quality System database. Exposure estimates were calculated for all birth defect cases and non-affected births during etiologically relevant time windows and multivariable regression models were used to obtain aPRs and 99% CIs comparing each quartile of exposure to the first.

For PM2.5, across all exposure-outcome pairs under study, the alternative choices for the decision of spatial scale resulted in the largest median percent difference in aPRs (e.g., between county and block group) when results were aggregated for all quartiles. This was followed by the temporal scale and spatial limit decisions. For benzene, spatial limit resulted in largest median percent difference in aPRs, followed by spatial scale and temporal scale. However, for both pollutants, when evaluating the agreement in aPRs between alternative choices on the direction and significance of the association (i.e., statistically significant increase, decrease, or no risk), the decision of spatial limit resulted in the lowest percent agreement (biggest impact). Temporal limit had the second lowest percent agreement for PM2.5, whereas spatial scale had the second lowest for benzene. Finally, the metric assessing the level of overlap in 95% CIs for the measures of association was inconsequential, suggesting little impact of any exposure assessment decision, and doing little to differentiate between the relative impacts of each decision. That is because, for all exposure-outcome pairs, when comparing alternative choices for each decision, there was 100% overlapping intervals (i.e., no aPRs were statistically significantly different from each other). Lastly, exposure assessment decisions impacted the analytic sample sizes, with some decisions resulting in a three-fold difference in the sample size alone.

Mothers of singleton infants exposed to higher levels of ambient PM2.5 were more likely than mothers with lower exposure levels to give birth to an infant with isolated anomalies of the skull and face bones, any rib and sternum anomalies, any skull and face anomalies, any spine anomaly, and other congenital anomalies of lower limb including pelvic girdle. Higher PM2.5 exposure was also associated with an increased risk of non-isolated anomalies of skull and face bones, any skull and face anomalies, and reduction deformities of the upper limb. Exposure to higher levels of benzene was associated with an increased risk of isolated congenital hip dislocation and congenital valgus deformities of feet as well as multiple inverse associations.

The findings presented here indicate that measures of association between maternal pollutant exposures (PM2.5 and benzene) and selected birth defect outcomes in offspring are sensitive to exposure assessment decisions, with some decisions more impactful than others. The findings can be used, not only to explain the lack of consistency in results across existing epidemiological studies, but to guide decision-making in future studies. This study also adds to the growing body of epidemiological studies suggesting an association between specific air pollutants and birth defects. In the current political climate, it is important that researchers continue to provide evidence of the detrimental health effects of air pollution in order to circumvent change in current policies established to regulate and reduce pollution emissions.

Available for download on Wednesday, August 01, 2018

Included in

Epidemiology Commons