Grantee Research Project Results
2009 Progress Report: Research Project A: Mapping Disparities in Birth Outcomes
EPA Grant Number: R833293C001Subproject: this is subproject number 001 , established and managed by the Center Director under grant R833293
(EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
Center: The Center for Study of Neurodevelopment and Improving Children's Health
Center Director: Murphy, Susan K.
Title: Research Project A: Mapping Disparities in Birth Outcomes
Investigators: Miranda , Marie Lynn , Gelfand, Alan , Swamy, Geeta , Maxson, Pamela , James, Sherman
Current Investigators: Miranda , Marie Lynn , Gelfand, Alan , James, Sherman , Swamy, Geeta , Maxson, Pamela
Institution: Duke University
EPA Project Officer: Callan, Richard
Project Period: May 1, 2007 through April 30, 2012 (Extended to April 30, 2014)
Project Period Covered by this Report: May 1, 2009 through April 30,2010
RFA: Centers for Children’s Environmental Health and Disease Prevention Research (2005) RFA Text | Recipients Lists
Research Category: Human Health , Children's Health
Objective:
The central objective of this project (R833293C001) is to determine whether and to what extent joint exposures to socioeconomic and environmental stressors contribute to racial and ethnic health disparities in fetal growth restriction.
Using a geographically based nested study design moving from analysis of births for the entire State of North Carolina to six demographically and geographically distinct counties to a single health center and state-of-the-art geographic information systems (GIS) applications with Bayesian spatial hierarchical modeling and other advanced spatial statistical approaches, the specific aims are to:
- Spatially link detailed birth record, fetal death certificates, socioeconomic, environmental, tax assessor, community-based, and clinical obstetric data at highly resolved scales for the State of North Carolina from 1990-2003;
- Refine the concept of fetal growth restriction by a) developing a joint distribution for birthweight and gestation using bivariate modeling for live births and fetal deaths – both separately and jointly, and b) defining it in terms of fetal and infant mortality, rather than percentile cut points; and
- Determine whether and to what extent differential exposures to both environmental and social stressors help explain health disparities in fetal growth restriction among a) African-American women compared to non-Hispanic white and Hispanic women, b) older African-American women compared to younger African-American women, c) Hispanic women compared to non-Hispanic white and African-American women, and d) foreign born Hispanic women compared to U.S. born Hispanic women.
This project evaluates a large number of factors in diverse populations, providing broad relevance for birth outcomes across time, space, and demography. Identifying social and environmental factors contributing to fetal growth restriction will improve our understanding of disease etiology and explain the racial disparity in disease incidence, leading to effective interventions against poor outcomes in all population groups. Of note, we have expanded our inquiry beyond fetal growth restriction to encompass a broader range of pregnancy outcomes.
Progress Summary:
Over the past year, the research team for Project R833293C001 has met both at full group level and in small groups to discuss new research ideas, review progress of current analysis, identify next steps, and work on manuscript preparation.
Air Pollution. We have spent considerable time linking the detailed birth record data to USEPA PM10, PM2.5, and ozone monitoring data in order to study the impact of maternal exposure to air pollution on birthweight. Initial work built customary regression models to assess the linkage and has been published in the Journal of Exposure Science, Environment, and Epidemiology. We are especially focused on refining exposure metrics to most effectively characterize meaningful exposures, as well as to capture any windows of vulnerability. Spatial and non-spatial hierarchical models have been explored, incorporating uncertainty in exposure as a function of distance from the nearest monitoring station. An associated manuscript has been submitted for publication.
Related work has studied the use of a PM2.5 exposure simulator to explain birthweight. In a recently submitted paper, a template is developed for using an environmental dose simulator to connect ambient exposure to personal exposure. Then, using various exposure metrics, calculated from these personal exposures that are clinically plausible over the course of a pregnancy, linkage is built to adverse birth outcomes. This work is forthcoming in Environmetrics.
In an extension of our previous work linking air quality monitoring data for PM10 and PM2.5 with the detailed birth record, we have been working on an analysis of the relationship between maternal exposure to particulate matter during pregnancy and the occurrence of pregnancy-induced hypertension. We have found that among women living within 20km of a monitor, higher levels of pregnancy-averaged PM10 and PM2.5 were associated with an increased risk of pregnancy-induced hypertension. This work, for which we are collaborating with investigators at the EPA, was presented at the Society for Epidemiologic Research in June 2010 and a manuscript currently is in preparation. In Year 4, we plan to extend this work to other criteria air pollutants and to include additional adverse birth outcomes.
