key: cord-0682579-3hldvsg7 authors: Li, D.; Gaynor, S. M.; Quick, C.; Chen, J. T.; Stephenson, B. J. K.; Coull, B. A.; Lin, X. title: Unraveling US National COVID-19 Racial/Ethnic Disparities using County Level Data Among 328 Million Americans date: 2020-12-04 journal: medRxiv : the preprint server for health sciences DOI: 10.1101/2020.12.02.20234989 sha: 1e039636e06f0f342b8053f2e51c59a4abb2131e doc_id: 682579 cord_uid: 3hldvsg7 Racial and ethnic disparities in COVID-19 outcomes reflect the unequal burden experienced by vulnerable communities in the United States (US). Proposed explanations include socioeconomic factors that influence how people live, work, and play, and pre-existing comorbidities. It is important to assess the extent to which observed US COVID-19 racial and ethnic disparities can be explained by these factors. We study 9.8 million confirmed cases and 234,000 confirmed deaths from 2,990 US counties (3,142 total) that make up 99.8% of the total US population (327.6 out of 328.2 million people) through 11/8/20. We found national COVID-19 racial health disparities in US are partially explained by various social determinants of health and pre-existing comorbidities that have been previously proposed. However, significant unexplained racial and ethnic health disparities still persist at the US county level after adjusting for these variables. There is a pressing need to develop strategies to address not only the social determinants but also other factors, such as testing access, personal protection equipment access and exposures, as well as tailored intervention and resource allocation for vulnerable groups, in order to combat COVID-19 and reduce racial health disparities. We performed several additional analyses to gain further insight into the findings 1 and investigate sensitivity of the findings to model assumptions. We examined 2 correlations between county race percentage and socioeconomic and health variables 3 to understand differences in univariate and multiple regression results. We performed a 4 sensitivity analysis investigating potential non-linear associations between county 5 race/ethnicity percentage and cases and deaths by categorizing county race/ethnicity 6 percentage into quartiles. We assessed adding county-level diabetes and kidney 7 disease rates to improve modeling. To gain additional insight into our results, we 8 investigated how the mobility measure using SafeGraph county resident percentage 9 time at home 17 was associated with demographic, socioeconomic, and health factors. 10 We studied the effects of monthly average percentage time at home between March 11 and August on cumulative case and death rates. Lastly, we performed exploratory Case (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint county case rates (RR: 1.03, 95% CI: 1.01-1.05); (Fig. 2a) . The associations of county 1 American Indian, Asian, and Native Hawaiian percentage and case/death rates that 2 became insignificant did so after controlling for other race/ethnicity percentages or 3 percentage with no health insurance (Table S3-S5) . Greater White/non-White 4 segregation was associated with higher county case (RR: 1.05, 95% CI: 1.03-1.07) and 5 death rates (RR: 1.07, 95% CI: 1.03-1.10). 6 After adjusting for demographic variables and prevalence of comorbid conditions, 7 county average household size, percentage in poverty, and percentage of individuals 8 with no high school diploma were associated with higher county case and/or death 9 rates. Adjusting for demographic and socioeconomic variables, the percentages of 10 individuals with heart failure, hypertension, and stroke were associated with higher 11 county case and/or death rates, and county-level asthma and chronic obstructive 12 pulmonary disease (COPD) rate were associated with lower county case and/or death 13 rates. 14 After adjusting for demographic and socioeconomic variables and county-level 15 comorbidity disease rates, there was a stronger association between American 16 Indian/Native Alaskan percentage death rates among counties in the top quartile of 17 White/non-White segregation (RR: 1.13, 95% CI: 1.07-1.20) compared to counties with 18 less White/non-White segregation (RR: 1.06, 95% CI: 1.00-1.12). The difference in 19 death rate associations was statistically significant (effect modification: 1.07, 95% CI: 20 1.00-1.14) (Fig. 2B) . 21 Estimated cumulative case and death rates for each US county were calculated 22 from Poisson mixed models (Figs. 3A and 3B) . Of the counties presented in Figs. 3C 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint and 3D, during the study period, those comprising the Navajo Nation, Miami, and New 1 York City had the highest estimated case rates. Counties comprising New York City, the 2 Navajo Nation, and Detroit had the largest estimated death rates. Observed and model-3 based rates had good agreement, suggesting adequacy of model fit (Fig. S3) . Results 4 for every US county with full covariates are provided online. 5 We calculated the correlation of demographic, socio-economic and existing 6 medical conditions to better understand the racial disparity findings and differences 7 between multiple regression and univariate regression results. We found different 8 county race/ethnicity percentages were correlated with different socioeconomic 9 variables and disease rates (Fig. S4) percentage was negatively associated with county smoking (R 2 = 0-0.40; 95% CI: -0.43, 18 0.37) and COPD (R 2 = -0.36; 95% CI: -0.39, -0.33). 