A census tract-level assessment of social determinants of health, traffic exposure, and asthma exacerbations in New York State's Medicaid Population (2005–2015)

Abstract

This study aims to evaluate the association between social determinants, environmental exposure metrics, and the risk of asthma emergency department (ED) visits in the New York State (NYS) Medicaid population using small-area analysis. Traffic densities for each census tract in NYS were calculated using the length of road segments within each tract and total area of the tract to produce a measure of average number of vehicles per square meter per day. Data on social determinants of health including internal and external environments and other demographic factors were obtained from various sources. Poisson regression analyses were conducted to identify significant factors associated with asthma ED visits in Medicaid claim and encounter data for years 2005–2015. High traffic density in NYS excluding New York City (NYC) correlated with increased risk of asthma ED visits (RR 1.69; 95% CI: 1.42, 2.00), mitigated by adjusting for environmental and social determinants (RR 1.00; 95% CI: 0.85, 1.19). Similar trends were observed in NYC only (RR 1.19; 95% CI: 1.00, 1.41), with the adjusted risk remaining elevated (RR 1.14; 95% CI: 0.98, 1.33) albeit not statistically significant. Living in census tracts with high concentrated disadvantage index, high proportions of minorities, and less green space predicted higher asthma ED visits. We mapped predicted rates and model residuals to identify areas of high risk. Our results support previous findings that environmental and social risk factors in poor and urban areas contribute to asthma exacerbations in the NYS Medicaid population, even if they may not necessarily contribute to its development.

Graphical abstract

Highlights

• •New York State Medicaid enrollees living in high traffic density census tracts had an increased risk of asthma emergency department visits compared to enrollees living in low traffic density census tracts.
• •The concentrated disadvantage index, a comprehensive neighborhood measure of socioeconomic status, is significantly associated with asthma emergency department visits.
• •Environmental and social risk factors contribute to asthma exacerbations, particularly in poor and urban areas.
• •Data mapping techniques, like those used in this study, can help policy makers target limited resources to areas of greatest need.

1. Introduction

An estimated 25 million (7.7%) Americans have asthma, of whom about 10 million experience an asthma exacerbation, resulting in 1 million emergency department (ED) visits each year [1]. Social determinants of health (SDOH), defined by the World Health Organization as nonmedical factors that influence health, are known to contribute to asthma morbidity [2]. Lower socio-economic status (SES) has been associated with higher morbidity in patients with chronic diseases [3,[4], [5]], particularly asthma [6]. Those with a lower SES may have higher exposures to both outdoor (e.g., PM_2.5, nitrogen oxides, sulfur dioxides, and volatile organic compounds) [[7], [8], [9]] and indoor (e.g., tobacco smoke, mold, cockroaches, crowding) triggers, thereby increasing their risk for exacerbations. In particular, prior studies have linked proximity to traffic to adverse asthma outcomes [[10], [11], [12]], but have also suggested that socioeconomic factors play a significant role in the associations observed [13]. These studies have identified people with lower incomes, low levels of English proficiency, and people of color as those more likely to live in areas with high traffic density [14] and thus at higher risk for adverse health outcomes related to traffic exposure. A disproportionate burden of air pollution exposures has been found in areas with lower income and/or minority populations, leading to environmental justice (EJ) concerns [8,9]. Urban communities, particularly those located centrally in a city, are usually characterized by low SES, and predominantly populated by race/ethnic minorities. These areas are usually referred to as “inner cities”. The increased burden of asthma morbidity in inner-city neighborhoods may stem from several factors, including proximity to pollution sources like traffic, industrial facilities, and waste disposal sites, as well as housing type and conditions, and other factors related to lower SES. Health disparities for outcomes such as asthma that are known to be impacted by SDOH have complex neighborhood-level drivers that are not well understood. Therefore, identifying contextual factors that place certain areas at higher risk is a goal for public health programs such as the Medicaid program.

In this study, we aim to use small-area analyses to evaluate the association between SDOH and environmental exposure metrics and the risk of asthma ED visits at the census tract level in the New York State (NYS) Medicaid population. This study of the Medicaid population provides the opportunity to independently evaluate risk factors among those who are most disadvantaged and have the highest burden of asthma to understand health inequities in asthma morbidity.

