Prevalence of SARS-CoV-2 Antibodies in New York City Adults, June–October 2020: A Population-Based Survey

Abstract

(See the Editorial Commentary by Rosenberg and Tesoriero, on pages 185–7.)

In March 2020, New York City (NYC) was the first epicenter of the coronavirus disease 2019 (COVID-19) pandemic in the United States [1]. Early in the outbreak there was limited testing capacity and health care was prioritized for people with severe illness [2]. Case surveillance data alone do not reflect the magnitude of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections in NYC. The presence of SARS-CoV-2 antibodies suggests prior infection and can contribute to an understanding of asymptomatic or mild infections otherwise not detected through traditional surveillance. Population-level seroprevalence can help determine the population proportion previously infected and the proportion with possible humoral immunity against SARS-CoV-2.

Past SARS-CoV-2 serological surveys of NYC residents have used convenience samples [3–6] or assessed seroprevalence in specific populations [7, 8]. Seroprevalence estimates ranged from 6.9% to 23.6% in the general NYC population [3–6] and 13.7% to 31.2% for NYC health care personnel [7, 8] from February to July 2020. While studies of clinical laboratory residual serum and other convenience-based sampling might be subject to selection bias and might have limited data for analysis by demographic and socioeconomic variables, they can be rapidly deployed to provide timely seroprevalence estimates during a public health emergency. Currently lacking from the literature are population-based, representative surveys that estimate antibody prevalence in the general NYC adult population overall and for priority demographic groups [9].

During 1 June to 9 October 2020, the NYC Department of Health and Mental Hygiene (DOHMH) conducted a representative cross-sectional survey using serosurvey data and self-reported test results to estimate SARS-CoV-2 antibody prevalence in NYC.

METHODS

Telephone Survey

Participants were recruited from the ongoing NYC Community Health Survey (CHS), a cross-sectional telephone survey used to assess health and risk behaviors of New Yorkers [10]. The CHS uses a disproportionate stratified random sample to help assure geographic representativeness across the city. Participation goals are set for each of 42 United Hospital Fund neighborhoods, which are defined by contiguous zip codes. Using random digit dialing, a sample of landline and cellular telephone numbers was created to reach noninstitutionalized adult NYC residents (≥18 years). Interviews were conducted in English, Spanish, Russian, and Chinese.

Respondents were asked about demographics, underlying health conditions, employment, and social distancing (Supplementary Table 1). Respondents were categorized into neighborhood poverty level based on the population percentage in a respondent’s zip code living below the federal poverty level (FPL) per the American Community Survey (ACS) 2014–2018, with imputation of missing cases [11]. Categories were low (<10% below FPL), medium (10%–20% below FPL), and high poverty (>20% below FPL). Respondents were asked if they experienced fever, cough, shortness of breath, sore throat, or loss of taste or smell within the past 30 days. If not, they were asked if they experienced these symptoms since February 2020. All respondents were asked if they believed they previously had COVID-19. All respondents were asked “There is a test to detect antibodies to the virus that causes COVID-19. The test is usually done with a blood sample. Have you ever had an antibody test for COVID-19?” If yes, we asked respondents what their prior SARS-CoV-2 antibody test result was. Irrespective of previous serological testing, we invited all respondents to participate in antibody testing.

Specimen Collection and Testing

For each consenting participant, a phlebotomist conducted an at-home blood draw to collect 5 mL of whole blood in a serum separator tube. On the same day, samples were transported at 4°C to the NYC Public Health Laboratory where serum was separated from the specimen and tested for SARS-CoV-2 immunoglobulin G (IgG) antibodies using the DiaSorin LIAISON SARS-CoV-2 S1/S2 IgG assay [12]. The test has a reported 97.6% positive and 99.3% negative percent agreement with reverse transcription polymerase chain reaction (RT-PCR) testing at the time of infection [12]. Peer reviewed literature on this assay described specificity ranging from 90.5% to 98.9% with varying sensitivity depending on the timeframe of antibody detection after positive PCR test for SARS-CoV-2 [13–15].

