key: cord-0932343-1tqw8ach
authors: Pennington, Audrey F; Kompaniyets, Lyudmyla; Summers, April D; Danielson, Melissa L; Goodman, Alyson B; Chevinsky, Jennifer R; Preston, Leigh Ellyn; Schieber, Lyna Z; Namulanda, Gonza; Courtney, Joseph; Strosnider, Heather M; Boehmer, Tegan K; Mac Kenzie, William R; Baggs, James; Gundlapalli, Adi V
title: Risk of clinical severity by age and race/ethnicity among adults hospitalized for COVID-19 — United States, March–September 2020
date: 2020-12-28
journal: Open Forum Infect Dis
DOI: 10.1093/ofid/ofaa638
sha: 94b03a2cba4820264b881ad7d38958f6d092d57a
doc_id: 932343
cord_uid: 1tqw8ach

BACKGROUND: Older adults and people from certain racial and ethnic groups are disproportionately represented in COVID-19 hospitalizations and deaths. METHODS: Using data from the Premier Healthcare Database on 181,813 hospitalized adults diagnosed with COVID-19 during March–September 2020 we applied multivariable log-binomial regression to assess the associations between age and race/ethnicity and COVID-19 clinical severity (intensive care unit [ICU] admission, invasive mechanical ventilation [IMV], and death); and determine whether the impact of age on clinical severity differs by race/ethnicity. RESULTS: Overall, 84,497 (47%) patients were admitted to the ICU, 29,078 (16%) received IMV, and 27,864 (15%) died in the hospital. Increased age was strongly associated with clinical severity when controlling for underlying medical conditions and other covariates; the strength of this association differed by race/ethnicity. Compared with non-Hispanic White patients, risk of death was lower among non-Hispanic Black patients (adjusted risk ratio [95% CI]: 0.96 [0.92, 0.99]), and higher among Hispanic/Latino patients (RR [95% CI]: 1.15 [1.09, 1.20]), non-Hispanic Asian patients (RR [95% CI]: 1.16 [1.09, 1.23]), and patients of other racial and ethnic groups (RR [95% CI]: 1.13 [1.06, 1.21]). Risk of ICU admission and IMV was elevated among some racial and ethnic groups. CONCLUSIONS: These results indicate that age is a driver of poor outcomes among hospitalized persons with COVID-19. Additionally, clinical severity may be elevated among patients of some racial and ethnic minority groups. Public health strategies to reduce SARS-CoV-2 infection rates among older adults and racial and ethnic minorities are essential to reduce poor outcomes.

People of older ages and from certain racial and ethnic minority groups, such as persons of Hispanic ethnicity or Black race, are overrepresented among coronavirus disease 2019 (COVID- 19) hospitalizations and deaths in the United States [1] [2] [3] [4] [5] . These trends could be due to differences in rates of SARS-CoV-2 infection, prevalence of underlying medical conditions, clinical severity, access to care, quality of care, or a mixture of multiple factors. Results from research on risk factors for COVID-19 clinical severity consistently indicate that age is a risk factor for hospitalizations and death from COVID-19, but are inconsistent on the extent to which race and ethnicity are also independently related to these serious outcomes [6] [7] [8] [9] [10] [11] [12] .

Attributing the contributions of age, race, and ethnicity to COVID-19 clinical severity is complicated because many of the underlying medical conditions that increase risk for severe illness from COVID-19 (e.g., obesity, heart disease, type 2 diabetes), are more common with increasing age and in people from certain racial and ethnic groups [1-3, 13, 14] . To date, many of the studies on these associations that have controlled for underlying medical conditions have been limited in sample size or restricted to people from specific geographic areas. Additionally, few studies have examined clinical severity among patients with COVID-19 of Hispanic or Latino ethnicity, and, to our knowledge, none have examined whether the effect of age on COVID-19 clinical severity differs by race or ethnicity. Understanding the relationships between age, race, and ethnicity and COVID-19 clinical severity is essential to guide the public health response to the ongoing COVID-19 pandemic including decision making on priority groups for vaccines and therapeutics.

