An official website of the United States government
Here's how you know
Official websites use .gov
A
.gov website belongs to an official
government organization in the United States.
Secure .gov websites use HTTPS
A lock (
) or https:// means you've safely
connected to the .gov website. Share sensitive
information only on official, secure websites.
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with,
the contents by NLM or the National Institutes of Health.
Learn more:
PMC Disclaimer
|
PMC Copyright Notice
. 2025 Feb 1;2025:223.
Impacts of Vehicle Emission Regulations and Local Congestion Policies on Birth Outcomes Associated with Traffic Air Pollution
Effectiveness of Vehicle Emission Regulations and Traffic Congestion Reduction Actions in Improving Birth Outcomes in Texas
This Statement, prepared by the Health Effects Institute, summarizes a research project funded by HEI and conducted by Dr. Perry Hystad at Oregon State University and his colleagues. Research Report 223 contains the detailed Investigators’ Report and a Commentary on the study prepared by the HEI Review Committee.
Air pollution accountability research evaluates whether regulatory actions, other interventions, or “natural” experiments yield improved air quality and public health. It is often difficult, however, to differentiate among the effects of policy interventions that take place over the same time frame. It is also difficult to disentangle the effects of policies from unrelated changes in background trends in air quality and health. It can be particularly challenging when policies target numerous pollutant sources, affect large geographic regions, and take several years to fully implement.
Dr. Perry Hystad of Oregon State University and colleagues proposed to examine the effects of both long-term and shorter-term air quality interventions on changes in traffic-related air pollution and birth outcomes in Texas over a 20-year study period. They addressed concurrent policy interventions by considering the cumulative effects of national vehicle emissions regulations in evaluating long-term trends in air quality and health outcomes. They also took advantage of diverse natural experiments, such as local traffic-congestion reduction actions that occurred over shorter time scales. In particular, they focused on changes in air quality and health outcomes before and after implementing tolling and projects to improve roadway capacity.
APPROACH
Hystad and colleagues used a research triangulation approach, which integrates multiple analytical methods, to answer their research questions. To assess the cumulative effects of long-term national regulations aimed at reducing motor vehicle emissions, they examined changes in measures of traffic-related air pollution and birth outcomes over a 20-year period (1996 to 2016). To assess the effects of local tolling implementation and roadway capacity improvement projects that might change air pollution exposures over shorter time periods, they evaluated changes in measures of traffic congestion and adverse birth outcomes before and after project implementation.
What This Study Adds.
This accountability study examined whether reduced traffic-related air pollution was associated with improved birth outcomes in Texas from 1996 to 2016, focusing on the long-term cumulative effects of national regulations to curb vehicle emissions and shorter-term effects of local actions to reduce traffic congestion.
The investigators used a research triangulation approach to evaluate long-term changes in air quality and quasi-experimental designs to evaluate the implementation of tolling systems and roadway improvement projects on birth outcomes.
Annual estimates of nitrogen dioxide exposures decreased by more than 50% for pregnant individuals over the course of the study period, while total vehicle miles traveled decreased by just over 9%. Measures of traffic congestion did not change significantly after toll implementation or roadway improvement projects.
The study found consistent associations between estimated exposure to traffic-related air pollution and adverse birth outcomes throughout the 20-year study period, but the magnitude of those associations decreased over time, parallelling air quality improvements. Local toll implementation and roadway improvement projects did not meaningfully change associations between congestion measures and adverse birth outcomes.
The study provides some evidence that the cumulative effects of long-term regulations aimed at reducing tailpipe emissions have been more effective at decreasing adverse birth outcomes than shorter-term localized programs aimed at reducing congestion.
The investigators assembled a population-based cohort of 8.1 million births that occurred across Texas from 1996 to 2016. The cohort included information on birth outcomes, residential addresses, and sociodemographic characteristics for pregnant individuals. The team specifically examined four birth outcomes: term birth weight (infants born after a minimum of 37 weeks pregnancy), term low birth weight (infants weighing less than 2,500 g at term), preterm birth (pregnancies lasting less than 37 weeks), and very preterm birth (pregnancies lasting less than 32 weeks).
For each birth, individuals were assigned indicators of traffic-related air pollution exposure at their residential addresses, including annual nitrogen dioxide concentrations as well as total and truck-specific vehicle miles traveled (within 500 m) and measures of traffic congestion (based on traffic counts and traffic delays within a range of distances). The team also developed a database of toll implementation and roadway improvement projects in Texas that took place during the study period.
To evaluate the effect of national-scale cumulative regulatory improvements targeting vehicle tailpipe emissions, Hystad and colleagues first examined whether annual nitrogen dioxide levels and the number of total vehicle miles traveled within 500 m of pregnant individuals’ residences changed over the study period. They then used linear and logistic regression models to characterize the association between these metrics and adverse birth outcomes. They also ran stratified models to assess whether the association between vehicle miles traveled on nearby roads and adverse birth outcomes changed over the two decades studied, hypothesizing that the magnitude of the association would decrease over time due to the reductions in the amount and potential toxicity of traffic-related air pollution. Additional stratified models were run to assess whether the change in association over time differed across sociodemographic groups.
To evaluate the effect of local-scale congestion reduction actions, the investigators assessed whether the implementation of tolling systems (switching a toll road or booth to electronic tolling, or adding tolled express lanes) and roadway improvement projects (road widening or intersection improvement projects) resulted in changes to traffic congestion. Using a quasi-experimental design with difference-in-differences methods, they evaluated whether birth outcomes had improved after project implementation in analyses of thousands of projects.
KEY RESULTS
From 1996 to 2016, estimates of annual exposures to nitrogen dioxide levels decreased by 59%, and the number of vehicle miles traveled within 500 m of residences decreased by 9.4% for pregnant individuals across Texas. Both metrics of traffic-related air pollution were consistently associated with adverse birth outcomes (e.g., term birth weight, term low birth weight, preterm birth, and very preterm birth). These associations were strongest for term birth weight and term low birth weight for individuals in the highest 20% of exposures. In other words, on average, pregnancies that experienced the highest level of estimated traffic-related air pollution (whether nitrogen dioxide or nearby vehicle miles traveled) were associated with lower birth weight for infants born at term.
Results from models that assessed changes in the relationship between measures of traffic-related air pollution and adverse birth outcomes over time showed that there remained elevated risks for adverse birth outcomes, even in the most recent years (i.e., 2014 to 2016). However, the magnitude of the risk became smaller over time (for all outcomes except term birth weight). For example, the association between total vehicle miles traveled and term low birth weight decreased by 60% over the course of the study period, meaning the association between total vehicle miles traveled and infants who are born at term with low birth weight decreased over time (see Statement Figure).
Associations from time-stratified models for total vehicle miles traveled within 500 m of pregnant individuals’ residences and term low birth weight (top) and preterm birth (bottom) from 1997 to 2015. The figures compare the 20% highest to the 20% lowest exposures to total vehicle miles traveled. Associations are represented as odds ratios. The percentage changes in the figure denote the percentage change between the predicted associations in 2016 compared to 1996 based on linear trends. (Adapted from Figure 7 in the Investigators’ Report.)
The investigators observed large differences in race, ethnicity, and sociodemographic factors when comparing the lowest and highest 20% of exposure for total vehicle miles traveled. For example, Black non-Hispanic pregnant individuals who lived near roadways had nearly double the amount of total vehicle miles traveled compared to White non-Hispanic pregnant individuals. Despite these large differences in exposure, higher overall associations between total vehicle miles traveled and adverse birth outcomes were reported for White non-Hispanic and higher socioeconomic status pregnant individuals. The investigators found no consistent pattern of change in these associations over time by sociodemographic group.
In analyses of congestion measures of traffic volume and delay, the investigators reported a decrease of 33–53% in congestion after roadway improvement (depending on the type of project) and a decrease in nitrogen dioxide levels up to 13%. They reported no difference in traffic volume or delay after the implementation of tolling. Little evidence was found of improvement in birth outcomes after the implementation of tolling and roadway improvement projects.
INTERPRETATION AND CONCLUSIONS
In its independent evaluation of the study, the Review Committee appreciated that this accountability study was thoughtfully designed. Strong aspects included the research triangulation approach to support the robustness of the findings and the assessment of both long-term cumulative effects of regulation and more localized congestion reduction actions on adverse birth outcomes. The Committee also appreciated the use of a large sample of recorded births over a 20-year period, the detailed measures of air pollution exposures over time, the construction of various novel metrics of congestion, and the compilation of a large database of tolling implementation and roadway improvement projects.
The Committee largely agreed with the interpretation of results reported by the investigators. First, the investigators reported that exposure to traffic-related air pollution was consistently associated with birth outcomes over the 20-year period of the study. During this period, pregnant individuals’ estimated annual exposures to nitrogen dioxide decreased by over 50%, while the estimated number of vehicle miles traveled decreased by 9.4%. Importantly, the magnitude of the associations between total vehicle miles traveled and several of the birth outcomes examined decreased over time, paralleling regulatory progress to reduce traffic-related air pollution. These results suggest that the same amount of traffic may be less toxic on birth outcomes. However, the decrease in the magnitude of these associations over time could also reflect changes in other unmeasured characteristics over time. Ultimately, this finding provides some evidence of improvement in birth outcomes that can potentially be attributed to the cumulative effects of long-term regulatory policies aimed at reducing tailpipe emissions.
Although there was some evidence of improved air quality and less congestion after roadway improvement projects, there was little evidence of such improvements after toll implementation. Likewise, there was no evidence of improved birth outcomes after toll implementation or roadway improvement projects.
However, the focus on major highways could have limited those findings if, for example, any vehicles avoiding tolls were routed onto surface streets in neighborhoods. Relatedly, this work suggests that relatively small-scale actions such as tolling might not result in meaningful reductions in traffic. It might be worthwhile in future studies to evaluate other kinds of congestion measures related to infrastructure improvements, such as specific congestion reduction programs (e.g., the US Department of Transportation’s Congestion Mitigation and Air Quality Improvement Program) or the implementation of intelligent transportation systems.
Overall, this study provides some evidence that reducing tailpipe emissions of the vehicle fleet was more effective at improving air quality and improving birth outcomes than local actions aimed at reducing congestion. This study provides an important contribution to accountability research because of its strong study design and simultaneous evaluation of long-term national-scale actions and shorter-term local actions. However, separating the effects of individual national and local-scale regulations on traffic-related air pollution exposures and health remains a challenge. Future work could benefit from improvements in data and exposure measurements and examine other types of infrastructure projects aimed at reducing traffic pollution.
Res Rep Health Eff Inst. 2025 Feb 1;2025:223.
Impacts of Vehicle Emission Regulations and Local Congestion Policies on Birth Outcomes Associated with Traffic Air Pollution
This Investigators’ Report is part of Health Effects Institute Research Report 223, which includes a Review Committee Commentary and an HEI Statement about the research project. Correspondence concerning the Investigators’ Report may be addressed to Dr. Perry Hystad, College of Health, Oregon State University, Milam Hall 20C, 2520 SW Campus Way, Corvallis, OR 97331; email: perry.hystad@oregonstate.edu. The authors reported no potential conflict of interest.
Although this document was produced with partial funding by the United States Environmental Protection Agency (EPA) under Assistance Award CR–83998101 to the Health Effects Institute, it has not been subjected to the Agency’s peer and administrative review. It may not necessarily reflect the views of the Agency, and no official endorsement by it should be inferred. The contents of this document also have not been reviewed by private party institutions, including those that support the Health Effects Institute; therefore, it may not reflect the views or policies of these parties, and no endorsement by them should be inferred.
In the United States, billions of dollars have been spent implementing interventions to reduce traffic-related air pollution (TRAP*). These interventions are usually regulatory actions focused on reducing tailpipe emissions. However, they also include local programs to reduce traffic congestion and excess vehicle emissions, such as electronic tolls and roadway capacity improvements. Few health studies have empirically evaluated the direct impact of air pollution exposure reductions from these emission regulations and congestion reduction programs; no studies have examined infant health, an important population health outcome linked to air pollution exposures.
Objective
Assess changes in birth outcomes for all recorded births in Texas from 1996 to 2016 associated with (1) long-term cumulative regulatory improvements of motor vehicle emissions and resulting TRAP change and (2) local congestion reduction programs that may yield localized TRAP changes over shorter time periods.
Methods
We used Vital Statistics data in Texas from 1996 to 2016 (n = 8.1 million recorded births; n = 6,158,518 births analyzed after exclusions). We calculated diverse traffic-related exposure measures using residential addresses at the time of delivery. We implemented research triangulation methods using different study design and analysis approaches to test our primary hypotheses on the effects of long-term cumulative regulatory improvements and local congestion reduction programs on birth outcomes.
Results
Traffic-related exposure measures (nitrogen dioxide [NO2] air pollution, traffic volume, congestion) were consistently associated with adverse birth outcomes over the 20-year study period. This finding is supported by an analysis of pregnant individuals living upwind versus downwind of the same major road, where living downwind within 500 m was associated with an 11.6-g decrease (95% CI: −18.01, −5.21) in term birth weight. For all pregnant individuals, NO2 exposures decreased 59% from 1996 to 2016, while the total vehicle miles traveled (VMT) within 500 m of residential addresses (VMT500m) remained relatively stable. We observed marked differences in TRAP exposure for pregnant individuals by sociodemographic characteristics. While levels of air pollution disparities reduced in absolute terms over the 20 years, relative disparities persisted, and large differences in traffic levels remained. The magnitude of associations between VMT500m and adverse birth outcomes decreased for term low birth weight (−60%, OR in 1996: 1.08, OR in 2016: 1.03 for the highest vs. lowest quintile) and preterm (−65%) and very preterm (−61%) births, but not for term birth weight. A direct analysis of congestion exposure for 2015–2016 births, measured for all roadways in Texas using connected device data, showed that congestion was associated with decreased term birth weight, background traffic, and TRAP levels. When we examined local projects designed to reduce congestion as a natural experiment and applied a difference-in-differences (DiD) study design, we found little evidence that the implementation of tolling projects was associated with improved birth outcomes. For roadway construction projects, we observed increased congestion during construction and decreased congestion post-construction. This dynamic translated into increased odds of term low birth weight (OR 1.19; 95% CI: 1.05, 1.36) for pregnant individuals living within 300 m during construction but no consistent improvements in birth outcomes post-construction.
Conclusions
TRAP is an important environmental health and justice issue that affects pregnancy. Our results provide some evidence supporting that cleaning up the vehicle fleet was more impactful at decreasing adverse pregnancy outcomes than local programs aimed at reducing congestion.
CHAPTER 1: INTRODUCTION
Traffic-related air pollution (TRAP) is a toxic subset of ambient air pollution, including byproducts of fossil fuel combustion, road dust, and brake wear.1 This type of pollution disperses along highly localized gradients, primarily concentrated within 500 m of major roads.2–5 Due to the historical siting of highways and major roadways and the strong correlation between socioeconomic status and TRAP exposure, this is a particularly relevant source of air pollution disparities for environmental justice.6
TRAP is a well-established hazard for population health.7–11 Pregnancy is a vulnerable time during which air pollution may have particularly deleterious effects,12,13 and TRAP has been associated with diverse adverse pregnancy outcomes, such as infertility, spontaneous abortion, gestational hypertension, preterm birth, and intrauterine growth restriction.9,12–17 However, there is mixed evidence linking TRAP to adverse birth outcomes,8 and the strong association among TRAP exposures, socioeconomic status,18 and other co-exposures (e.g., noise)19,20 creates a challenging source of confounding (by environmental co-exposures, socioeconomic status, or both) that is often an issue in epidemiological studies of traffic air pollution. Nevertheless, the body of evidence linking TRAP to population health was strong enough to support regulations to reduce the amount and toxicity of tailpipe emissions from vehicles.21
In the United States, billions of dollars have been spent implementing interventions to reduce TRAP. These have generally taken the form of regulatory actions to reduce tailpipe emissions. Figure 1 illustrates the major accomplishments in transportation, air pollution, and climate regulations summarized by the Environmental Protection Agency (EPA).22 These regulatory actions broadly encompass both technological innovations in engine design and enhancements in fuel quality and policies aimed at increasing vehicle fleet turnover. This comprehensive approach to managing emissions from vehicles and other mobile sources required a collaborative effort between the EPA, manufacturers of vehicles, engines, and fuels, and state and local government entities.22 Given the overlap between these distinct regulatory actions and the long periods for implementation and vehicle fleet turnover, it is difficult to isolate the influence of individual regulations on emission reductions and population health. In this report, we use an accountability perspective to examine the cumulative impact of these regulations on reduced TRAP and subsequent infant health.
Timeline of Key US National Policies to Reduce Traffic-Related Air Pollution, 1990 to 2016.22 (CO = carbon monoxide; HC = hydrocarbon; NO = nitric oxide; NOx = nitrogen oxides; PM = particulate matter, SUVs = sport utility vehicles.)
In addition to emission regulations that reduce the pollution emitted from vehicles, local construction and policy programs have been implemented to reduce traffic levels and congestion. For example, many cities implement managed lanes, electronic tolling systems, or roadway capacity expansion to reduce congestion. While these types of studies may have cobenefits of reducing air pollution, few health studies have empirically evaluated the direct impact of air pollution exposure reductions from these types of congestion programs,23 and limited studies have examined infant health, which is a crucial susceptible period for air pollution exposures with far-reaching population health outcomes.
Here, we assessed changes in birth outcomes for 8.1 million births recorded in Texas Vital Statistics data from 1996 to 2016 associated with (1) long-term cumulative regulatory improvements of motor vehicle emissions (and resulting TRAP change) and (2) local congestion reduction programs that could yield TRAP changes in shorter periods. Adverse birth outcomes provided a unique context to examine how these policies changed maternal exposures to TRAP during pregnancy and resulting birth outcomes. Texas was an ideal study location to examine these changes because ~1.7 million pregnant individuals lived within 500 m of a highway or expressway during this period. TRAP measurements there indicated a reduction of over 50% from 1996 to 2016 (using NO2 as a marker for the TRAP mixture). Also, detailed longitudinal road network information (with volumes) was available for Texas, and many congestion reduction programs had been implemented in different metropolitan areas, providing diverse natural experiments to examine changes in birth outcomes for pregnant individuals living in impacted areas before and after implementation. The goal of this research was to examine to what extent vehicle emission regulations and local congestion programs resulted in improvements in birth outcomes through reductions in TRAP exposures.
CHAPTER 2: SPECIFIC AIMS AND OVERARCHING APPROACH
RESEARCH OBJECTIVES
We implemented research triangulation methods using different study designs and analysis approaches to test our primary hypotheses about the impacts on birth outcomes from long-term cumulative regulatory improvements and local congestion reduction programs. Below are the specific aims of our research (see Figures 2 and 3) and an overview of the Texas Birth Cohort, traffic, spatial exposure data linkages, and research triangulation methods implemented. Each subsequent chapter provides additional details for the specific analyses conducted.
Aim 2 conceptual diagram and chain of accountability.
Aim 1: Examine the impact of vehicle emission regulations on TRAP exposures and associated changes in adverse birth outcomes.
Hypothesis: The magnitude of the association between adverse birth outcomes and pregnant individuals exposed to TRAP will decrease over the 1996 to 2016 period, and the benefits of TRAP reductions will not be evenly distributed across socioeconomic and demographic characteristics.
Aim 2: Assess how local congestion programs change local TRAP and adverse birth outcomes.
Hypothesis: Implementing specific traffic reduction programs (e.g., managed lanes, electronic tolling systems, capacity expansion) will be associated with decreased risks of adverse birth outcomes among pregnant individuals who resided near a major road during pregnancy after implementation.
STUDY POPULATION
We leveraged vital statistics data (from the Vital Statistics Unit of the Texas Department of Health and Human Services) to assess all recorded births in Texas from 1996 to 2016. This study was approved by the institutional review board at Oregon State University (0947) and the Texas Department of State Health Services (15-029). The vital statistics data we used contained birth outcome information, full residential locations, and socioeconomic variables for parents of infants born in Texas between January 1, 1996, and December 31, 2016 (n = 8,114,440 recorded births). Vital Statistics data provided sociodemographic characteristics of pregnant people at the individual level. This included information on infant sex, maternal age, maternal race and ethnicity, whether the pregnant individual was born outside of the US, maternal smoking status during pregnancy, maternal weight gain, and the month prenatal care began. Because the vital statistics questionnaire changed in 2005, we relied on variables consistently collected over the entire 1996 to 2016 study period or variables that could be standardized between questionnaires. Additional variables (e.g., payment source used for delivery and WIC [Special Supplemental Nutrition Program for Women, Infants, and Children] use during pregnancy) are available after 2005 and included in analyses covering only the post-2005 period.
We relied on recorded birth information from the vital statistics records to define our main outcomes of interest (Table 1). These were based on birth weight in grams and gestational age. We examined term birth weight (i.e., birth weight among only births whose gestational age was 37–42 weeks), term low birth weight (i.e., birth weight less than 2,500 g among only births whose gestational age was 37–42 weeks), preterm birth (i.e., birth before an estimated gestational age of 37 completed weeks), and very preterm birth (i.e., birth before an estimated gestational age of 32 completed weeks). Measures for birth weight were outcomes related to the pathway of intrauterine growth restriction, while measures of preterm birth were outcomes related to triggers for premature labor.24,25 Ambient air pollution has been associated with all of these outcomes in the literature, although the association is more consistent for birth weight than preterm birth.26,27
Table 1.
Birth Outcomes and Operational Definitions
Outcome
Definition
Term birth weight
Infant weight at delivery among births whose gestational age was 37–42 weeks
Term low birth weight
Infant weight less than 2,500 g among births whose gestational age was 37–42 weeks
Preterm birth
Birth before an estimated gestational age of 37 completed weeks
Very preterm birth
Birth before an estimated gestational age of 32 completed weeks
We acknowledge the complex etiology of birth outcomes, including the substantial debate in the literature on how best to examine these outcomes.28–32 To distinguish these pathways, we examined preterm births (operationalized as preterm birth and very preterm birth) and birth weight restricted to term births (operationalized as continuous term birth weight and term low birth weight). In particular, we restricted our analyses of birth weight to births born at term (≥37 weeks’ gestation). We conceptualized term birth weight as an independent outcome measure for adverse pregnancy outcomes. Birth weight, independent of adjustment for gestational duration, represents the combined influence of potential biological pathways on gestational length and growth restriction.24 This decision followed the inclusion criteria of the Health Effects Institute’s (HEI) Systematic Review and Meta-analysis of Selected Health Effects of Long-Term Exposure to Traffic-Related Air Pollution,25 where only studies restricting analyses to term birth weight and term low birth weight were included. There are concerns with conditioning on gestational age as an intermediate and the potential for collider bias26,27,29,31; however, there is considerable debate in the literature on this issue.25,30,32 We restricted our analyses of birth weight to term births or comparability to other air pollution studies to follow the HEI TRAP review inclusion criteria and to examine infant growth and gestational duration separately. We conducted sensitivity analyses of term birth weight for all births to examine potential collider bias and to determine whether associations were similar to those observed for term births. We did not examine “small for gestational age” (SGA) because this metric conflates the biological pathways of growth and gestational length, complicating the isolation of their individual impacts. The decision to exclude SGA from our analysis is supported by literature indicating that SGA can conflate these distinct pathways and poorly identifies at-risk infants.33 By concentrating on term low birth weight and term birth weight, we aimed to understand better how environmental factors, such as roadway construction and the associated changes in air pollution, specifically impacted infant growth and gestational duration.
TRAFFIC EXPOSURE ASSESSMENT
We derived several traffic exposure assessment measures and additional spatial data linkages from the residential address reported on the birth certificate at time of delivery. Figure 4 summarizes the data linkages created and each measure’s available years. Variables are described in detail within subsequent analyses. Briefly, we used nitrogen dioxide (NO2) air pollution as a marker of the overall tailpipe emission exposure mixture8,10,21 to determine the extent to which TRAP is associated with adverse birth outcomes in this study population. We assigned annual NO2 (ppb) concentrations to each participant’s address using an existing hybrid land use regression model for the years 1996 to 2016.34 Our primary traffic metrics to evaluate whether decreased tailpipe air pollution emissions resulted in better birth outcomes were total and truck-specific vehicle miles traveled, or VMT, near residential addresses. Examining VMT, rather than NO2, allowed us to determine whether being exposed to the same amount of traffic had a smaller influence on adverse birth outcomes over time due to reduced tailpipe emissions. We also developed and implemented an exposure assessment approach to calculate hourly wind rows to estimate the percentage of time each pregnancy was downwind of roadways and matched maternal residences predominantly upwind and downwind of the same high-traffic road, considering building and tree shielding. In addition, we examined vehicle congestion measures directly from a unique “connected device” dataset available for recent periods (2012 to 2016). This database was created for the Texas’ Most Congested Roadway project by researchers at the Texas A&M Transportation Institute.35,36 Congestion metrics were derived down to the road section level based on congestion performance measure calculations from the state’s roadway inventory and a proprietary data source of global positioning system (GPS) speed reports generated from individual vehicles and cell phones (i.e., connected devices). Specific to Aim 2, we developed spatial-temporal databases for toll implementation and roadway construction projects.
Summary of traffic exposure assessment data and additional data linkages conducted for residential addresses at time of delivery.
ADDITIONAL DATA LINKAGES
We conducted extensive data linkages based on residential addresses to further characterize environmental exposures, neighborhood contextual factors, and household characteristics. Each variable used in subsequent analyses is described in more detail in that chapter. The data availability section of the report provides information on accessing the exposure data used in this report.
RESEARCH TRIANGULATION APPROACH
This report leverages research triangulation techniques to address our two overarching research aims. Research triangulation seeks to obtain reliable answers to research questions by integrating results from several different approaches, where each approach has different potential biases unrelated to one another.37Figure 5 illustrates how this method allowed our research team to use data from various times and locations, involve multiple researchers with expertise in diverse disciplines, apply different theoretical perspectives, and use different methodologies to examine our research questions. The results gave us a more robust understanding of (1) TRAP effects on adverse birth outcomes and (2) to what extent vehicle emission regulations and local congestion programs resulted in improvements in birth outcomes through reductions in TRAP exposures.
Research triangulation methodology used to address our research aims.
This report combines seven papers, led by different research team members, into five chapters that comprehensively examine vehicle emission regulations and local congestion policies’ impacts on birth outcomes. In Table 2, we list our main analyses, general methods used, and analysis goals. Within Chapter 3: “Assessing the Effect of Decreased Tailpipe Emissions on Adverse Birth Outcomes Over 20 Years in Texas,” we directly examine Aim 1. Few studies have explicitly tried to overcome the potential confounding of TRAP with lower socioeconomic status and other environmental exposures (e.g., noise). In Chapter 4: “Living Downwind of High-Traffic Roads and Adverse Birth Outcomes,” we present an analysis that compares birth outcomes of maternal neighbors living upwind and downwind of the same high-traffic road. In Chapter 5: “Isolating the Additional Influence of Traffic Congestion on Infant Health,” we leverage connected device data to assess congestion exposures for 2015–2016 births and associations with term birth weight. This analysis is an important link to Aim 2, which examines how congestion reduction methods (tolls and roadway improvements) may improve pregnancy outcomes for nearby individuals. Chapter 6: “Transition to Electronic Tolling and Associations with Adverse Birth Outcomes,” and Chapter 7: “Highway Construction Projects, Changes in Congestion and Traffic Air Pollution, and Adverse Birth Outcomes,” both use quasi-experiments and a DiD design to isolate the causal associations between these local interventions and changes in birth outcomes.
Table 2.
Summary of Analyses Conducted, General Methods Used, and Main Analysis Goal
Analysis
Time-Period
General Methods Used
Primarily TRAP Exposure
Analysis Goals
Chapter 3: Impact of Vehicle Emissions Regulations on Traffic-Related Air Pollution Exposures and Associated Changes in Adverse Birth Outcomes
1996 to 2016
Linear regression models of change in associations over time
Vehicle miles traveled within 500 m of residences
Aim 1: Change in TRAP exposures and adverse birth outcome associations over time
Chapter 4: Living Downwind of High-Traffic Roads and Adverse Birth Outcomes
2007 to 2016
Linear and logistic regression of neighbors matched by wind
Living downwind (within 500 m) of major roads (>25,000 AADT) compared to upwind of the same road
Reduce the potential impact of residual and unmeasured confounding
Chapter 5: Isolating the Additional Influence of Traffic Congestion on Infant Health
2015 to 2016
Linear regression models of congestion exposure
Roadway congestion levels within 500 m of residences measured from connected device/vehicle data
Examine direct associations between congestion and term birth weight
Chapter 6: Transition to Electronic Tolling and Associations with Adverse Birth Outcomes
1999 to 2016
DiD design
Quasi-experiment examining residences before and after toll changes and near and far from toll road locations
Aim 2: Impact of transition to tolling and changes in birth outcomes
Chapter 7: Roadway Capac-ity Improvements, Changes in Congestion and Traffic Air Pollution, and Adverse Birth Outcomes
2009 to 2016
DiD design
Quasi-experiment examining residences before, during, and after roadway construction and near and far from construction locations
Aim 2: Impact of roadway construction and changes in birth outcomes
AADT = annual average daily traffic; DiD = difference-in-differences; TRAP = traffic-related air pollution.
CHAPTER 3: IMPACT OF VEHICLE EMISSION REGULATIONS ON trap EXPOSURES AND ASSOCIATED CHANGES IN ADVERSE BIRTH OUTCOMES
INTRODUCTION
TRAP is a well-established hazard for population health.7–11 This pollution is a heterogeneous mixture of NO2, particulate matter, carbon monoxide, and other volatile organic compounds.8,11 Due in part to several landmark epidemiological studies in the 1990s.38–42 US policymakers implemented regulations to reduce the amount and toxicity of tailpipe emissions,21 costing billions of dollars in state and federal funds.43,44 Many of these regulatory measures (e.g., the Clean Air Act) are specifically designed to improve population health,45 but the extent to which the totality of these regulatory measures succeeded in their health-related goals is typically not evaluated.
Accountability studies (i.e., the assessments of past environmental policies) can determine whether and to what extent regulatory measures were effective at improving population health outcomes and explore what sources, exposures, and mechanisms may be responsible for this change.46–50 This process is somewhat straightforward when the goal is to understand a single regulatory action, such as the shutdown of an industrial plant, a temporary traffic reduction for a large event, or a regional shift away from coal use.46,51–56 However, multiple TRAP policies are often implemented simultaneously (e.g., limits on diesel fuel sulfur content, criteria pollutant emissions standards, Corporate Average Fuel Economy (CAFE) [I] standards) making it difficult to isolate the influence of a single regulatory decision.21 Further, many regulations create complex, long-term programs with benefits that may not be immediately evident.43
To overcome this challenge, epidemiological studies of TRAP regulations require health data sources spanning long periods. Several studies established that decades of reductions in TRAP could improve mortality, even for improvements at very low ambient exposure levels.57–59 Further studies found that the benefits of TRAP reductions were not equitably distributed across sociodemographic characteristics,60,61 and persistently marginalized populations are still often exposed to worse air quality than the general population.2,62–65 Other than mortality, few health outcomes have been evaluated in an accountability study framework, primarily due to challenges related to sample size and data quality.46–49,57–59
Adverse birth outcomes are a critical public health issue that can be accessed via population-based data over extended periods in the United States. Preterm birth (<37 weeks’ gestation) and low birth weight (<2,500 g) are immediate risk factors for maternal-infant mortality and long-term risk factors for chronic health conditions later in life.66–70 The link between TRAP and adverse birth outcomes is well-established, although the magnitude of associations varies between study settings globally.7,8,71 A large body of evidence has established associations between vehicle exposure measures (e.g., proximity, traffic density, vehicle miles traveled) and adverse birth outcomes.71 To date, few accountability studies have focused on the influence of TRAP regulations on birth outcomes,55,72,73 a highly sensitive topic with stark socioeconomic gradients that can be measured at the population level.7,71,74,75 Our previous research showed that the magnitude of association between residence within 300 m of a major road and term birth weight has decreased over time,72 lending credibility to the hypothesis that regulation can reduce the health impact of TRAP.
In this study, we leveraged a population-based vital statistics cohort of geocoded births in Texas, spanning 20 years (1996 to 2016) to evaluate associations between VMT, an indirect measure of TRAP often used by urban planners and traffic engineers), roadway proximity, NO2, and adverse birth outcomes. We hypothesized that the associations between adverse birth outcomes and both VMT and roadway proximity would decrease in magnitude over time, paralleling regulatory progress to reduce TRAP (NO2) concentrations.
STUDY DESIGN AND METHODS
Study Population
We used birth certificate data from the Texas Department of Health and Human Services to identify all pregnancies recorded in Texas between 1996 and 2016, resulting in a cohort of 8,114,440 recorded births. We excluded records without street-level residential address data (n = 890,842, 11.0%), missing information related to delivery (i.e., birth weight, gestational age, maternal age, or birth type) (n = 47,096, 0.6%), birth weight outside of reasonable bounds (i.e., <500 g or >5,000 g) (n = 17,591, 0.2%), gestational age outside of reasonable bounds (i.e., <22 weeks or >42 weeks) (n = 10,447, 0.1%), deliveries with multiple infants (n = 213,250, 2.6%), and missing data for key covariates (n = 776,696, 9.6%), yielding an analytic sample of 6,158,518 births.
Trap Exposure Assessment
We first examined NO2 air pollution exposure, a marker of the overall tailpipe emission exposure mixture,8,10,21 to determine the extent to which TRAP is associated with adverse birth outcomes in this cohort. We assigned annual NO2 (ppb) concentrations to each participant’s address using an existing hybrid land use regression model for the years 1996 to 2016.34
Traffic Count Exposure Assessment
Our primary traffic metric to evaluate whether decreased tailpipe air pollution emissions resulted in better birth outcomes was VMT near residential addresses. Examining VMT, rather than NO2, allowed us to determine if the association between exposure to the same amount of traffic has a smaller influence on adverse birth outcomes over time due to reduced tailpipe emissions. We hypothesized that living near a major road with 20,000 annual average daily traffic (AADT) in 1996 should be associated more strongly with adverse birth outcomes compared to living near a major road with 20,000 AADT in 2016, primarily due to the reductions in the amount and potential toxicity of TRAP released from these vehicles.
We created a database of traffic volumes using historical Texas roadway inventory databases.76 AADT for all vehicles and trucks was available for the Texas Department of Transportation (TxDOT)-managed roadways from 1999 to 2016. We used linear extrapolation at the street segment level to estimate AADT from 1996 to 1999 and any other missing AADT values for road segments from 1999 to 2016. Next, we calculated the VMT for all roads within 300, 500, and 1,000 m of residential addresses (Equation 1).
(1)
where
VMTx = VMT for all roads within x meters of residential address
nx = number of road segments within x meters of residential address
lengthi = length in meters of road segment i
AADTi = AADT of road segment i
We also estimated truck-specific VMT using AADT restricted to vehicles classified as single-unit or combination trucks.77 A priori, we examined the 500 m buffer distance as the best measure of TRAP exposure.10 We classified birth addresses as being within 300 m of a highway (defined using the Census TIGER Road shapefiles in 2010), corresponding to the highway proximity exposure measure from our prior study.72
Additional Contextual Variable
We linked additional data to residential locations to quantify environmental and social contextual factors: percentage tree cover within 500 m buffer areas from Landsat Vegetation Continuous Fields tree cover product (2000, 2005, 2010, and 2015)78; percentage impervious surface within 500 m from the USGS National Land Cover Database (2001, 2006, 2011, 2016)79; annual maximum normalized difference vegetation index (NDVI) within 500 m buffer areas, standardized across Landsat 5, 7, and 8 imagery; and monthly mean and maximum temperature for the month of delivery using PRISM climate data.80 We also linked census tract socioeconomic data (race and ethnicity profiles, percentage of homes owned and rented, and median household income) from the American Community Survey for 2007 to 2016. We applied linear interpolation models on the census tract level Decennial Census and American Community Survey data to births from 1996 to 2006.
Birth Outcome Measures
We examined term birth weight in grams (delivery at 37–42 weeks’ gestation), term low birth weight (term birth weighing less than 2,500 g), preterm birth (<37 weeks’ gestational age), and very preterm birth (<32 weeks’ gestational age).
