Seroprevalence of SARS-CoV-2-Specific IgG Antibodies Among Adults Living in Connecticut: Post-Infection Prevalence (PIP) Study

Abstract

Background

A seroprevalence study can estimate the percentage of people with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibodies in the general population; however, most existing reports have used a convenience sample, which may bias their estimates.

Methods

We sought a representative sample of Connecticut residents, ages ≥18 years and residing in noncongregate settings, who completed a survey between June 4 and June 23, 2020, and underwent serology testing for SARS-CoV-2-specific immunoglobulin G (IgG) antibodies between June 10 and July 29, 2020. We also oversampled non-Hispanic black and Hispanic subpopulations. We estimated the seroprevalence of SARS-CoV-2-specific IgG antibodies and the prevalence of symptomatic illness and self-reported adherence to risk-mitigation behaviors among this population.

Results

Of the 567 respondents (mean age 50 [± 17] years; 53% women; 75% non-Hispanic white individuals) included at the state level, 23 respondents tested positive for SARS-CoV-2-specific antibodies, resulting in weighted seroprevalence of 4.0 (90% confidence interval [CI] 2.0-6.0). The weighted seroprevalence for the oversampled non-Hispanic black and Hispanic populations was 6.4% (90% CI 0.9-11.9) and 19.9% (90% CI 13.2-26.6), respectively. The majority of respondents at the state level reported following risk-mitigation behaviors: 73% avoided public places, 75% avoided gatherings of families or friends, and 97% wore a facemask, at least part of the time.

Conclusions

These estimates indicate that the vast majority of people in Connecticut lack antibodies against SARS-CoV-2, and there is variation by race and ethnicity. There is a need for continued adherence to risk-mitigation behaviors among Connecticut residents to prevent resurgence of COVID-19 in this region.

Clinical Significance.

• •Our results show that despite Connecticut having an early outbreak of coronavirus disease 2019 (COVID-19), a majority of people in Connecticut lack detectable antibodies to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and, as such, remain vulnerable to infection.
• •There is continued need for strong public health efforts encouraging Connecticut residents to continue adherence to risk-mitigation behaviors so as to prevent resurgence of the virus in the region.

Alt-text: Unlabelled box

Introduction

Connecticut was one of the first states in the United States to be severely affected by coronavirus disease 2019 (COVID-19), with its first confirmed case of COVID-19 in early March. While almost 43,000 cases and 4000 deaths were reported by June,¹ a seroprevalence study, which estimates the percentage of people with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibodies, may provide a more accurate estimate of the percentage of Connecticut population with evidence of a prior infection from COVID-19.

Prior seroprevalence studies have estimated the spread of COVID-19 in the United States.2, 3, 4, 5, 6, 7, 8 However, the majority have taken advantage of blood samples collected for other reasons or used a convenience sample, which limits their representativeness. The Centers for Disease Control and Prevention (CDC) conducted a seroprevalence survey in Connecticut using blood specimens collected at commercial laboratories.⁸ However, these specimens were produced as part of routine or sick visits, representing a biased sample. Moreover, this effort did not provide the reason for the blood collection or information about recent symptomatic illness, underlying conditions, or relevant risk-mitigation behaviors, which may help predict detection of antibodies against SARS-CoV-2.

Accordingly, with support from the Connecticut Department of Public Health (DPH) and the CDC, we conducted the Post-Infection Prevalence (PIP) Study, a public health surveillance project to determine the seroprevalence of SARS-CoV-2 among adults residing in community noncongregate settings in Connecticut before June. Specifically, we sought to understand prior spread at the state level; collect information about symptomatic illness, risk factors for virus infection, and self-reported adherence to risk-mitigation behaviors; compare our seroprevalence estimates to available Connecticut estimates; and provide targeted estimates for the non-Hispanic black and Hispanic populations.

Methods

Study Cohort

For the state-level seroprevalence estimate, from June 4 to June 23, 2020, we enrolled 735 adults residing in noncongregate settings (ie, excluding individuals living in long-term care facilities, assisted living facilities, nursing homes, and prisons or jails) in Connecticut, ages ≥18 years, using a dual-frame Random Digit Dial (RDD) methodology.⁹ Additionally, from June 23 to July 22, 2020, we oversampled non-Hispanic black (n = 269) and Hispanic (n = 341) individuals to provide more accurate estimates for these subpopulations. Details of the sample size calculation and RDD methodology are described in eMethods 1, available online. Details of participant recruitment are described in eMethods 2, available online. We contacted a total of 7305 respondents at the state level and successfully completed 735 interviews. We contacted a total of 12,508 respondents for the oversampled subpopulations, of whom 457 completed interviews.

The study was deemed not to be research by the institutional review board at Yale University because of the public health surveillance activity exclusion and was approved by the institutional review board at Gallup.

Survey Components

Individuals selected were provided study details, and informed consent was obtained from all participants by trained interviewers. Participants were interviewed using a questionnaire that collected information on demographics, social determinants of health, history of influenza-like-illness, symptoms experienced, and other COVID-19-related topics. The average survey time was 15 minutes.

Specimen Collection and Serology Testing

Within 24-48 hours of completing the interview, respondents were contacted to schedule their blood draw appointment at their nearest Quest Diagnostics Patient Service Center (PSC). Up to 5 attempts were made to each household where the participant agreed to be tested. On confirmation that the participant had completed the test, an incentive payment of $50 was sent as a gift card via email or mail. Beginning July 17, 2020, we offered participants an additional $50 (for a total compensation of $100) to incentivize completion of the serology test.

Of the 735 participants enrolled in the state-level estimate, 25 participants refused to participate when recontacted for scheduling and 567 participants completed serology testing at 93 Quest Diagnostics PSCs throughout Connecticut between June 10 and July 29, 2020 (eFigure 1 , available online). Of the total 341 Hispanic and 269 non-Hispanic black participants enrolled for the oversample estimate, 171 and 148 participants, respectively, completed serology testing (eFigure 2 , available online). The distribution of the timing of the blood draws is shown in eFigure 3 , available online.

eFigure 1.

Flowchart showing sample selection for the state-level estimate.

eFigure 2.

Flowchart showing sample selection for the oversample estimate.

eFigure 3.

Distribution of the timing of the blood draws between June 10 and July 29, 2020, for the state-level estimate (A) and the subpopulation estimate (B).

Sera were obtained from samples collected in BD Hemogard serum separator tubes. All samples were processed at the Quest Diagnostics Marlborough Laboratory. Samples were run at room temperature using the primary collection tube. We measured immunoglobulin G (IgG) SARS-CoV-2 antibodies using the Ortho-Clinical Diagnostics Vitros anti-SARS-CoV-2 IgG test, which detects antibodies against the spike glycoprotein of the virus.¹⁰ Antibody levels were expressed as the ratio of the chemiluminescence signal over the cutoff value, with a value ≥1.00 reported as positive.¹¹ The Ortho Vitros IgG test had a reported sensitivity and specificity of 90% and 100%, respectively.¹⁰ We validated the sensitivity of this test in a small subset of patients who were positive for SARS-CoV-2 (n = 36) with variable disease severity, using reverse transcription polymerase chain reaction testing as the gold standard.¹²

Additionally, given the concern about the accuracy of serology tests,¹³ we retested the negative samples from 5 high-risk cities of Connecticut (ie, Bridgeport, Hartford, New Haven, Stamford, and Waterbury) with the Abbott Architect SARS-CoV-2 IgG test that detects antibodies aimed at a different SARS-CoV-2 antigen (nucleocapsid protein).¹⁴

Finally, Quest Diagnostics provided results for all SARS-CoV-2 serology tests conducted throughout Connecticut in the same time period (ie, June 10 and July 29, 2020) for comparison.

