Who Gets Sick From COVID-19? Sociodemographic Correlates of Severe Adult Health Outcomes During Alpha- and Delta-Variant Predominant Periods: September 2020–November 2021

Abstract

While COVID-19 has widely affected US communities, some sociodemographic groups have been more severely affected. Prior studies have found older adults and males showing a disproportionate risk of severe outcomes [1–4]. Elevated case rates, hospitalizations, and mortality have been reported for persons who are American Indian and Alaska Native (AI/AN), Black non-Hispanic (Black), and Hispanic or Latino (Hispanic) as compared with those who are White non-Hispanic (White) [2, 5–8], and areas with greater social vulnerability had higher rates of COVID-19 cases and mortality during certain pandemic periods [9, 10].

Previous reports of differences in COVID-19 morbidity and mortality have evaluated risks per case [1, 4]. However, infections may not always be reported, and some communities may test less [11]. US seroprevalence studies have estimated 1 report per 2 or 3 infections depending on calendar period and census region [11, 12]. Seroprevalence can estimate the overall proportion of people infected with SARS-CoV-2 [13]. By combining seroprevalence data with additional surveillance data, this article examines the relationship between sociodemographic variables and risk of COVID-19 outcomes per infection. Specifically, it assesses the likelihood per SARS-CoV-2 infection of case reporting, emergency department (ED) visits, hospitalizations, and deaths across 6 sociodemographic factors: race and ethnicity, sex, age, county-level social vulnerability, county-level urban-rural classification, and region.

METHODS

Data sets were merged from 6 surveillance systems for persons aged ≥18 years in matched counties. Methods and statistical considerations for each of these systems are described in Supplementary Table 1. Population denominators were drawn from the 2020 census to estimate the proportion experiencing each event. Individual race and ethnicity, sex, and age were directly collected by surveillance systems. Race and ethnicity were categorized by the alternative mutually exclusive method described by Yoon et al to maximize available data [14]. Social Vulnerability Index (SVI) [15], urban-rural classification [16], and US Census region [17] were calculated by county of residence. Except as specified, the geographic catchment was based on the maximum overlap of the surveillance systems and included 1490 US counties from 50 states and the District of Columbia, representing 71% of the US population (Supplementary Table 2, Supplementary Figure 1).

Data were analyzed for infections occurring from 16 September 2020 to 15 November 2021. The starting date maximized the stability of data source methods and collection areas. The end point fell in the trough between the Delta- and Omicron-variant predominant periods in the pandemic curve. The Omicron-predominant period was not included due to higher reinfections [18]. To assess differences by pandemic period, a cut point of 15 June 2021 was employed, as the midpoint of the month when the pandemic shifted from Alpha- to Delta-variant predominance. Alpha and other pre-Delta variants predominated during the Alpha period from 16 September 2020 to 15 June 2021 and the Delta period from 16 June 2021 to 15 November 2021 [19]. Change in risk between Alpha and Delta periods was calculated by risk difference, and change in relative risk (RR) was calculated with the ratio of RRs. To generate the analogous periods applied to ED visits, hospitalizations, and deaths, 3, 3, and 15 days, respectively, were added to the infection date periods to account for the median time between events based on Centers for Disease Control and Prevention (CDC) planning values [20].

Given that most infections are neither diagnosed nor reported as cases [12], it was important to have a separate estimate of the number of infections. Incidence of infections was estimated by increases in cross-sectional seroprevalence from monthly cross-sectional seroprevalence of blood donation specimens from a national blood donor study, weighted to account for demographic differences between the blood donor population and the overall US population [12]. Only infection-related seropositivity is described, as differentiated from vaccine-related seropositivity through a previously published testing algorithm [21].

COVID-19 cases (confirmed or probable) were obtained from aggregate and line-level surveillance systems [22, 23]. For SVI, urban-rural classification, and census region analyses, aggregate case reports collected by the CDC were used [24]. For race and ethnicity, sex, and age, which were unavailable in aggregate case reports, individual cases were obtained from the CDC's case-based surveillance system as reported by jurisdictional health departments. Following previously published methods [25, 26], data were restricted to health jurisdictions where individual case reports made up ≥30% of total aggregate case reports, and ≥70% of observations had complete race and ethnicity data; this included 1329 counties in 45 states and the District of Columbia, making up 51% of the US population (Supplementary Figure 2).