DATA ANALYSIS
We first calculated descriptive statistics for our birth cohort (n = 6,158,518) by the lowest and highest quintiles of total VMT500m for the entire study period (1996 to 2016), as well as early (1996 to 1998) and late (2014 to 2016) study years. We visually examined disparities in these TRAP exposures by sociodemographic characteristics from 1996 to 2016 and summarized exposure levels, using the group with the highest socioeconomic positioning as the reference (e.g., White non-Hispanic, completed more than high school, US-born, high-income neighborhood). We computed each group’s absolute differences in exposure (via subtraction) and relative differences in exposure (via a percentage difference). These analyses are presented in full in a separate manuscript focused on TRAP exposure inequalities.81
Next, we implemented a set of linear and logistic regression models to estimate the influence of traffic-related exposures (NO2, total and truck-specific VMT, and highway proximity) on our four adverse birth outcome measures (term birth weight, term low birth weight, preterm birth, very preterm birth) for our entire 20-year study period. We ran nonlinear models (SAS Proc GAMPL) to visually examine the concentration-response relationship between NO2 and VMT and adverse birth outcomes measures, which informed our decisions to examine quintile-based exposure categories. We implemented basic and adjusted models using a-priori selected covariates that have been consistently collected from 1996 to 2016 through vital statistics records. The basic model included infant sex (female, male), maternal age (linear), birth month (categorical) and year (categorical), and county (categorical). The adjusted models additionally included maternal race (White, Black, Native American, Asian, Pacific Islander, Other) and ethnicity (Hispanic/Latina, Non-Hispanic/Latina), whether the pregnant individual was born outside of the US (yes, no), educational attainment of the pregnant individual (<8th grade, up to high school diploma, up to bachelor’s degree, more than bachelor’s degree), maternal smoking status during pregnancy (yes, no), maternal weight gain (linear), month of pregnancy prenatal care began (none, months 1–9), and census tract median household income (tertiles by year).
We created models for our entire study period (1996 to 2016) and time-stratified models for 3-year rolling periods (e.g., 1996 to 1998 through 2014 to 2016) to evaluate association trends. For all analyses, we examined models stratified by key race/ethnicity and sociodemographic characteristics to determine the potential effect modification of TRAP on adverse birth outcomes and differential changes in associations over time. We examined stratified models for race and ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic or Latina), educational attainment of the pregnant individual (high school graduate or less, greater than a high school diploma), and neighborhood income status (highest income tertile vs. lowest income tertile). All analyses were run in SAS 9.4.
We conducted several sensitivity analyses to evaluate the robustness of our models. First, we examined different buffer distances (300 m, 1,000 m) for all vehicle and truck-specific VMT. Second, we ran models that included individuals dropped from our main analysis (complete case analysis) using missing categories for categorical covariate data. Third, we added additional contextual variables to the model, including particulate matter ≤2.5 μm in aerodynamic diameter (PM2.5) (1999 to 2016 only), particulate matter ≤10 μm in aerodynamic diameter (PM10), ozone (O3) air pollution estimates,34 NDVI, tree canopy cover and impervious surface area in 500 m, monthly mean temperature, and census tract % non-White residences and % renters as independent covariates. Fourth, we included an independent covariate for census tract (i.e., a categorical variable for each tract) to examine potentially unmeasured geographic confounding. Finally, we examined the sensitivity of birth weight associations when restricting the subset to full-term births and adjusting for gestational duration.
Finally, in addition to global models for all of Texas, we explored geographic differences in the effects of TRAP on adverse birth outcomes using a spatial smoothing Bayesian multilevel modeling framework. The basic model (Equation 2) involves, for region i (census tract, denoted as CT, or county) at time point j for individual k, a binary indicator of an adverse birth outcome yijk, and an estimate of exposure indicated as xijk. As per Bell and colleagues,82 we examined within-region effects of traffic based on deviations of individual-level traffic exposure from region-level means and adjusted for region-level average traffic exposure , to inform estimation of intercept terms. Given the set of individual-level confounders (same as in the global model) and associated regression coefficients, Vηk, our model was:
(2)
where “intercepts” or baseline levels of log-odds of term low birth weight, αij, and “slopes,” or associations with traffic, βij, vary over space and time.
Combinations of spatially smoothing intrinsic conditional autoregressive (SAR) and random walk (RW1) constructs were used to model the main effects for space and time, with interaction terms incorporated as described elsewhere.83,84 These models were estimated in a unified manner via the INLA software package.85,86 Additional details of the spatially varying modeling approach are described in Appendix A (available on the HEI website).
RESULTS
For our sample of 6,158,518 births, mean term birth weight was 3,366 (SD: 448) g, with 134,695 (2.4%) births classified as term low birth weight (<2,500 g at term), 518,063 (8.4%) births preterm (<37 weeks’ gestation), and 66,334 (1.2%) births very preterm (<32 weeks’ gestation) (Table 3). Mean birth outcome measures were similar in the early (1996 to 1998) and later (2014 to 2016) study years.
Table 3.
Descriptive Characteristics of Study Population with TRAP Exposures from 1996 to 2016 by the Lowest and Highest Quintiles of Total Vehicle Miles Traveled Within 500 m
NO2 = nitrogen dioxide; ppb = parts per billion; Q = quintile; SD = standard deviation; TRAP = traffic-related air pollution; VMT = vehicle miles traveled.
Over the 20-year study period, large differences in race, ethnicity, and sociodemographic factors were observed between the lowest and highest total VMT500m exposure quintiles (Table 3). We previously published a full analysis and description of the environmental disparities associated with TRAP in Texas and the changes from 1996 to 2016.81 Differences in exposure disparities changed over time, but these changes were less pronounced than the absolute between-group exposure differences. For example, in the early years (1996 to 1999), the lowest quintile of VMT500m was comprised of 7.5% black pregnant individuals, compared to 14.4% in the highest exposure quintile, while for White non-Hispanic pregnant individuals, the lowest quintile of VMT500m was 90.3%, compared to 80.5% in the highest exposure quintile.
We observed large decreases in NO2 exposures (–59%) for all births to pregnant individuals from 1996 to 2016, with smaller decreases in total VMT500m (–9.4%). Mean NO2 over the early years (1996–1998) was 13.7 (SD: 4.6) ppb, and VMT500m was 16,198 (23,779), compared to 6.5 (2.6) ppb and 15,976 (26,084) over the later years (2014–2016). The percentage of pregnant individuals living within 300 m of a highway in early years was 19.6%, compared to 17.1% in later years. Spatial patterns of NO2 exposures, total VMT500m, and road proximity for 2010 in Houston, Texas, are shown in Figure 6. The Spearman correlation between total VMT500 and NO2 was 0.38, and between total VMT500 and truck VMT500 was 0.89.
Traffic-related exposure measures (NO2, total vehicle miles traveled within 500 m, and highway proximity) for Houston, Texas, 2010. (Hystad et al. 2025; Creative Commons License CC BY-NC 4.0.)
For all birth years, most TRAP and adverse birth outcome associations demonstrated nonlinear associations (Appendix A, Figure A1). Strong associations were observed for all outcomes and exposure measures in the base models (Appendix A, Table A1). For NO2, we observed strong adjusted associations with term birth weight and term low birth weight, with smaller associations with preterm and very preterm birth (Table 4). For example, the highest quintile of NO2 exposure, compared to the lowest, was associated with a 16.6-g decrease (95% CI: –18.3, –14.8) in term birth weight and an odds ratio (OR) of 1.03 (95% CI: 1.02, 1.05) for preterm birth. We observed consistently adverse associations between traffic exposure and all adverse birth outcomes. For example, the highest quintile of total VMT500 and truck VMT500, compared to the lowest respective quintiles, was associated with an 11.5-g decrease (95% CI: −12.7, −10.4) and 8.1 g (95% CI: −9.2, −6.9) in term birth weight. We observed a weaker association between traffic exposures and preterm birth (<37 weeks) compared to very preterm birth (<32 weeks) (Table 4). For very preterm birth, the highest quintiles of total VMT500 and truck VMT500 compared to the lowest had an OR of 1.09 (95% CI: 1.06, 1.12) and 1.07 (95% CI: 1.05, 1.10), respectively. Highway proximity measures showed smaller associations compared to those from our VMT metrics.
Table 4.
Adjusted Associations Between NO2 Air Pollution, Total VMT500m of Pregnant Individual’s Address, Truck VMT in 500 m, and Highway Proximity and Birth Outcome Measures
NO2 = nitrogen dioxide; ppb = parts per billion; Q = quintile; VMT = vehicle miles traveled.
Adjusted models: Maternal age, infant sex, maternal race and ethnicity, maternal educational attainment, pregnant individual foreign-born, prenatal care received, maternal weight gain during pregnancy, smoking during pregnancy, neighborhood median income, birth month and year, county.
We then examined changes in the associations between traffic exposures and adverse birth outcomes over time (Figure 7). For total VMT500 and truck VMT500m, we observed increases over time in the magnitude of the association with term birth weight. For term low birth weight, associations became smaller (the OR decreased from 1.07 in 1996 to 1.03 in 2016), although this association varied substantially over time. We also observed decreases in the magnitude of associations for preterm birth and very preterm birth. For example, the OR decreased for very preterm birth from 1.12 in 1996 to 1.05 in 2016. For highway proximity, we observed similar temporal association patterns with VMT500m.
Adjusted model results for 3-year rolling stratified populations from 1997 to 2015 of VMT500m of residences. Graphs present the highest versus lowest quintile of total VMT. Change estimates represent the percentage change between the predicted associations in 2016 compared to 1996 based on linear trends. PTB = preterm birth; TLBW = term low birth weight; VMT = vehicle miles traveled; VPTB = very preterm birth. (Hystad et al. 2025; Creative Commons License CC BY-NC 4.0.)
In models stratified by race/ethnicity, maternal education attainment, and neighborhood income tertiles (Appendix A, Table A2), associations were generally larger for White non-Hispanic pregnant individuals, higher-educated individuals, and individuals living in higher-income neighborhoods. For example, no association was observed between VMT500 and term birth weight for Black non-Hispanic pregnant individuals, compared to a 15.57-g reduction (95% CI: −17.3, −13.6) for White non-Hispanic pregnant individuals exposed to the highest quintile of VMT500m compared to the lowest. For very preterm birth, associations observed for Black non-Hispanic pregnant individuals were 1.08 (95% CI: 1.02, 1.15), compared to 1.15 (95% CI: 1.10, 1.21) for White non-Hispanic pregnant individuals.
In terms of change in associations over time, we examined trends in total VMT500m by race/ethnicity, maternal education attainment, and neighborhood median household income (Figure 8). Term birth weight and term low birth weight demonstrated inverse patterns, especially for White non-Hispanic pregnant individuals for whom associations between VMT500m and term birth weight became larger over time, compared to decreasing effect sizes for term low birth weight. For preterm birth and very preterm birth, associations over time were decreasing for all subpopulations, except for Black non-Hispanic pregnant individuals. For this population, term low birth weight and preterm birth effect estimates increased slightly over time.
Adjusted model results stratified by race/ethnicity, educational attainment, and neighborhood income for 3-year rolling stratified populations from 1997 to 2015 of VMT500m of residences. Graphs present the highest versus lowest quintile of total VMT.
Change estimates represent the percentage change between the predicted associations in 2016 compared to 1996 based on linear trends. PTB = preterm birth; TLBW = term low birth weight; VMT = vehicle miles traveled; VPTB = very preterm birth. (Hystad et al. 2025; Creative Commons License CC BY-NC 4.0.)
In sensitivity analyses, we examined VMT300m and VMT1000m at residential addresses, which showed similar associations to our VMT500m estimates. We included individuals who dropped from our main analysis by coding missing covariates into their own category (i.e., a missing data category), and model results were nearly identical to our complete case analysis (Appendix A, Table A3). Adding extensive additional social and environmental contextual variables to the model, including PM10, O3, NDVI, tree canopy cover, impervious surface area within 500 m, monthly mean temperature, and census tract percentage visible minority and percentage rental attenuated annual model results but patterns over time remained stable (Appendix A, Table A4). Models that included PM2.5 from 1999 to 2016 also did not change results. Inclusion of a census tract covariate, which restricts comparisons to locations within the same census tract to further control for potential unmeasured and residual confounding, attenuated point estimates for term birth weight but did not change other adverse birth outcome measures substantially (Appendix A, Table A5). Finally, associations for birth weight were similar across the analysis when restricted to term births (37–42 weeks), all births, and all births with no adjustment for gestational weeks (Appendix A, Table A6).
Model estimates specific to counties and census tracts demonstrated spatial and temporal variability, but overall trends matched the overall associations observed for Texas. For example, Figure 9 illustrates county-level associations for VMT500m and term low birth weight, or TLBW. Most counties show that an increase in VMT500m is associated with increasing probabilities of term low birth weight, except for certain years (2008 to 2013) in less populated counties in the southwest of Texas. Yearly trends also indicate decreased overall associations over time. Model estimates for census tract also revealed spatial and temporal variability, with more uncertainty than county estimates due to smaller sample sizes (Appendix A, Figure A2).
Adjusted county-level associations between VMT500m and term low birth weight for each year from 1996–2016. Positive slopes represent the probability of term low birth weight increasing with increasing VMT500m.
DISCUSSION AND CONCLUSIONS
In this population-based cohort (n = 6,158,518 births), average NO2 exposures decreased by 59% from 1996 to 2016 despite relatively stable VMT500m near residential addresses of pregnant individuals in Texas. We observed that exposure to vehicle traffic was consistently associated with adverse birth outcomes over our 20-year period. However, the magnitude of these associations was reduced over time for term low birth weight, preterm birth, and very preterm birth. This trend generally paralleled the decrease observed in NO2 concentrations over this period. However, trends were inconsistent for term birth weight, and no consistent patterns in the magnitude of change were observed when we evaluated trends by race/ethnicity, educational attainment, or neighborhood income level.
Similar to results in other published reports,45 our results demonstrated that outdoor ambient NO2 concentrations around residences were reduced by 59% over our study period (1996 to 2016), while contemporaneous VMT500m trends around the same residences decreased by only 9%. These decreases represented changes in TRAP exposure to pregnant individuals, not overall population NO2 and VMT500m trends. Interestingly, VMT for the entire state of Texas road network increased over our study period, suggesting that pregnant individuals were either choosing to live in areas with lower VMT or that new developments were occurring in areas of lower VMT, such as suburbs.
While pregnant individual exposures to TRAP substantially decreased overall, the magnitude of these improvements was consistently lower for pregnancies among historically marginalized populations and in lower-income neighborhoods.81 In addition, residential exposure to traffic has increased over time for Black non-Hispanic pregnancies, individuals with less than a high school diploma, foreign-born individuals, and individuals living in historically redlined or low-income neighborhoods. These results demonstrate the persistent legacy of structural sociodemographic segregation in Texas, highlighting the ongoing need for equitable implementation of environmental policy.
We hypothesized that the association between VMT500m and adverse birth outcomes would decrease in magnitude over time, paralleling regulatory improvements that have reduced vehicle tailpipe emissions (shown here via NO2). Our results demonstrated some evidence to support this hypothesis, especially for term low birth weight (−60% reduction in ORs for VMT500 from 1996 to 2016), preterm birth (−65%), and very preterm birth (−61%). For term birth weight, we observed increased magnitudes of associations over time, potentially suggesting that NO2 may not accurately represent the responsible TRAP exposure. Additional analyses in this report (Chapter 5) show that traffic congestion had a unique influence on reduced term birth weight beyond the influence of TRAP alone87; other studies have demonstrated that road noise contributes toward adverse infant health outcomes beyond TRAP itself. While the percentages of reduction in estimated effect sizes over time were large (~60%), they represented small changes in the actual ORs over time, and substantial heterogeneity existed in associations over the 20-year study period.
Nevertheless, the totality of evidence suggests that regulations targeting vehicle tailpipe emissions decreased the impacts of TRAP on some adverse birth outcomes (term low birth weight, preterm birth, very preterm birth) but were less effective at protecting against other measures of adverse birth outcomes (term birth weight). Notably, most model results for NO2 and traffic exposure continued to indicate adverse associations with birth outcomes, even in recent study years (e.g., 2016). We did not directly test whether NO2 associations with adverse birth outcomes changed over time, as this would have addressed a different question (i.e., whether the toxicity of NO2 changed over time), which requires different analytical approaches. Further research is needed to determine the specific TRAP exposure(s) that may be driving these differences and contributing to these persistent associations with adverse birth outcomes.
In addition, we hypothesized that decreases in TRAP concentrations and adverse birth outcomes over time would differ by race/ethnicity and socioeconomic status, indicators corresponding to characteristics of persistently marginalized populations. Surprisingly, we observed larger overall associations for White non-Hispanic and higher socioeconomic status pregnant individuals, with no consistent pattern of change over time in marginalized subpopulations. These associations are opposite to the exposure level experienced by these populations, with Black non-Hispanic pregnant individuals having nearly double the exposure to VMT500 (mean: 20,797) compared to White non-Hispanics (mean: 12,053). This same exposure disparity holds for NO2 and other markers of TRAP exposure. The smaller association in these subgroups may be due to competing risks and higher levels of baseline risk. For example, for Black non-Hispanic pregnancies, there was a much higher rate of term low birth weight (4.3%), preterm birth (12.0%), and very preterm birth (2.6%) compared to White non-Hispanic births classified as term low birth weight (1.8%), preterm birth (7.7%), and very preterm birth (1.0%), respectively.
In terms of trends over time, for White non-Hispanic pregnant individuals, associations between VMT500 and term birth weight became larger, compared to smaller associations with term low birth weight over time. For preterm birth and very preterm birth, association sizes decreased over time for all race/ethnic, education, and neighborhood income groups, except for Black non-Hispanic pregnant individuals, where term low birth weight and preterm birth increased slightly over the study period.
We also examined spatially varying patterns of associations by counties and census tracts. We observed differences across space and time in the associations between VMT500m and term low birth weight. Yet these differences generally followed the global associations observed for all of Texas and did not reveal clear patterns at the census tract level. Further research will explore these patterns and potential factors leading to higher and lower associations in specific regions and neighborhoods across Texas.
Reducing tailpipe air pollution emissions has been a major regulatory undertaking for the last five decades and should yield direct benefits for population health. Billions of dollars have been spent on technological improvements to reduce tailpipe emissions since the 1970s.43,44 Accountability analyses to determine whether, and how much, these regulations have changed air pollution emissions and concomitant exposures and health outcomes would be essential to evaluating the efficacy of these policies and guiding policies to be more protective of vulnerable populations. To date, few studies have been able to directly evaluate this full chain of accountability for TRAP and any health outcome,46,49 let alone complex regulatory interventions, as we examined here.48,88
We examined birth outcomes as the health endpoint of interest, given the potential link between TRAP and adverse birth outcomes, consistent population-level recording of adverse birth outcomes on vital statistics records, and the large population (almost 7 million births over 20 years). The relatively short exposure period for gestation also helps isolate association changes over time. However, there are major challenges in evaluating the full scope of the accountability chain, such as collecting policy-relevant exposure data over long periods, accessing biological data to determine the extent to which reduced environmental exposures yielded changes in human exposures, finding sufficiently large cohorts to ensure statistical power to detect associations, and estimating air pollution exposures (and potentially changing air pollution mixtures) over time.
There are some key limitations to keep in mind when interpreting our results. First, our analysis examined the totality of policy efforts related to TRAP reductions; thus, we cannot attribute our results to any specific regulatory action. Second, our exposure measures were predicated on maternal address at delivery, and we lacked information on residential mobility and daily time-activity patterns that could influence exposures.89 Third, our analysis leveraged vital statistics records, which only included live-born infants.26 As TRAP could influence fertility and pregnancy loss,14,90,91 our associations may have underestimated the magnitude of association that TRAP had on reproductive and pregnancy health, including birth outcomes. We also restricted birth weight analyses to term births to separate the potential biological pathways for gestational length and infant growth. However, sensitivity analyses showed similar associations for birth weight when examining all births and births ≥37 weeks of gestation. For example, term birth weight analyses for VMT within 500 m yielded a decrease in birth weight of −11.5 g (95% CI: −12.7, −10.4) for the highest compared to the lowest quintile, compared to −11.7 g (95% CI: −13.1, −10.3) for all births. Similarly, for NO2, term birth weight analyses yielded a decrease in birth weight of −16.6 g (95% CI: −18.3, −14.8) for the highest compared to the lowest quintile, compared to −14.5 g (95% CI: −16.7, −12.4) for all births. Fourth, our measures of TRAP were at the annual level, and we could not identify potential susceptible periods of exposure during pregnancy. Fifth, data quality may be improving over the study period. However, we conducted a 3-year rolling average analysis, controlling for the year. Improving data quality would likely attenuate the changes observed in our study over time.
In conclusion, in Texas data spanning 20 years, we observed some evidence of improvements in birth outcomes associated with decreasing traffic volumes near the homes of pregnant individuals, paralleling decreases in NO2 levels, and other policies related to tailpipe emissions. We also found that TRAP, and exposure to traffic in general, remained associated with adverse birth outcomes, showing that the broader public health issue has not been fully resolved.
CHAPTER 4: HIGH-TRAFFIC ROADS AND ADVERSE BIRTH OUTCOMES: COMPARING BIRTHS UPWIND AND DOWNWIND OF THE SAME ROAD
INTRODUCTION
TRAP emissions have been associated with numerous adverse birth and early life health outcomes, including low term birth weight,19,92 increased odds of preterm birth,92,93 cardiac94,95 and neural tube birth defects,94 and increased rates of childhood asthma incidence96 and hospital admissions.97 With a projected 22% increase in VMT over the next 30 years,98 traffic is likely to be a persistent concern for perinatal health for the foreseeable future.
While there is evidence that TRAP harms perinatal health,8 many challenges remain in quantifying its causal relationship with adverse birth outcomes. Traffic-related exposures, in general, represent a diverse combination of air pollutants99 (e.g., brake dust,19 nitrogen oxides,92,93,95 particulates) and other environmental noise,19,20 light pollution,100 neighborhood context.20 These exposures are negatively correlated with distance to roadways,101 positively correlated with VMT,102 and highly correlated spatially and temporally with each other. Road density, VMT, and traffic emissions are also correlated with socioeconomic status,101 with persistently marginalized communities (e.g., lower-income, race/ethnicity) experiencing disproportionately greater exposures.18 These co-occurring conditions create a challenging source of confounding (by either co-exposure or socioeconomic status) that is often an issue in epidemiological studies of traffic air pollution. Given the rapid conversion of the vehicle fleet to electric vehicles over the next decade,103 determining how much air pollution emissions contribute to perinatal health outcomes is central to policy and planning.
Wind can be used to disentangle TRAP effects on adverse birth outcomes from other traffic co-exposures and sociodemographic confounders. Neighbors living on the same street but in opposite wind directions (i.e., upwind/downwind) are likely to experience similar nonair pollution exposures, such as noise, light, green space, and neighborhood socioeconomic status, but significantly differing air pollution exposures, such as nitrogen oxides. Wind determines a component of TRAP exposure and is not directly related to adverse birth outcomes. While wind has previously been used as an instrumental variable,104–107 few TRAP studies have integrated wind into study designs due primarily to computational complexity. One study examined living downwind of highways in Los Angeles (determined from the nearest meteorological station) and mortality rates at the census block level and observed that doubling the percentage of time spent downwind of a highway increased mortality among individuals 75 or older by 3.9–6.4%.107 Integrating wind direction with the spatial layout of traffic sources and other built environment factors that modify dispersion (e.g., building and tree shielding) near the vicinity of individual residences remains computationally challenging for large cohorts. Still, it is necessary when matching high and low-exposure residences in the same neighborhood.
To disentangle TRAP from socioeconomic status and other environmental exposures, we matched maternal residence neighbors living upwind and downwind (within 500 m) of the same high-traffic road (over 25,000 AADT) from 2007 to 2016 (n = 388,316 births). We developed and implemented an exposure assessment approach to calculate hourly wind rows to estimate the percentage of time each pregnancy was downwind of roadways. Maternal residences were matched as predominantly upwind or downwind of the same high-traffic road, considering building and tree shielding. We then determined differences in term birth weight, low term birth weight, preterm birth, and very preterm birth.
STUDY DESIGN AND METHODS
Study Population
We examined all births in Texas from 2007 to 2016 using vital statistical records (n = 3,570,272 births). Individual data from the vital statistics records were used to assess maternal sociodemographic characteristics, risk and protective factors for birth outcomes, and birth outcome measures (birth weight, estimated gestational length). We used residential addresses at delivery to assess TRAP exposures and additional environmental and contextual exposure measures.
High-Traffic Roads
We created a database of annual traffic volumes using data from historical Texas roadway inventory databases,108 where AADT was available. We restricted our analyses to births with residential addresses within 500 m of roadways with AADT ≥25,000. These roadways represent large highways, expressways, and arterial roads with substantial vehicle traffic and associated TRAP levels. High-traffic roads were divided into 10-m road segments for subsequent wind analyses.
Wind Exposure Measures
We downloaded hourly u- and v-wind vectors covering Texas from 2007 to 2016 from the European Centre for Medium-Range Weather Forecasts (ERA5 reanalysis v5).109 For each maternal residence, we calculated the distribution of wind directions during pregnancy at 1 radial degree resolution from the estimated conception date until birth, including both end dates. We then applied a ± 15-degree radial interval and estimated the number of hours during pregnancy the maternal residence was downwind of all roads within each one-degree radial segment (Equation 3, Figure 10A). This process was performed in parallel for 360 radial segments, thus providing complete 360-degree coverage of each maternal residence (Figure 10B).
Estimating wind exposure and building shielding at a maternal residence. (A) Wind direction is converted into a 31-degree interval. All objects within the interval are classified as upwind of the residence. (B) Hourly intervals are summed over the length of pregnancy, creating 360 radial segments. To protect cohort privacy, Figure 10 shows intervals summed over 2018 for EPA air monitor EP_10-3-2004-1. (C) Radial segments were spatially joined to road networks to estimate the hours each road segment was upwind from a residence. (D) Buildings between roads and the residence contributed to building shielding. The radii for all circles in Figure 10 are 500 m. Variables in panels B–D are colored by quantile. (Larkin et al. 2024; Creative Commons License CC BY 4.0.)
where
= number of hours maternal residence i was downwind of high-traffic roads located within radial segment j during pregnancy
ki = kth hour of pregnancy for maternal residence i
mi = number of hours during pregnancy for pregnant woman at residence i
Hours Downwind from High-Traffic Roads
For each residence, we spatially joined the wind direction radial segments with high-traffic roads (AADT ≥25,000 during the birth year) to estimate the number of hours the maternal residence was downwind of each 10-m high-traffic road segment within 500 m (Figure 10C). We then calculated the mean hours and percentage of each pregnancy when the residence was downwind from high-traffic roads (Equation 4), identified the road segment with the most hours upwind from the residence (max upwind road, Equation 6), and calculated the hours and percentage pregnancy the residence was downwind from the max upwind road (Equation 5).
(4)
where
mean hours downwindi = mean number of hours maternal residence i was downwind of 10-m road segments within 500 m
ni = number of high-traffic road segments within 500-m of maternal residence i
li = lth 10-m high-traffic road segment within 500-m of maternal residence i
= number of hours during pregnancy maternal residence i was downwind of li
(5)
where
{dli} is the set of dli values described in Equation 4 for all road segments l within 500 m of maternal residence i
(6)
where
max roadi = road segment with the greatest number of hours upwind from residence i
f(x) = function that returns the id for road segment li given input value dli as described in Equation 4.
If multiple road segments are associated with input x, f(x) returns the id of the road segment closest to the maternal residence.
Building and Tree Shielding
In addition to wind direction, shielding by buildings or trees may influence air pollution dispersion. Annual Texas building footprints from 2007 to 2016 were developed by combining Microsoft Bing building footprints,110 land parcel records from the Texas Natural Resource Information System,111 and building records from CoreLogic,112 which we used to identify building construction dates, building types, and building heights (based on number of stories). Using the road segments described above, we calculated building shielding metrics, which consisted of the percentage of area between the road segment and maternal residence occupied by a building footprint (Figure 10D).
DATA ANALYSIS
Matching Downwind (Exposed) and Upwind (Control) Maternal Residences
Residents in the top and bottom quartile of hours downwind from max road (Equation 5) were categorized as exposed (i.e., top quartile downwind, n = 97,024) and control groups (i.e., bottom quartile downwind, also referred to in this manuscript as the upwind group, n = 97,026). Exposed residences were then matched with up to four control residences (median = 2), prioritizing matches with the lowest match score, calculated using Equations 7 to 10 (Figure 11A). The match score estimates the difference in road and neighborhood-level exposures between exposed and control addresses, except for wind direction (a lower score is a better match). Approximately 98% of matches were neighbors on opposite sides of high-traffic roads (Figure 11B).
Matching downwind and upwind perinatal residences in Texas from 2007 to 2016. (A) Matching maximizes the similarity in distance to road between exposed (downwind) and control (upwind) residences. (B) Matched residences are mostly neighbors on opposite sides of a high-traffic road. (Larkin et al. 2024; Creative Commons License CC BY 4.0.)
where
locality score = difference in distance to the matched road between exposed and control
distmaxee = distance from exposed maternal residence to the max road segment (Equation 5) for exposed residence e
distmaxec = distance from control maternal residence to the max road segment (Equation 5) for exposed residence c
(8)
where
near road score = similarity between exposed and control in distance to nearest road
distneare = distance from nearest road to exposed maternal residence e
distnearc = distance from nearest road to control maternal residence c
(9)
where
birth yeare = birth year at the exposed residence
birth yearc = birth year at the control residence
birth yearc must be ± 4 years of birth yeare
A final 100-m distance-based inclusion criteria was applied to matches before proceeding to epidemiological analyses (Equation 10). Matches were excluded if exposed and control were too dissimilar (delta greater than 100 m) in either distance to the nearest road or distance to the exposed max downwind road.
(10)
Epidemiological Analysis
We tested for differences between matched upwind and downwind maternal residences in term birth weight (delivery at 37–42 weeks’ gestation), term low birth weight (term birth weighing less than 2,500 g), preterm birth (22–37 weeks’ gestational age), and very preterm birth (22–32 weeks’ gestational age) using linear and logistic regression. For term birth weight and term low birth weight outcomes, both exposed and control records were restricted to 37–42 weeks of estimated gestational age before creating matched pairs. For very preterm births, we compared all births >32 weeks to 42 weeks’ gestation to maximum matched pairs. For each outcome, we created two regression models using two covariate sets. The “base” set of covariates included infant sex (female, male), maternal age (linear), birth month (categorical) and year (categorical), distance to nearest road (linear), and county (categorical). The base model for term birth weight also included estimated gestational weeks (categorical) to further separate fetal growth from gestational length. The “full” set of covariates included the base model plus maternal race and ethnicity (see Table 5 for all categories), whether the pregnant individual was born outside of the US (yes, no), source of payment used for delivery (Medicaid, private insurance, self-pay or other), maternal use of the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) during pregnancy (yes, no), maternal smoking status during pregnancy (yes/no), weight gain (linear), month pregnancy prenatal care began (none, months 1–9), and neighborhood income (tertiles by year). We ran models for all matched pairs and tested for interactions between wind direction and air pollution decay gradients by stratifying the fully adjusted regression models by distance to the nearest road.
Table 5.
Descriptive Statistics for Matched Pairs of Pregnant Individuals (n = 41,912 pairs) Living Downwind and Upwind of a Major Road
We conducted several sensitivity analyses to examine our wind exposure measures, matching methods, and potential effect modification by environmental and individual measures:
We performed stratified analyses by race, ethnicity, income, education, and whether the pregnant individual was born in the US.
We substituted the categorical exposed/control variable with the continuous variable of percentage downwind from max road (Equation 5).
We included our building and shielding metrics to examine how wind direction estimates changed.
We implemented more stringent inclusion matching criteria (15, 25, and 50 m) than the 100-m used in our main analyses.
We conducted analyses using 3-year rolling averages to evaluate temporal trends in our estimated associations with living downwind compared to upwind of major roads.
RESULTS
Descriptive Statistics
The selection of our study population by residence within 500 m of a high-traffic road and matched upwind and downwind births is summarized in Appendix B, Figure B1 (available on the HEI website). Characteristics of matched exposed and control birth addresses are shown in Table 5. Exposed residences were, on average, downwind from the max downwind road (Equation 5) 3.5 times longer than their upwind counterparts (24.6% vs. 7.0%) during pregnancy (from the estimated date of conception until birth). Exposed residences were similarly downwind from the average road segment 2.5 times longer than upwind matches (11.3% vs. 4.6%). Building and tree shielding were similar between matched downwind and upwind residences. The percentage of Black, Hispanic, or lower education pregnant people is slightly greater among paired downwind vs paired upwind residences. Descriptive statistics of matched pairs restricted to 37–42 weeks are shown in Appendix B, Table B1).
Linear and Logistic Regression Models
Results of the base and fully adjusted linear and logistic regression models are shown in Table 6. Living downwind was associated with a 13.8 g (95% CI: −20.03, −7.63) and 11.6 g (95% CI: −18.01, −5.21) decrease in term birth weight in the base and fully adjusted models, respectively. There were no significant associations between living downwind and term low birth weight, preterm birth, or very preterm birth.
Table 6.
Relationships Between Birth Outcomes and Wind Direction in Base and Adjusted Models
Base model: Wind exposure (upwind is reference), birth sex, maternal age, birth month and year, distance to nearest road, and a fixed effect for county. The base model for term birth also included estimated gestational age. Full model: Base model + maternal race and ethnicity, whether pregnant individual was born outside of the US, source of payment used for delivery, maternal WIC use during pregnancy, maternal smoking status during pregnancy, maternal weight gain, month of pregnancy prenatal care began, and neighborhood income.
aEvents are the total number of events, not the pairs with an event. Term birth weight: β in grams (95% CI). Low term birth weight, preterm birth, and very preterm birth: OR (95% CI).
Stratifying by Distance to Nearest Road
Associations between birth outcomes and wind direction when stratifying by distance to the nearest road are summarized in Table 7. Residences downwind and within 50 m of high-traffic roads had significantly lower term birth weight by −36.3 g (95% CI: −67.74, −4.93) and statistically nonsignificant increased odds of term low birth weight (OR = 1.95 95% CI: 0.58, 6.51) and preterm birth (OR = 1.25, 95% CI: 0.92, 1.69) compared to upwind residences similarly within 50 m of the road. Models for very preterm birth did not converge for the 0–50 m group, but for 51 to 100 m, the OR for living downwind compared to upwind was 3.68 (95% CI: 1.71, 7.90).
Table 7.
Relationships Between Wind Direction and Birth Outcomes, Stratified by Distance to the Nearest Major Road
Associations between birth outcomes and wind direction stratified by select sociodemographic characteristics are summarized in Appendix B, Table B2. Living downwind was associated with 1.21 (95% CI: 0.69, 2.12), 1.21 (95% CI: 0.94, 1.56), and 2.92 (95% CI: 1.00, 8.53) increased odds of low-weight term birth, preterm birth, and very preterm birth for Black non-Hispanic births. However, for Black non-Hispanic births, living downwind was also associated with a statistically nonsignificant 20.7 g (95% CI: −10.80, 52.21) increased term birth weight. Lower education (high school or less) was also associated with increased odds of preterm birth (OR = 2.69, 95% CI: 0.52, 13.88), low term birth weight (OR = 1.85, 95% CI: 1.18, 2.88), and decreased birth weight (−23 g, 95% CI: −43.51, −2.54). Residents with a high income and living downwind had decreased term birth weight (−13.59 g, 95% CI: −27.55, 0.36) but also reduced odds of very preterm birth (OR = 0.53, 95% CI: 0.29, 0.95) and low-weight term birth (OR = 0.77, 95% CI: 0.58, 1.03). When we examined different match criteria, we generally observed larger associations with more restrictive matches (i.e., <25 m) (Appendix B, Table B3). When replacing the exposure group with a continuous wind variable, for every 10% increase in the days spent downwind of a major road, the term birth weight decreased by −4.96 g (95% CI: −8.43, −1.50). Including the shielding metrics did not change our wind estimates. We conducted analyses using 3-year rolling averages to evaluate temporal trends in our estimated associations with living downwind compared to upwind of major roads, but sample sizes were small, and no clear patterns were seen (Appendix B, Table B4).
DISCUSSION AND CONCLUSIONS
This study contributes to the body of studies using causal inference methods for estimating associations between TRAP and population health outcomes. Specifically, we leveraged wind to remove socioeconomic factors and other traffic-related environmental exposures (e.g., noise) as confounders from the associations between TRAP and adverse birth outcomes.
We observed a positive association between TRAP exposure and adverse birth outcomes, corresponding to existing literature findings. However, the interpretation of our findings is somewhat different. The body of literature often compares traffic air pollution measures at residential addresses across entire cities, assessing exposure variation driven by local, neighborhood, and regional factors. Here, we restricted our study population to birth addresses within 500 m of the same major roads (AADT >25,000) and used wind to differentiate TRAP exposure levels. Our finding of an 11.6-g decrease (95% CI: −18.01, −5.21) in term birth weight for pregnant individuals living predominantly downwind compared to upwind of a matched road represented the impact of wind dispersion of air pollution alone within this heterogeneous group. We also found a strong wind-distance interaction, with larger associations for individuals living downwind and within 100 m of the major road. A recent systematic review of 26 studies that used traffic metrics (e.g., road proximity or traffic density) and adverse birth outcomes113 found that the risk of term low birth weight was associated with traffic density in 500 m, with an estimated increased risk of 1.06 (95% CI: 1.002, 1.121) in higher quality studies. Studies have also observed that individuals living very close to roads (<50 and <100 m) have greater risks, for example, term low birth weight OR of 1.13 (95% CI: 1.04–1.21) for traffic density increases of 100,000 vehicles within 50 m,114 which corresponds roughly to the wind-distance decay findings of our current study. Alternatively, a recent systematic review of TRAP concentrations found that a 5 μg/m3-increase in PM2.5 was associated with a mean difference in term birth weight of −17.3 g per 5-μg/m3. However, no association was observed for NO2, which is surprising as NO2 is often used as a marker of traffic air pollution emissions.25 Our results add new information to this mixed evidence base by demonstrating robust associations between TRAP exposure and term birth weight, especially very near large roadways.