2. Methods

2.1. Study population

NYS Medicaid encounter data, submitted by Medicaid managed care (MMC) health plans, and post-adjudicated fee-for-service (FFS) claim data were used to identify Medicaid enrollees less than 65 years of age who had ED visits with a primary diagnosis of asthma (ICD-9 code 493.xx, ICD-10 code J45; CPT Codes: 99281–99285; Revenue Codes: 0450–0459, and 0981) between 2005 and 2015 [[15], [16]]. Medicaid enrollees with a diagnosis of chronic obstructive pulmonary disease (COPD) (ICD-9 code 493.2x, ICD-10 code J44.0, J44.1, and J44.9) at any time during the study period were excluded based on HEDIS specifications and also because the etiology, symptoms, and treatment of COPD differ from asthma. 99.04% of enrollee home addresses at the time of the ED visit were successfully geocoded. Enrollees were further categorized as living in NYS excluding NYC [Rest of State (ROS)] or New York City (NYC).

Fig. 1 describes the inclusion criteria for the final dataset used. We started out with 713,245 ED visits in 4,806 census tracts after excluding all COPD cases. Further exclusion criteria included zero-population census tracts and greater than 50% tract population in group quarters. The final study population included 710,554 ED visits (99.6%) within 4,732 census tracts (96%).

Fig. 1.

Flowchart of patient distribution in the medicaid population (2005–2015).

2.2. Outcome

The total number of ED visits per census tract per year was calculated and served as the outcome for this assessment.

2.3. Traffic exposure

Traffic density, the average number of vehicles per square meter per day, was used as a proxy for exposure to traffic-related emissions across the census tract [17,18]. Using revised 2016 Annual Average Daily Traffic (AADT) data from the New York State Department of Transportation [19], three measures of traffic density were calculated for each census tract after excluding tracts that contained only water bodies. Traffic density was calculated by multiplying the length of each road segment within the tract by the AADT for the full road, then summing the results for the tract and dividing by the total area of the census tract to produce a measure of the average number of vehicles per square meter per day. An unweighted measure (UWTD) was calculated using the 2016 AADT data and includes the total number of vehicles on a segment of road each day. Next, truck traffic density (TTD), an estimate of truck emissions exposure, was obtained using the reported percentage of trucks on a road segment each day. Trucks were defined as vehicles belonging to Vehicle Classification Codes F04 (Buses) through F13 (Seven-or-More Axle, Multi-Trailer Trucks) [20]. Finally, weighted traffic density (WTD)–the weighted estimate of traffic density was calculated by multiplying the number of trucks by 3.6 to account for higher toxic emissions produced by heavy-duty, typically diesel-fueled vehicles compared with gasoline-fueled passenger vehicles [21] and adding the result with the number of non-truck vehicles on that road segment. For each measure of traffic density, census tracts were categorized into terciles of high, medium, or low traffic density based on the distribution of all traffic densities across all census tracts.

2.4. Social determinants

Information on SDOH was derived from the American Community Survey (ACS) 2008–2012 dataset [22]. We calculated the Concentrated Disadvantage Index as the sum of the z-scores of five socio-economic status variables: percent of total population under age 18, percent of those in the civilian labor force who are unemployed, percent of households with female-headed families (no spouse present), percent of the population in poverty, and percent of households with public assistance income or food stamps/Supplemental Nutrition Assistance Program in the past 12 months [23]. Positive scores on this index indicate higher levels of Concentrated Disadvantage while negative scores indicate lower levels of Concentrated Disadvantage. Other demographic variables were % minority (mean % Asian + % non-Hispanic Black + % Hispanic) and % female.

2.5. Other environmental exposures

We also included built environment information from the ACS such as household crowding (total occupied housing with more than 1.5 persons per room), older housing (houses built before 1980), and dirty home fuel sources. The primary source of fuel for most households in NYS was utility gas (mean of 58%), followed by kerosene/fuel oil (mean of 27%), electric (mean of 8.6%), and wood (mean of 1.8%). Of the home fuel sources classified as “dirty”, kerosene/fuel oil was the major contributing source, so proportion of houses with kerosene/fuel oil as fuel source was used to indicate “dirty” fuel use in a census tract (data not shown).

