Accuracy of Case-Based Seroprevalence of SARS-CoV-2 Antibodies in Maricopa County, Arizona

Abstract

We conducted a community seroprevalence survey in Arizona, from September 12 to October 1, 2020, to determine the presence of antibodies to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). We used the seroprevalence estimate to predict SARS-CoV-2 infections in the jurisdiction by applying the adjusted seroprevalence to the county’s population. The estimated community seroprevalence of SARS-CoV-2 infections was 4.3 times greater (95% confidence interval = 2.2, 7.5) than the number of reported cases. Field surveys with representative sampling provide data that may help fill in gaps in traditional public health reporting. (Am J Public Health. 2022;112(1):38–42. https://doi.org/10.2105/AJPH.2021.306568)

Although vaccination programs against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) continue to expand in scope, an unknown proportion of individuals in the United States continue to rely upon natural postinfection immunity for protection from reinfection. Reliable estimates of the number of people who have been infected with SARS-CoV-2 are therefore of substantial value for public health practice.

Traditional public health reporting may undercount COVID-19 cases.¹ Serological surveys to detect anti-SARS-CoV-2 antibodies can provide an estimate of the true population prevalence of past infection, including those missed by traditional public health reporting, because of asymptomatic infections for which health care or testing was not sought, or symptomatic infections in persons who did not seek care or on whom SARS-CoV-2 testing was not performed.² However, most previous US serosurveys of SARS-CoV-2 infection have examined the prevalence of SARS-CoV-2 antibodies in convenience samples or high-risk populations, which do not provide an accurate estimate of the prevalence of SARS-CoV-2 infection in a target population.³ Accurate estimates of the cumulative incidence of SARS-CoV-2 infection require minimally biased, population-based seroprevalence studies.

INTERVENTION

We used a 2-stage cluster probabilistic sampling design to conduct a community-level seroprevalence survey using the Community Assessment for Public Health Emergency Response (CASPER) toolbox from the US Centers for Disease Control and Prevention (CDC), a validated method for drawing a random sample of the population during public health emergencies.⁴ Census blocks were randomly selected with probability proportional to the number of occupied households (per 2010 US Census) without substitution (for methodology, see Appendix A, available as a supplement to the online version of this article at http://www.ajph.org). Household sampling was conducted in the field to draw a systematic sample of households within each census block. Selected households were approached and invited to participate in the serosurvey, which consisted of a standardized household questionnaire and blood sample for serology. More than 300 field volunteers were recruited and trained including Spanish speakers, nurses, public health staff, and student volunteers from academic programs that provide disciplinary attention to concepts of institutional and structural racism and bias and their impact on underserved and underrepresented communities.

We used a Roche Elecsys Anti-SARS-CoV-2 S assay to determine the presence of antibodies to SARS-CoV-2. This assay uses spike protein (total immunoglobulin) as the antigen for the detection of antibodies and has a reported specificity of 99.5% and sensitivity of 99.8%.⁵

PLACE AND TIME

The serosurvey was conducted in Maricopa County, Arizona, between September 12 and October 1, 2020, and excluded persons who were living on tribal lands or in congregate settings. Ninety-two percent of all reported cases in the county occurred within 12 weeks of the survey (see Appendix B, available as a supplement to the online version of this article at http://www.ajph.org, for sampling period).

PERSON

A total of 791 households were approached across 30 sampling blocks. Of the 587 households where contact was made, 173 households agreed to participate, resulting in a cooperation rate of 29.5%. A total of 260 persons from 169 households consented to serology testing. Compared with census data for the county, participants differed in terms of age distribution, language spoken at home, and household size (Table 1).

TABLE 1—

Unweighted Demographic Characteristics of Survey Participants With a SARS-CoV-2 Serology Test Result Compared With 2019 Postcensal Estimates for the Overall Catchment Area: Maricopa County, AZ, September 12–October 1, 2020