The NYC DOHMH Institutional Review Board determined this as public health surveillance. The Centers for Disease Control and Prevention (CDC) reviewed this activity; it was conducted consistent with applicable federal law and CDC policy, eg, 45 C.F.R. part 46.102(I)(2), 21 C.F.R. part 56; 42 U.S.C.§241(d); 5 U.S.C.§552a; 44 U.S.C.§3501 et seq. Informed consent was obtained from participants before specimen collection.

Power Calculations

Power calculations were performed using population estimates from the 2019 CHS to assure ability to reliably detect antibody prevalence. We calculated the sample size necessary to achieve a relative standard error (RSE) < 0.30 for a range of hypothetical prevalence rates of 5%–30%. For our survey, an RSE > 0.30 indicates an estimate is potentially unreliable. Accounting for survey design artifacts and assuming ≥10% antibody prevalence, with 1000 participants, we could make a citywide antibody prevalence estimate with a RSE < 0.30. Assuming ≥15% antibody prevalence, with 2200 participants, we could make antibody prevalence estimates stratified by race/ethnicity with a RSE < 0.30 for white, black, and Latino New Yorkers. We aimed to recruit between 1000 and 2200 participants with the knowledge that with higher antibody prevalence, a smaller sample size would be necessary to achieve a stable estimate overall or in stratified analyses.

Data Analysis

To generate a representative estimate of antibody prevalence among the NYC noninstitutionalized adult residential population, CHS data were weighted adjusting for varying selection probabilities and potential overlapping landline and cell phone sampling frames. Survey weights were adjusted to 2014–2018 ACS [11] population control totals using SAS code rake_and_trim_G4_V5.sas [16]; final weights were scaled to adjust for potential nonresponse bias.

We used American Association for Public Opinion Research standard definitions (revised 2016) to calculate annual CHS cooperation rate No. 3 and response rate No. 3 [17]. Cooperation rate No. 3 is the number of survey participants, divided by the number in the sample who were contacted and determined as eligible. Response rate No. 3, a more conservative estimate, is the number of survey participants who complete the survey, divided by those who completed plus partial completes, refusals, noncontacts, and cases of unknown eligibility. People with unknown eligibility lacked live contact by an interviewer to determine eligibility. These could have included business or nonworking numbers (both ineligible). People with unknown eligibility typically comprise most of the denominator; the denominator was adjusted to estimate the proportion of unknown but likely to be eligible people. The 2020 CHS cooperation and response rates were 74.4% and 7.4%, respectively.

Univariate prevalence estimates and 95% confidence intervals (CI) were generated. Combined antibody test results, including self-reported test results, were used for all analyses to estimate citywide and stratified prevalence. For those who provided both a blood specimen and a self-reported test result, only serosurvey specimens tested by DOHMH were used in analyses. Data were stratified by demographics. t tests were used to compare antibody prevalence by sex, age, race/ethnicity, borough of residence, place of birth, language of interview, neighborhood poverty level, and health insurance status. We assessed estimate reliability based on RSE, sample size, and CI width. To determine if prevalence differed between the 2 groups, a multivariable logistic regression model was constructed to see if there were increased odds of having a positive test result if the respondent self-reported versus provided a blood specimen for the serosurvey, while controlling for sex, race/ethnicity, age, borough of residence, language of interview, and neighborhood poverty level. We used SAS EG version 7.15 and SUDAAN 11.0.1 accounting for weight and complex survey design. Two-sided P values ≤ .05 were statistically significant.

RESULTS

From June to October 2020, 1074 respondents completed the survey; of these, 497 provided whole blood and 577 provided only self-reported antibody test results. Of 1074 respondents, 442 were males and 628 females. Respondent ages (in years) varied: 18–44 (n = 458), 45–64 (n = 406), and ≥65 (n = 210). Respondents were Asian or Pacific Islander (n = 112), black (n = 194), Latino (n = 309), other (n = 29), and white (n = 428). They were distributed geographically and by neighborhood poverty level: Bronx (n = 222), Brooklyn (n = 330), Manhattan (n = 226), Queens (n = 232), and Staten Island (n = 64); low (n = 217), medium (n = 425), and high neighborhood poverty (n = 415) (Table 1).