Statistical Analysis

The sample data were weighted to approximate the Connecticut population (details described in eMethods 3, available online). Briefly, the base weight assigned to each completed survey was derived as the product of the inverse of the probability of selection and nonresponse adjustment. Next, poststratification weighting adjustments were made to account for residual nonresponse and to match the weighted sample estimates to known population characteristics for Connecticut. Poststratification weighting for the state-level sample was carried out using ranking (or Iterative Proportional Fitting) procedures to adjust for age, gender, race or ethnicity, and education. The categories chosen for weighting the oversample subpopulations were different from what was used for the state-level adjustments due to lower available sample sizes. To reduce the effect of extreme weights on sampling variance, final weights were trimmed. The margin of error (MOE) for this study was calculated at the 90% confidence level (CI) taking into consideration the design effect introduced by variability of weights on each survey estimate. Overall study design effect as estimated by the Kish approximation equals 1.83; however, it varies by each survey estimate.

Next, the unweighted seroprevalence was calculated for both the overall state-level sample and the oversampled non-Hispanic black and Hispanic subgroups. Finally, we estimated the weighted state-level seroprevalence and the MOE of these estimates, both overall and for subgroups with sufficient sample size. Subgroups with sample sizes < 30 were too small to calculate accurate estimates and were thus not reported. We also estimated the MOE at 95% CI for the state-level estimates as a secondary outcome. We reported the weighted seroprevalence for non-Hispanic black and Hispanic subgroups separately.

All statistical analyses were performed using SPSS 24.0 (SPSS, Inc. Chicago, Ill.) and R version 4.0.2. We considered 2-sided P values < 0.05 as statistically significant.

Results

Population Characteristics for the State-Level Sample

The final state-level sample included 567 respondents who completed both the survey and the serology test. The mean age of the weighted sample was 50.1 (± 17.2) years, 53% were women, and the majority (75%) were non-Hispanic white individuals. Other weighted and unweighted characteristics of the study sample are reported in Table 1 .

Table 1.

Sociodemographic and Clinical Characteristics of Adults Included in the Study for the State-Level Estimate

Characteristics	Unweighted N	Unweighted Proportion, %	Weighted Proportion, %	Target Percentage,* %
Overall	567	—	567	—
Age group, years
18-29	41	7.2	13.1	19.9
30-44	90	15.9	26.7	22.9
45-54	113	19.9	18.6	17.5
55-64	134	23.6	18.6	18.1
≥65	187	33.0	23.0	21.6
Sex
Men	244	43.0	46.6	48.1
Women	323	57.0	53.4	51.9
Race/ethnicity
Hispanic	49	8.6	13.0	14.4
Non-Hispanic White	470	82.9	74.9	69.4
Non-Hispanic Black	37	6.5	9.6	9.8
Non-Hispanic Asian	9	1.6	1.2	4.7
Non-Hispanic Other	5	0.9	1.7	1.7
Education level
Less than high school	7	1.2	3.7	9.3
High school or GED	79	13.9	33.2	27.4
Some college	131	23.1	23.9	26.5
Bachelor's degree or more	350	61.7	39.2	36.8
Income level
<$24,000	40	7.1	11.3	—
$24,000-$59,999	104	18.3	25.0	—
$60,000-$119,999	178	31.4	30.1	—
$120,000+	195	34.4	26.8	—
Don't know/Refused	50	8.8	6.8	—
Health insurance
Yes	554	97.7	95.3	94.0
No	13	2.3	4.7	6.0
Employment status
Employed full-time	263	46.4	45.2	63.8
Employed part-time	56	9.9	10.0
Unemployed	43	7.6	10.6	3.5
Retired/Student/Homemaker	175	30.9	25.5	—
Disabled	0	0.0	0.0	—
Unknown	30	5.3	8.6	—
Essential job (exempt from stay-at-home orders)
Yes	140	24.7	27.5	—
No	169	29.8	24.8	—
Don't know/Refused/Not employed	258	45.5	47.7	—
Region/county
Fairfield	126	22.2	25.2	25.8
Hartford	157	27.7	24.1	24.9
Litchfield	42	7.4	5.5	5.2
Middlesex	34	6.0	5.0	4.7
New Haven	131	23.1	24.3	24.1
New London	41	7.2	7.8	7.6
Tolland	20	3.5	4.5	4.4
Windham	16	2.8	3.6	3.3
Type of home
Mobile home	2	0.4	1.0%	—
Single family house or townhouse	447	78.8	69.7%	—
Apartment or condo	112	19.8	28.1	—
Group facility	2	0.4	0.3	—
Don't know/Refused	4	0.7	0.9	—
Self-reported health status				—
Excellent	177	31.2	29.7	—
Very good	223	39.3	32.0	—
Good	128	22.6	26.9	—
Fair	33	5.8	9.2	—
Poor	6	1.1	2.3	—
Chronic conditions				—
Diabetes	64	11.3	12.2	—
Asthma, COPD, or another lung disease	63	11.1	16.7	—
Heart disease	37	6.5	6.9	—
Cancer	72	12.7	10.7	—
High blood pressure	171	30.2	30.5	—
Immune compromised	46	8.1	8.5	—
Lived in Connecticut in past 12 weeks
<6 weeks	9	1.6	1.0	—
6-10 weeks	13	2.3	1.9	—
11-12 weeks	543	95.8	96.5	—
Don't know/Refused	2	0.4	0.6	—

^⁎Source for age, sex, race, ethnicity, education, employment, county targets: American Community Survey 2018. Source for health insurance: the Current Population Survey estimates, 2018. Target percentage is based on expected proportions for a perfectly random sample, based on credible external sources.

COPD = chronic obstructive pulmonary disease; GED = general educational development test.

Comparison of the unweighted demographic distribution of individuals who completed only the survey with those who completed both the survey and the antibody test has been provided in eTable 1, available online. Although the 2 groups were not significantly different in regional representation, a significantly higher number of younger, Hispanic and non-Hispanic black individuals did not complete blood testing. However, our weighted study sample was closer to the target sample in the distribution of subgroups by age, sex, race and ethnicity, education level, and health insurance (Table 1, available online).

eTable 1.

Comparison of Demographics (Age-Group, Race/Ethnicity, and Geographic Region) of Those Who Completed Only the Survey with Those Who Completed the Survey and the Serology Test for the State-Level Population

	Completed survey but not blood test (N = 168)		Completed survey and blood test (N = 567)
	N	Unweighted %	N	Unweighted %	P Value†
Region
Total	168	—	567	—	—
Fairfield	38	22.6	126	22.2	0.91
Hartford	41	24.4	157	27.7	0.40
Litchfield	10	6.0	42	7.4	0.52
Middlesex	12	7.1	34	6.0	0.59
New Haven	41	24.4	131	23.1	0.73
New London	16	9.5	41	7.2	0.33
Tolland	9	5.4	20	3.5	0.28
Windham	1	0.6	16	2.8	0.09
Race/Ethnicity
Total*	176	—	570	—
Hispanic	36	21.4	49	8.6	<0.001
Non-Hispanic White	94	56.0	470	82.9	<0.001
Non-Hispanic Black	32	19.0	37	6.5	<0.001
Non-Hispanic Asian	10	6.0	9	1.6	0.002
Non-Hispanic Other	4	2.4	5	0.9	0.12
Age
Total	167	—	565	—
18-29 years	34	20.2	41	7.2	<0.001
30-44 years	48	28.6	90	15.9	<0.001
45-54 years	23	13.7	113	19.9	0.07
55-64 years	26	15.5	134	23.6	0.02
≥ 65 years	36	21.4	187	33.0	0.004

^⁎Can have multiresponse for race/ethnicity so this sum may be higher than the total N.

