Assessing the effectiveness of vehicle emission regulations on improving perinatal health: a population-based accountability study

Abstract

Background

Since the 1990s, extensive regulations to reduce traffic-related air pollution (TRAP) have been implemented, yet the effectiveness of these regulations has not been assessed with respect to improving infant health. In this study, we evaluate how infant health risks associated with maternal residences near highways during pregnancy have changed over time.

Methods

We created a population-based retrospective birth cohort with geocoded residential addresses in Texan metropolitan areas from 1996 through 2009 (n = 2 259 411). We compared term birthweight (37–42 weeks of gestation) among maternal residences <300 m from a highway (high TRAP exposure) (n = 394 346) and 500–3500 m from a highway (comparison group) (n = 1 865 065). We implemented linear regressions to evaluate interactions between high TRAP exposure and birth year, adjusting for demographics, socioeconomic status and neighbourhood context. In addition, we used propensity score matching to further reduce residual confounding.

Results

From 1996 to 2009, outdoor NO₂ decreased by 51.3%, based on regulatory monitoring data in Texas. Among pregnant women who resided in the high TRAP zone during pregnancy, interaction terms between residential location and birth year show that birthweight increased by 1.1 g [95% confidence interval CI): 0.7, 1.5) in unadjusted models and 0.3 g (95% CI: 0.0, 0.6) in matched models. Time-stratified models also show decreasing impacts of living in high TRAP areas on birthweight when comparing infants born in 1996–97 with 2008–09. Sensitivity analyses with alternative exposure and control groups show consistent results.

Conclusions

Infant health risks associated with maternal residence near highways have reduced over time, paralleling regulatory measures to improve exhaust pipe emissions.

Key Messages

• Outdoor NO₂ decreased by 51.3% between 1996 and 2009, based on regulatory monitoring data in Texas.

• Term birthweight among infants exposed to living within 300 m of a highway [high traffic-related air pollution (TRAP) areas] has increased by 0.3–1.1 g per year.

• Time-stratified models demonstrate the decreasing impacts of living within 300 m of a highway (high TRAP areas) on birthweight when comparing infants born in 1996–97 with 2008–09.

• Evidence suggests that the impact of maternal exposure to high TRAP during pregnancy (as measured by residential proximity to highways) has a decreasing association with term birthweight over time.

Background

In North American cities, 38–45% of residents live within 500 m of a major road.¹ Thus, a large segment of the population is at risk for adverse health outcomes associated with exposure to traffic-related air pollution (TRAP).^1,2 TRAP is a heterogeneous mix of air pollutants including carbon monoxide (CO), carbon dioxide (CO₂), particulate matter (PM) and nitrogen oxides (NO_x),^3–9 many of which are associated with adverse environmental¹⁰ and human health outcomes.^2,11,12 Over the past two decades, federal regulations have been incrementally implemented to reduce TRAP via fuel economy goals, emissions standards and mobile source air toxic regulations (Figure 1). These traffic policies cost an estimated 2.3 billion dollars to implement between 1990 and 2010.¹³ Recent evidence suggests that the near road TRAP environment is less polluted by conventional metrics, such as nitrogen dioxides (NO₂), than shown in previous studies.^14,15 As regulations have been implemented, the primary methods used to assess their effectiveness have been ongoing air pollution monitoring, environmental damage and economic impact studies.¹³ However, one of the Environmental Protection Agency’s (EPA) official goals is to protect local populations from environmental exposures,¹⁶ so the effectiveness of TRAP regulations should be assessed with respect to human health.

Figure 1.

