Persistence of racial/ethnic and socioeconomic status disparities among non‐institutionalized patients hospitalized with COVID‐19 in Connecticut, July to December 2020

Abstract Background COVID‐19 hospitalizations of non‐institutionalized persons during the first COVID‐19 wave in Connecticut disproportionately affected the elderly, communities of color, and individuals of low socioeconomic status (SES). Whether the magnitude of these disparities changed after the initial lockdown and before vaccine rollout is not well documented. Methods All first‐time hospitalizations with laboratory‐confirmed COVID‐19 during July to December 2020, including patients' geocoded residential addresses, were obtained from the Connecticut Department of Public Health. Those living in congregate settings, including nursing homes, were excluded. Community‐dwelling patients were assigned census tract‐level poverty and crowding measures from the 2014–2018 American Community Survey by linking their geocoded addresses to census tracts. Age‐adjusted incidence and relative rates were calculated across demographic and SES measures and compared with those from a similar analysis of hospitalized cases during the initial wave. Results During July to December 2020, there were 5652 COVID‐19 hospitalizations in community residents in Connecticut. Incidence was highest among those >85 years, non‐Hispanic Blacks and Hispanic/Latinx compared with non‐Hispanic Whites {relative rate (RR) 3.1 (95% confidence interval [CI] 2.83–3.32) and 5.9 (95% CI 5.58–6.28)}, and persons living in high poverty and high crowding census tracts. Although racial/ethnic and SES disparities during the study period were substantial, they were significantly decreased compared with the first wave of COVID‐19. Conclusions The finding of persistent, if reduced, large racial/ethnic disparities in COVID‐19 hospitalizations 2–7 months after the initial lockdown was relaxed and before vaccination was widely available is of concern. These disparities cause a challenge to achieving health equity and are relevant for future pandemic planning.


| INTRODUCTION
Coronavirus disease 2019 (COVID- 19), caused from infection with SARS-CoV-2, is a highly contagious, viral disease that can lead to severe health outcomes that may require hospitalization and intensive care. 1 According to COVID-NET estimates, at the end of 2020, the cumulative incidence rate of COVID-19 hospitalizations in the United States was 369.3 hospitalizations per 100,000 population. 2 Hospitalizations are valuable to study from an epidemiological perspective because they are more likely to accurately reflect who is getting infected with COVID-19 compared with viral testing that can be prone to testing biases.
Over the course of the pandemic, it has become evident that certain people are hospitalized with COVID-19 at disproportionately higher rates than others, including the elderly and people with underlying health conditions. 3,4 People of color, particularly Black and African American communities, have also faced an increased risk of

COVID-19 infection and hospitalization compared with non-Hispanic
White communities, 5-10 as have Hispanic and Latinx patients who in some cases have experienced increased in-hospital mortality. 7 Additionally, there is increasing evidence that low socioeconomic status (SES) is an important risk factor for hospitalization and thus, antecedent infection. 5,6 Individual-level measures of SES are not typically obtained or available through public health surveillance programs, so instead, census tract-level measures of poverty and crowding from the US Census can be linked to patients' residential addresses as a way to assess SES disparities. 11 Census-tract-based metrics have been valuable to determining the role SES plays in influenza in Connecticut and in other jurisdictions contributing to FluSurv-Net. [12][13][14] To date, we are not aware of studies that analyze COVID-19 hospitalizations and disparities solely among non-institutionalized individuals in the community (unlike congregate settings which are mostly closed environments) throughout an entire state using public health surveillance data. In Connecticut, the geographical focus of this analysis, disparities in COVID-19 hospitalizations that occurred during the state's initial "Stay Safe, Stay Home" lockdown period have been previously described but were limited to those in New Haven and Middlesex counties. 10 In this analysis, we aim to describe Connecticut's statewide trends in COVID-19 hospitalization among community members after the first, initial wave of COVID-19 and before the effect vaccinations would have on epidemiology-a time when most individuals had potential for COVID-19 exposure-in order to help determine the magnitude and persistence of disparities in COVID-19 hospitalizations. In addition, we compare the magnitude of racial/ethnic and SES disparities from the initial lockdown period 10 to those found in this analysis.

| Surveillance data
We used statewide surveillance data collected by the Connecticut Department of Public Health (DPH) to monitor COVID-19 hospitalizations beginning on July 1, 2020. Hospitalizations on or after this point were required to be reported to the DPH by hospital staff completing a case report form, which included relevant information such as the patient's age, sex, and race/ethnicity, along with the COVID-19 case classification, date of admission, whether the patient resided in a congregate setting, and the patient's residential address.
All patients' residential addresses were automatically geocoded by the DPH, assigning each its census tract identification number. For those addresses that could not be automatically geocoded, the DPH manually geocoded them. Addresses unable to be geocoded included those with PO boxes or those deemed erroneous.