^†P value at 95% confidence level.

Symptoms and Risk-Mitigation Behaviors at the State Level

As shown in Table 2 , fever, cough, sore throat, diarrhea, and new loss of taste or smell was reported by 9%, 18%, 10%, 16%, and 5% respondents, respectively, at some point between March and June. About 16% of individuals reported being tested for coronavirus previously, and of these, 12% reported testing positive.

Table 2.

Prevalence of Symptomatic Illness, Risk Factors for Possible Exposure, and Adherence to Social-Distancing Behaviors Since March 1, 2020, Among the State-Level Population

Characteristics	Unweighted N	Unweighted Proportion, %	Weighted Proportion, % (MOE)
Symptoms
Fever	43	7.6	8.7 (±2.7)
Cough	84	14.8	18.3 (±3.5)
Sore throat	54	9.5	10.1 (±2.7)
New loss of taste or smell	25	4.4	4.8 (±2.0)
Diarrhea	70	12.3	15.9 (±3.2)
Risk Factors/Behaviors
Received coronavirus test	90	15.9	15.7 (±3.5)
Tested positive for coronavirus	11	1.9	1.9 (±1.4)
Anyone in household (other than respondent) had symptoms of coronavirus	54	9.5	10.2 (±2.7)
Anyone in household (other than respondent) tested positive for coronavirus	16	2.8	3.7 (±1.8)
Avoided going to public places, such as stores or restaurants	422	74.4	72.8 (±4.1)
Avoided small gatherings of people, with family or friends	426	75.1	75.3 (±4.0)
Worked from home (among all respondents, regardless of employment status)	223	39.3	31.4 (±4.1)
Worn a mask on your face when outside your home	557	98.2	96.9 (±1.6)
Traveled by airplane	39	6.9	6.5 (±2.6)
Traveled using public transportation, such as bus or train	19	3.4	5.2 (±2.0)

MOE = margin of error at the 90% confidence level.

The majority of respondents reported following risk-mitigation practices, at least some of the time, since March, with 73% reporting having avoided public places and 75% reporting having avoided gatherings of family and friends. Notably, 97% of respondents reported wearing a mask outside their home at least part of the time. About 31% of all respondents reported having worked from home at least part of the time, representing 57% of working respondents. We compared the prevalence of symptomatic illness and risk-mitigation behaviors among individuals who completed only the survey with those who completed the survey and the antibody test in eTable 2.

eTable 2.

Comparison of Prevalence of Symptomatic Illness, Risk Factors for Possible Exposure, and Adherence to Social-Distancing Behaviors Since March 1, 2020, Among Those Who Completed Only the Survey with Those Who Completed the Survey and the Serology Test for the State-Level Population

	Completed survey but not blood draw, N = 168		Completed survey and blood draw, N = 567
Characteristics	Unweighted N*	Unweighted %	Unweighted N*	Unweighted %	P Value†
Symptoms
Fever	16	9.5	43	7.6	0.42
Cough	20	11.9	84	14.8	0.34
Sore throat	18	10.7	54	9.5	0.65
New loss of taste or smell	3	1.8	25	4.4	0.12
Diarrhea	19	11.3	70	12.3	0.72
Risk Factors/Behaviors
Received coronavirus test	32	19.0	90	15.9	0.33
Tested positive for coronavirus	7	4.2	11	1.9	0.10
Anyone in household (other than respondent) had symptoms of coronavirus	17	10.1	54	9.5	0.82
Avoided going to public places, such as stores or restaurants	129	76.8	422	74.4	0.54
Avoided small gatherings of people, with family or friends	129	76.8	426	75.1	0.66
Worked from home (among all respondents, regardless of employment status)	35	20.8	223	39.3	<0.001
Worn a mask on your face when outside your home	159	94.6	557	98.2	0.01
Traveled by airplane	15	8.9	39	6.9	0.37
Traveled using public transportation, such as bus or train	15	8.9	19	3.4	<0.001

^⁎N for “yes” response.

^†P value at 95% confidence level.

Seroprevalence of SARS-CoV-2 Antibodies at the State Level

Seroprevalence estimates are shown in Table 3 . Overall, 23 respondents tested positive for SARS-CoV-2 antibodies, yielding a weighted seroprevalence of 4.0% (90% CI 2.0-6.0). Among individuals who reported having symptomatic illness, those with fever, cough, sore throat, and diarrhea had a weighted seroprevalence of 32.4% (90% CI 15.1-49.7), 11.4% (90% CI 2.8-20.0), 10.3% (90% CI 0.0-21.0), and 6.9% (90% CI 0.0-14.4), respectively. Among the 25 individuals who reported loss of taste or smell, 14 tested positive for SARS-CoV-2-specific antibodies.

Table 3.

Unweighted and Weighted State-Level Seroprevalence of SARS-CoV-2-Specific IgG Antibodies Among Adults in Connecticut, Overall and by Symptoms and Risk Factors and Behaviors

Characteristics	Sample Size, N	Unweighted Seroprevalence, N (%)	Weighted Seroprevalence, % (MOE)
Overall	567	23 (4.1)	4.0 (±2.0)
Race/Ethnicity
Hispanic	49	3 (6.1)	12.8 (±8.0)
Non-Hispanic White	470	16 (3.4)	2.7 (±1.7)
Non-Hispanic Black	37	3 (8.1)	2.6 (±4.7)
Non-Hispanic Asian	9	*	*
Non-Hispanic Other	5	*	*
Symptoms
Fever	43	14 (32.6)	32.4 (±17.3)
Cough	84	11 (13.1)	11.4 (±8.6)
Sore throat	54	5 (9.3)	10.3 (±10.7)
New loss of taste or smell†	25	*	*
Diarrhea	70	5 (7.1)	6.9 (±7.5)
Symptoms Aggregate
Asymptomatic	410	5 (1.2)	0.6 (±0.7)
1 or more symptoms	157	18 (11.5)	11.3 (±5.9)
2 or more symptoms	67	13 (19.4)	16.1 (±11.2)
Risk Factors/Behaviors
Received coronavirus test	90	13 (14.4)	19.5 (±9.5)
Tested positive for coronavirus†	11	*	*
Anyone in household (other than respondent) had symptoms of coronavirus	54	12 (22.2)	19.8 (±11.8)
Anyone in household (other than respondent) tested positive for coronavirus	16	*	*
Avoided going to public places, such as stores or restaurants	422	17 (4.0)	4.8 (±2.4)
Avoided small gatherings of people, with family or friends	426	17 (4.0)	4.6 (±2.4)
Worked from home (among all respondents, regardless of employment status)	223	14 (6.3)	4.2 (±2.3)
Worn a mask on your face when outside your home	557	23 (4.1)	4.1 (±2.0)
Traveled by airplane	39	0 (0.0)	0.0
Traveled using public transportation, such as bus or train	19	*	*

^⁎Sample size is < 30 and too small to report.