Timeline of key environmental protection agency programmes and policies to reduce traffic-related air pollution between 1996 and 2010. SUVs, sport utility vehicles; PM, particulate matter

Infant health outcomes, such as birthweight, may be particularly sensitive to TRAP and therefore represent an important population that may be positively affected by air pollution regulations. Lower birthweight represents immediate health consequences (e.g. early onset neonatal sepsis, respiratory distress syndrome) and a long-term risk factor for chronic conditions (e.g. cardiovascular disease, developmental delays, metabolic syndrome).^17,18 Many cases of lower birthweights are attributed to preterm birth (<37 weeks of gestation), but some mothers with otherwise healthy pregnancies may give birth to smaller infants at full term (37–42 weeks of gestation) due to conditions related to intrauterine growth restriction.^19,20 Costs for conditions related to lower birthweight total $5.8 billion per year, thus placing a considerable burden on the medical system.²¹

TRAP has been associated with adverse infant health outcomes including decreased birthweight, but the magnitude of associations substantially varies across studies.^8,22–31 A meta-analysis (which covers studies published before 2010) demonstrates that a 20 ppb increase in NO₂ exposure, a marker for TRAP mixtures, during pregnancy is associated with a 28.1 g (95% CI: −44.8, −11.5) reduction in infant birthweight.⁸ However, this review shows that the magnitude of association varies across studies, ranging from –81.0 g (95% CI: −102.8, −59.2) to 102.9 g (95% CI: −70.0, 275.7) per 20 ppb increase in NO₂ exposure.^8,28,32 One potential reason for heterogeneity among cohorts is that incremental improvements in TRAP driven by federal regulatory actions may be reducing the infant health risks among mothers who reside near roadways. An updated systematic review shows similarly mixed results.³³ To date, no studies have examined associations between TRAP and adverse infant health outcomes with respect to variation over time periods corresponding to regulatory measures to reduce TRAP.

Using a diverse population-based vital statistics birth cohort between 1996 and 2009 in Texas, this study assesses changes in infant health risk associated with regulatory improvements in motor vehicle emissions (and resulting TRAP). We test our working hypothesis that the series of emission regulations and vehicle efficiency policies implemented between 1996 and 2009 will yield improved infant health outcomes among mothers who reside within 300 m of a highway (a high TRAP exposure zone) during pregnancy. That is, regulations on motor vehicle emissions have reduced air pollution exposures for pregnant women and their infants who reside in high TRAP exposure zones between 1996 and 2009. This analysis provides insights into the magnitude of improvement to infant health provided by regulations to reduce TRAP from on-road motor vehicles.

Methods

Study population

We obtained geocoded maternal address data at delivery for all births in Texas from 1 January 1996 through 31 December 2009 through the Texas Vital Statistics programme. Since tetraethyl leaded petrol for on-road vehicles was banned on 1 January 1996, our study encompasses smaller TRAP-related regulatory changes (Figure 1). We restricted this analysis to singleton births to reduce confounding from risk factors associated with multiple births, and excluded implausible observations based on maternal ages (<10 and >65 years old) and birthweights (<500 and >5000 g). Finally, we limited our sample to mothers living in metropolitan areas with >10 000 term births within 1500 m of a primary or secondary road to ensure our analysis contains sufficient sample size in all subgroups (exposure details described below). This research has been approved by the Texas State Department of Health and Human Services (#15-063) and the Oregon State University Institutional Review Board (#6692).

Traffic-related air pollution exposure metric

We examined trends in mean daily maximum 1-h outdoor NO₂ concentrations from EPA air monitors across Texas to verify our baseline assumption that concentrations have decreased between 1996 and 2009 in our study region. NO₂ is widely considered to be a reasonable surrogate for TRAP mixtures, but there are other components of TRAP (e.g. diesel particular matter, benzene) that are often highly correlated with NO₂^2,34 and regional sources (e.g. power plants) are also responsible for some NO_2.³⁵ We anticipate that other components of TRAP will be captured in our residential proximity metric (described below), and regional sources of NO₂ should not differentially affect our study population.