19 We investigated potential non-linear associations between county race/ethnicity 20 percentage and cases/deaths. Controlling for demographic and socioeconomic 21 variables and county disease rates, we found counties with Black/African American 22 percentages in the top quartile had greater case and death rates than those in the 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint bottom quartile (Table S6) . Similarly, counties with Hispanic/Latino percentages in the 1 top quartile had greater case and death rates than those in the bottom quartile. To gain additional insight into observed health disparities, we used mobility data 3 to study whether time spent at home due to stay-at-home orders and social distancing 4 were associated with demographic and socioeconomic variables. Controlling for all 5 other county level variables, adjusted effects for the change in county percentage time 6 spent at home were calculated for a one standard deviation increase in a county-level 7 variable (Fig. S5) . Counties with greater percentage of residents ages 20-29 years (-8 1.39; 95% CI: -1.68, -1.10) and greater no health insurance (-0.63, 95% CI: -1.11, -0.15) 9 were associated with spending less time at home. Higher county population density, 10 metro > 1 million people, metro/near metro 20,000 to 1 million people (relative to 11 nonmetro <20,000 people), and higher average household size were associated with 12 county residents spending more time at home. Including average percentage time at 13 home as a covariate also did not change adjusted cumulative case and death relative 14 risks (Fig. S6) . 15 We performed exploratory Case Fatality Rate (CFR) analyses and Infection 16 Fatality Rate (IFR) analyses (Fig. S7) . Assuming the ascertainment rates of reported 17 cases vary by state and counties modeled using fixed and random effects respectively, 18 the CFR and IFR regression analyses produced identical results (see Methods for 19 further discussion). These results had similar directions to the primary death rate results 20 but there were fewer significant associations. These analyses are likely subject to bias 21 due to several factors, such as differential underestimation of the total number of cases 22 (including asymptomatic and mildly symptomatic cases) by race and ethnicity 23,24 , 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint selection bias of subjects who have been tested (e.g. symptomatic subjects and 1 vulnerable subjects were more likely to be tested), insufficient testing capacity in many 2 areas, and variable testing rates between and within states (Methods). Our ecological study used US county-level data to investigate US health 6 disparities by studying the joint effects of demographic, socioeconomic, and health 7 variables on COVID-19 cumulative case and death rates as of 11/8/20 using 8 representative national data. We found racial and ethnic health disparities among 9 Black/African American, Hispanic/Latino, American Indian/Native Alaskan, and Native 10 Hawaiian/Pacific Islander communities persisted after adjusting for county-level 11 socioeconomic and prevalence of comorbid conditions. We also observed that 12 increased racial residential segregation increased COVID-19 case and death rates, with 13 different impacts across racial and ethnic communities. Lastly, our estimated COVID-19 14 case and death rates account for potential instability in observed rates from counties 15 with small populations or few confirmed cases/deaths and can assist in identifying 16 counties with the greatest total COVID-19 burden. Univariate associations between race/ethnic composition and COVID-19 18 outcomes were considerably stronger prior to adjustment for socioeconomic factors and (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. exposure to infected individuals within households and in communities, Personal 1 Protection Equipment (PPE) access, use of public transportation, access, quality, and 2 utilization rates of available healthcare facilities/resources, access to living resources 3 (such as a lack of access to clean water in many households of American 4 Indian/Alaskan Native communities), and health literacy. Since many of these measures 5 were either not available or quantifiable at the county-level, we were unable to control 6 for them in analyses. Counties with increased racial residential segregation experienced increased Our county-level results on racial and ethnic disparities also reinforce and 19 expand findings reported from existing individual-level studies. Single institution studies 20 in the US have also found that Black COVID-19 patients were more likely to be 21 hospitalized, enter the intensive care unit, and die 9,30 . United Kingdom Biobank and 22 electronic health record studies looking at individual-level data have also found that 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. Black and Asian individuals have an increased risk of COVID-19 hospitalization and 1 deaths after adjusting for covariates 31-33 . Our county-level results in the other domains are consistent with those from 3 several smaller scale individual-level studies. We found counties with a greater 4 proportion of individuals with ages 60+ years tended to have increased death rates. Individual-level studies also reported that older patients were more likely to develop 6 severe COVID-19 symptoms and have greater mortality rates 34 . We found county 7 average household size was associated with increased case and death rates. Household size is known to affect COVID-19 contact and transmission rates 35 . We 9 found county level rates of heart failure, hypertension, and stroke were associated with 10 case and/or death rates. These pre-existing health conditions are important biological (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. essential workers who are unable to work from home, may be more likely to take public 1 transportation, and may be more susceptible to contracting COVID-19 41 . Such areas 2 may require additional attention and interventions. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint imply causality. It is of interest to in the future conduct studies on COVID-19 disparities 1 using individual-level data with additional information on household and community 2 exposures to COVID-19 cases, occupation and work conditions, housing conditions, 3 public transportation usage, basic living resources, and COVID-19 treatments. Despite 4 these limitations, our US county-level ecological study identified elevated risks of 5 COVID-19 cases and deaths in areas with substantial non-White populations after 6 adjusting for socioeconomic and disease prevalences. Multi-faceted efforts are needed to combat the pandemic by addressing these 8 COVID-19 health disparity issues. Increased resources, such as testing priority and 9 accessible points of care, should be allocated to counties with more racial/ethnic 10 minority populations or residential racial segregation, as well as those counties with 11 more crowded housing, more elderly residents, less education infrastructure, greater 12 prevalences of hypertension, and less living resources, such as a lack of clean water. Intervention measures can include policies requiring face coverings, guaranteeing 14 workers can take paid sick leave, providing personal protective equipment to essential 15 workers, and ensuring prioritized and robust testing, tracing, and isolation infrastructure. 16 Outreach efforts can include transportation assistance, social and community support, 17 and increased accessibility and affordability of health care. Data and code availability 20 All materials and code for analysis are available on https://github.com/lin-lab. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint Acknowledgments: Resource support was provided by the Cannon cluster supported Series 19, (2020). 27 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. Washington Post 5/9/2020. 27 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. https://www.countyhealthrankings.org/explore-health-rankings/measures-data-sources/county-health-rankings-11 model/health-factors/social-and-economic-factors/family-social-support/residential-segregation-blackwhite. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint includes all other variables in the primary models. The coefficients in the fully adjusted 4 models are the same as those in Figure 2A . Bold 1.18, 1.24) 1.05 (1.02, 1.08) 1.24 (1.19, 1.30) 1.13 (1.07, 1.19 ) Hispanic (%) 1 .30 (1.27, 1.33) 1.09 (1.06, 1.12) 1.23 (1.18, 1.28) 1.08 (1.02, 1 .14) American Indian (%) 1.05 (1.03, 1.07) 1.01 (0.99, 1.04) 1.12 (1.07, 1.17) 1.10 (1.05, 1.15 .11 (1.09, 1.13) 1.03 (1.01, 1.05) 1.04 (1.00, 1.09 County-level population distribution by race/ethnicity, including Black/African American, 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. Table S1 . (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint Models included fixed effects for covariates and indicator variables for each state to 1 account for differences in state geography, testing rates, and other sources of 2 variability. County specific random effects were included to account for overdispersion. Log transformations were applied to racial percentage and population density to reduce 4 the influence of outlier values (Table S1 ). All continuous variables were scaled to have 5 mean 0 and standard deviation 1. 6 We explored if race relative risks varied by residential segregation by including 7 race percentage and White/non-White segregation interaction terms. Case and death 8 rates were fit with one model each, and results controlled for all covariates in the 9 primary models. We were interested in these interactions because we hypothesized 10 certain counties may have a lower percentage of certain racial/ethnic populations but a 11 large amount of racial segregation that would predispose these racial/ethnic 12 communities to worse COVID-19 outcomes. Predicted rates were calculated from the cumulative case and death rate models Various exploratory and sensitivity analyses were run. We compared multiple 18 regression results with univariate/minimally adjusted regression results. We obtained 19 univariate/minimally adjusted relative rates adjusting for a log population size offset, 20 state fixed effects, and county random effects. Multiple regression/adjusted relative risks 21 additionally controlled for all other covariates in the primary model (Extended Table 22 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint S2). We examined Pearson's correlation between county race percentage and 1 socioeconomic and disease rate variables. 2 We investigated potential non-linear associations between county race/ethnicity 3 percentages and cases and deaths by categorizing county race/ethnicity percentages 4 into quartiles. We partly performed this analysis to explore the absence of an adjusted 5 association between county Hispanic/Latino percentage and deaths in the primary 6 analyses. We obtained minimally adjusted relative rates adjusting for a log population 7 size offset, state fixed effects, and county random effects. Adjusted relative risks 8 additionally controlled for all other covariates in the primary model (Extended Table 9 S2), though the county race/ethnicity variables were categorized into quartiles instead of 10 continuous. We also assessed adding county-level diabetes and kidney disease 11 percentage to improve modeling. Minimally adjusted and adjusted relative risks 12 obtained from the diabetes and kidney disease analyses were similar to the above 13 analyses. Bayesian information criterion (BIC) were compared to evaluate model fit. (Table S2 ) without the county-specific random effects or log population size offset; 19 fixed effects for weekday were also added. A cubic spline basis for time with knots 20 every 14 days and an auto-regressive-1 working correlation structure were used to 21 account for serial correlation among repeated measurements over time in each county. Robust sandwich standard errors were calculated. We also included average 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint percentage time home from 3/8/20-9/30/20 as a covariate in modeling case and death 1 rates. 