ED visits with a COVID-19 diagnosis were obtained from the National Syndromic Surveillance Program based on the following billing codes: U07.1 or J12.82 (ICD-10-CM) or 840539006, 840544004, or 840533007 (Systematized Nomenclature of Medicine) [1, 27].

Hospital data were based on laboratory-confirmed COVID-19, which was also the reason for admission in >80% of patients [28]. All analyses used the Unified Hospital Data Surveillance System [22], except for those including race/ethnicity and sex, which used hospital data drawn from COVID-NET (Coronavirus Disease 2019–Associated Hospitalization Surveillance Network) [29]. Analyses comparing hospitalizations from COVID-NET with infections were restricted to the 95 counties in 14 states, covering 9% of the US population in which both these data sources were available. Hospitalized patients did not necessarily visit the ED. Persons who visited the ED and were hospitalized would be expected to be included in both data sets.

COVID-19–associated deaths were deaths for which COVID-19 was listed on the death certificate as the confirmed or presumed underlying cause or the contributing cause of death (U07.1; ICD-10-CM) in the National Vital Statistics System [30]. Data from 2020 are final, and data from 2021 are provisional. To assess the impact of health disparities, deaths averted were estimated for the hypothetical condition in which the fatality rate for all sociodemographic subgroups was no higher than the age-standardized fatality rate for a selected comparison subgroup.

These activities were reviewed by the CDC and conducted consistent with applicable federal law and CDC policy. (See e.g., 45 C.F.R. part 46, 21 C.F.R. part 56; 42 U.S.C. §241(d); 5 U.S.C. §552a; 44 U.S.C. §3501 et seq).

Outcomes per 100 000 infections were compared across sociodemographic strata with RRs and 95% CIs. The number of infections occurring during a given period was estimated by taking the observed seroprevalence during the month whose midpoint fell on the end date of the period under analysis and subtrating the observed seroprevalence during the month whose midpoint fell on the start date, and then multiplying by the underlying census population. The complex sampling design of weighted seroprevalence data was handled by using previous methods [21]. All other data sources were subject to random variation with an underlying Poisson distribution. To achieve SEs across sums and ratios, SEs were estimated with propagation of error–defined linear formulas while assuming independence [31]. Resulting error estimates were combined with the critical value from the 95% CIs of a standard normal distribution to achieve final CIs. Given the strong relationship between age and COVID-19 morbidity and mortality, statistics for all sociodemographic variables, with the exception of age, were directly standardized against the 2000 US Census age distribution [32]. To maintain consistency with prior publications for each data source, complete case analysis was used for seroprevalence data; mortality and census data underwent imputation by the National Center for Health Statistics and the US Census Bureau, respectively, with methods such as characteristic or “hot deck” imputation to create complete data sets without substantially altering variability [33, 34]; for other data sources, all available cases were included in each analysis. All analyses were conducted in Python version 3.9.7 (Python Software Foundation) with the Pandas and NumPy libraries alongside Excel (Microsoft Corporation).

RESULTS

Data Set Description

Data on 312 765 positive results among 1 900 595 total specimens, 20 167 253 cases, 2 254 758 ED visits, 2 067 947 hospitalizations, and 348 835 deaths were analyzed with respect to an underlying population of 183 725 389 persons aged ≥18 years in our primary catchment area (Supplementary Table 3). For analyses concerning data on race and ethnicity, sex, and age, our data set contained 19 468 700 case reports, and for analyses based on hospitalization data on race/ethnicity and sex, our data set contained 183 906 hospitalizations. Characteristics of counties included in and excluded from the analysis based on the catchment areas (Supplementary Figures 1 and 2) are described in Supplementary Table 4. Included counties tended to have lower social vulnerability and higher urbanicity. Counties in the Northeast were most likely to be included (74.2%), whereas Southern counties were least likely (37.0%). Missingness of sociodemographic data by data source and variable is described in Supplementary Table 5. Data were <9% missing for all sources and variables, except race and ethnicity for case report data, where 32.4% were missing.

