Assessment of Seroprevalence of SARS-CoV-2 and Risk Factors Associated With COVID-19 Infection Among Outpatients in Virginia

Key Points Question What percentage of the Virginia population had been exposed to severe acute respiratory syndrome coronavirus 2 after the first wave of coronavirus disease 2019 (COVID-19) infections in the US? Findings In this cross-sectional study of 4675 adult outpatients presenting for non–COVID-19–associated health care in Virginia, a seroprevalence of approximately 2% was found, with an estimated 66% of seropositive results associated with asymptomatic infections. Hispanic ethnicity, residence in a multifamily unit, and contact with an individual with confirmed COVID-19 infection were risk factors significantly associated with exposure. Meaning This study found that, as of August 2020, the population of Virginia remained largely immunologically naive to the virus.


Introduction
The coronavirus disease 2019 (COVID-19) pandemic, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has disrupted the US. Government agencies, health care systems, and the general public track confirmed case counts in US states to guide decision-making. Most of the confirmed case counts have been based on positive viral detection on the reverse transcriptasepolymerase chain reaction (PCR) test among individuals with symptoms. These case counts are underestimates of the total burden of SARS-CoV-2 infection because of incomplete testing availability and the substantial fraction of asymptomatic individuals with SARS-CoV-2 infection who never receive testing. Therefore, antibody evidence of previous SARS-CoV-2 exposure can be used as a complementary tool to measure the total burden of infection in the population.
There was initial concern that serologic assays would be subject to a high rate of false-positive results (ie, low specificity) because of exposure to other cross-reacting coronaviruses. False-negative results (ie, low sensitivity) were also a concern given the time required for the immune response to generate antibodies and the potential for dampened immune responses with asymptomatic infection. A number of studies, however, have indicated that the performance of many serologic assays is reasonable with regard to sensitivity and specificity. For example, the Abbott SARS-CoV-2 immunoglobulin G assay test has indicated 92.7% to 100% sensitivity and 99.4% to 100% specificity 1-3 against a criterion standard of PCR-confirmed infection 14 days after symptom onset.
Although sensitivity may be lower among asymptomatic individuals (78% sensitivity was reported in 1 study 3 ), data regarding this population are incomplete.
Several serologic studies have been performed in the US to date. These studies have included large multiregional seroprevalence analyses that use residual clinical specimens from convenience samples 4 ; recruitment of self-selected individuals within a particular geographic region 5,6 ; surveys of high-risk individuals, such as first responders or health care workers; and community-level seroprevalence data from randomized households. 7,8 Each strategy has limitations, such as generalizability to the public when participants are self-selected, inability to ascertain risk factors when no questionnaire is administered, and variable geographic reach and sample size constraints depending on the study design.
We conducted a statewide serologic survey of Virginia to examine the commonwealth's total burden of previous COVID-19 infection. We enrolled participants at outpatient clinics or outpatient medical laboratories in 5 geographically disparate regions. To improve generalizability, we stratified enrollment to meet the age, race, and ethnicity profile of each region and did not include self-referred participants.

Study Design
The study was approved by the institutional review board of the University of Virginia, with a waiver of informed consent because the study did not constitute human subject research and because the study was requested by the Virginia Department of Health and therefore considered public health surveillance according to 45 CFR §46.102. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies. Adults 18 years or older who presented in person for scheduled outpatient clinic or outpatient laboratory appointments were eligible. All outpatient sites conducted prescreening of individuals to ensure no COVID-19-like symptoms; at all sites, patients had options to convert in-person visits to telehealth visits. Enrollment occurred from June 1 to August 14, 2020, at 5 geographically diverse health system sites: the University of Virginia Health System in the northwest, INOVA Health System in the north, Sentara Healthcare in the east, Carilion Clinic in the southwest, and Virginia Commonwealth University in the central region. Each site enrolled up to 1000 Virginia residents, with up to 1 resident per household, and excluded individuals who were being evaluated for active COVID-19 illness. To improve generalizability, we stratified enrollment to meet the age, racial, and ethnic demographic profile of the region. Clinic sites were chosen to obtain this diversity. Population estimates for age, race, and ethnicity were obtained from the University of Virginia Weldon Cooper Center for Public Service based on the population as of July 1, 2019 9 (population by zip code is shown in eFigure 1 in the Supplement). Race and ethnicity were defined by participants. If an individual's characteristics aligned with the region's demographic quota and the individual consented to participate in the study, research coordinators administered an electronic questionnaire and obtained a blood sample via venipuncture for SARS-CoV-2 serologic testing. Electronic informed consent was obtained from each participant.
All blood samples were tested at a central laboratory (University of Virginia Medical Laboratories, Charlottesville). Plasma (in lithium heparin tubes) was tested on the Architect i2000 analyzer (Abbott) using the SARS-CoV-2 immunoglobulin G antibody immunoassay approved via emergency use authorization. Participants were provided with their serology test results by mail.