Living downwind may increase exposure to traffic air pollution in general, but susceptibility and competing risks may differ across race, ethnicity, income, education, and other sociodemographic characteristics. We conducted exploratory stratified analyses to examine differences by key individual and neighborhood factors. These analyses suggested that Black non-Hispanic individuals may be particularly susceptible to increased risk from this exposure on preterm (OR 1.21, 95% CI: 0.94, 1.56) and very preterm birth (OR 2.92, 95% CI: 1.00, 8.53). Competing risks may explain the increased term birth weight we also observed among Black non-Hispanic residents living downwind (i.e., if susceptible Black non-Hispanic babies have an increased risk of being born early (and thus excluded from the term birth weight analysis), then there is a lower percentage of susceptible Black non-Hispanic babies that make it to full gestational age (and thus included in the term birth weight analysis). For individuals with a high school diploma or less, we saw larger associations (e.g., −23.0 g (95% CI: −43.51, −2.54) compared to individuals with greater than a high school education (−8.70 g (95% CI: −17.32, −0.08)). Surprisingly, individuals living in high-income neighborhoods had larger associations for term birth weight but not for preterm or very preterm births.
This study has several limitations that should be acknowledged when interpreting our findings:
Wind exposures were derived from meteorological data representing regional wind patterns rather than localized wind direction. High-resolution models incorporating local effects such as wind tunnels and turbulence may improve wind exposure estimates, especially within 50 m of high-traffic roads.
Shielding effects, chemical transformations, and air mixing are potential explanations for decreasing associations between birth outcomes and wind direction with increasing distance from the road. In this study, shielding covariates did not change associations between birth outcomes and living downwind. However, our shielding metrics were calculated independently of wind metrics and did not consider detailed street-level characteristics that may determine air dispersion.
Some exposure datasets (e.g., AADT) were based on annual average measures and did not capture trimester-specific exposures.
Our analysis relied on residential addresses at the time of delivery, and we did not have information on residential mobility during pregnancy, whether women were at home or work, and the infiltration properties of residential homes.115
In conclusion, this study provides strong evidence that pregnant individuals living downwind of high-traffic roads have an increased risk of adverse birth outcomes compared to individuals living upwind of the same roads. Thus, TRAP is associated with adverse birth outcomes, with steep distance decay gradients around major roads.
CHAPTER 5: ISOLATING THE ADDITIONAL INFLUENCE OF TRAFFIC CONGESTION ON INFANT HEALTH
INTRODUCTION
More than 11 million people in the United States live within 150 meters of a major highway and are exposed to elevated levels of traffic-generated air pollution.116 Traffic congestion, defined as roads operating at lower than free-flow speeds due to excess vehicles,117 further contributes to this problem. Congestion is a modern lifestyle inconvenience of great interest to policymakers, researchers, and the public for various health and nonhealth reasons.118 Traffic congestion has increased consistently from 1982 through 2019, costing up to 190 billion dollars annually in delay time and wasted fuel.119
Traffic congestion increases motor vehicle emissions, resulting in higher levels of traffic-related air pollutants. TRAP is a heterogeneous mixture,120 and the concentrations and exact composition of air pollution will vary based on a range of parameters, including the number of vehicles, driving conditions and vehicle speed, fuel combustion, and vehicle fleet characteristics (i.e., age of cars, proportions of cars vs. trucks).117,121 Generally, a higher number of vehicles on the road increases TRAP concentrations, with an exponential decrease in concentrations away from roadways but with a continued elevation above background levels up to 500 meters.122 Congestion can dramatically increase vehicle emissions and local air pollutant concentrations123,124; for instance, emissions measured in passenger vehicles increased by 200% when comparing rush-hour driving to free-flow driving conditions.125
A large body of epidemiological literature addresses the influence of TRAP on reproductive and infant health outcomes.7,126 However, only limited work has examined the potential additional influence of traffic congestion, primarily due to the challenges of measuring congestion accurately for large geographic areas.127 Across a wide variety of countries and settings, living near a major road during pregnancy, as well as exposure to specific traffic air pollutants, has been associated with decreased birth weight and increased risk of preterm birth.7,71,72,120,126,128,129 Notably, most of the exposure assessments used in these studies were based on proximity to major roads or models of specific traffic pollutants (e.g., NO2). Very little of this evidence incorporated traffic congestion in the exposure measures.130 No studies have examined the impact on adverse birth outcomes when congestion is added to traffic volume and “normal” air pollution levels from background traffic. This impact has important policy implications, as congestion can be modified through policy and infrastructure changes potentially independent of those targeting vehicle volumes or tailpipe emissions (e.g., electronic tolling, congestion pricing, and roadway capacity improvements).
Here, we leveraged novel congestion measurement data derived from connected device information, in the form of driving volumes and speeds, linked to a population-based retrospective cohort of births in Texas. Using this novel database, we examined associations between metrics of congestion and term birth weight. By systematically examining exposure to both vehicle volume and congestion, and controlling for background air pollution levels, transportation noise, and other environmental co-exposures, this study provides important, policy-relevant insights into the extent that traffic congestion might contribute to adverse reproductive health outcomes, and whether health impacts should be included in evaluations of the benefits of policies aiming to reduce congestion.
STUDY DESIGN AND METHODS
Study Population
We leveraged birth certificate data from the Texas Department of Health and Human Services to extract information on residential birth address, demographics, risk factors, and birth outcomes. Each maternal address at time of delivery was geocoded to examine births between January 1, 2015, and December 31, 2016 (n = 820,328). We removed records for which the maternal address at delivery could not be precisely geocoded (n = 43,306, 5.3%), and any births that were missing one or more key fields: birth weight, gestational age, maternal age, and number of fetuses in this pregnancy (n = 1,082, <0.1%). Births were excluded based on improbable birth weight (<500 g or >5,000 g, n = 1,989, <0.1%), maternal age (<10 years old or >65 years old, n = 0, 0%), and any deliveries with multiple births (n = 24,834, 3.0%). Because our outcome was term birth weight, we removed births with gestational age <37 weeks or >42 weeks (n = 63,217, 7.7%). We then removed births with maternal residences >1,000 m from at least one road in the congestion database with >500 vehicles per day. We used these births to derive exposure measures (n = 67,788, 8.2%). This ensured our exposed and unexposed populations were similar regarding geographic distribution within Texas. Note that this database contained congestion measurements for road segments smaller than major roadways; thus, we could retain most of the births in the cohort. In addition, we removed births in counties with fewer than 100 term births (n = 2,673, <0.1%). Finally, we removed births (n = 36,317, 4.4%) that were missing covariates needed for our adjusted regression models (i.e., we conducted a complete case analysis). Applying all these criteria yielded 579,122 births for analysis.
Traffic Congestion Exposure Assessment
We used the database Texas’ Most Congested Roadways, developed by researchers at the Texas A&M Transportation Institute.35,36 We relied on data for the 2015 to 2016 period. Each annual database contained detailed information related to traffic volume (including trucks), delays, fuel type, and emissions. Congestion metrics were derived down to the road section level based on congestion performance measure calculations, based on the state’s roadway inventory and a proprietary data source of GPS speed reports generated from individual vehicles and cell phones (i.e., connected vehicle data). We examined several different measures of congestion. Vehicle miles traveled, or VMT, is the number of vehicles traveling a given road (i.e., a metric for overall traffic), multiplied by the length of each road segment. Annual delay-per-mile is a metric in person-hours of delay that occurs per mile along a given road (i.e., a metric for traffic congestion). Total person-hours of delay were calculated based on traffic volumes for each 15-minute time interval, calculated from average daily traffic counts using hourly volume profiles, and corresponding travel speed, measured from the vehicle movement database. We then compared peak morning and evening commute travel speeds to free-flow (low volume) travel speeds using speeds from 10 pm to 6 am, or the speed limit for each road section as an upper limit. Delay metrics were calculated for all vehicles and trucks only. Parsing congestion related to trucks versus cars provided additional insights into truck-specific emissions, such as diesel particulate matter and benzene, both of which are known reproductive toxicants. We also examined a metric that measured the greenhouse gas emissions (carbon dioxide [CO2] equivalents) from all traffic on a road segment, as well as from congestion-specific traffic. Greenhouse gas emissions were calculated as pounds of CO2 using the EPA’s Motor Vehicle Emission Simulator (MOVES-2010) model that incorporates vehicle volumes, vehicle emission rates, climate data, and vehicle speeds. We ran the model for each 15-minute period for the measured speed and the corresponding free-flow speed to calculate the excess CO2 produced during congestion. Full details of our congestion calculation methods are published elsewhere.36
Each maternal residence was assigned congestion values based on all roads within a given buffer of a home (100 m, 300 m, 500 m). To account for multiple roads and segment lengths, we calculated the total length of each road segment that fell into the buffer and weighted the road length by each congestion metric in separate exposure estimates. Our metrics can be interpreted as the total VMT, person-hours of delay, and greenhouse gas emissions within 100, 300, and 500 m of a home address. We present exposure measures for the 500 m buffer in the main manuscript, while the influence of the 100 m and 300 m buffer distances are in supplemental tables.
Infant Health Outcome Assessment
We ascertained infant health outcomes from the birth certificate. Term birth weight (primary outcome) is the weight reported at birth among infants born at 37 to 42 weeks of estimated gestation. Although we hypothesized that other infant health outcomes might be influenced by traffic congestion (e.g., preterm birth, term low birth weight), we focused on term birth weight as a sensitive marker of growth restriction25 that could be assessed using a continuous model. Due to measured congestion data being available only for 2015 and 2016 and limited residential proximity to highly congested roadways, we did not have sufficient power to examine other outcomes.
Covariate Assessment
We used covariates from the birth certificate to ascertain additional information on the pregnant individual-infant dyad. For this analysis, we examined the following covariates: county of maternal residence at delivery, birth year, maternal age, infant sex, maternal race, maternal ethnicity, maternal educational attainment, method of payment for delivery, maternal cigarette usage, month of prenatal care initiation, prepregnancy body mass index, and infant gestational age at delivery.
In addition to the detailed data provided on the birth certificates, we integrated external data sources to estimate household characteristics, residential mobility, neighborhood context, and background (noncongestion) levels of noise and air pollution. To determine whether the pregnant individual lived in a single-family home, mobile home, or multifamily home (e.g., apartments), we linked the maternal address to housing valuation data.131 We used the housing valuation data to check whether there was a housing transaction during the pregnancy period.131 For residential locations with a transaction, the pregnant individual likely changed residences during pregnancy. In such cases, the exposure assessment for early pregnancy could be inaccurate. We accounted for the neighborhood’s deprivation status via the 2015 Area Deprivation Index, which provides a metric ranking census tracts within a state on socioeconomic characteristics.132,133 We included residential green space using the yearly NDVI from Landsat 8 images within 500 m of maternal addresses. The transportation noise level was assessed using 2016 US Department of Transportation national transportation noise map predictions, which we assigned to maternal addresses.134 We also linked annual NO2 and PM2.5 in the year before the delivery and ultrafine particles (UFPs) in 2017 (due to data limitations) using existing hybrid land use regression model estimates.135–137 Many existing air pollution exposure models miss sharp changes in the exposure gradient.138,139 Therefore, we controlled for background exposure levels so our regression results could parse out the additional influence of congestion.
DATA ANALYSIS
We first calculated descriptive statistics for the cohort by levels of traffic delay exposure and examined the relationships between different traffic-related exposure measures. We also built restricted cubic splines to visually examine the relation between metrics of traffic delay and term birth weight by percentile of exposure in the cohort. This examination informed our decisions regarding cut points in the quintile models. These models contained a covariate for the VMT within the respective buffer distance of the residence, allowing us to disentangle the influence of traffic congestion from vehicle volume.
Next, we implemented a set of linear regressions to estimate to what extent, if any, congestion exposures were associated with term birth weight. We ran an unadjusted model (Model 1), a model adjusted for individual covariates (Model 2), and a model adjusted for individual covariates and other environmental co-exposures (Model 3). Model 1 only contained a covariate for the VMT within the respective buffer distance of the residence, allowing us to isolate the influence of traffic congestion from vehicle volume. Model 2 contained the following covariates: county of maternal residence at delivery (indicator for each county), birth year (indicator for each year), maternal age (continuous), infant sex (male, female), maternal race (White, Black, American Indian or Alaskan Native, Asian, Pacific Islander, other race), maternal ethnicity (Hispanic/Latina, not Hispanic/Latina), maternal educational attainment (8th grade or less, 9th–12th grades without a diploma, high school graduate or equivalent, some college credit but no degree, associate’s, bachelor’s, master’s, doctorate), the method of payment for delivery (Medicaid, private insurance, self-pay, other), maternal cigarette usage (yes, no), month of prenatal care initiation (indicator for “no reported care” or each month 0–9), prepregnancy body mass index (underweight, normal weight, overweight, obese), and infant gestational age (indicator for each week). Model 3 contained the covariates from Model 2, plus the following covariates: VMT within the respective buffer distance of the residence, area deprivation index (state ranking for Texas), transportation noise, PM2.5, NO2, UFPs, and green space. We included VMT within the same buffer distance as the main congestion exposure measure, such that we estimate only the effects from congestion metrics for traffic delay, trucks-only traffic delay, and greenhouse gas emissions. We included annual concentrations of NO2, PM2.5, and UFP air pollution concentrations, which can be interpreted as noncongestion-related ambient concentrations, because these predictive models do not include congestion as a predictor.135,136 Model 3 thus isolated the impact of congestion on term birth weight, in addition to vehicle volume and background air pollution and noise levels.
Sensitivity Analyses
We conducted several sensitivity analyses to better understand which sources of bias were present in our data. First, we restricted our sample to pregnant individuals without a housing transaction during their pregnancy period, removing pregnant individuals who likely moved during pregnancy and could introduce exposure misclassification for whole pregnancy exposures. Second, we restricted analyses to pregnant individuals who did not report a labor induction, as several medical reasons for induction are likely unrelated to TRAP.140 Third, we stratified the population by reported occupational status (homemaker or unemployed vs. currently employed), as traffic congestion can negatively affect individuals who commute during their pregnancy, relative to women who stay home. At the same time, our TRAP exposure metrics for traffic congestion would be more accurate for pregnant individuals who spent more time at home. Fourth, we stratified the population by household type (single-family vs. multifamily home) to determine how housing type influenced our results. Housing type could operate as an effect modifier for air pollution exposures141 or could be an additional surrogate for socioeconomic status.142 Fifth, we restricted the population by maternal birth location to only pregnant individuals who reported being born in the United States, reducing measurement bias that could stem from timing and access to prenatal care as well as difficulties related to accurate gestational age dating. Sixth, we examined the influence of socioeconomic and demographic disparities on our effect estimates by implementing models restricted by education (high school or less, some college or higher), payment mechanism for delivery (private insurance, Medicaid), WIC usage (no, yes), and maternal race and ethnicity (non-Hispanic White, Hispanic/Latina). While further disaggregating these groups would be ideal, we were limited to broad categories due to sample size.
RESULTS
Descriptive Statistics
Figure 12 illustrates AADT and the percentage of greenhouse gas emissions from congestion for Houston during 2016. Traffic volume and total delay were highly correlated (0.81 for 500 m buffers around a pregnant individual’s home address), as were traffic volume and congestion-related emissions (r = 0.75 for 500 m buffers). However, comparing volume and congestion yielded unique geographic patterns. In addition, these data showed that congestion-related emissions on specific road segments could exceed 50% of total emissions. The correlations between total delay and background levels of PM2.5, NO2, and UFP air pollution were 0.18, 0.42, and 0.48 for the 500-m buffer area, respectively. This geographic variation provided an opportunity to isolate the unique contributions of congestion, in addition to vehicle volume and background air pollution levels, on infant health.
A comparison of traffic characteristics in Houston, Texas, 2016. Displayed data show road segments by traffic volume and percentage of total greenhouse gas emissions due to congestion. (Willis et al. 2022; Creative Commons License CC BY-NC 4.0.)
In total, 579,122 term births were included in our analysis (Table 8). While mean gestational ages were similar across quintiles of exposures for traffic delay within 500 m of their maternal residence at delivery, birth weights were on average 29 g lower in the highest quintile of exposure compared to the lowest quintile of exposure, and trends corresponded to the low birth weight percentages. When comparing the lowest to the highest quintile of traffic delay, we observed that pregnant individuals were more likely to be non-White race, Hispanic ethnicity, use WIC, and have normal weight in the prepregnancy period. These individuals were also less likely to be highly educated or to report smoking during pregnancy. In the group experiencing the highest quintile of traffic delay, pregnant individuals were less likely to live in a single-family home and more likely to live in a structure built before 1978, although we noted that these housing-related data were unavailable for pregnant individual-infant dyads. Increasing levels of all environmental co-exposures corresponded to higher tertiles of traffic delay.
Table 8.
Study Characteristics of Pregnant Individuals and Infants Born Between 37–42 Weeks Gestation, Texas, USA, 2015 to 2016
GED = general equivalency diploma; NVDI = normalized difference vegetation index; UFP = ultrafine particles; WIC = Special Supplemental Nutrition Program for Women, Infants, and Children.
a 509,794 records for this characteristic due to only a subset of linked tax records containing these data from CoreLogic.
b 474,148 records for this characteristic due to only a subset of linked tax records containing these data from CoreLogic.
c Derived from the 2015 Area Deprivation Index.132,133
d Derived from the Center for Air, Climate, and Energy Solutions land use regression model.135,137
e Derived from the Center for Air, Climate, and Energy Solutions land use regression model.136
f Derived from the National Noise Transportation Map.134
Traffic Congestion and Birth Weight
In restricted cubic splines, we observed a nonlinear association between congestion metrics and term birth weight (Appendix C, Figure C1 {available on the HEI website}). We determined that our exposure metrics should be divided into quintiles for linear regression models based on a visual examination of where the spline knots fell. We found consistent associations in the models between congestion metrics and term birth weight in unadjusted (Model 1), individual-variable (birth certificate) adjusted (Model 2), and environmental co-exposure adjusted (Model 3) (Table 9). In models adjusted for individual variables, environmental co-exposures, associations were –3.43 (95% CI: –7.24, .38), –6.55 (95% CI: –10.35, –2.75), –5.79 (95% CI: –9.72, –1.86), –5.48 (95% CI: –9.63, –1.32), and –8.93 (95% CI: –14.08, –3.79) across quintiles. Similar associations were found for truck delay and congestion emissions. We observed slightly larger associations when we examined pregnant individuals living within 100 m and 300 m of major roadways (Table 9; Appendix C, Tables C1–2). We also saw strong associations between an interquartile range increase in PM2.5, NO2, and UFP concentrations and reduced term birth weight in adjusted models (Appendix C, Table C3).
Table 9.
Associations Between Metrics of Traffic Congestion Within 500 m of a Maternal Residence and Term Birth Weight, Texas, USA, 2015 to 2016
We found similar magnitude results as described for our main adjusted regression model when we conducted extensive sensitivity analyses (summarized in Figure 13). Traffic exposure misclassification due to a pregnant individual moving during pregnancy is a common concern in birth cohort studies that rely on home addresses at the time of delivery. We used property data linkage to identify movers and then restricted analyses to pregnant individuals who lived at an address without a housing transaction during pregnancy (these individuals were less likely to have moved during pregnancy). This restriction did not change the overall interpretation of our results. Similarly, among pregnant individuals who reported being homemakers or unemployed (and therefore likely to spend more time at the home location used for exposure assignment), we observed a larger estimated effect size than those presented as our main adjusted results above. Yet for pregnant individuals who reported being employed, associations were largely similar, although with wider confidence intervals. Among pregnant individuals whose addresses correspond to a single-family home, we found results similar to the adjusted main model. However, we found no association for pregnant individuals who lived in multifamily homes or apartments. For pregnant individuals born in the United States with births not induced, the results were similar to those in our adjusted main model. For sociodemographic characteristics such as payment mechanism for delivery, WIC usage, education level, and maternal race and ethnicity, the results were also similar to the main adjusted model.
Adjusted associations between the highest quartile of total delay within 500 meters of a pregnant individual’s address and term birth weight for different sensitivity analyses. Model 1 contained a covariate for vehicle miles traveled within the respective buffer of the residence. Model 2 contained the following covariates: county of maternal residence at delivery (indicator for each county), birth year, birth month, maternal age, infant sex, maternal race, maternal ethnicity, maternal educational attainment, method of payment for delivery, maternal cigarette usage, month of prenatal care initiation, prepregnancy body mass index, infant gestational age, and total vehicle miles traveled within the respective buffer of the residence. Model 3 contained the covariates from Model 2, plus the following: area deprivation index (state ranking for Texas), transportation noise, previous year concentration of PM2.5, previous year concentration of NO2, previous year concentration of UFPs, and green space within 500 m of the residence (measured by NDVI). All restricted models use the covariates from the fully adjusted model (i.e., Model 3), except the covariate related to the restriction was removed if included in fully adjusted model specifications. (Willis et al. 2022; Creative Commons License CC BY-NC 4.0.)
DISCUSSION AND CONCLUSIONS
Our analysis of term births suggests that traffic congestion, as measured by delay-per-mile and greenhouse gas emissions from congestion, might adversely affect term birth weight, beyond the impacts of traffic volume, background air pollution, and noise on nearby roads. Specifically, traffic delay within 500 m of a maternal residence at delivery was associated with an estimated mean decrease of 9 g when comparing the highest to the lowest quintile of exposure after adjusting for individual covariates and environmental co-exposures. Pregnant individuals who lived closer to these roadways (i.e., within 100 m or 300 m) experienced slightly larger impacts. Although congestion and total traffic volumes were highly coupled, these results suggest that there were additional health impacts from congestion-related air pollution emissions — separate from traffic volume. This has two important policy implications: (1) congestion can be reduced through specific infrastructure and policy changes independent of those targeting vehicle volumes and tailpipe emissions, and (2) health impacts should be included when evaluating the benefits of policies aiming to reduce congestion.
Our present results do not suggest that redesigning highways and expressways with more lanes and higher throughput is the solution to improving population health outcomes associated with TRAP. Rather, we argue that vehicle congestion is an understudied and quantified component of TRAP that can be easily intervened upon at a local level. Reducing traffic congestion could provide provisional population health benefits as other fundamental modifications to the road system and transportation sector are designed and implemented. These modifications include highway reclamation projects, active transportation and public transportation, and a transition to electric vehicles. Researchers predict that the prevalence of internal combustion vehicles on the road might dramatically decrease over the next few decades.143,144 A change in the vehicle fleet mix would reduce the toxicity of tailpipe emissions because battery electric vehicles do not produce incomplete combustion byproducts when burning gasoline or diesel. However, a fully electric vehicle fleet would not entirely remove the hazards related to traffic congestion and resulting air pollution. Tailpipe emissions represent only one component of the complex mixture of TRAP that could be exacerbated by traffic congestion. For instance, electric vehicles are, on average, 197 to 362 kg heavier than their internal combustion counterparts, largely due to battery size.145,146 The extra battery weight causes additional wear on the vehicle’s brakes and tires, producing air pollution at higher concentrations (e.g., PM2.5).146,147 Therefore, unless substantial investment is made in reducing traffic volume and delay, the impending transition away from internal combustion vehicles would likely attenuate, but not remove, the adverse associations between traffic congestion and infant growth presented in our analysis.
This study is among the first epidemiological studies to leverage detailed traffic congestion metrics at a large geographic scale, and our results align with previous health impact assessments of traffic congestion, environmental pollution, and human health outcomes.7,126 Thanks to substantial investments in the Texas’ Most Congested Roads database,35 we could quantify congestion metrics and exposures for a large population-based cohort of pregnant individual-infant dyads across the entire state for singleton births. The use of connected device data to quantify vehicle travel patterns, types, volumes, and speeds is becoming more accessible to exposure scientists. We recommend further incorporating connected device data into air pollution exposure models and other epidemiological studies to confirm our findings. Current approaches to estimate TRAP exposures using road proximity or density, VMT, and land use regression models, do not capture all air pollution exposure contributions from congestion, and regional air pollution monitoring data does not address localized traffic-related exposure hotspots.
Long-term improvements have reduced vehicle exhaust over the last few decades,148 and previous work has shown that air pollution reductions paralleled improvements in infant birth weight.72 Therefore, focusing on improving traffic congestion could benefit population health and help planners reduce local exposure inequalities.149,150
Our results align with literature demonstrating that TRAP and proximity to major roads are associated with adverse reproductive and infant health outcomes, including reduced birth weight.7,126 For example, a meta-analysis found a –25 g reduction in birth weight (95% CI: –11.5, –4.8) per 20-ppb increase in NO2, showing a substantial association between markers of TRAP and infant health.7 In comparison, in our current analysis, we observed a –25 g reduction in term birth weight for a 20 ppb increase in NO2. Also, we found an association with the additional influence of traffic congestion. Notably, this association remained after adjusting for vehicle volume and NO2, PM2.5, and UFP background concentrations. This result suggests that some local impacts near congested roadways are not captured in air or noise pollution models138,139 and that interventions to reduce congestion could provide cobenefits for infant health. A few studies have indirectly examined traffic congestion on reproductive outcomes using natural experiments.73,151 For example, one study used the conversion from tollbooths to electronic tolling to examine how reductions in traffic congestion could influence infant health; the authors hypothesized that the switch from a stop-and-go toll to an overhead toll would reduce traffic congestion.73 They found a large reduction in low birth weight infants among pregnant individuals who resided within 2 km of a toll plaza during pregnancy, comparing the 3 years before and after the switch. Our present analysis expands upon this body of work by directly measuring traffic congestion and conducting a spatially based comparison of pregnant individuals living closer to or farther away from congestion. More research is needed, yet these results suggest that the pregnancy period might be a particularly vulnerable time and that adverse birth outcomes are a sensitive marker for the impacts of traffic congestion-related pollution.
We conducted several novel sensitivity analyses that addressed the potential influence of unmeasured confounders on our results. For example, we used maternal occupational data from the birth certificate to examine whether a pregnant individual was likely to not be away from home for most of the working day. We also used property data to identify housing characteristics and home sales data to determine whether the pregnant individual likely lived at this address throughout the pregnancy. These potential confounders are well-documented in existing literature,152–154 but few studies have quantified this source of exposure misclassification, and none on a population-level scale as we accomplished here. We found that addressing these hypothesized sources for exposure misclassification did not change the meaning of our results overall. However, the magnitude of the estimated effect sizes was somewhat larger among pregnant individuals who did not report being employed at the time of delivery, aligning with the notion that exposure misclassification is reduced among pregnant individuals who spend more time at home. We also observed large differences by housing type, with much larger associations for manufactured and mobile homes, associations consistent with the overall results reported for single-family homes. Yet we found less evidence of an association for multifamily or apartment buildings. Further research is needed to determine if household characteristics influence the exposure to TRAP (such as the indoor/outdoor ratios of pollutants) or capture unmeasured socioeconomic influences that modify or confound the associations between traffic air pollution and adverse birth outcomes.
Our study had several limitations. First, our traffic congestion data were derived at an annual temporal scale. Therefore, it represents longer-term exposure to traffic congestion. Although there are seasonal patterns to traffic,155 we assigned annual estimates to the entire pregnancy and were unable to examine trimester-specific congestion measures. Second, vehicle fleet mixtures vary regionally, including the types of vehicles on the road (e.g., sedans, sport utility vehicles [SUVs], diesel trucks) and their age (i.e., model year). We hypothesize that there might be exposure misclassification by the type and age of the fleet mixture, which we could not assess in this analysis. The EPA MOVES-2010 model used default parameters for these characteristics to estimate total and congestion-specific CO2 emissions. Thus, it did not consider exposure variation stemming from the fleet mixture. Our models included a county-fixed effect, which removed the potential exposure misclassification from fleet differences across regions of Texas but did not alleviate problems related to fleet heterogeneity within a given county. Third, the birth certificate data did not capture residential changes occurring during pregnancy, which could have introduced misclassification into our exposure assessment.152,153,156 We overcame this limitation to some degree using housing transaction data and showed that our results were similar when we restricted the sample to pregnant individual-infant dyads without a housing transaction at their residential address during the pregnancy period. Further, maternal addresses at delivery are likely to be most accurate during the third trimester, which is the period during which air pollution likely has the greatest effect on term birth weight.55,157 Fourth, given the nature of administrative data, we lacked information on some potential individual confounding factors that could have influenced this association, such as nutrition or lifestyle data — assuming these are indeed related to living near congestion. We included detailed individual information on the sociodemographic characteristics of pregnant individuals, which also might capture some major lifestyle factors. When we restricted our sensitivity analyses to individual race and ethnic groups, education levels, insurance type, and WIC use, they were similar to our overall results. Traffic congestion might also operate as an instrumental variable for personal air pollution exposures (individuals are less likely to select houses based on traffic congestion than on distance to major roads), thus reducing the potential influence of confounding from individual behavioral differences.158 Nevertheless, residual and unmeasured confounding cannot be ruled out in observational studies. Fifth, our analyses controlled for gestational age among term infants. This decision means that our results cannot be interpreted by gestation length (i.e., we can only make conclusions about infant growth); however, it also reduces the potential for bias in our analysis.26,159 With these limitations in mind, we note that our results were highly robust to several sensitivity analyses.
In conclusion, our study provides important new evidence that traffic congestion is associated with adverse infant health outcomes, as measured by reductions in term birth weight. This effect was in addition to the effects of total traffic volume on nearby roads and background levels of air pollution and noise. Therefore, programs and policies to reduce traffic congestion might have positive cobenefits for infant health regarding birth weight. In the subsequent two chapters, we evaluate two different methods for congestion reduction: implementation of tolling and roadway capacity improvements.
CHAPTER 6: TRANSITION TO ELECTRONIC TOLLING AND ASSOCIATIONS WITH ADVERSE BIRTH OUTCOMES
INTRODUCTION
Traffic congestion (i.e., vehicle delay caused by excess vehicles on a portion of roadway at a particular time) results in over 4.3 billion hours of commuter time lost each year, costing over 101 billion dollars in lost productivity.119 The amount of traffic delay quadrupled in the United States from 1982 to 2019.117,119 Additionally, traffic congestion poses hazards to population health that compound those from free-flowing vehicles.87,160–162 In particular, vehicle air pollution emissions and local air pollution concentrations are generally higher when traffic is slowed down and with stop-and-go traffic, compared to free-flowing speeds.123,124 For instance, passenger vehicle emissions increase by approximately 200% during rush-hour driving.125 Policymakers have addressed the issue of congestion through investments in roadway infrastructure — by adding lanes, improving intersections, or switching to electronic tolling. However, whether these investments reduce congestion and ameliorate the health effects of congestion-related air pollution has not been well-evaluated in previous research.163
Tolling (i.e., a usage fee for driving on a specific road) is a program aimed at reducing traffic congestion. Traditionally, tolls were paid as vehicles stopped at a booth. By the mid-1990s, several states began distributing electronic transponders to drivers that would allow them to simply slow down when passing under electronic tolling mechanisms to pay the toll, or else the mechanisms would take photos of license plates of vehicles traveling full speed and collect the tolls by mail.164 The introduction of tolling in general might decrease local TRAP levels due to reductions in on-road congestion; electronic tolling specifically might reduce delays at tollbooths,73,165–167 both of which might reduce adverse health impacts to local populations.
Few studies have explicitly examined the health impacts of electronic tolling. A previous study showed that switching from tollbooths to an overhead electronic system (i.e., E-ZPass, a multistate electronic tolling program in the eastern United States) was associated with a reduction in preterm birth and low birth weight infants born to pregnant people who resided within 2 km of toll plazas in New Jersey and Pennsylvania.73 Similarly, another study showed that congestion pricing (a time-varying charge for entering an area) yielded improvements in the rates of acute asthma attacks among children living near roads in Stockholm, Sweden.165 While the results of these studies are compelling, this work only examined one specific implementation of electronic tolling. Further, no research has evaluated the effects on congestion and human health caused by other tolling modifications, such as adding toll lanes to an existing highway to improve traffic flow and eliminate excess air pollution caused by congestion. Beyond the lens of TRAP, many studies have taken advantage of natural experiments, such as air pollution reductions from Olympic events or shutdowns of coal-fired power plants. These natural experiments have shown that large-scale reductions in local air pollution were associated with improvements in pregnancy and pediatric outcomes.53,55,168–171 Given that electronic tolling is being implemented around the world, quantifying its impact on population health is key to understanding its potential health benefits.
In this analysis, we conducted a population-based birth cohort study to determine the associations between implementing a variety of tolling systems and changes in traffic congestion and adverse birth outcomes across Texas from 1999 to 2016. We built upon the existing literature by examining several different types of real-world tolling implementations and following the chain of accountability as closely as possible given data availability.46,49 By systematically examining a wide range of tolling projects across Texas, our study adds to the body of literature that examines the extent to which congestion reduction projects might yield health-protective benefits for local populations.
STUDY DESIGN AND METHODS
Study Design
We examined toll implementation in Texas using an accountability framework to leverage available birth data (1999 to 2016), roadway vehicle volume (1999 to 2019), and congestion (2015 to 2019).172 We used the publicly available shapefile of all toll road segments in Texas from the Texas Department of Transportation (TxDOT) to ascertain where current tolling roads were located at the end of 2020 (Figure 14).173 However, these data did not contain information regarding when the tolling systems were implemented, and tolling data are not centrally located in a single database for the state. We used internet-based resources (e.g., TxDOT reports, local news archives, construction documentation) to find spatial-temporal information on each tolling construction project and systematically built a database at the road segment level that connected to the primary shapefile. When possible, we extracted distinct data related to the toll construction and toll opening dates. For a small subset of tolling systems that switched from tollbooths to electronic systems (Figure 14), we used the geocoded location (i.e., a point instead of a line segment) with exact dates that tollbooths were removed in favor of an overhead electronic system (data are exclusive to Houston). In total, we examined three distinct types of tolling implementations: (1) switching an existing toll road to exclusively electronic tolling (e.g., the whole road segment now uses electronic tolling, with no cash payments), (2) adding tolled express lanes to an existing road (e.g., adding toll lanes to an existing highway), and (3) switching an existing tollbooth to electronic tolling (i.e., the physical booth was replaced with overhead sensors and cars were no longer required to pay at the booths). We followed the “chain of accountability” framework,49 examining how toll implementation resulted in changes to (1) roadway traffic and congestion and (2) local adverse birth outcomes. Limited data availability precluded analyses of outdoor air pollution changes near tolled roads.
Maps of tollways and booths in selected metropolitan areas of Texas. White lines denote roadways. Red lines denote tollways. Yellow point markers denote the subset of tollbooths with geocoded locations. (Willis et al. 2024; Creative Commons License CC BY 4.0.)
Measurement of Traffic Volume and Congestion
We linked each toll road to the TxDOT Roadway Inventory files, which contained roadway segment-level information for major roads managed by TxDOT. The data included historic measures of AADT for trucks and passenger vehicles from 2000 to 2019, the associated highway system and functional classification (i.e., major arterial road, local road), the number of lanes, and the segment length. We imputed missing values of AADT over time via linear interpolation at the road segment level. For a supplementary analysis, we also linked road segments to an annual measure of delay, measured in lost person-hours, from the Texas’ Most Congested Roadways database developed by researchers at the Texas A&M Transportation Institute.35,36
Traffic Volume Analyses
We first identified roadways that experienced one of the three types of tolling implementations (the “treated roads”) based on the previously described toll database. Because roads with added toll infrastructure might be systematically different than roadways without any toll infrastructure, we identified a comparable control group of road segments that did not experience a change in tolling infrastructure that were more than 2 km from a tolled road but were otherwise like the road segments that transitioned to tolling. To understand the effect of tolling on traffic, we retained all three categories of tolling changes and matched all treated segments with control segments using a combination of Mahalanobis Distance Matching and exact matching. For each treated segment, we exact-matched to one control segment on categories of highway system classification (a categorical classification of roads that includes US highways, state highways, state loops, business state roads, and ranch-to-market roads), and a variable indicating which of the 25 TxDOT regional districts the segment is located in. We then used Mahalanobis Distance Matching to match on the following characteristics: segment length (continuous), functional system (interstate, other freeway or expressway, other principal arterial, minor arterial, major collector), proportion of commercial trucks in the AADT (continuous), average of AADT in the past 3 years (continuous), and average change in AADT in the past 3 years (continuous). The control segments were matched based on the values for the year before treatment began on the treated segment. Control segments selected via the matching process were weighted based on the number of treated segments to which they were matched. In a supplemental analysis, we repeated this matching process separately for roads that were not treated but intersected a treated road or were within 2 km of a treated road, allowing us to explore potential spillover effects to nearby areas from tolling.