We derived exposure data on TRAP via residential proximity to a highway, which is a proxy for where TRAP exposures should be the highest. For this proximity metric, we obtained data on primary and secondary road locations (e.g. major highways, expressways and interstates) from the 2010 US Census TIGER file.³⁶ Generally, these roads are distinguished by multiple lanes of traffic in each direction with interchanges or at-grade intersections.³⁶ There were no large-scale highway infrastructure projects in Texas during our study period,³⁷ so using a single time point road network should not substantially affect exposure assessment. We define high TRAP exposure as maternal residence within 300 m of a highway, which aligns with existing literature indicating that TRAP largely dissipates beyond this distance.^2,38,39 To ensure that our comparison group is likely unexposed to the same high levels of TRAP but similar in other sociodemographic characteristics, we defined the primary comparison population as maternal residence from within 500 to 3500 m of a highway and exclude births within 300 to 500 m of a highway to reduce potential exposure misclassification.² This metric provides a fixed exposure metric over time that is easily used by community groups to translate into relevant policy decisions.

Neighbourhood exposure metrics

We linked maternal residences to census tract sociodemographic information to establish neighbourhood context.^40,41 Births before 2005 were assigned 2000 census information and births in 2005 and after were assigned 2010 census information. Since highways are more likely to run through neighbourhoods with certain characteristics, neighbourhood context is important information to incorporate into our regression models.

Term birthweight metric

Our metric to assess infant health was term birthweight (37–42 weeks of gestation). We accessed detailed information on maternal sociodemographic and clinical characteristics to use in our models. The birth certificate form is different for the 1996–2004 birth years compared with the 2005–09 birth years, so we harmonized important covariates, including maternal education and smoking status. Missing categorical covariates were coded as a separate category as appropriate and missing continuous covariates were removed from our sample. We assessed term birthweight as a continuous outcome.

Statistical analysis

We examined the changes in NO₂ using EPA air monitor data in Texas from 1996 through 2009 to verify that TRAP was decreasing in our study period. To examine changes, we graphically depicted the annual mean daily 1-h maximum NO₂ concentrations over this period, and examine the mean term birthweight by distance from a highway for the 1996–97 births compared with the 2008–09 births. Further graphs show distance gradients of key maternal characteristics (e.g. race and ethnicity, educational attainment) for the 1996–97 births compared with the 2008–09 births.

Our first models examine the overall associations between our exposure metrics and term birthweight across our study period. We assessed proximity as a binary variable for maternal residence within 300 m versus 500–3500 m of a highway. For each model specification, we included an interaction term between birth year (continuous variable for year since 1996) and road exposure (indicator variable for residence less than 300 m from a major highway), which allowed us to ascertain the impact of temporal changes in TRAP exposure on our sample. Covariates were selected a priori via literature review and directed acyclic graphs (Supplementary Figure S1, available as Supplementary data at IJE online). We examined four distinct models: Model 1, unadjusted; Model 2, adjusted for infant sex (male, female), maternal age (continuous), maternal race and ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, other), maternal educational attainment (less than high school, high school graduate, some college education, bachelor’s degree, postgraduate education, missing), nulliparous (yes, no), prenatal care received (yes, no), maternal weight gain during pregnancy (continuous), diabetes (yes, no), gestational hypertension (yes, no) and smoking during pregnancy (yes, no); Model 3, adjusted with the same covariates as Model 2 with neighbourhood median household income (continuous), neighbourhood percentage unemployment (continuous) and neighbourhood percentage White population (continuous); and Model 4, propensity score-matched sample within birth year (described below) using the same covariates as Model 3 with the addition of birth city (categorical—i.e. Austin, Houston) and frequency weighting (details of matching provided below). We ran subsequent model sets where we changed the high TRAP zone to be 100 m, 200 m and 500 m from a highway, respectively, to better understand the sensitivity of our primary results. In addition, we assessed the robustness of our results with alternative comparison groups at 500–1500 m, 1500–2500 m and 2500–3500 m, to ensure that distant air pollution exposures and changing neighbourhood compositions were not impacting on our results. We also tested our models with and without the gestational hypertension and diabetes covariates to understand if these pathways were mediating our associations.

