Global seroprevalence of SARS-CoV-2 antibodies: A systematic review and meta-analysis.

BACKGROUND: Many studies report the seroprevalence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibodies. We aimed to synthesize seroprevalence data to better estimate the level and distribution of SARS-CoV-2 infection, identify high-risk groups, and inform public health decision making.
METHODS: In this systematic review and meta-analysis, we searched publication databases, preprint servers, and grey literature sources for seroepidemiological study reports, from January 1, 2020 to December 31, 2020. We included studies that reported a sample size, study date, location, and seroprevalence estimate. We corrected estimates for imperfect test accuracy with Bayesian measurement error models, conducted meta-analysis to identify demographic differences in the prevalence of SARS-CoV-2 antibodies, and meta-regression to identify study-level factors associated with seroprevalence. We compared region-specific seroprevalence data to confirmed cumulative incidence. PROSPERO: CRD42020183634.
RESULTS: We identified 968 seroprevalence studies including 9.3 million participants in 74 countries. There were 472 studies (49%) at low or moderate risk of bias. Seroprevalence was low in the general population (median 4.5%, IQR 2.4-8.4%); however, it varied widely in specific populations from low (0.6% perinatal) to high (59% persons in assisted living and long-term care facilities). Median seroprevalence also varied by Global Burden of Disease region, from 0.6% in Southeast Asia, East Asia and Oceania to 19.5% in Sub-Saharan Africa (p<0.001). National studies had lower seroprevalence estimates than regional and local studies (p<0.001). Compared to Caucasian persons, Black persons (prevalence ratio [RR] 3.37, 95% CI 2.64-4.29), Asian persons (RR 2.47, 95% CI 1.96-3.11), Indigenous persons (RR 5.47, 95% CI 1.01-32.6), and multi-racial persons (RR 1.89, 95% CI 1.60-2.24) were more likely to be seropositive. Seroprevalence was higher among people ages 18-64 compared to 65 and over (RR 1.27, 95% CI 1.11-1.45). Health care workers in contact with infected persons had a 2.10 times (95% CI 1.28-3.44) higher risk compared to health care workers without known contact. There was no difference in seroprevalence between sex groups. Seroprevalence estimates from national studies were a median 18.1 times (IQR 5.9-38.7) higher than the corresponding SARS-CoV-2 cumulative incidence, but there was large variation between Global Burden of Disease regions from 6.7 in South Asia to 602.5 in Sub-Saharan Africa. Notable methodological limitations of serosurveys included absent reporting of test information, no statistical correction for demographics or test sensitivity and specificity, use of non-probability sampling and use of non-representative sample frames. DISCUSSION: Most of the population remains susceptible to SARS-CoV-2 infection. Public health measures must be improved to protect disproportionately affected groups, including racial and ethnic minorities, until vaccine-derived herd immunity is achieved. Improvements in serosurvey design and reporting are needed for ongoing monitoring of infection prevalence and the pandemic response.

Introduction

Over one year has passed since the World Health Organization announced on January 30, 2020 that COVID-19 was a public health emergency of international concern, yet many questions persist about the spread and impact of the virus driving this crisis [1]. As of May 15, 2021, there were over 160 million confirmed cases of SARS-CoV-2 infection and 3.3 million deaths worldwide [2]. However, these case counts inevitably underestimate the true cumulative incidence of infection [3] because of limited diagnostic test availability [4], barriers to testing accessibility [5], and asymptomatic infections [6]. As a consequence, the global prevalence of SARS-CoV-2 infection remains unknown.

Serological assays identify SARS-CoV-2 antibodies, indicating previous infection in unvaccinated persons [7]. Population-based serological testing provides better estimates of the cumulative incidence of infection by complementing diagnostic testing of acute infection and helping to inform the public health response to COVID-19. Furthermore, as the world moves through the vaccine and variant era, synthesizing seroepidemiology findings is increasingly important to track the spread of infection, identify disproportionately affected groups, and measure progress towards herd immunity.

SARS-CoV-2 seroprevalence estimates are reported not only in published articles and preprints, but also in government and health institute reports, and media [8]. Consequently, few studies have comprehensively synthesized seroprevalence findings that include all of these sources [9, 10]. Describing and evaluating the characteristics of seroprevalence studies conducted over the first year of the pandemic may provide valuable guidance for serosurvey investigators moving forward.