We implemented a linear regression in an event-study framework to examine the impact of tolling infrastructure on traffic volumes, measured by AADT per lane, for the directly affected roadways. The event-study design compared the difference in traffic on the treated roads and the matched control group of roads in each year-long period relative to the year the tollway implementation process began, allowing for temporal variation in the impact of tolls on traffic. The event study effectively controlled for any pre-existing differences in traffic between control and treated roads that were not accounted for by our matching exercise. Even so, we further included the covariates of segment length and functional system with year and county fixed effects in our event study.
We included a supplementary analysis similar to the main specifications. In this analysis, we redefined the exposure groups to be an indicator equal to one for road segments with a tollbooth that switched to electronic tolling (i.e., treated) or equal to zero for segments without a tollbooth on them but that were within 2 km of a booth that switched (i.e., the control).
Measurement of Birth Outcomes
We leveraged birth certificate data from the Texas Department of Health and Human Services, where we obtained information on residential birth addresses, demographics, risk factors, and birth outcomes. Between 1996 and 2016, there were 8,114,440 recorded births in Texas. We excluded participants based on the following criteria: imprecise geocode for the maternal address at delivery (n = 890,842, 11.0%), missing information related to delivery (i.e., birth weight, gestational age, maternal age, or birth type) (n = 47,096, 0.6%), birth weight outside of reasonable bounds (i.e., less than 500 g or greater than 5,000 g) (n = 17,591, 0.2%), gestational age outside of reasonable bounds (i.e., less than 22 weeks or greater than 42 weeks) (n = 10,447, 0.1%), deliveries with multiple infants (n = 213,250, 2.6%), maternal addresses more than 5 km from a toll road (n = 4,315,775, 53.2%), births in counties with too few deliveries for analysis (i.e., no exposure or outcome variation) (n = 84,150, 1.0%), births outside a 5-year span before or after the toll road was implemented (n = 1,872,078, 23.1%), and births missing analytic covariates (i.e., complete case analysis) (n = 37,932, 0.5%). In total, these criteria yielded 625,279 births for analysis, spanning 1999 to 2016.
We obtained data on birth weight in grams and gestational age from the birth record. For our outcomes, we examined term birth weight (i.e., birth weight among only births whose gestational age was 37–42 weeks), term low birth weight (i.e., birth weight less than 2,500 g among only births whose gestational age was 37–42 weeks), preterm birth (i.e., birth before an estimated gestational age of 37 completed weeks), and very preterm birth (i.e., birth before an estimated gestational age of 32 completed weeks). Measures for birth weight were outcomes related to the pathway of intrauterine growth restriction, while measures of preterm birth were outcomes related to triggers for premature labor.67,174 Ambient air pollution had been associated with all of these outcomes in previous literature, although the consistency was somewhat stronger for birth weight than preterm birth.7,175
DATA ANALYSIS
We used a quasi-experimental design with a DiD estimator to compare the risk of adverse pregnancy outcomes, before and after the implementation of a tolling project, relative to a spatial-temporal control group where tolling never occurred (Figure 15). An advantage of the DiD framework is that it accounts for the secular trends in infant health that are contemporaneous with our traffic-related exposures of interest but not directly influenced by traffic-related exposures.176 Births were defined as exposed if there was a tolling project within 2 km of the maternal address before birth. We examined exposed distances of 0–0.5 km, 0.5–1 km, 1–1.5 km, and 1.5–2 km separately to evaluate air pollution distance decay gradients. Births were defined as a part of the control groups if they were born prior to tolling within 5 km of the maternal address or if the closest tolling project at the time of birth was within 2–5 km of the maternal address. Each birth was linked to the nearest road segment with a toll project and the nearest tollbooth conversion point, where the earliest date was used for the temporal term in our primary analyses.
Conceptual schematic of the DiD study design for the influence of tolling on traffic congestion and infant health. The figure visually depicts the spatial structure of the main analysis. For the traffic analyses, roads in red were considered treated, while roads in the red buffer were intersecting or partially treated. Examples of matched roads are shown in orange. For the birth outcome analyses, residences in the red buffer were considered treated, while residences in the blue buffer were controls. The supplemental tollbooth analysis used roads and births within the black 2-km buffer of a tollbooth. (Willis et al. 2024; Creative Commons License CC BY 4.0.)
We used covariates from the birth certificate for sociodemographic characteristics of the pregnant individual and their infant: birth year, birth month, county of residence, maternal age, maternal race, maternal ethnicity, maternal educational attainment, maternal foreign-born status, month of prenatal care initiation, maternal smoking status, maternal weight gained during pregnancy, and infant sex. We also linked census tract median household income to each birth, where we interpolated estimates via linear regression for intercessional years. The term birth weight analyses also included a covariate for infant gestational age at delivery. To describe local context, we linked each residential location to data for outdoor ambient air quality, green space, and AADT counts.76,177,178 We did not adjust for these variables in our analyses as they could mediate the association between electronic tolling and birth outcomes.
We tested these key assumptions required to pursue a DiD analysis176,179–181: (1) stable population composition in the near (i.e., 0–0.5 km) and far groups (i.e., 2–5 km) before and after the tolling implementation (i.e., no spillover effects where tolling differentially affected the demographics of the far distance group); and (2) parallel trends in our near (i.e., 0–0.5 km) and far groups (i.e., 2–5 km) before the commencement of the tolling construction (e.g., term birth outcomes were not differentially changing in areas without tolling). To test the stable population assumption, we compiled descriptive statistics for each of our population characteristics in the near and far groups before and after tolling begins to examine differences among groups. To test the parallel trends assumption, we expanded our simple DiD design into an event-study design by replacing the “after” indicator with indicators for each year to get an annual average treatment effect and controlling for county of residence at birth and birth year. We considered the parallel trends assumption to be valid if the annual average effect before treatment is not statistically differentiable from the null.
Using the DiD design, we implemented a series of regression models to examine associations of residential proximity to a toll road project during pregnancy and adverse birth outcomes, broadly aligning with methods from a similar project.73 Each model included an indicator term for residential proximity: 1 if the birth occurred within 0–0.5 km, and 0 if the birth occurred within 2–5 km of an active or future toll road for our primary models. The models also included an indicator term for the time period: 1 if the birth occurred after the toll road was opened for operation, or 0 if the birth occurred beforehand. They also included a product interaction term between the proximity and time period indicator variables.
The primary estimate of interest was the coefficient for the product interaction term, which represents the difference in the association between proximity and the outcome once the toll road was opened relative to the association between proximity and the outcome before the toll road was operational. We used separate models to estimate interactions with further proximities defined as 0.5–1 km, 1–1.5 km, and 1.5–2 km, assuming observed associations would grow increasingly small at further distances. We also explored the association between each of the three types of tolling implementations and infant health outcomes to examine potential heterogeneity in effect estimates. For the tollbooth removal subtype, we combined the 0–0.5 km and 0.5–1.0 km groups due to small sample sizes. For a sensitivity analysis, we examined the exposed distance of 0–150 m (i.e., the hypothetical highest exposure zone), where there could be a stronger overall association.
RESULTS
Traffic Volume and Congestion
We observed minimal traffic reductions after the tolling implementation (Appendix D, Figure D1). In the years before switching an existing toll road to electronic tolling (i.e., the whole road segment now uses electronic tolling) or adding tolled express lanes to an existing road, we found little evidence of differential trends over time in traffic volume per lane. Once tolling began, traffic on the toll road itself exhibited minimal changes. For instance, traffic in the first year after tolling initiation increased by a statistically insignificant 0.19 log points on the toll roads, while by Year 3, the point estimate declined to 0.03 log points. Our supplemental analysis showed no significant rerouting when we examined intersecting roadways and roadways partially affected by toll changes, although due to pretrends, the results were less reliable (Appendix D, Figure D2). When we specifically examined tollbooths, we found a similar lack of change in traffic volume and delay (Appendix D, Figure D3).
Birth Outcomes
Among 625,279 births included in the analysis, 8.0% were preterm, and 1.1% were very preterm. Among 575,148 term births, mean term birth weight was 3,366 g, of which 2.3% were low birth weight. Table 10 summarizes the characteristics of our study population by near and far groups. While we found some evidence of changes in infant health outcomes before and after the implementation of tolling among births within 500 m (e.g., 8.3% vs. 7.5% for preterm births), the magnitude of change was similar in the 2–5 km group (e.g., 8.3% vs. 7.7%). Trends in maternal characteristics were somewhat similar between the births near to and far from tolled roads, except for prenatal care, although less consistent in the further groups (e.g., 1.0–1.5 km, 1.5–2.0 km). Environmental characteristics followed expected patterns, where we found higher levels of annual average air pollution and traffic congestion among births in closest proximity to tolling. After the tolled roads were implemented, both annual averages of air pollution and traffic congestion near the birth residences were slightly reduced. When we examined trends in infant health outcomes before and after the implementation of tolling, we found little evidence of an association before tolling began (i.e., the parallel trends assumption was not violated). However, estimates were not precise for all years (Appendix D, Figures D4–6).
Table 10.
Selected Descriptive Statistics of Infant Health, Maternal Sociodemographic, and Environmental Characteristics by Timing and Location of Tolling Relative to the Reported Address at Delivery, Texas, 1999 to 2016
b Data from the Center for Air, Climate and Energy Solutions.
c Data from Landsat 8 images.
d Data from the Texas Department of Transportation.
For term birth weight outcome measures, we found that residence within 500 m of a tolled road after implementation was associated with a 4.5-g reduction (95% CI: –11.7, 2.6) in term birth weight and an OR of 1.00 (95% CI: 0.89, 1.13) for term low birth weight in fully adjusted models (i.e., Model 3), compared to term births with residences 2–5 km from a tolled road before and after tolling implementation (Table 11). Similar results were found as far out as 2 km. For term birth weight, we found a similar lack of evidence of an association in fully adjusted models when we separately examined new tolled lanes (0.9 g, 95% CI: –11.8, 13.6), and booth to overhead (–2.6 g, 95% CI: –23.7, 18.5) (Appendix D, Table D1). However, we saw some evidence of a decrease in term birth weight for added overhead tolls (–10.3 g, 95% CI: –23.1, 2.5). Results were largely similar at other buffer distances. For term low birth weight, we found similarly null results in fully adjusted models when we explored added overhead tolls (0.92, 95% CI: 0.75, 1.14), new tolled lanes (1.07, 95% CI: 0.87, 1.31), and booth to overhead 1.12, 95% CI: 0.81, 1.56), although estimates were somewhat imprecise (Table 12; Appendix D, Table D2).
Table 11.
Associations Between the Implementation of Tolling by Distance from the Road and Adverse Birth Outcomes in a DiD Study Design, Texas, 1999 to 2016
CI = confidence interval; DiD = difference-in-differences.
Results correspond to logistic regressions for binary outcomes (term low birth weight, preterm birth, very preterm birth) and linear regressions for continuous outcomes (term birth weight). Coefficient of interest is a product interaction term between proximity and time period. Model 1 was unadjusted. Model 2 was adjusted for year, county, and birth month. Model 3 was adjusted for year, county, birth month, maternal age, maternal race, maternal ethnicity, maternal education, maternal foreign-born status, month of prenatal care start, maternal smoking status, maternal weight gained during pregnancy, median household income, and infant sex. All term birth weight models also include gestational length (weeks).
Table 12.
Associations Between the Implementation of Tolling in the Near Distance from the Road and Adverse Birth Outcomes in a DiD Study Design, Disaggregated Tolling Types, Texas, 1999 to 2016
CI = confidence interval; DiD = difference-in-differences.
Results corresponded to logistic regressions for binary outcomes (term low birth weight, preterm birth, very preterm birth) and linear regressions for continuous outcomes (term birth weight). Coefficient of interest was a product interaction term between proximity and time period. Model 1 was unadjusted. Model 2 was adjusted for year, county, and birth month. Model 3 was adjusted for year, county, birth month, maternal age, maternal race, maternal ethnicity, maternal education, maternal foreign-born status, month of prenatal care start, maternal smoking status, maternal weight gained during pregnancy, median household income, and infant sex.
For preterm birth outcome measures, we found that residence within 500 m of a tolled road after implementation in fully adjusted models (i.e., Model 3) was associated with an OR of 0.99 (95% CI: 0.92, 1.05) for preterm birth and 1.00 (95% CI: 0.84, 1.18) for very preterm birth, compared to term births with residences 2–5 km from a tolled road before and after tolling implementation (Table 2). Similar results were found as far away as 2 km. For preterm births, we found a similar lack of evidence of an association in fully adjusted models at 500 m when we examined added overhead tolls (1.03, 95% CI: 0.92, 1.15) and new tolled lanes (1.01, 95% CI: 0.90, 1.13), with similar results at further buffer distances (Table 12; Appendix D, Table D3). However, we found a borderline inverse association in fully adjusted models changing from booth to overhead at 1 km (0.84, 95% CI: 0.70, 1.00) and 1.5 km buffers (0.85, 95% CI: 0.73, 1.00). Associations dissipated at further buffer distances. For very preterm birth, we found null results in fully adjusted models when we explored added overhead tolls (0.87, 95% CI: 0.63, 1.19), new tolled lanes (1.09, 95% CI: 0.80, 1.48), and booth to overhead (0.71, 95% CI: 0.43, 1.18), although estimates were imprecise (Table 2; Appendix D, Table D4).
In a sensitivity analysis of the main model, we found that residence within 150 m of any type of tolled road in fully adjusted models (i.e., Model 3) was associated with an 8.3-g reduction (95% CI: –25.1, 8.4) in term birth weight and an OR of 0.91 (95% CI: 0.70, 1.20) for term low birth weight, compared to term births with residences 2–5 km from a tolled road before and after tolling implementation (Appendix D, Table D5). For preterm birth outcome measures, we found that residence within 150 m of a tolled road after implementation in fully adjusted models (i.e., Model 3) was associated with an OR of 0.96 (95% CI: 0.82, 1.11) for preterm birth and 1.19 (95% CI: 0.81, 1.75) for very preterm birth, compared to births with residences 2–5 km from a tolled road before and after tolling implementation.
DISCUSSION AND CONCLUSIONS
In this large population-based retrospective cohort study of births in Texas, we found little evidence that the implementation of tolling overall was associated with improvements in infant health outcomes. However, our exploratory results suggest that replacing tollbooths with overhead electronic tolling might yield reductions in preterm birth among infants born to individuals residing within 1 km of a booth (i.e., a protective effect). The results also suggest that adding toll lanes could decrease term birth weight (i.e., an adverse effect). These results suggest that tolled roads might not broadly produce the hypothesized cobenefits for infant health.
Our present study was partially motivated by a highly influential 2011 paper on replacing toll plazas with overhead tolling in a relatively rural part of Pennsylvania and New Jersey.73 Here, the implementation of E-ZPass (an example of a booth to overhead tolling implementation) yielded a moderate reduction in preterm birth and low birth weight among pregnant individuals residing within 2 km of a toll plaza, compared to those with residences at 2–10 km from the toll plaza. Our results demonstrated similar (but not identical) reductions in preterm birth, and this association was concentrated within 1,000 m of the tollbooth location. In contrast to the 2011 results, we did not find evidence of a protective association between tollbooths and clinically relevant metrics of infant growth. We also extended our exploration of tolling implementations other than changing from booths to an electronic tolling scenario. We did so by examining adding overhead tolls and adding a new tolled lane, neither of which produced the hypothesized health benefits. Also, we examined pregnant individuals living within 300 m of tolling (where traffic air pollution levels are highest and where the most change was expected to occur). However, the associations were null for all birth outcomes. Our results might differ from the previous study as our population lived in a more urban context, as tolling in Texas is mostly concentrated in large metropolitan areas. To our knowledge, these two studies were the only ones that directly examined the potential health benefits of changes to electronic tolling using empirical data. Because electronic tolling is being implemented in several metropolitan areas, future work is needed to understand the extent to which different tolling scenarios could yield improvements in population health, which is critical to determining which new programs or policies could tout health-protective benefits.
We could not reliably assess changes in air pollution that might be associated with tolling changes due to a lack of monitors near roadways. There are few outdoor air monitors in relevant locations with temporal alignment. This omission meant that we could not complete the full chain of accountability in our current study.46,49 However, we examined traffic congestion, roadway construction, and air pollution in a separate analysis of roadway construction, where we broadly found reductions in outdoor ambient concentrations of NO2 but not fine particulate matter.163 Previous work used a single monitor located within 2 km of a toll plaza in Pennsylvania, finding reductions in NO2 compared to randomly selected control monitors (10.8% decrease) and no change in sulfur dioxide (SO2) (a negative control as this pollutant was not expected to be associated with road TRAP).73 However, we found little evidence of an association between electronic tolling and adverse birth outcomes compared to the previous seemingly robust negative associations. This suggests that the contextual exposures or air pollution mixture might differ between Pennsylvania and Texas.
When interpreting the results of this study, there are some key limitations to consider:
The detailed data on tolls was collected from various primary sources and was unavailable, incomplete, or at a coarse level for some tolls in Texas. Therefore, our definition of tolling exposure on road segments was more aggregated than ideal, and we could not capture potentially interesting variations in traffic and pollution exposures at different stages of the tolling process (e.g., during construction, staggered implementation). However, we noted that the tollbooth analysis used more precise data, yielding broadly null findings in most scenarios.
Our analysis used birth certificate data, where the only location information was maternal address at delivery. Pregnancy is a highly mobile period in the life course, meaning that many couples move residences.153,182 Further, we could not consider time-activity patterns in subjects’ daily lives,154 as we only knew where the pregnant individuals lived. Residential mobility and time-activity patterns could introduce a degree of misclassification into our analyses if movement differed by exposure status, although we expected this bias to be small or null.
Birth certificate data were predicated on the pregnancy lasting long enough to yield a vital statistic record, meaning that we could not consider early pregnancy losses. This situation could have introduced a degree of live birth bias,26,183,184 which could underestimate the association if the exposure (e.g., air pollution, noise) yielded a higher incidence of unmeasured pregnancy loss.
Some of our event-study graphs were quite imprecise, as evidenced by the relatively wide confidence intervals. The possibility remains that there could be pretrends that we could not measure due to limited data.
For the DiD model, we could only test according to the data we had access to. Thus, we acknowledge that there might have been unmeasured confounding factors relevant to the exposure-outcome relationship.
Despite these limitations, accountability studies, such as the present analysis, are critical tools for understanding to what extent actions translate into the promised societal benefits (e.g., reductions in traffic, improvements in population health).49 Many congestion reduction initiatives are complex and cover long implementation periods. Therefore, robust results depend on leveraging a study design that can control for temporal trends unrelated to any changes in exposure.46 In our present analysis, among births within 500 m of a tolled road, the prevalence of preterm birth was 8.31% and 7.50% before and after tolling was implemented, respectively. If we only compared the prepost data near tolled roads, we would erroneously conclude that there was a 0.81% difference in preterm birth prevalence. However, when we controlled for temporal trends, in general, using a control group living within 2–5 km of the tolled roads, we observed a similar magnitude difference in preterm birth prevalence of 0.63% in both groups pre- and post-treatment. This was the case although the control group further from these roads would not have been expected to have experienced substantial changes in air pollution exposure due to the toll implementation. By applying a DiD study design, we demonstrated no consistent evidence of an association between overall tolling and preterm birth in our main models. Such confounding by temporal trends has been demonstrated in other contexts. For instance, the first study on a coal ban in Dublin showed a 10.3% reduction in cardiovascular-related mortality.52 However, in a follow-up reanalysis that controlled for temporal trends in cardiovascular-related mortality in other cities, this association was reduced to essentially zero.51 These applications provide context to demonstrate that methods, such as DiD analysis, are important in the body of accountability literature.
In conclusion, we found little evidence of a broad association between residential proximity to a tolled road and improved infant health outcomes using a DiD study design. In a subgroup analysis, we observed a small reduction in preterm birth (i.e., a protective effect) among infants born to individuals who resided within 1,000 m of a tollbooth that was switched to an electronic overhead toll. However, we also found some evidence of decreased term birth weight (i.e., an adverse effect) among infants born to individuals who resided within 500 m of a road that added overhead tolling. Overall, our results demonstrated that tolled roads might not produce cobenefits for infant health, highlighting the need for further accountability research.
CHAPTER 7: ROADWAY CAPACITY IMPROVEMENTS AND CHANGES TO TRAP AND ADVERSE BIRTH OUTCOMES
INTRODUCTION
Between 1982 and 2019, the number of hours Americans spent sitting in traffic increased four-fold from 1.7 billion to 8.9 billion.185 In addition to the monetary costs of wasted travel time,186 increased traffic congestion creates excess air pollution beyond emissions from free-flowing traffic.187–189 For example, vehicle emissions increase by up to 200% during rush hour driving.125 Roadway capacity enhancements offer an alternative approach to tolling to increase vehicle flow and reduce congestion. Suppose roadway construction (e.g., widening roads to add capacity or improving intersections to increase flow) effectively reduced congestion. In that case, such projects could also create health-related benefits to nearby residents from reduced TRAP.
Few studies have used natural experiment designs to address potential causal relations between roadway infrastructure improvements and population health outcomes. Several studies suggested that public transportation infrastructure was associated with improved local air quality.190–192 A subset of these studies linked the magnitude of improvements in air quality to pre-existing mortality estimates and health benefits.191,193–195 To our knowledge, only two studies have directly examined the impact of infrastructure improvements on population health using primary estimates for health outcomes.196,197 The earlier of these two studies examined the conversion of cash tollbooths to overhead electronic tolling systems and found small reductions in low birth weight among infants born to persons who resided within 2,000 m of a toll plaza during pregnancy after the conversion occurred,196 lending credibility to our hypothesis that reduced congestion could improve infant health. However, the later study demonstrated that the tollbooth type could influence the magnitude of congestion reduction and its cascading impact on health, finding little evidence of a clear association between electronic tolling and birth outcomes.197
Using a novel dataset of traffic construction projects between 2007 and 2016 in Texas, we examined the extent to which a large sample of diverse roadway construction projects affected changes in traffic volume, congestion, and local air quality and subsequent impacts on infant birth outcomes. We also examined how this influence might differ during construction (when congestion has been found to increase) and after construction (when congestion has been found to decrease).198 Our study adds to the small body of literature examining how roadway construction might reduce vehicle congestion and the cobenefits of reduced TRAP to local population health.
STUDY DESIGN AND METHODS
Study Setting
Texas was ideal for this study as it contains many congested metropolitan areas and spends more money annually on road construction than any other state.185,199 Prior research by our team showed that in Texas, traffic construction projects reduced congestion by around 33% to 52% post-construction and reduced congestion-related NO2 by around 13%,198 suggesting a plausible causal mechanism that could connect road construction projects to birth outcomes. Texas has a large population (approximately 3.5 million births over our study period) that could have experienced TRAP changes from roadway construction.
Road Construction Projects
The TxDOT Project Tracker database contains all roadway construction projects (contracts for roadway improvement and maintenance) on the State Highway System between 2004 and 2020 (n = 48,521 construction projects).200 The database describes the construction activities, construction costs, project location, and start and end dates of each project. We used the TxDOT’s Roadway Inventory for geospatial data on the locations and attributes of roads, including variables describing each segment’s length, number of lanes, highway system classification, and functional classification. Building upon our prior work,198 we defined six broad categories of projects that could affect congestion: widening existing roads, building new roads, improving bridges, installing intelligent transportation systems, improving intersections, and installing or upgrading traffic signals. We restricted the database to construction projects likely to influence congestion (e.g., those with a total project cost of over $5 million; excluded landscaping and routine road maintenance, repair, or reconstruction without meaningful modification) that could be matched to a major road with traffic volume data.198 Due to sample size constraints for specific project types, we examined roadway widening and intersections in our stage one analysis and combined all project types in our main analyses (Appendix E, Table E1 [available on the HEI website]).
Congestion Change Study Design
A full description of our Stage 1 analysis of changes in traffic congestion and air pollution due to major roadway infrastructure improvements in Texas was described in a previously published paper.163 Briefly, we created a dataset of construction projects that could be linked to existing EPA air pollution monitoring data for NO2 and PM2.5 and three variables operationalizing congestion: average annual daily traffic (AADT), AADT per lane, and delay in hours. We used DiD methods to estimate the effect of widening and intersection improvements on roadway congestion and local air pollution.
Birth Outcome Study Design
Our main analytic approach leveraged large roadway construction in a quasi-experimental research design with a DiD estimator.176,201 The DiD approach is commonly used in the epidemiological literature to estimate the effects of exposure changes over time while also controlling for differences between the control and treatment groups.171,197,202 Briefly, we compared birth outcomes of pregnant people living closer to those living further from a roadway project before, during, and after construction. The DiD approach addressed pre-existing differences in risk for the outcome studied when comparing mothers who lived near construction projects to those who lived slightly further away. The DiD estimator does not require the absence of baseline differences in maternal characteristics or levels of exposure to traffic, but rather that the trend in risk over time would be similar in both groups if the time-varying exposure (shock) did not affect the risk in the index group. A key advantage of DiD is that it does not require comparable baseline exposures or characteristics in levels. Instead, selection over time changes similarly in both groups.203
Thus, the DiD estimator allowed other secular or seasonal differences to affect both groups as long as they similarly affected health outcomes over time. We controlled for county, birth month, and birth year in our regression models to ensure that comparisons were made within the same policy environment, season, and birth cohort.
We started with the population of all births in Texas from 2007 to 2016 using Vital Statistics records from the Texas Department of Health and Human Services (n = 4,031,366 births). This period corresponds to the timing of construction projects in our database (the majority occurred between 2010 and 2019) and allows for several years of preconstruction for all births. We excluded births based on the following criteria: imprecise geocode for the address at delivery (n = 327,092, 8.11%), missing information related to delivery (i.e., birth weight, gestational age, pregnant individual age, or birth type) (n = 2,921, 0.07%), birth weight outside of reasonable bounds (i.e., less than 500 g or greater than 5,000 g) (n = 9,563, 0.24%), gestational age outside of reasonable bounds (i.e., less than 22 weeks or greater than 42 weeks) (n = 317,232, 7.87%), and deliveries with multiple infants (n = 117,041, 2.90%). These restrictions yielded 3,257,517 births available for analysis.
We used the residential address at delivery to calculate the distance to the nearest major road and the distance to all road construction projects before, during, and after the gestation period within 1,000 m. Major roads are defined by the highway system in TxDOT’s Roadway Inventory file and include federal and state-managed “on-system” roads that include US highways, state highways, state loops, business state highways, and ranch-to-market highways.
Our main analysis was limited to births within 5 years before and 3 years after construction starts (analysis of during construction) and 5 years before and up to 2 years after construction ends (for the analysis of after construction). This restriction retained years around construction start and end with a sufficient number of births, as this number declined in the after-period due to project starts or completions that occur near the end of our study period in 2016. Finally, we made two distance restrictions: (a) residences within 1,000 m of one road construction project (n = 261,442 births) and (b) residences within 1,000 m of one or more road construction projects (n = 408,979 births). Our main analysis was restricted to residences that were exposed to only one construction project within 1,000 m of the residence, thus avoiding the challenges of isolating the influence of a single project within a given distance when projects were spatially colocated. However, we also present results for multiple projects as many of the births were within 1,000 m of multiple roadway construction projects. In this case, an eligible birth was considered to occur in the after-period if all projects within 300 m were completed (for the treatment group) and all projects within 1,000 m were completed (for the control group).
The birth outcomes studied were the following: term birth weight (i.e., birth weight among births with gestational age 37–42 weeks), term low birth weight (i.e., birth weight less than 2,500 g among births with gestational age 37–42 weeks), preterm birth (i.e., birth before an estimated gestational age of 37 completed weeks), and very preterm birth (i.e., birth before an estimated gestational age of 32 completed weeks). We assessed the sociodemographic characteristics of pregnant individuals, risk factors, and birth outcomes from the vital statistical records. We conceptualized term birth weight as an independent outcome measure for adverse pregnancy outcomes. Birth weight, independent of adjustment for gestational duration, represents the combined influence of potential biological pathways on gestational length and growth restriction.24 To distinguish these pathways, we examined preterm births (including very preterm births) and birth weight (including low birth weight) restricted to term births.
DATA ANALYSIS
We used a DiD quasi-experimental design to isolate construction projects’ impact on congestion during and after construction. Because road segments with construction projects were likely different from the segments not experiencing construction, we used a combination of exact and Mahalanobis distance matching to select a group of control road segments to estimate the trends that road segments with construction would have followed in the absence of construction.204 See previous publication163 for a full description of the DiD methods used.
For the health analysis, we compiled descriptive statistics comparing the characteristics and birth outcomes for individuals living close to a project (0–300 m) to those living further away (300–1,000 m), before and after a project construction start date (Table 13). This analysis was restricted to one project within 1,000 m of the residence with a start or end date within 5 years of birth.
Table 13.
Descriptive Statistics of Birth Outcomes, Pregnant Individual Characteristics, and Traffic Volume Near House by Timing and Location of Construction Project Relative to Reported Address at Delivery
AAPI = Asian American and Pacific Islanders; VMT = vehicle miles traveled; WIC = Special Supplemental Nutrition Program for Women, Infants, and Children.
Our main regression model estimate was the DiD estimator, which included a binary variable for close proximity to a project (300 m), binary variables indicating time (during and after), and interactions between the indicators for “close and during” and “close and after.” The main analysis’s comparison group comprised those living within 300–1,000 m (Figure 16). In models adjusted for the pregnant person’s characteristics, we included infant sex (male/female), age at delivery (linear), race and ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, Asian non-Hispanic, other race non-Hispanic), educational attainment (less than high school, high school graduate, some college, college graduate, college plus), born outside of the US (yes/no), the source of payment used for delivery (Medicaid, private insurance, self-pay or other), Special Supplemental Nutrition Program for WIC use during pregnancy (yes/no), smoking status during pregnancy (yes/no), weight gain during pregnancy (linear in lb), month that pregnancy prenatal care began (none, months 1–9), and neighborhood income in the lowest tertile (yes/no). To maintain the same set of included births, we included a missing data indicator for each covariate. We also included the distance to the nearest major road and the distance to the closest major road squared. The regression model for term birth weight also included estimated weeks of gestation (indicator for each week) to further separate fetal growth from gestational length. We included county, birth month, and year covariates to control for unobserved differences between counties, seasons of birth, and birth cohort, and standard errors were clustered to the residential county.
Example map of construction projects from Houston illustrating treatment (exposed) and control groups used for the main specification. Brighter colors denote overlapping projects. (Hill et al. 2024; Creative Commons License CC BY 4.0.)
The quasi-experimental design and DiD estimator required several assumptions, including no compositional changes in the groups compared and parallel trends before the exposure began.176 To test for compositional changes in characteristics that are coincident with road construction projects, we estimated regressions with the pregnant individual characteristics as the dependent variable and the DiD estimator as the independent variables, also controlling for county, birth month, and birth year. To test for parallel trends, we employed an event-study design comparing the difference in birth outcomes near treated roads (within 300 m of a project) and the control group (residences 300–1,000 m from a project) in each 6-month period, relative to the project start date for during and project end date for after (excluding the “during” period). The differences are presented relative to an omitted period (6 months before the project start date). The event-study regressions also controlled for characteristics of the pregnant individual, county, birth month, and birth year.
We performed multiple sensitivity analyses. First, we conducted regression modeling for two additional distances, 0–150 m and 0–500 m, to assess the robustness of our main results. Second, we altered our main model in three ways: (1) restricting the sample to include only individuals without missing covariates; (2) restricting the comparison group to within 1,000 m of a project and 500 m of a major road; and (3) comparing to a fixed comparison group of pregnant people living within 500–1,000 m of a single construction project. In addition, we implemented models where we removed the temporal restriction by including births that occurred 5 years after construction ended (for the after-period).
RESULTS
Stage 1 Analysis of Congestion and Air Pollution Change
On average, over the construction period, we found that widening increased delay by 42% (95% CI: 30, 56%) (Figure 17), but intersection projects did not affect delay. On average and over the first 3 years post-construction, we found that widening reduced delay by 33% (95% CI: –41, –24%) and reduced NO2 levels within 500 meters by 13% (95% CI: –22%, –2%). After construction, intersection projects reduced delay by 52% (95% CI: –65, –35%) and NO2 levels within 500 meters by 12% (95% CI: –18%, –5%). The Stage 1 analysis results are published elsewhere.142
The effect of widening construction projects on AADT, AADT per lane, and delay.
Birth Analysis Descriptive Statistics
Table 13 shows birth outcomes, population characteristics, and traffic volumes for the exposed and control groups before and after construction start dates for those exposed to a single roadway construction project. Birth outcomes and characteristics of pregnant individuals were generally similar between the exposed (within 300 m) and control (300–1,000 m) groups, but there are a few differences to note. There were more White non-Hispanic births during (34.4% within 300 m, 36.1% 300–1,000 m) and after construction (39.4% within 300 m, 34.2% 300–1,000 m), than before for both distances (30.5% within 300 m, 31.8% 300–1,000 m). Private insurance status was lower within 300 m for all three timepoints (29.1% before, 32.7% during, and 34.5% after) than for 300–1,000 m (35.9% before, 40.2% during, 38.6% after). For all groups, approximately 6% of births had any missing covariates. As expected, traffic volume vehicle miles traveled, or (VMT), within 300 m of the residence was much higher for the exposed group (VMT of 20,919) than for the control group (VMT of 4,851).
We further examined compositional changes for select sociodemographic characteristics. We did not observe any differences by sociodemographic characteristics, insurance status, WIC, or prenatal care during the construction period. However, in the after-construction periods among those living within 300 m of a construction project, we observed differences relative to before construction and compared to those living within 300–1,000 m of a construction project. The population was comprised of 18% (95% CI: 1.07, 1.31) more White non-Hispanics, 17% (95% CI: 0.73, 0.95) fewer Black non-Hispanics, and 12% (95% CI: 1.02, 1.23) more privately insured individuals (Table 13).
Event Studies
We present event studies for the main model for births exposed to only a single construction project (Figures 18 and 19) and for births exposed to one or more construction projects (Appendix E, Figures E1–2). Preperiod trends were relatively stable for term birth and term low birth weights during and after construction periods, while the estimates for preterm and very preterm births were less precise. We observed increasing odds of term low birth weight during the construction period, with estimates reaching a maximum of 59% (95% CI: 0.97, 2.63) increased odds in the second year after construction began. We also observed a decrease in term birth weight in Years 2 and 2.5 after construction started, with estimates reaching a maximum of –37.6 g (95% CI: 0.01, –80.6) in the second year. However, we did not observe consistent patterns for preterm or very preterm birth. In the after-construction period, we continued to see increased odds of term low birth weight, but estimates were imprecise. This pattern disappeared when we examined multiple projects in supplemental analyses (Appendix E, Figure E3).
Associations between exposure to single construction projects and adverse birth outcomes at different points in time from before to during construction periods. Results correspond to a linear regression (risk estimates and 95% CI) for term birth weight and logistic regressions for term low birth weight, preterm birth, and very preterm birth. The event-study estimator contained a dummy for within 300 m, a dummy for during construction, and the interactions between within 300 m and dummies for each 6-month period relative to the project start date. The omitted category was 6 months before the project started. The models adjusted for infant sex, pregnant individual age, race, ethnicity, educational attainment, foreign-born status, source of payment used for delivery, WIC use, smoking status, weight gained during pregnancy, the month of prenatal care start, and neighborhood income in the lowest tertile. We also adjusted for county, birth month, and birth year. The term birth weight models also included gestational length (weeks). (Hill et al. 2024; Creative Commons License CC BY 4.0.)
Associations between exposure to single construction projects and adverse birth outcomes from before to after construction. Results correspond to a linear regression (risk estimates and 95% CI) for term birth weight and logistic regressions for term low birth weight, preterm birth, and very preterm birth. The event-study estimator contained a dummy for within 300 m, a dummy for each 6-month period relative to the start (for the before period) and after the construction end date (for the after-period), and the interactions between within 300 m and dummies for each 6-month period relative to the project start/end dates. The omitted category was 6 months before the project start date. These models excluded the during-period (project start date through project end date). The models adjusted for infant sex, pregnant individual age, race, ethnicity, educational attainment, foreign-born status, source of payment used for delivery, WIC use, smoking status, weight gained during pregnancy, the month of prenatal care start, and neighborhood income in the lowest tertile. We also adjusted for county, birth month and birth year. The term birth weight models also included gestational length (weeks). (Hill et al. 2024; Creative Commons License CC BY 4.0.)
Main Model
Broadly, unadjusted and adjusted models yielded similar results (Table 14). During project construction, we observed an OR of 1.19 (95% CI: 1.05, 1.36) for term low birth weight in the adjusted model. The period during construction is associated with a 6.44-g decrease (95% CI: –17.18, 4.30) in term birth weight in the adjusted model. Estimates for preterm birth and very preterm birth were imprecise. In the 2 years after construction ended, we observed an OR of 1.24 (95% CI: 0.95, 1.63) in adjusted models for the prevalence of term low birth weight among pregnant individuals within 300 m compared with those 300 to 1,000 m from a project (Table 14). We implemented a model allowing multiple projects within 1,000 m (compared to our main model, which was limited to one project) (Table 14). We observed similar increased odds of term low birth weight during construction as described for our main model, but no evidence of associations after construction ended.
Table 14.