In this study, we examined the associations between age and race/ethnicity and the outcome of COVID-19 clinical severity using electronic health record data from over 180,000 hospitalized COVID-19 patients from across the United States. We aimed to determine first, the extent to which age and race/ethnicity are independently related to COVID-19 clinical severity, and second, whether the effect of age on COVID-19 clinical severity differs by race/ethnicity. 

The Premier Healthcare Database comprises de-identified hospital-based discharge data of inpatient and outpatient medical visits from non-governmental, teaching, and community hospitals [15] . It includes 25% of annual U.S. inpatient hospital admissions. Compared with hospitals that are part of the American Hospital Association, hospitals contributing data to the Premier Healthcare Database are slightly less urban (71% vs. 75%), less likely to have <100 beds (30% vs. 51%) and more likely to be in the Southern U.S. Census region (44% vs. 38%). The Premier Healthcare Database has been used widely in health research [16] [17] [18] [19] [20] .

The outcomes of interest for this study were three markers of clinical severity: intensive care unit (ICU) admission, invasive mechanical ventilation (IMV), and death. ICU admission and IMV were determined using patient billing records, and death was defined using patient discharge status.

Clinical severity outcomes were not mutually exclusive.

The primary predictors of interest were age and race/ethnicity (non-Hispanic Black, non-Hispanic White, non-Hispanic Asian, Hispanic or Latino, and other race). The "other race" category included non-Hispanic people of races other than Black, White, or Asian such as Native American, Alaska Native, Native Hawaiian, or other Pacific Islander. If Hispanic or Latino ethnicity information was unknown, people were considered not Hispanic or Latino and race information alone was used to classify race/ethnicity. Additional covariates of interest were sex, U.S. Census region, insurance type, discharge month, and underlying medical conditions. Underlying medical conditions were A c c e p t e d M a n u s c r i p t 5 classified using ICD-10-CM diagnosis codes from January 2019 up to, and including, diagnoses during a patient's first COVID-19 hospitalization. They were categorized according to a revised version of the Elixhauser Comorbidity Index [21] using the Healthcare Cost and Utilization Project Elixhauser Comorbidity Software [22] via the R package 'icd' [23] . Of the 30 conditions included in the revised Elixhauser Comorbidity Index, we excluded fluid and electrolyte disorders because they are not generally chronic and can be a clinical feature of patients with COVID-19 [24] . All conditions considered are listed in the Table 2 footnote.

We first used frequencies and percentages to describe characteristics of the patient population, the medical facilities visited by patients, and clinical severity among patients. We also compared hospitalized adults with COVID-19 to all hospitalized adults in the Premier Healthcare Database. Next, we used multivariable log-binomial regression to estimate the associations between age and race/ethnicity and the outcomes of ICU admission, IMV, and death among patients hospitalized with COVID-19 when controlling for covariates; a separate model was used for each outcome. Generalized estimating equations with an exchangeable correlation structure were used to account for potential correlation between patients at the same hospital. Modification of the effect of age on clinical severity by race/ethnicity was evaluated by stratified analyses and by assessing the addition of product terms to statistical models using generalized score tests. Due to the many underlying medical conditions considered, a backward elimination approach was used to determine which conditions to include in regression models; only conditions that impacted the main effect estimates of interest (for age and race/ethnicity) ≥10% were retained in final adjusted models [25] .

All other covariates were selected a priori. In the event of model convergence issues, the Poisson distribution with robust standard errors was used instead of the binomial distribution [26, 27] .

We completed a supplementary analysis to describe factors that might affect ICU admission and receiving IMV. We examined differences by age and race/ethnicity in discharges to hospice and presence of a do not resuscitate (DNR) order (ICD-10-CM code Z66) coded as present on admission.

All analyses were completed in SAS 9.4 (SAS Institute, Cary, NC) and R 4.02 [28] . M a n u s c r i p t 7 of the most common underlying medical conditions were hypertension (68%), diabetes (42%), and chronic pulmonary disease (23%) (data not shown). Clinical severity generally increased with age and presence of underlying medical conditions.