As part of our larger efforts exploring the relationship of air pollution exposure and pregnancy outcomes, we sought to consider a relatively simple metric for assessing risk of exposure to air pollution, specifically traffic-related air pollution which includes particulate matter and diesel exhaust, both of which are being investigated within Project R833293C003. We utilized the statewide GIS layer of street-geocoded 2005-2007 births to calculate the proximity of each geocoded birth to the nearest primary and secondary roadway. While controlling for all standard covariates, we incorporated measures of air pollution exposure as dichotomous variables indicating residence within 500, 250, 150, 100, or 50 m of a primary or secondary roadway into models for birthweight, low birth weight (LBW), very low birth weight (VLBW), preterm birth (PTB), very preterm birth (VPTB), and any hypertensive disorder. Our findings, which are presented in a manuscript that will be submitted in the coming year, indicate a significant dose-response relationship between proximity to a primary or secondary road and the adverse outcomes of PTB, VPTB and hypertension—for example, the probability of hypertension is increased by living within 500 m of a primary or secondary roadway, with this probability being even higher at 250 m, and still higher at each of 150, 100, and 50 m.
Work continues on building spatial downscalers. Such modeling strategies enable the fusion of monitoring station data with computer model output to better assess environmental exposure. Then, we can utilize improved exposure assessment to examine linkage between exposure and adverse birth outcomes. A first paper on univariate exposure has been accepted at the Journal of Agriculture, Biological and Environmental Statistics. A follow-on paper considers downscaling for co-pollutants and reveals the benefits of studying exposures jointly. This work is forthcoming in the Annals of Applied Statistics. Recognizing the potential for “modeling” error in computer model output, we are completing a third manuscript on enabling local directionality in spatial structure to recalibrate the model output as well as to improve the fusion with station data.
Racial Residential Segregation. Our project on racial residential segregation now has seen the near completion of one paper (currently in preparation), which enables quantification of racial exposure/isolation at finer spatial scales within SMSAs. Such a measure can be connected to measures of social and economic disadvantage at these scales to gain insight into how racial residential segregation has manifested itself across urban landscapes. In turn, this promises to reveal key insights into how to think about the spatial aspects of the social factors influencing health disparities. We are working to determine which facets of segregation best characterize the way community-level racial residential segregation acts to promote health disparities in birth outcomes. Although our initial efforts were statewide, we have since decided that, given the significantly more detailed data available for Durham County, we will focus on this area while we work to determine what variables are most important to characterizing racial residential segregation in terms of its health consequences.
Nulliparous Women. In Year 3, we submitted a manuscript describing analysis exploring the observed association between parity and risk of adverse birth outcomes (i.e., women having their first child are at increased risk of adverse outcomes compared to women who already have had at least one child). We linked births in the North Carolina Detailed Birth Record 1990-2007 with previous and subsequent births to the same mother using deterministic techniques that evaluated various combinations of maternal identifying variables to link births, including full name, maiden name, date and state of birth, parity, and date of last birth. We employed statistical and modeling-based analyses to estimate first birth outcome rate differences between nulliparas who did have a subsequent pregnancy versus those who did not. Among nulliparas that were not linked to a second birth, maternal-age-adjusted rates of multiple measures of adverse outcomes, including maternal medical complications, were almost all statistically higher compared to rates for linked women. This work suggests that the observed differences in rates of adverse outcomes between nulliparas and multiparas are partly attributable to higher risk women not having a subsequent pregnancy (either by choice or due to fecundity differences).
Community Assessment Project/Built Environment. Analyses of the built environment data are under way. Seven scales (housing damage, property disorder, security measures, tenure, vacancy, violent crime, and nuisances) have been constructed at five levels of geography (census block, primary adjacency neighborhood, census block group, census tract, and city-defined neighborhoods). The continuous and categorical scale variables have been merged with the Durham birth records. A comparison of the results obtained from models comparing the relationship between the built environment scales and preterm birth at two distinct units of aggregation (the city-defined neighborhoods and census block groups) will be presented at the Society for Epidemiologic Research in June 2010. This work will be expanded to consider all five units of aggregation early in Year 4. Using the built environment scales constructed at the primary adjacency neighborhood unit, a paper assessing the associations between these scales and five birth outcomes (PTB, LBW, small for gestational age, continuous birth weight, and birth weight percentage for gestational age) has been drafted. It will be submitted early in Year 4.