^†Though the sample size was too small to report seroprevalence estimates, all 11 of these individuals tested positive for SARS-CoV-2-specific IgG antibodies. Among the 25 individuals who reported loss of taste or smell, 14 tested positive for SARS-CoV-2-specific IgG antibodies.

IgG = immunoglobulin G; MOE = margin of error at the 90% confidence level.

Asymptomatic individuals had significantly lower weighted seroprevalence 0.6% (90% CI 0.0-1.3) compared with the overall state estimate, whereas those with ≥ 1 and ≥ 2 symptoms had a seroprevalence of 11.3% (90% CI 5.4-17.2) and 16.1% (90% CI 4.9-27.3), respectively (Table 3). The comparisons between other subgroups and the state estimates are presented in eTable 3, available online. Additionally, seroprevalence estimates at 95% MOE have also been shown in eTable 3, available online.

eTable 3.

Unweighted and Weighted State-Level Seroprevalence of SARS-Cov-2-Specific IgG Antibodies Among Adults in Connecticut, by Sociodemographic and Clinical Subgroups

Characteristics	Sample Size, N	Unweighted Seroprevalence, N (%)	Weighted Seroprevalence %, (MOE at 90% CI)	Weighted Seroprevalence %, (MOE at 95% CI)
Overall	567	23 (4.1)	4.0 (±2.0)	4.0 (±2.3)
Age group, years
18-29	41	2 (4.9)	6.4 (±7.7)	6.4 (±9.2)
30-44	90	4 (4.4)	4.9 (±4.6)	4.9 (±5.5)
45-54	113	9 (8.0)	6.6 (±4.7)	5.5 (±5.6)
55-64	134	6 (4.5)	2.6 (±2.4)	2.6 (±2.9)
≥65	187	2 (1.1)	0.8 (±1.2)	0.8 (±1.4)
Sex
Men	244	8 (3.3)	2.5 (±2.4)	2.5 (±2.9)
Women	323	15 (4.6)	5.3 (±2.9)	5.3 (±3.5)
Race/ethnicity
Hispanic	49	3 (6.1)	12.8 (±8.0)	12.8 (±9.6)
Non-Hispanic White	470	16 (3.4)	2.7 (±1.7)	2.7 (±2.0)
Non-Hispanic Black	37	3 (8.1)	2.6 (±4.7)	2.6 (±5.6)
Non-Hispanic Asian	9	*	*	*
Education level
Less than high school	7	*	*	*
High school or GED	79	3 (3.8)	5.0 (±4.1)	5.0 (±4.9)
Some college	131	4 (3.1)	4.0 (±3.6)	4.0 (±4.3)
Bachelor's degree or more	350	16 (4.6)	3.6 (±1.7)	3.6 (±2.1)
Income level
<$24,000	40	2 (5.0)	8.5 (±7.4)	8.5 (±8.8)
$24,000 to $59,999	104	2 (1.9)	3.0 (±3.7)	3.0 (±4.4)
$60,000 to $119,999	178	9 (5.1)	4.7 (±3.5)	4.7 (±4.2)
$120,000+	195	10 (5.1)	3.3 (±2.2)	3.3 (±2.6)
Don't know/Refused	50	0 (0.0)	0.0	0.0
Health insurance
Yes	554	20 (3.6)	3.1 (±1.8)	3.1 (±2.2)
No	13	*	*	*
Employment status
Employed full-time	263	13 (4.9)	2.6 (±1.7)	2.6 (±2.0)
Employed part-time	56	7 (12.5)	15.1 (±10.1)	15.1 (±12.0)
Unemployed	43	1 (2.3)	5.4 (±5.7)	5.4 (±6.8)
Retired	152	1 (0.7)	0.2 (±0.6)	0.2 (±0.7)
Homemaker	15	*	*	*
Student	8	*	*	*
Disabled	0	*	*	*
Unknown	30	*	*	*
Essential job (exempt from stay-at-home orders)
Yes	140	8 (5.7)	5.3 (±4.3)	5.3 (±5.2)
No	169	12 (7.1)	5.0 (±2.8)	5.0 (±3.4)
Don't know/refused	10	*	*	*
Not employed	248	3 (1.2)	3.0 (±2.1)	3.0 (±2.5)
Region/County
Fairfield	126	9 (7.1)	5.7 (±5.4)	5.7 (±6.4)
Hartford	157	5 (3.2)	4.0 (±3.6)	4.0 (±4.3)
Litchfield	42	2 (4.8)	1.6 (±3.2)	1.6 (±3.8)
Middlesex	34	*	*	*
New Haven	131	5 (3.8)	3.4 (±3.2)	3.4 (±3.8)
New London	41	1 (2.4)	1.7 (±3.3)	1.7 (±4.0)
Tolland	20	*	*	*
Windham	16	*	*	*
Type of home
Mobile home	2	*	*	*
Single family house/townhouse	447	20 (4.5)	4.1 (±2.2)	4.1 (±2.6)
Apartment or condo	112	2 (1.8)	3.9 (±3.1)	3.9 (±3.7)
Group facility	2	*	*	*
Unknown	4	*	*	*
Self-reported health status
Excellent	177	5 (2.8)	1.3 (±1.5)	1.3 (±1.7)
Very good	223	11 (4.9)	3.3 (±2.1)	3.3 (±2.5)
Good	128	4 (3.1)	3.4 (±3.5)	3.4 (±4.1)
Fair	33	1 (3.0)	4.3 (±5.8)	4.3 (±6.9)
Poor	6	*	*	*
Chronic conditions
Diabetes	64	2 (3.1)	4.3 (±4.7)	4.3 (±5.6)
Asthma, COPD, or another lung disease	63	2 (3.2)	4.2 (±4.9)	4.2 (±5.9)
Heart disease	37	0 (0.0)	0.0	0.0
Cancer	72	3 (4.2)	7.5 (±7.9)	7.5 (±9.4)
High blood pressure	171	4 (2.3)	2.9 (±2.8)	2.9 (±3.4)
Immune compromised	46	1 (2.2)	8.4 (±6.7)	8.4 (±8.0)
Individual symptoms
Fever	43	14 (32.6)	32.4 (±17.3)	32.4 (±20.6)
Cough	84	11 (13.1)	11.4 (±8.6)	11.4 (±10.3)
Sore throat	54	5 (9.3)	10.3 (±10.7)	10.3 (±12.7)
New loss of taste or smell	25	*	*	*
Diarrhea	70	5 (7.1)	6.9 (±7.5)	6.9 (±9.0)
Symptoms aggregate
Asymptomatic	410	5 (1.2)	0.6 (±0.7)	0.6 (±0.8)
1 or more symptoms	157	18 (11.5)	11.3 (±5.9)	11.3 (±7.0)
2 or more symptoms	67	13 (19.4)	16.1 (±11.2)	16.1 (±13.4)
Risk factors/behaviors
Received coronavirus test	90	13 (14.4)	19.5 (±9.5)	19.5 (±11.3)
Tested positive for coronavirus†	11	*	*	*
Anyone in household (other than respondent) had symptoms	54	12 (22.2)	19.8 (±11.8)	19.8 (±14.0)
Anyone in household (other than respondent) tested positive for coronavirus	16	*	*	*
Avoided going to public places, such as stores or restaurants	422	17 (4.0)	4.8 (±2.4)	4.8 (±2.8)
Avoided small gatherings of people, with family or friends	426	17 (4.0)	4.6. (±2.4)	4.6 (±2.8)
Worked from home (among all respondents, regardless of employment status)	223	14 (6.3)	4.2 (±2.3)	4.2 (±2.8)
Worn a mask on your face when outside your home	557	23 (4.1)	4.1 (±2.0)	4.1 (±2.4)
Traveled by airplane	39	0 (0.0)	0.0	0.0
Traveled using public transportation, such as bus or train	19	*	*	*

^⁎Sample size is < 30 and too small to report

^†Though the sample size was too small to report seroprevalence estimates, all 9 of these individuals tested positive for SARS-Cov-2-specific IgG antibodies.