Subsequent models quantified associations between maternal exposure to TRAP during pregnancy and term birthweight, in stratified 2-year increments (e.g. 1996–97, 1998–99). This method allowed us to examine temporal policy trends of infant birthweight in high TRAP zones, without external factors that might change term birthweight over time. Smaller time increments yielded lower power for detecting an association in our stratified populations, so we leveraged the 2-year temporal groups across all models. For each set of time-stratified models, we ran all four iterations of the regression adjustments that we used for the overall associations.

Since previous literature establishes that populations living in close proximity to highways are generally different in sociodemographic characteristics,^1,42,43 we implemented stratified propensity score-matching using the same birth year categories as our time-stratified models (e.g. 1996–97, 1998–99). Propensity score-matching creates statistically comparable populations that differ only on exposure status. Interpretation of the matched sample requires that two assumptions be met: (i) no unmeasured confounding in our analysis; and (ii) no spillover effects between the exposed group and the comparison group. By adding stratification by birth year to our matching schema, we accounted for potential changes over time in the population characteristics for residents near major highways. Our matching scheme included matching on birth city, maternal age, maternal race and ethnicity, maternal education, maternal weight gain during pregnancy and neighbourhood household income, based on a priori knowledge of how communities living near highways likely differ from communities further away and on differences between our exposed and comparison groups in the data. We employed a 1:3 nearest-neighbour matching scheme with replacement (e.g. each exposed mother is matched to three comparison mothers and a comparison mother may match to multiple exposed mothers). Within the matched sample, we ran identical linear regressions to our Model 3 with the addition of birth city. We provided: robustness checks for our matches through a table of covariate distribution before and after matching; a figure of the propensity score distribution before and after matching; and a figure of how risk estimates changed based on what matching scheme we selected (e.g. 1:2 matching).

Results

Study population and descriptive statistics

Our population of geocoded maternal addresses contains 4 567 714 births in Texas between 1 January 1996 and 31 December 2009. After excluding births on account of clinical characteristics, residential locations, and missing covariates (Supplementary Figure S2, available as Supplementary data at IJE online), our sample contains 2 259 411 singleton term births. Within this sample, there were 394 346 maternal addresses within 300 m of a highway (high TRAP exposure group) and 1 865 065 maternal addresses 500–3500 m from a highway (comparison group) (Table 1). Term birthweights were 20 g smaller on average among infants in the high TRAP group compared with the comparison group. Since the Texas population grew over our study period, the number of births per year increased annually, but proportions of births in each group remained similar. When comparing the high- and low-TRAP-exposed infants, we observed that mothers vary on characteristics including race and ethnicity, educational attainment and neighbourhood median household income. We also noted some variation in the proportion of mothers residing in high-TRAP zones versus low-TRAP zones by birth city (Supplementary Table S1, available as Supplementary data at IJE online).

Table 1.

Sociodemographic characteristics of term births (37–42 weeks of gestation) in sample

Characteristics	Residence 0–300 m of a highway	Residence 500–3500 m of a highway
Total births (n)	394 346	1 865 065
Birthweight (mean)	3356	3 376
Gestation length (mean weeks)	39.0	39.0
Birth year (n)
1996	20 903	108 299
1997	21 648	112 302
1998	21 608	113 864
1999	22 636	116 424
2000	24 044	122 807
2001	31 839	127 115
2002	33 624	135 335
2003	33 205	136 467
2004	31 322	126 293
2005	36 149	148 076
2006	29 703	155 733
2007	29 568	156 614
2008	29 362	153 708
2009	28 735	152 028
Female sex (%)	49.1	49.1
Nulliparous (%)	41.5	41.0
Maternal age (mean)	25.4	26.7
Maternal weight gain during pregnancy (kg.)	13.8	13.7
Maternal education (%)
Less than high school	37.1	29.5
High school graduate	31.1	28.1
Some college education	18.1	20.2
Bachelor’s degree	8.2	13.4
Postgraduate education	4.7	8.1
Maternal race and ethnicity (%)
White non-Hispanic	26.1	33.1
Black non-Hispanic	12.5	12.2
Hispanic	57.8	49.7
Other	3.7	5.0
Diabetes (%)	3.1	3.3
Gestational hypertension (%)	3.4	3.2
No prenatal care received (%)	3.9	2.9
Smoking during pregnancy (%)	6.4	5.2
Neighbourhood characteristics a
Median household income (US$)	37 886	46 233
Unemployment (%)	6.2	5.6
White population (%)	64.3	63.8