Data from the National Land Cover Database 2011, were used to create the proportion of green space (external environment) variable for each census tract [24,25]. To calculate the proportion of green space within a census tract, we summed the proportion of area within a tract classified as developed (open space), deciduous forest, evergreen forest, mixed forest, shrub/scrub, grassland/herbaceous, pasture/hay, cultivated crops, woody wetland, and emergent herbaceous wetlands to account for all classifications where significant plant growth occurs. Urbanicity was determined using the rural-urban commuting area (RUCA) codes (Classification C), which classify US census tracts using measures of population density, urbanization, and daily commuting [26].

In the analysis restricted to NYC census tracts, a measure of indoor air quality was calculated from the NYC Department of Housing Preservation and Development's Housing Maintenance Code Violations dataset [27]. The dataset includes all violations open as of October 1, 2012 and is updated daily with newly issued violations and status changes. Class C violations issued for conditions known to exacerbate asthma and other respiratory issues including water leaks, presence of mold or mildew, and presence of rodents, cockroaches, and other pests were extracted for years 2013–2015 to capture all violations issued within a full calendar year during the period of study. The number of violations was averaged by census tract to produce a yearly average number of violations related to indoor air quality. Such data were not available for the ROS.

2.6. Statistical analysis

This paper uses small area analysis (SAA) approach to explore associations between traffic density and the risk of asthma ED visits at the census tract level while controlling for SDOH and other environmental exposures. Small area analysis (SAA) describes statistical methods or assessment procedures that focuses on small specific geographic areas or populations (in this case, NYS census tracts) in order to identify disparities or differences between and/or amongst them and any statistical pattern. Poisson regression analyses were conducted to identify significant factors associated with rate of asthma ED visits in each census tract. The total Medicaid enrollee population in December 2012 was calculated for each census tract as an offset variable in the models [[16], [28]]. To evaluate the robustness of the findings for traffic measures, we performed the multivariate regression analysis with each traffic measure separately. We first fit unadjusted models with each predictor. Model 1 included traffic density, SDOH such as concentrated disadvantage index, and % minority and % female demographics. Model 2 additionally included predictors related to the environment (% crowding, % old homes, % kerosene/fuel oil, % green space, and average building violations for NYC). The results were represented as the rate ratio (RR) and 95% confidence intervals (CI).

Statistical analyses were performed using SAS version 9.4. Spatial patterns of residual variation were analyzed to identify unique risk factors in areas of significant deviation from model predictions. Predicted rates and model residuals from model 2 were mapped to identify these areas of high risk.

3. Results

Medicaid enrollees with addresses geocoded to census tracts that had zero population or had >50% population in group quarters according to the 2010 US Decennial Census were excluded (Fig. 1). There were 295,765 Medicaid enrollees with at least one ED visit for asthma-related diagnoses resulting in 710,554 ED visits between 2005 and 2015 in NYS geocoded to 4,732 census tracts after exclusions. Most census tracts were in ROS (55.85%) compared to NYC (44.15%). The average number of vehicles per square kilometer (km) per day was significantly higher in NYC compared to ROS for all three traffic density classifications (Table S1). Approximately 33.64% of all NYS census tracts were classified as having high traffic density, while 33.66% and 32.69% had medium and low traffic density, respectively.

However, a higher proportion of enrollees were classified as living in high traffic density area in NYC (57.78%) compared to ROS (14.49%) in the weighted model. Similar distributions were observed in both the UWTD and TTD models for ROS and NYC (Table S2). NYC had a higher mean percentage of minorities, Concentrated Disadvantage Index, crowding, and developed area compared to ROS (Table S2). Since the distribution of covariates was very different between NYC and ROS, we conducted separate analyses to characterize risks separately for these areas.

In bivariate analysis for WTD measures (Table 1), we observed a significantly increased risk of asthma ED visits in the overall NYS Medicaid population from 2005 through 2015 among enrollees living in census tracts with medium (RR 1.36; 95% CI: 1.15, 1.62) or high traffic density (RR 1.62; 95% CI: 1.21, 2.16). The estimates were similar in magnitude for ROS in both medium (RR 1.43; 95% CI: 1.25, 1.63) and high (RR 1.69; 95% CI: 1.42, 2.00) traffic density tracts. In a NYC only assessment, the risk ratios for medium (RR 0.99; 95% CI: 0.79, 1.25) and high traffic density tracts (RR 1.19; 95% CI: 1.00, 1.41) were not statistically significant and were smaller compared to ROS or NYS. Similar risks were observed in the unweighted and trucks only models.