Note 1: The rows highlighted in grey indicate estimates where we are confident at the 90% level that there is not a null result (estimate is not equal to 0). Though the results for the other rows are presented, the sample size is inadequate to be able to detect any significance.

CI = confidence interval; COPD = chronic obstructive pulmonary disease; GED = general educational development test; IgG = immunoglobulin G; MOE = margin of error.

Among the 143 negative samples from 5 high-risk cities in Connecticut that were retested with Abbott Architect serology assay, 142 (99.3%) samples tested negative. Additionally, of the total 25,274 antibody tests conducted by Quest Diagnostics in Connecticut during this time period, 2072 (8.4%) samples tested positive. Of the 11 respondents who reported testing positive for coronavirus, all tested positive for antibodies.

Characteristics and Seroprevalence Estimates Among Non-Hispanic Black and Hispanic Subpopulations

For the subpopulation estimate, the final sample included 171 Hispanic (39.9 [± 15.5] years and 51% women) and 148 non-Hispanic Black (46.4 [± 13.0] years and 56% women) adults (eTable 4, available online). Fever, cough, sore throat, diarrhea, and new loss of taste or smell was reported by 11%, 17%, 15% 10%, and 8%, respectively, of Hispanic participants and 4%, 10%, 5%, 4%, and 6%, respectively, of black participants (Table 4 ). About 37% of Hispanic and 31% of non-Hispanic black individuals reported receiving a coronavirus test previously and nearly 6% of Hispanic and 4% non-Hispanic black individuals reported testing positive for coronavirus. The prevalence of symptomatic illness and risk-mitigation behaviors among individuals who completed only the survey has been compared with those who completed both the survey and the antibody test in eTable 5 , available online.

eTable 4.

Sociodemographic and Clinical Characteristics of the Non-Hispanic Black and Hispanic Subpopulations Included in the Study

	Hispanic Subpopulation				Non-Hispanic Black Subpopulation
Characteristics	Unweighted N	Unweighted Proportion, %	Weighted Proportion, %	Target Percentage,* %	Unweighted N	Unweighted Proportion, %	Weighted Proportion, %	Target Percentage,* %
N	171				148
Age group, years
18-29	35	20.5	28.6	29.5	4	2.7	9.1	26.9
30-44	54	31.6	29.0	33.5	28	18.9	37.7	26.6
45-54	28	16.4	18.2	17.5	41	27.7	20.9	18.3
55-64	35	20.5	17.8	10.9	56	37.8	28.2	14.5
≥65	18	10.5	4.9	8.7	19	12.8	4.1	13.7
Sex
Men	68	39.8	48.6	49.4	53	35.8	44.4	46.5
Women	103	60.2	51.4	50.6	95	64.2	55.6	53.5
Don't know/Refused	0	0.0	0.0	—	0	0.0	0.0	—
Education level
Less than high school	20	11.7	19.0	26.5	2	1.4	1.0	12.4
High school or GED	45	26.3	44.4	33.1	31	20.9	30.6	35.0
Some College	40	23.4	21.1	25.6	50	33.8	47.1	33.2
Bachelor's degree or more	65	38.0	15.1	14.9	63	42.6	20.2	19.4
Don't know/Refused	1	0.6	0.4	—	2	1.4	1.0	—
Income level
<$24,000	37	21.6	30.6	—	15	10.1	13.9	—
$24,000 to $59,999	59	34.5	37.9	—	50	33.8	41.5	—
$60,000 to $119,999	33	19.3	12.1	—	45	30.4	22.7	—
$120,000+	28	16.4	9.7	—	29	19.6	12.5	—
Don't know/Refused	14	8.2	9.7	—	9	6.1	9.3	—
Health insurance
Yes	155	90.6	80.5	81.4	141	95.3	91.8	91.6
No	16	9.4	19.5	18.6	7	4.7	8.2	8.4
Unknown	0	0.0	0.0	—	0	0.0	0.0	—
Employment status
Employed full-time	80	46.8	40.1	64.7	72	48.6	39.3	62.1
Employed part-time	21	12.3	17.3		19	12.8	13.2
Unemployed	24	14.0	18.1	6.6	18	12.2	25.2	7.4
Retired/Student/Homemaker	33	19.3	16.7	—	21	14.2	8.7	—
Disabled	0	0.0	0.0	—	0	0.0	0.0	—
Unknown	13	7.6	7.8	—	18	12.2	13.7	—
Essential job (exempt from stay-at-home orders)
Yes	65	38.0	37.9	—	43	29.1	27.6	—
No	33	19.3	17.9	—	47	31.8	24.4	—
Don't know/Refused/Not Employed	73	42.7	44.2	—	58	39.2	48.1	—
Region/county
Fairfield	40	23.4	29.8	33.2	25	16.9	22.1	27.7
Hartford	63	36.8	28.4	27.9	64	43.2	34.2	32.2
Litchfield	3	1.8	4.2	1.9	1	0.7	1.8	0.9
Middlesex	2	1.2	0.3	1.7	0	0.0	0.0	2.3
New Haven	49	28.7	26.9	26.5	55	37.2	41.3	29.7
New London	5	2.9	0.8	4.9	1	0.7	0.4	4.9
Tolland	4	2.3	1.2	1.6	2	1.4	0.1	1.5
Windham	5	2.9	8.3	2.3	0	0.0	0.0	0.8
Unknown	0	0.0	0.0	—	0	0.0	0.0	—
Type of home
Mobile home	0	0.0	0.0	0.1	1	0.7	0.4	0.1
Single family house or townhouse	89	52.0	42.7	41.7	95	64.2	49.3	44.8
Apartment or condo	78	45.6	55.9	54.8	46	31.1	43.7	48.3
Group facility	1	0.6	0.2	3.3	1	0.7	0.4	6.8
Don't know/Refused	3	1.8	1.2	—	5	3.4	6.2	—
Self-reported health status
Excellent	41	24.0	26.7	—	25	16.9	25.2	—
Very good	46	26.9	20.8	—	50	33.8	28.5	—
Good	64	37.4	35.3	—	60	40.5	37.2	—
Fair	17	9.9	15.5	—	12	8.1	7.1	—
Poor	3	1.8	1.8	—	1	0.7	2.0	—
Unknown	0	0.0	0.0	—	0	0.0	0.0	—
Chronic conditions
Diabetes	34	19.9	19.3	—	40	27.0	26.6	—
Asthma, COPD, or another lung disease	28	16.4	15.6	—	29	19.6	21.8	—
Heart disease	10	5.8	4.1	—	8	5.4	3.0	—
Cancer	8	4.7	5.6	—	7	4.7	4.1	—
High blood pressure	42	24.6	25.4	—	69	46.6	44.7	—
Immune compromised	12	7.0	6.9	—	7	4.7	6.1	—
Lived in Connecticut in past 12 weeks
<6 weeks	1	0.6	0.2	—	1	0.7	0.5	—
6-10 weeks	6	3.5	1.8	—	0	0.0	0.0	—
11-12 weeks	162	94.7	97.3	—	145	98.0	98.5	—
Don't know/Refused	2	1.2	0.7	—	2	1.4	1.0	—

^⁎Source for age, sex, race, ethnicity, education, employment, county targets: American Community Survey 2018. Source for health insurance: Reference information for health insurance coverage is obtained from the Current Population Survey estimates, 2018. Target percentage is based on expected proportions for a perfectly random sample, based on credible external sources.

