Toxic air pollution and concentrated social deprivation are associated with low birthweight and preterm Birth in Louisiana

Previous studies indicate that pollution exposure can increase risks of adverse birth outcomes, but Black communities are underrepresented in this research, and the potential moderating role of neighborhood context has not been explored. These issues are especially relevant in Louisiana, which has a high proportion of Black residents, an entrenched history of structural racism, the most pounds of toxic industrial emissions annually, and among the nation’s highest rates of low birthweight (LBW), preterm birth (PTB), and infant mortality. We investigated whether air pollution and social polarization by race and income (measured via the index of concentration at the extremes [ICE]) were associated with LBW and PTB among Louisiana census tracts (n = 1101) using spatial lag models. Data sources included 2011–2020 birth records, U.S. Census Bureau 2017 demographic data, and 2017 respiratory hazard (RH) from the U.S. Environmental Protection Agency. Both RH and ICE were associated with LBW (z = 4.4, P < 0.0001; z = −27.0, P < 0.0001) and PTB (z = 2.3, P = 0.019; z = −16.7, P < 0.0001), with no interaction. Severely polluted tracts had 25% higher and 36% higher risks of LBW and PTB, respectively, versus unpolluted tracts. On average, 2166 low birthweight and 3583 preterm births annually were attributable to pollution exposure. Tracts with concentrated social deprivation (i.e. low ICE scores) had 53% higher and 34% higher risks of LBW and PTB, respectively, versus intermediate or mixed tracts. On average, 1171 low birthweight and 1739 preterm births annually were attributable to concentrated deprivation. Our ecological study found that a majority of adverse birth outcomes in Louisiana (i.e. 67% of LBW and PTB combined) are linked to air pollution exposure or disadvantage resulting from social polarization. These findings can inform research, policy, and advocacy to improve health equity in marginalized communities.


Introduction
Environmental health research has long documented disproportionate pollution exposures among black, indigenous, other people of color, and low income people in the U.S. (e.g.[1][2][3][4][5][6][7][8][9][10][11]).Numerous studies have identified corresponding burdens of cancer, respiratory disease, or premature death (e.g.[12][13][14]), but less is known about how racial disparities in pollution exposure affect reproductive health.Such knowledge is a long history of systemic and structural racism, persistently high rates of adverse birth outcomes, and a complete lack of peer-reviewed research addressing pollution impacts on reproductive health [58,60,61].In this analysis, we sought to estimate the degree to which two residential factors, i.e. air pollution exposure and social polarization by race and income, were associated with elevated risk of low birthweight and preterm birth from 2011 to 2020.We hypothesized that census tracts with higher burdens of air pollution and concentrated social deprivation would have higher incidences of these adverse birth outcomes.Because place-based psychosocial stressors are known to exacerbate the effects of chemical exposures [55], we hypothesized that any association between pollution and adverse birth outcomes would be more severe in places of concentrated social deprivation.