Associations Between Single Construction Projects Within 300 m of Homes and Adverse Birth Outcomes in a DiD Study Design
CI = confidence interval; DiD = difference-in-differences.
Results correspond to linear regressions for term birth weight and logistic regressions for term low birth weight, preterm birth, and very preterm. The DiD estimator contained a dummy for within 300 m, a dummy for during construction, a dummy for after construction, and the interactions between within 300 m and during and after. The left side shows the unadjusted model (including controls for county, birth month, and birth year), and the right side shows the model adjusted for infant sex, pregnant individual age, race, ethnicity, educational attainment, foreign-born status, source of payment used for delivery, WIC use, smoking status, weight gained during pregnancy, month of prenatal care start, and neighborhood income in the lowest tertile. We also adjusted for the distance to the nearest major road, distance to the nearest major road squared, county, birth month, and birth year. The term birth weight models also included gestational length (weeks).
Sensitivity Analysis
We conducted several sensitivity analyses:
We examined additional distances to define our exposed group (using 150 m and 500 m proximity to a road construction project) (Appendix E, Table E2). For the 150 m model, our results yielded similar magnitudes and precision for term birth weight and term low birth weight during construction, with no associations for the after-construction period. For the 500 m model, we observed no evidence of associations during or after construction.
We restricted the population to those without missing covariates and observed estimated associations consistent with our main results (Appendix E, Table E3).
We restricted the control group to include only individuals within 500 m of a major road and observed slightly larger associations for the during and after construction estimates. For example, term low birth weight showed 32% increased odds (95% CI: 1.01, 1.72) after construction ended (Appendix E, Table E4).
We observed associations consistent with our main model results when we used a fixed comparison of 500–1,000 m and a 300-m treatment group (Appendix E, Table E5).
When we examined models without temporal restrictions, we observed associations similar to our main model results, for which we had limited the observation period to 2 years after the project’s end date (Appendix E, Table E6).
DISCUSSION AND CONCLUSIONS
We leveraged diverse roadway construction projects to examine the influence of roadway infrastructure improvements on adverse infant health outcomes using a quasi-experimental design with a DiD estimator. During the construction period, we found 19% increased odds in the prevalence of term low birth weight among pregnant people living within 300 m of a single project during gestation, which was robust to several model specifications. We did not find an effect at a distance further than 300 m from the residential address, suggesting that the effects of road construction are highly localized. Contrary to our hypothesis, we did not observe consistent improvements in adverse birth outcomes post-construction for pregnant individuals living within 300 m compared to 300–1,000 m of projects. These results suggest that large roadway construction projects might not have clear benefits for local population health and that the construction process itself might create hazardous conditions for pregnant people and their infants. This finding is especially important for TRAP, which is a highly localized pollution source that disproportionally affects low-income communities and communities of color.205
Our observation of consistent increases in term low birth weight during the construction period aligns with our finding that active roadway construction created up to a 42% increase in congestion and up to a 13% increase in local NO2 during the construction period.198 Given the evidence linking air pollution to decreased term birth weight and increases in term low birth weight,206–208 our finding of a 19% increased odds of term low birth weight among pregnant individuals living within 300 m was not surprising. In addition to the increased TRAP from roadway congestion concurrent with construction, there might have been additional air pollution emissions from construction equipment and road dust.209–211 However, we could not isolate these emission sources and their contributions in this analysis. In addition, living within 300 m of a construction site likely means that pregnant individuals could see or hear this construction, which might have created stressful conditions with psychological impacts that influenced adverse birth outcomes.212 Future research examining infrastructure change should attempt to understand these different pathways and how they contribute toward adverse birth outcomes, which could inform health-protective policy.
We did not observe consistent changes in adverse birth outcomes after construction was completed. Our Stage 1 analysis of construction projects in Texas showed traffic congestion reductions of up to 53% on average and improved air pollution by up to 13% (measured by NO2) on average, up to 3 years after construction ended.198 While this construction project sample was slightly different than what we used in our birth analysis due to the vital statistics coverage, we had hypothesized that a reduction in TRAP would translate into improved birth outcomes. There are several potential reasons for this finding. First, the 13% reduction of TRAP (measured by NO2) was small and might not have been large enough to translate into health benefits. In addition, reducing congestion and NO2 might not have reduced other important pollutants that harm infant health. Our previous work examining roadway construction projects in Texas found that PM2.5 concentrations did not change after construction ended,198 and other work showed that UFP levels did not differ between congested rush-hour periods and less congested non-rush-hour periods.213 Similarly, reducing congestion might not improve nonexhaust pipe emissions such as brake wear, tire wear, or resuspended particulate matter,214,215 or other non-TRAP health risks associated with living near roads, such as traffic noise.216 Previous work by our team showed a unique additional influence of traffic congestion on reduced term birth weight, highlighting the importance of delay and congestion as hazards in their own right.217 We also showed that TRAP reductions did not necessarily benefit all communities, particularly neighborhoods with persistently marginalized populations.205 Regardless, our analyses demonstrated no benefits to birth outcomes for pregnant individuals living near roadway construction projects designed to reduce congestion.
Our study had several key strengths. First, we examined the influence of a large sample of diverse roadway projects over a vast geographic area in Texas. The scale of our construction project sample allowed us to produce more externally valid findings than similar studies relying on a few projects of the same type, such as subway expansions or adding overhead electronic tolling.21–23,28,29 Second, our study used more recent construction and health outcome data than previous studies. This point is important because although tailpipe emissions have decreased over time, we still found damaging impacts of congestion on nearby infant health outcomes.217,218 Third, our results corresponded to a rigorous analysis of the influence of construction projects on traffic volume, congestion, and local TRAP. Our health analysis used similar projects to evaluate the association with birth outcomes. Because we identified a clear association between construction and TRAP, we are confident that the relationship between active construction and term low birth weight identified was likely nonspurious for our data. Fourth, unlike many studies on improvements to public transportation and health that used a risk assessment framework, we relied on primary measures of health outcomes of individuals living near roads. This approach is distinct from measuring air quality changes near roads and applying an approximate relationship between air quality changes and health outcomes (primarily mortality) estimated in separate environmental contexts.
There are several limitations to interpreting the results of our study. First, birth certificate data contained maternal address at the time of delivery, which could have caused exposure misclassification if pregnant persons moved during pregnancy. This limitation might have been mitigated for our study because air pollution has the strongest effect on the birth outcomes we studied during the third trimester of pregnancy, when the address is more likely to be unchanged at delivery.55,157 Second, we could not quantify the impact of individual project types such as widening projects, intersection projects, bridges, new roads, intelligent transportation systems, and traffic signals due to sample sizes and subsequent changes to statistical power as well as nonparallel trends in the “before” period that violate the assumptions of a DiD design. However, roadway widening and intersection construction are the most common project types and are likely to have had the greatest overall impact on congestion and air quality. Third, our analysis did not differentiate between the highly correlated co-occurring exposures related to TRAP (e.g., transportation noise) that could have influenced the associations between project types and birth outcomes.216 Fourth, our sample of births in the after-construction period was about one-third of the size of the sample during the construction period, which limited the statistical precision of our after-construction analysis. Fifth, as in every application of the DiD design, our sample size declined considerably as we moved forward in time beyond the treatment period, limiting our ability to produce externally valid findings about the relationship between construction and the health of infants born many years after construction ended. Sixth, our data for checking the parallel trends assumption of the DiD design for preterm birth and very preterm birth outcomes was imprecise, limiting confidence in our null findings for these outcomes. Seventh, we acknowledge the complexity of birth outcomes, including the substantial debate in the literature on how best to examine these outcomes.26,27,219 We made inferences for term birth weight only and not for birth weight with preterm births. When we removed additional adjustments for gestational length from our term birth weight models, the results did not change.
In conclusion, health cobenefits are often included in the outcomes of roadway construction projects designed to reduce vehicle congestion. In this study of 1,360 diverse roadway projects, we did not observe evidence of consistent benefits to local infant health after construction ended. In fact, we found increased risks to infant growth during the construction period. Our results highlight the need for further accountability research on infrastructure projects designed to have localized health benefits.
CHAPTER 8: SYNTHESIS, INTERPRETATION, AND IMPLICATIONS OF FINDINGS
SYNTHESIS OF RESEARCH FINDINGS
Figure 20 and Table 15 summarize our research triangulation framework and the main findings from each chapter. Each analysis contributed new information to our overall study objective and specific research aims. Overall, our results show that cleaning up the vehicle fleet was more significant indecision decreasing adverse pregnancy outcomes than local programs aimed at reducing congestion.
Summary of Analyses Conducted, General Methods Used, and Main Study Findings
Analysis
Time Period
General Methods Used
TRAP Exposure
Study Finding
Chapter 3: Impact of Vehicle Emissions Regulations on Traffic-Related Air Pollution Exposures and Associated Changes in Adverse Birth Outcomes
1996 to 2016
Linear regression models of change in associations over time
Vehicle miles traveled within 500 m of residences
Improvements in birth outcomes associated with decreasing traffic volumes near homes of pregnant individuals, paralleling decreases in NO2 levels and other policies related to tailpipe emissions.
Chapter 4: Living Downwind of High-Traffic Roads and Adverse Birth Outcomes
2007 to 2016
Linear and logistic regression of neighbors matched by wind
Living downwind (within 500 m) of major roads (>25,000 AADT) compared to upwind of the same road
Pregnant individuals living downwind of high-traffic roads had increased risk of adverse birth outcomes compared to individuals residing upwind of the same roads. This suggests that TRAP is causally associated with adverse birth outcomes.
Chapter 5: Isolating the Additional Influence of Traffic Congestion on Infant Health
2015 to 2016
Linear regression models of congestion exposure
Roadway congestion levels within 500 m of residences measured from connected device/vehicle data
Traffic congestion was associated with reductions in term birth weight when added to total traffic volume on nearby roads and background levels of air pollution and noise. This finding suggests that programs and policies to reduce traffic congestion might benefit infant health positively.
Chapter 6: Transition to Electronic Tolling and Associations with Adverse Birth Outcomes
1999 to 2016
DiD design
Quasi-experiment examining residences before and after toll changes and near and far from toll road locations
Little evidence of improved pregnancy outcomes for residential proximity to a tolled road after transition to electronic tolling.
Chapter 7: Roadway Capacity Improvements, Changes in Congestion and TRAP, and Adverse Birth Outcomes
2009 to 2016
DiD design
Quasi-experiment examining residences before, during, and after roadway construction and near and far from construction locations
Little evidence of improved pregnancy outcomes for residential proximity to roadway construction after construction ended. Increased risk of term low birth weight during the construction period.
AADT = annual average daily traffic; DiD = difference-in-differences; TRAP = traffic-related air pollution; VMT = vehicle miles traveled.
Our research strengthens the overall body of evidence showing that TRAP is associated with adverse pregnancy outcomes, even in recent periods. For example, analyses of births from 2014 to 2016 showed that the highest versus the lowest quintile of VMT within 500 m of the residence, VMT500m was associated with a 13.9 g (95% CI: –16.7, –11.2) reduction in term birth weight, an increased OR of 1.08 (95% CI: 1.03, 1.13) for term low birth weight, and 1.08 (95% CI: 1.01, 1.15) for very preterm birth. Analyses of pregnant individuals living predominantly downwind, compared to upwind, of the same major road between 2007 and 2016 showed that individuals who lived downwind had term birth weight –11.6 g (95% CI: –18.0, –5.2) lower than matched upwind individuals. These associations also had a steep distance decay gradient, with larger associations for individuals living downwind and closer to roads. Given the substantial heterogeneity in the research on TRAP and adverse birth outcomes,25 these results provided crucial new information on the potential risks posed by TRAP specifically, and vehicle exposure in general, to birth outcomes.
Second, our research highlights the importance of TRAP as an environmental justice issue. TRAP substantially decreased between 1996 and 2016 among pregnant people in Texas (over 59%), a major success for regulations targeting vehicle emissions. However, the magnitude of these improvements was consistently lower for pregnancies among historically marginalized populations and in lower-income neighborhoods.220 Additionally, residential exposure to traffic volume increased over time for Black non-Hispanic pregnancies, individuals with less than a high school diploma, foreign-born individuals, and individuals living in historically redlined and low-income neighborhoods. For example, truck VMT within 500 m was 43% higher for Black non-Hispanic pregnant people, compared to White non-Hispanics, in 1996, and increased to a 59% difference in 2016. Many detrimental exposures, in addition to tailpipe emissions, are associated with living in areas of higher traffic. For example, increased traffic noise has been associated with various adverse health effects, including adverse birth outcomes.221 There is increased air pollution from brake and tire wear, especially in “stop-and-go” traffic.222,223 Pedestrian accidents increase with increasing vehicle traffic,224,225 and high-traffic streets and neighborhoods are associated with less outdoor physical activity, pedestrian activity and a lower sense of belonging in community members.226 These exposure disparities demonstrate the persistent legacy of structural sociodemographic segregation in Texas, highlighting the ongoing need for equitable implementation of environmental policy. Given that infrastructure locations rarely change, environmental justice concerns must be considered before breaking ground on new transportation projects, as the burden of pollution resulting from these projects lasts generations.
Third, we added to the accountability literature by evaluating the impacts of the multiple complex regulations targeting vehicle tailpipe emissions implemented over the last three decades. We hypothesized that the associations of VMT and roadway proximity with adverse birth outcomes would decrease in magnitude over time, paralleling regulatory progress to reduce TRAP concentrations. Figure 21 illustrates that we identified decreased tailpipe emissions and ambient TRAP levels (through secondary data) and decreased TRAP exposures for pregnant individuals from 1996 to 2016 (primary data). Among exposures for 6,158,518 births, NO2 exposures decreased 59% over time, but VMT500m remained relatively stable. We also found that the magnitude of associations decreased for total VMT500m and term low birth weight (–60%, OR1996: 1.08 to OR2016: 1.03 for the highest vs. lowest quintile) and preterm birth (–65%) and very preterm birth (–61%), but not for term birth weight. These decreases parallel those seen for NO2 levels and provide some evidence that regulations aimed at reducing tailpipe emissions and associated health effects were successful. While we could not fully measure the “chain of accountability” for specific regulations, we did measure the impact of cumulative regulations targeting vehicle emissions, which can take decades to implement. This research is a clear example of moving from a chain to a “web of accountability,” as discussed in a recent air pollution accountability research commentary.227
Aim 1 conceptual diagram and evidence found for each stage of the chain of accountability. Darker shades of green illustrate strong evidence observed in our chain of accountability analysis.
This study has contributed important new information on how vehicle congestion (in and above normal traffic volumes) might affect birth outcomes and how targeted local interventions (tolls and roadway improvements) might reduce congestion, reduce excess air pollution emissions, and benefit birth outcomes. Our study is among the first epidemiological studies to leverage detailed traffic congestion metrics from connected devices at a large geographical scale. Our analysis of term births suggests that traffic congestion, as measured by delay-per-mile and greenhouse gas emissions from congestion, might adversely affect term birth weight beyond the impacts of traffic volume, background air pollution, and noise on nearby roads. Specifically, traffic delay within 500 m of a maternal residence at delivery was associated with a mean decrease in term birth weight of –8.93 g (95% CI: –14.08, –3.79), comparing the highest to the lowest quintile of exposure, after adjusting for individual covariates and comprehensive environmental co-exposures. This finding suggests that programs and policies to reduce traffic congestion might have positive cobenefits for infant health, which we directly examined in Aim 2 of this study.
We leveraged the large number of roadway infrastructure improvements conducted in Texas as natural experiments to examine the chain of accountability for tolls and roadway capacity improvements to birth outcomes. We hypothesized that implementing these specific traffic reduction programs would be associated with decreased risks of adverse birth outcomes among pregnant individuals who resided near a major road during pregnancy after implementation. Figure 22 summarizes the overall findings of our analyses.
Aim 2 conceptual diagram of evidence found for each stage of the chain of accountability for congestion reduction programs (tolls and roadway improvements). Darker shades of green illustrate strong evidence observed in our chain of accountability analysis.
For toll implementation, we observed minimal changes in local traffic after the implementation of tolling. Among births within 500 m of a tolled road, we found little evidence of an association between the implementation of tolling and adverse birth outcomes. In subanalyses, we found some evidence of a reduced association between tollbooth removal and preterm birth (OR: 0.84, 95% CI: 0.70, 1.01) but no evidence for other outcomes or tolling types. Overall, we found little evidence that the implementation of tolling was consistently associated with improvements in local infant health outcomes.
For large roadway construction projects focused on capacity improvements, we observed that, during the construction period, congestion increased by 42% (95% CI: 30, 56%) for roadway widening projects. For the 3 years post-construction, congestion was reduced by 33% (95% CI: –41, –24%), and NO2 levels within 500 meters were reduced by 13% (95% CI: –22%, –2%). During the construction period, we found a 19% (95% CI: 1.05, 1.36) increased odds of term low birth weight among pregnant people living within 300 m of a single project during gestation, which aligned with the observed increases in congestion during construction. However, we did not observe any evidence of improvements in adverse birth outcomes after construction was completed.
Overall, we did not find evidence that implementing traffic reduction programs decreased risks of adverse birth outcomes among pregnant individuals who resided near a major road during pregnancy after project implementation. This finding highlights the need for further accountability research of infrastructure projects that claim to produce health cobenefits.
RESEARCH TRIANGULATION APPROACH
We used research triangulation to obtain more reliable answers to our overarching accountability research questions. We conducted and integrated results from several different population groups, time periods, TRAP exposure measures, study designs, and analytical approaches, where each approach had different potential biases. Given the challenges of assessing the full chain of accountability228 and the diverse analyses required, a research triangulation approach was important in addressing our main hypotheses. It has been proposed that a minimum set of criteria for using triangulation in etiological epidemiology is to explicitly describe key sources of bias of different approaches; different approaches would be expected to bias the actual causal effect in different directions.37 While accountability research and the chain of accountability framework differ slightly from pure etiological epidemiology, the goals are similar to research triangulation.
Our different analyses and potential associated biases meet the criteria for a research triangulation approach. Within Chapter 3: “Impact of Vehicle Emissions Regulations on Traffic-Related Air Pollution Exposures and Associated Changes in Adverse Birth Outcomes,” we directly examined Aim 1. The major potential source of bias for traditional TRAP epidemiological studies is residual or unmeasured confounding, which can bias associations away from the null. Here, we examined how associations between traffic exposure and adverse birth outcomes changed over time, which has a different confounding structure. Confounding is still a concern; however, this would need to take the form of characteristics that change differently for pregnant individuals in high-traffic areas compared to those in low-traffic areas. In Chapter 4: “Living Downwind of High-Traffic Roads and Adverse Birth Outcomes: Comparing Births Upwind and Downwind of the Same Road,” the primary potential bias was exposure misclassification, which would bias associations toward the null. In Chapter 5: “Isolating the Additional Influence of Traffic Congestion on Infant Health,” the primary potential source of bias was residual or unmeasured confounding, which can bias associations away from the null. However, we controlled for an extensive number of individual, household, and neighborhood variables through innovative data linkages. Congestion could have served as an instrumental variable (i.e., individuals might not select residential locations based on congestion levels (rather, road proximity), and congestion is associated with excess vehicle air pollution emissions). In Chapter 6: “Transition To Electronic Tolling And Associations With Adverse Birth Outcomes,” and Chapter 7: “Roadway Capacity Improvements, Changes In Congestion And Traffic Air Pollution, And Adverse Birth Outcomes,” we used quasi-experiments and a DiD design to isolate the causal associations between these local interventions and changes in birth outcomes. The major potential source of bias for these analyses was unmeasured trends between exposed and unexposed populations, which could potentially bias associations toward or away from the null. However, we tested the DiD model assumptions using the detailed individual information available in the vital statistics data, examined multiple distances to define exposure and control groups, and conducted extensive sensitivity analyses.
LIMITATIONS
While we used a research triangulation approach to obtain more robust conclusions to our hypotheses across all individual studies, it is essential to highlight the main limitations of our research. These limitations were (1) TRAP exposure misclassification, (2) birth outcome measurements, and (3) unmeasured and residual confounding.
While we developed several innovative research methods to estimate exposure to TRAP, there are several limitations to note. First, all exposure assessment approaches relied on full residential addresses reported on the vital statistics data at the time of birth. We did not have information on residential mobility, which can be high during pregnancy but less so during the third trimester.153 Therefore, our TRAP estimates were likely to be most accurate during the third trimester, a period of pregnancy that might be especially susceptible to the impacts of air pollution exposure.55,157 We also lacked information on daily time-activity patterns and residential infiltration. Like most other large-scale air pollution health studies,25 we assumed that TRAP concentrations outdoors at residences were a good proxy for personal exposures. We relied on annual average measures for traffic volume, traffic congestion, and ambient air pollution concentrations and linked data to births based on residential address and corresponding birth year. Relying on annual averages likely resulted in exposure misclassification, and we could not examine specific exposure periods during pregnancy. However, annual estimates were the only data available for the 1996 to 2016 study period. From an accountability perspective, we examined the totality of policy efforts related to TRAP reductions (Aim 1) but could not attribute our results to any specific regulatory action. Similarly, we used several different exposure metrics for TRAP, but we could not isolate the specific air pollution or component of the TRAP mixture affecting adverse birth outcomes. Despite these limitations, we developed and applied several new TRAP exposure assessment approaches that considerably improved over prior research methods.153
We relied on recorded birth information from the vital statistics records to define our main outcomes of interest. Importantly, vital statistics data provides reliable information only for live-born infants,88 and this data source is largely lacking in clinical information (e.g., biomarkers). As TRAP might influence fertility and pregnancy loss,14,90,91 our associations might have underestimated TRAP’s magnitude of association with reproductive and pregnancy health, including birth outcomes. We examined preterm and term birth weight measures as our main outcome variables. Still, these were the only two measures representing the potential biological impacts of TRAP exposure on gestational length and growth restriction.
There is debate in the literature on how best to examine birth outcomes, how to isolate the effects of gestational duration and fetal growth, and the importance of small changes in gestational length and birth weight to future health outcomes.28–32 We chose to restrict our birth weight analyses to term infants to distinguish the impacts of TRAP on fetal growth from gestational length. Our results were, therefore, specific to term births and did not capture the influence of TRAP on birth weight for preterm births.
However, in sensitivity analyses (Appendix F, Table F1), we observed highly similar associations between birth weight and all TRAP exposure measures and births when restricted to term (37–42 weeks’ gestation), all births (22–42 weeks’ gestation) with an adjustment for gestational duration, and all births (22–42 weeks’ gestation), with no adjustment for gestational duration. For example, for all births from 1996 to 2016, term birth weight analyses for VMT within 500 m yielded a decrease in birth weight of –11.5 g (95% CI: –12.7, –10.4) for the highest compared to the lowest quintile, compared to –11.7 g (95% CI: –13.1, –10.3) for all births with no adjustment for gestational duration. Similarly, for NO2, term birth weight analyses yielded a decrease in birth weight of –16.6 g (95% CI: –18.3, –14.8) for the highest compared to the lowest quintile, compared to –14.5 g (95% CI: –16.7, –12.4) for all births with no gestational adjustment. This shows that our effect estimates were consistent regardless of the method used for handling gestational duration. While collider bias remains a possibility when estimating influences of air pollution on birth weight when restricting to term births, our analyses showed that this bias was slight and minimally affected our results.
The observed changes in birth outcome associations we observed across different analyses were small in magnitude. For example, pregnant individuals living upwind, compared to those living downwind, of the same major road, had an 11.6-g decrease (95% CI: –18.01, –5.21) in term birth weight. The magnitude of associations between VMT500m and adverse birth outcomes decreased for term low birth weight (–60%, OR in 1996: 1.08 to OR in 2016: 1.03 for the highest vs. lowest quintile), and pregnant individuals with the highest exposure quintile for NO2, compared to the lowest had an increased OR of 1.06 (95% CI: 1.04, 1.08) for term low birth weight. The magnitudes of these changes were small and likely not clinically significant to individuals. By comparison, in a recent systematic review, maternal smoking during pregnancy was associated with an OR of 1.89 (95% CI: 1.80, 1.98) for low birth weight.229 However, TRAP exposures are widespread in the US population. Even small effect sizes might result in a shift in the health profile of newborns at the population level, resulting in more high-risk births that could have immediate and long-term health consequences. The small associations observed between TRAP and adverse birth outcomes could have reflected changes to biological mechanisms that might have other influences on future morbidity. Additional research is needed to examine the potential impacts of perinatal TRAP exposure and environmental exposures in general on future health trajectories and outcomes.
While we used several causal inference methods to examine the impact of TRAP on adverse birth outcomes, residual and unmeasured confounding cannot be ruled out in observational studies. Even with the large sample size of all births in Texas, power became an issue when we examined fine spatial-temporal exposure measures (e.g., wind matching) and quasi-experiments (e.g., roadway construction, toll implementation). In our DiD analyses of tolls and roadway construction, some of our event-study graphs were imprecise due to small sample sizes of exposed pregnant individuals living near construction and toll changes. The possibility remains that there were pretrends that we could not measure in our dataset. Similarly, due to power issues, we could not quantify the impact of individual project types, such as widening projects, intersection projects, bridges, new roads, intelligent transportation systems, and traffic signals. Our DiD analyses could not differentiate among the highly correlated co-occurring exposures related to TRAP (e.g., transportation noise) that could have influenced the associations between construction and toll changes and birth outcomes.216 Nevertheless, our analyses of wind matching and quasi-experiments of roadway construction and toll implementation provided new information on how TRAP exposures influenced adverse birth outcomes and an accountability perspective on the effectiveness of projects to reduce vehicle congestion and subsequent changes to local birth outcomes.
FUTURE RESEARCH DIRECTIONS
While we conducted comprehensive analyses to answer our primary research questions, several additional questions emerged that should be examined in future research.
We had hypothesized that TRAP was an important environmental justice issue for pregnant individuals and that the benefits of TRAP reductions from emission regulations would not be evenly distributed across race/ethnicity and socioeconomic status, indicators corresponding to characteristics of persistently marginalized populations. We measured very large exposure disparities. For example, Black non-Hispanic pregnant individuals experienced nearly double the exposure to VMT500m (mean: 20,797) compared to White non-Hispanics (mean: 12,053). Yet, we did not observe larger associations between TRAP exposures and adverse birth outcomes. Surprisingly, we generally observed larger overall associations for White non-Hispanic and higher socioeconomic status pregnant individuals, with no consistent pattern of change in these associations over time for marginalized subpopulations. This finding of a smaller association in these subgroups might have been due to competing risks and higher levels of baseline risk. For example, for Black non-Hispanic pregnancies, there was a much higher rate of term low birth weight (4.3%), preterm birth (12.0%), and very preterm birth (2.6%), compared to White non-Hispanic births classified as term low birth weight (1.8%), preterm birth (7.7%), and very preterm birth (1.0%), respectively. Thus, other larger risk factors might have obscured the influence of TRAP exposures. Another potential contributing factor might have been live birth bias, a phenomenon of conditioning a cohort on an infant being alive at delivery, meaning that cases of infertility or early pregnancy loss are excluded.26,230–232 By definition, a vital statistics cohort is conditioned on the infant being alive at delivery, so this source of bias could have been present. In populations with the highest TRAP exposures, more early pregnancy losses could have occurred, leading to fewer conceptions that yielded a live birth. As a result, this would have attenuated associations within the largest exposure groups, as we saw in our analyses. Future research should focus on understanding TRAP risk in highly exposed populations and the potential for live birth bias.
We focused on large roadway construction projects implemented to improve vehicle capacity (and reduce congestion) as a quasi-experiment to understand how birth outcomes changed for pregnant individuals living nearby. We examined over 1,500 specific roadway construction projects with a mean cost of approximately $26 million. We did not conduct a broader analysis of roadway expansion’s overall public health impact; however, roadway expansion is increasingly seen as at odds with a people-focused approach to urban planning that prioritizes investments that improve public health and safety, wellness, equity, and carbon reduction. These goals are now considered of equal or greater importance than moving cars and trucks around as quickly as possible. Future research could replicate our analysis approach for specific public transportation projects to quantify potential health benefits and compare cost-benefits broadly to roadway expansion initiatives. Large-scale highway reclamation projects offer an alternative perspective on examining large-scale highway infrastructure change.
We developed several new innovative research methods that moved the field of TRAP epidemiology and accountability research forward. Data availability, however, restricted the full integration of all methods into one analysis. Future research that uses more recent periods (2013 to present) could integrate wind measurements, congestion measures from connected devices, housing data from CoreLogic, new spatial-temporal models of TRAP (e.g., UFPs, noise), spatially varying modeling approaches, and data on local infrastructure improvements and policy as quasi-natural experiments to conduct future accountability studies.
Data set contains all births in Texas for the 1996 to 2016 period, including geocoded residential address data. All data related to the individual maternal-infant pair is from this source. Data cannot be shared but can be obtained by request on the linked website for similar research purposes.
TxDOT publishes its Roadway Inventory data in various common GIS and tabular formats annually. Data includes GIS linework and all roadway inventory attributes. TxDOT submits this data annually to the Federal Highway Administration as part of the Highway Performance Monitoring System Program. Data are available on the website from 2005 to 2022 and can be requested back to 1999.
Data set contains all roads in Texas with a calculated traffic congestion measure for the 2013 to 2019 period. Publicly available data are a subset of the full source used in this analysis. Data can be obtained for similar research purposes via request on the linked website.
NO2, PM2.5, and O3 data are available for download on the linked website. UFP data are available via request following the instructions at the linked website.
GFCC Tree Cover Multi-Year Global 30 m estimates of Landsat Vegetation Continuous Fields (VCF) tree cover layers of the percentage of horizontal ground in each 30 m pixel covered by woody vegetation greater than 5 meters in height
We used the TxDOT Project Tracker database (linked website) that contains all roadway construction projects (contracts for roadway improvement and maintenance) on the State Highway System between 2004 and 2020 (n = 48,521 construction projects). The database describes the construction activities, construction costs, project location, and start and end dates of each project. We used the TxDOT’s Roadway Inventory for geospatial data on the locations and attributes of roads including variables that describe each segment’s length, number of lanes, highway system classification, and functional classification. Final database available upon request.
We used the publicly available shapefile of all toll road segments in Texas from the TxDOT (linked website) to ascertain where current tolling roads were located at the end of 2020. We used internet-based resources (e.g., TxDOT reports, local news archives, and construction documentation) to find spatial-temporal information on each tolling construction project. We systematically built a database at the road segment level that connected to the primary shapefile. Final database available upon request.
AADT = annual average daily traffic; NDVI = normalized difference vegetation index; TxDOT = Texas Department of Transportation; UFP = ultrafine particle.
ACKNOWLEDGMENTS
We would like to thank Evan Volkin, Lena Harris, Erin J. Campbell, Grace Sventek, Mira Chaskes, Ethan Sawyer, and Max Harleman for their excellent contributions to specific research analyses conducted within this HEI grant.
Footnotes
* A list of abbreviations and other terms appears at the end of this volume.
HEI QUALITY ASSURANCE STATEMENT
The conduct of this study was subjected to independent audits by RTI International staff members Dr. Linda Brown, Dr. David Wilson, and Mr. Ryan Chartier. These staff members are experienced in quality assurance (QA) oversight for air quality monitoring, modeling and exposure assessment, epidemiological methods, and statistical modeling.
The QA oversight program consisted of a remote audit of the final report and the data processing steps. Key details of the dates of the audit and the reviews performed are listed below.
Audit: Final Remote Audit
Date: June 2024 – August 2024
Remarks: The final remote audit consisted of two parts: (1) a review of the final project report and (2) an audit of data processing steps. The review of the final report focused on ensuring that the methods are well documented, and the report is easy to understand. The review also examined if the report highlighted key study findings and limitations. The data audit included review of the datasets and codes for data reduction, processing and analysis, and comparison of the data outputs with reported data. This portion of the audit was restricted to the key components of the study and associated findings. Selected codes for exposure and epidemiological model development were sent to RTI. No raw health data were sent to RTI due to data confidentiality restrictions.
The codes were reviewed at RTI to verify, to the extent feasible, linkages between the various scripts; confirm the models and model variables reported; and verify key tables, figures, and data outputs. The codes appear to be largely consistent with the models described in the report and follow the overall model development procedure described. The values themselves were verified by RTI using the data and scripts provided by the investigators.
Except for a few minor discrepancies, no major quality-related issues were identified from the review of the codes, data, and report. Recommendations were made to address noted discrepancies and typographical errors and included general edits for improved clarity. Those recommendations were addressed in the final report.
A written report was provided to HEI. The QA oversight audit demonstrated that the study was conducted according to the study protocol. The final report appears to be representative of the study conducted.
Perry Hystad is the principal investigator of this study, an associate professor in the College of Health at Oregon State University, and the lead of OSU’s Spatial Health Lab. He received his MSc in Geography from the University of Victoria and his PhD in Epidemiology from the University of British Columbia. Hystad oversaw all aspects of the research project and led analyses of Aim 1. His research focuses on environmental exposure science and epidemiology, with applications to air pollution, healthy built environments, and climate change.
Mary Willis is a co-investigator of this study and an assistant professor in the Department of Epidemiology at the Boston University School of Public Health. She received her MPH from the University of Rochester and her PhD from Oregon State University. Willis led analyses on traffic congestion and electronic tolling and contributed to all other research chapters. Her research focuses on the design and conduct of policy-relevant epidemiology, primarily focusing on exposures related to the energy and transportation sectors.
Elaine Hill is a co-investigator on this project and an associate professor in the Department of Public Health Sciences and Economics at the University of Rochester. She received her PhD in Applied Economics and Management from Cornell University. Hill led analyses of roadway construction and contributed to all other research chapters. Her research interests are in health economics and environmental economics, focusing on the intersection between health, health policy, the environment, and human capital formation.
David Schrank is a co-investigator on this project. He received his PhD in Urban and Regional Science from Texas A&M University. Schrank contributed expertise in urban mobility and Texas data to all analyses. He has been extensively involved in urban mobility research for over 30 years, including contributing to research that has developed and applied a methodology to assess areawide traffic congestion levels and congestion costs.
John Molitor is a co-investigator on this project and an associate professor in the College of Health at Oregon State University. He received his PhD in Statistics from the University of Missouri. Molitor led analyses of spatially varying associations between TRAP measures and adverse birth outcomes at the County and Census Tract levels. His research interests focus on Bayesian methods and spatial modeling applications to air pollution and other environmental exposures.
Andrew Larkin is a co-investigator on this project and an assistant research professor in the Spatial Health Lab within the College of Health at Oregon State University, where he received his PhD in molecular toxicology. Larkin led the development of the wind exposure assessment, upwind and downwind matching, and epidemiological analyses. His research focuses on the intersection of environmental health and data science.
Beate Ritz was a consultant on this project and contributed to the design of epidemiological analyses, interpretation, and manuscript preparation. She is currently a professor at the UCLA Brain Research Institute. Ritz received her MD and a PhD in Medical Sociology from the University of Hamburg Germany. She has extensive experience in environmental epidemiology, with expertise specific to this research of examining the biological mechanisms that might influence the relationship between in-utero exposure to traffic-related air pollution and adverse birth outcomes.
OTHER PUBLICATIONS RESULTING FROM THIS RESEARCH
Hystad P, Hill EL, Larkin A, Schrank D, Harleman M, Volkin E, et al. 2025. Changes in traffic-related air pollution exposures and associations with adverse birth outcomes over 20 years in Texas. Int J Epidemiol 54(1):dyae178, https://doi.org/10.1093/ije/dyae178.
Hill E, Harleman M, Harris L, Sventek G, Ritz B, Campbell EJ, et al. 2024. Roadway construction as a natural experiment to examine air pollution impacts on infant health. Environ Res 252(2):118788, https://doi.org/10.1016/j.envres.2024.118788.
Larkin A, Willis MD, Harris L, Ritz B, Hill EL, Hystad P. 2024. High traffic roads and adverse birth outcomes: comparing births upwind and downwind of the same road. Am J Epidemiol 193(12):1720–1728, https://doi.org/10.1093/aje/kwae120.
Willis MD, Harris L, Campbell EJ, Chaskes M, Sawyer E, Harleman M, et al. 2024. A population-based cohort study of electronic tolling, traffic congestion, and adverse birth outcomes. Environ Int 183:108355, https://doi.org/10.1016/j.envint.2023.108355.
Harleman M, Harris L, Willis MD, Ritz B, Hystad P, Hill EL. 2023. Changes in traffic congestion and air pollution due to major roadway infrastructure improvements in Texas. Sci Total Environ 10:898:165463, https://doi.org/10.1016/j.scitotenv.2023.165463.
Willis MD, Hill EL, Ncube CN, Campbell EJ, Harris L, Harleman M, et al. 2023. Changes in socioeconomic disparities for traffic-related air pollution exposure during pregnancy over a 20-year period in Texas. JAMA Netw Open 6(8):e2328012, https://doi.org/10.1001/jamanetworkopen.2023.28012.
Willis MD, Schrank D, Xu C, Harris L, Ritz B, Hill EL, et al. 2022. A population-based cohort study of traffic congestion and infant growth using connected vehicle data. Sci Adv 8(43):eabp8281, https://doi.org/10.1126/sciadv.abp8281.