COPD = chronic obstructive pulmonary disorder; GED = general educational development test.

Table 4.

Prevalence of Symptomatic Illness, Risk Factors for Possible Exposure, and Adherence to Social-Distancing Behaviors Since March 1, 2020, Among Non-Hispanic Black and Hispanic Subpopulation

	Hispanic Subpopulation			Non-Hispanic Black Subpopulation
Characteristics	Unweighted N	Unweighted Proportion, %	Weighted Proportion, % (MOE)	Unweighted N	Unweighted Proportion, %	Weighted Proportion, % (MOE)
Overall	171	—	—	148	—	—
Symptoms
Fever	16	9.4	10.8 (±5.6)	7	4.7	3.9 (±3.6)
Cough	31	18.1	17.4 (±6.4)	18	12.2	10.1 (±6.4)
Sore throat	30	17.5	15.0 (±6.1)	8	5.4	4.7 (±4.6)
New loss of taste or smell	15	8.8	7.8 (±4.3)	8	5.4	4.2 (±3.8)
Diarrhea	25	14.6	10.2 (±5.5)	11	7.4	5.7 (±4.7)
Risk Factors/Behaviors
Received coronavirus test	64	37.4	36.9 (±9.6)	54	36.5	31.0 (±10.5)
Tested positive for coronavirus (out of all participants, regardless of prior testing)	10	5.8	6.2 (±3.8)	9	6.1	3.9 (±4.7)
Anyone in household (other than respondent) had symptoms of coronavirus	28	16.4	20.6 (±6.8)	6	4.1	2.1 (±2.1)
Anyone in household (other than respondent) tested positive for coronavirus	15	8.8	9.3 (±4.6)	3	2.0	2.0 (±2.7)
Avoided going to public places, such as stores or restaurants	139	81.3	79.2 (±7.6)	96	64.9	63.8 (±10.5)
Avoided small gatherings of people, with family or friends	140	81.9	81.6 (±6.9)	108	73.0	75.4 (±9.2)
Worked from home (among all respondents, regardless of employment status)	42	24.6	11.8 (±5.7)	48	32.4	17.8 (±9.0)
Worn a mask on your face when outside your home	168	98.2	97.7 (±2.8)	145	98.0	96.5 (±4.0)
Traveled by airplane	11	6.4	4.8 (±3.3)	6	4.1	4.0 (±4.8)
Traveled using public transportation, such as bus or train	11	6.4	13.1 (±5.5)	16	10.8	23.7 (±7.5)

MOE = margin of error at the 90% confidence level.

eTable 5.

Comparison of Prevalence of Symptomatic Illness, Risk Factors for Possible Exposure, and Adherence to Social-Distancing Behaviors Since March 1, 2020, Among Those Who Completed Only the Survey with Those Who Completed the Survey and the Serology Test for the Hispanic and Non-Hispanic Black Subpopulations

	Hispanic Subpopulation					Non-Hispanic Black Subpopulation
	Completed survey but not blood draw, N = 170		Completed survey and blood draw, N = 171			Completed survey but not blood draw, N = 121		Completed survey and blood draw, N = 148
Characteristics	Unweighted N*	Unweighted %	Unweighted N*	Unweighted %	P value†	Unweighted N*	Unweighted %	Unweighted N*	Unweighted %	P value†
Symptoms
Fever	14	8.2	16	9.4	0.71	8	6.6	7	4.7	0.50
Cough	19	11.2	31	18.1	0.07	15	12.4	18	12.2	0.95
Sore throat	19	11.2	30	17.5	0.09	10	8.3	8	5.4	0.35
New loss of taste or smell	11	6.5	15	8.8	0.42	7	5.8	8	5.4	0.89
Diarrhea	13	7.6	25	14.6	0.04	17	14.0	11	7.4	0.08
Risk Factors/Behaviors
Received coronavirus test	40	23.5	64	37.4	0.01	32	26.4	54	36.5	0.08
Tested positive for coronavirus	6	3.5	10	5.8	0.31	4	3.3	9	6.1	0.29
Anyone in household (other than respondent) had symptoms of coronavirus	14	8.2	28	16.4	0.02	7	5.8	6	4.1	0.51
Avoided going to public places, such as stores or restaurants	134	78.8	139	81.3	0.57	79	65.3	96	64.9	0.94
Avoided small gatherings of people, with family or friends	139	81.8	140	81.9	0.98	88	72.7	108	73.0	0.96
Worked from home (among all respondents, regardless of employment status)	31	18.2	42	24.6	0.15	22	18.2	48	32.4	0.01
Worn a mask on your face when outside your home	165	97.1	168	98.2	0.47	117	96.7	145	98.0	0.51
Traveled by airplane	8	4.7	11	6.4	0.49	8	6.6	6	4.1	0.35
Traveled using public transportation, such as bus or train	19	11.2	11	6.4	0.12	17	4.0	16	10.8	0.42

^⁎N for “yes” response.

^†P value at 95% confidence level.

The weighted seroprevalence among the Hispanic and non-Hispanic black subpopulation, derived from both the random state sample and the oversample, was 19.9% (90% CI 13.2-26.6) and 6.4% (90% CI 0.9-11.9), respectively. The seroprevalence estimate for the Hispanic group was significantly higher than the overall state-level estimate.

Discussion

Our study primarily shows that despite Connecticut being an early COVID-19 hotspot, the vast majority of people in Connecticut lack detectable antibodies to SARS-CoV-2. In addition, individuals who reported having symptomatic illness between March and June of 2020 had higher seroprevalence rates, but more than 90% of these individuals did not have SARS-CoV-2-specific IgG antibodies. Also, a high percentage of people interviewed reported following risk-mitigation strategies, which may be partly responsible for the reduction in the number of new COVID-19 cases being reported in Connecticut. Finally, the Hispanic subpopulation had a higher prevalence of SARS-CoV-2-specific antibodies as compared with the overall state-level estimate, suggesting that the burden of disease was higher in this subgroup.

Our findings are consistent with other reports of more selected Connecticut populations. The CDC conducted a seroprevalence study using commercial laboratory data and reported a seroprevalence of 4.9% (95% CI 3.6-6.5) between April 26 and May 3 and 5.2% (95% CI 3.8-6.6) between May 21 and May 26 in Connecticut.^{2 , 8} However, these estimates were from people who had blood specimens tested for reasons unrelated to COVID-19, such as for a routine or sick visit, and as such would be expected to be biased higher than estimates for the general population. Similarly, data for all antibody tests conducted by Quest Diagnostics in Connecticut between June 10 and July 29 showed a seropositivity rate of 8.4%. Because these estimates were also among people who had a serology test done at a commercial laboratory, it is likely that these specimens were drawn from individuals who were more likely to suspect prior disease exposure than the general population.