| Study population
The study population included all Connecticut residents who were hospitalized at an acute care facility with COVID-19 for the first time

| Census data
Area-based SES measures of poverty and household crowding for each patient were determined by matching each patient's census tract of residence with the corresponding census tract estimate of poverty and crowding from the 2014-2018 American Community Survey (ACS) 5-Year Estimates from the US Census (https://data.census.gov/ cedsci/). Both SES measures were stratified into four levels based on precedent in Connecticut. 9,11,12 Poverty, defined as the percentage of households living below the federal poverty level, was categorized as very low (<5%), low (5% to <10%), medium (10% to <20%), and high (≥20%). Crowding, defined as the percentage of households with more than one occupant per room, was categorized as very low (<0.9%), low (0.9% to <2.5%), medium (2.5% to <5%), and high (≥5%).

| Statistical analysis
Although we described the overall epidemiology of COVID-19 hospitalizations in Connecticut, our analyses placed emphasis on patients who resided in the community, as opposed to congregate settings.
Crude and age-adjusted incidence rates of COVID-19 hospitalizations were calculated by dividing the case counts by the total population estimates for each age group, gender, race/ethnicity group, poverty level, and crowding level. Age adjustments, used to account for potential age-related confounding, were based on the 2000 US Standard Population proportions. Chi-square tests were used to compare hospitalization incidence between demographic and SES strata.
Mantel-Haenszel chi-square tests for trend were used to determine whether there were significant associations between increasing poverty and crowding levels with hospital incidence, both alone and within age, gender, and race/ethnicity groups.
Additionally, we split these data into two groups:

| RESULTS
There were 7062 first-time COVID-19 hospitalizations among Connecticut residents from July 1 to December 31, 2020. Approximately 98% (6901) of patients' residential addresses were successfully geocoded by the DPH. Of these, 294 patients were excluded from analyses because they did not meet this study's criteria and/or were missing data ( Figure 1).

| Characteristics of all patients hospitalized with COVID-19
After exclusion criteria, there were 6607 first-time COVID-19 hospitalizations between July 1 to December 31, 2020, that were confirmed with a positive molecular or antigen-based SARS-CoV-2 test and had a geocodable residential address (Table 1)

| Demographic-based disparities in hospitalization incidence
After excluding 188 (3.3%) patients in the community whose race, ethnicity, and/or gender were unknown, there were 5464 noninstitutionalized patients included in the analysis. Incidence and trends of COVID-19 hospitalization significantly varied by age and race/ ethnicity groups (Table 2). Elderly persons were disproportionately hospitalized; 75-to 84-year-old and ≥85-year-old patients were hospitalized at rates 8.4 (95% confidence interval [CI] 7.70-9.12) and 9.9 (95% CI 9.01-10.95) times higher, respectively, compared with 18-to 49-year-old patients. There were also significantly higher rates of hospitalization among patients of color, except for non-Hispanic Asian patients. The age-adjusted relative rates among non-Hispanic Black and Hispanic/Latinx cases compared with non-Hispanic White cases were 3.1 (95% CI 2.83-3.32),and 5.9 (95% CI 5.58-6.28), respectively.

| SES-based disparities in hospitalization incidence
When assessing census tract poverty and crowding levels as measures of SES, patients living in high poverty and crowding census tracts were hospitalized at an age-adjusted rate approximately three times higher (poverty 95% CI 2.88-3.30, crowding 95% CI 2.63-3.05) than patients living in very low poverty and crowding tracts ( Table 2). As census tract poverty and crowding levels increased, there were strong and statistically significant trends of increased, age-adjusted hospitalization incidence (P < 0.001 chi-square for trend for each) (Figure 2A,B).
Across increasing census tract poverty levels, there were statistically highly significant trends (P < 0.001) of increasing hospitalization within each race/ethnicity group ( Figure 3A), except for non-Hispanic Blacks (P = 0.008).. For increasing census tract crowding levels, statistically insignificant findings were only observed among non-Hispanic Black patients (P = 0.167 chi-square for trend) ( Figure 3B).