In adjusted regression models, after dropping underlying medical conditions that did not influence the main effect estimates of interest, the following variables were included: age, race/ethnicity, sex, insurance type, discharge month, U.S. Census region, anemia, heart disease, diabetes, obesity, renal failure, and coagulopathy. There was a strong positive association between age and clinical severity when controlling for covariates ( In stratified analyses, the effect of age on clinical severity differed some across racial and ethnic groups (Figure 1) . Increasing age had a slightly stronger effect on IMV risk among Hispanic or Latino patients than among other racial and ethnic groups. For example, the risk of IMV in the 65-74

year age group compared with the 18-39 year age group was 3.11 times higher (95% CI 2.80, 3.47) among Hispanic or Latino patients and 2.11 times higher (95% CI 1.91, 2.33) among non-Hispanic White patients (eTable 1). The effect of increasing age was also stronger on risk of death among patients of non-Hispanic Asian race/ethnicity; however estimates were imprecise. When assessing statistical interaction, there were statistically significant differences in the effect of age on ICU admission, IMV, and death by race/ethnicity (P interaction <0.05).

In the supplementary analysis of discharges to hospice and DNR orders, it was observed that among patients aged 65-74 years, 3% were discharged to hospice and 10% had a DNR order. Among patients aged ≥75 years, 9% were discharged to hospice and 31% had a DNR order; these outcomes were more common among non-Hispanic White patients (11% discharged to hospice, 36% DNR) and less common among non-Hispanic Black patients (7% discharged to hospice, 22% DNR), non-Hispanic Asian patients (7% discharged to hospice, 27% DNR), Hispanic or Latino patients (6% discharged to hospice, 21% DNR), and patients of other racial groups (7% discharged to hospice, 30% DNR).

In this study of over 180,000 hospitalized patients with COVID-19, we observed that increasing age is a strong predictor of COVID-19 clinical severity when controlling for underlying medical conditions and other factors. Compared with adults aged <40 years, the increased risk for ICU admission, IMV, and death was notable in patients in their 40s and continued to increase with age. This strong effect of age was apparent across all racial and ethnic groups examined. The findings from this large study of electronic health record data confirm results from previous studies that age is a strong risk factor for death from COVID-19 even after controlling for race/ethnicity and underlying medical conditions [7, 8, 29] .

A c c e p t e d M a n u s c r i p t 9 While risk of death was highest among patients aged ≥75 years, the risk of ICU admission and IMV peaked earlier among patients aged 65-74 years. Whether or not a patient is admitted to the ICU or mechanically ventilated is impacted by the severity of their illness, advance directives or current wishes regarding end-of-life care, and other factors. Patients in the oldest age group might be more medically fragile and less likely to survive invasive procedures such as IMV. The increase in discharges to hospice and DNR orders observed among this age group suggests that palliative care might replace more invasive procedures for some older patients with the same clinical severity.

There was variation in the association between age and clinical severity by race/ethnicity, but the pattern of this association (i.e., generally increasing effect estimates with increasing age) was similar across racial and ethnic groups.

We observed higher risks of both ICU admission and IMV among patients of Hispanic or Latino ethnicity, non-Hispanic Asian race/ethnicity, and other racial groups compared with non-Hispanic White patients. Research suggests that, compared with non-Hispanic White patients, patients of racial and ethnic minority groups are less likely to receive palliative and hospice care at the end of life [30, 31] which may lead to more use of invasive procedures. The smaller percentage of discharges to hospice and DNR orders among patients in the oldest age group of Hispanic or Latino ethnicity, non-Hispanic Asian race/ethnicity, and other racial groups than among non-Hispanic White patients suggests that the observed increased risk of ICU admission and IMV may be due to factors other than clinical severity. However, the differences in discharges to hospice and DNR orders are not expected to entirely explain the increased risk of death observed among non-Hispanic Asian, Hispanic or Latino patients, and patients of other racial groups compared with non-Hispanic White patients.