Overall, our findings are consistent with other reports of population-level seroprevalence of SARS-CoV-2 in Europe and the United States, although the burden of disease in these regions may have varied. Recent data report from Spain¹⁵ indicated a seroprevalence of 4.6% (95% CI 4.3-5.0), and a population-based study from Switzerland¹⁶ reported SARS-CoV-2 antibodies in <10% of the population. Reports from regions within the United States have also shown similar numbers. A recent report from Indiana⁵ found a seropositivity rate of 1.0% (95% CI, 0.8-1.5), and a community seroprevalence survey from Atlanta⁴ estimated seroprevalence of 2.5% (95% CI 1.4-4.5). Our findings of a higher burden of SARS-CoV-2 antibodies among Hispanic subgroups is also consistent with reports demonstrating that minority populations have been disproportionately affected by COVID-19.^{5 , 17}

There are several explanations for why our state-level estimates are lower than what one might expect given that Connecticut had nearly 43,000 positive cases and 4000 COVID-19 deaths by June 1, 2020. First, the majority of those deaths were among residents of congregate facilities. Second, the response and serology testing rates may have influenced the result. Only 7% of those contacted by phone completed the survey and blood test, and the recruited population differed from the targets. However, this is a standard response rate in studies seeking representative populations and was considered in weighting the data. It is also possible that those who were more likely to have a positive test failed to complete the blood draw in higher proportions. However, this nonresponse was taken into account while weighting the sample. Third, there is some evidence suggesting a short-lived antibody response, especially among individuals with mild or asymptomatic illness,^{18 , 19} and it is possible that more people were infected who lost antibodies over time. However, recent studies suggest that the decline in this timeframe is small and antibody levels can remain stable for up to 120 days,^{20 , 21} and all 11 people who reported receiving a positive coronavirus test previously, tested positive for antibodies in our study. Fourth, the accuracy of the serology tests has been a concern.¹³ However, 99% of the negative serology samples from the highest-risk regions of Connecticut that we retested with Abbott Architect serology assay tested negative a second time.

Nevertheless, our findings are concordant with other studies in indicating that the vast majority of the population in Connecticut does not have detectable levels of antibodies against SARS-CoV-2. At present, we do not know whether anti-SARS-CoV-2 antibodies confer immunity. If such antibodies, as detected by enzyme-linked immunosorbent assay (ELISA), are a marker of immunity, then more than 95% of the people in Connecticut would be susceptible to the virus. Given low infection rates over the summer, these general estimates are still reasonable. As such, there is continued need for strong public health efforts encouraging Connecticut residents to adhere to risk-mitigation behaviors so as to prevent a second wave of spread in the region.

Conclusion

Our findings indicate that even in one of the early hotspots of the SARS-CoV-2 outbreak in the United States, most of the population does not have detectable antibodies against SARS-CoV-2 and, as such, remains vulnerable to infection. Also, there is notable variation by race and ethnicity. People likely need to continue to be vigilant about practices that can slow the spread to prevent resurgence of the virus in these regions.

Supplemental Appendix

eMethods 1. Calculation of sample size.

In designing the study, sample size calculation has an important role to detect an effect and to achieve the desired precision in estimates of the prevalence of (SARS-CoV-2). Early research has shown that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is more prevalent in minority groups and in older populations, which needs to be accounted for while estimating sample size requirements. For the purposes of baselining, we estimated having to recruit a total of 1460 respondents to go through testing. The assumptions underlying the state-level estimate is:

• •For state-level estimates: required sample size = 609 assuming state-level prevalence of 10%, precision requirement of 2%, and a confidence level of 90%.
• •For estimates among non-Hispanic black and Hispanic subpopulations: combined required sample size of 960, including those completed as part of the state sample, assuming a subpopulation prevalence of 20%, precision requirement of 3%; and confidence level of 90%.

eMethods 2. Details of data source and participant enrollment.

Data source: To achieve the required sample size for generating state-level estimates of seroprevalence, Gallup used a dual-frame Random Digit Dial (RDD)⁹ methodology to select a random sample of adults residing in noncongregate settings (ie, excluding individuals living in long-term care facilities, assisted living facilities, nursing homes, and prisons or jails), ages 18 years and older, living in Connecticut. Briefly, this involves drawing a random sample of landline and cell phone numbers from among all potential landline and cell phone numbers with valid area codes assigned to Connecticut. The source of the telephone numbers was the North American Numbering Plan that specifies which area codes and exchanges are assigned to Connecticut. A simple random sample of telephone numbers was drawn from the landline and cellphone frame with a goal to complete roughly 30% of the sample from the landline frame and the remaining 70% from the cellphone frame. The proportion in which landline and cell phone interviews were completed was designed to maximize coverage and representation of those with access to landline telephone only, cell phone only, and those having access to both landline and cell phones. Interviews were available to be conducted in either English or Spanish depending on the preference of the respondent.

To minimize the undercoverage resulting from residents having cell phone numbers that are not assigned to Connecticut, Gallup supplemented the RDD sample with a random sample of listed cellular numbers of residents with an address in Connecticut but a cell phone number with area code that is not assigned to Connecticut. This was done to ensure that residents of Connecticut who moved from another area in the country and ported their cellphone number were still eligible to participate in this survey. The combination of the different sample types was designed to ensure high coverage and representativeness of the state population and subgroups in an efficient manner.

To ensure adequate sample sizes for subpopulations of interest (black and Hispanic subpopulations), additional RDD samples were drawn from the cellphone frame of geographic areas with relatively higher expected incidence of these racial/ethnic groups. The incidence of these groups in the statewide RDD samples was not high enough to produce the corresponding targeted number of interviews, and hence, it was necessary to carry out targeted oversampling of these ethnic groups. Respondents recruited through these oversamples were screened to confirm that they actually belonged to these targeted groups. Although the plan was to complete 430 surveys for the black subpopulation and 320 surveys for the Hispanic subpopulation, the study was closed once 269 non-Hispanic black interviews and 341 Hispanic interviews were completed by cellphone using the oversamples.

Participant enrollment: Within households reached by landline, a random adult 18 or older from among all eligible adults in that household was selected using the “next birthday” method. For those reached via cell phone, we confirmed that we are speaking with an adult 18 years or older, living in Connecticut, and for the oversample that they meet the race and ethnicity requirements. No respondent selection was done in the case of cell phone respondents, where 1 user per number was assumed.

To maximize the opportunity for the inclusion of harder to reach individuals, a multicall design was implemented whereby up to 5 attempts were made to each randomly selected telephone number, spread over different days of the week, including the weekend and different times of day to ensure we achieved a representative sample of adults. Scheduled “call-backs” to complete an interview with an eligible respondent at a later time or date agreed on between the interviewer and respondent were also executed.

If after 5 attempts, we were unable to make a human contact or encountered a refusal to participate, the number was retired. New sample replicates were released as numbers were retired and the selection of the respondent was started afresh with the new number.

For the state-level sample, out of the 7305 participants contacted, 1875 were via landline and 5430 via cell phone. Of the 735 participants who completed the survey, 224 were contacted via landline and 511 via cell phone. Out of 567 participants who completed the survey and the blood draw, 186 were contacted via landline and 381 via cell phone.

For the non-Hispanic black and Hispanic subpopulations, out of the 12,508 participants contacted, all 12,508 were over cell phone.

State-level response rate: We contacted a total of 7305 respondents at the state-level between June 4 and June 23 of 2020, and successfully completed 735 interviews resulting in a combined dual-frame Response Rate of 7.3%. Response rate calculated was RR3 as specified by the American Association of Public Opinion Research standard definition (p. 61): https://www.aapor.org/AAPOR_Main/media/publications/Standard-Definitions20169theditionfinal.pdf.