We conducted a systematic review and meta-analysis of SARS-CoV-2 seroprevalence studies published in 2020. We aimed to: (i) describe the global prevalence of SARS-CoV-2 antibodies based on serosurveys; (ii) detect variations in seroprevalence arising from study design and geographic factors; (iii) identify populations at high risk for SARS-CoV-2 infection; and (iv) evaluate the extent to which surveillance based on detection of acute infection underestimates the spread of the pandemic.

Methods

Data sources and searches

This systematic review and meta-analysis was registered with PROSPERO (CRD42020183634), reported per PRISMA guidelines [11] (S1 File in S1 Materials), and will be regularly updated on an open-access platform (SeroTracker.com) [12].

We searched Medline, EMBASE, Web of Science, and Europe PMC, using a search strategy developed in consultation with a health sciences librarian (DL). The strategies for MEDLINE and EMBASE were an expanded version of the published COVID-19 search strategies created by OVID librarians for these databases [13]. Search terms related to serologic testing were identified by infectious disease specialists (MC, CY, and JP) [7] and expanded using Medical Subject Heading (MeSH) or Emtree thesauri. These searches were adapted for the other databases. The full search strategy can be found in S2 File in S1 Materials.

Given that many serosurveys are not reported in these databases [8] we used four additional search approaches to identify serosurveys reported in the grey literature. First, we searched for reports from national and international health agencies using their website search functions and examining their recurring COVID-19 reports (World Health Organization, European Centres for Disease Control, Centres for Disease Control, National Institutes of Health). Second, we searched Google News for reports of seroprevalence studies. When we encountered reports of potentially eligible government, non-governmental organizations (NGO), or academic studies, we conducted a targeted Google search to locate and include the full study. Updates of routinely reported NGO and government studies (e.g., Public Health England’s weekly COVID-19 serosurveillance reports) were screened after the date they first appeared in the Google News search. Third, we consulted with international experts via e-mail to identify additional literature after all other sources had been searched. Fourth, we invited submission of seroprevalence study results on our live dashboard—SeroTracker.com.

Our search dates were from January 1, 2020 to December 31, 2020. MedRxiv pre-print articles that were updated or published as peer-review articles between January 1, 2021 and February 28, 2021, according to the MedRxiv website, were also included. No restrictions on language were applied.

Study selection

We included SARS-CoV-2 serosurveys in humans. We defined a single serosurvey as the serological testing of a defined population over a specified time period to estimate the prevalence of SARS-CoV-2 antibodies [14, 15]. To be included, studies had to report a sample size, sampling date, geographic location of sampling, and prevalence estimate. Articles not in English or French were included if they could be fully extracted using machine translation [16]. Articles that provided information on two or more distinct cohorts (different sample frames or different samples at different time points) without a pooled estimate were considered to be multiple studies.

If multiple articles provided unique information about a study, both were included. Articles reporting identical information to previously included articles were excluded as duplicates–this rule extended to pre-print articles that were subsequently published are peer-reviewed journals. In these cases, the peer-reviewed articles were considered the definitive version.

We excluded studies conducted only in people previously diagnosed with COVID-19 using PCR, antigen testing, clinical assessment, or self-assessment; dashboards that were not associated with a defined serology study; and case reports, case-control studies, randomized controlled trials, and reviews.

Data extraction and quality assessment

Two authors independently screened articles. Data were extracted by one reviewer and verified by a second. We extracted characteristics of the study, sample, antibody test, and seroprevalence. We extracted sub-group seroprevalence estimates when they were stratified by one variable (e.g., age) but not two variables (e.g., age and sex). Antibody isotype and time period were not considered as stratifying variables. We contacted study authors to request missing sub-group seroprevalence data.

A modified Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Prevalence Studies was used to assess study risk of bias [17]. Studies were classified by overall risk of bias: low, moderate, high, or unclear (detailed criteria in S3 File in S1 Materials).

Data synthesis and analysis

Evaluation of seroprevalence studies and estimates

The intended geographic scope of each estimate was classified as (A) national; (B) regional (e.g., province-level); (C) local (e.g., county-level, city-level); or (D) sublocal (e.g., one hospital department). Countries were classified according to Global Burden of Disease (GBD) region, and country income status classified by distinguishing the high-income GBD region from other regions [18, 19].

Seroprevalence studies were grouped as providing either population-wide or population-specific estimates. Population-wide studies included those using household or community sampling frames as well as convenience samples from blood donors or residual sera used for monitoring other conditions in the population. Population-specific studies were those sampling from well-defined population sub-groups, such as health care workers or long-term care residents.