Statistical Analysis
All analyses were conducted using R software, version 3.6.3 (R Foundation for Statistical Computing). Sociodemographic characteristics were tabulated by region. To estimate the prevalence of seropositivity that was representative of the regional distributions of age, sex, race, ethnicity, and insurance status, we used an iterative proportional fitting procedure (also termed raking) to estimate sampling weights. Age was categorized into 18 to 29 years, 30 to 39 years, 40 to 49 years, 50 to 59 years, 60 to 69 years, 70 to 79 years, and 80 years or older. Race was defined according to census definitions of White, Black or African American, Asian, other race (specific races not queried on survey), and 2 or more races. Health insurance on January 1, 2020, was categorized as Medicaid, Medicare, private, military, and none or uninsured. Virginia population estimates by age, sex, race, and ethnicity at the regional level as of July 1, 2019, were collected. 9 Virginia population estimates by insurance status at the regional level were collected from 2018 US Census Bureau data. 10 The raked weights were estimated using the R software survey package 11 to match the cross-classified distribution of age and sex and the marginal distribution of race, ethnicity, and health insurance at the regional level.
The raw prevalence of SARS-CoV-2 seropositivity by region and subgroup was estimated as the proportion of samples with positive results. The weighted seropositivity prevalence was estimated, accounting for the estimated raked weights using the R survey package. Prevalence was further corrected for imperfect sensitivity and specificity of the serology assay using the formula: where P is the weighted prevalence, Se is the sensitivity of the assay, and Sp is the specificity of the assay. Sensitivity and specificity were assumed to be 92.7% (95% CI, 90.2%-94.8%) and 99.9% (95% CI, 99.4%-100%), respectively. 3 Variability around these estimates was incorporated into the calculation of 95% CIs for the corrected prevalence rates using the delta method. Subgroups categorized by age, sex, race, ethnicity, insurance status, high-risk health condition, and week of collection were assessed.
To estimate rates of underascertainment, we estimated the expected total number of SARS-CoV-2 infections per region by multiplying the prevalence by the regional population of individuals 18

JAMA Network Open | Infectious Diseases
Seroprevalence and Risk Factors Associated With COVID-19 Infection in Virginia years and older. We compared this number to the number of cases per region among individuals 18 years and older, excluding cases among adults living in institutional facilities (long-term care, correctional, or congregate facilities), as reported by the Virginia Department of Health. We calculated the ratio of estimated infections to reported cases with ranges derived from the 95% CIs of the seroprevalence estimates. We identified risk factors associated with SARS-CoV-2 seropositivity using logistic regression analysis. We assessed region, sociodemographic characteristics, living environment, working environment, contact with someone with confirmed COVID-19 infection, perception of risk, perception of adherence to prevention behaviors, and travel since January 1, 2020. Risk factors with univariable odds ratios (ORs) that were statistically significant (P < .05) or univariable ORs that were less than 0.67 or greater than 1.50 were included in the multivariable model.  We then examined the prevalence of SARS-CoV-2 immunoglobulin G antibodies across the commonwealth and by subgroup ( Table 1). Overall, the unadjusted prevalence was 101 of 4675 participants (2.2%). Of 101 participants with seropositivity, 42 participants (41.6%) were Hispanic.