Our finding that non-Hispanic Black hospitalized patients had a lower mortality risk than non-Hispanic White hospitalized patients (RR [95% CI] 0.96 [0.92, 0.99]) when accounting for differences in age and underlying medical conditions provides more evidence of differences in mortality between these groups than previous studies that presented effect estimates in the same A c c e p t e d M a n u s c r i p t 10 direction. One study of 3,626 patients with COVID-19 in a single healthcare system in the Southern United States observed a hazard ratio of 0.89 (95% CI 0.68, 1.17) comparing in-hospital death among non-Hispanic Black patients with non-Hispanic White patients [8] . A second study of 11,210 patients from 12 states observed a hazard ratio of 0.93 (95% CI 0.80, 1.09) for Black patients compared with White patients hospitalized with COVID-19 [7] . Our findings also suggest that after adjusting for confounding there may be minimal differences in ICU admission between non-Hispanic White patients and non-Hispanic Black patients. These findings do not negate that non-Hispanic Black persons are over-represented in COVID-19 hospitalizations and deaths in the United States [3, 32] .

However, they suggest that higher national rates of hospitalization and death among persons of Black race may be due to differences in underlying rates of COVID-19 infection (due to increased exposure) or chronic medical conditions. The large, racially diverse, patient population enabled a thorough assessment of the effect of race and ethnicity on COVID-19 clinical severity. Unlike in many of the previous studies using electronic health record data to examine race and clinical severity among COVID-19 patients, this study included patients from across the United States and was able to assess more separate categories of race/ethnicity. To evaluate the potential representativeness of this sample, we compared the characteristics of this cohort to two external datasets. First, we compared deaths with COVID-19 deaths captured in death certificate data from the National Center for Health Statistics We acknowledge several imitations. First, laboratory test results for SARS-CoV-2 infection were only available for a small subset of patients so were not used to define study inclusion. Among the 12,779 patients in this cohort with results from a SARS-CoV-2 RNA test during their hospitalization, 90% tested positive. These results, combined with the possibility that individuals could also have tested positive for COVID-19 before hospital admission, suggest that the use of ICD-10-CM diagnoses to determine study inclusion resulted in a low percentage of individuals without COVID-19 in the cohort. Second, this analysis was not able to capture deaths that occurred after hospital discharge so mortality prevalence might be underestimated. Third, only four categories of race were available in the Premier database which precluded examining clinical severity separately among other racial groups. Some of the races included in the other race category, such as American Indian/Alaska Native people, have a high burden of COVID-19 highlighting the importance of future studies of these populations [34] . Fourth, several underlying medical conditions were classified using patient diagnoses, but some of them, such as obesity, are often underrepresented in diagnostic records [35] , which would prevent fully controlling for their impact on the associations of interest.

Fifth, we were unable to adjust for precise metrics of socioeconomic status leaving the potential for residual confounding by factors such as occupation, income, education, and access to healthcare.

Lastly, although this study included patients from across the United States, they were not selected in a way to make these results nationally representative.

This study demonstrates that among patients hospitalized with COVID-19, increasing age is a strong independent risk factor for COVID-19 clinical severity and suggests that there are some differences in clinical severity between racial and ethnic groups. Public health strategies to protect older adults and reduce SARS-CoV-2 infection rates among racial and ethnic minorities are essential to reduce COVID-19-associated poor outcomes. M a n u s c r i p t 12

The authors thank Sachin Agnihotri, Indira Srinivasan, David Nitschke, and Kimberly Riggle for facilitating access to these data, Greg Armstrong for his contributions to the initial conceptualization of this study, and Zhaohui Cui for his contributions to this analysis. They also thank 

The authors report no conflicts of interest.

This activity was reviewed by CDC and was conducted consistent with applicable federal law and Numeric results corresponding to this figure are listed in Table 3 and eTable 1. Note: each plot uses a different y axis.