We prioritized estimates based on more accurate laboratory-based assays (e.g. ELISA, CLIA), as opposed to rapid diagnostic tests. We also prioritized estimates based on IgG and anti-spike antibodies, as non-IgG and anti-nucleocapsid antibodies appear to decline more rapidly than anti-spike/RBD IgG antibodies [20-25].

Data processing and descriptive statistics were conducted in Python. p-values less than 0.05 were considered statistically significant.

Correcting seroprevalence estimates

To account for imperfect test sensitivity and specificity, seroprevalence estimates were corrected using Bayesian measurement error models, with binomial sensitivity and specificity distributions [26]. The sensitivity and specificity values for correction were derived, in order of preference, from: (i) the FindDx -McGill database of independent evaluations of serological tests [27]; (ii) independent test evaluations conducted by serosurvey investigators and reported alongside serosurvey findings; (iii) manufacturer-reported sensitivity and specificity (including author evaluated in-house assays); (iv) published pooled sensitivity and specificity by immunoassay type [25]. If uncorrected estimates were not available, we used author-reported corrected seroprevalence estimates. Details of these evaluations are located in S4 File in S1 Materials.

We presented corrected and uncorrected estimates for all studies. Subsequent analyses were done using corrected seroprevalence estimates. To assess the impact of correction, we calculated the absolute difference between seroprevalence estimates before and after correction. We also conducted each analysis with uncorrected data.

Global seroprevalence and associated factors

To examine study-level factors affecting population-wide seroprevalence estimates, we constructed a multivariable linear meta-regression model. The outcome variable was the natural logarithm of corrected seroprevalence. Independent predictors were defined a priori. Categorical covariates were encoded as indicator variables, and included: study risk of bias (reference: low risk of bias), GBD region (reference: high-income); geographic scope (reference: national); and population sampled (reference: household and community samples). The sole continuous covariate was the cumulative number of confirmed cases in the country of the study. We obtained data on total confirmed SARS-CoV-2 infections [28, 29] and population size [30] that geographically matched the study populations nine days before the study end date, to reflect the time period between COVID-19 diagnosis and seroconversion (S5 File in S1 Materials) [31-33]. A quantile-quantile plot and a funnel plot were generated to visually check normality and homoscedasticity. All meta-analysis and meta-regression were done using the meta package in R [34].

Population differences in seroprevalence

To quantify population differences in SARS-CoV-2 seroprevalence, we identified subgroup estimates within population-wide studies that stratified by sex/gender, race/ethnicity, contact with individuals with COVID-19, occupation, and age groups. We calculated the ratio in prevalence between groups within each study (e.g., prevalence in males vs. females) then aggregated the ratios across studies using inverse variance-weighted random-effects meta-analysis (S4 File in S1 Materials). Heterogeneity was quantified using the I2 statistic [35].

Comparisons of seroprevalence and confirmed SARS-CoV-2 infections

To measure how much confirmed SARS-CoV-2 infections detected using RT-PCR underestimate seroprevalence, we calculated the ratio between population-wide seroprevalence estimates and the cumulative incidence of confirmed SARS-CoV-2 infections.

Results

Characteristics of included studies

We screened 24,999 titles and abstracts and 1,830 full text articles (Fig 1). We identified 968 unique seroprevalence studies in 605 articles. These studies included 9,329,185 participants.

Fig 1

PRISMA flow diagram of study inclusion.

There were 590 (61%) population-wide studies and 378 (39%) population-specific studies (Table 1). Characteristics of individual studies are reported in S1 and S2 Tables in S1 Materials. Study sampling dates ranged from September 1, 2019 to December 31, 2020.

Table 1

Summary characteristics of included articles.

Characteristic	Studies n (%)
Geographic scope
National	116 (12%)
Regional	347 (36%)
Local	277 (29%)
Sublocal	228 (24%)
Age groupsa
Children and Youth (0–17 years)	28 (3%)
Adults (18–64 years)	268 (28%)
Seniors (65+ years)	7 (0.7%)
Multiple age groups	609 (63%)
Population
Studies reporting population-wide estimates	590 (61%)
Studiesreporting population-specific estimatesb	378 (39%)
County income levelc
High income	747 (77%)
Low/middle income	221 (23%)
Sampling method
Probability sampling	209 (22%)
Non-probability sampling	759 (78%)
Antibody testsd
ELISA	242 (25%)
CLIA	409 (42%)
LFIA	137 (14%)
Other	10 (1%)
Neutralization	4 (0.4%)
Multiple types	37 (4%)
Antibody isotypes reportedd
IgG	845 (87%)
IgM	227 (24%)
IgA	47 (5%)
Risk of bias
Low	28 (3%)
Moderate	443 (46%)
High	424 (44%)
Unclear	73 (8%)

aWhen the age range for participants in a study overlapped multiple age categories by > = 30% then the study was counted as examining multiple age groups.