JAMA Network Open | Infectious Diseases
Seroprevalence and Risk Factors Associated With COVID-19 Infection in Virginia during the spring of 2020 have been within this 1.1% to 3.0% range. [5][6][7] Among frontline health care personnel, seropositivity has been higher, with 1 study reporting seropositivity of 7%. 13 In contrast, New York City reported the results of more than 1 million antibody tests, finding an estimated 27% seropositivity rate. 14 Based on the estimated weighted seroprevalence of 2.4%, it is estimated that approximately 2.8-fold more infections occurred than were ascertained by confirmed case counts. This ratio is relatively low compared with the underascertainment estimated by some previous studies, which have ranged from 6-fold to 53-fold more infections than ascertained by confirmed case counts. 4,5 There are 2 possibilities for the lower underascertainment: (1) the current seroprevalence study having a high-risk health condition, which is similar to reported national estimates. 16 It also appeared that participants' level of concern about becoming ill with COVID-19 infection was similar to national norms, with 65% of the US population very or somewhat concerned. 17 The seroprevalence among individuals aged 18 to 39 years in the study's outpatient population was low relative to confirmed case counts in the state, 18 suggesting that this study may underestimate seroprevalence in this group. It is possible that the true seroprevalence in the state is marginally higher than the 2.4% estimate.

Strengths and Limitations
This study has several strengths. The study had broad geographic reach, which enabled a comparison between regions, and the study had the ability to administer a questionnaire to assess demographic characteristics, risk factors, and previous illness. By enrolling community-dwelling adults receiving outpatient services, the study ensured sufficient enrollment of older adults and individuals with chronic health conditions to assess seroprevalence in this subset of the population who are at high risk of severe illness and hospitalization should they become infected with COVID-19. The highest seropositivity was in the northern region, which is unsurprising given the population density and the location of reported case counts. Notably, seroprevalence was inconsistent, with many zip codes exhibiting 0% seropositivity and others 20% seropositivity, often among adjacent zip codes. The rate of seropositivity was substantially high among Hispanic individuals, at 10.2%. Latinx individuals have been reported to account for a disproportionate number of COVID-19 cases, including in Virginia, in which they account for 33.8% 18 of cases but only 9% of the general population. 9 However, it is notable that the seroprevalence in the current study indicated that Hispanic individuals constituted an even higher proportion of seropositive results at 41.6%. Further investigation and preventive strategies targeting Hispanic individuals are needed.
Because the study assessed previous COVID-19 symptoms, it was estimated that at least 50% and more likely 66% of seropositive cases were associated with asymptomatic infection. The asymptomatic rate has been assessed in a number of case series (summarized in a systematic review 19 ) and has ranged from 6.3% to 96.0%; however, these case series did not include representative samples from the population. In contrast, community studies from Iceland and Italy estimated the asymptomatic rate to be approximately 40%. 19,20 These PCR-based studies to define positivity had shorter follow-up periods than the current study. It is important to precisely document the asymptomatic rate because a rate of approximately 66% vs 40% produces large differences in the estimated transmission from asymptomatic individuals. 21 This study's data, which indicated a high rate of asymptomatic infection and a higher rate of seropositivity among those who had contact with individuals with confirmed COVID-19 infection, support the testing of asymptomatic individuals who are close contacts as a means of identifying cases and reducing transmission. Although outside the scope of this study, another strength of the study design is that it will be possible to assess this cohort for seroconversion and seroreversion in the future and evaluate whether individuals with seropositivity manifest protection from subsequent COVID-19-like illness.
This study also has several limitations. These limitations include the outpatient participant study design, which may limit generalizability to the general population. In addition, data regarding risk factors and previous testing and illnesses were based on self-reporting and thus subject to participant recall. Notably, however, the most important factors identified (ethnicity, age, and type and geographic location of residence) are not likely to be subject to recall bias. The test sensitivity and specificity of any serologic test is not perfect, and false-positive results increase in a

JAMA Network Open | Infectious Diseases
Seroprevalence and Risk Factors Associated With COVID-19 Infection in Virginia