| County-level comparisons
Among non-institutionalized patients, 37.2% resided in New Haven and Middlesex Counties, whereas the remaining 62.8% resided in the other six counties (Table 3A,B). The age-adjusted incidence in New Haven and Middlesex Counties was approximately 43.6% higher than in the rest of the state; however, percentages of patients and relative incidence of COVID-19 hospitalization by demographic subgroups were comparable between these two county-based groups with some exceptions. Disparities were primarily found among patients characterized by low SES after adjusting for age. New Haven and Middlesex County patients living in high poverty and crowding were hospitalized at rates 2.5 (95% CI 2.20-2.81) and 2.1 (95% CI 1.79-2.36) times higher, respectively, than patients living in low poverty and crowding.
These disparities were stronger in magnitude for patients of the other six counties, with the high poverty and crowding groups hospitalized at similar rates of 3.4 (95% CI 3.09-3.73) and 3.4 (95% CI 3.13-3.74) times higher than the low poverty and crowding groups, respectively.  Stay Home" lockdown period from March to May, they were generally much lower. Additionally, the finding that racial/ethnic disparities in hospitalization were stronger than SES ones during the "Stay Safe, Stay Home" period 10 remained true throughout the July to December months. Of interest, with influenza hospitalizations in Connecticut, SES disparities have been generally larger than racial/ ethnic ones, and both have been lower than the ones found in this analysis. 12,14 We postulate several explanations for the disparities in COVID-19 hospitalizations found in this analysis. From March to May 2020, adult, public-facing essential workers (e.g., health aides, childcare workers, bus drivers, cashiers, factory workers, farm workers, and custodial staff), 16 disproportionately Black and Hispanic, many without personal protective equipment (PPE), were exposed occupationally, bringing infection into their home and largely segregated neighborhoods, resulting in the high racial/ethnic disparities seen not just in working age adults but also across all age groups. From July to December, with a lifting of restrictions on non-essential businesses, gatherings, and activities outside the home including camps, sports, and school, a broader spectrum of the population left their homes than during the initial lockdown, resulting in more diversity of potential exposure across age, racial/ethnic, and SES groups. In addition, PPE shortages were largely resolved. These may account in part for the smaller disparities seen July through December than found during the initial wave, a trend that was also observed nationally. 17 However, despite being smaller, substantial racial/ethnic disparities persisted and remained larger than those seen in Connecticut for influenza hospitalizations. These disparities were particularly large for mission. In addition, essential workers often face economic vulnerability not only due to low wages but also due to these jobs sometimes being part-time. Individuals who are not able to work from home may also need to use daycare for their children, as daycares have been shown to be a facilitator of SARS-CoV-2 transmission from children to their families. 18 With household crowding as an additional obstacle for isolation and quarantine practices, one new infection can be quickly amplified and reach those who are medically more vulnerable.

| Time period comparisons
The historic cause underlying the occupational, household, and community transmission dynamics is the systemic racism that has denied opportunity, generational education and wealth, and community integration to people of color, particularly non-Hispanic Black and Hispanic/Latinx patients. In addition, discrimination causes social and economic stress that is associated with a higher frequency and severity of chronic health conditions such as obesity, diabetes and heart disease that predispose to more serious COVID-19. Furthermore, people of color, especially those of low SES or living in low-income neighborhoods, may have inadequate access to care, which might result in delayed medical attention and increase an individual's chances of being hospitalized. 9 For immigrants and undocumented individuals, fear of culturally incompetent providers, language barriers, or deportation may also result in apprehension towards seeking care until their condition becomes critical. 9 Our findings have important implications for pandemic planning in Connecticut and, likely, other states. It is necessary to understand why marginalized communities (people of color, those living in poverty T A B L E 1 Characteristics of all patients with geocodable residential addresses hospitalized with laboratory-confirmed COVID-19 in CT, July to December 2020 T A B L E 2 Characteristics, crude and age-adjusted incidence, and relative rates (RR) for all non-institutionalized patients hospitalized with COVID-19 in CT, July to December 2020 tests. However, because this study relied on public health surveillance data, there were missing data components from the initial case reporting, resulting in several "unknowns" for race/ethnicity, type of residence, and ICU admission, in which further analysis could not be done. In addition, surveillance was not active in six counties, and there may have been some underreporting of hospitalizations, potentially contributing to the differences in incidence between the two regions in this analysis. The census tract-level poverty and crowding measures only characterize SES at the neighborhood level and do not necessarily apply to all individuals or households, although neighborhoods are considered a social determinant of health. Further, the ACS, from where the poverty and crowding measures were obtained, is also based on random sampling of the population, with the potential for misclassification of poverty and crowding levels in some census tracts.
Grouping them into four categories, however, likely minimized the potential for bias in misclassification. Most importantly, we were unable to exclude those living in institutions from census tract denominators, leading to community rates that were underestimates, particularly for the age groups living in institutional settings (long-term care facilities, corrections, etc.). These need to be considered when planning for the response to a future pandemic caused by a communicable virus.