^aDerived from the United States Census Bureau for the years 2000 and 2010. Births before 2005 were attached to 2000 census information and births in 2005 and after were attached to 2010 census information

Outdoor NO₂ trends

When we examined the outdoor NO₂ monitors in Texas (Figure 2), we confirmed our assumption that mean daily 1-h maximum NO₂ concentrations have decreased between 1996 and 2009. In 1996, the mean daily maximum concentration was 34.1 ppb, but it was 16.6 ppb by 2009. Results indicated a 51.3% decrease in NO₂ over our study period.

Figure 2.

Mean daily maximum 1 hour NO2 concentrations (ppb) by year from EPA outdoor air monitors in Texas, 1996–2009; ppb, parts per billion. Nitrogen dioxide (NO₂) concentrations are from outdoor air monitors (n = 67) run by the Environmental Protection Agency (EPA) in Texas. This figure is provided only for illustrative purposes

Overall associations between high-TRAP exposure and term birth weight

When we examined the mean term birthweight by distance from a highway in 1996–97 compared with 2008–09 (Figure 3), we note that mean term birthweights increased as residential distance from a highway increased. Birthweights were smaller in the 2008–09 compared with the 1996–97 group. In addition, we confirmed our assumption that the slope of the line from 0–300 m in the 1996–97 group was steeper than the same slope for the 2008–09 group, but slopes for 500–3,500 m were similar for both groups. This figure provides preliminary confirmation that our selection of a 0-300m zone for high TRAP exposure and a 500–3500m zone for the comparison group was reasonable. Additional analysis demonstrated that population composition changes for maternal race, ethnicity and educational attainment were minimal over this time frame, but that the mothers residing in the high-exposure zones are substantially different in some characteristics compared with the mothers living in the low-exposure zones (Figure 4).

Figure 3.

Mean term birthweight by distance from a major highway in 1996–97 vs. 2008–09. TRAP, traffic-related air pollution; exc, excluded. Results are unadjusted mean term birthweights for maternal residences at each 50-m interval between 0 and 3500 m from a major highway, with a fractionated polynomial regression to create smoothed lines. For this analysis, births to mothers at between 0 m and 300-m are the high-TRAP exposure zone, and births to mothers at between 500 m and 3500 m are the comparison group, which is assumed to be a low-TRAP exposure zone

Figure 4.

Maternal sociodemographic characteristics by distance from a major highway in 1996–97 vs. 2008–09. Results are unadjusted proportions of each characteristic for maternal residences at each 50-m interval between 0 m and 3500 m from a major highway, with a fractionated polynomial regression to create smoothed lines. For this analysis, births to mothers at between 0 –m and –300 m are the high-TRAP exposure zone, and births to mothers at between 500 m and –3500 m are the comparison group, which is assumed to be a low-TRAP exposure zone

