Assessment of Racial/Ethnic Disparities in Hospitalization and Mortality in Patients With COVID-19 in New York City

Key Points Question Do outcomes among patients with coronavirus disease 2019 (COVID-19) differ by race/ethnicity, and are observed disparities associated with comorbidity and neighborhood characteristics? Findings This cohort study including 9722 patients found that Black and Hispanic patients were more likely than White patients to test positive for COVID-19. Among patients hospitalized with COVID-19 infection, Black patients were less likely than White patients to have severe illness and to die or be discharged to hospice. Meaning Although Black patients were more likely than White patients to test positive for COVID-19, after hospitalization they had lower mortality, suggesting that neighborhood characteristics may explain the disproportionately higher out-of-hospital COVID-19 mortality among Black individuals.


Outcomes: Testing Positive, Hospitalization, and Critical Illness
Testing was performed for patients presenting to outpatient offices, employee health offices, or the emergency department as clinically indicated. COVID-19 was defined as a positive result on realtime reverse-transcriptase-polymerase chain reaction assay. Patients were classified as hospitalized if they were admitted to NYULH after testing positive for COVID- 19. Critical illness was defined as a composite of care in the intensive care unit, use of mechanical ventilation, discharge to hospice, or death. For patients with multiple encounters, the most severe outcome that occurred across encounters was assigned.

Predictor Variables: Race and Ethnicity
Race and ethnicity were categorized using self-reported EHR data. Patients who self-reported Hispanic ethnicity were classified as Hispanic regardless of race. If ethnicity was missing, selfreported race was assigned. For patients with multiple entries for race, the entry with the most information was chosen. For example, if a patient had their race recorded as unknown in one clinical encounter and Black in another, that individual was classified as Black. Patients who reported more than 1 race (eg, White and Asian), were categorized as multiracial/other. Patients missing both ethnicity and race were classified as unknown. This coding process resulted in 6 groups: non-Hispanic White, non-Hispanic Black, Hispanic, Asian, multiracial/other, and unknown.

Covariates: Patient Characteristics, Comorbidity, and Socioeconomic Status
For patients with COVID-19 test results, we extracted EHR data including age, sex, obesity (defined by body mass index [BMI] >30 [calculated as weight in kilograms divided by height in meters squared]), history of smoking (current, former, or never), and medical comorbidity (diabetes, hypertension, hyperlipidemia, coronary artery disease, CKD, heart failure, chronic obstructive pulmonary disease or asthma, and cancer). Comorbidity was defined based on diagnostic codes in the EHR. Vital signs and laboratory results obtained upon admission were also extracted where available.
Neighborhood SES was determined by geocoding home addresses and zip codes in the EHR and matching to census tracts using ArcGIS software version 10.0 (ESRI). Confidence of the geocoding accuracy was classified, and only matches with high confidence of accuracy were maintained for this analysis; addresses were then spatially joined to census tracts. 15 The Agency for Healthcare Research and Quality (AHRQ) SES Index was calculated using American Community Survey data to assign an SES score to each patient. 16 The AHRQ SES index is a validated, weighted composite of 7 indicators, including: percentage of population within a patient's census tract in the labor force who are unemployed; percentage living below poverty level; median household income; median value of owner-occupied dwellings; percentage of residents aged 25 years or older with less than a 12th grade education; percentage who are aged 25 years or older completing 4 or more years of college; and the percentage of households that average 1 or more persons per room. 17,18 This methodology was previously used in NYC to augment EHR data. 17

Statistical Analysis
Descriptive statistics were used to characterize each racial/ethnic group of patients: COVID-test result status (ie, negative and positive), hospitalization status, and with or without critical illness (ie, received care in the ICU, on mechanical ventilation, discharged to hospice, or died). Demographic and clinical characteristics were described and then stratified by race and ethnicity. Multivariable logistic regression models were conducted to identify factors independently associated with the following outcomes: testing positive for COVID-19, hospitalization, and critical illness. Competing risk survival analyses for the outcome of mortality or discharge to hospice, where discharge home alive was the competing risk with time from first positive test as the start point, were constructed including only hospitalized patients. Patients still hospitalized as of May 13, 2020, were counted as censored. The model was fitted with the R library cmprisk 19 and the proportionality assumption was checked with the goffte library. 20 All selected exposures (race/ethnicity) and covariates (patient characteristics, comorbidities, neighborhood SES) were included based on a priori clinical significance after testing for collinearity using the variance inflation factor to ensure none had a quotient above 2. 21   For the model with testing positive for COVID-19 as outcome, all patients who had COVID-19 test results were included. For the model with hospitalization as the outcome, only patients who tested positive for COVID-19 were included. The critical illness model excluded 4 patients who were still hospitalized as of May 13, 2020, and had not met criteria for critical illness because the final outcome of hospitalization was not yet determined. Odds ratios (ORs) were obtained from each model, and confidence intervals for the ORs were bootstrapped using the approach of Venables and Ripley 22 because assuming normality of the maximum likelihood estimate to estimate confidence intervals can be biased. 23 All logistic regression models were conducted with R, version 3.6.3 (R Project for Statistical Computing). All analyses used 2-sided statistical tests and a P value <.05 was considered to be statistically significant. were discharged alive; 36.3% experienced critical illness; 24.7% died or were discharged to hospice; and 4.5% remained hospitalized as of May 13, 2020 (see eFigure in the Supplement). Table 1 shows the proportion of COVID-19-positive patients by race/ethnicity and the outcomes by race/ethnicity among COVID-19-positive patients who were hospitalized.

JAMA Network Open | Infectious Diseases
Racial/Ethnic Disparities in Hospitalization and Mortality in Patients With COVID-19 in New York City

Risk of Death After Hospitalization Among Patients Who Tested Positive for COVID-19
In unadjusted models, Black patients (hazard ratio [HR], 0.6; 95% CI, 0.5-0.8) and Hispanic patients (HR, 0.7; 95% CI, 0.6-0.9) were less likely than White patients to die or be discharged to hospice.

Discussion
We compared racial/ethnic differences in the likelihood of testing positive for COVID-19, hospitalization, critical illness, and death, and also assessed whether any differences in these outcomes were associated with comorbidity, age, sex, insurance, and neighborhood SES. Black and Hispanic patients were more likely than White patients to test positive for COVID-19. Contrary to our

Strengths and Limitations
The strengths of our study include a large sample size, a high follow-up rate, and full adjustment of our findings for insurance status, neighborhood SES, and comorbidity using data from an integrated health system. Unlike prior studies, we adjusted for effects of crowding, educational status, and unemployment. Other prior studies are largely based on publicly available data sets that do not account for individual patient comorbidity and insurance status.
This study had the following limitations. Some patients who tested positive at NYULH could have been hospitalized subsequently at another medical center. However, we have no reason to believe this would differ markedly by race or ethnicity. We did not have complete data on the final outcome of 4% of hospitalized patients, although this study had better follow up than other COVID-19 case series. 6

Conclusions
Our findings support the notion that Black and Hispanic populations are not inherently more susceptible to having poor COVID-19 outcomes than other groups and, more importantly, that if they make it to the hospital they fare as well as or better than their White counterparts. This supports the assertion that existing structural determinants-including inequality in housing, access to care, differential employment opportunities, and poverty-that remain pervasive in Black and Hispanic communities should be addressed in order to improve outcomes in COVID-19-related mortality.
Future research should explore the direct impact of structural inequities on racial and ethnic disparities in COVID-19 related hospitalization, morbidity, and mortality.