2 We also performed exploratory Case Fatality Rate (CFR) and Infection Fatality 3 Rate (IFR) analyses to characterize associations of death rates with covariates among 4 infected cases (elaborated on below). 5 We also explored including state testing rates as a covariate for modeling. 6 However, we found this did not change any estimated effects adjusted for demographic, 7 socioeconomic, and comorbidity variables because we already controlled for fixed state 8 effects using state dummy variables. Adding state testing rates only changed the 9 estimated state fixed effects through re-parametrization. All analyses were conducted in R. The following packages were used in 11 formatting data: data.table, dplyr. The following packages were used in formatting 12 results and creating plots: ggplot2, usmap, gridExtra, tidyverse, plyr. The following 13 packages were used in modeling: glmnet, geepack, geeM, lme4, splines. Code for 14 these analyses is available as described in the code availability section. using an offset for log(total reported cases) instead of log(population size). IFR were 23 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint calculated by dividing the number of deaths by the number of infected cases in each 1 county. Since county-specific number of total infected cases were not observed and 3 would likely be underestimated by using the numbers of reported cases, we estimated 4 the county-specific total number of infected cases by dividing the number of total 5 reported cases by a constant ascertainment rate of cases. Since the county-specific 6 testing data and ascertainment rates were not available, we modeled them using state-7 specific fixed effects and county-specific random effects and allowed the ascertainment 8 rates to vary between states and counties. To define the CFR model, assume (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The CFR/IFR results had similar directions as the primary death rate results but 10 there were few significant variables, possibly because of differential underestimation of 11 asymptomatic and mildly asymptomatic cases by race/ethnicity 23 , selection bias 12 associated with both the subjects who were tested, e.g., symptomatic subjects were 13 more likely to be tested and to test positive, large fluctuations in the numbers of tests 14 from county to county, and insufficient testing capacity. Additional data collection, such as county-level test data and race/ethnicity 16 specific case and death counts, is needed to better estimate the number of infected 17 cases by estimating county-specific ascertainment rates more accurately, in order to 18 make IFR analysis results more reliable. Limitations and Strengths 21 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; As an ecological study, there are some limitations to our study. We analyzed 1 aggregated county-level data which were subject to confounding bias and bias due to 2 data aggregation 49 . Individual level data from electronic health records or case control 3 or community studies would provide valuable information 31, 50 . Additional variables such 4 as household and community exposures to COVID-19 cases, occupation (e.g., health 5 care workers or essential workers), public transportation usage, COVID-19 symptoms, 6 living resources, and medications could improve modeling for studying disparities and 7 provide more control of confounders 51 . 8 However, there are also many strengths to our study. We used nationally 9 compiled county data from various reputable sources, and we showed that many 10 findings from previous studies and reports on the city and state level generalize to the 11 national level. Our analyses also adjust for regional and state-to-state variation, and 12 while many studies adjust for demographic and either socioeconomic or health 13 variables, we present results jointly adjusting for all three domains, as well as county 14 resident mobility. Many other studies have focused solely on Black/African American or a more 16 general non-White ethnicity, but we also incorporate and study Hispanic/Latinos, 17 Asians, and American Indians/Native Alaskan effects separately. We also explore the 18 effects of residential racial segregation indices of dissimilarity calculated from US 19 Census tract level data. We also studied the effects of race/ethnicity and other 20 demographic variables, socioeconomic and comorbidity factors on social distancing 21 using mobility-based time at home measures. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint Case and death counts from USA Facts and race/ethnicity percentages from US Census estimates were available for these counties 3,113 counties COVID-19 data repository contained information for these counties 29 counties COVID-19 data repository did not contain information for these counties 2,990 counties Had complete data for all covariates in analyses 123 counties Missing either prevalence of asthma, cancer, chronic obstructive pulmonary disease, heart failure, or stroke All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint 1.12 (1.10, 1.14) 5 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20234989 doi: medRxiv preprint Hospitalization Rates and Characteristics of Patients Hospitalized with Laboratory-Confirmed 18 Coronavirus Disease 2019 -COVID-NET, 14 States The COVID-19 Pandemic: a Call to Action to Identify and Address Racial 21 and Ethnic Disparities