Unadjusted regressions (Model 1 demonstrated a −29.5 g (95% CI: −33.1, −26.0) decrease in term birthweight among infants whose mothers resided within 300 m of a highway at delivery compared with infants whose mothers resided between 500 m and 3500 m of a highway at delivery (Table 2). In Model 1, we found that the interaction between TRAP exposure status and birth year yields a 1.1-g (95% CI: 0.7, 1.5) increase in birthweight per year from 1996 to 2009 among infants born to mothers with residences in the high-TRAP exposure zone. The magnitude of this association was attenuated in adjusted models by individual (birth certificate) covariates (Model 2 and neighbourhood (Model 3) covariates. In Model 4 (matched), we demonstrated a −3.0 g (95% CI: −5.5, −0.5) decrease in term birthweight, and the interaction term between TRAP exposure status and birth year shows a 0.3-g (95% CI: 0.0, 0.6) increase in birthweight per year. Diagnostic tests for our propensity-score matching are presented in the Supplementary Table S2 and Supplementary Figures S2 and S3, available as Supplementary data at IJE online.

Table 2.

Associations between maternal residence within 300 m of a highway and term birthweight

Model specification	n	Road exposure(95% CI)	Road exposure*year(95% CI)
1. Unadjusted	2 259 411	−29.5 (−33.1, −26.0)	1.1 (0.7, 1.5)
2. Individual covariates	2 259 411	−9.6 (−13.0, −6.2)	0.6 (0.2, 1.0)
3. Neighbourhood covariates	2 259 411	−7.5 (−10.9, −4.1)	0.6 (0.3, 1.0)
4. Matched	2 366 028	−3.0 (−5.5, −0.5)	0.3 (0.0, 0.6)

For each model specification listed above, the reported coefficients are from regressions that include an interaction between road exposure (binary) and birth year. Road exposure is an indicator variable defined as maternal residence 0–300 m of a highway compared with maternal residences 500–3500 m of a highway. Model 1 is the unadjusted model that only includes birth year. Model 2 is the adjusted model for covariates on the birth certificate (infant sex, maternal age, maternal race and ethnicity, maternal educational attainment, prenatal care received, diabetes, gestational hypertension and smoking status). Model 3 includes the covariates listed in Model 2 with the addition of neighbourhood census tract variables (median household income, unemployment and percentage White population). Model 4 includes the covariates listed in Model 3 with the addition of a fixed effect for city and propensity-score matching at a 1:3 ratio with frequency weighting.

We also examined alternative high-TRAP groups at 100 m, 200 m and 500 m (Supplementary Table 3, available as Supplementary data at IJE online). Generally, we found associations similar to our primary specifications with evidence of an exposure-response relationship, where the magnitude of association between distance between a residence and a highway and term birthweight is larger for mothers who resided closer to the highways. For example, we observed a consistent decreasing exposure-response relationship in our matched models from 100 m (road exposure: −7.9, 95% CI: −12.6, −3.2; year*road exposure: 0.8, 95% CI: 0.3, 1.3) out to 500 m (road exposure: −3.2, 95% CI: −5.1, −1.3; year*road exposure: 0.3, 95% CI: 0.0, 0.5). When we examined alternative comparison groups for our 0–300-m exposure group, we found associations similar to our main analyses as well, except in our matched models (Model 4 with the 1500–2500-m and the 2500–3500-m comparison groups, where we lost statistical precision (Supplementary Table S4, available as Supplementary data at IJE online). We also tested our Model 2 covariates without gestational hypertension and diabetes, both of which are potential mediators of the association between TRAP and term birthweight, where we demonstrated little change in our model results (Supplementary Table S5, available as Supplementary data at IJE online).

Temporal trends in associations between TRAP exposure and term birth weight

Our subsequent analyses examined associations between TRAP exposure and term birthweight through regressions stratified in 2-year increments (Figure 5; Supplementary Table S6, available as Supplementary data at IJE online). Unadjusted regressions (Model 1) showed a decrease in term birthweight of −27.7 g (95% CI: −32.5, −23.0) in the 1996–97 group compared with a decrease in term birthweight of −16.9 g (95% CI: −20.9, −13.0) in the 2008–09 group. Adjusted regressions for birth certificate covariates (Model 2) and neighbourhood covariates (Model 3) substantially temper results (Figure 5; Supplementary Table S6, available as Supplementary data at IJE online). After implementing propensity-score matching on our sample (Model 4), results changed to a −0.8-g (95% CI: −4.1, 2.6) reduction in term birthweight in 1996–97 and a 0.0-g (95% CI: −2.8, 2.8) reduction in 2008–09 (Figure 5; Supplementary Table 6, available as Supplementary data at IJE online).