bStudies sampling from well-defined population sub-groups.

cClassified according to the WHO global burden of disease region groupings (high vs other—low/middle).

dStudies could have met multiple criteria so the sum of percentages may exceed 100%. Abbreviations: ELISA = enzyme-linked immunosorbent assay; CLIA = chemiluminescence immunoassay; LFIA = lateral flow immunoassay.

Seventy-four countries across all GBD regions were represented among identified serosurveys (Fig 2; S1 Fig in S1 Materials). A minority of studies were conducted in low- and middle-income countries (n = 221, 23%).

Fig 2

Map of national seroprevalence studies reporting population-wide estimates.

Countries with national-level seroprevalence studies reporting population-wide estimates are coloured on the map, based on the seroprevalence reported in the most recent such study in each country. Countries with no such national serosurveys but with “other serosurveys” are coloured in grey; this includes local and regional studies, as well as studies in specific populations. Map data reprinted from Natural Earth under a CC BY license, with permission from Natural Earth, original copyright 2009.

Many studies were at moderate (n = 443, 46%) or high risk of bias (n = 424, 44%), owing primarily to the absence of statistical correction either for population demographics or test sensitivity and specificity, using non-probability sampling methods, and using non-representative sample frames (Fig 3, S3 Table in S1 Materials).

Fig 3

Study risk of bias summary.

Item 1: Was the sample frame appropriate to address the target population? Item 2: Were study participants recruited in an appropriate way? Item 3: Was the sample size adequate? Item 4: Were the study subjects and setting described in detail? Item 5: Was data analysis conducted with sufficient coverage of the identified sample? Item 6: Were valid methods used for the identification of the condition? Item 7: Was the condition measured in a standard, reliable way for all participants? Item 8: Was there appropriate statistical analysis? Item 9: Was the response rate adequate, and if not, was the low response rate managed appropriately? Item 10: Overall risk of bias.

Correction of estimates for test sensitivity and specificity

In order to improve comparability between data and correct for misclassification error, we corrected seroprevalence values for imperfect sensitivity and specificity. To do so, we sourced additional evaluation data as described in the methods. Overall, there were 795 studies (82%) for which test sensitivity and specificity values were reported or located (S5 Table in S1 Materials). Authors reported sensitivity and specificity data in 229 studies, with reported sensitivity values ranging from 35–100% and specificity between 87–100%.

Independent evaluation data from the FindDx initiative were available for 359 studies (37%), manufacturer evaluations were available for 182 studies (19%), and published pooled sensitivity and specificity results for ELISAs, LFIAs, and CLIAs, based on the test type known to have been used, and using the definitions for these test types provided by Bastos et al. [25], were available for 101 studies (10%). Between FindDx, manufacturer evaluations, and published pooled results, test sensitivity ranged from 9–100% and specificity from 0–100%.

Estimates from 587 studies (61%) were corrected for imperfect sensitivity and specificity. We corrected seroprevalence estimates from 290 studies (30%), while author-corrected estimates were used in 297 (31%) studies as uncorrected estimates were not available for our analysis. The median absolute difference between corrected and uncorrected seroprevalence estimates was 1.1% (IQR 0.6–2.3%).

Of the 381 studies for which estimates were not corrected, data were insufficient to inform the correction analysis in 118 studies (12%). Corrected seroprevalence estimates could not be determined for 261 studies (27%), most of which were population-specific studies using small sample sizes and low test sensitivity and specificity. In these studies, the model used to correct for test sensitivity and specificity often failed to converge to a reasonable adjusted prevalence value.

Population-wide seroprevalence estimates

In studies reporting population-wide seroprevalence estimates, median corrected seroprevalence was 4.5% (IQR 2.4–8.4%, Table 2). These studies included household and community samples (n = 125), residual sera (n = 248), and blood donors (n = 54), with median corrected seroprevalence of 6.0% (IQR 2.8–15.1%), 4.0% (IQR 2.4–6.8%), and 4.7% (IQR 1.4–6.8%), respectively (Table 3).