Table 1.

Risk ratios for the association between traffic density at the census tract and asthma emergency department visits.

Model	Unadjusted		IHSTa
	Medium traffic density	High traffic density	Log traffic density
Weighted
NYS	1.36 (1.15, 1.62)	1.62 (1.21, 2.16)	1.12 (1.05, 1.20)
NYS excl.NYC	1.43 (1.25, 1.63)	1.69 (1.42, 2.00)	1.15 (1.10, 1.20)
NYC	0.99 (0.79, 1.25)	1.19 (1.00, 1.41)	1.02 (0.99, 1.06)
Unweighted
NYS	1.36 (1.15, 1.60)	1.61 (1.22, 2.13)	1.12 (1.05, 1.20)
NYS excl.NYC	1.44 (1.26, 1.64)	1.70 (1.43, 2.03)	1.15 (1.10, 1.20)
NYC	0.99 (0.76, 1.28)	1.17 (0.99, 1.39)	1.02 (0.99, 1.06)
Trucks only
NYS	1.40 (1.15, 1.70)	1.72 (1.25, 2.36)	1.13 (1.05, 1.21)
NYS excl.NYC	1.46 (1.28, 1.66)	1.63 (1.40, 1.91)	1.14 (1.10, 1.19)
NYC	1.04 (0.83, 1.30)	1.17 (1.03, 1.34)	1.03 (0.995, 1.06)

The bolded numbers show statistical significance of the risk ratios.

^aInverse hyperbolic sine transformation.

We therefore used the WTD model for multivariate analysis. Higher risk was associated with areas with medium or high traffic density, higher disadvantage index, and higher proportions of minorities, females, or older homes. A lower risk of asthma ED visits was found in areas with more green space. In the multivariate analysis for ROS adjusted for SDOH, an increased risk of asthma ED visits was observed in census tracts with high traffic density (RR 1.11; 95% CI: 0.95, 1.31) compared to those in low traffic density tracts (Table 2, Model 1), but it was not statistically significant. Higher proportion of minorities was a significant predictor of asthma ED visits in the area (Table 2). On further adjustment for other internal and external built environment factors (Table 2, Model 2), the risk due to traffic was further ameliorated. Census tracts with higher proportions of older homes were at 1.01% greater risk (RR 1.01; 95% CI 1.00, 1.01) asthma (Table 2).

Table 2.

Risk ratios and 95% confidence intervals for associations between weighted traffic density, social determinants, built environment and asthma ED visits for Medicaid enrollees in ROS (2005–2015).

Variable	Unadjusted (ROS)	Model 1a Adjusted (ROS)	Model 2b Adjusted (ROS)
	RR (95% CI)	RR (95% CI)	RR (95% CI)
Traffic density
Medium	1.43 (1.25, 1.63)	1.10 (0.97, 1.24)	1.01 (0.89, 1.16)
High	1.69 (1.42, 2.00)	1.11 (0.95, 1.31)	1.00 (0.85, 1.19)
Social determinants
Concentrated disadvantage indexc	1.05 (1.04, 1.07)	1.02 (0.997, 1.05)	1.02 (1.00, 1.04)
% Minority	1.03 (1.02, 1.04)	1.02 (1.01, 1.03)	1.02 (1.01, 1.03)
% Female	1.02 (1.01, 1.03)	1.00 (0.99, 1.01)	1.00 (0.99, 1.01)
Built Environment
% Total Occupied Housing > 1.5 Persons/room	1.02 (0.98, 1.07)		0.98 (0.94, 1.01)
% Old home	1.02 (1.01, 1.02)		1.01 (1.00, 1.01)
% Fuel oil/kerosene	1.00 (0.99, 1.00)		1.00 (0.997, 1.00)
External environment
% Green spaced	0.99 (0.99, 0.99)		1.00 (0.997, 1.00)

^aModel 1: traffic density, social determinants.