	Unweighted Participants	Weighted Sample	Catchment Areaa
Overall no.	260	4 090 940	4 485 414
Individual characteristics (n = 260)
Age group, y, no. (%)
5–19b	20 (7.7)	881 189 (21.5)	895 544 (20.0)
20–44	80 (30.8)	1 493 193 (36.5)	1 539 171 (34.3)
45–64	87 (33.5)	1 051 372 (25.7)	1 078 113 (24.0)
≥ 65	73 (28.1)	662 732 (16.2)	696 467 (15.5)
Gender, no. (%)
Male	105 (40.4)	2 016 833 (49.3)	2 217 116 (49.4)
Female	154 (59.5)	2 072 061 (50.7)	2 268 298 (50.6)
Household characteristics (n = 169)
Household size, mean	2.59	2.50	2.75
Non-English language spoken at home, no. (%)	35 (20.7)	1 278 191 (28.8)	1 210 791 (26.9)
Urbanicity,c no. (%)
Urban	166 (98.3)	3 951 848 (96.5)	3 725 506 (97.6)
Rural	3 (1.7)	143 183 (3.5)	91 611 (2.4)

Note. SARS-CoV-2 = severe acute respiratory syndrome coronavirus 2.

^a Age, gender, household size, and language spoken at home were obtained from American Community Survey 2019 1-year estimates.

^b Our sample included children aged > 6 years, and census data are only available for children aged 5–19 years.

^c Urbanicity was estimated using American Community Survey 2010 data.

PURPOSE

This serological survey was conducted to detect the presence of anti‒SARS-CoV-2 antibodies to estimate the true population prevalence of past infection and assess the accuracy of case-based surveillance.

IMPLEMENTATION

The intervention included both a serological sample obtained from all consenting household members aged older than 6 years and a questionnaire administered to the self-identified head of household or delegate. The questionnaire included questions about household and demographic characteristics, chronic medical conditions, recent illnesses and associated symptoms, previous testing for SARS-CoV-2, mental and financial impacts from the pandemic, and occupational exposures. Blood samples were collected onsite using standard venipuncture techniques, and samples were centrifuged in the field within 30 to 120 minutes of collection.

EVALUATION

Overall, 30 (11.5%) of 260 blood samples collected from 169 households were seropositive. The overall weighted seroprevalence of SARS-CoV-2 antibodies across Maricopa County was 14.0% (95% confidence interval [CI] = 7.2%, 24.0%) through August 27, 2020, in persons aged 7 years or older, corresponding to an estimated 589 156 (95% CI = 302 994, 1 009 982) individuals having been infected with SARS-CoV-2 through the time of the survey. As of September 9, 2021, a total of 136 193 cases of SARS-CoV-2 infection had been confirmed by conventional testing strategies and reported in Maricopa County resulting in a 4.3-fold difference (95% CI = 2.2, 7.5) between the estimated number of infections based on seroprevalence and reported case counts in line with other population-based estimates.^6–8

As compared with seronegative households, seropositive households were more likely to report a non-English language spoken in the home (weighted % = 37.5%; 95% CI = 18.3%, 61.7% vs weighted % = 16.5%; 95% CI = 9.1%, 27.9%; P = .02) and more likely to report that an occupant had been told by a health professional that they previously had COVID-19 compared with seronegative households (weighted % = 45.8%; 95% CI = 26.1%, 66.8% vs weighted % = 4.0%; 95% CI = 1.7%, 9.2%; P = <.001; Table 2). Mean household size was slightly larger for seropositive households as compared with seronegative households, although this was not statistically significant (3.1 vs 2.4; P = .18).

TABLE 2—

Weighted Demographic Characteristics of Households With and Without SARS-CoV-2 Antibodies: Maricopa County, AZ, September 12‒October 1, 2020

	Households With SARS-CoV-2 Antibodiesa (n = 23)		Households Without SARS-CoV-2 Antibodies (n = 146)		P b
No.	Weighted Proportion (95% CI)	No.	Weighted Proportion (95% CI)
Household characteristic
Non-English language spoken in home	8	37.5 (18.3, 61.7)	27	16.5 (9.1, 27.9)	.021
Illness history
COVID-19‒like Illness reported in householdc	14	59.4 (38.3, 77.5)	39	23.2 (15.6, 33.1)	.002
Proportion of households with symptomatic family members who were tested when ill	10	54.1 (21.2, 83.8)	16	37.1 (18.0, 61.4)	.038
Medical history
Presence of chronic conditions in householdd	10	50.0 (25.9, 74.1)	67	51.6 (39.7, 63.3)	.91
Previously tested for SARS-CoV-2
None	11	43.7 (24.2, 65.5)	97	66.4 (54.9, 76.2)	.09
Swab test	11	52.5 (31.6, 72.5)	44	31.4 (21.8, 42.9)	.1
Positive test reported in swab-tested household members	9	76.2 (35.3, 94.9)	4	7.1 (2.8, 16.9)	< .001
Household member ever told by physician that they had COVID-19	10	45.8 (26.1, 66.8)	7	4.0 (1.7, 9.2)	< .001
Work industry of immediate household members
Health care	4	15.0 (5.2, 36.2)	19	10.5 (7.2, 15.0)	.48