Table 2

Summary of seroprevalence data from studies reporting population-wide estimates by global burden of disease region, geographic scope, and risk of bias.

Characteristic	No. studies	No. countries	Median sample size (IQR)	Median uncorrected seroprevalence (IQR)	No. studies with correctable data	Median corrected seroprevalence (IQR)	Risk of bias
Population-wide studies	590	57	987 (786–2639)	4.6% (2.2–8.5%)	427	4.5% (2.4–8.4%)	L: 4%, M: 62%, H: 27%, U: 6%
GBD region
Central Europe, Eastern Europe, and Central Asia	14	6	2681 (992–3037)	7.8% (2.3–20.5%)	9	12.2% (4.5–25.4%)	L: 7%, M: 43%, H: 43%, U: 7%
High-income	453	28	985 (786–1709)	4.4% (2.2–7.2%)	339	4.1% (2.4–6.9%)	L: 3%, M: 65%, H: 27%, U: 5%
Latin America and Caribbean	57	10	900 (832–1968)	6.8% (2.6–19.5%)	37	10.6% (3.0–46.5%)	L: 5%, M: 70%, H: 18%, U: 7%
North Africa and Middle East	5	4	1212 (600–3530)	12.9% (0.8–19.3%)	4	8.2% (0.1–17.7%)	L: 20%, M: 40%, H: 20%, U: 20%
South Asia	35	2	3000 (502–15625)	17.6% (8.8–26.8%)	25	17.1% (8.7–25.0%)	L: 17%, M: 43%, H: 20%, U: 20%
Southeast Asia, East Asia, and Oceania	20	2	2192 (434–18024)	1.0% (0.4–2.9%)	8	0.6% (0.3–1.4%)	L: 0%, M: 35%, H: 60%, U: 5%
Sub-Saharan Africa	6	5	528 (214–2282)	14.6% (8.0–24.0%)	5	19.5% (9.0–26.0%)	L: 17%, M: 33%, H: 50%, U: 0%
Scope
National	83	32	4297 (1200–24926)	4.5% (1.9–6.1%)	51	3.5% (1.2–6.0%)	L: 10%, M: 64%, H: 19%, U: 7%
Regional	312	19	980 (802–1106)	4.3% (2.3–7.6%)	276	4.5% (2.5–7.6%)	L: 4%, M: 73%, H: 21%, U: 3%
Local	167	31	1000 (752–2547)	5.5% (2.1–14.8%)	87	6.7% (2.6–21.9%)	L: 4%, M: 46%, H: 38%, U: 13%
Sub-local	28	14	500 (357–928)	7.2% (1.4–15.1%)	13	8.7% (0.6–15.1%)	L: 0%, M: 36%, H: 61%, U: 4%
Risk of bias
Low	25	13	4151 (2203–9922)	8.2% (2.9–13.6%)	20	10.3% (3.3–18.9%)	..
Moderate	367	42	985 (900–1545)	4.7% (2.6–7.9%)	307	4.5% (2.5–7.9%)	..
High	161	30	731 (313–2415)	3.9% (1.2–9.4%)	93	3.9% (0.9–8.2%)	..
Unclear	37	15	1709 (774–8006)	3.3% (1.5–11.0%)	7	11.7% (4.8–24.6%)	..

Abbreviations: No. = number; IQR = interquartile range; L = low; M = moderate; H = high; U = unclear; GBD = global burden of disease region.

Table 3

Summary of seroprevalence data by study sampling frame.