RR of Infection and COVID-19 Outcomes per Infection

In unadjusted analysis, an estimated 21.7% of the population was infected with COVID-19 from 16 September 2020 to 15 November 2021. Of those, 50.7% were reported as cases, 5.7% visited the ED, 5.2% were hospitalized, and 0.9% died. In the age-standardized analyses given in Tables 1 and 2 and Figure 1, Hispanic persons were more likely than White persons to be infected (RR, 1.11; 95% CI, 1.01–1.21), whereas non-Hispanic Asian or Pacific Islander (Asian/PI) persons were less likely (RR, 0.50; 95% CI, .41–.60). Per infected person, AI/AN, Black, and Hispanic persons were more likely than White persons to visit the ED or die, and Hispanic and Black persons were about twice as likely as White persons to be hospitalized. Asian/PI persons were also more likely to die per infection (RR, 1.56; 95% CI, 1.31–1.81). AI/AN persons had the highest infection fatality ratio (IFR), with 1.4% of infections resulting in death vs 0.7% among White persons.

Figure 1.

Age-standardized relative risk of infection and outcomes per infection by sociodemographic factors: 1490 US counties in 50 states and the District of Columbia, September 2020–November 2021.^aa Race/ethnicity and sex data on hospitalizations were taken from the 95 counties in 14 states in which COVID-NET and seroprevalence data were available. Relative risk calculations involving these cells used denominators from a matched geographic area. Reference values for relative risks are given in parentheses on the y-axis. ^b Race and ethnicity include American Indian (AI), Alaska Native (AN), Pacific Islander (PI), and non-Hispanic (NH). ^c A logarithmic scale is used for the x-axis. COVID-NET, Coronavirus Disease 2019–Associated Hospitalization Surveillance Network; ED, emergency department.

Open in new tabDownload slide

Females were slightly less likely to be infected than males (RR, 0.92; 95% CI, .87–.97), although they had higher case reports per infection vs males (RR, 1.18; 95% CI, 1.11–1.24; Table 1). Per infection, they were a third less likely to die (RR, 0.69; 95% CI, .66–.72; Table 2).

Older persons were less frequently infected than younger (Table 1, Figure 2). The risks of ED visit, hospitalization, and death per infection increased in a stepwise manner with age, and these differences were progressively larger from ED visit to hospitalization to death (Table 2). IFR for those aged ≥70 years was 6.7%, as opposed to 0.02% for those 20 to 29 years, resulting in an RR of 273 (95% CI, 242–305).

Figure 2.

Age-standardized relative risk of infection and outcomes per infection by age: 1490 US counties in 50 states and the District of Columbia, September 2020–November 2021.^aa Reference values for relative risks are given in parentheses on the y-axis. ^b A logarithmic scale is used for the x-axis. ED, emergency department.

Open in new tabDownload slide

While residents of more socially vulnerable counties were more likely to be infected with COVID-19 than those residing in the least socially vulnerable counties, case reports per infection were lower. Once infected, residents of more socially vulnerable counties were more likely to experience an ED visit, hospitalization, or death.

Residents of small metro and rural areas (micropolitan and noncore) [16] were more likely to be infected than residents of large central metro areas. However, they were less likely to be reported as a case, and they were less likely to experience an ED visit, hospitalization, or death. Residents of large fringe metro (suburban) areas were also less likely to be hospitalized or die per infection than residents of large central metro areas.

Midwest and Southern residents were more likely to be infected with COVID-19 than Northeastern residents. However, case reports per infection were lower. Per infection, residents of the South and West were slightly more likely to die than residents of the Northeast.

These analyses have focused on RR for COVID-19 outcomes among infected persons. For comparison, Supplementary Table 6 presents RR estimates for each outcome studied by sociodemographic group, with infected and uninfected persons included in the denominator.

Comparison of Alpha and Delta Periods

The overall risk of death per infection was higher during the Delta period vs the Alpha period (815 vs 714 per 100 000 infections) but not among all subgroups (Table 3). During the Alpha period, Asian/PI and Hispanic persons had substantially higher risks of death per infection than White persons. However, during the Delta period, the absolute risk of death among Asian/PI persons became lower than that during the Alpha period, while the risk among White persons became higher. In the Delta period, the risk of death per infection among Asian/PI persons became less than that among White persons. During the Delta period, the risk of death among those aged ≥70 years was half that during the Alpha period, while it was higher for all other age groups. The RR of death among those aged ≥70 years as opposed to 20 to 29 years went from 639 (95% CI, 541–737) to 70 (95% CI, 44–96). Counties in the lowest and middle SVI tertiles, but not the highest SVI tertile, had higher deaths per infection during the Delta period. As a result, the risk of death among residents of higher SVI counties per infection was no longer significantly different from that of lower SVI counties during the Delta period. Among large central metro counties, the risk of death per infection during the Delta period (704) was lower than during the Alpha period (851); in contrast, for most other urban-rural classes, the risk of death per infection was higher during the Delta period than during the Alpha period. Consequently, the elevated risk of death per infection initially seen in large central metro areas was no longer present during the Delta period. Changes in risk of infection, case report, ED visit, and hospitalization are shown in Supplementary Tables 7 to 10.