^bModel 2: traffic density, social determinants, built environment, external environment.

^cConcentrated disadvantage index: summation of z-scores of percent [total population under 18, unemployed civilian labor force, household with female-headed families (no spouse present), population in poverty, households with public assistance income or food stamps/SNAP in past 12 months].

^dGreen space: summation of proportion of tract classified as developed (open space), deciduous forest, evergreen forest, mixed forest, shrub/scrub, grassland/herbaceous, pasture/hay, cultivated crops, woody wetland, emergent herbaceous wetlands.

In the NYC only models, the associations were similar to those observed in ROS with a few exceptions. Notably, the risk associated with high traffic density remained elevated although not statistically significantly, even after adjustment for SDOH and built environment (Table 3, Model 2). In contrast to ROS, there was reduced risk of asthma ED visits for census tracts with higher proportions of older housing (RR 0.99; 95% CI: 0.99, 0.998) (Table 3). We also observed a small increase in risk for asthma ED visits in NYC census tracts with a higher proportion of homes with kerosene/fuel oil home fuel sources (RR 1.01; 95% CI: 1.00, 1.01). The overall risk for asthma ED visits in NYS census tracts adjusted for SDOH and built environment was also assessed with concentrated disadvantage index, and % minority as significant predictors (Table S3).

Table 3.

Risk ratios and 95% confidence intervals for associations between weighted traffic density, social determinants, built environment and asthma ED visits for Medicaid enrollees in NYC (2005–2015).

Variable	Unadjusted (NYC)	Model 1 Adjusted (NYC)	Model 2 Adjusted (NYC)
	RR (95% CI)	RR (95% CI)	RR (95% CI)
Traffic density
Medium	0.99 (0.79, 1.25)	0.95 (0.84, 1.07)	0.999 (0.89, 1.12)
High	1.19 (1.00, 1.41)	1.12 (0.94, 1.34)	1.14 (0.98, 1.33)
Social determinants
Concentrated disadvantage index	1.09 (1.05, 1.13)	1.05 (1.00, 1.10)	1.03 (0.99, 1.08)
% Minority	1.05 (1.02, 1.09)	1.03 (0.998, 1.07)	1.03 (1.00, 1.07)
% Female	1.05 (1.02, 1.07)	1.02 (0.999, 1.04)	1.01 (0.999, 1.03)
Built environment
% Total occupied housing > 1.5 Persons/room	0.97 (0.93, 1.01)		0.96 (0.94, 0.98)
% Old home	0.99 (0.99, 1.00)		0.99 (0.99, 0.998)
% Fuel oil/kerosene	1.01 (1.01, 1.02)		1.01 (1.00, 1.01)
Average yearly building violations	1.00 (1.00, 1.01)		1.00 (0.999, 1.00)
External environment
% Green space	1.00 (0.99, 1.02)		1.00 (0.999, 1.01)

Model predicted rates and residuals were mapped to identify areas of elevated asthma risk in NYS. Fig. 2 shows a thematic map for ROS with predicted number of ED visits per 1,000 Medicaid enrollees. Areas with darker colors depict higher predicted rates. Most areas of high risk of asthma ED visits are in inner city urban areas (NYC–Fig. 3, Buffalo area–Figure S1, Capital District area–Figure S2). We used hatches to show areas where the model residuals suggest significant underprediction (blue hatches) or overprediction (purple hatches) (Fig. 2, Fig. 3, S1–S4). When we assessed spatial model fit using residuals in NYC, only 60 census tracts had significant overprediction [residuals < 1.5 Standard Deviation (SD)], while 147 census tracts had significant underprediction (residuals > 1.5 SD). Only 83 census tracts in ROS had observed rates lower than the model predicted rates (residuals < 1.5 SD). There were 179 census tracts where observed rates were even higher than predicted rates, mostly in inner city urban areas (residuals > 1.5 SD) (Fig. 3, S1–S4).

Fig. 2.

Predicted rates of emergency department visits for asthma by medicaid recipients by census tract in NYS excluding NYC (2005–2015). Data Source: New York State Department of Health, Office of Health Insurance Programs Medicaid Data Mart, accessed October 11, 2022.