Population	No. of studies	Median sample size (IQR)	Median uncorrected seroprevalence (IQR)	No. of studies with correctable data	Median corrected seroprevalence (IQR)	Risk of Bias
Population-wide studies	590	987 (786–2639)	4.6% (2.2–8.5%)	427	4.5% (2.4–8.4%)	L: 4%, M: 62%, H: 27%, U: 6%
Residual sera	289	980 (804–1043)	4.1% (2.2–7.1%)	248	4.0 (2.4–6.8)	L: 0%, M: 72%, H: 28%, U: 0%
Household and community samples	228	1530 (615–4889)	5.7% (2.4–12.0%)	125	6.0 (2.8–15.1)	L: 10%, M: 49%, H: 26%, U: 14%
Blood donors	73	1110 (881–7389)	4.0% (1.8–10.3%)	54	4.7 (1.4–11.1)	L: 1%, M: 66%, H: 29%, U: 4%
Population-specific studies	378	634 (200–1694)	5.3% (1.7–14.0%)	160	3.6% (0.9–12.3%)	L: 1%, M: 20%, H: 70%, U: 10%
Health care workers and caregivers	191	801 (242–2420)	5.0% (1.7–12.0%)	66	3.6 (0.8–11.0)	L: 1%, M: 23%, H: 68%, U: 9%
Patients seeking care for non-COVID-19 reasons	46	229 (94–560)	3.6% (1.5–9.2%)	24	2.7 (1.1–7.4)	L: 0%, M: 7%, H: 83%, U: 11%
Multiple populations	41	1159 (276–4656)	5.5% (1.5–14.8%)	23	3.2 (0.3–11.3)	L: 2%, M: 17%, H: 71%, U: 10%
Essential non-healthcare workers	27	405 (239–992)	4.3% (2.2–14.8%)	11	7.5 (2.4–29.9)	L: 0%, M: 15%, H: 78%, U: 7%
Contacts of COVID patients	18	178 (71–302)	17.7% (1.3–35.2%)	11	31.5 (2.7–49.5)	L: 0%, M: 33%, H: 61%, U: 6%
Pregnant or parturient women	17	433 (169–1000)	5.8% (2.1–8.3%)	8	3.7 (1.7–5.8)	L: 0%, M: 24%, H: 76%, U: 0%
Non-essential workers and unemployed persons	13	2500 (1007–2715)	2.6% (1.0–20.0%)	8	1.5 (0.8–7.7)	L: 0%, M: 38%, H: 54%, U: 8%
Assisted living and long-term care facilities	9	291 (150–371)	23.6% (17.3–39.0%)	2	59.2 (39.7–78.8)	L: 0%, M: 0%, H: 78%, U: 22%
Persons who are incarcerated	4	1034 (664–1213)	50.3% (29.3–72.2%)	0	-	L: 0%, M: 0%, H: 0%, U: 100%
Family of essential workers	3	849 (484–920)	7.7% (5.4–15.6%)	0	-	L: 0%, M: 33%, H: 67%, U: 0%
Students and day-cares	2	900 (845–954)	7.0% (5.5–8.4%)	2	4.6 (4.3–4.9)	L: 0%, M: 50%, H: 50%, U: 0%
Persons experiencing homelessness	2	474 (301–646)	28.4% (16.5–40.2%)	1	2.8 (2.8–2.8)	L: 0%, M: 0%, H: 100%, U: 0%
Persons living in slums	2	2131 (1096–3166)	45.0% (40.5–49.6%)	2	41.7 (40.0–43.4)	L: 50%, M: 0%, H: 50%, U: 0%
Tissue donor	1	235 (235–235)	0.9% (0.9–0.9%)	0	-	L: 0%, M: 0%, H: 100%, U: 0%
Perinatal	1	1206 (1206–1206)	1.4% (1.4–1.4%)	1	0.6 (0.6–0.6)	L: 0%, M: 0%, H: 100%, U: 0%
Hospital visitors	1	1188 (1188–1188)	2.7% (2.7–2.7%)	1	1.5 (1.5–1.5)	L: 0%, M: 100%, H: 0%, U: 0%

Abbreviations: No. = number; IQR = interquartile range; L = low; M = moderate; H = high; U = unclear; GBD = global burden of disease region.

Among high-income countries, the median corrected seroprevalence in studies reporting population-wide estimates 4.1% (IQR 2.4–6.9%). In the low- and middle-income GBD regions, median corrected seroprevalence ranged from 0.6% (IQR 0.3–1.4%) in Southeast Asia, East Asia, and Oceania to 19.5% (IQR 9.0–26.0%) in South Asia (Table 2).

Population-specific seroprevalence estimates

The median corrected seroprevalence in studies reporting population-specific seroprevalence estimates was 3.6%, (IQR 0.9–12.3%, Table 4) however, there was wide variation (0.6–59%) between different populations (Table 3). Notably, the median corrected seroprevalence was 3.6% (IQR 0.8–11.0%, n = 66 studies) in healthcare workers and caregivers and 2.7% (IQR 1.1–7.4%, n = 24 studies) in specific patient groups (e.g., cancer patients). Essential non-healthcare workers (e.g., first responders) had a median seroprevalence of 7.5% (IQR 2.4–29.9%, n = 11 studies, Table 3). Higher seroprevalence estimates were reported in studies of contacts of COVID-19 patients (median 31.5%, IQR 2.7–39.9%, n = 11 studies), persons living in slums (median 41.7%, IQR 40.0–43.4%, n = 2 studies), and persons in assisted living and long-term care facilities (median 59.2%, IQR 39.7–78.8%, n = 2 studies).