2.Chambliss SE, Pinon CPR, Messier KP, LaFranchi B, Upperman CR, Lunden MM, et al. 2021 Local- and regional-scale racial and ethnic disparities in air pollution determined by long-term mobile monitoring. Proc Natl Acad Sci
118(37):e2109249118; https://doi.org/10.1073/pnas.2109249118.
[DOI] [PMC free article] [PubMed] [Google Scholar]
3.Health Effects Institute (HEI). 2010. Traffic-related air pollution: A critical review of the literature on emissions, exposure, and health effects. Boston, MA: Health Effects Institute. [Google Scholar]
4.Zwack LM, Paciorek CJ, Spengler JD, Levy JI. 2011. Modeling spatial patterns of traffic-related air pollutants in complex urban terrain. Environ Health Perspect
119(6):852–859, https://doi.org/10.1289%2Fehp.1002519.
[DOI] [PMC free article] [PubMed] [Google Scholar]
5.Patton AP, Perkins J, Zamore W, Levy JI, Brugge D, Durant JL. 2014. Spatial and temporal differences in traffic-related air pollution in three urban neighborhoods near an interstate highway. Atmos Environ (1994)
99:309–321, https://doi.org/10.1016/j.atmosenv.2014.09.072.
[DOI] [PMC free article] [PubMed] [Google Scholar]
7.Stieb DM, Chen L, Eshoul M, Judek S. 2012. Ambient air pollution, birth weight and preterm birth: a systematic review and meta-analysis. Environ Res
117:100–111, https://doi.org/10.1016/j.envres.2012.05.007.
[DOI] [PubMed] [Google Scholar]
9.National Toxicology Program. 2019. NTP Monograph on the Systematic Review of Traffic-related Air Pollution and Hypertensive Disorders of Pregnancy. NTP Monograph 7. Research Triangle Park, NC: National Toxicology Program. https://doi.org/10.22427/ntp-mgraph-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.HEI Panel on the Health Effects of Traffic-Related Air Pollution. 2010. Traffic-Related Air Pollution: A Critical Review of the Literature on Emissions, Exposure, and Health Effects. Special Report 17. Boston, MA: Health Effects Institute. Available: https://www.healtheffects.org/system/files/SR17TrafficReview.pdf [accessed 1 Dec 2021]. [Google Scholar]
12.Klepac P, Locatelli I, Korošec S, Künzli N, Kukec A. 2018. Ambient air pollution and pregnancy outcomes: a comprehensive review and identification of environmental public health challenges. Environ Res
167:144–159, https://doi.org/10.1016/j.envres.2018.07.008.
[DOI] [PubMed] [Google Scholar]
13.Sram RJ, Binkova B, Dejmek J, Bobak M. 2005. Ambient air pollution and pregnancy outcomes: a review of the literature. Environ Health Perspect
113(4): 375–382, https://doi.org/10.1289/ehp.6362.
[DOI] [PMC free article] [PubMed] [Google Scholar]
14.Carré J, Gatimel N, Moreau J, Parinaud J, Léandri R. 2017. Does air pollution play a role in infertility? A systematic review. Environ Health Glob Access Sci Source
16(1):82, https://doi.org/10.1186/s12940-017-0291-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Wesselink AK, Kirwa K, Hatch EE, Hystad P, Szpiro AA, Kaufman JD, et al. 2020. Residential proximity to major roads and fecundability in a preconception cohort. Environ Epidemiol
4(6):e112, https://doi.org/10.1097%2FEE9.0000000000000112.
[DOI] [PMC free article] [PubMed] [Google Scholar]
16.Conforti A, Mascia M, Cioffi G, De Angelis C, Coppola G, De Rosa P, et al. 2018. Air pollution and female fertility: a systematic review of literature. Reprod Biol Endocrinol RBE
16:117, https://doi.org/10.1186/s12958-018-0433-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Boyles AL, Beverly BE, Fenton SE, Jackson CL, Jukic AMZ, Sutherland VL, et al. 2021. Environmental factors involved in maternal morbidity and mortality. J Women’s Health
30(2):245–252, https://doi.org/10.1089%2Fjwh.2020.8855. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Rowangould GM. 2013. A census of the US near-roadway population: public health and environmental justice considerations. Transp Res Part Transp Environ
25:59–67, https://doi.org/10.1016/j.trd.2013.08.003. [Google Scholar]
19.Smith RB, Fecht D, Gulliver J, Beevers SD, Dajnak D, Blangiardo M, et al. 2017. Impact of London’s road traffic air and noise pollution on birth weight: retrospective population based cohort study. BMJ
359:j5299, 10.1136/bmj.j5299.
[DOI] [PMC free article] [PubMed] [Google Scholar]
20.Dzhambov AM, Markevych I, Lercher P. 2019. Associations of residential greenness, traffic noise, and air pollution with birth outcomes across Alpine areas. Sci Total Environ
678:399–408, https://doi.org/10.1016/j.scitotenv.2019.05.019.
[DOI] [PubMed] [Google Scholar]
24.Hunter PJ, Awoyemi T, Ayede AI, Chico RM, David AL, Dewey KG, et al. 2023. Biological and pathological mechanisms leading to the birth of a small vulnerable newborn. Lancet Lond Engl
401(10389):1720–1732, https://doi.org/10.1016/s0140-6736(23)00573-1. [DOI] [PubMed] [Google Scholar]
26.Neophytou AM, Kioumourtzoglou MA, Goin DE, Darwin KC, Casey JA. 2021. Educational note: addressing special cases of bias that frequently occur in perinatal epidemiology. Int J Epidemiol
50(1):337–345, https://doi.org/10.1093/ije/dyaa252.
[DOI] [PMC free article] [PubMed] [Google Scholar]
27.Wilcox AJ, Weinberg CR, Basso O. 2011. On the pitfalls of adjusting for gestational age at birth. Am J Epidemiol
174(9):1062–1068, https://doi.org/10.1093/aje/kwr230.
[DOI] [PMC free article] [PubMed] [Google Scholar]
28.Wilcox AJ. 2001. On the importance—and the unimportance—of birthweight. Int J Epidemiol
30(6):1233–1241, https://doi.org/10.1093/ije/30.6.1233.
[DOI] [PubMed] [Google Scholar]
29.Delbaere I, Vansteelandt S, De Bacquer D, Verstraelen H, Gerris J, De Sutter P, et al. 2007. Should we adjust for gestational age when analysing birth weights? The use of z-scores revisited. Hum Reprod
22(8):2080–2083, https://doi.org/10.1093/humrep/dem151.
[DOI] [PubMed] [Google Scholar]
30.Hertz-Picciotto I. 2001. Commentary: when brilliant insights lead astray. Int J Epidemiol
30(6):1243–1244, https://doi.org/10.1093/ije/30.6.1243.
[DOI] [PubMed] [Google Scholar]
31.VanderWeele TJ, Mumford SL, Schisterman EF. 2012. Conditioning on intermediates in perinatal epidemiology. Epidemiol Camb Mass
23(1):1–9, https://doi.org/10.1097%2FEDE.0b013e31823aca5d. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Ashorn P, Ashorn U, Muthiani Y, Aboubaker S, Askari S, Bahl R, et al. 2023. Small vulnerable newborns—big potential for impact. The Lancet
401(10389):1692–1706, https://doi.org/10.1016/s0140-6736(23)00354-9. [DOI] [PubMed] [Google Scholar]
33.Wilcox AJ, Cortese M, McConnaughey DR, Moster D, Basso O. 2021. The limits of small-for-gestational-age as a high-risk category. Eur J Epidemiol
36(10):985–991, https://doi.org/10.1007/s10654-021-00810-z.
[DOI] [PMC free article] [PubMed] [Google Scholar]
34.Kim SY, Bechle M, Hankey S, Sheppard L, Szpiro AA, Marshall JD. 2020. Concentrations of criteria pollutants in the contiguous US, 1979 – 2015: role of prediction model parsimony in integrated empirical geographic regression. PLOS ONE
15(2):e0228535, https://doi.org/10.1371/journal.pone.0228535.
[DOI] [PMC free article] [PubMed] [Google Scholar]
37.Lawlor DA, Tilling K, Davey Smith G. 2016. Triangulation in aetiological epidemiology. Int J Epidemiol
45(6):1866–1886, https://doi.org/10.1093/ije/dyw314.
[DOI] [PMC free article] [PubMed] [Google Scholar]
38.Dockery DW, Pope CA, 3rd, Xu X, Spengler JD, Ware JH, Fay ME, et al. 1993. An association between air pollution and mortality in six US cities. N Engl J Med
329(24):1753–1759, https://doi.org/10.1056/nejm199312093292401.
[DOI] [PubMed] [Google Scholar]
39.Ritz B, Yu F. 1999. The effect of ambient carbon monoxide on low birth weight among children born in southern California between 1989 and 1993. Environ Health Perspect
107(1):17–25, https://doi.org/10.1289/ehp.9910717.
[DOI] [PMC free article] [PubMed] [Google Scholar]
40.Woodruff TJ, Grillo J, Schoendorf KC. 1997. The relationship between selected causes of postneonatal infant mortality and particulate air pollution in the United States. Environ Health Perspect
105(6):608–612, https://doi.org/10.1289/ehp.97105608.
[DOI] [PMC free article] [PubMed] [Google Scholar]
41.Schwartz J. 1994. Air pollution and daily mortality: a review and meta-analysis. Environ Res
64(1):36–52, https://doi.org/10.1006/enrs.1994.1005.
[DOI] [PubMed] [Google Scholar]
42.Pope CA, 3rd, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, et al. 1995. Particulate air pollution as a predictor of mortality in a prospective study of US adults. Am J Respir Crit Care Med
151(3 Pt 1):669–674, https://doi.org/10.1164/ajrccm/151.3_pt_1.669.
[DOI] [PubMed] [Google Scholar]
44.Krupnick A, Morgenstern R. 2002. The future of benefit-cost analyses of the Clean Air Act. Annu Rev Public Health
23:427–448, https://doi.org/10.1146/annurev.publhealth.23.100901.140516.
[DOI] [PubMed] [Google Scholar]
46.Rich DQ. 2017. Accountability studies of air pollution and health effects: lessons learned and recommendations for future natural experiment opportunities. Environ Int
100:62–78, https://doi.org/10.1016/j.envint.2016.12.019.
[DOI] [PMC free article] [PubMed] [Google Scholar]
47.Bell ML, Morgenstern RD, Harrington W. 2011. Quantifying the human health benefits of air pollution policies: Review of recent studies and new directions in accountability research. Environ Sci Policy
14(4):357–368, 10.1016/j.envsci.2011.02.006. [DOI] [Google Scholar]
48.Henneman LRF, Liu C, Mulholland JA, Russell AG. 2017. Evaluating the effectiveness of air quality regulations: a review of accountability studies and frameworks. J Air Waste Manag Assoc
67(2):144–172, https://doi.org/10.1080/10962247.2016.1242518.
[DOI] [PubMed] [Google Scholar]
49.Boogaard H, van Erp AM, Walker KD, Shaikh R. 2017. Accountability studies on air pollution and health: the HEI experience. Curr Environ Health Rep
4(4):514–522, https://doi.org/10.1007/s40572-017-0161-0.
[DOI] [PubMed] [Google Scholar]
50.Yu W, Thurston GD. 2023. An interrupted time series analysis of the cardiovascular health benefits of a coal coking operation closure. Environ Res Health
1(4):045002, 10.1088/2752-5309/ace4ea. [DOI] [Google Scholar]
51.Dockery DW, Rich DQ, Goodman PG, Clancy L, Ohman-Strickland P, George P, et al. 2013. Effect of air pollution control on mortality and hospital admissions in Ireland. Res Rep Health Eff Inst (176):3–109. [PubMed] [Google Scholar]
52.Clancy L, Goodman P, Sinclair H, Dockery DW. 2002. Effect of air-pollution control on death rates in Dublin, Ireland: an intervention study. Lancet Lond Engl
360(9341):1210–1214, https://doi.org/10.1016/s0140-6736(02)11281-5. [DOI] [PubMed] [Google Scholar]
53.Friedman MS, Powell KE, Hutwagner L, Graham LM, Teague WG. 2001. Impact of Changes in Transportation and Commuting Behaviors During the 1996 Summer Olympic Games in Atlanta on Air Quality and Childhood Asthma. JAMA
285(7):897–905, https://doi.org/10.1001/jama.285.7.897.
[DOI] [PubMed] [Google Scholar]
54.Peel JL, Klein M, Flanders WD, Mulholland JA, Tolbert IPEHEI Health Review Committee. 2010. Impact of improved air quality during the 1996 Summer Olympic Games in Atlanta on multiple cardiovascular and respiratory outcomes. Res Rep Health Eff Inst (148):3–23; discussion 25–33. [PubMed] [Google Scholar]
55.Rich DQ, Liu K, Zhang J, Thurston SW, Stevens TP, Pan Y, et al. 2015. Differences in birth weight associated with the 2008 Beijing Olympics air pollution reduction: results from a natural experiment. Environ Health Perspect
123(9):880–887, https://doi.org/10.1289/ehp.1408795.
[DOI] [PMC free article] [PubMed] [Google Scholar]
56.Zhang J, Zhu T, Kipen H, Wang G, Huang W, Rich D, et al. 2013. Cardiorespiratory biomarker responses in healthy young adults to drastic air quality changes surrounding the 2008 Beijing Olympics. Res Rep Health Eff Inst (174):5–174. [PMC free article] [PubMed] [Google Scholar]
57.Weichenthal S, Pinault L, Christidis T, Burnett RT, Brook JR, Chu Y, et al. 2022. How low can you go? Air pollution affects mortality at very low levels. Sci Adv
8(39):eabo3381, https://doi.org/10.1126%2Fsciadv.abo3381.
[DOI] [PMC free article] [PubMed] [Google Scholar]
58.Strak M, Weinmayr G, Rodopoulou S, Chen J, de Hoogh K, Andersen ZJ, et al. 2021. Long term exposure to low level air pollution and mortality in eight European cohorts within the ELAPSE project: pooled analysis. BMJ
374:n1904, https://doi.org/10.1136/bmj.n1904.
[DOI] [PMC free article] [PubMed] [Google Scholar]
59.Di Q, Wang Y, Zanobetti A, Wang Y, Koutrakis P, Choirat C, et al. 2017. Air Pollution and Mortality in the Medicare Population. N Engl J Med
376(26):2513–2522, https://doi.org/10.1056/nejmoa1702747.
[DOI] [PMC free article] [PubMed] [Google Scholar]
60.Wang Y, Apte JS, Hill JD, Ivey CE, Patterson RF, Robinson AL, et al. 2022. Location-specific strategies for eliminating US national racial-ethnic PM2.5 exposure inequality. Proc Natl Acad Sci. 2022 Nov;119(44):e2205548119, https://doi.org/10.1073/pnas.2205548119.
[DOI] [PMC free article] [PubMed] [Google Scholar]
61.Wang Y, Apte JS, Hill JD, Ivey CE, Johnson D, Min E, et al. 2023. Air quality policy should quantify effects on disparities. Science
381(6655):272–274, https://doi.org/10.1126/science.adg9931.
[DOI] [PubMed] [Google Scholar]
62.Josey KP, Delaney SW, Wu X, Nethery RC, DeSouza P, Braun D, et al. 2023. Air Pollution and Mortality at the Intersection of Race and Social Class. N Engl J Med
388(15):1396-1404, https://doi.org/10.1056/nejmsa2300523.
[DOI] [PMC free article] [PubMed] [Google Scholar]
63.Clark L, Millett D, Marshall J. 2017. Changes in Transportation-Related Air Pollution Exposures by Race-Ethnicity and Socioeconomic Status: Outdoor Nitrogen Dioxide in the United States in 2000 and 2010. Environ Health Perspect
125(9):097012, https://doi.org/10.1289/ehp959.
[DOI] [PMC free article] [PubMed] [Google Scholar]
64.Liu J, Clark LP, Bechle MJ, Hajat A, Kim SY, Robinson AL, et al. 2021. Disparities in Air Pollution Exposure in the United States by Race/Ethnicity and Income, 1990–2010. Environ Health Perspect
129(12):127005, https://doi.org/10.1289/ehp8584.
[DOI] [PMC free article] [PubMed] [Google Scholar]
65.Lane HM, Morello-Frosch R, Marshall JD, Apte JS. 2022. Historical redlining is associated with present-day air pollution disparities in US cities. Environ Sci Technol Lett
9(4):345–350, https://doi.org/10.1021%2Facs.estlett.1c01012.
[DOI] [PMC free article] [PubMed] [Google Scholar]
67.Beck S, Wojdyla D, Say L, Betran AP, Merialdi M, Requejo JH, et al. 2010. The worldwide incidence of preterm birth: A systematic review of maternal mortality and morbidity. Bull World Health Organ
88(1):31–38, https://doi.org/10.2471/blt.08.062554.
[DOI] [PMC free article] [PubMed] [Google Scholar]
68.Hack M, Klein NK, Taylor HG. 1995. Long-term developmental outcomes of low birth weight infants. Future Child
5(1):176–196.
[PubMed] [Google Scholar]
69.Howland RE, Angley M, Won SH, Wilcox W, Searing H, Liu SY, et al. 2019. Determinants of Severe Maternal Morbidity and Its Racial/Ethnic Disparities in New York City, 2008–2012. Matern Child Health J
23(3):346–355, https://doi.org/10.1007/s10995-018-2682-z.
[DOI] [PubMed] [Google Scholar]
70.Admon LK, Winkelman TNA, Zivin K, Terplan M, Mhyre JM, Dalton VK. 2018. Racial and Ethnic Disparities in the Incidence of Severe Maternal Morbidity in the United States, 2012–2015. Obstet Gynecol
132(5):1158–1166, https://doi.org/10.1097/aog.0000000000002937.
[DOI] [PubMed] [Google Scholar]
71.Wang L, Guo P, Tong H, Wang A, Chang Y, Guo X, et al. 2020. Traffic-related metrics and adverse birth outcomes: a systematic review and meta-analysis. Environ Res
188:109752, https://doi.org/10.1016/j.envres.2020.109752.
[DOI] [PubMed] [Google Scholar]
72.Willis MD, Hill EL, Kile ML, Carozza S, Hystad P. 2021. Assessing the effectiveness of vehicle emission regulations on improving perinatal health: a population-based accountability study. Int J Epidemiol
49(6):1781–1791, https://doi.org/10.1093/ije/dyaa137.
[DOI] [PMC free article] [PubMed] [Google Scholar]
73.Currie J, Walker R. 2011. Traffic congestion and infant health: evidence from E-ZPass. Am Econ J Appl Econ
3(1):65–90. [Google Scholar]
74.Ferguson KK, O’Neill M, Meeker J. 2014. Environmental contaminant exposures and preterm birth: a comprehensive review. J Toxicol Environ Health B Crit Rev
16(2):69–113, https://doi.org/10.1080/10937404.2013.775048. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Blumenshine P, Egerter S, Barclay CJ, Cubbin C, Braveman PA. 2010. Socioeconomic disparities in adverse birth outcomes: a systematic review. Am J Prev Med
39(3):263–272, https://doi.org/10.1016/j.amepre.2010.05.012.
[DOI] [PubMed] [Google Scholar]
80.PRISM Climate Group. PRISM Climate Group at Oregon State University. Available: https://prism.oregonstate.edu/ [accessed 9 Aug 2023].
81.Willis MD, Hill EL, Ncube CN, Campbell EJ, Harris L, Harleman M, et al. 2023. Changes in socioeconomic disparities for traffic-related air pollution exposure during pregnancy over a 20-year period in Texas. JAMA Netw Open
6(8):e2328012, https://doi.org/10.1001/jamanetworkopen.2023.28012.
[DOI] [PMC free article] [PubMed] [Google Scholar]
83.Knorr Held L. 2000. Bayesian modelling of inseparable space-time variation in disease risk. Stat Med
19:17–18, https://doi.org/10.1002/1097-0258(20000915/30)19:17/18%3C2555::aid-sim587%3E3.0.co;2-#. [DOI] [PubMed] [Google Scholar]
84.Urdangarin A, Goicoa T, Dolores Ugarte M. 2022. Space-time interactions in Bayesian disease mapping with recent tools: making things easier for practitioners. Stat Methods Med Res
31(6):1085–1103, https://doi.org/10.1177/09622802221079351.
[DOI] [PubMed] [Google Scholar]
86.Bakka H, Rue H, Fuglstad GA, Riebler A, Bolin D, Illian J, et al. 2018. Spatial modeling with R-INLA: a review. Wiley Interdiscip Rev Comput Stat
10(6):e1443, 10.1002/wics.1443. [DOI] [Google Scholar]
87.Willis M, Schrank D, Xu C, Harris L, Ritz B, Hill E, et al. 2022. A population-based cohort study of traffic congestion and infant health using connected vehicle data. Sci Adv
8(4):eabp8281, https://doi.org/10.1126/sciadv.abp8281.
[DOI] [PMC free article] [PubMed] [Google Scholar]
88.Henneman LRF, Shen H, Hogrefe C, Russell AG, Zigler CM. 2021. Four Decades of United States Mobile Source Pollutants: Spatial–Temporal Trends Assessed by Ground-Based Monitors, Air Quality Models, and Satellites. Environ Sci Technol
55(2):882–892, https://doi.org/10.1021%2Facs.est.0c07128.
[DOI] [PMC free article] [PubMed] [Google Scholar]
89.Lane KJ, Levy JI, Scammell MK, Patton AP, Durant JL, Mwamburi M, et al. 2015. Effect of time-activity adjustment on exposure assessment for traffic-related ultrafine particles. J Expo Sci Environ Epidemiol
25(5):506–516, https://doi.org/10.1038/jes.2015.11.
[DOI] [PMC free article] [PubMed] [Google Scholar]
90.Butler L, Gallagher L, Winter M, Fabian MP, Wesselink A, Aschengrau A. 2021. Residential proximity to roadways and placental-associated stillbirth: a case-control study. Int J Environ Health Res
31(4):465–474, https://doi.org/10.1080%2F09603123.2019.1673882.
[DOI] [PMC free article] [PubMed] [Google Scholar]
91.Wesselink AK, Wang TR, Ketzel M, Mikkelsen EM, Brandt J, Khan J, et al. 2022. Air pollution and fecundability: results from a Danish preconception cohort study. Paediatr Perinat Epidemiol
36(1):57–67, https://doi.org/10.1111/ppe.12832.
[DOI] [PMC free article] [PubMed] [Google Scholar]
92.Wilhelm M, Ritz B. 2003. Residential proximity to traffic and adverse birth outcomes in Los Angeles county, California, 1994-1996. Environ Health Perspect. 2003 Feb;111(2):207–216, https://doi.org/10.1289%2Fehp.5688.
[DOI] [PMC free article] [PubMed] [Google Scholar]
93.Wilhelm M, Ghosh JK, Su J, Cockburn M, Jerrett M, Ritz B. 2011. Traffic-related air toxics and preterm birth: a population-based case-control study in Los Angeles county, California. Environ Health
10(1):89, https://doi.org/10.1186/1476-069x-10-89.
[DOI] [PMC free article] [PubMed] [Google Scholar]
94.Girguis MS, Strickland MJ, Hu X, Liu Y, Bartell SM, Vieira VM. 2016. Maternal exposure to traffic-related air pollution and birth defects in Massachusetts. Environ Res
146:1–9, https://doi.org/10.1016/j.envres.2015.12.010.
[DOI] [PMC free article] [PubMed] [Google Scholar]
95.Padula AM, Tager IB, Carmichael SL, Hammond SK, Yang W, Lurmann F, et al. 2013. Ambient Air Pollution and Traffic Exposures and Congenital Heart Defects in the San Joaquin Valley of California. Paediatr Perinat Epidemiol
27(4):329–339, https://doi.org/10.1111/ppe.12055.
[DOI] [PMC free article] [PubMed] [Google Scholar]
96.Achakulwisut P, Brauer M, Hystad P, Anenberg SC. 2019. Global, national, and urban burdens of paediatric asthma incidence attributable to ambient NO2 pollution: estimates from global datasets. Lancet Planet Health
3(4):e166–178, https://doi.org/10.1016/s2542-5196(19)30046-4.
[DOI] [PubMed] [Google Scholar]
97.Zhang Y, Ni H, Bai L, Cheng Q, Zhang H, Wang S, et al. 2019. The short-term association between air pollution and childhood asthma hospital admissions in urban areas of Hefei City in China: a time-series study. Environ Res
169:510–516, https://doi.org/10.1016/j.envres.2018.11.043.
[DOI] [PubMed] [Google Scholar]
99.Willis MD, Schrank D, Xu C, Harris L, Ritz BR, Hill EL, et al. 2022. A population-based cohort study of traffic congestion and infant growth using connected vehicle data. Sci Adv
8(43):eabp8281, https://doi.org/10.1126/sciadv.abp8281.
[DOI] [PMC free article] [PubMed] [Google Scholar]
100.Schirmer AE, Gallemore C, Liu T, Magle S, DiNello E, Ahmed H, et al. 2019. Mapping behaviorally relevant light pollution levels to improve urban habitat planning. Sci Rep
9(1):11925, https://doi.org/10.1038/s41598-019-48118-z.
[DOI] [PMC free article] [PubMed] [Google Scholar]
101.Généreux M, Auger N, Goneau M, Daniel M. 2008. Neighbourhood socioeconomic status, maternal education and adverse birth outcomes among mothers living near highways. J Epidemiol Community Health
62(8):695–700, https://doi.org/10.1136/jech.2007.066167.
[DOI] [PubMed] [Google Scholar]
102.Fang K, Volker J. 2017. Cutting Greenhouse Gas Emissions Is Only the Beginning: A Literature Review of The Co-Benefits of Reducing Vehicle Miles Traveled. A white paper from the National Center for Sustainable Transportation. Available: https://rosap.ntl.bts.gov/view/dot/32254 [accessed 15 March 2023].
103.Rietmann N, Hügler B, Lieven T. 2020. Forecasting the trajectory of electric vehicle sales and the consequences for worldwide CO2 emissions. J Clean Prod
261:121038, 10.1016/j.jclepro.2020.121038. [DOI] [Google Scholar]
104.Zabrocki L, Alari A, Benmarhnia T. 2022. Improving the design stage of air pollution studies based on wind patterns. Sci Rep
12(1):7917, 10.1038/s41598-022-11939-6.
[DOI] [PMC free article] [PubMed] [Google Scholar]
105.Knittel C, Miller D, Sanders N. 2016. Caution, Drivers! Children Present: Traffic, Pollution, and Infant Health. Rev Econ Stat
98(2):350–366, https://doi.org/10.1162/REST_a_00548. [Google Scholar]
106.Deryugina T, Heutel G, Miller NH, Molitor D, Reif J. 2019. The mortality and medical costs of air pollution: evidence from changes in wind direction. Am Econ Rev
109(12):4178–4219, https://doi.org/10.1257/aer.20180279.
[DOI] [PMC free article] [PubMed] [Google Scholar]
107.Anderson ML. 2019. As the wind blows: the effects of long-term exposure to air pollution on mortality. J Eur Econ Assoc
18(4):1886–1927, https://doi.org/10.1093/jeea/jvz051.
[DOI] [PMC free article] [PubMed] [Google Scholar]
109.Hersbach H, Bell B, Berrisford P, Hirahara S, Horányi A, Muñoz-Sabater J, et al. 2020. The ERA5 global reanalysis. Q J R Meteorol Soc
146(730):1999–2049, https://doi.org/10.1002/qj.3803. [Google Scholar]
113.Wang L, Guo P, Tong H, Wang A, Chang Y, Guo X, et al. 2020. Traffic-related metrics and adverse birth outcomes: A systematic review and meta-analysis. Environ Res
188:109752, https://doi.org/10.1016/j.envres.2020.109752.
[DOI] [PubMed] [Google Scholar]
114.Laurent O, Hu J, Li L, Kleeman MJ, Bartell SM, Cockburn M, et al. 2016. Low birth weight and air pollution in California: Which sources and components drive the risk?
Environ Int
92–93:471–477, https://doi.org/10.1016/j.envint.2016.04.034. [DOI] [PMC free article] [PubMed] [Google Scholar]
115.Bell ML, Belanger K. 2012. Review of research on residential mobility during pregnancy: consequences for assessment of prenatal environmental exposures. J Expo Sci Environ Epidemiol
22(5):429–438, https://doi.org/10.1038%2Fjes.2012.42.
[DOI] [PMC free article] [PubMed] [Google Scholar]
116.Boehmer TK, Foster SL, Henry JR, Woghiren-Akinnifesi EL, Yip FY; Centers for Disease Control and Prevention (CDC). 2013. Residential proximity to major highways - United States, 2010. MMWR Suppl
62(3):46–50.
[PubMed] [Google Scholar]
117.Cambridge Systematics, Texas Transportation Institute. 2004. Traffic Congestion and Reliability: Linking Solutions to Problems. Available: https://ops.fhwa.dot.gov/congestion_report_04/.
118.Litman T. 2013. Transportation and Public Health. Annu Rev Public Health
34(1):217–233, https://doi.org/10.1146/annurev-publhealth-031912-114502.
[DOI] [PubMed] [Google Scholar]
122.HEI Panel on the Health Effects of Traffic-Related Air Pollution. 2010. Traffic-Related Air Pollution: A Critical Review of the Literature on Emissions, Exposure, and Health Effects. Special Report 17. Boston, MA: Health Effects Institute. [Google Scholar]
123.Zhang K, Batterman S, Dion F. 2011. Vehicle emissions in congestion: comparison of work zone, rush hour and free-flow conditions. Atmos Environ
45(11):1929–1939, 10.1016/j.atmosenv.2011.01.030. [DOI] [Google Scholar]
124.Smit R, Brown AL, Chan YC. 2008. Do air pollution emissions and fuel consumption models for roadways include the effects of congestion in the roadway traffic flow?
Environ Model Softw
23(10):1262–1270, 10.1016/j.envsoft.2008.03.001. [DOI] [Google Scholar]
125.De Vlieger I, De Keukeleere D, Kretzschmar JG. 2000. Environmental effects of driving behaviour and congestion related to passenger cars. Atmos Environ
34(27):4649–4655, 10.1016/S1352-2310(00)00217-X. [DOI] [Google Scholar]
126.Klepac P, Locatelli I, Korošec S, Künzli N, Kukec A. 2018. Ambient air pollution and pregnancy outcomes: a comprehensive review and identification of environmental public health challenges. Environ Res. 2018 Nov 1;167:144–159, 10.1016/j.envres.2018.07.008.
[DOI] [PubMed] [Google Scholar]
127.Levy JI, Buonocore JJ, von Stackelberg K. 2010. Evaluation of the public health impacts of traffic congestion: a health risk assessment. Environ Health
9(1):65, https://doi.org/10.1186/1476-069x-9-65.
[DOI] [PMC free article] [PubMed] [Google Scholar]
128.Wilhelm M, Ritz B. 2003. Residential proximity to traffic and adverse birth outcomes in Los Angeles county, California, 1994-1996. Environ Health Perspect
111(2):207–216, https://doi.org/10.1289%2Fehp.5688.
[DOI] [PMC free article] [PubMed] [Google Scholar]
129.Wilhelm M, Ghosh JK, Su J, Cockburn M, Jerrett M, Ritz B. 2011. Traffic-related air toxics and preterm birth: a population-based case-control study in Los Angeles county, California. Environ Health
10:89, https://doi.org/10.1186/1476-069x-10-89.
[DOI] [PMC free article] [PubMed] [Google Scholar]
130.Batterman S, Burke J, Isakov V, Lewis T, Mukherjee B, Robins T. 2014. A comparison of exposure metrics for traffic-related air pollutants: application to epidemiology studies in Detroit, Michigan. Int J Environ Res Public Health
11(9):9553–9577, https://doi.org/10.3390%2Fijerph110909553.
[DOI] [PMC free article] [PubMed] [Google Scholar]
132.Center for Health Disparities Research. University of Wisconsin School of Medicine and Public Health. 2015 Area Deprivation Index. Available: https://www.neighborhoodatlas.medicine.wisc.edu/ [accessed 4 Oct 2018].
133.Kind AJH, Buckingham WR. 2018. Making neighborhood-disadvantage metrics accessible — the Neighborhood Atlas. N Engl J Med
378(26):2456–2458, https://doi.org/10.1056%2FNEJMp1802313.
[DOI] [PMC free article] [PubMed] [Google Scholar]
135.Kim SY, Bechle M, Hankey S, Sheppard L, Szpiro AA, Marshall JD. 2020. Concentrations of criteria pollutants in the contiguous US, 1979 – 2015: role of prediction model parsimony in integrated empirical geographic regression. PLOS ONE
15(2):e0228535, https://doi.org/10.1371/journal.pone.0228535.
[DOI] [PMC free article] [PubMed] [Google Scholar]
136.Saha PK, Hankey S, Marshall JD, Robinson AL, Presto AA. 2021. High-Spatial-Resolution Estimates of Ultrafine Particle Concentrations across the Continental United States. Environ Sci Technol
55(15):10320–10331, https://doi.org/10.1021/acs.est.1c03237.
[DOI] [PubMed] [Google Scholar]
137.CACES. CACES RCM/LUR Data Download. Available: https://www.caces.us/data [accessed 23 Feb 2022].
138.Zhou Y, Levy JI. 2007. Factors influencing the spatial extent of mobile source air pollution impacts: a meta-analysis. BMC Public Health
7:89, https://doi.org/10.1186/1471-2458-7-89.
[DOI] [PMC free article] [PubMed] [Google Scholar]
139.Arunachalam S, Valencia A, Akita Y, Serre ML, Omary M, Garcia V, et al. 2014. A method for estimating urban background concentrations in support of hybrid air pollution modeling for environmental health studies. Int J Environ Res Public Health
11(10):10518–10536, https://doi.org/10.3390/ijerph111010518.
[DOI] [PMC free article] [PubMed] [Google Scholar]
141.Pickett AR, Bell ML. 2011. Assessment of Indoor Air Pollution in Homes with Infants. Int J Environ Res Public Health
8(12):4502–4520, https://doi.org/10.3390%2Fijerph8124502.
[DOI] [PMC free article] [PubMed] [Google Scholar]
142.Makri A, Stilianakis NI. 2008. Vulnerability to air pollution health effects. Int J Hyg Environ Health
211(3):326–336, https://doi.org/10.1016/j.ijheh.2007.06.005.
[DOI] [PubMed] [Google Scholar]
145.Hooftman N, Messagie M, Joint F, Segard JB, Coosemans T. 2018. In-life range modularity for electric vehicles: the environmental impact of a range-extender trailer system. Appl Sci
8(7):1016, 10.3390/app8071016. [DOI] [Google Scholar]
146.Liu Y, Chen H, Gao J, Li Y, Dave K, Chen J, et al. 2021. Comparative analysis of non-exhaust airborne particles from electric and internal combustion engine vehicles. J Hazard Mater
420:126626, https://doi.org/10.1016/j.jhazmat.2021.126626.
[DOI] [PubMed] [Google Scholar]
149.Jbaily A, Zhou X, Liu J, Lee TH, Kamareddine L, Verguet S, et al. 2022. Air pollution exposure disparities across US population and income groups. Nature
601(7892):228–233, https://doi.org/10.1038%2Fs41586-021-04190-y.
[DOI] [PMC free article] [PubMed] [Google Scholar]
150.Liu J, Clark LP, Bechle MJ, Hajat A, Kim SY, Robinson AL, et al. 2021. Disparities in air pollution exposure in the United States by race/ethnicity and income, 1990–2010. Environ Health Perspect
129(12):127005, https://doi.org/10.1289/ehp8584.
[DOI] [PMC free article] [PubMed] [Google Scholar]
151.Knittel CR, Miller DL, Sanders NJ. 2015. Caution, Drivers! Children Present: Traffic, Pollution, and Infant Health. Rev Econ Stat
98(2):350–366, 10.1162/REST_a_00548. [DOI] [Google Scholar]
152.Canfield MA, Ramadhani TA, Langlois PH, Waller DK. 2006. Residential mobility patterns and exposure misclassification in epidemiologic studies of birth defects. J Expo Sci Environ Epidemiol
16(6):538–543, https://doi.org/10.1038/sj.jes.7500501.
[DOI] [PubMed] [Google Scholar]
153.Bell ML, Belanger K. 2012. Review of research on residential mobility during pregnancy: consequences for assessment of prenatal environmental exposures. J Expo Sci Environ Epidemiol
22(5):429–438, https://doi.org/10.1038%2Fjes.2012.42.
[DOI] [PMC free article] [PubMed] [Google Scholar]
154.Lane KJ, Kangsen Scammell M, Levy JI, Fuller CH, Parambi R, Zamore W, et al. 2013. Positional error and time-activity patterns in near-highway proximity studies: an exposure misclassification analysis. Environ Health
12(1):75, https://doi.org/10.1186%2F1476-069X-12-75.