Fig. 3.

Predicted rates of emergency department visits for asthma by medicaid recipients by census tract in NYC (2005–2015). Data Source: New York State Department of Health, Office of Health Insurance Programs Medicaid Data Mart, accessed October 11, 2022.

4. Discussion

The relationships between air pollution due to traffic density, SDOH, built environment, and adverse health outcomes such as asthma, are complex and often contribute to inequity in health care, especially in low-income populations [29]. Our study used Medicaid claim and encounter data to assess area level risk factors that may contribute to asthma ED visits among low-income populations. Like our study, a nationwide study of Medicaid members showed that asthma morbidity as characterized by ED visits is higher among minorities and inner city/poor urban areas [30]. Use of geospatial tools and SDOH and environmental variables to predict areas of high risk can help inform targeted policy interventions [31,32]. Based on our model, these areas have higher levels of risk factors such as neighborhood deprivation, traffic, and other built environment variables. Although these models simplify the complex factors that interplay to determine environmental equity, when used along with community specific program information, they can be useful tools to determine target areas for intervention.

Spatial patterns of residual variation can be analyzed to identify unique risk factors in areas of significant deviation from model predictions. Model fit was assessed by analyzing residuals for the model. In areas that have residuals > 1.5 SD, the observed rates are significantly higher than predicted and these areas likely have additional, unmeasured risk factors that are contributing to more asthma ED visits. Conversely, in communities where residuals are <1.5 SD, there may be unmeasured protective factors that contributed to fewer asthma ED visits even though risk factors are prevalent in the community. As is evident, significant underprediction of the model is more frequent than significant overprediction in both ROS and NYC models. Although this may be attributable to factors such as lower asthma counts in some tracts, overall, there was no apparent link between lower asthma counts per tract and underprediction/overprediction. A plausible reason may be that our models are unable to capture other significant triggers of asthma including point source pollution [33]. Other nonpoint sources of pollution related to transportation including shipping, diesel trains, and airplanes, indoor air quality, and other naturally occurring allergens such as pollen may also be contributing to these observed high risks. Traffic density may also have been underestimated when using AADT data. On examination of the AADT dataset, we found records with zero (0) AADT counts for road segments in areas where traffic exists especially in NYC, Albany, Buffalo, and Syracuse. Accuracy of geocoding likely affects the estimation of number of ED visits and estimated Medicaid population within a census tract leading to overprediction or underprediction depending on the tract. AADT data also do not account for elevation and other geographic characteristics of roadway or the residential dwelling that may lead to misclassification of traffic exposure. However, using individual level data, we have assessed the association between traffic density and recurrence of asthma ED visit in this study population in a previous study [34], where traffic densities were calculated and assigned based on the individual's home addresses [34].

NYC has the highest traffic density census tracts in the state. Previous studies have linked exposures to traffic emissions with multiple adverse health effects [[35], [36], [37]]. Traffic is known to produce a mixture of gaseous and particulate air pollution such as volatile organic compounds, black carbon, PM_2.5, and ozone. Asthma exacerbations and severity have been associated with ambient pollutants including benzene concentrations [38] and proximity to traffic [39]. Similar to what has been reported in areas across the world, our results suggest that Medicaid enrollees in New York are at higher risk of asthma ED visits in areas with high traffic exposure [[40], [41], [42]]. Although we observed higher risk of asthma ED visits in NYC, which are associated with high traffic exposure, this risk also persisted in ROS tracts despite lesser traffic-related exposures. This suggests that factors, such as SDOH and built environment, play a larger role in asthma exacerbation in high pollution neighborhoods than in low pollution neighborhoods. In a study that used high neighborhood asthma prevalence as a proxy measure for SES and built environment, Lovinsky-Desir et al. reported similar findings of lower risk of urgent asthma medical visits associated with air pollution in high-risk neighborhoods [42]. Our nested models reveal that increased risk of asthma due to high traffic density persists in NYC after adjustment of SDOH and built environment variables, although not statistically significant. In ROS, the risk is attenuated after adjustment for green space and built environment.