Response rate formula was defined as “completes” divided by “eligible” plus “presumed eligible” participants, and as such, the denominator contained more individuals than those actually contacted. Among those whom we could not contact, we estimate a certain proportion to be eligible based on what we observed among those we contacted. So, the actual denominator is more than 7305. We estimate, had we been able to continue calling indefinitely, we would have found a total of 9959 individuals eligible to participate in the study, given the number of phone numbers we were dialing.

eMethods 3. Details of weighting of study sample.

The sample data were weighted to minimize bias and thereby ensure that the results were representative of the target population of all adults in Connecticut. The base weight (or selection probability weight) assigned to each completed survey was derived as the inverse of the probability of selection of that respondent in the sample.

As described in the section on Sampling Methodology, a dual-frame (landline and cell phone) sample design was employed for this study. In addition, Gallup supplemented the Random Digit Dial (RDD) sample with a random sample of listed cellular numbers (hereafter referred to as the listed sample) of residents with an address in Connecticut but a cell phone number with area code that is not assigned to Connecticut.

The calculation of selection probabilities for each respondent was done taking into consideration his or her access to landline or cell phones. If someone had access to landline only, then selection probability was based on the landline frame count and the sample size drawn from that frame. If someone had access to cell phone only, similar calculations based on cell frame count and sample size were used. For dual users (with access to both landline and cell phones), their chances of selection from both frames were taken into consideration. If a respondent from the listed cell sample had access to landline phones, the selection probability calculation for that respondent included his or her chances of selection from the landline frame also. In other words, the selection probability was dependent on all possible ways that person could be selected in the sample and not based on how that respondent was actually selected in this sample.

The oversamples were weighted to minimize bias in the survey-based estimates. As described, specific geographic areas with relatively higher incidence of these groups were targeted for sampling, and hence, the selection probabilities were unequal. For every respondent, the selection probability was calculated taking into consideration the various ways that a respondent could be selected in the oversample. For any specific area, the total number of cell phone numbers in the sampling frame (frame count) and the corresponding sample size actually selected was known, and hence, it was possible to calculate the selection probability as the ratio of the sample size and the frame count. The base weight assigned to each respondent was the inverse of the selection probability.

Because not every respondent who completed the survey ended up going through the test, a separate nonresponse weighting step was first carried out (after computation of base weights) before using the poststratification process. For all respondents who completed the survey, a response propensity model was used to estimate the probability of them completing the test using a logistic regression model. The dependent variable in the logistic regression set up was the probability of completing the test (P) whereas the independent (or explanatory) variables included demographic and other survey variables such as symptoms, risk factors and behaviors, self-reported results of polymerase chain reaction (PCR) test, etc. For the state sample consisting of 735 respondents, 567 took the test and hence the information on whether they took the test or not was available for each respondent. Once the model was fit (using the stepwise regression method with inclusion criteria of 0.10 P value and stay criteria of 0.15 P value), the inverse of the estimated probability (P) was used as the nonresponse adjustment weighting factor. For each respondent who took the test, the weight up to this point (at the end of nonresponse adjustment process) was the product of base weight and the nonresponse adjustment factor and that was used as the input weight for the final poststratification weighting step.

For the Hispanic subpopulation consisting of 341 respondents, 171 took the test, and for the non-Hispanic black subpopulation consisting of 269 respondents, 148 took the test, and hence, information on whether they took the test or not was available for each respondent. Separate response propensity models were fit for the Hispanic and non-Hispanic black subpopulations and following the previously given description, the nonresponse adjustment weighting factor was estimated for each subpopulation.

The next step involved poststratification weighting adjustments to account for any residual nonresponse and to match the weighted sample estimates to known population characteristics for the state of Connecticut. For the state-level sample, poststratification weighting was carried out using raking (or iterative proportional fitting) procedures to adjust for demographic variables such as age, gender, race and ethnicity, and education. The different categories for these variables that were used for poststratification were as follows: age group (18-39, 40-49, 50-59, 60-69, and 70+), gender (male, female) race/ethnicity, (Hispanic, non-Hispanic white/Other, non-Hispanic black), education (high school or less, some college/no 4-year college degree, college graduate, postgraduate degree), and county of residence.

The distribution of the final weights (obtained after the poststratification weighting) were examined and some trimming of weights (7th percentile at the bottom and 97th percentile at the top) was carried out to avoid extreme weights and thereby reduce the effect of such weights on sampling variance.

For the oversamples, poststratification weighting adjustments were carried out to account for any residual nonresponse and to match the weighted sample estimates to known population characteristics for those subpopulations in the state of Connecticut. The categories chosen are different from what was used for the state-level adjustments due to lower available sample sizes.

For the Hispanic oversample, poststratification weighting was carried out using ranking as described previously. The categories that were used for poststratification were: age group (18-29, 30-39, 40-49, 50-59, 60+), gender (male, female), race (white Hispanic, other), education (high school graduate or less, some college/no 4-year college degree, college graduate and higher) and county of residence. The distribution of the final weights (obtained after the poststratification weighting) was examined and some trimming of weights (3rd percentile at the bottom and 98th percentile at the top) was carried out to avoid extreme weights and thereby reduce the effect of such weights on sampling variance.

For the non-Hispanic black oversample, poststratification weighting was carried out using ranking as described previously. The categories that were used for poststratification were: age group (18-39, 40-49, 50-59, 60+), gender (male, female), education (college degree or higher, rest), and county of residence. The distribution of the final weights (obtained after the poststratification weighting) was examined and some trimming of weights (3rd percentile at the bottom and 98th percentile at the top) was carried out to avoid extreme weights and thereby reduce the effect of such weights on sampling variance.

The margin of error (MOE) for a simple random sample, also known as precision for estimating the unknown population proportion, P, at the 90% confidence level can be derived based on the following formula:

$$\text{MOE}=\pm 1.645*\sqrt{P*\left(1-P\right)/n}$$

where n is the sample size (ie, the number of individuals who completed the survey and the blood test).

The MOE at the 95% confidence level can be derived based on the following formula:

$$\text{MOE}=\pm 1.96*\sqrt{P*\left(1-P\right)/n}$$

where n is the sample size (ie, the number of individuals who completed the survey and the blood test).

The weighting process resulted in unequal weights to correct for household selection with unequal probability of selection and nonresponse adjustments through poststratification weighting. This introduces a design effect that needs to be taken into account while computing precision of estimates. The design effect is defined as the ratio of the design-based sample variance to the sample variance obtained from a simple random sample of the same size. Overall study design effect as estimated by the Kish approximation equals 1.83. In addition, the design effect for every estimate is calculated and used in the construction of 90% confidence intervals using the below formula.

$$\text{MOE}=\pm 1.645*\sqrt{P*\left(1-P\right)/n}*\sqrt{d\mathrm{e}signeffect}$$

The design effect for every estimate was calculated and used in the construction of 95% confidence intervals using the below formula:

$$\text{MOE}=\pm 1.96*\sqrt{P*\left(1-P\right)/n}*\sqrt{d\mathrm{e}signeffect}$$

The design effect was defined formally by Kish as “the ratio of the actual variance of a sample to the variance of a simple random sample of the same number of elements.” Based on Kish's approximate formula:

$$\left\{\text{design}\text{effect}=n*\Sigma ·{\left(w_t\right)}^2/{\left(\Sigma w_t\right)}^2\right\}\oplus$$