Table 4

Summary of seroprevalence data from studies reporting population-specific estimates by global burden of disease region, geographic scope, and risk of bias.

Characteristic	No. studies	No. countries	Median sample size (IQR)	Median uncorrected seroprevalence (IQR)	No. studies with correctable data	Median corrected seroprevalence (IQR)	Risk of bias
Population-specific studies	378	53	634 (200–1694)	5.3% (1.7–14.0%)	160	3.6% (0.9–12.3%)	L: 1%, M: 20%, H: 70%, U: 10%
GBD region
Central Europe, Eastern Europe, and Central Asia	12	7	512 (354–1611)	2.8% (1.2–10.7%)	5	10.6% (8.8–14.4%)	L: 0%, M: 33%, H: 42%, U: 25%
High-income	294	24	611 (188–1662)	5.1% (1.8–12.1%)	125	3.2% (0.9–10.0%)	L: 0%, M: 19%, H: 71%, U: 9%
Latin America and Caribbean	12	6	378 (275–1820)	9.8% (5.7–13.7%)	7	10.7% (4.6–16.5%)	L: 8%, M: 25%, H: 58%, U: 8%
North Africa and Middle East	16	7	434 (223–2991)	16.8% (3.8–38.7%)	7	29.4% (20.0–45.8%)	L: 0%, M: 25%, H: 75%, U: 0%
South Asia	14	2	1006 (671–1537)	16.6% (11.3–30.7%)	2	28.1% (19.6–36.6%)	L: 7%, M: 29%, H: 50%, U: 14%
Southeast Asia, East Asia, and Oceania	26	3	1024 (346–4418)	1.9% (0.3–5.3%)	13	0.3% (0.2–3.5%)	L: 0%, M: 19%, H: 69%, U: 12%
Sub-Saharan Africa	4	4	452 (320–614)	20.2% (12.8–24.3%)	1	11.3% (11.3–11.3%)	L: 0%, M: 0%, H: 100%, U: 0%
Scope
National	33	24	1150 (525–4234)	3.8% (1.7–11.6%)	19	4.5% (0.5–12.1%)	L: 0%, M: 24%, H: 55%, U: 21%
Regional	35	14	1671 (320–4814)	3.1% (1.5–13.5%)	15	3.7% (1.9–19.4%)	L: 3%, M: 37%, H: 57%, U: 3%
Local	110	28	681 (206–1654)	5.1% (1.9–14.4%)	49	3.0% (0.8–11.5%)	L: 2%, M: 20%, H: 71%, U: 7%
Sub-local	200	33	376 (174–1156)	6.0% (1.9–14.0%)	77	4.0% (0.9–12.0%)	L: 0%, M: 16%, H: 74%, U: 10%
Risk of bias
Low	3	3	4202 (2770–16497)	29.1% (16.6–41.6%)	3	45.1% (24.7–56.6%)	..
Moderate	76	27	1808 (922–4127)	5.1% (2.3–11.3%)	34	3.4% (1.4–8.6%)	..
High	263	42	320 (152–1002)	5.4% (1.7–15.1%)	113	3.4% (0.8–13.4%)	..
Unclear	36	16	1098 (354–2880)	3.8% (0.9–10.0%)	10	4.6% (2.7–7.4%)	..

Abbreviations: No. = number; IQR = interquartile range; L = low; M = moderate; H = high; U = unclear; GBD = global burden of disease region.

Seroprevalence by population sub-groups (meta-analysis)

Within studies, seroprevalence was significantly lower for seniors 65+ compared to adults 18–64 (prevalence ratio [PR]: 0.79 [95% CI: 0.69–0.90]). Seroprevalence was significantly higher for Black persons, Asian persons, Indigenous persons, and other groups compared to Caucasian persons (PRs from 1.89–5.74), and in health care workers with close contact with COVID-19 patients compared to those with no close contact (PR 2.10 [1.28–3.44]). Seroprevalence differences approached significance for individuals in the community with close contact with COVID-19 patients (PR 1.85 [0.99–3.44]) and for health care workers compared to members of the community (PR 1.45 [0.99–2.14]). There were no differences in infection risk based on sex and gender. Full results are reported in Table 5, and results for uncorrected prevalence estimates are reported in S4 Table in S1 Materials.