Most studies use surrogates of traffic exposure such as traffic counts, multiple distances within a near-road exposure zone, and proximity to major roads for epidemiological studies of health effects [12,43,44]. These measures, however, are less suitable for aggregate area analysis, thus we calculated traffic density from the AADT as a proxy for near road exposure to vehicle emissions within a defined geographic area. While this method is not as rigorous as taking actual air quality measurements from indoor/outdoor monitors, it is a more cost-effective approach to determining exposures to pollution. This method assumes equal exposure to pollutants to all residents of the census tract and does not include spillover pollutants from neighboring tracts. The calculated traffic density captures the density of roadways within a tract, traffic levels on these roads, and vehicle composition to provide a more comprehensive measure of the burden of traffic emissions within a census tract. Based on evidence that truck traffic is a strong influence on near-road pollutant levels including black carbon and PM_2.5, we included a measure of truck traffic density, which may be of particular concern in EJ areas [[45], [46], [47]].

We found that even among a low-income population, a comprehensive measure of neighborhood SES, the concentrated disadvantage index, showed a significant association with asthma ED visits. Neighborhood disadvantage can lead to an interplay of daily and chronic stressors for its residents including poverty, poor infrastructure leading to crowding and noise, unemployment, low social capital, exposure to violence and crime, and inadequate access to care [29,48,49]. Stress is also associated with increased asthma severity [50]. Neighborhood disadvantage may also influence behavioral risk factors if residents adopt dysfunctional coping strategies (e.g., indoor smoking) that may lead to increased asthma severity [51]. We found that areas with a high proportion of minorities were more likely to have higher asthma ED visits even after adjusting for neighborhood disadvantage and environmental factors. The historical legacy of neighborhood segregation and discrimination may contribute to such persistent disparities. We used the concentrated disadvantage index to capture multiple aspects of SDOH, but found that racial disparities persisted even after adjustment by this index.

Poor quality housing can harbor allergens and triggers such as mites, mold, and cockroaches [52]. Inadequate ventilation can also result in higher concentrations of asthma triggers such as allergens and tobacco smoke in the home [53]. We found significant associations with older housing and asthma ED visits in ROS but not in NYC. A recent research paper suggests that age of housing and SES may be important risk factors for rural and suburban areas [54]. The difference in the association of old housing and asthma ED visits in NYC compared to ROS suggests that older housing in NYC may not equate to substandard housing and/or indoor environmental pollutants which have been linked to increased sensitization and asthma morbidity for low-income racial/ethnic groups and children living in urban areas [55,56]. Although older and poor housing is an important risk factor for asthma exacerbation, NYC's targeted interventions such as the Healthy Homes Program (HHP) [57] and perhaps gentrification of certain neighborhoods may have contributed to lack of significant associations observed in this study. The lower association of older housing and risk of asthma ED visits in NYC may be linked to efforts by the New York City Council to improve housing conditions in the city, such as the Underlying Conditions Program (LLP) which allows the NYC Department of Housing Preservation (HPD) to issue an administrative order to residential building owners to fix/correct underlying conditions that have caused, or are causing, a violation of the Housing Maintenance Code [58]. Other efforts at improving housing conditions in NYS include the New York State Healthy Neighborhoods Program (HNP) [59]. However, only 18 counties out of the 57 outside of NYC currently participate in the program. Research has shown that HNP is making a positive impact in improving the living conditions and health of participating communities [60,61]. Expanding HNP to include more communities may lead to lower effects of old housing in ROS.

Kerosene/fuel oil are diesel-based fuels obtained from crude oil after gasoline and other distillates have been removed [62]. Fuel oils produce the highest levels of particulate matter (PM) of any diesel type due to its high sulfur composition and combustion/firing practices [63], which in turn could contribute to adverse health effects such as asthma exacerbation. The significant risk of asthma ED visits in association with use of kerosene/fuel oil in NYC but not ROS may be related to the slow phasing out of the use of diesel fuel oil #4 in NYC. Since 2007, NYC has made strides in improving air quality for residents through governmental policy changes and programs. One such program is the NYC Clean Heat Program (CHP), wherein policy mandated transitions from residual diesel fuel oils (#6 by 2015 and #4 by 2030) to cleaner-burning alternatives are in place [64]. Zhang et al. [65] evaluated the CHP outcomes and quantified the CHP-attributable air pollution reductions between 2012 and 2016. The authors found significant reductions in air pollutants (SO₂, PM_2.5, and NO₂) per 10 buildings converted from heating oil #2, #4 and #6 based on spatial lag models [65]. However, a recent study assessing the unintended consequences of NYC's Clean Heat Transition aimed at reducing air pollution by banning residual diesel fuel oils #6 in 2015 and #4 by 2030, showed that approximately 53% of all residential residual fuel burning buildings can be found in Northern Manhattan and the Bronx [64]. This is of particular concern because of the social disadvantage and poorer health experienced by residents of Northern Manhattan and South Bronx. Despite the compliance of most buildings in NYC with the ban on residual diesel fuel oil #6, many still opt to burn residual fuel oil #4.