Data sources and calculation of study variables
We obtained birth record data (2011-2020) from the Louisiana Office of Vital Records and geocoded the data by maternal census tract of residence using ArcPro (Esri, Inc.).For each census tract, we calculated rates of low birthweight (<2500 g; LBW) and preterm birth (<37 weeks gestation; PTB) as percentages of total births for the 10 yr study period.We excluded tracts with no birth data (n = 20) or fewer than 100 total births (n = 27), resulting in an analytic sample of 1101 census tracts.
We calculated the index of concentration at the extremes (ICE) for each census tract using 5 yr demographic estimates from the US Census Bureau's 2017 American Community Survey (ACS), following established methodology [62].The ICE quantifies the degree to which an area's residents are concentrated into groups of deprivation or privilege based on race and income [62].For each tract, we took the count of White, non-Hispanic persons with annual household income above the 80th national percentile (i.e.$100 000 or above, based on ACS income categories), subtracted the count of Black, non-Hispanic persons with annual household income below the 20th national percentile (i.e.$24 999 or below), and divided the resulting value by the total population of the tract.Possible resulting values of the ICE range from +1 (concentrated privilege) to −1 (concentrated deprivation).
We obtained respiratory hazard (RH) values for Louisiana census tracts from EPA's 2017 AirToxScreen.This indicator of overall pollution burden is calculated from the estimated concentrations of diesel particulate matter and more than 40 different toxic air pollutants (table A1; [63]]).Estimates of RH are only available for certain years from 2011-2020 (i.e.2011, 2014, 2017, 2018 and 2019) [64,65].We chose 2017 RH values because this year is as an approximate midpoint for our births dataset.We conducted an exploratory analysis with 2014 RH values, which yielded results that were consistent with those presented here.Because EPA uses a different, updated methodology with each air toxics assessment, we cannot average pollution data across multiple years.Further, we cannot make meaningful comparisons with a single year of data because of the small number of annual births among census tracts (median, 43 births annually).As described below, we include a sensitivity analysis to account for the possibility that 2017 air quality data may not be representative of exposures in prior years.
To account for shifting demographics during our study period, we calculated the absolute change in the percentage of residents of color (∆Demo) as a nuisance variable for each census tract using preliminary data from the 2020 Census [66].These data were not available for 270 of the 1101 census tracts due to changes in tract boundaries.In order to retain these tracts in the spatial regression model, we assigned them the median ∆Demo value from the overall dataset (3.74%).Given this limitation, we also explored the use of an analogous ∆Demo variable calculated from American Community Survey (ACS) data, which are published yearly using information collected over the preceding 5 yr.This alternative ∆Demo, calculated using 2019 and 2013 ACS data (representing the periods 2015-2019 and 2009-2013), yielded findings that were highly consistent with our main analysis, but with less model support (∆AIC ⩾ 5.7).Thus, we retained the original ∆Demo values (i.e.calculated from preliminary 2020 census data) as a nuisance variable in our analysis.

Spatial lag modeling
We conducted all statistical analyses using R statistical software [67].We report summary statistics as median ± standard deviation.All variables were normally distributed, except RH, which we log-transformed prior to analysis, resulting in an approximately normal distribution.We scaled and centered all predictor variables prior to analysis.
Exploratory analysis revealed significant spatial autocorrelation (P < 0.001) in an ordinary least-squares regression of LBW rates, as tested by a simulation with 10 000 replicates and measured by Moran's i [68].Therefore, we fitted spatial lag models of LBW and PTB as functions of social inequality (ICE), pollution burden (RH), and shifting demographics (∆Demo), using the lagsarlm function from the spatialreg package [69].The lagsarlm function accounts for spatial autocorrelation by estimating the autoregressive lag coefficient, ρ, associated with an n-by-n matrix of spatial weights, W. The final models were: LBW ∼ ICE + log (RH) + ∆Demo + ρW (LBW) PTB ∼ ICE + log (RH) + ∆Demo + ρW (PTB) .
We estimated W using the poly2nb function and the nb2listw function from the spdep package [70].To distinguish direct effects from the indirect influence of neighbors, we calculated model coefficients using the impacts function in the spatialreg package [71].
We generated predicted LBW and PTB rates for each census tract from the respective fitted spatial models using the predict.Sarlm method in the spatialreg package of R [69].We used these modeled rates to calculate relative risk (RR) of each birth outcome (LBW or PTB) associated with each RH and ICE using the formula: We calculated ŷgroup as the mean modeled birth outcome rate among census tracts grouped by RH or ICE percentile (90th, 75th, 50th, 25th, or 10th).For RH, the reference group (ŷ ref ) corresponded to the mean modeled birth outcome rate among the least polluted tracts (10th percentile).For ICE, ŷref corresponded to the mean modeled birth outcome rate among census tracts with a relatively even distribution of race and income (i.e.−0.1 < ICE < 0.1).In calculating ŷgroup , we used ICE > 0.1 and ICE < −0.1 instead of the top and bottom 50th percentiles to avoid overlap with our reference group.
We used the same spatial models and new input data to predict rates of LBW and PTB under theoretical reference conditions, corresponding to scenarios with low pollution (i.e.minimal RH) and a lack of social polarization (i.e.ICE = 0).This approach enabled us to reliably evaluate effect size for our two variables of interest (RH and ICE), while accounting for the log-transformation of RH and the scaling and centering of both variables.For the minimal-RH scenario, we assigned all census tracts an RH value of −4.287, corresponding to the minimum untransformed RH value in the original modeled dataset (i.e.0.193), with all other predictor variables (ICE, ∆Demo) unchanged.For the non-polarized scenario, we assigned all census tracts an ICE value of 0.065 (corresponding to ICE = 0 after scaling and centering), with all other predictors unchanged.For each variable of interest (RH and ICE), we calculated the number of annual LBW cases attributable to that variable by taking the difference in outcome rates (i.e.observed condition minus reference condition) for each census tract, dividing by 100 (to convert from percentage to proportion), multiplying by total births for that tract, summing the values across all tracts, and dividing the resulting value by 10 to obtain the annual average (since total births represent a 10 yr study period).We used the same approach for calculating annual PTB cases attributable to each variable.Importantly, changing the ICE score to 0 corresponds to the elimination of deprivation or the elimination of privilege, depending on the initial ICE value of the census tract.Therefore, we only included tracts that were deprived (ICE < 0) in these calculations, to avoid confounding from the elimination of privilege.

