Low birth weight, defined as less than 2,500 grams (g) at birth, is associated with immediate and long-term health effects, including death. Low birth weight affects around 8% of all births in the United States (US) with over two-thirds of those cases associated with premature birth [1
]. For term births who result in low birth weight, fetal growth restriction is thought to be the major contributing factor, associated with a number of factors including smoking during the pregnancy, alcohol/drug abuse, birth defects, and certain socioeconomic factors [2
]. Long-term health effects of low birth weight include type-2 diabetes, high blood pressure, heart disease, hearing/vision problems, and intellectual disabilities.
Previous ambient air pollution/birth weight epidemiologic studies have focused on low birth weight as a binary variable as well as working with continuous birth weight directly. Common analyses involve calculating pollution exposures based on active pollution monitors near the residence at delivery of the mother. Trimester averages are the most common time period of interest in these studies [5
], while some studies have incorporated monthly exposure averages throughout the pregnancy [13
]. A recent literature review by Šrám et al. [15
] concluded that for birth weight, there was a need for future studies “to clarify the most vulnerable periods of pregnancy and the role of individual pollutants.” In this paper, we thoroughly investigate these vulnerable periods in terms of ambient ozone exposure.
Previous studies have examined a number of pollutants and time frames with varying results. Carbon monoxide exposure during multiple periods of the pregnancy was shown to adversely affect the birth weight of the child in multiple studies [5
], with most of the results indicating the first trimester as the most susceptible time. Sulfur dioxide was also consistently found to decrease birth weight in multiple studies during various trimesters and months of pregnancy [6
]. A number of studies have also uncovered an adverse association between total suspended particles and birth weight [6
]. The relationship between birth weight and ozone exposure is less clear.
In this paper, we utilize a Bayesian statistical model for low birth weight, which has the ability to more accurately identify critical periods during the pregnancy where increased exposure to ambient ozone concentrations significantly increases the probability of low birth weight of the child. This model was originally developed for a preterm birth analysis by Warren et al. [16
]. We allow for a more continuous form of pollution exposure throughout the pregnancy than typically considered in the low birth weight context through the use of weekly ambient averages. Working in the Bayesian setting allows us to properly characterize the uncertainty in our model parameters while also helping to control the multicollinearity introduced by jointly considering weekly effects in the model.
We begin the analysis by assigning ozone pollution exposures to each woman in a subdomain of Texas, based on residence at delivery, during each week of the pregnancy for three different pollution metrics. These metrics include the standard Air Quality System (AQS) data, the Community Multiscale Air Quality (CMAQ) chemistry model output, and the newly developed downscaler (DS) pollution model output. The AQS data are the most commonly used in epidemiologic studies and represent observed monitoring data across the state, while the CMAQ output is based on a deterministic model which relies on meteorological input and pollutant specific chemistry expertise. The DS process calibrates the CMAQ output by statistically combining it with the AQS data in order to provide estimates of the AQS observations in regions lacking spatial-temporal monitor coverage. We investigate the use of the CMAQ and DS metrics with respect to risk assessment estimation and compare their results with results seen using the AQS data.
This metric comparison analysis will help to inform future studies which may require the use of alternative pollution output in rural or undeveloped geographical regions of interest. The alternate pollution metrics have complete spatial-temporal coverage over a geographic domain of interest and can therefore be very useful in determining exposures for people in areas not represented by AQS pollution monitors. It is important to assess the impact these metrics have on risk assessment estimation in the environmental health settings, as they can lead to the analysis of populations of people who have previously been excluded from most typical analyses due to their large distances from a nearby monitor. Comparing the health effect estimation results from use of these metrics with respect to the AQS data results has not been considered previously in this setting but can potentially have a large impact on the direction of future research.
For each exposure dataset, we fit the continuous pollution exposure model, and critical weeks during the pregnancy are analyzed. We compare the results from each metric, and the similarities and differences between the identified critical periods of the pregnancy are discussed. Based on these results, we extend the dataset to incorporate women who did not reside near an active pollution monitor within the region through use of the DS output, which includes ozone estimates across the region in space and time from 2001–2004. Susceptible periods during the entire pregnancy are then analyzed using the full birth dataset from this region.
In Section 2
, we discuss the data used in the analysis, the statistical model, and the application method. The results and discussion are presented in Section 3
. We close in Section 4
with the conclusions.