A c c e p t e d M a n u s c r i p t 

Characteristics of persons who died with COVID-19 -United States

Coronavirus disease 2019 case surveillance -United States

Hospitalization rates and characteristics of patients hospitalized with laboratory-confirmed coronavirus disease 2019 -COVID-NET, 14 States

Centers for Disease Control and Prevention. COVID-19 hospitalization and death by race/ethnicity

Race, ethnicity, and age trends in persons who died from COVID-19 -United States

Disparities in outcomes among COVID-19 patients in a large health care system in California

Association of race with mortality among patients hospitalized with coronavirus disease 2019 (COVID-19) at 92 US hospitals

Hospitalization and mortality among black patients and white patients with Covid-19

Association of black race with outcomes in COVID-19 disease: a retrospective cohort study

Characteristics associated with hospitalization among patients with COVID-19 -metropolitan

The association of race and COVID-19 mortality

Risk factors for intensive care unit admission and inhospital mortality among hospitalized adults identified through the U.S. Coronavirus Disease 2019 (COVID-19)-Associated Hospitalization Surveillance Network (COVID-NET)

Prevalence of uncontrolled risk factors for cardiovascular disease: United States

Deptartment of Health and Human Services

Premier Healthcare Database white paper: data that informs and performs

Multidrug-resistant bacterial infections in U.S. hospitalized patients

The evolving landscape of impella use in the United States among patients undergoing percutaneous coronary intervention with mechanical circulatory support

Epidemiology of meningitis and encephalitis in the United States

Epidemiology and costs of sepsis in the United States -an analysis based on timing of diagnosis and severity level

Centers for Disease Control and Prevention. New ICD-10-CM code for the 2019 novel coronavirus (COVID-10)

Comorbidity measures for use with administrative data

Agency for Healthcare Research and Quality Elixhauser Comorbidity Software for ICD-10-CM (beta version). Healthcare Cost and Utilization Project (HCUP)

icd: Comorbidity Calculations and Tools for ICD-9 and ICD-10 Codes

Prevalence and impact of hyponatremia in patients with coronavirus disease

Hertz-Picciotto I. Methods of covariate selection: directed acyclic graphs and the change-in-estimate procedure

Easy SAS calculations for risk or prevalence ratios and differences

A modified poisson regression approach to prospective studies with binary data

R: A language and environment for statistical computing

Why does COVID-19 disproportionately affect older people?

Methods for overcoming barriers in palliative care for ethnic/racial minorities: a systematic review

Racial and ethnic differences in end-of-life costs: why do minorities cost more than whites?

COVID-NET: A weekly summary of U.S. COVID-19 hospitlaization data

COVID-19 among American Indian and Alaska Native persons -23 States

Coding of obesity in administrative hospital discharge abstract data: accuracy and impact for future research studies

A c c e p t e d M a n u s c r i p t 17 1 Presence of any of the following conditions from the Elixhauser Comorbidity Index: alcohol abuse, blood loss anemia, chronic pulmonary disease, coagulopathy, congestive heart failure, deficiency anemias, depression, complicated diabetes, uncomplicated diabetes, drug abuse, HIV/AIDS, complicated hypertension, uncomplicated hypertension, hypothyroidism, liver disease, lymphoma, metastatic cancer, obesity, other neurological disorders, paralysis, peptic ulcer disease excluding bleeding, peripheral vascular disorders, psychoses, pulmonary circulation disorders, renal failure, rheumatoid arthritis/collagen vascular diseases, solid tumor without metastasis, valvular disease, weight loss A c c e p t e d M a n u s c r i p t 19 group, ICU = intensive care unit, IMV = invasive mechanical ventilation. All models account for correlation by medical facility. Unadjusted models were fit separately for race/ethnicity and age. Adjusted models control for age, race/ethnicity, sex, health insurance, discharge month, U.S. Census region, and underlying medical conditions (anemia, heart disease [combining the Elixhauser categories of congestive heart failure and valvular disease], diabetes, obesity, renal failure, and coagulopathy).A c c e p t e d M a n u s c r i p t