Potential Deaths Averted

During the study period, an estimated 53 123 deaths (95% CI, 52,270–53,977) may have been averted if the fatality rate for persons of any race or ethnicity had been no higher than that among White persons (Supplementary Table 11).

DISCUSSION

This article describes COVID-19 outcomes from infection through death using surveillance data with broad US coverage. RRs of COVID-19 outcomes per infection allow a clearer understanding of the separate relationships between sociodemographic factors and infection and those between sociodemographic factors and downstream events: case reporting, ED visits, hospitalizations, and death.

When a population has a lower risk of infection but a higher risk of morbidity or mortality, this distinction can be particularly important. For instance, Asian/PI persons had a lower risk of infection than White persons (RR, 0.50) but a higher IFR (RR, 1.56). This results in an overall lower risk of death due to COVID-19 among Asian/PI persons than among White persons (RR, 0.81). Considered in isolation, the total risk of death obscures the elevated IFR and may even lead one to erroneously suspect a lower IFR among Asian/PI persons. Similarly, the lower risk of infection among those aged ≥70 years (RR, 0.47) is counterbalanced by a much higher risk of death per infection (RR, 273) so that the overall risk of death per population underestimates the risk of death once infected.

For an infection to result in a case, a person must be infected, diagnosed, and meet the case-reporting definition, and the report must be transmitted. Asymptomatic infections, untested infections, or infections diagnosed by self-test will often not meet reporting criteria [35]. Reviewing case rates in the absence of infection rates does not allow one to separate these events. In our analysis, there were an estimated 2.0 infections per report. Although routine case reports consistently show higher rates among females than males [22], it is more informative to state that the risk of infection is lower among females (RR, 0.92) but that the risk of report per infection is elevated (RR, 1.18). In contrast to analyzing events per case report, the approach taken here of estimating events per infection is meaningfully different and provides additional insights to guide prevention.

While earlier US literature showed higher infections and reports among AI/AN, Black, and Hispanic persons as compared with White persons [5, 12], disparities have reduced over time [25]. In this analysis, the risk of infection was elevated only among Hispanic persons as compared with White persons (RR, 1.11). Consistent with prior literature on risk of adverse COVID-19 outcomes per population, our analysis showed elevated risks of COVID-19 morbidity and mortality per infection among AI/AN, Black, and Hispanic persons vs White persons [2, 4, 6], although disparities were no longer statistically significant in the later period. While noting that the overall COVID-19 mortality rate among Asian/PI persons was not elevated as compared with White persons [2, 36], Yan et al described an elevated case fatality ratio [37]. Our analysis takes this a step further in demonstrating an IFR that was higher among Asian/PI persons than White persons during the Alpha period and lower during the Delta period. Of possible explanatory factors [38], changes in testing and care-seeking patterns, vaccination rates, discrimination, and xenophobia could have applied.

The increase in mortality per total population observed in this analysis as age increases has been well described [39, 40], though it potentially underestimates mortality per infection. Additionally, the marked drop in IFR among those aged ≥70 years between the Alpha and Delta periods is notable. This change may be due to the more durable effect of vaccines on prevention of severe disease rather than mild [41], improvement in treatment of COVID-19, early mortality among those with increased COVID-19 susceptibility, and less scarcity of health care resources [42]. That this improvement in IFR was not observed in other age groups may be due to initiation of vaccination, the emphasis of vaccine campaigns and clinical interventions, and vaccine acceptance [43], all favoring older persons as overlaid on the greater virulence [44] of the virus during the Delta period. A limitation of this aggregate data analysis is that the individual contributions of these factors cannot be separated.