We leveraged several data sources to comprehensively evaluate the association between SDOH and environmental exposure metrics and the risk of asthma ED visits in the NYS Medicaid population. With over 6.3 million enrollees as of December 2015, the NYS Medicaid population is large and racially/ethnically diverse with a significant asthma burden. Asthma prevalence rates in the MMC population, which covered 4.8 million or approximately 76% of Medicaid enrollees in December 2015, ranged between 10.1 and 11.1 per 100 MMC enrollees from 2006 to 2013 [66]. We were also able to successfully geocode over 99% of enrollees' home addresses, with the overwhelming majority of addresses used in this study representing valid locations. However, home addresses for Medicaid recipients often include non-residential address including addresses for social service organizations, homeless shelters, hospitals, and charities which decreases accuracy of estimated exposure.

We used the ACS data to determine housing conditions and built environment characteristics. The building violations data were only available for NYC. There is a paucity of data on housing conditions that can be readily used for health studies [67]. Additionally, to generate adequately robust datasets, several years of data were combined, covering discordant time periods. The use of ACS data from 2008 to 2012 provided additional geographic detail on the SDOH. However, the ACS is a sample dataset for the US population and some census tracts may have large margins of error in their estimates. The ACS data is also subject to errors common in survey data, namely instrument bias and response errors. While underreporting of Medicaid ED claims and encounter data are possible, the effects are likely small as Medicaid reimbursement is determined, in part, by enrollee acuity of illness. Additionally, the NYS Medicaid population, and in particular residents of NYC are known to move often, which may lead to differential errors in exposure assessment statewide. Another limitation is lack of information on other sources of pollution, or more specific data regarding components of road traffic-related pollution for a particular location. These other sources of pollution might contribute to and/or confound the association between traffic and increased risk of asthma ED visits. Use of the land cover variable in the analysis may serve as an indicator of some of the spatial variation in other sources of pollution, but future studies could explore the use of air pollution/toxics data to capture some of these contributions. Our assessment of the external environment included green space defined as the summation of proportion of tracts classified as developed (open space), deciduous forest, evergreen forest, mixed forest, cultivated crops, grassland/herbaceous, shrub/scrub, pasture/hay, woody wetland, and emergent herbaceous wetlands. The narrow CI of the risk associated with proportion of green space suggests precise estimates of the effect of green space on asthma ED visits. However, it may also be result of lack of variability in the underlying data. Finally, we used area level variables to characterize risk, but such ecological analysis may not truly reflect risk at an individual level. Nonetheless, these variables closely reflect SDOH and environmental variables identified at an individual level [34].

5. Conclusion

We used small area analysis techniques to identify neighborhood-level risk factors for asthma exacerbations. These results do not discuss or evaluate causes of asthma in this population. Our findings are consistent with reports that environmental and social risk factors concentrated in poor and urban areas contribute to asthma exacerbations in the NYS Medicaid population. Risk maps using predicted rates and model residual may be useful in furthering asthma mitigation and traffic volume and patterns discussions, both locally and regionally.

CRediT authorship contribution statement

T.A.: formal analysis, writing–original draft, and project administration; T.Z.I.: conceptualization, methodology, writing–review & editing; C.A.: writing–review & editing, visualization, data curation relating to geographical information systems and mapping; A.P.: data curation and writing–review & editing. V.W. and S.M.: investigation and writing–review & editing. All authors contributed substantial edits to the final version of the article. All authors read and approved the final manuscript.

Declaration of competing interests

The authors have declared no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: