Article Text

Download PDFPDF

Unequal burdens: assessing the determinants of elevated COVID-19 case and death rates in New York City’s racial/ethnic minority neighbourhoods
Free
  1. D. Phuong Do1,
  2. Reanne Frank2
  1. 1 Public Health Policy & Administration, Zilber School of Public Health, University of Wisconsin-Milwaukee, Milwaukee, Wisconsin, USA
  2. 2 Department of Sociology, Ohio State University, Columbus, Ohio, USA
  1. Correspondence to D. Phuong Do, University of Wisconsin-Milwaukee; dphuong{at}uwm.edu

Abstract

Background The disproportionate burden of the COVID-19 pandemic on racial/ethnic minority communities has revealed glaring inequities. However, multivariate empirical studies investigating its determinants are still limited. We document variation in COVID-19 case and death rates across different racial/ethnic neighbourhoods in New York City (NYC), the initial epicentre of the U.S. coronavirus outbreak, and conduct a multivariate ecological analysis investigating how various neighbourhood characteristics might explain any observed disparities.

Methods Using ZIP-code-level COVID-19 case and death data from the NYC Department of Health, demographic and socioeconomic data from the American Community Survey and health data from the Centers for Disease Control’s 500 Cities Project, we estimated a series of negative binomial regression models to assess the relationship between neighbourhood racial/ethnic composition (majority non-Hispanic White, majority Black, majority Hispanic and Other-type), neighbourhood poverty, affluence, proportion of essential workers, proportion with pre-existing health conditions and neighbourhood COVID-19 case and death rates.

Results COVID-19 case and death rates for majority Black, Hispanic and Other-type minority communities are between 24% and 110% higher than those in majority White communities. Elevated case rates are completely accounted for by the larger presence of essential workers in minority communities but excess deaths in Black neighbourhoods remain unexplained in the final model.

Conclusions The unequal COVID-19 case burden borne by NYC’s minority communities is closely tied to their representation among the ranks of essential workers. Higher levels of pre-existing health conditions are not a sufficient explanation for the elevated mortality burden observed in Black communities.

  • Epidemics
  • Ethnicity
  • Health inequalities
  • Neighbourhood/place
  • Social inequalities

This article is made freely available for use in accordance with BMJ's website terms and conditions for the duration of the covid-19 pandemic or until otherwise determined by BMJ. You may use, download and print the article for any lawful, non-commercial purpose (including text and data mining) provided that all copyright notices and trade marks are retained.

https://bmj.com/coronavirus/usage
View Full Text

Statistics from Altmetric.com

INTRODUCTION

New York City (NYC), the initial epicentre of the U.S. coronavirus outbreak, has experienced nearly a quarter million COVID-19 cases and over 22 000 deaths in the five months since the city recorded its first positive confirmed case.1 Nearly as quickly as the pandemic unleashed its impact on the region, the unequal burden of the disease became clear. In a pattern that has now been repeated across the country, data from New York suggested that racial/ethnic minority communities were disproportionately affected, both in terms of COVID-19 cases and mortality.2

Evidence of COVID-19’s unequal impact on racial/ethnic minority communities is continuing apace,3 yet the reasons for these disparities rest on a much smaller evidence base.4–6 From a fundamental cause perspective, racial/ethnic health disparities are caused by inequalities in social and economic conditions that put certain racial/ethnic communities at higher risk for negative health outcomes, including mortality.7 8 In the case of COVID-19, a growing number of commentaries underscore the potential role of societal inequalities in contributing to community-based disparities.9–11 The ranks of essential and front-line workers, which disproportionately consist of members of racial/ethnic minority groups, are among the most well-cited concerns.12 Unequal representation in these ‘telecommute unfriendly’ occupations is frequently proposed as the key factor in contributing to higher prevalence of COVID-19 in minority communities.13 Others have pointed to the potential role of the geography of pre-existing health conditions, such as diabetes and heart disease, which are disproportionately present in lower income and racial/ethnic minority communities.14 Relatedly, unequal access to quality healthcare has been cited as a factor, with disadvantaged racial/ethnic minority communities underserved both in terms of quality and access.2 15

At present, however, there remains a significant imbalance between the large number of commentaries theorising on the reasons minority communities are being disproportionately impacted by COVID-19 and the number of empirical studies assessing the links between hypothesised determinants and COVID-19 outcomes. A major obstacle to understanding disparities in COVID-19 burden has been a lack of data. Individual data on COVID-19 infection and death linked to an individual’s socioeconomic and health characteristics are still not routinely available in the United States.16 However, analysis of group-level data can provide meaningful insights by identifying the characteristics of neighbourhoods that make specific communities more vulnerable to, and at higher risk for, adverse COVID-19 outcomes.

For example, concentrated poverty, alongside the impact of a lifetime exposure to the noxious consequences of racism, has been argued to play a key role in driving the elevated COVID-19 burden observed within minority communities.17 18 Conversely, concentrated affluence might be reflective of a host of salutary features, including access to high-quality healthcare facilities and the ability of residents to engage in protective actions (eg, social distancing through working from home and ordering groceries and other essential items for delivery). Importantly, neighbourhood poverty and affluence may tap into different risk and protective features, as the absence of poverty does not necessarily indicate affluence.

Accordingly, we investigate the roles of neighbourhood poverty and affluence in shaping community-level racial/ethnic disparities in both COVID-19 cases and deaths in NYC. We further examine the extent to which these associations can be explained by neighbourhood proportion of residents with pre-existing health conditions and essential workers, respectively.

METHODS

Data

Analyses are based on aggregate data at the ZIP Code Tabulation Area (ZCTA) geographic level, our operationalisation of neighbourhoods/communities. Created by the Census, ZCTAs are representations of United States Postal Service ZIP code service areas.19

COVID-19 case and death data are from the NYC Department of Health and Mental Hygiene’s repository on Github.1 Demographic and health data are from the 2014–2018 American Community Survey’s (ACS’s) 5 year estimates and the Centers for Disease Control and Prevention’s (CDC’s) 500 Cities Project 2019 data, respectively.20 21 Both NYC’s COVID-19 data and the ACS’s demographic data were released at the ZCTA level. The CDC’s 500 Cities data on neighbourhood prevalence of health conditions, released at the Census tract level, was converted to ZCTA-level data via the Census’ 2010 ZCTA crosswalk file.22 Because the NYC Health Department combined ZCTAs with smaller populations to other ZCTAs in their COVID-19 data file, we replicated the ZCTA modification for the ACS and CDC data using the NYC Health Department’s crosswalk file.23 All data were then merged via (modified) ZCTA identifiers. In most cases, ZCTAs are equivalent to postal ZIP codes; thus, we henceforth refer to ZCTAs as ZIP codes due to higher familiarity with the latter term.

Outcome and predictor variables

Our outcome variables reflect NYC’s COVID-19 ZIP code level confirmed case and death counts as of 19 July 2020. COVID-19 deaths are cumulative counts reported since 11 March 2020, the date of the first confirmed COVID-19 death. Because the neighbourhood determinants of nursing home deaths may differ from those in the general community, we subtracted the number of nursing home deaths in each ZIP code from the total death counts.24

Our key predictor variable of interest is neighbourhood racial/ethnic composition, divided into four categories. Majority non-Hispanic White, majority non-Hispanic Black and majority Hispanic neighbourhoods are defined as neighbourhoods with at least 50% of the specified racial/ethnic group. The remaining neighbourhoods are classified as ‘Other-type’, which are composed of 71% racial/ethnic minority residents, on average. In our study, we focus on majority non-Hispanic Black and Hispanic neighbourhoods (henceforth Black and Hispanic neighbourhoods for brevity) and their case and death burdens as compared to majority non-Hispanic White neighbourhoods (henceforth White neighbourhoods).

We investigate the role of four neighbourhood factors: percentage poor, percentage affluent, percentage essential worker and percentage with pre-existing health conditions. Percentage poor is the proportion of residents with an income that falls below the federal poverty level. Percentage affluent reflects the proportion of households with incomes above $200 000, representing the top 5% to 8% of the U.S. household income distribution in 2018.25 Percentage essential worker is defined as the proportion of workers who are in occupations that were deemed ‘essential’, including healthcare practitioners, healthcare service support, community and social services, food preparation, protective services (eg, law enforcement), natural resources (eg, farming), construction, maintenance, production, transportation and material moving occupations. Percentage with pre-existing health condition is derived by averaging the proportion of residents who have any of the health conditions that have been listed by the CDC as having the strongest and most consistent evidence of being associated with severe illness from COVID-19, including cancer, chronic kidney disease, chronic obstructive pulmonary disease, obesity and diabetes.26

Neighbourhood controls include percentage of residents aged 65 years old or older and population size. The latitude and longitude of Zip code centroids were also included to explicitly account for possible spatial autocorrelation. Lastly, models of cases and mortality were adjusted for COVID-19 testing rate and COVID-19 case rate, respectively, to adjust for variation in testing rates by place.27

Statistical analysis

We estimate a series of negative binomial regression models for both neighbourhood case and death counts.

For case counts, we first estimate a base model that controls for the percentage of residents aged 65 years and older, COVID-19 testing rate, latitude, longitude and neighbourhood population size. We then separately add neighbourhood percentage poor, affluent and essential worker to examine the role of each factor in explaining neighbourhood racial/ethnic disparities observed in the base model. To assess their independent associations, all three neighbourhood characteristics are included concurrently in the final model.

We employ a similar strategy for death counts. The base model adjusts for percentage of residents that are 65 years and older, COVID-19 case rate, latitude, longitude and neighbourhood population size. Neighbourhood percentage poor, affluent and residents with pre-existing health conditions are added as covariates separately and then included concurrently.

Moran’s I tests on residuals were conducted to detect for residual spatial autocorrelation and were found to not be statistically significant.

RESULTS

Bivariate descriptives

Table 1 presents neighbourhood characteristics by neighbourhood racial/ethnic composition. Black and Hispanic neighbourhoods in NYC have borne the brunt of the COVID-19 pandemic, with disproportionately higher case and death rates, compared to White neighbourhoods. Notably, death rates are approximately twice as high in Black and Hispanic neighbourhoods. Hispanic neighbourhoods are the hardest hit with respect to COVID-19 infections. The prevalence of risk factors is also higher in minority neighbourhoods; Black and Hispanic neighbourhoods are poorer, less affluent and have a higher prevalence of prior health conditions and percentage of workers in essential industries. On the other hand, predominately White neighbourhoods tend to have an older age demographic, particularly compared to Hispanic neighbourhoods.

Table 1

Zip code level characteristics of New York City by racial/ethnic composition

Maps of NYC depicting neighbourhood racial/ethnic composition and their corresponding COVID-19 case and death rates (top and bottom panels, respectively) are presented in figure 1. Consistent with table 1, Black and Hispanic neighbourhoods tend to have higher case and death rates (darker shades of grey) than White neighbourhoods, which have the lowest case and death rates. Other-type neighbourhoods have a wide distribution of cases, with some having relatively low and others having relatively high case and death rates.

Regression models

Cases

Figure 1

Neighbourhood racial/ethnic composition and COVID-19 case and death rates in New York City.

Table 2 presents the results from the ZIP code-level COVID-19 case count models. Significant racial/ethnic disparities across NYC neighbourhoods are evident in the incidence rate ratios presented in Model 1. Black, Hispanic and Other-type neighbourhoods have approximately 50% to 110% higher case rates than White neighbourhoods, with the greatest disparity between Hispanic and White neighbourhoods. Models 2–5 assess the extent to which neighbourhood poverty, affluence and proportion of essential workers explain the observed disparities, separately and then concurrently.

Table 2

Negative binomial regression results for COVID-19 cases in New York City, 19 July 2020

Neighbourhood poverty is predictive of COVID-19 neighbourhood case rates in NYC, net of baseline control variables; a 10-percentage point increase in poverty is associated with a 6% increase in case rates. With its addition to the model, disparities in cases across neighbourhoods are reduced modestly. Neighbourhood affluence is associated with significantly lower case rates (IRR=0.80, 95% CI=[0.72, 0.88]). Moreover, after adjusting for neighbourhood affluence, disparities across neighbourhood type are reduced substantially. Compared to White neighbourhoods, the excess COVID-19 case rates for Black and Hispanic neighbourhoods are attenuated, respectively, from 63% (95% CI=[1.42, 1.86]) to 18% (95% CI=[1.01, 1.37]) and 110% (95% CI=[1.81, 2.43]) to 34% (95% CI=[1.11,1.62]). The proportion of essential workers in a ZIP Code is also found to be highly predictive of neighbourhood case rates (IRR=1.29, 95% CI=[1.25,1.33]) and explained the entirety of disparities in case rates across different racial/ethnic neighbourhoods. With the inclusion of all three risk and protective factors concurrently, neighbourhood affluence is no longer predictive of COVID-19 case rates whereas the extent of essential workers in a community remains a strong positive predictor. Interestingly, neighbourhood poverty becomes protective after accounting for both neighbourhood affluence and essential workers.

Deaths

Similar to the differences observed in COVID-19 case rates across NYC neighbourhoods, minority communities are also characterised by higher death rates, with Hispanic and non-Hispanic Black communities exhibiting over 40% higher death rates than those observed in White neighbourhoods (table 3, Model 1). Again, neighbourhood poverty is associated with elevated death rates (IRR=1.18, 95% CI=[1.13, 1.25], table 3, Model 2) and neighbourhood affluence is protective (IRR=0.84, 95% CI=[0.72,0.97]). Higher prevalence of pre-existing health conditions, on the other hand, is associated with a significantly higher death rate (IRR=1.88, 95% CI=[1.37, 2.58]). Even after accounting for differences in neighbourhood poverty, affluence and health conditions, Black neighbourhoods display death rates that are still 28% higher than those observed in White communities. In the case of Hispanic neighbourhoods, their higher death rates are attenuated in the final model and no longer statistically significant.

Table 3

Negative binomial regression results for COVID-19 deaths in New York City, 19 July 2020

DISCUSSION

For COVID-19 case rates, the NYC experience strongly implicates two key factors; the level of neighbourhood affluence and the proportion of essential workers in a community. Affluent neighbourhoods are uniquely positioned to shield themselves from virus exposure, perhaps because higher income individuals tend to work in occupations that are more amendable to stay-at-home orders, including jobs that can be performed remotely.28 The high negative correlation between neighbourhood affluence and essential workers (Pearson’s correlation=−0.86) as well as the complete attenuation of the protective association between neighbourhood affluence and case rates after accounting for essential worker presence, are consistent with residents’ ability to stay at home as a primary mechanism through which neighbourhood affluence works to reduce exposure to COVID-19. The disproportionate number of minorities among the ranks of essential workers places their neighbourhoods at higher risk and completely accounts for the elevated case rates observed among Black, Hispanic and Other-type communities relative to White ones.

Results of models investigating the relationship between neighbourhood factors and the disproportionate COVID-19 deaths in majority minority communities in NYC paint a less clear picture. Even when neighbourhood poverty, affluence and prevalence of pre-existing health conditions are accounted for, they remain insufficient to fully explain the excess COVID-19 mortality burden in Black neighbourhoods. Further, although not statistically significant, the estimated mortality disparity for Hispanic neighbourhoods is still substantively high (IRR=1.13, 95% CI=[0.96, 1.33]). This is in contrast to the estimates for the final COVID-19 case model, which had risk ratios for Hispanic neighbourhoods close to one. Hence, failure to detect statistically significant disparities in COVID-19 deaths for Hispanic neighbourhoods when all three neighbourhood factors are included may be due to lack of power.

We performed sensitivity analyses with respect to our categorisation of neighbourhoods to assess the robustness of our inferences across different methods in classifying communities. Our alternative delineation of neighbourhoods includes (1) >60% specific racial/ethnic composition and (2) one SD greater than NYC’s ZIP code mean of a specific racial/ethnic group. Both of these alternative definitions effectively increase the number of zip codes classified as ‘Other-type’. Inferences for the three predictor variables of interest remain the same and all neighbourhood disparities in case rates were again fully explained after accounting for the proportion of workers who worked in essential occupations. For models estimating COVID-19 death rates, the unexplained disparity for Black neighbourhoods remained substantively and statistically significant in all models for both alternative specifications. However, there were two differences in inferences for neighbourhood disparities. In contrast to results in our main model, the disparity in Other-type neighbourhoods (IRR=1.17, 95% CI=[1.04,1.33]) and Hispanic neighbourhoods (IRR=1.26, 95% CI=[1.05,1.52]) remained statistically significant after adjusting for all three community factors in alternative neighbourhood specification one and two, respectively.

One additional pattern deserves further comment. Although neighbourhood poverty is initially associated with higher case rates when it is the only neighbourhood predictor, it becomes protective after adjusting for community affluence and proportion of essential workers. The reasons why neighbourhood poverty, net of these two factors, might confer protection from coronavirus infection are not clear. We performed sensitivity analyses to test the robustness of the finding. First, we estimated different specifications, including categorising neighbourhood poverty into different discrete levels to assess whether the protective association was an artefact of a linearity constraint. Second, to minimise bias due to extrapolation across regions of non-overlap, we restricted the full model to ranges of overlap in neighbourhood poverty across neighbourhood type. Inferences remained unchanged. More detailed information on social distancing and mask usage might be helpful in trying to explain this pattern, which is completely at odds with how neighbourhood poverty typically operates in shaping health outcomes.

Limitations

There are limitations to this study. Variation in undercounts in COVID-19 case and death data across areas are well known.29 Although our study minimises these issues by focusing only on one region (New York City) and adjusting for zip code testing rates, residual variation in undercounts may still remain.

