We enrolled 26,201 adults and children in the study from January 12, 2021, through August 12, 2021; median recruitment period per site was 164 days (range 84–199 days, IQR 150–185 days). During that time, ≈69,000 persons were approached from a cumulative total of >450 unique community venues surrounding the 15 clinical research sites; 22,284 (≈32%) participants enrolled (Figure 1), resulting in a median enrollment per community of 1,246 (range 508–2,924, IQR 997–1,682) participants. Sites intended to enroll one quarter of the community cohort into each of the 4 age groups; however, only 2,113 (9.5%) enrolled participants were <18 years, and the median number of children per site was 48 (IQR 24–100). Most sites enrolled similar proportions of male and female participants from community venues (Appendix Table 2). A total of 3,111 participants were enrolled in the clinical cohort and 806 in the nursing home cohort (Appendix Tables 1, 3, 4, Figures 1, 2).

A total of 22,284 participants of the targeted 43,800 were enrolled from community venues; of those, 95% provided a blood sample for SARS-CoV-2 antibody testing and 99% provided a nasal swab sample for SARS-CoV-2 PCR testing (Figure 1). Complete data were available for 21,189 (95%) community cohort enrollees for analysis of a combined endpoint of Ab+, PCR+, or both (i.e., evidence of prior or active infection). The remainder of the findings about the community cohort pertain to the group of 21,189.

The unadjusted demographic profile for sex and ethnicity of the community cohort was similar to the estimated demographic profiles for most communities according to corresponding county-level data from the American Community Survey (Figure 2); however, the profile by race differed at several sites. In addition, prevalence of race varied widely, from 9% to 89% for Black persons (median 37% [IQR 24%–62%] across sites) and from 6% to 80% for White persons (median 28% [IQR 15%–60%] across sites). Hispanic ethnicity varied from 3% to 98% (median 33% [IQR 9%–47%] across sites).

Median seropositivity of the community cohort (Appendix Table 5, Figure 3), determined according to 2,272 Ab+ participants, was 12.4% (IQR 9.1%–14%); seropositivity was similar for the clinical cohort (11.3% [IQR 7.7%–15.5%]) but lower for the nursing home cohort (3.3% [IQR 2.4%–7.7%]). Of note, 50% of the nursing home cohort participants were recruited in Puerto Rico, where seroprevalence for all age groups was 3%–5% (Figure 3, panel B). The median overall prevalence of active infection (PCR positivity) in the community cohort across all sites (Appendix Table 5, Figure 3), determined according to 189 PCR+ participants, was 0.8% (IQR 0.2%–1.5%) A total of 64 participants enrolled in COMPASS were both PCR+ and Ab+ (52 from the community cohort and 12 from the clinical cohort).

In the community cohort, overall median prevalence across all 15 site-level estimates of the combined endpoint, based on PCR+ or Ab+ status, was 12.9% (range 4.9%–18%, IQR 9.2%–14%) (Figure 3, panel A). Although prevalence varied widely, when we looked at demographic differences within each site, we found no difference in average prevalence of prior or active infection by age, whether considering the 4 age groups at all sites (Figure 3, panel A) or comparing children (<18 years of age) with adults (>18 years of age) at each of the 10 sites that enrolled >25 children (Figure 3, panel B). We also found no difference in average prevalence by participant sex or ethnicity (Figure 3, panel A). Within each site, however, the average prevalence estimate for Black participants was 3 percentage points higher than for White participants (p<0.01) and 2.4 percentage points higher than for those with race identified as other (p = 0.11). Among participants >18 years of age, we found a nonsignificant association between a higher prevalence of active or prior SARS-CoV-2 infection and a lower level of attained education (Appendix Figure 4). We found no association between COVID-19 and income level; however, many (32.8%) participants reported their household income as don’t know or prefer not to answer (data not shown).

About half of seropositive persons in the community cohort reported that they had not ever experienced COVID-19 symptoms (median 50.0% [IQR 45.3%–63.8%]; data not shown). Similarly, about half of PCR+ persons in the community cohort reported not having experienced symptoms within the 14 days before their participation in the survey (median 52.4% [IQR 50.0%–62.4%]; data not shown). In contrast, among the seronegative persons in the community cohort, a median of 80.9% (IQR 75.7%–83.4%; data not shown) reported they had not ever experienced COVID-19 symptoms. Similarly, a median of 79.8% of PCR-negative persons (IQR 75.7– 83.1%; data not shown) in the community cohort reported not having experienced symptoms within the 14 days before the survey.

In a separate analysis of self-reported PCR and antibody results among the 21,940 persons who enrolled and tested antibody negative in COMPASS (18,629 from the community cohort), prevalence of a self-reported prior positive COVID-19 test (nasal swab sample) was 4.9% (n = 1,067) and prevalence of a self-reported prior positive antibody test was 2.4% (n = 526) (data not shown). By contrast, among the 2,792 who enrolled and tested antibody positive in COMPASS (2,380 from the community cohort), prevalence of a self-reported prior positive COVID-19 test (nasal swab sample) was 33.4% (n = 933) and prevalence of a self-reported prior positive antibody test was 7.9% (n = 221) (data not shown).