Figure 5.

Temporal associations between maternal residence within 300 m of a highway and term birthweight. For each model specification listed above, the temporal stratified models include all births from 2 consecutive years (e.g. 1996–97; 1998–99). Model 1 is the unadjusted model. Model 2 is the adjusted model for covariates on the birth certificate (infant sex, maternal age, maternal race and ethnicity, maternal educational attainment, prenatal care received, diabetes, and gestational hypertension). Model 3 includes the covariates listed in Model 2 with the addition of neighbourhood census tract variables (median household income, unemployment and percentage White population). Model 4 includes the covariates listed in Model 3 with the addition of a fixed effect for city and nearest neighbour propensity-score matching at a 1:3 ratio with frequency weighting

Discussion

We leveraged 15 years of geocoded birth certificate information on 2 259 411 infants to assess the impact of TRAP changes on term birthweight. Results indicate that the regulations to reduce TRAP have provided a direct benefit to infant birthweight for those mothers who lived near major roadways (<300 m). Our findings add to the body of literature linking TRAP to adverse birth outcomes, and more importantly show that the magnitude of this association has decreased over time, highlighting that regulatory efforts to reduce TRAP directly benefit infant health.

From 1996 to 2009, outdoor NO₂ decreased by 51.3%, based on regulatory monitoring data in Texas. Among pregnant women who resided in the high-TRAP zone during pregnancy, interaction terms between residential location and birth year show that birthweight increased by 1.1 g (95% CI: 0.7, 1.5) in unadjusted models and 0.3 g (95% CI: 0.0, 0.6) in matched models per year between 1996 and 2009. Time-stratified models also show decreasing impacts of living in high-TRAP areas on birthweight when comparing infants born in 1996–97 with 2008–09. These results were robust to different exposure measures, control groups and comprehensive adjustment using propensity score matching methods. To our knowledge, there are no other studies that examine the influence of TRAP changes over time on adverse birth outcomes.

Although we focus our analysis around NO₂ reductions, a traditional marker of TRAP, the exposure pathways that may be responsible for the infant health improvements may be any component (or components) of this mixture. For instance, recent literature discusses the complex mixture of pollutants that are found in exhaust pipe emissions in addition to NO₂, including carbon monoxide, black carbon, benzene and diesel particulate matter.^3–9 Many of these pollutants have been consistently associated with adverse birth outcomes in single and multipollutant TRAP models.^33,44 TRAP exposure also includes traffic noise, but the evidence thus far shows mixed results for associations between traffic noise and birth outcomes after accounting for TRAP.^23,45–49 Our present analysis does not differentiate between any air or noise pollution components; rather we examine whether the mixture is becoming less toxic for infant health over time.