The overall weighted SARS-CoV-2 antibody prevalence, including those who provided a blood specimen and those who only provided a self-reported result, was 24.3% (95% CI, 20.7%–28.3%); it was 21.2% (95% CI, 16.6%–26.6%) among respondents who provided blood for this serosurvey and 26.9% (95% CI, 21.7%–32.8%) among respondents only with a self-reported result. Participants who only self-reported a test result had increased nonsignificant odds of having a reported positive test (odds ratio = 1.47; 95% CI, .95–2.27).

Ninety-one respondents provided both a blood sample and self-reported antibody result. Among these, 81 had concordant results (89.0%) and 10 (11.0%) had discordant results: 4 respondents self-reported a positive test result and were negative for antibodies upon testing; 6 respondents self-reported a negative test result and were positive for antibodies upon testing.

When examining combined weighted results from the serological tests and self-reported data, antibody prevalence was similar by sex. Respondents aged 18–44 years had higher prevalence (26.4%; 95% CI, 21.0%–32.7%; P = .018) compared with respondents aged ≥ 65 (15.1%; 95% CI, 9.2%–24.0%). Latino (30.7%; 95% CI, 24.1%–38.2%; P = .021) and black (30.7%; 95% CI, 21.9%–41.2%; P = .004) respondents had a significantly higher prevalence compared with white respondents (17.4%; 95% CI, 12.5%–23.7%). Antibody prevalence among respondents living in the Bronx was nearly double compared with respondents living in Manhattan (32.4%; 95% CI, 23.6%–42.6%; P = .009 vs 17.1%; 95% CI, 11.6%–24.4%). Antibody prevalence was similar among respondents born outside of the US (26.0%; 95% CI, 20.7%–32.0%; P = .440) compared with US-born respondents (23.0%; 95% CI, 18.2%–28.6%). Non-English–speaking respondents (35.1%; 95% CI, 26.6%–44.6%; P = .013) had higher prevalence than English-speaking respondents (22.3%; 95% CI, 18.4%–26.8%). Respondents living in neighborhoods with high neighborhood poverty had higher prevalence compared with those living in low neighborhood poverty (31.6%; 95% CI, 25.4%–38.6%; P = .029 vs 19.8%; 95% CI, 12.8%–29.4%). Respondents without health insurance had a higher prevalence compared with those with health insurance (37.6%; 95% CI, 23.8%–53.7%; P = .078 vs 23.3%; 95% CI, 19.6%–27.5%; Table 1).

Antibody prevalence varied among respondents with different underlying conditions. Respondents with asthma had a lower prevalence than those without it (11.7%; 95% CI, 6.6%–19.9%; P < .001 vs 26.8%; 95% CI, 22.7%–31.4%). Respondents with obesity (body mass index [BMI] ≥ 30 and ≤100) had a higher prevalence compared with those who had BMI < 25 (34.0%; 95% CI, 26.8%–42.1%; P = .008 vs 20.5%; 95% CI, 15.0%–27.4%; Table 2).

There were 681 respondents currently employed, self-employed, or who recently lost their job because of the pandemic. Among these, antibody prevalence of respondents mostly working outside of the home was higher but not significantly different (30.0%; 95% CI, 23.5%–37.4%; P = .059) compared with the prevalence of those mostly working from within the home (20.9%; 95% CI, 15.1%–28.1%). Among all respondents, there was a higher prevalence among those who reported staying at home none or some of the time, avoiding interacting with others outside except for essential needs (31.8%; 95% CI, 25.3%–39.2%; P = .004) compared with respondents who reported staying at home all or most of the time (19.7%; 95% CI, 15.8%–24.3%; Table 3).