[DOI] [PMC free article] [PubMed] [Google Scholar]
155.Batterman S, Cook R, Justin T. 2015. Temporal variation of traffic on highways and the development of accurate temporal allocation factors for air pollution analyses. Atmos Environ
(1994)
107:351–363, 10.1016/j.atmosenv.2015.02.047. [DOI] [PMC free article] [PubMed] [Google Scholar]
156.Lupo PJ, Symanski E, Waller DK, Chan W, Langlois PH, Canfield MA, et al. 2011. Maternal exposure to ambient levels of benzene and neural tube defects among offspring: Texas, 1999-2004. Environ Health Perspect
119(3):397–402, https://doi.org/10.1289%2Fehp.1002212.
[DOI] [PMC free article] [PubMed] [Google Scholar]
157.Fleischer NL, Merialdi M, van DA Vadillo-Ortega Felipe, Martin RV, Betran AP, et al. 2014. Outdoor air pollution, preterm birth, and low birth weight: Analysis of the World Health Organization Global Survey on Maternal and Perinatal Health. Environ Health Perspect
122(4):425–430, https://doi.org/10.1289/ehp.1306837.
[DOI] [PMC free article] [PubMed] [Google Scholar]
158.Weisskopf MG, Kioumourtzoglou MA, Roberts AL. 2015. Air pollution and autism spectrum disorders: Causal or confounded?
Curr Environ Health Rep
2(4):430–439, https://doi.org/10.1007/s40572-015-0073-9.
[DOI] [PMC free article] [PubMed] [Google Scholar]
159.Wilcox AJ. 2001. On the importance—and the unimportance— of birthweight. Int J Epidemiol
30(6):1233–1241, https://doi.org/10.1093/ije/30.6.1233.
[DOI] [PubMed] [Google Scholar]
160.Levy JI, Buonocore JJ, von Stackelberg K. 2010. Evaluation of the public health impacts of traffic congestion: a health risk assessment. Environ Health
9(1):65, https://doi.org/10.1186/1476-069x-9-65.
[DOI] [PMC free article] [PubMed] [Google Scholar]
161.Pedde M, Szpiro AA, Adar SD. 2017. Traffic congestion as a risk factor for mortality in near-road communities: a case-crossover study. Am J Epidemiol
186(5):564–572, https://doi.org/10.1093/aje/kwx130.
[DOI] [PubMed] [Google Scholar]
162.Zhang K, Batterman S. 2013. Air pollution and health risks due to vehicle traffic. Sci Total Environ
450–451:307–316, https://doi.org/10.1016%2Fj.scitotenv.2013.01.074. [DOI] [PMC free article] [PubMed] [Google Scholar]
163.Harleman M, Harris L, Willis MD, Ritz B, Hystad P, Hill EL. 2023. Changes in traffic congestion and air pollution due to major roadway infrastructure improvements in Texas. Sci Total Environ
898:165463, https://doi.org/10.1016/j.scitotenv.2023.165463.
[DOI] [PubMed] [Google Scholar]
165.Simeonova E, Currie J, Nilsson P, Walker R. 2021. Congestion pricing, air pollution, and children’s health. J Hum Resour
56(4):971–996, https://doi.org/10.3368/jhr.56.4.0218-9363R2. [Google Scholar]
166.Bartin B, Mudigonda S, Ozbay K. 2007. Impact of electronic toll collection on air pollution levels: estimation using micro-scopic simulation model of large-scale transportation network. Transp Res Rec
2011(1):68–77, https://doi.org/10.3141/2011-08. [Google Scholar]
167.Lin MY, Chen YC, Lin DY, Hwang BF, Hsu HT, Cheng YH, et al. 2020. Effect of implementing electronic toll collection in reducing highway particulate matter pollution. Environ Sci Technol
54(15):9210–9216, https://doi.org/10.1021/acs.est.0c00900.
[DOI] [PubMed] [Google Scholar]
168.Rich DQ, Kipen HM, Huang W, Wang G, Wang Y, Zhu P, et al. 2012. Association between changes in air pollution levels during the Beijing Olympics and biomarkers of inflammation and thrombosis in healthy young adults. JAMA. 307(19):2068–2078, https://doi.org/10.1001/jama.2012.3488.
[DOI] [PMC free article] [PubMed] [Google Scholar]
169.Huang W, Wang G, Lu SE, Kipen H, Wang Y, Hu M, et al. 2012. Inflammatory and oxidative stress responses of healthy young adults to changes in air quality during the Beijing Olympics. Am J Respir Crit Care Med
186(11):1150–1159, https://doi.org/10.1164/rccm.201205-0850oc.
[DOI] [PMC free article] [PubMed] [Google Scholar]
170.Tolbert PE, Mulholland JA, Macintosh DL, Xu F, Daniels D, Devine OJ, et al. 2000. Air quality and pediatric emergency room visits for asthma and Atlanta, Georgia. Am J Epidemiol
151(8):798–810, https://doi.org/10.1093/oxfordjournals.aje.a010280.
[DOI] [PubMed] [Google Scholar]
171.Casey JA, Karasek D, Ogburn EL, Goin DE, Dang K, Braveman PA, et al. 2018. Retirements of coal and oil power plants in California: association with reduced preterm birth among populations nearby. Am J Epidemiol
187(8):1586–1594, https://doi.org/10.1093/aje/kwy110.
[DOI] [PMC free article] [PubMed] [Google Scholar]
172.Heale R, Forbes D. 2013. Understanding triangulation in research. Evid Based Nurs
16(4):98–98, 10.1136/eb-2013-101494.
[DOI] [PubMed] [Google Scholar]
174.Romo A, Carceller R, Tobajas J. 2009. Intrauterine growth retardation (IUGR): epidemiology and etiology. Pediatr Endocrinol Rev 6 Suppl
3:332–336. [PubMed] [Google Scholar]
175.Bekkar B, Pacheco S, Basu R, DeNicola N. 2020. Association of air pollution and heat exposure with preterm birth, low birth weight, and stillbirth in the US: A Systematic Review. JAMA Netw Open
3(6):e208243, https://doi.org/10.1001/jamanetworkopen.2020.8243.
[DOI] [PMC free article] [PubMed] [Google Scholar]
176.Wing C, Simon K, Bello-Gomez RA. 2018. Designing difference in difference studies: Best practices for public health policy research. Annu Rev Public Health
39(1):453–469, https://doi.org/10.1146/annurev-publhealth-040617-013507.
[DOI] [PubMed] [Google Scholar]
177.CACES [cited 2021 Dec 30]. CACES RCM/LUR Data Download. Available: https://www.caces.us/data [accessed 30 Dec 2021].
180.Kahn-Lang A, Lang K. 2019. The promise and pitfalls of differences-in-differences: reflections on 16 and pregnant and other applications. J Bus Econ Stat. 38(3):1–14, 10.1080/07350015.2018.1546591. [DOI] [Google Scholar]
181.Goodman-Bacon A. 2018. Difference-in-differences with variation in treatment timing. J Econom
225(2):254–277, https://doi.org/10.1016/j.jeconom.2021.03.014. Available: http://www.nber.org/papers/w25018 [accessed 16 April 2019]. [Google Scholar]
182.Canfield M, Ramadhani T, Langlois P, Waller DK. 2006. Residential mobility patterns and exposure misclassification in epidemiologic studies of birth defects. J Expo Sci Environ Epidemiol
16:538–543, https://doi.org/10.1038/sj.jes.7500501.
[DOI] [PubMed] [Google Scholar]
183.Liew Z, Olsen J, Cui X, Ritz B, Arah OA. 2015. Bias from conditioning on live birth in pregnancy cohorts: an illustration based on neurodevelopment in children after prenatal exposure to organic pollutants. Int J Epidemiol
44(1):345–354, https://doi.org/10.1093/ije/dyu249.
[DOI] [PMC free article] [PubMed] [Google Scholar]
184.Leung M, Kioumourtzoglou MA, Raz R, Weisskopf MG. 2021. Bias due to selection on live births in studies of environmental exposures during pregnancy: a simulation study. Environ Health Perspect
129(4):47001, https://doi.org/10.1289/ehp7961.
[DOI] [PMC free article] [PubMed] [Google Scholar]
187.Frey HC, Rouphail NM, Unal A, Colyar J. 2001. Emissions reduction through better traffic management. An empirical evaluation based upon on-road measurements 2001. Available: https://trid.trb.org/view/661327.
188.Shah SD, Cocker DR, Miller JW, Norbeck JM. 2004. Emission rates of particulate matter and elemental and organic carbon from in-use diesel engines. Environ Sci Technol, 38(9):2544–2550, https://doi.org/10.1021/es0350583.
[DOI] [PubMed] [Google Scholar]
189.Cocker DR, Shah SD, Johnson K, Miller JW, Norbeck JM. 2004. Development and application of a mobile laboratory for measuring emissions from diesel engines. 1. Regulated gaseous emissions. Environ Sci Technol
38(7):2182–2189, https://doi.org/10.1021/es034888d.
[DOI] [PubMed] [Google Scholar]
190.Meinardi S, Nissenson P, Barletta B, Dabdub D, Sherwood Rowland F, Blake DR. 2008. Influence of the public transportation system on the air quality of a major urban center. A case study: Milan, Italy. Atmos Environ
42(34):7915–7923, https://doi.org/10.1016/j.atmosenv.2008.07.046. [Google Scholar]
191.Chen Y, Whalley A. 2012. Green Iinfrastructure: the effects of urban rail transit on air quality. Am Econ J Econ Policy
4(1):58–97, 10.1257/pol.4.1.58. [DOI] [Google Scholar]
192.Titos G, Lyamani H, Drinovec L, Olmo FJ, Močnik G, Alados-Arboledas L. 2015. Evaluation of the impact of transportation changes on air quality. Atmos Environ
114:19–31, https://doi.org/10.1016/j.atmosenv.2015.05.027. [Google Scholar]
193.Li S, Liu Y, Purevjav AO, Yang L. 2019. Does subway expansion improve air quality?
J Environ Econ Manag
96:213–235, https://doi.org/10.1016/j.jeem.2019.05.005. [Google Scholar]
194.Zheng S, Zhang X, Sun W, Wang J. 2019. The effect of a new subway line on local air quality: a case study in Changsha. Transp Res Part Transp Environ
68:26–38, https://doi.org/10.1016/j.trd.2017.10.004. [Google Scholar]
195.Gendron-Carrier N, Gonzalez-Navarro M, Polloni S, Turner MA. 2022. Subways and urban air pollution. Am Econ J Appl Econ
14(1):164–196, https://doi.org/10.1257%2Fapp.20180168.
[DOI] [PMC free article] [PubMed] [Google Scholar]
196.Currie J, Walker R. 2011. Traffic congestion and infant health: evidence from E-ZPass. Am Econ J Appl Econ
3(1):65–90. [Google Scholar]
197.Willis MD, Harris L, Campbell EJ, Chaskes M, Sawyer E, Harleman M, et al. 2024. A population-based cohort study of electronic tolling, traffic congestion, and adverse birth outcomes. Environ Int
183:108355, https://doi.org/10.1016/j.envint.2023.108355.
[DOI] [PubMed] [Google Scholar]
198.Harleman M, Harris L, Willis MD, Ritz B, Hystad P, Hill EL. 2023. Changes in traffic congestion and air pollution due to major roadway infrastructure improvements in Texas. Sci Total Environ
898:165463, https://doi.org/10.1016/j.scitotenv.2023.165463
[DOI] [PubMed] [Google Scholar]
199.Pierson K, Hand ML, Thompson F. 2015. The government finance database: a common resource for quantitative research in public financial analysis. PloS One
10(6):e0130119, https://doi.org/10.1371/journal.pone.0130119.
[DOI] [PMC free article] [PubMed] [Google Scholar]
202.Benmarhnia T, Bailey Z, Kaiser D, Auger N, King N, Kaufman JS. 2016. A difference-in-differences approach to assess the effect of a heat action plan on heat-related mortality, and differences in effectiveness according to sex, age, and socioeconomic status (Montreal, Quebec). Environ Health Perspect
124(11):1694–1699, https://doi.org/10.1289/ehp203.
[DOI] [PMC free article] [PubMed] [Google Scholar]
203.Wing C, Simon K, Bello-Gomez RA. 2018. Designing difference in difference studies: best practices for public health policy research. Annu Rev Public Health
39(1):453–469, https://doi.org/10.1146/annurev-publhealth-040617-013507.
[DOI] [PubMed] [Google Scholar]
204.King G, Nielsen R. 2019. Why propensity scores should not be used for matching. Polit Anal
27(4):435–454, https://doi.org/10.1017/pan.2019.11. [Google Scholar]
205.Willis MD, Hill EL, Ncube CN, Campbell EJ, Harris L, Harleman M, et al. 2023. Changes in socioeconomic disparities for traffic-related air pollution exposure during pregnancy over a 20-year period in Texas. JAMA Netw Open
6(8):e2328012, https://doi.org/10.1001/jamanetworkopen.2023.28012.
[DOI] [PMC free article] [PubMed] [Google Scholar]
206.Chen Y, Hodgson S, Gulliver J, Granell R, Henderson AJ, Cai Y, et al. 2021. Trimester effects of source-specific PM10 on birth weight outcomes in the Avon Longitudinal Study of Parents and Children (ALSPAC). Environ Health
20:4, https://doi.org/10.1186/s12940-020-00684-w.
[DOI] [PMC free article] [PubMed] [Google Scholar]
207.Rodríguez-Fernández A, Ramos-Castillo N, Ruiz-De la Fuente M, Parra-Flores J, Maury-Sintjago E. 2022. Association of prematurity and low birth weight with gestational exposure to PM2.5 and PM10 particulate matter in Chilean newborns. Int J Environ Res Public Health
19(10):6133, https://doi.org/10.3390/ijerph19106133.
[DOI] [PMC free article] [PubMed] [Google Scholar]
208.Dadvand P, Ostro B, Figueras F, Foraster M, Basagaña X, Valentín A, et al. 2014. Residential proximity to major roads and term low birth weight: the roles of air pollution, heat, noise, and road-adjacent trees. Epidemiol Camb Mass
25(4):518–525, https://doi.org/10.1097/ede.0000000000000107. [DOI] [PubMed] [Google Scholar]
209.Wieser AA, Scherz M, Passer A, Kreiner H. 2021. Challenges of a healthy built environment: air pollution in construction industry. Sustainability
13(18):10469, 10.3390/su131810469. [DOI] [Google Scholar]
210.Reff A, Bhave PV, Simon H, Pace TG, Pouliot GA, Mobley JD, et al. 2009. Emissions Inventory of PM2.5 trace elements across the United States. Environ Sci Technol
43(15):5790–5796, https://doi.org/10.1021/es802930x.
[DOI] [PubMed] [Google Scholar]
211.Giunta M. 2020. Assessment of the environmental impact of road construction: modelling and prediction of fine particulate matter emissions. Build Environ
176:106865, 10.1016/j.buildenv.2020.106865. [DOI] [Google Scholar]
212.Ding X, Lian Zhao T, Qu G, Zhang J, et al. 2021. The impact of prenatal stressful life events on adverse birth outcomes: a systematic review and meta-analysis. J Affect Disord
287:406–416, https://doi.org/10.1016/j.jad.2021.03.083.
[DOI] [PubMed] [Google Scholar]
213.Bergmann ML, Andersen ZJ, Amini H, Khan J, Lim YH, Loft S, et al. 2022. Ultrafine particle exposure for bicycle commutes in rush and non-rush hour traffic: a repeated measures study in Copenhagen, Denmark. Environ Pollut
294:118631, https://doi.org/10.1016/j.envpol.2021.118631.
[DOI] [PubMed] [Google Scholar]
214.Piscitello A, Bianco C, Casasso A, Sethi R. 2021. Non-exhaust traffic emissions: Sources, characterization, and mitigation measures. Sci Total Environ
766:144440, https://doi.org/10.1016/j.scitotenv.2020.144440.
[DOI] [PubMed] [Google Scholar]
215.Fussell JC, Franklin M, Green DC, Gustafsson M, Harrison RM, Hicks W, et al. 2022. A review of road traffic-derived non-exhaust particles: emissions, physicochemical characteristics, health risks, and mitigation measures. Environ Sci Technol
56(11):6813–6835, https://doi.org/10.1021/acs.est.2c01072.
[DOI] [PMC free article] [PubMed] [Google Scholar]
216.Dzhambov AM, Lercher P. 2019. Road Traffic Noise Exposure and Birth Outcomes: An Updated Systematic Review and Meta-Analysis. Int J Environ Res Public Health
16(14):2522, https://doi.org/10.3390%2Fijerph16142522.
[DOI] [PMC free article] [PubMed] [Google Scholar]
217.Willis MD, Schrank D, Xu C, Harris L, Ritz BR, Hill EL, et al. 2022. A population-based cohort study of traffic congestion and infant growth using connected vehicle data. Sci Adv
8(43):eabp8281, https://doi.org/10.1126/sciadv.abp8281.
[DOI] [PMC free article] [PubMed] [Google Scholar]
218.Willis MD, Hill EL, Kile ML, Carozza S, Hystad P. 2020. Assessing the effectiveness of vehicle emission regulations on improving perinatal health: a population-based accountability study. Int J Epidemiol
49(6):1781–1791, https://doi.org/10.1093/ije/dyaa137. [DOI] [PMC free article] [PubMed] [Google Scholar]
219.Hertz-Picciotto I. 2003. Is it time to abandon adjustment for birth weight in studies of infant mortality?
Paediatr Perinat Epidemiol
17(2):114–116, https://doi.org/10.1046/j.1365-3016.2003.00483.x.
[DOI] [PubMed] [Google Scholar]
220.Willis MD, Hill EL, Ncube CN, Campbell EJ, Harris L, Harleman M, et al. 2023. Changes in socioeconomic disparities for traffic-related air pollution exposure during pregnancy over a 20-year period in Texas. JAMA Netw Open
6(8):e2328012, https://doi.org/10.1001/jamanetworkopen.2023.28012.
[DOI] [PMC free article] [PubMed] [Google Scholar]
221.Nieuwenhuijsen MJ, Ristovska G, Dadvand P. 2017. WHO Environmental Noise Guidelines for the European Region: a systematic review on environmental noise and adverse birth outcomes. Int J Environ Res Public Health
14(10):1252, https://doi.org/10.3390/ijerph14101252.
[DOI] [PMC free article] [PubMed] [Google Scholar]
222.Askariyeh MH, Venugopal M, Khreis H, Birt A, Zietsman J. 2020. Near-road traffic-related air pollution: resuspended PM2.5 from highways and arterials. Int J Environ Res Public Health
17(8):2851, https://doi.org/10.3390/ijerph17082851.
[DOI] [PMC free article] [PubMed] [Google Scholar]
223.Harrison RM, Jones AM, Gietl J, Yin J, Green DC. 2012. Estimation of the contributions of brake dust, tire wear, and resuspension to nonexhaust traffic particles derived from atmospheric measurements. Environ Sci Technol
46(12):6523–6529, https://doi.org/10.1021/es300894r.
[DOI] [PubMed] [Google Scholar]
224.Noland RB, Quddus MA. 2005. Congestion and safety: a spatial analysis of London. Transp Res Part Policy Pract
39(7):737–754, https://doi.org/10.1016/j.tra.2005.02.022. [Google Scholar]
225.Congiu T, Sotgiu G, Castiglia P, Azara A, Piana A, Saderi L, et al. 2019. Built environment features and pedestrian accidents: an Italian retrospective study. Sustainability
11(4):1064, 10.3390/su11041064. [DOI] [Google Scholar]
226.Mullan E. 2003. Do you think that your local area is a good place for young people to grow up? The effects of traffic and car parking on young people’s views. Health Place
9(4):351–360, https://doi.org/10.1016/s1353-8292(02)00069-2.
[DOI] [PubMed] [Google Scholar]
227.Ebelt S, Baxter L, Erickson HS, Henneman LRF, Lange S, Luben TJ, et al. 2023. Air pollution accountability research: moving from a chain to a web. Glob Epidemiol
6:100128, https://doi.org/10.1016/j.gloepi.2023.100128.
[DOI] [PMC free article] [PubMed] [Google Scholar]
228.Boogaard H, van Erp AM, Walker KD, Shaikh R. 2017. Accountability studies on air pollution and health: the HEI experience. Curr Environ Health Rep
4(4):514–522, https://doi.org/10.1007/s40572-017-0161-0.
[DOI] [PubMed] [Google Scholar]
229.Di HK, Gan Y, Lu K, Wang C, Zhu Y, Meng X, et al. 2022. Maternal smoking status during pregnancy and low birth weight in offspring: systematic review and meta-analysis of 55 cohort studies published from 1986 to 2020. World J Pediatr
18(3):176–185, https://doi.org/10.1007/s12519-021-00501-5.
[DOI] [PubMed] [Google Scholar]
230.Wilcox AJ, Weinberg CR, O’Connor JF, Baird DD, Schlatterer JP, Canfield RE, et al. 1988. Incidence of early loss of pregnancy. N Engl J Med
319(4):189–194, https://doi.org/10.1056/nejm198807283190401.
[DOI] [PubMed] [Google Scholar]
231.Raz R, Kioumourtzoglou MA, Weisskopf MG. 2018. Live-birth bias and observed associations between air pollution and autism. Am J Epidemiol
187(11):2292–2296, https://doi.org/10.1093%2Faje%2Fkwy172.
[DOI] [PMC free article] [PubMed] [Google Scholar]
232.Kioumourtzoglou MA, Raz R, Wilson A, Fluss R, Nirel R, Broday DM, et al. 2019. Traffic-related air pollution and pregnancy loss. Epidemiol Camb Mass
30(1):4–10, https://doi.org/10.1097%2FEDE.0000000000000918. [DOI] [PMC free article] [PubMed] [Google Scholar]
Res Rep Health Eff Inst.
Commentary by Review Committee
Research Report 223, Impacts of Vehicle Emission Regulations and Local Congestion Policies on Birth Outcomes Associated with Traffic Air Pollution, P. Hystad et al.
Dr. Perry Hystad’s 3-year study, “Impacts of Vehicle Emission Regulations and Local Congestion Policies on Birth Outcomes Associated with Traffic Air Pollution,” began in April 2020. Total expenditures were $957,174. The draft Investigators’ Report from Hystad and colleagues was received for review in September 2023. A revised report, received in March 2024, was accepted for publication in April 2024. During the review process, the HEI Review Committee and the investigators had the opportunity to exchange comments and clarify issues in both the Investigators’ Report and the Review Committee’s Commentary.
This document has not been reviewed by public or private party institutions, including those that support the Health Effects Institute; therefore, it may not reflect the views of these parties, and no endorsements by them should be inferred.
Air quality regulations are essential for protecting environmental quality and human health but typically incur an economic cost. It is, therefore, essential to understand whether air quality regulations result in the intended improvements by comparing the predicted benefits to air quality and health with those actually achieved. This area of study known as environmental accountability research (sometimes also referred to as intervention research) evaluates the extent to which environmental regulations, other interventions, or “natural” experiments have yielded improved air quality and public health. A major challenge in this research field is isolating changes that can be attributed to the policy in question from improvements that might be due to other unrelated regulations or long-term trends. This challenge is a particular concern when policies target numerous pollutant sources, affect large geographic regions, and take several years to fully implement.
Over the past two decades, the Health Effects Institute (HEI) has supported air pollution accountability research, contributing to research design, method development, and evidence synthesis (see Preface). Through a series of Requests for Applications (RFAs*), HEI has now funded an extensive program of more than 20 studies to date that have assessed a wide variety of air quality actions that have targeted both point and mobile sources of air pollution. Early studies often focused on local-level actions that were implemented over a relatively short time frame and were sometimes temporary in nature. HEI later solicited research that evaluated actions with a larger geographical scope and were implemented over longer periods.
In its RFA 18-1, “Assessing Improved Air Quality and Health from National, Regional, and Local Air Quality Actions,” HEI aimed to fund research studies that would assess the health effects of air quality actions, with a particular interest in (1) national- or regional-scale regulatory actions implemented over multiple years; (2) local actions targeted at improving air quality in urban areas, with well-documented air quality problems and programs to address them; and (3) regulatory programs to improve air quality around major ports and transportation hubs and corridors. HEI additionally aimed to fund research studies that would develop methods required for, and specifically suited to, conducting research to assess the health effects of air quality actions and make those methods accessible and available to other researchers.
In response to the RFA, Dr. Perry Hystad of Oregon State University submitted an application titled “Impacts of Vehicle Emission Regulations and Local Congestion Policies on Birth Outcomes Associated with Traffic Air Pollution.” Dr. Hystad proposed to assess changes in traffic-related air pollution and birth outcomes in a population-based cohort of 8 million births in Texas from 1996–2016 associated with long-term cumulative regulatory improvements of motor vehicle emissions and shorter-term local congestion-reduction programs. HEI’s Research Committee recommended funding Dr. Hystad’s proposed study because it was highly innovative, with particular strengths being the large sample size, the causal modeling analyses, and the potential policy relevance of the results in evaluating both long-term regulations and shorter-term actions.
This Commentary provides the HEI Review Committee’s evaluation of the study. It is intended to aid HEI’s sponsors and the public by highlighting the study’s strengths and limitations as well as placing the Investigators’ Report into a broader scientific and regulatory context.
SCIENTIFIC AND REGULATORY BACKGROUND
EMISSIONS AND CONCENTRATIONS OF TRAFFIC-RELATED AIR POLLUTION
Traffic-related air pollution (TRAP) is a complex mixture of gases and particles emitted from motor vehicles. It includes nitrogen oxides (NOx), elemental carbon (EC), and particulate matter ≤2.5 μm in aerodynamic diameter (PM2.5). TRAP includes tailpipe emissions from the exhaust of on-road vehicles and nontailpipe emissions generated, for example, from road, brake, and tire wear.1
In many high-income countries, TRAP concentrations have decreased over recent decades due to both air quality regulations and improvements in vehicular emission control technologies.2 Using NO2 as a marker for the TRAP mixture, the United States has seen a 60% decrease in annual average NO2 concentrations since 1990.3 In Texas, annual NO2 concentrations declined by 53.8% from 1996 to 2016, while other states, such as California and Massachusetts, saw 50.7% and 64.4% respective declines over the same period.4 Despite the downward trend in emissions associated with TRAP, there remain concerns about TRAP exposures, given the increase in traffic volume and congestion due to population growth, urbanization, and economic activity.5 For example, annual vehicle miles traveled (VMT) increased by 194% from 1970 to 2023.3
At the same time, TRAP exposures remain a source of inequality across the population. Racial, ethnic, and socioeconomic disparities in air pollution exposures have been widely documented throughout the United States6–9, and these disparate exposures often coincide with living close to busy roadways.10
HEALTH EFFECTS OF TRAP
A recent systematic review from HEI of over 350 studies on long-term exposure to TRAP and select adverse health outcomes found a high level of confidence in the evidence that strong associations exist between TRAP and premature mortality due to cardiovascular diseases1 (and Commentary Figure 1). High confidence in an association was also found between TRAP and lung cancer mortality, asthma onset in children and adults, and acute lower respiratory infections in children. Other reviews have arrived at similar conclusions, including the evaluation of the carcinogenic effect of diesel and gasoline exhaust by the International Agency for Research on Cancer11, the Integrated Science Assessment of NO2 by the US Environmental Protection Agency12, and an umbrella review of TRAP from Health Canada.13
Associations between long-term exposure to TRAP and selected health outcomes. Health outcomes for which the overall confidence in the evidence was low to moderate, low, or very low are not in the figure (Source: HEI 2022).
Although air pollution exposures have been associated with various adverse birth outcomes14–17, HEI’s 2022 systematic review reported mixed evidence linking TRAP exposures to adverse birth outcomes. Exposure to PM2.5 over the entire course of pregnancy was found to be most clearly associated with measures of fetal growth restriction (term low birth weight and small for gestational age [SGA]). However, the evidence for associations with other traffic-related air pollutants (e.g., NO2, NOx, and EC) was largely inconclusive for other birth outcomes, including term birth weight, term low birth weight, SGA, and preterm birth (see Box 1 for an explanation of birth outcomes and how they are assessed).1 Meanwhile, other reviews have reported associations between proximity to roads and preterm birth, term low birth weight, and SGA.17,46
TRAP REGULATIONS
A suite of regulatory actions to reduce air pollution, including TRAP, have been implemented in the United States over recent decades. These actions include national regulations that target emissions from on-road vehicles, such as the Tier regulations that set limits on tailpipe emissions of NOx, PM2.5, CO (carbon monoxide), and VOCs (volatile organic compounds) in a tiered progression of increasing stringency.18 Other regulations include programs and standards that target fuel composition and quality, such as the Renewable Fuel Standard (RFS) program and fuel economy standards (corporate average fuel economy, or CAFE).19 In addition, many local programs and policies designed to reduce traffic congestion and excess vehicle emissions (e.g., electronic toll implementation and roadway capacity improvements) can contribute to reductions in air pollution. For example, the Texas Clear Lanes is a statewide strategic plan targeting traffic congestion in the five major metropolitan areas in the state.20
Examining the effectiveness of air quality interventions is challenging. Common difficulties include differentiating between the effects of simultaneous policy interventions and disentangling air quality improvements among concurrent changes in background trends in air quality and health.21 For practical reasons, many early accountability studies focused on short-term, local air quality actions or natural experiments, such as factory closures or congestion charging schemes. An increasing number of accountability studies have since focused on the effectiveness of long-term emission regulations on population health.22,23 However, choosing a study design and analytical methods to address the possibility of residual confounding and identifying appropriate control populations remain difficult in establishing causal relationships for all types of accountability studies.24,25 Recent reviews have recommended using multiple statistical methods and a multidisciplinary approach to help researchers design and conduct more robust studies to examine current and future air pollution regulations.26,27
In this study, Hystad and colleagues leverage an interdisciplinary team and multiple statistical methods and study designs to examine the impact of different types of air quality interventions. They address simultaneous policy interventions by considering the cumulative effects of vehicle emissions regulations in evaluating long-term trends in air quality and health outcomes. They also take advantage of the diverse natural experiments that the implementation of local congestion programs can provide in assessing changes in air pollution and health outcomes before and after program implementation. Texas provides an ideal study location to examine these changes because about 1.7 million pregnant individuals lived within 500 m of a highway or expressway during the 1996 to 2016 time period.
STUDY OBJECTIVES AND OVERALL STUDY DESIGN
The study aimed to assess changes in birth outcomes for all recorded births (8.1 million) in Texas from 1996 to 2016 associated with the following:
Long-term changes in TRAP that might result from the cumulative effects of national vehicle emissions regulations
Localized changes in TRAP over shorter periods that might result from the implementation of local congestion-reduction programs
To address Aim 1, Hystad and colleagues evaluated the association between metrics of TRAP exposure near residential addresses and birth outcomes and changes in these associations over a 20-year period using linear and logistic regression models. They hypothesized that the magnitude of the association between adverse birth outcomes and pregnant individuals exposed to TRAP would decrease over the 1996 to 2016 time period, paralleling regulatory progress aimed at reducing TRAP. They also hypothesized that the benefits of TRAP reductions would not be evenly distributed across socioeconomic and demographic characteristics. To address the potential confounding of TRAP exposures with sociodemographic characteristics and other environmental exposures, the investigators conducted an additional analysis under this aim that used linear and logistic regressions to compare birth outcomes of matched pregnant individuals living upwind or downwind of the same high-traffic road.
To address Aim 2, the investigators first evaluated the association between metrics of traffic congestion and term birth weight using linear regression models. They next assessed whether the implementation of tolling systems and roadway improvement projects resulted in changes in birth outcomes among pregnant individuals living nearby using a quasi-experimental design and applying difference-in-differences (DiD) methods. Hystad and colleagues hypothesized that the implementation of such local congestion-reduction programs would be associated with a reduction in the risks of adverse birth outcomes among those pregnant individuals living near a major road during their pregnancy.
The investigators assembled a cohort of all recorded births in Texas from 1996 to 2016, a total of 8.1 million births, using data from the Texas Department of Health and Human Services. The birth cohort included information on birth outcomes, complete residential locations, and sociodemographic characteristics for pregnant individuals. Residential addresses were assigned several measures of traffic exposure, including annual NO2 concentrations, traffic volume and congestion markers, and wind metrics. Residential addresses were also linked to data on environmental measurements (e.g., tree cover, temperature, and impervious surface), neighborhood characteristics, and household characteristics. Specific to Aim 2, the investigators also developed a spatial-temporal database for toll implementation projects and roadway construction projects to improve congestion.
SUMMARY OF METHODS AND STUDY DESIGN
Hystad and colleagues used a research triangulation approach to address the two aims of their study. Research triangulation integrates results from multiple approaches, each subject to various unrelated biases, to obtain more robust answers to research questions of interest.28,29 In this study, the investigators used a variety of TRAP metrics and subsets of the study population; involved multiple researchers with diverse expertise; applied theoretical perspectives across epidemiology, exposure science, and econometrics; and used different methods to examine their two aims.
STUDY POPULATION
Hystad and colleagues created a cohort of 8.1 million recorded births in Texas between January 1, 1996, and December 31, 2016, using Vital Statistics data from the Texas Department of Health and Human Services. These data included information on birth outcomes and infant sex and detailed individual information related to the infants’ parents, including demographic, socioeconomic, and health-related characteristics.
The investigators were interested in examining birth outcomes related to the pathway of intrauterine growth restriction and birth outcomes related to triggers for premature labor (see Box 1). To distinguish these pathways, they assessed four birth outcomes: term birth weight, term low birth weight, preterm birth, and very preterm birth.
Box 1: Birth Outcomes.
Birth outcomes can be characterized in several ways that provide important indicators of infant health. Intrauterine growth restriction (IUGR) is a measure that indicates whether a fetus does not meet its growth potential due to a variety of maternal (e.g., age, race, smoking, alcohol consumption), fetal, or placental factors that lead to low birth weight.30 Triggers for premature labor (labor that occurs before a certain gestational age) can be attributed to multiple pathological processes, of which genetic and environmental factors are likely contributors.31 To distinguish the pathways of IUGR and triggers for premature labor, Hystad and colleagues assessed four outcomes based on birth weight (grams) and gestational age (weeks):
Term birth weight: birth weight among births whose gestational age was 37–42 weeks
Term low birth weight: birth weight less than 2,500 g among births whose gestational age was 37–42 weeks
Preterm birth: birth before an estimated gestational age of 37 completed weeks
Very preterm birth: birth before an estimated gestational age of 32 completed weeks
The “small for gestational age” outcome was initially included in this study but ultimately not examined because its definition conflates the biological pathways of growth during gestation and length of gestation and complicates the isolation of the individual factors that affect these two pathways.32
EXPOSURE ASSESSMENT
Hystad and colleagues derived several metrics of TRAP exposure, which they assigned to residential addresses reported at delivery time. As a marker of the overall tailpipe emissions exposure mixture, residential addresses were assigned annual NO2 concentrations (ppb) from 1996 to 2016 using an existing hybrid land use regression model developed by researchers at the University of Washington.4 As markers of traffic volume, the investigators constructed estimates of total and truck-specific VMT for all roads within 500 m of residential addresses using historical Texas roadway inventory databases. As markers of traffic congestion, the investigators constructed estimates of AADT counts, AADT per lane, and person-hours of delay due to all traffic and truck-specific traffic within a range of distances (see Commentary Table) from a residential address using the Texas’ Most Congested Roadways database from the Texas A&M Transportation Institute. Wind metrics were computed as the hours spent upwind or downwind of a high-traffic road segment within 500 m of a residential address (considering shielding from buildings and trees) using meteorological and building data. Hystad and colleagues also assembled a spatial-temporal database of toll implementation and roadway construction projects using various Texas Department of Transportation datasets.
Commentary Table.
Summary of Main Analyses Conducted in This Study
Chapter
Aim
Time Period
Study Population
TRAP Exposure
Birth Outcome
Methods
3
Aim 1: Assess changes in associations of TRAP exposures and adverse birth outcomes over time
1996 to 2016
N = 6,158,518
Annual concentrations of NO2, total, and truck VMT within 500 m of residences
Term birth weight Term low birth weight Preterm birth Very preterm birth
Linear and logistic regression models of associations and of change in associations over time
4
Aim 1: Address potential effect of residual and unmeasured confounding of TRAP exposures with other environmental exposures and sociodemographic characteristics
2007 to 2016
N = 81,386
Living downwind (within 500 m) of high traffic (≥25,000 of AADT counts) roads compared to upwind of the same road
Term birth weight Term low birth weight Preterm birth Very preterm birth
Linear and logistic regression of neighbors matched by wind
5
Aim 2: Examine direct associations between congestion and term birth weight
2015 to 2016
N = 579,122
Total and truck-specific person-hours of delay within 500 m of a residential address
Term birth weight
Linear regression models of congestion exposure
6
Aim 2: Assess effect of tolling on changes in adverse birth outcomes
1999 to 2016
N = 625,279
AADT per lane; near (500 m) and far (2–5 km) from toll road locations
Term birth weight Term low birth weight Preterm birth Very preterm birth
Quasi-experimental design with DiD estimator comparing before and after tolling changes with linear and logistic regression
7
Aim 2: Assess effect of roadway construction on changes in adverse birth outcomes
2009 to 2016
N = 261,442
AADT, AADT per lane, and person-delay hours; near (300 m) and far (300–1,000 m) from construction locations
Term birth weight Term low birth weight Preterm birth Very preterm birth
Quasi-experimental design with DiD estimator comparing before, during, and after roadway improvement projects with linear and logistic regression
AADT = annual average daily traffic; DiD = difference-in-differences; TRAP = traffic-related air pollution; VMT = vehicles miles traveled.