Overall, across 15 site-level estimates, the median percentage of persons in the community cohort who reported being willing to receive a COVID-19 vaccine was 78% (IQR 72%–82%); the percentage was higher among participants >60 years of age (median 89%, IQR 84%–92%) compared with those 18–39 years of age (median 74%, IQR 67%–83%) and 40–59 years of age (median 76%, IQR 74%–81%) (Figure 4). The median percentage of participants who reported being willing to receive a vaccine was 71.4% (IQR 64.8%–76.3%) for Black, 78% (IQR 68.8%–82.5%) for other race, and 84.2% (IQR 73.6%–87.0%) for White participants (Appendix Table 6). We found no differences in willingness to receive a vaccine by sex or ethnicity across the sites. When looking at demographic differences within each site, we found that vaccine willingness among those >60 years of age was an average of 10 percentage points higher than among those 40–59 (p<0.01) years of age, 11 percentage points higher than those 18–39 (p<0.01) years of age, and 22 percentage points higher than those <18 years of age (p<0.01). Similarly, the average difference among participants who reported being willing to receive a vaccine was 10 percentage points lower for Black compared with White participants (p<0.01) and 7 percentage points lower for Black participants than for persons in other racial groups (p<0.01).

COMPASS, a population-based cross-sectional serosurvey, enrolled ≈22,000 adults and children from community venues in 15 largely urban US settings in the first half of 2021 and found that the overall prevalence of prior and active SARS-CoV-2 infection was 12.9%. TLS, typically used to recruit hard-to-reach populations (19), was an innovative way to randomly sample populations representative of diverse communities amid pandemic restrictions to access and movement, based on demographic comparisons with the American Community Survey. Contrasting with initial reports of COVID-19 (22–24), population-based prevalence did not vary by age, indicating that acquisition was similar for all age groups. In addition, despite prevalence of prior or active COVID-19 being higher among Black participants, fewer Black participants in the sampled communities were willing to receive an approved COVID-19 vaccine compared with participants who were from White or other racial groups, a finding that may have magnified the disparate burden of COVID-19 and associated outcomes in these communities.

Compared with our findings, studies conducted in healthcare settings may have arrived at higher estimates of COVID-19 prevalence (3,4), reflecting the symptomatic status, healthcare access, and healthcare-seeking behavior of clinical populations. For example, a commercial laboratory seroprevalence study conducted in the United States during the second half of 2021 assessed at 4-week intervals convenience samples of blood specimens collected for clinical testing and found an overall US seroprevalence of 33.5% in December 2021 (4). Similar to our findings with regard to age, the study of commercial laboratory samples also found that seroprevalence among children and adults did not differ. Whether for COVID-19 or other widespread outbreaks, such as mpox, seroprevalence estimates from population-based studies that include those who are less likely to engage in care, combined with estimates from healthcare settings, may offer the most comprehensive picture of outbreak effects at that point in time.

COMPASS found that willingness to receive a vaccine during the first half of 2021, during the earliest stages of vaccine availability in the United States, was 10 percentage points lower among Black participants and 7 percentage points lower among other race participants than among White participants. Monthly telephone surveys of immunization practices, conducted during December 2020–November 2021, found a similar level of racial disparity in April 2021 (25). By November 2021, racial disparities in COVID-19 vaccine uptake in the United States were no longer evident (26); nonetheless, 1 year later, by December 2022, Black race and Hispanic ethnicity remained associated with higher rate ratios of age-adjusted hospitalization and death compared with White race (26,27), indicating that vaccine uptake alone did not explain disparities in outcomes.

Strengths of our study included collection of biological samples and detailed demographic, behavioral, and self-reported clinical data from randomly sampled persons from the general population, including children. COMPASS used a centrally managed approach to TLS to ensure a participant population that reflected the communities surrounding clinical research sites participating in COVID-19 vaccine trials. The weekly systematic updating of community venue sampling frames was a novel way to adapt to changing COVID-19 guidelines and restrictions as well as to fluctuating weather (18). The prevalence of active infection was based on validated laboratory assays performed on approved platforms. Although we did not assess serologic evidence of vaccine immunity (SARS-CoV-2 spike antibody), the serologic assessment of prior infection through a single assay conducted at a centralized laboratory yielded rigorous seroprevalence findings.

Limitations of our study included conducting it in the first half of 2021, just before the Delta variant became dominant in the United States, when vaccines were becoming available, home-based rapid test kits were not yet widely available, and even PCR testing was often not easily available. Although enrollment targets, especially for children, were not fully achieved, the study had sufficient power to compare both seroprevalence and the combined endpoint of seroprevalence, PCR positivity, or both, by age group. The results do not reflect the entire US population but do reflect persons who were well enough to attend commonly frequented venues in diverse, largely urban communities surrounding the participating clinical research sites, who represent those who had more exposure to COVID-19 compared with those who were homebound. Sampling largely urban settings was relevant, given that the US population is 80% urban (28) and that, globally, COVID-19 cases had a large effect in urban settings, especially in the earlier stages of the epidemic (29). Serologic testing may have underestimated seroprevalence because of either absent or low titers after a mild infection (30) or waning antibody titers after a distant (>6 months) infection (31). Last, recall and social desirability biases may have affected the accuracy of some replies to some sections of the questionnaire.

In our population-based survey, we used TLS to recruit members of the general population who were relatively hard to reach because of the status of the COVID-19 pandemic. Our findings demonstrate that prior and active SARS-CoV-2 infection varied widely by community but, contrasting with initial reports, not by age. Half of Ab+ and PCR+ participants reported no symptoms, underscoring the limitations of case-based reporting and the potential for asymptomatic transmission (32,33). Our findings of higher prevalence of prior or active COVID-19 among persons who were less willing to get vaccinated highlight the value of tailoring public health efforts to the communities most likely to continue to experience disparate effects COVID-19 and its complications.