Sensitivity analysis
Because our study period was relatively long (i.e.births from 2011 to 2020), we conducted a sensitivity analysis with births from 2017 to 2020.We focused on this four-year period because it represented the most recent data and still overlapped with the vintage of our pollution dataset (2017) and ICE dataset (2013)(2014)(2015)(2016)(2017).This limited study period resulted in a smaller sample size (n = 890 tracts), since fewer tracts met the criteria of at least 100 births.We excluded four census tracts that had no neighboring tracts (from which to measure indirect effects in the spatial lag model), resulting in an analytic sample of 886 tracts (80% of the main dataset).Using this limited dataset, we repeated the methods described above, except that we did not include demographic change from 2010-2020 (∆Demo) in the models.

Spatial lag models for rates of low birth weight (LBW) and pre-term birth (PTB)
Modeled LBW and PTB rates were relatively high throughout most of northwest Louisiana and along much of the Mississippi River, including the area known as the 'Industrial Corridor' or 'Cancer Alley' (figures 3 and 4; note that LBW and PTB were analyzed in separate spatial models).Higher LBW and PTB rates were associated with higher RH, lower ICE, and higher ∆Demo (tables 1 and 2).For each outcome (LBW and PTB), there was no significant interaction between RH and ICE (model estimates = −0.09and −0.004, P = 0.15 and 0.95, respectively), so we excluded this interaction term from the final models.As expected, we detected spatial autocorrelation within the LBW dataset (ρ = 0.25, P < 0.0001) and within the PTB dataset (ρ = 0.46, P < 0.0001); however, there was no significant residual autocorrelation in the respective spatial lag models (P = 0.92 and P = 0.15).These spatial models performed better than corresponding ordinary least-squares regressions (∆AIC = 57 for LBW and ∆AIC = 185 for PTB) and better than analogous spatial lag models without: ICE (∆AIC = 611 and 292, respectively), RH (∆AIC = 16 and 3, respectively), or ∆Demo (∆AIC = 4 and 11, respectively).
RR of each outcome variable increased with increasing RH and decreasing ICE (tables 3 and 4).The most polluted tracts had a 36% higher risk of LBW and a 25% higher risk of PTB compared to the least polluted tracts (table 3).On average, 2166 out of 5935 LBW cases and 3583 out of 6931 PTB cases in Louisiana each year were attributable to air pollution exposure.The most deprived census tracts had a 53% higher risk of LBW and a 34% higher risk of PTB compared to reference tracts (i.e.those with an ICE score near zero; table 4).On average, 1171 LBW cases and 1739 PTB cases in Louisiana annually were attributable to social deprivation (i.e.ICE < 0).