As in all classifications of racial/ethnic neighbourhoods, the criteria for how a community qualifies for a specific characterisation are arbitrary. Sensitivity analyses using different definitions of racial/ethnic neighbourhoods reveal that inferences for the extent to which neighbourhood poverty, affluence and prevalence of pre-existing conditions explain excess COVID-19 deaths in Hispanic neighbourhoods are somewhat sensitive to how neighbourhoods are defined. This sensitivity, in addition to the estimated risk ratio for Hispanic neighbourhood being substantively higher than one in the final COVID-19 mortality model, suggests that consideration of other neighbourhood factors may be required before elevated COVID-19 deaths in Hispanic neighbourhoods can be decidedly accounted for.

Our neighbourhood measures were limited to those that were available from the Census and CDC. Hence, due to unavailability, our list of health conditions omitted sickle cell disease and solid organ transplantation, two conditions included under the CDC’s list of health conditions that had the most consistent evidence of severe COVID-19 outcomes.26 In alternative specifications, we added additional health conditions, including hypertension, smoking, asthma and liver disease. Inferences were unchanged. Additionally, we were unable to account for neighbourhood-level variation in healthcare access and quality. Significant differences exist across New York communities in the availability of high-quality care and these differences may be a factor in understanding the mortality disparities we document here.30 Lastly, ecological analyses preclude any inferences to relationships at the individual level. However, our analysis focuses on why predominately minority communities in NYC have experienced disproportionately higher case and death rates, which lends itself to an ecological model. Whether and how the explanatory factors we have identified in our analysis apply to individual-level relationships is a separate question that will shed further light on why members of certain groups are bearing a disproportionate burden of COVID-19.

Public health implications

Our results provide evidence of the high price being paid by working-class minority communities for providing essential services early in the pandemic. Affluent communities have been much more successful in limiting COVID-19 transmission, largely due to their lower representation among the ranks of essential workers. As the country moves forward in reopening, public health and policy interventions need to be focused on mitigating the risk of exposure and infection for communities where essential workers reside. A central component must involve robust worker protection policies. At the federal level, worker protection falls under the jurisdiction of the Occupational Health and Safety Administration (OSHA) which, as of July 21, had issued only three citations on a total 7943 COVID-19-related complaints filed.31 Our analysis suggests that stronger labour protection policies should be targeted to reduce minority communities’ excess exposure to COVID-19. As the pandemic’s epicentre shifts to other areas of the country, we must work hard to avoid continued repeat performances of disproportionate harm.

What is already known on this subject

  • Early on, as the COVID-19 pandemic began to unfold in communities across the U.S., it became clear that racial/ethnic minority communities were bearing a disproportionate share of the COVID-19 burden, both in case counts and death rates. There have been a large number of published commentaries hypothesising on the possible explanations for the documented racial/ethnic disparities in COVID-19, but, so far, a dearth of multivariate empirical assessments. The need for empirical investigations into the factors contributing to COVID-19-related racial/ethnic disparities is urgent as only they can help us avoid the very real risk of pathologising precisely those populations experiencing the brunt of negative outcomes.

What this study adds

  • We focus on New York City, the initial epicentre of the U.S. pandemic, and find elevated COVID-19 case rates in minority communities are fully accounted for by their high ranks of essential workers. Our results are suggestive of a high price being paid by working-class minority communities for providing essential services early in the pandemic. Contingent on infection, our analysis demonstrates that minority communities in New York City have experienced unacceptably elevated COVID-19 mortality rates, particularly its majority Black neighbourhoods, where excess risk persisted net of pre-existing conditions, poverty and affluence levels. Given our findings that link prevalence of essential workers to disproportionate exposure to COVID-19 in minority communities, we end by urging more attention be given to labour protection policies which so far have been only weakly enforced in this pandemic.

REFERENCES

View Abstract

Footnotes

  • Contributors DPD designed and directed the project and carried out the analysis and directed the writing of the manuscript. RF assisted with the interpretation of the results and the writing of the manuscript.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Map disclaimer The depiction of boundaries on the map(s) in this article does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. The map(s) are provided without any warranty of any kind, either express or implied.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval The data for the analysis are all publicly available de-identified data at the ZCTA-level (from the U.S. Census Bureau, the Centers for Disease Control and the NYC Department of Health and Mental Hygiene’s repository on Github).

  • Provenance and peer review Not commissioned; internally peer reviewed.

  • Data availability statement All data relevant to the study are included in the article.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.