Our study design fits into the chain of accountability to assess the population health impacts of air quality regulations proposed by the Health Effects Institute,⁵⁰ and contributes to the growing body of accountability literature that examines how complex air quality regulations impact on population health across the USA. To date, most of these studies focus on asthma and mortality,^51–56 and no accountability study has focused on TRAP regulations and birth outcomes. The Health Effects Institute accountability framework ensures that we are providing high-quality evidence on the associations between TRAP regulations and term birthweight. Key aspects of the chain of accountability for TRAP include: identifying regulatory changes; examining exhaust pipe emissions modifications; assessing ambient air quality changes; determining population exposures; and investigating the human health responses in the exposed populations.⁵⁰ We first describe sentinel regulatory changes at the federal level which are designed to reduce TRAP (Figure 1). Although we cannot test specific regulations, due to their overlapping implementation, we show that concentrations of ambient NO₂, a key marker for the TRAP mixture, have dropped by over 50% as these regulations were implemented (Figure 2). This provides some direct evidence of improvements to outdoor air quality and indirect evidence of changes to exhaust pipe emissions. Studies outside our analysis provide additional evidence that exhaust pipe emissions are improving due to TRAP regulations, and that the near-road environment contains less air pollution over time.^{11,14,15,43,57} We use an existing expert panel review² to determine that the highest TRAP exposure zones are within 300 m of a highway, where we hypothesized that the largest magnitudes of changes in birth outcomes are among pregnant women residing these areas. Finally, our study results align with existing literature on the impacts of reducing traffic-related air pollution on population health via regulatory measures.^58,59

A substantial proportion of the associations between residence within 300 m of a highway and term birthweight which we see in unadjusted models are attenuated by maternal sociodemographic characteristics, particularly educational attainment as well as race and ethnicity. In addition, including neighbourhood sociodemographic characteristics further attenuates our risk estimates. These attenuations suggest that high-TRAP exposure zones are home to environmental justice communities, which is well established in previous articles.^60–63 Our results are further diminished by the inclusion of birth city in our matched models, showing that some areas are experiencing differential impacts of TRAP reductions. This study provides preliminary evidence that the benefits of improved TRAP policies may not extend to all exposed populations in the same way. Future research in this birth cohort will attempt to determine which populations have benefited the most, or not at all, from regulatory measures to reduce TRAP.

There are several limitations of our study to consider when interpreting our results. First, the vital statistics data source only provides reported maternal addresses at delivery, so our analyses are predicated on the assumption that mothers lived at their reported residence for the entire pregnancy period, which inherently cannot consider residential changes during the pregnancy period nor daily variations in time-activity patterns. It is also possible that mothers who resided in our comparison group spent time in the high-TRAP group during their pregnancies by exposure pathways such as cycling, walking or working. Second, we assess TRAP exposures using a residential proximity metric instead of an air pollution model. Since our primary question assesses how exposures in high-TRAP zones change over time, we chose to use a proximity metric to ensure that our high-exposure zones are the same across our study period and allow our results to be easily interpretable. This choice, however, may result in exposure misclassification, because we do not consider heterogeneity among highway exposures (e.g. multilevel road structures, congested intersections), non-exhaust pipe emissions (e.g. tyre dust, brake wear), meteorological variability (e.g. predominant wind direction from the household relative to the highway), noise co-exposures (e.g. pavement type) and personal time-activity patterns (e.g. walking or cycling near highways). Third, our results with spatial covariates demonstrate that there may be variation in the impact of TRAP on infant health by community in our study period. Many local and regional policies (e.g. electronic tolling, capacity expansion) have been implemented in recent years to reduce traffic in urban areas which may also yield improvements to infant health in high TRAP exposure zones. We did not include such information, but future research will examine whether local congestion reduction practices also benefit birth outcomes.

Our present analysis provides insight into the effectiveness of regulatory changes targeting TRAP from 1996 to 2009 and associated benefits in reducing adverse birth outcomes in Texas. Infant health risks associated with maternal residence near highways have reduced over this time period, paralleling regulatory measures to improve exhaust pipe emissions. Since many TRAP policy improvements occurred simultaneously, we cannot isolate impacts of specific legislative measures, but the collective effect of these policies is what is most important for reducing TRAP exposures and associated health effects. Our results suggest that the totality of regulatory changes towards TRAP have been effective in protecting infant health.

Supplementary data

Supplementary data are available at IJE online.

Funding

This work is supported by the National Institute of Environmental Health Sciences, National Institutes of Health (grant number F31 ES029801) and the National Center for Advancing Translational Sciences of the National Institutes of Health (grant number TL1 TR002371). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of interest

None declared.