Discussion
In this population-based observational study, we found evidence that the majority of adverse birth outcomes in Louisiana (i.e.67% of low birthweights and preterm births combined) are attributable to air pollution exposure or to social deprivation resulting from structural racism, as measured by RH and ICE, respectively.Importantly, our reference population for understanding the impacts of social deprivation was intermediate/mixed census tracts, meaning that the burden of adverse birth outcomes would be more extreme if we used areas of concentrated privilege (high-income, predominantly White neighborhoods) as the comparison group.Our findings are broadly consistent with other studies that have linked adverse birth outcomes with air pollution exposure (reviewed in [17]) or structural racism measured by ICE [72][73][74][75][76][77].Yet our findings also suggest that the magnitude of pollution effects may be underestimated by studies that rely on data for a single pollutant.Specifically, the estimated number of annual preterm births attributable to air pollution exposure in our study (n = 3583) was more than 15 times higher than the annual number of preterm births (n = 218) that a previous study attributed to fine particulate matter (PM2.5)exposure in Louisiana [45].A possible explanation for this difference is that our study used a more comprehensive measure of air pollution exposure, resulting in a larger exposed population.(Respiratory Hazard is calculated by EPA's AirToxScreen Program and is based on modeled exposures to diesel particulate matter and 44 hazardous air pollutants (table A1; [51]).Regardless, the consistencies within our study suggest that Louisiana's burden of air pollution may influence birthweight and gestational period through common biological mechanisms that could have broader health effects than those assessed here.Thus, understanding the full extent of these impacts, as well as their mechanistic basis, are priorities for future research.
The lack of significant interaction between RH and ICE among census tracts was unexpected, since numerous studies indicate that socioeconomic stressors can exacerbate the harm of chemical exposures on birth outcomes (reviewed in [55,57]).However, our ability to detect interactive effects between RH and ICE may be limited because our dataset contained relatively few census tracts that were both polluted and highly privileged (only 60 of the 1101 tracts were in the 75th percentile for both factors; figures 1 and 2).It is also possible that social polarization at smaller geographic scales (e.g.census blocks) interacts with air pollution to influence birth outcomes.Notably, an ICE value of zero can indicate that no resident is classified as either extremely privileged or deprived, but can also mean that there are similar numbers of residents in each group.Unfortunately, the lack of corresponding pollution data prevented us from investigating these relationships at geographic scales smaller than the census tract.Alternatively, social privilege may not protect against pollution-related adverse birth outcomes in Louisiana.For example, higher geographic mobility in privileged areas may protect against harms from chronic pollution exposure (e.g.cancer), but not against shorter-term harms, such as adverse birth outcomes.Another unexpected finding was the relatively modest association between RH and ICE, which was somewhat surprising given the severe racial disparity in pollution exposure among Louisiana census tracts [44].This finding may be partly explained by the prevalence of concentrated deprivation in very rural parts of the state, particularly along the upper Mississippi River and in central Louisiana, where there is less air pollution (figures 1 and 2).
This study appears to be the first peer-reviewed research focused on how pollution exposure influences birth outcomes in Louisiana, despite public outcries about this issue (e.g.[36]) and the longstanding availability of relevant datasets (e.g.[78]).This lack of peer-reviewed research, combined with the history of misguided public health investigations in Louisiana (e.g.[37,40]), raises serious concerns about the role that the scientific community has played in enabling health disparities to persist in this state.These concerns are reinforced by our current findings, which suggest that toxic air pollution is a major driver of the exceptionally high rates of low birthweight and preterm births in Louisiana.Our findings also reveal that these health burdens are far more extreme in certain neighborhoods; some census tracts had low birthweight rates that were more than double the Louisiana average and more than triple the U.S. average [79], and the disparity in preterm birth rates was nearly as severe.Notably, these averages obscure the even more extreme (∼10-fold) disparity in low birthweight and preterm birth among Louisiana's census tracts.Many of the census tracts with the highest rates of adverse birth outcomes in our spatial lag models were in northern Louisiana, including the Shreveport-Bossier City metropolitan area, and in southeast Louisiana, including the area known as the Industrial Corridor or Cancer Alley (figures 3 and 4).This industrialized area has  received considerable media attention not only for pollution-related health concerns, but also for environmental justice advocacy, which has been largely led by Black women [80].Like all epidemiological research, our study has limitations to consider.First, it is ecological and cross-sectional, which allows us to draw conclusions about relationships but not causality.Second, we used an area-level indicator of total pollution burden (i.e.respiratory hazard; RH), so we cannot distinguish the effects of individual pollutants, or the corresponding biological mechanisms.Despite this limitation, RH provides a useful measure of real-world risk because it accounts for the combined exposure to a large number of pollutants (table A1).We considered using developmental hazard (DH) from the 2017 AirToxScreen; however, DH is based on a relatively small number of pollutants (n = 8) and does not include certain pollutants (e.g.benzene) that are known to have direct developmental effects but which lack quantitative, developmental-specific risk values for chronic exposure [81].We also recognize that RH is a 'snapshot' of pollution levels in 2017, whereas air quality inevitably varied over the study period (2011-2020).However, our sensitivity analysis, which limited the dataset to 2017-2020 births, yielded findings that were consistent with our main analysis.Furthermore, relative levels of RH tend to remain stable over time, because the most polluted neighborhoods generally remain the most polluted.