Seasonality. We have examined the relationship between seasonality and pregnancy outcomes. Our initial aspatial models indicated that the effect of season was most apparent among non-Hispanic white women. We currently are working on spatial models to better understand what factors of season of conception or birth are influencing pregnancy outcomes.
Environmental Contributions to Disparities in Pregnancy Outcomes. We published a review article on social and environmental contributors to disparities in birth outcomes based on both national and North Carolina data, as a way of compiling the literature we have accessed throughout our work on this project. The manuscript, published in Epidemiologic Reviews, reviews research on how environmental exposures affect pregnancy outcomes and how these exposures may be embedded within a context of significant social and host factor stress.
Racial Disparities in Maternal Hypertensive Disorders. We analyzed data from North Carolina to determine how the pattern of maternal hypertensive disorders differs among non-Hispanic white, non-Hispanic black, and Hispanic women across the range of maternal ages. In addition we explored whether rates of poor birth outcomes, including LBW and PTB, among hypertensive women differed by race. This manuscript is forthcoming in Public Health Reports in 2010.
Maternal Age and Birth Order. Investigations of maternal age, birth order, and birthweight have not delineated the relative contributions of each factor to birthweight, especially as they may differ by race. Using the NC DBR data from 1999-2003, we modeled maternal age and birth order on birthweight, adjusting for infant sex, education, marital status, and race. Birth order exerts greater influence on birthweight than maternal age, with significantly different effects across racial subgroups. A manuscript on this work is in submission.
Statistical Methods Development. Out of efforts to develop new spatial methodologies for addressing health disparities, additional methodological work on disaggregated spatial modeling for areal unit categorical data went forward. This work uses innovative statistical methodology that extends spatial disease mapping techniques to model subgroups within areal units using a spatially smoothed, multilevel loglinear model. This work appeared in the Journal of the Royal Statistical Society, Series C. An attractive feature of this methodology for public health applications is the possibility to elucidate health disparities across space, across subgroups, and space-subgroup interactions.
Bivariate Normal Mixture Models. Another completed manuscript builds joint models for birthweight and gestational age using bivariate normal mixtures. Such joint modeling adjusts for maternal risk factors and provides mixture analysis of the residuals to help illuminate further subpopulations with differential risk for adverse joint birth outcomes. It also avoids potential causal inference issues. Modeling of the mixture components is done through gestational age and then birthweight given gestational age. This work is forthcoming in Statistics and Medicine. There is continuing work on this project. Our initial effort was non-spatial, ignoring the geo-coded locations of the births. It also was atemporal, ignoring the year of the birth, and it addresses only individual level risk factors but no environmental risk factors. Also, all of this work was confined to finite bivariate normal mixture models. Through the use of Dirichlet process mixed models, we can allow the data to inform regarding the number of mixture components, as well as associated clustering of births.
Spatial Quantile Regression. Novel work on spatial quantile regression has made substantial progress. We want to understand how dependence of response (birth weight) varies with quantile. Does the regression on median or mean birthweight look different from that for the 0.1 quantile? This is important given the greater interest in explaining low birthweight rather than in explaining average birthweight. Also, we expect quantile regressions to be more similar to each other when they are closer spatially than when they are farther apart. This requires the development of spatial quantile processes. Two manuscripts are in development summarizing this work.
Flexible Bayesian Spatial Discrete-time Survival Model. In addition, we have developed a flexible Bayesian spatial discrete-time survival model to estimate the effect of environmental exposure on the risk of preterm birth. We view gestational age as time-to-event data where each pregnancy enters the risk set at a pre-specified time (e.g., the 32th week). The pregnancy then is followed until either (1) a birth occurs before the 37th week (preterm); or (2) it reaches the 37th week and a full-term birth is expected. As preliminary analysis, the methodology was applied to a dataset of geo-coded births in North Carolina in 2002. We estimated the risk of preterm birth associated with short-term exposure to fine particulate matter using air quality metrics derived from the EPA’s Statistically Fused Air Pollution Database. We also conducted a simulation study and compared the proposed approach to the standard case-control and time series design. Two associated manuscripts are in preparation.