Antibody prevalence was significantly greater, more than 3 times higher, among respondents who reported COVID-19–like illness symptoms at some point between February 2020 until survey administration (43.6%; 95% CI, 36.7%–50.7%; P < .001) compared with those who did not report any of these symptoms (13.0%; 95% CI, 9.3%–17.9%). People who thought they had experienced COVID-19 (not mutually exclusive with those who reported COVID-19–like illness) had higher prevalence (81.2%; 95% CI, 72.8%–87.4%; P < .001) compared with those who did not think they had COVID-19 (9.5%; 95% CI, 6.5%–13.6%; Table 4).

DISCUSSION

Nearly 1 in 4 adult NYC residents had evidence of SARS-CoV-2 infection by October 2020. This is among the highest antibody prevalence reported for a US jurisdiction during the first wave of the COVID-19 pandemic and it is the first population-based survey conducted in NYC of SARS-CoV-2 antibody prevalence [3–6, 18–22]. These data represent an important contribution to understanding the true extent of the pandemic in NYC, particularly by allowing for stratification by race/ethnicity and neighborhood poverty.

Our study suggests disparities in SARS-CoV-2 antibody prevalence across racial/ethnic subgroups in NYC, consistent with other NYC and US-based studies reporting COVID-19 infection by race/ethnicity [4, 6, 23–25]. Nearly 1 in 3 black and 1 in 3 Latino respondents had SARS-CoV-2 infection by October 2020. Respondents without health insurance had higher prevalence than those with it; however, wide CIs limit interpretation. While sample sizes are small, among those without health insurance, 58.8% (40/68) were Latino, and 52.5% (21/40) of Latino respondents without health insurance had SARS-CoV-2 antibodies. We also observed prevalence differences by borough of residence, language of interview, and neighborhood poverty level. Considering the varied demographic composition of NYC boroughs [26], our findings are congruent with other data indicating higher SARS-CoV-2 infection rates in areas of concentrated poverty due to multiple factors that increase exposure risk [27–28].

Respondents who worked mostly outside the home had a higher prevalence compared with those who worked mostly from within the home. As black and Latino people are overrepresented in several essential industries, workplace-related exposures might be contributing to higher antibody prevalence among black and Latino respondents [6, 29]. Overall, our findings illustrate how the COVID-19 pandemic unequally impacted NYC residents, with black and Latino New Yorkers and those from poorer neighborhoods more likely to have had previous SARS-CoV-2 infection. These inequities result from myriad structural, racial, and economic inequalities that drive NYC health disparities [25, 28–30].

Our combined weighted citywide seroprevalence estimate is consistent with New York State Health Department data of NYC residents visiting grocery stores, which found 22.7% seropositivity during 19–28 April 2020 [4]. A cross-sectional serosurvey of routine care patients at Mt Sinai Hospital in NYC found 19.1% prevalence in the week ending on 19 April, and up to 61.7% among urgent care patients [5]. A NYC serosurvey of adult residents found a 23.6% prevalence from 13 May to 21 July 2020 [6]. Using residual commercial laboratory specimens, a CDC nationwide serosurvey estimated 25.1% of metro New York residents had antibodies during 27 July to 13 August 2020, similar to our citywide estimate [20]. While prior studies differ in methodology and sample, our study provides important confirmation of prior estimates in NYC during this time and also offers the ability to examine estimates based on demographic and socioeconomic variables.

We chose to conduct phlebotomy in participants’ homes to reduce potential biases associated with sampling in specific locations that would require participants to travel to or only testing persons potentially more likely to be ill. Home specimen collection enabled us to include respondents who lacked access to transportation or health care, or who were uncomfortable leaving their home during the pandemic. We combined antibody test results for this survey with self-reported antibody results collected via telephone survey to increase our sample size to estimate stratified antibody prevalence with greater precision. Including self-reported test results also helped to minimize selection bias in 2 ways. First, it included individuals who previously had an antibody test and did not want the test to be repeated. During the time of this serosurvey, antibody testing was publicly available, including at several city-run mass serology sites and some people did not want to undergo repeat phlebotomy. Second, including self-reported results allowed us to survey individuals who preferred not having a phlebotomist visit their home during a pandemic, which was a common reason respondents declined specimen collection. While controlling for demographic variables, we found that respondents with only self-reported results had elevated yet nonsignificant odds of having a positive result compared with respondents that provided blood. We hypothesize that individuals who self-reported a positive antibody test might have been less inclined to repeat serology compared with those who self-reported a negative antibody test.