Though IFRs were elevated in socially vulnerable counties and large central metro counties, disparities by SVI and urbanicity decreased over time. Disparities were independent of differences in infection risk but may have been related to health system overloading, which had a greater impact on socially vulnerable and urban counties and improved over time [45]. As a composite index, SVI captures many factors that may be related to poorer and less timely access to health care services, such as poverty, limited English proficiency, and poor housing conditions. Lower vaccine uptake has been demonstrated in higher SVI counties and rural areas, which may have also contributed to disparate outcomes [46–48].

While important for comparison among subgroups, age-standardized estimates should not be reported in isolation as they are normalized to the year 2000 overall age structure; for this reason, unadjusted top-line estimates are presented separately.

This analysis has several limitations. This ecologic study compared data from different systems using shared variables. Variables such as race may have been collected differently across systems. Race and ethnicity data were also aggregated in some systems, precluding deeper examination. The merging of data sets depended on matching county catchment areas, an imperfect process when county coverage is incomplete. Certain geographic areas were not covered by overlapping catchments areas, potentially limiting generalizability. We were additionally limited to variables that were common across the surveillance system. Data on factors such as medical comorbidities and vaccination status, while important predictors of COVID-19 outcomes, were unfortunately not available. We did not adjust for variables besides age; presented estimates cannot be considered independent, and relationships should be considered descriptive rather than causal. Race and ethnicity, for instance, vary by census region, with the largest minority populations residing in the West, followed by the South [49]—the 2 regions that showed elevated IFR. Bias may be present due to missingness and could not be quantified due to nonequivalence of missing data by variable and source. This is particularly relevant for case report race data, in which nearly a third of the observations lacked race and ethnicity data (Supplementary Table 4). Seroconversions are used to estimate infections, assuming that seroconversion is universal and persistent and that people will be infected only once. It was important to truncate the analysis prior to introduction of the Omicron variant, when reinfections were more common [18]. To the degree that these assumptions do not always hold, they would lead to an undercounting of total infections. Yet, the fact that a person may go to the ED or be hospitalized more than once with COVID-19 would lead to an overcounting of persons experiencing ED visits or hospitalizations. Certain hard-to-reach populations may be underrepresented in this analysis. For instance, while undocumented immigrants are not excluded from any of the data systems, they may be less likely to donate blood or access health care than others residing in the United States. Although the blood donor data set used to estimate infections was weighted to make it more nationally representative, documentation status is an example of a factor that could not be fully adjusted for, because of the lack of available data.

This article provides a detailed portrait of the US relationship between sociodemographic factors including race and COVID-19 outcomes per infection from September 2020 to November 2021. The estimates come from the comprehensive surveillance systems and establish a frame of reference for further inquiry. While numbers of additional hospitalizations and deaths quantify somewhat the additional suffering among populations disproportionately affected by structural inequalities, the full extent of this burden goes beyond these figures. If the disparities described here had been eliminated, tens of thousands of deaths may have been averted. Although some improvement was seen, continuing efforts are needed, such as ensuring equitable access to vaccines and antivirals and building partnerships between public health and communities to advance health equity [50].

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We thank the many collaborators at federal, state, and local levels who have contributed to the collection and analysis of these COVID-19 surveillance data. We add particular thanks to Robert Anderson of the National Vital Statistics System and Karl Soetebier of the National Syndromic Surveillance Program for their contributions.

Author contributions. S. C. W., K. E. N. C.: methodology, supervision, writing–original draft, writing–review and editing. D. F.: formal analysis, methodology, supervision, visualization. A. H.: formal analysis, methodology. K. B.: visualization, methodology, writing–original draft, writing–review and editing. A. V. G., A. J. H.: supervision, methodology, writing–original draft, writing–review and editing. C. H.: formal analysis. M. E. V. D., J. A., F. B. A., T. J. B., F. H., S. J. H., D. J., J. D. O., H. T. P., M. J. S., C. A. T.: methodology, writing–original draft, writing–review and editing. J. M. J.: conceptualization, methodology, supervision, writing–original draft, writing–review and editing.

Data access. Serologic data may be accessed on the CDC website. Case and unified hospitalization data may be accessed through the HHS Protect Public Data Hub. Emergency department patient visit data may be accessed via the CDC COVID Data Tracker: Trends in ED Visits. COVID-NET hospitalization data may be accessed through the CDC COVID Data Tracker. Mortality data may be accessed via CDC Wonder.

Disclaimer. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the CDC.

Financial support. Data analyzed in this manuscript were collected through COVID-19 surveillance systems funded by the US government at federal, state, and local levels.

References

Author notes