Statistical Methods for Multivariate Spatial Data Measured on Different Scales. We currently are developing multivariate spatial models for birth outcomes measured on different quantitative scales. These outcomes include continuous variables, such as birthweight and gestational age; categorical variables, such as preterm birth and small for gestational age (SGA); and zero-inflated count variables, such as occurrences of medical complications. The multivariate approach allows us to explore geographic variation among several variables at once, rather than focusing on one variable at a time. Moreover, by taking into account the correlation between various outcomes, multivariate models improve the precision of regression estimates and the ability to detect exposure effects. As part of our analysis, we are considering a variety of multivariate conditionally autoregressive (CAR) models, including multivariate intrinsic CAR models, multivariate proper CAR models, finite mixtures of CAR models, and Dirichlet process CAR models. This last approach accommodates uncertainty in the underlying CAR specification for the spatial random effects. In Year 4, we plan to apply these models to the Detailed Birth Record (DBR) data, as part of an ongoing effort to examine ways in which environmental, host, and psychosocial factors affect birth outcomes across different regions of North Carolina.
Collaborations with other SCEDDBO Components
We have worked closely with the investigators for Project R833293C003 to design analysis looking at the same pollutants at comparative levels of exposure from different methodological perspectives. Our discussions with the investigators of Project R833293C003 help inform our methods for framing ozone and particulate matter exposures in our models, as well as help refine the planning and implementation of future animal models in Project R833293C003. As the dataset being collected under Project R833293C002 reaches a size and completeness suitable for analysis, we plan to bring some of the methodological strategies developed under Project R833293C001 to this dataset including synthesis with the Detailed Birth Record data, the mixture modeling for birthweight and gestational age, and the refined environmental exposure approaches. Arguably, synthesizing the DBR data with the clinical OB database is our biggest challenge. We understand the issues here - the clinical OB births are contained in the DBR births but are not sampled randomly; the clinical OB dataset is much more complete but, due to the biased sampling of this dataset, relationships within this dataset (joint probabilities of events) will not necessarily extrapolate to the DBR data. However, with knowledge regarding the nature of the bias in recruitment for the clinical OB dataset, we will attempt to revise these joint probabilities to be applicable to the DBR data. This effort extends across both Projects R833293C001 and R833293C002.
Meanwhile, we have begun examining residential mobility by linking women who are in both Projects R833293C001 and R833293C002. This linkage will allow us to determine who is moving during pregnancy (by comparing the address at enrollment and the DBR address at delivery) and the nature of those moves, including the quality of the new location compared to the previous location (and thus changes in environment or exposure).
Future Activities:
We plan to continue working on each of the areas described in the progress report's summary of accomplishments section. Achieving a better understanding of exposure to air toxins, as well as particulate matter and ozone, is a central focus of our future efforts. Areas of investigation will include space time analysis of trends in births across North Carolina, an investigation of linked births (same mother) using suitable random effects models, and a more thorough investigation of the impact of introducing spatial random effects in regression modeling to explain birth outcomes.
Community assessment/built environment. We plan on continuing our analyses in Year 4 to further explore the impact of the built environment on birth outcomes and to examine associations between the built environment and racial residential segregation.
Residential Mobility. As indicated above, we recently began the process of linking participants in Project R833293C002 with their associated birth certificate record. We are excited to begin exploring the additional insights into the detailed birth record data that can be gleaned by linking these data with the rich dataset collected in Project R833293C002. This linkage will not only allow us to explore issues of data accuracy in the detailed birth record, but will also allow us to begin implementing the methods of synthesizing categorical data discussed above.
Survival Approach with Air Pollution and Preterm Birth. We plan to examine the risks of preterm birth due to long-term and short-term exposures to ambient fine particulate matter and its chemical constituents. We will apply the survival approach to geo-coded births in North Carolina for the period 2001 to 2005 separately for each county. County-specific risks will be combined via a two-stage Bayesian hierarchical model where effect modification due to county characteristics also will be explored. Moreover, we plan to extend the analysis to (1) account for exposure measurement error, (2) differentiate spontaneous and medically indicated preterm births, and (3) jointly model gestational age and birth weight.
Spatial Analysis. We have made one of our primary objectives for Year 4 the development of spatial analogue models for each of the non-spatial analyses we carry out. The motivation for this is to test whether parameter estimates derived from non-spatial models are biased due to spatial autocorrelation in standard regression model error terms. Spatial analysis will account for unobserved characteristics related to the study outcome that may be spatially patterned.
The specific spatial model, known as the Conditional Autoregressive (CAR) model, that has been our focus assumes that study outcomes in neighborhoods that share a boundary with each other are not spatially independent. In effect, we are able to estimate a spatial random effect associated with each neighborhood unit, where the value of this effect is equal to the mean value over adjacent areal units. Estimation also provides us with a spatial variance, which speaks to overall variability in the outcome across the study region. This spatial approach not only has the potential to improve parameter estimates but also allows us to identify areas with increased risk of a poor outcome.