Residential addresses were also linked to contextual factors related to the environment, neighborhood, and household where the infants’ parents lived at birth that might modify traffic exposures.
MAIN ANALYSES
Hystad and colleagues used a research triangulation approach to address their study aims. For brevity, see Commentary Table for a summary of the analyses used in their study.
Effects of National-Scale Cumulative Regulatory Improvements
The investigators conducted two analyses under Aim 1. In the first analysis, Hystad and colleagues used estimates of annual total VMT over the study period to indicate the cumulative effect of national regulations targeting vehicle tailpipe emissions. Using linear and logistic regression models, they then examined the association between total VMT and other TRAP exposure metrics (truck-specific VMT, concentrations of NO2) and birth outcomes across the 20-year study period. In these models, the metrics of TRAP exposure were examined using quintile-based exposure categories, and the effect on birth outcomes was expressed as changes in weight (g) and odds ratios (ORs).
The investigators explored base models that adjusted for a basic set of covariates and adjusted models with added complexity that included additional individual-level and some neighborhood-level covariates. All term birth weight models included a covariate for the length of gestation. Base and adjusted models were run over the entire study period and for 3-year rolling periods to evaluate whether the associations between birth outcomes and differences in air pollution trends changed over time. The investigators also explored base and adjusted models stratified by race, ethnicity, educational attainment, and census tract-level income to evaluate the potential effect modification of TRAP on adverse birth outcomes and differential changes in associations over time.
In a second analysis, Hystad and colleagues investigated birth outcomes over the 2007 to 2016 time period for matched pregnant individuals who were living within 500 m upwind or downwind of the same major (i.e., high-traffic, ≥25,000 AADT) road segment to disentangle the effects of TRAP exposures on birth outcomes from the confounding effects of other traffic-related exposures and sociodemographic characteristics. Differences in birth outcomes for matched downwind (exposed) and upwind (control) residences of pregnant individuals were evaluated using linear and logistic regression models and expressed as changes in weight (g) and ORs. The investigators explored base and adjusted models as well as stratified models by distance to major roads.
Effects of Local Congestion-Reduction Actions
Hystad and colleagues conducted three analyses under Aim 2. First, the investigators evaluated changes in term birth weight (g) and metrics of traffic congestion. The investigators ran base models and two sets of adjusted models. Base models adjusted for VMT within the respective buffer distance of the residence, and adjusted models included individual-level and environmental covariates.
In subsequent analyses, Hystad and colleagues evaluated whether the implementation of tolling systems and roadway improvement projects resulted in changes in birth outcomes among pregnant individuals. Both analyses used a quasi-experimental design with DiD methods. The quasi-experimental design aims to establish the causal effect of interventions without randomization. DiD methods are commonly used in epidemiological studies to estimate the effects due to changes in exposure over time while also controlling for differences between the control and treatment groups. Both analyses reported associations as changes in term birth weight (g) grams and ORs.
In the analysis of tolling system projects, the investigators examined three distinct types of tolling implementation: (1) switching an existing toll road to electronic tolling (i.e., cash payments were removed), (2) adding tolled express lanes to an existing road, and (3) switching an existing tollbooth on a toll road to electronic tolling (i.e., the physical booth was replaced with overhead sensors). They first explored whether toll implementation resulted in changes to roadway traffic volume from 1999 to 2016 by comparing the difference in traffic on treated roads and matched control groups of roads in each year-long period relative to the year the toll project began. Next, Hystad and colleagues assessed the difference in the association between the proximity of the pregnant individual’s residential address to the project and adverse birth outcomes once the toll road was opened relative to the association between this proximity and adverse birth outcomes before the toll road was operational.
In analyzing roadway improvement projects, Hystad and colleagues defined six categories of projects that might affect congestion: widening existing roads, building new roads, improving bridges, installing intelligent transportation systems (ITS), improving intersections, and installing or upgrading traffic signals. They used a DiD estimator with linear and logistic regression to assess traffic congestion and NO2 changes for road segments that experienced road widening or intersection improvement projects between 2010 and 2019 relative to matched control road segments that did not experience a roadway improvement project. They next compared adverse birth outcomes of pregnant individuals living near any of the six categories of roadway improvement projects to those living farther from a roadway improvement project before, during, and after construction from 2007 to 2016. The analysis was restricted to assessing proximity to a single project within 1,000 m of the residence with a start or end date within 5 years of the infant’s birth date.
ADDITIONAL ANALYSES
To examine the potential effect of restricting their analyses to births that made it to full term, Hystad and colleagues conducted sensitivity analyses that examined birth weight associations for all births and for all births with no adjustment for gestational duration. These analyses were intended to check for bias that might occur because traffic is a risk factor for preterm birth. There may be underlying conditions that influence birth weight that also affect being born at term.
Specific to the analysis of tolling implementation, the investigators conducted a separate analysis in which the road matching process was repeated for roads that are not treated but intersect a treated road or are within 2 km of a treated road to examine the potential for rerouting traffic from tolling.
SUMMARY OF KEY RESULTS
EFFECTS OF NATIONAL-SCALE CUMULATIVE REGULATORY IMPROVEMENTS
Over the 20-year study period (1996 to 2016), estimated NO2 exposures decreased by 59% for pregnant individuals, and total VMT (within 500 m) decreased by 9.4%. Mean annual NO2 exposure for pregnant individuals for the early years of the study period (1996 to 1998) was 13.7 ppb, and total VMT was 16,198 compared to 6.5 ppb and 15,976 VMT for the later years (2014 to 2016). Hystad and colleagues observed consistent adjusted associations between TRAP exposures and birth outcomes (Commentary Figure 2). Higher levels of NO2 and total VMT were associated with decreases in term birth weight and increases in the ORs for term low birth weight, with weaker associations for preterm and very preterm birth. Associations between birth outcomes and truck-specific VMT were similar to the associations for total VMT.
Associations between quintiles of exposure for (top) NO2 air pollution and (bottom) total VMT within 500 m of pregnant individual’s addresses and birth outcome measures. Associations for term birth weight are reported as a change in grams. Associations for all other outcomes are reported as ORs. Quintile one is the reference category. Results presented are from adjusted models that include adjustment for individual (e.g., infant sex, maternal age, smoking, maternal weight gain) and some neighborhood (e.g., income) characteristics.
Hystad and colleagues used time-stratified models (3-year rolling periods from 1996 to 2016) to evaluate whether results differed between the beginning and end of the study period. Over those two decades, they observed that the magnitude of the ORs for associations between total and truck VMT and term low birth weight, preterm birth, and very preterm birth became smaller over time; this result indicated that the highest (vs. lowest) quintile of VMT was associated with fewer infants being born at term with low birth weight and fewer born prematurely by the end of the study period. For example, the odds ratio (OR) for preterm birth decreased from 1.08 (comparing the highest vs. lowest quintile of total VMT) in 1996 to 1.03 in 2016, a 60% reduction in the trends in OR for premature delivery. In contrast, they observed increases in the magnitude of the association between total and truck VMT and term birth weight in time-stratified models (i.e., infant weight at birth declined over time). The association for term birth weight declined from –8.7 g to –11.9 g for total VMT and from –4.1 g to –9.3 g for truck VMT, comparing the start of the study period to the end of the study period (see Investigators’ Report Figure 7).
Large differences in race, ethnicity, and sociodemographic factors were observed when comparing the lowest and highest quintiles of exposure to total VMT. In some cases, the difference between the lowest and highest quintiles of exposure was as high as 60% (for Black pregnant individuals). Although the differences in quintiles of exposure became smaller over time within groups, the pattern of disparity in exposure between groups remained. In models stratified by race, ethnicity, maternal education, and neighborhood income, Hystad and colleagues surprisingly found larger adjusted associations between total VMT and birth outcomes for White non-Hispanic pregnant individuals, higher-educated individuals, and individuals living in higher-income neighborhoods. Throughout the study period, VMT exposure increasing over time was associated with decreasing term birth weight (i.e., infants were born at lower weights than in earlier years) for White non-Hispanic pregnant individuals. The odds of low birth weight for infants born at term and for premature delivery (comparing the highest vs. lowest quintile of total VMT) decreased over time for all racial, ethnic, education, and neighborhood income groups, except for Black non-Hispanic pregnant individuals. For this subgroup, the ORs (comparing the highest vs. lowest quintile of total VMT) for term low birth weight and preterm birth became slightly larger.
In analyses testing for differences between matched upwind (control) and downwind (exposed) residential addresses in birth outcomes, Hystad and colleagues found that living downwind of the same major road was associated with an 11.6 g (95% confidence interval [CI]: –18.01, –5.21) decrease in term birth weight in adjusted models, but no significant associations with living downwind of the same major road and term low birth weight, preterm birth, or very preterm birth were observed, except at very close distances. For example, in models stratified by distance to a major road, living downwind and within 100 m of a major road was associated with an OR of 3.68 (95% CI: 1.71–7.90) for very preterm birth.
EFFECTS OF LOCAL CONGESTION-REDUCTION ACTIONS
In analyses of traffic congestion measures, Hystad and colleagues observed consistent associations between congestion metrics and reductions in term birth weight. In adjusted models that controlled for both individual characteristics and environmental co-exposures, a decrease in term birth weight of 8.9 g was found for the highest (vs. lowest) quintile of person-hours of traffic delay and a decrease of 8.4 g for the highest (vs. lowest) quintile of person-hours of truck traffic delay.
The investigators reported minimal reductions in traffic volume and delay after the implementation of tolling (not shown). For roadway improvement projects, road widening reduced congestion by 33% and reduced NO2 exposures by 13% after implementation. Intersection projects reduced congestion by 52% and NO2 by 12%. Little evidence was found of improvement in birth outcomes (i.e., increases in the weight of infants born at term or fewer premature deliveries) for analyses of toll implementation and roadway improvement projects (Commentary Figure 3).
Adjusted associations after (left) tolling implementation within 500 m distance of pregnant individuals’ addresses and (right) roadway improvement projects within 300 m of pregnant individuals’ addresses and adverse birth outcomes using DiD methods. Associations for term birth weight are reported as a change in grams. Associations for all other outcomes are reported as ORs. Results presented are from adjusted models that include adjustment for individual (e.g., infant sex, maternal age, smoking, maternal weight gain) and some neighborhood (e.g., income) characteristics.
In analyses of toll implementation projects, residence within 500 m of a tolled road after implementation was found to be associated with a small decrease in term birth weight (–4.5 g, 95% CI: –11.7, 2.6), and no association was found with the odds of term low birth weight in adjusted models (compared to term births with residences 2–5 km from a tolled road before and after project implementation). Adjusted ORs exhibited a similar lack of association for preterm birth outcome measures.
In analyses of roadway improvement projects, Hystad and colleagues report a 19% increase in the odds of infants who were born at term with low birth weight and a small, nonsignificant reduction in term birth weight in adjusted models for pregnant individuals living within 300 m (compared with those within 300–1,000 m) of a project for the period during project construction. They report null associations for both preterm and very preterm births. Contrary to their hypothesis that birth outcomes would improve in the years after construction ended, the investigators also reported null associations for all outcomes for the period after construction. For example, the odds of term low birth weight was 1.24 (95% CI: 0.95, 1.63) for the period after construction.
ADDITIONAL ANALYSES
In sensitivity analyses that examined the associations for birth weight for all births and for all births with no adjustment for gestational age, Hystad and colleagues report similar associations to those reported for term births. The effect estimate for birth weight for all births (adjusting for gestational age) and VMT only differed from the effect estimate for term birth weight and VMT by 0.4 g. Without adjusting for gestational age, the effect estimate for birth weight for all births and VMT only differed from the effect estimate for term birth weight and VMT by 0.2 g. A comparison of birth weight associations and NO2 exposures yielded similarly marginal differences.
In an analysis examining the potential for rerouting away from tolls, no significant rerouting was evident in examining intersecting roadways and roadways partially affected by toll changes. However, Hystad and colleagues note that these results might not be as reliable due to preexisting differences in congestion trends.
HEI REVIEW COMMITTEE’S EVALUATION
In its independent evaluation of the study, the Review Committee appreciated that this accountability study was thoughtfully designed (1) by using a research triangulation approach to support the robustness of the findings and (2) by including an assessment of cumulative effects of national-scale regulation and localized congestion-reduction actions on adverse birth outcomes. The Committee commends Hystad and colleagues for carefully thinking through their research questions and designing a set of complementary analyses that evaluated infant health, which has not yet been as widely examined in accountability research as other health outcomes and is an outcome that can be consistently tracked over time.
The Committee largely agreed with the interpretation of results reported by the investigators. First, the investigators reported that TRAP exposure measures were consistently associated with adverse birth outcomes over the 20 years of their study. During this period, NO2 exposures for pregnant individuals decreased by over 50%, while the total amount of VMT within 500 m of residential addresses of pregnant individuals decreased by a smaller amount (9.4%). The magnitude of the associations between these markers of TRAP and several of the adverse birth outcomes examined also decreased over time. Collectively, these results suggest that there should be improvements in community health because of reduced exposure over time and the potential toxicity of those exposures. However, these improvements could also reflect changes in other unmeasured characteristics over time. Such improvement in birth outcomes could potentially be attributed to the cumulative effects of long-term regulatory policies aimed at reducing tailpipe emissions. The investigators also observed relative disparities in TRAP exposures for pregnant individuals by sociodemographic characteristics that persisted throughout the study period, especially for Black pregnant individuals. In contrast to the findings related to traffic volumes and emissions, actions aimed at reducing traffic congestion seemed to suggest little improvement in birth outcomes, as is evident in the analyses of toll implementation and roadway improvement projects.
Overall, the Committee found the results interesting and agreed with the study’s conclusion that the cumulative effects of national regulations aimed at reducing motor vehicle tailpipe emissions were more successful at decreasing adverse birth outcomes compared to local actions aimed at reducing traffic congestion. However, the Committee thought some aspects of the congestion exposure metrics might have limited those findings. Below, we highlight the strengths and limitations of the study.
STUDY DESIGN, DATASETS, AND ANALYTICAL APPROACHES
The Committee noted that integrating analyses using research triangulation helped to enable a robust study design. The focus on Texas for this set of research questions also provided a compelling case study — the investigators reported that between 1996 and 2016, about 1.7 million pregnant individuals lived within 500 m of a Texas highway or expressway. Using a large sample of recorded births over 20 years allowed the investigators to examine the cumulative effects of regulations targeting tailpipe emissions over the last three decades. The Committee also appreciated the detailed measures of air pollution exposures over time, the construction of various novel congestion metrics, and the extensive linkages with contextual data of other environmental, sociodemographic, and neighborhood characteristics.
One noted limitation was that both NO2 concentrations and congestion exposures comprise annual estimates that are assigned based on the year of the infant’s birth; that approach limits the definition of the exposure window depending on the timing of pregnancy. Several studies have shown that the susceptible windows of exposure associated with adverse birth outcomes, such as preterm birth and low birth weight, can vary by trimester and during the period before pregnancy.33–35 This limitation would only be very important if there were strong fluctuations in traffic volumes by time of year such that the annual estimate was not a strong reflection of the key windows of susceptibility. The investigators noted the potential for exposure misclassification as a limitation of their study, and the Committee recommends that future studies explore if and how various TRAP exposures, including congestion metrics, vary throughout different windows of susceptibility during pregnancy.
The investigators also constructed a spatial-temporal database of tolling project implementation and roadway improvement construction projects across the state, a unique and innovative approach to evaluating potential reduced TRAP exposures in those locations. However, the Committee noted that the spatial scale of each type of project likely differs. These differences might result in a heterogeneous mix of projects in their exposure categories with different influences on congestion. Indeed, the analysis of roadway improvement projects was restricted to projects with a total cost of over $5 million because those were considered the most likely to influence traffic volume and congestion levels. The temporal scale of each project type also likely differs in terms of implementation and construction timing, and these two sets of actions spanned different periods (1996 to 2016 for tolling implementation vs. 2007 to 2016 for roadway improvement projects). The influence of these differing spatial and temporal scales on the analyses is an open question, and the Committee wondered if these differences in scales might have contributed to the lack of an effect in some of the analyses.
The Committee also appreciated the choice of adverse birth outcomes as a health endpoint that could be consistently tracked over time and has important short-term and long-term implications. Nonetheless, they noted an important debate within the field of perinatal epidemiology on the use of term birth weight in analyses of adverse birth outcomes given the potential for collider bias (see Box 2).36,37 The investigators recognized the complex etiology of birth outcomes and explained in their report that the focus on term birth weight in their analyses allowed them to distinguish between fetal growth and gestational duration. They also reported additional sensitivity analyses that examine the effect of restricting the study population to term births and adjusting for gestational duration on the results reported for analyses of birth weight.
Box 2: Conditioning on Intermediates in Analyses of Birth Outcomes.
It has been recognized that conditioning on intermediate parameters in analyses of the potential effects of environmental exposures on adverse birth outcomes has the potential to introduce bias.36,37 Here, the evaluation of birth weight among infants born at term or “term birth weight” necessarily restricts, or conditions, the analysis of the timing of birth (i.e., on births whose gestational length exceeds 37 weeks). This type of bias is known as collider bias.
Collider bias can occur if the timing of birth is itself affected by TRAP, and there are unmeasured common causes that affect both the timing of birth and birth weight. In this situation, investigating associations only among term births (a form of conditioning on term births) can artificially induce a correlation between TRAP and birth weight that can bias estimates in either positive or negative directions.
This issue gained traction when Hernandez-Diaz et al.38 illustrated that the use of birth weight as an intermediate for the effect of smoking on infant mortality introduced collider bias. Notably, it produced the contradictory result that smoking was associated with lower infant mortality rates among infants who were born with low birth weight. Others have shown how the potential for collider bias occurs specifically for birth weight as the outcome of interest.39 The potential for collider bias has been shown to occur when conditioning or adjusting for the intermediate, regardless of the statistical adjustment method (e.g., regression adjustment or stratification).37,39
To address this issue, it is important to conduct various sensitivity analyses and include a clear statement of the specific question of interest, a discussion of the biological plausibility underlying the relationship of interest, and a discussion of the potential biases that might occur for epidemiological investigations using this outcome.36,37 For example, choosing to condition on an intermediate in the relationship between an exposure and an outcome results in an assessment of the direct effect of the exposure rather than the total effect, which should be clearly stated if this is the question of interest.36 Here, Hystad and colleagues stated their interest in distinguishing between the pathways of intrauterine growth restriction and triggers for premature labor by examining birth outcomes related to (1) preterm birth and (2) birth weight for infants born at term. They also conducted sensitivity analyses to explore the potential for collider bias in their analyses that could occur by conditioning on gestational age as an intermediate (either through restriction or adjustment).
The Committee appreciated these additional analyses, which showed similar effect estimates for TRAP exposures and birth weight among term births (analysis restricted to births 37–42 weeks gestational length), all births (no restriction) with an adjustment for gestational duration (inclusion of a covariate for length of gestation in weeks), and all births with no covariate adjustment for gestational duration. The investigators appropriately acknowledged that, while collider bias remains a possibility when estimating influences of air pollution on birth weight when restricting the study population to term births, in this study, the bias is small, with only minimal effects on their main results.
FINDINGS AND INTERPRETATION
The Committee generally agreed with the interpretation of the findings presented in this study. The investigators concluded that the cumulative effects of regulatory improvements over the long term were likely contributors to improvements in TRAP exposures that were linked to improvements in adverse birth outcomes, even though adverse associations between TRAP exposures and birth outcomes were evident even in the most recent study years (i.e., 2016). The investigators appropriately recognized that their results cannot be attributed to any specific regulatory action, a common challenge in accountability studies. A previous HEI-funded study similarly provided some evidence of long-term cumulative regulations in California paralleling decreases in concentrations of several outdoor pollutants (PM2.5 and NO2) and associated improvements in children’s health.21
In their analysis of tolling implementation and roadway improvements, the investigators observed only minimal congestion reductions after tolling implementation. They also observed increases in congestion during construction years for roadway improvement projects accompanied by decreases in congestion and NO2 concentrations in the post-construction years. Overall, there was little evidence for associations between implementing congestion-reduction actions and improvement in adverse birth outcomes.
For the analyses of types of tolling implementation, the Committee thought that the focus on major roadways likely had implications for whether congestion truly changed and whether traffic was rerouted to alternate nearby roadways. Supplemental analyses showed no significant rerouting, but these analyses also focused on intersections with major roadways and did not capture more local roads. If tolling on major roadways simply shifted traffic into the local communities, it would not be surprising that there were no improvements in birth outcomes with these actions. Nonetheless, the Committee still thought this work was an important addition to the literature, as only one other study has similarly studied the effect of tolling implementation on birth outcomes. That work focused on implementing E-ZPass, a multistate electronic tolling program, in the eastern United States.40 More work is needed to draw definitive conclusions on the effects of this specific type of local congestion-reduction action on birth outcomes.
The Committee thought this work added nicely to other studies that have examined actions such as the effect of congestion pricing41,42 or low-emission zones.43,44 These studies were broader than the congestion reduction actions considered in this work, which were largely specific to infrastructure improvements. Although the evidence for the effectiveness of congestion reduction actions in improving public health is somewhat mixed, future work could focus on different potential infrastructure-related actions targeting congestion than were considered here. For example, one could evaluate specific actions such as optimization of traffic signals (i.e., implementing ITS control strategies) and other operational improvement strategies or examine projects that can be linked to a specific congestion reduction program, such as those implemented through the Federal Highway Administration’s Congestion Mitigation and Air Quality (CMAQ) Improvement Program.
More broadly, the Committee agreed with the investigators that the results of this study provide some evidence that the cumulative effect of regulatory improvements aimed at reducing vehicle tailpipe emissions that improve air quality over the long term was more meaningful for improving adverse birth outcomes than local congestion reduction actions that could change air quality over shorter periods. The Committee also noted that the investigators were suitably cautious in their conclusions.
This study’s results contributed to the mixed evidence base regarding congestion reduction actions and highlighted the need for further evaluation of, for example, infrastructure improvement projects, in particular. Although a recent review of low-emission zones and congestion charging schemes illustrated a positive effect on reducing adverse cardiovascular outcomes, the evidence for improved birth outcomes remained unclear.45 A different review of the effect of built environment characteristics on adverse birth outcomes found evidence for a small to moderate association between roadway proximity or traffic density and adverse birth outcomes.46 Overall, mixed evidence remains for the association between TRAP exposures and adverse birth outcomes (e.g., as reported in HEI 2022).1
Issues common to accountability studies, such as imprecise exposure assessment and the potential for residual confounding, remain in this work. However, the Committee thought that using multiple statistical approaches and data sources, as well as using DiD methods, strengthened the body of work presented. The investigators noted that they were not able to assess the “chain of accountability” (see Preface figure) in all aspects of this study but instead highlighted their work as an example of moving toward a “web of accountability” that captures factors that might be adjacent and related to various links in the chain from policy to health response.27 The Committee noted that future accountability studies could focus on the effects of community-driven actions and explore recent advances in data availability and measurement that can help investigators conduct robust accountability studies in the future.
SUMMARY AND CONCLUSIONS
Hystad and colleagues evaluated changes in adverse birth outcomes from 1996 to 2016 in Texas associated with both national-scale cumulative regulatory improvements aimed at decreasing vehicle tailpipe emissions and local congestion-reduction actions. They found that TRAP exposures were consistently associated with adverse birth outcomes throughout the 20 years. During this timeframe, pregnant individuals’ exposures to annual NO2 concentrations decreased markedly, with smaller decreases in VMT on nearby roadways. Reductions in VMT near the mothers’ residences were also associated with decreased risks of adverse birth outcomes, consistent with their hypothesis. These findings suggest that the totality of regulatory improvements implemented during this time had some success. Analyses that evaluated the effect of tolling implementation and roadway capacity improvement projects resulted in minimal changes in congestion and showed little evidence of any improvement in adverse birth outcomes, contrary to their hypothesis for this research aim.
In its independent evaluation of the study, the Review Committee commented on the strength of using a research triangulation approach to answer the study’s research questions. They also appreciated the large study population that was followed over a long period, the use of novel indicators and analyses of congestion, and the extensive supplementary data on contextual factors. However, they noted the substantial debate surrounding one of the birth outcomes used (term birth weight; see Box 2). This issue was mitigated by robust findings in secondary analyses investigating all births.
The Committee found the reduction in the size of the associations between VMT on nearby roadways and birth outcomes over two decades very interesting and agreed with the investigators’ interpretation that long-term regulations likely had some success in improving health. The Committee thought that the lack of evidence in the analyses of local actions aimed at reducing congestion was a feature of both the actions examined (i.e., tolling implementation and roadway improvement projects) and the spatial and temporal scale of the associated congestion metrics.
Ultimately, the Committee agreed with the investigators’ overall conclusions that this study provides some evidence that reducing the vehicle fleet’s tailpipe emissions was more effective at improving air quality and decreasing adverse birth outcomes than local actions aimed at reducing congestion. This study provides an important contribution to accountability research in the strengths associated with the study design and its comparison of long-term national-scale regulations versus shorter-term local actions. Separating the different effects of individual regulations on TRAP exposures and health remains a challenge. Future work could benefit from improvements in data and exposure measurements and examination of other infrastructure improvement actions.
ACKNOWLEDGMENTS
The HEI Review Committee thanks the ad hoc reviewers for their help in evaluating the scientific merit of the Investigators’ Report. The Committee is also grateful to Hanna Boogaard for her oversight of the study, to Yasmin Romitti for assistance with reviewing the report and preparing its Commentary, to Anne Connor for editing this Report and its Commentary, and to Kristin Eckles and Hope Green for their roles in preparing this Research Report for publication.
Footnotes
* A list of abbreviations and other terms appears at the end of this volume.
REFERENCES
1.HEI Panel on the Health Effects of Long-Term Exposure to Traffic-Related Air Pollution. 2022. Systematic Review and Meta-analysis of Selected Health Effects of Long-Term Exposure to Traffic-Related Air Pollution. Special Report 23. Boston, MA: Health Effects Institute. [Google Scholar]
2.Frey HC. 2018. Trends in onroad transportation energy and emissions. J Air Waste Manag Assoc
68:514–563, https://doi.org/10.1080/10962247.2018.1454357.
[DOI] [PubMed] [Google Scholar]
4.Kim SY, Bechle M, Hankey S, Sheppard L, Szpiro AA, Marshall JD. 2020. Concentrations of criteria pollutants in the contiguous US, 1979–2015: role of prediction model parsimony in integrated empirical geographic regression. PLoS ONE
15:e0228535, https://doi.org/10.1371/journal.pone.0228535.
[DOI] [PMC free article] [PubMed] [Google Scholar]
6.Clark LP, Millet DB, Marshall JD. 2017. Changes in transportation-related air pollution exposures by race-ethnicity and socioeconomic status: outdoor nitrogen dioxide in the United States in 2000 and 2010. Environ Health Perspect
125:097012, https://doi.org/10.1289/ehp959.
[DOI] [PMC free article] [PubMed] [Google Scholar]
7.Rosofsky A, Levy JI, Zanobetti A, Janulewicz P, Fabian MP. 2018. Temporal trends in air pollution exposure inequality in Massachusetts. Environ Res
161:76–86, https://doi.org/10.1016/j.envres.2017.10.028.
[DOI] [PMC free article] [PubMed] [Google Scholar]
8.Liu J, Clark LP, Bechle MJ, Hajat A, Kim SY, Robinson AL, et al. 2021. Disparities in air pollution exposure in the United States by race/ethnicity and income, 1990–2010. Environ Health Perspect
129:127005, https://doi.org/10.1289/ehp8584.
[DOI] [PMC free article] [PubMed] [Google Scholar]
9.Kerr GH, Goldberg DL, Harris MH, Henderson BH, Hystad P, Roy A, et al. 2023. Ethnoracial disparities in nitrogen dioxide pollution in the United States: comparing data sets from satellites, models, and monitors. Environ Sci Technol
57:19532–19544, https://doi.org/10.1021/acs.est.3c03999.
[DOI] [PMC free article] [PubMed] [Google Scholar]
10.Hajat A, Hsia C, O’Neill MS. 2015. Socioeconomic disparities and air pollution exposure: a global review. Curr Environ Health Rep
2:440-450, https://doi.org/10.1007/s40572-015-0069-5.
[DOI] [PMC free article] [PubMed] [Google Scholar]
11.IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. 2014. Diesel and gasoline engine exhausts and some nitroarenes: IARC monographs on the evaluation of carcinogenic risks to humans (volume 105). Geneva, Switzerland: World Health Organization. [PMC free article] [PubMed] [Google Scholar]
12.US EPA (US Environmental Protection Agency). 2016. Integrated Science Assessment for Oxides of Nitrogen–Health Criteria. EPA/600/R-15/068. Research Triangle Park, NC: US EPA. [Google Scholar]
14.Klepac P, Locatelli I, Korošec S, Künzli N, Kukec A. 2018. Ambient air pollution and pregnancy outcomes: a comprehensive review and identification of environmental public health challenges. Environ Res
167:144–59, https://doi.org/10.1016/j.envres.2018.07.008.
[DOI] [PubMed] [Google Scholar]
15.National Toxicology Program. 2019. NTP Monograph on the Systematic Review of Traffic-related Air Pollution and Hypertensive Disorders of Pregnancy. Report No.: NTP Monograph 7. Research Triangle Park, NC: National Institute of Environmental Health Sciences, National Institutes of Health. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Fussell JC, Jauniaux E, Smith RB, Burton GJ. 2024. Ambient air pollution and adverse birth outcomes: a review of underlying mechanisms. BJOG
131:538–550, https://doi.org/10.1111/1471-0528.17727.
[DOI] [PMC free article] [PubMed] [Google Scholar]
17.Wang L, Guo P, Tong H, Wang A, Chang Y, Guo X, et al. 2020. Traffic-related metrics and adverse birth outcomes: a systematic review and meta-analysis. Environ Res
188:109752, https://doi.org/10.1016/j.envres.2020.109752.
[DOI] [PubMed] [Google Scholar]
21.Boogaard H, van Erp AM, Walker KD, Shaikh R. 2017. Accountability studies on air pollution and health: the HEI experience. Curr Environ Health Rep
4:514-22, https://doi.org/10.1007/s40572-017-0161-0.
[DOI] [PubMed] [Google Scholar]
22.Gilliland F, Avol E, McConnell R, Berhane K, Gauderman WJ, Lurmann FW, et al. 2017. The Effects of Policy-Driven Air Quality Improvements on Children’s Respiratory Health. Research Report 190. Boston, MA: Health Effects Institute. [PMC free article] [PubMed] [Google Scholar]
23.Russell AG, Tolbert P, Henneman LRF, Abrams J, Liu C, Klein M, et al. 2018. Impacts of Regulations on Air Quality and Emergency Department Visits in the Atlanta Metropolitan Area, 1999–2013. Research Report 195. Boston, MA: Health Effects Institute. [PMC free article] [PubMed] [Google Scholar]
24.Rich DQ. 2017. Accountability studies of air pollution and health effects: lessons learned and recommendations for future natural experiment opportunities. Environ Int
100:62–78, https://doi.org/10.1016/j.envint.2016.12.019.
[DOI] [PMC free article] [PubMed] [Google Scholar]
25.Burns J, Boogaard H, Polus S, Pfadenhauer LM, Rohwer AC, Van Erp AM, et al. 2020. Interventions to reduce ambient air pollution and their effects on health: an abridged Cochrane systematic review. Environ Int
135:105400, https://doi.org/10.1016/j.envint.2019.105400.
[DOI] [PubMed] [Google Scholar]
26.Henneman LR, Liu C, Mulholland JA, Russell AG. 2017. Evaluating the effectiveness of air quality regulations: a review of accountability studies and frameworks. J Air Waste Manag Assoc
67:144-172, https://doi.org/10.1080/10962247.2016.1242518.
[DOI] [PubMed] [Google Scholar]
27.Ebelt S, Baxter L, Erickson HS, Henneman LR, Lange S, Luben TJ, et al. 2023. Air pollution accountability research: moving from a chain to a web. Global Epidemiology
6:100128, https://doi.org/10.1016/j.gloepi.2023.100128.
[DOI] [PMC free article] [PubMed] [Google Scholar]
28.Lawlor DA, Tilling K, Davey Smith G. 2016. Triangulation in aetiological epidemiology. Int J Epidemiol
45:1866–1886, https://doi.org/10.1093/ije/dyw314.
[DOI] [PMC free article] [PubMed] [Google Scholar]
29.Pearce N, Vandenbroucke JP, Lawlor DA. 2019. Causal inference in environmental epidemiology: old and new approaches. Epidemiology
30:311–316, https://doi.org/10.1097%2FEDE.0000000000000987.
[DOI] [PMC free article] [PubMed] [Google Scholar]
30.Suhag A., Berghella V. 2013. Intrauterine growth restriction (IUGR): etiology and diagnosis. Curr Obstet Gynecol Rep
2:102–111, https://doi.org/10.1007/s13669-013-0041-z. [Google Scholar]
32.Wilcox AJ, Cortese M, McConnaughey DR, Moster D, Basso O. 2021. The limits of small-for-gestational-age as a high-risk category. Eur J Epidemiol
36:985–991, https://doi.org/10.1007/s10654-021-00810-z.
[DOI] [PMC free article] [PubMed] [Google Scholar]
33.Li C, Yang M, Zhu Z, Sun S, Zhang Q, Cao J, et al. 2020. Maternal exposure to air pollution and the risk of low birth weight: a meta-analysis of cohort studies. Environ Res
190:109970, https://doi.org/10.1016/j.envres.2020.109970.
[DOI] [PubMed] [Google Scholar]
34.Chen J, Fang J, Zhang Y, Xu Z, Byun HM, Li PH, et al. 2021. Associations of adverse pregnancy outcomes with high ambient air pollution exposure: results from the Project ELEFANT. Sci Total Environ
761:143218, https://doi.org/10.1016/j.scitotenv.2020.143218.
[DOI] [PubMed] [Google Scholar]
35.Ju L, Li C, Yang M, Sun S, Zhang Q, Cao J, et al. 2021. Maternal air pollution exposure increases the risk of preterm birth: evidence from the meta-analysis of cohort studies. Environ Res
202:111654, https://doi.org/10.1016/j.envres.2021.111654.
[DOI] [PubMed] [Google Scholar]
39.Delbaere I, Vansteelandt S, De Bacquer D, Verstraelen H, Gerris J, De Sutter P, et al. 2007. Should we adjust for gestational age when analysing birth weights? The use of z-scores revisited. Hum Reprod
22:2080-2083, https://doi.org/10.1093/humrep/dem151.
[DOI] [PubMed] [Google Scholar]
40.Currie J, Walker R. 2011. Traffic congestion and infant health: evidence from E-ZPass. Am Econ J Appl Econ
3:65–90, https://doi.org/10.1257/app.3.1.65. [Google Scholar]
41.Kelly F, Anderson HR, Armstrong B, Atkinson R, Barratt B, Beevers S, et al. 2011. The Impact of the Congestion Charging Scheme on Air Quality in London. Research Report 155. Boston, MA: Health Effects Institute. [PubMed] [Google Scholar]
42.Simeonova E, Currie J, Nilsson P, Walker RC. 2021. Congestion pricing, air pollution, and children’s health. J Hum Res
56:971–996, https://doi.org/10.3368/jhr.56.4.0218-9363R2. [Google Scholar]
43.Kelly F, Armstrong B, Atkinson R, Anderson HR, Barratt B, Beevers S, et al. 2011. The London Low Emission Zone Baseline Study. Research Report 163. Boston, MA: Health Effects Institute. [PubMed] [Google Scholar]
44.Cesaroni G, Boogaard H, Jonkers S, Porta D, Badaloni C, Cattani G, et al. 2012. Health benefits of traffic-related air pollution reduction in different socioeconomic groups: the effect of low-emission zoning in Rome. Occup Environ Med
69:133–139, https://doi.org/10.1136/oem.2010.063750.
[DOI] [PubMed] [Google Scholar]
45.Chamberlain RC, Fecht D, Davies B, Laverty AA. 2023. Health effects of low emission and congestion charging zones: a systematic review. Lancet Public Health
8:e559-574, https://doi.org/10.1016/s2468-2667(23)00120-2.
[DOI] [PubMed] [Google Scholar]
46.Woods N, Gilliland J, Seabrook JA. 2017. The influence of the built environment on adverse birth outcomes. J Neonatal-Perinatal Med
10:233–248, https://doi.org/10.3233/npm-16112.
[DOI] [PubMed] [Google Scholar]
, to inform estimation of intercept terms. Given the set of individual-level confounders (same as in the global model) and associated regression coefficients, Vηk, our model was:
= number of hours maternal residence i was downwind of high-traffic roads located within radial segment j during pregnancy