Note. CI = confidence interval; SARS-CoV-2 = severe acute respiratory syndrome coronavirus 2.

^a Seropositive households were defined as households with at least 1 participating household member who was antibody positive.

^b Characteristics of seropositive and seronegative households were compared using the Pearson χ² test with design-based correction for categorical variables and Wald test (F ratio) for continuous variables.

^c An illness was categorized as compatible with COVID-19 if participants responded “yes” to the following question: “Since January 1, 2020, has anyone in your household experienced an illness with fever, cough, difficulty breathing/shortness of breath, and/or loss of taste or smell?”

^d Chronic conditions defined as diabetes, pregnancy, chronic lung disease, moderate to severe asthma, obesity, chronic kidney disease, serious heart condition, or immunocompromised state (including cancer, HIV, transplants, and immunosuppressive medications).

Among households with anti‒SARS-CoV-2 antibodies, 59.4% (95% CI = 38.3%, 77.5%) reported a history of COVID-19‒like illness. Only 54.1% (95% CI = 21.2%, 83.8%) of those households who reported illness had been tested for SARS-CoV-2 infection. The most commonly reported reasons for not being tested were concerns about the cost (24%), uncertainty about where to get a test (24%), inaccessibility of testing sites (19%), difficulty scheduling (19%), concerns about eligibility for testing given reported scarcity (5%), and concerns about citizenship status (2%).

Limitations include the potential for self-selection bias, whereby some individuals (e.g., those with previous symptoms or a previous positive SARS-CoV-2 infection) may have elected to participate in no-cost serology testing at a higher rate than the general population. Second, our estimate of exposure to SARS-CoV-2 infection is likely an underestimate of the true infection rate given that the sensitivity of SARS-CoV-2 immunoassays varies depending on the severity of the initial infection and the timing of collection.⁹ However, the Roche assay has been shown to be highly sensitive (91.4%) up to 8 months after asymptomatic or mildly symptomatic infection with SARS-CoV-2.¹⁰ Third, because the CASPER methodology is designed to collect household-level data, we could not correlate individual-level illness history with individual serological results.

ADVERSE EFFECTS

We did not observe any adverse effects.

SUSTAINABILITY

In the absence of national coordination, state and local health officials have turned to a variety of sampling approaches to determine SARS-CoV-2 seroprevalence, which may under- or overestimate seroprevalence. National coordination would enable better resource utilization, improve efficiency, foster data harmonization, and facilitate formation of best practices.

PUBLIC HEALTH SIGNIFICANCE

Confirmed COVID-19 case counts do not capture the total burden of the pandemic. Only 54% of households in this sample with COVID-19‒like illness reported that family members were tested, suggesting that testing was not widely accessible, or testing was not perceived to be affordable to a significant proportion of the community. Our findings may help explain why persons who are members of minority, rural, and other underserved communities are often undercounted in traditional case-based public health surveillance. Our study and others¹¹ have shown that door-to-door COVID-19 interventions help reach groups disproportionately affected by COVID-19.

The COVID-19 pandemic is strongly shaped by deep social disparities, adverse living and working conditions, and structural inequities that drive household and occupational transmission opportunities and access to testing and medical care.¹² The disproportionate impact of COVID-19 that we observed in non‒English-speaking households highlights the importance of tailoring communication strategies to the cultures and languages of local communities. The finding that a significant proportion of households experiencing COVID-19 symptoms did not seek testing also suggests a need to engage communities to better inform COVID-19 decisions about how to reduce disparities in the allocation of publicly available testing and vaccination resources. Field surveys with representative sampling provide data that may help fill in gaps in traditional public health reporting to better contextualize and inform COVID-19 mitigation strategies.