We currently are developing a spatial model that mimics our seasonality analysis and have plans to do the same for other non-spatial analyses currently under way.
Racial Residential Segregation. Several analyses are planned for Year 4 including: (1) determining whether the effect of neighborhood level racial isolation on birth outcomes varies by the geographic scale used to proxy a neighborhood unit; (2) investigating whether the racial isolation effect varies by maternal SES factors such as education level and marital status; (3) extending the black-black isolation and white-white isolation analyses to a multiple county analysis, in order to examine the robustness of the association across study areas; and (4) determining whether neighborhood scale racial isolation relates to maternal level psychosocial factors (e.g., depression and other mental well-being indicators), which may then in turn impact birth outcomes.
Dissemination. We continue to target various professional audiences for dissemination of our work. Recent presentations have been at conferences under the auspices of the Joint Statistical Meetings, the American Public Health Association, the Society of Epidemiological Research, the International Biometric Society, and the Society of Maternal and Fetal Medicine.
Journal Articles on this Report : 6 Displayed | Download in RIS Format
| Other subproject views: | All 66 publications | 37 publications in selected types | All 36 journal articles |
|---|---|---|---|
| Other center views: | All 163 publications | 77 publications in selected types | All 76 journal articles |
| Type | Citation | ||
|---|---|---|---|
|
|
Berrocal VJ, Gelfand AE, Holland DM. A spatio-temporal downscaler for output from numerical models. Journal of Agricultural, Biological, and Environmental Statistics 2010;15(2):176-197. |
R833293 (2009) R833293 (Final) R833293C001 (2009) R833293C001 (Final) |
Exit |
|
|
Gray SC, Edwards SE, Miranda ML. Assessing exposure metrics for PM and birth weight models. Journal of Exposure Science and Environmental Epidemiology 2010;20(5):469-477. |
R833293 (2008) R833293 (2009) R833293 (2010) R833293 (Final) R833293C001 (2009) R833293C001 (2010) R833293C001 (Final) |
Exit |
|
|
Miranda ML, Swamy GK, Edwards S, Maxson P, Gelfand A, James S. Disparities in maternal hypertension and pregnancy outcomes: evidence from North Carolina, 1994-2003. Public Health Reports 2010;125(4):579-587. |
R833293 (2008) R833293 (2009) R833293 (2010) R833293 (Final) R833293C001 (2009) R833293C001 (2010) R833293C001 (Final) |
Exit Exit |
|
|
Miranda ML, Maxson P, Kim D. Early childhood lead exposure and exceptionality designations for students. International Journal of Child Health and Human Development 2010;3(1):77-84. |
R833293 (2008) R833293 (Final) R833293C001 (2009) R833293C001 (Final) |
Exit |
|
|
Schwartz SL, Gelfand AE, Miranda ML. Joint Bayesian analysis of birthweight and censored gestational age using finite mixture models. Statistics in Medicine 2010;29(16):1710-1723. |
R833293 (2008) R833293 (2009) R833293 (2010) R833293 (Final) R833293C001 (2009) R833293C001 (2010) R833293C001 (Final) R833293C002 (Final) |
Exit Exit |
|
|
Tassone EC, Miranda ML, Gelfand AE. Disaggregated spatial modelling for areal unit categorical data. Journal of the Royal Statistical Society--Series C (Applied Statistics) 2010;59(1):175-190. |
R833293 (2007) R833293 (2008) R833293 (2009) R833293 (Final) R833293C001 (2009) R833293C001 (Final) |
Exit Exit Exit |
Supplemental Keywords:
Data fusion, meta analysis, disparities, spatial disaggregation, spatial interpolation, spatial modeling, racial residential segregation
Progress and Final Reports:
Original AbstractMain Center Abstract and Reports:
R833293 The Center for Study of Neurodevelopment and Improving Children's Health Subprojects under this Center: (EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
R833293C001 Research Project A: Mapping Disparities in Birth Outcomes
R833293C002 Research Project B: Healthy Pregnancy, Healthy Baby: Studying Racial Disparities in Birth Outcomes
R833293C003 Research Project C: Perinatal Environmental Exposure Disparity and Neonatal Respiratory Health
R833293C004 Community Outreach and Translation Core
R833293C005 Geographic Information System and Statistical Analysis Core
The perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.
Project Research Results
66 publications for this subproject
36 journal articles for this subproject
Main Center: R833293
163 publications for this center
76 journal articles for this center
36 journal articles for this subproject
Main Center: R833293
163 publications for this center
76 journal articles for this center