Conclusion
Our findings suggest that efforts to limit air pollution exposure and expand social services in Louisiana could improve health outcomes for thousands of newborns each year, as well as the future adult population.Such efforts should be a major public health priority, given the state's exceptionally high incidence of adverse birth outcomes and associated adult diseases [26,27,82,83].Our current study also adds to the growing evidence that Louisiana's extreme burden of industrial pollution is a significant driver of disease in the state [14].Yet more attention from the research community is needed to fully understand the extent of Louisiana's pollution-related health burden.Until science catches up with this need, decision-makers should exercise the precautionary principle, especially regarding Louisiana census tracts with extreme burdens of pollution, social deprivation, and/or disease.Our findings also emphasize the importance of including underrepresented demographics in environmental health research, as well as the need to account for both multi-pollutant exposures and fine-scale variation in risk factors.Future research should consider cumulative pollution exposures in the context of structural racism and should focus on populations that continue to be underrepresented in reproductive/environmental health studies.These populations deserve attention from the scientific community, and, in turn, they can help us understand how state and local policies contribute to health disparities [56,84].

Figure 1 .
Figure 1.Index of concentrations at the extreme (ICE) among Louisiana census tracts (n = 1125) based on 2017 data from the U.S. Census Bureau's American Community Survey.This measure accounts for social polarization based on race and income, with negative values indicating concentrated deprivation and positive values indicating concentrated privilege (see methods).Black lines indicate parish boundaries.The Mississippi River and the Lake Maurepas-Pontchartrain-Borgne system are illustrated in blue.The map inset illustrates the southeast Louisiana Industrial Corridor, also known as Cancer Alley.Data were not available for 23 of the 1148 tracts in Louisiana.

Figure 2 .
Figure 2. Respiratory Hazard (RH) among Louisiana census tracts (n = 1128), as estimated by the U.S. Environmental Protection Agency's 2017 AirToxScreen.Black lines indicate parish boundaries.The Mississippi River and the Lake Maurepas-Pontchartrain-Borgne system are illustrated in blue.The map inset illustrates the southeast Louisiana Industrial Corridor, also known as Cancer Alley.Data were not available for 20 of the 1148 tracts in Louisiana.

Figure 3 .
Figure 3.Rates of low birthweight (LBW) among Louisiana census tracts (n = 1101) predicted by a spatial lag model using birth data from 2011 to 2020.Black lines indicate parish boundaries.The Mississippi River and the Lake Maurepas-Pontchartrain-Borgne system are illustrated in blue.The map inset illustrates the southeast Louisiana Industrial Corridor, also known as Cancer Alley.Census tracts with no birth records (n = 20) or fewer than 100 births over the study period (n = 27) were excluded from the model.

Figure 4 .
Figure 4. Rates of pre-term births (PTB) among Louisiana census tracts (n = 1101) predicted by a spatial lag model using birth data from 2011 to 2020.Black lines indicate parish boundaries.The Mississippi River and the Lake Maurepas-Pontchartrain-Borgne system are illustrated in blue.The map inset illustrates the southeast Louisiana Industrial Corridor, also known as Cancer Alley.Census tracts with no birth records (n = 20) or fewer than 100 births over the study period (n = 27) were excluded from the model.

Table 1 .
Impact measures from spatial lag model of low birthweight rate among Louisiana census tracts (n = 1101) a .Index of Concentrations at the Extremes (ICE) is a measure of social polarization, calculated from race and income.Values >1 indicate concentrated privilege, <1 indicate concentrated deprivation, and 0 indicates an intermediate or mixed population.Respiratory hazard (RH) is an estimate of overall air pollution burden derived from modeled exposures to 45 different air pollutants.Change in the percent residents of Color (∆Demo) accounts for shifting demographics between 2010 and 2020.See Methods for full descriptions of model variables. b

Table 2 .
Impact measures from spatial lag model of pre-term birth rate among Louisiana census tracts (n = 1101) a .
bSee table 1 and methods for a description of predictor variables.

Table 3 .
Relative risks of low birth weight (LBW) and preterm birth (PTB) among census tracts grouped by Respiratory Hazard (RH).Mean of fitted values from spatial lag model (n = 1101 census tracts; see tables 1 and 2).c Relative to modeled rates among the least polluted census tracts (i.e.10th percentile RH). b

Table 4 .
Relative risks of low birth weight (LBW) and preterm birth (PTB) among census tracts grouped by ICE score.