The validity of self-reported medical test results varies in the literature according to the disease of interest and the survey population [31–33]. In our study, among respondents with both serosurvey and self-reported antibody data, we found substantial agreement (89%) between an individual’s self-reported result and serosurvey tested result. The handful of discordant results had plausible biological and epidemiological explanations, including waning antibodies [34–36], COVID-19 infection after the initial self-reported test, reporting bias, or incorrect recall.

While initially SARS-CoV-2 seroprevalence was believed to be a proxy for cumulative infections [9], it is now well-documented that individuals with asymptomatic or mild SARS-CoV-2 infections might not produce antibodies, have limited antibody response, or have waning antibodies [20, 34, 36]. Other studies have tracked the decline in seroprevalence in NYC from April to July 2020 [5, 7]. Our survey did not require respondents with self-reported antibody results to recall the date of their previous antibody testing, nor the manufacturer of the serological assay. Survey respondents who self-reported a test result that was conducted between March 2020, when commercial SARS-CoV-2 assays became available [37], and our study period in June 2020, might have included individuals who were positive at the time of testing but seroconverted by June. Our antibody prevalence estimates likely underestimate cumulative infections overall; however, the self-reported data might overestimate antibody prevalence by including individuals who by June through October were no longer antibody positive. Additionally, considering the varied performance of different SARS-CoV-2 serologic assays, respondents with self-reported antibody results might have been tested with a low performing assay, increasing potential for false-positive or -negative test results [37].

Telephone survey response rates have been declining for years, which is consistent with the low 2020 CHS response rate [38]. One limitation of a low response is the potential for nonresponse bias; the prevalence among those that declined survey participation is unknown. While declining survey response rates indicate potential nonresponse bias, they are not a direct measure of nonresponse bias [39].

Additional limitations include the following: aside from respondents who provided both a specimen and self-reported result, the remaining self-reported test results were not further verified, so recall bias could not be assessed. Among those who provided a blood specimen, the median time between survey administration and blood draw was 16 days. Some subgroups with small sample sizes should be interpreted with caution. Finally, this survey included only noninstitutionalized adults; results cannot be extrapolated to other populations like children or those residing in congregate settings.

At the population level, representative cross-sectional surveys provide important data about the extent of the pandemic. While other research is ongoing to understand immunity against the SARS-CoV-2 virus, our population-based prevalence estimates help inform an understanding of which NYC populations are most at risk and those that might still be susceptible to SARS-CoV-2 infection. This work offers an important baseline of SARS-CoV-2 antibody prevalence in NYC following the first wave of the COVID-19 outbreak and before mass vaccination. A unique feature of our survey is that it uses a hybrid approach to estimate SARS-CoV-2 antibody prevalence using a representative population-based sample of NYC residents. The findings highlight the considerable SARS-CoV-2 transmission in NYC, particularly in black and Latino populations, and communities with high neighborhood poverty, strengthening evidence of how structural racism led to an unequal burden of COVID-19 in NYC. Future analyses should further examine individual and neighborhood characteristics associated with having SARS-CoV-2 antibodies to identify upstream policy and public health levers for more equitable and targeted public health interventions.

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

Acknowledgment. The authors acknowledge all survey participants for their time and participation.

Disclaimer. The contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC or the Department of Health and Human Services.

Financial support. This work was supported in part by the Epidemiology and Laboratory Capacity for Infectious Diseases Cooperative Agreement (grant number ELC CARE 6 NU50CK000517-01-09) funded by the Centers for Disease Control and Prevention.

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References

Author notes