Table 5

Differences in seroprevalence estimates by demographic characteristics within studies.

Factor	Reference Group	Comparison Group	Number of Studies	Risk Ratio (95% CI)a	Heterogeneity (I2)
Age	Adults (18–64)	Youth (0–17)	82	0.92 (0.81–1.04)	90.7%
	Adults (18–64)	Seniors (65+)	127	0.79 (0.69–0.90)	93.9%
Sex/Gender	Female	Male	129	1.03 (0.98–1.08)	79.1%
Race	Caucasian	Black	19	3.37 (2.64–4.29)	85.7%
	Caucasian	Asian	17	2.47 (1.96–3.11)	88.9%
	Caucasian	Indigenous	8	5.74 (1.01–32.6)	75.3%
	Caucasian	Multiple/other	18	1.89 (1.60–2.24)	64.0%
Close contact with COVID-19 patients	Individuals with no close contact	Individuals with close contact	35	1.85 (0.99–3.44)	97.4%
	Health care workers with no close contact	Health care workers with close contact	44	2.10 (1.28–3.44)	89.4%
Health care worker status	Non-health care workers and caregivers	Health care workers and caregivers	19	1.45 (0.99–2.14)	98.3%

aUsing corrected seroprevalence estimates. Abbreviations: CI = confidence interval.

Seroprevalence by study and geographic factors (meta-regression)

On multivariable meta-regression, studies at low risk of bias reported higher corrected seroprevalence estimates relative to studies with moderate risk of bias (prevalence ratio 1.67, 95% CI 1.22–2.27, p = 0.001), high risk of bias (1.54, 95% CI 1.11–2.13, p = 0.01), and unclear risk of bias (2.63, 95% CI 1.54–4.55, p<0.001)(S6 Table in S1 Materials). Blood donors and residual sera groups, both used as proxies for the general population, reported similar corrected seroprevalence estimates compared to household and community samples (blood donors: 0.96, 95% CI 0.76–1.22, p = 0.77; residual sera: 1.12, 95% CI 0.94–1.35).

National studies reported lower seroprevalence estimates compared to regional studies (0.61, 95% CI 0.48–0.77, p<0.001), local studies (0.47, 95% CI 0.37–0.60, p<0.001) and sublocal studies (0.52, 95% CI 0.33–0.81, p = 0.004). Finally, compared to high-income countries, higher seroprevalence estimates were reported by countries in Sub-Saharan Africa (5.01, 95% CI 2.89–8.69, p<0.001), South Asia (2.84, 95% CI 2.09–3.85, p<0.001), Central Europe, Eastern Europe, and Central Asia (2.83, 95% CI 1.75–4.55, p<0.001), and Latin America and Caribbean (2.71, 95% CI 2.07–3.54, p<0.001), while countries in Southeast Asia, East Asia, and Oceania (0.18, 95% CI 0.09–0.34) reported lower seroprevalence estimates. Visual checks confirmed that model assumptions of normality and homoscedasticity were met.

Ratio of seroprevalence to cumulative case incidence

The median ratio between corrected seroprevalence estimates from national studies and the corresponding cumulative incidence of SARS-CoV-2 infection nine days prior was 18.1 (IQR 5.9–38.7, n = 49 studies; Table 6, S2 Fig in S1 Materials), indicating a median of 18.1 serologically identified infections per 1 confirmed case globally. Stratifying by risk of bias and GBD showed variation in median ratios between seroprevalence and cumulative incidence (Table 6).

Table 6

The median ratio between corrected seroprevalence estimates from national studies and the corresponding cumulative incidence of SARS-CoV-2 infection from nine days prior.

Characteristics	Number of studies	Ratio of seroprevalence to cumulative incidence
National studies with correctable estimates and matching case data available	49	18.1 (5.9–38.7)
Risk of bias
Low	6	19.9 (11.2–111.7)
Moderate	31	12.1 (5.3–32.9)
High	10	19.4 (18.8–39.3)
Unclear	2	0.4 (0.3–0.5)
Global burden of disease regions
Central Europe, Eastern Europe, and Central Asiaa	-	-
High-income	41	15.2 (5.9–24.2)
Latin America and Caribbean	3	49.5 (46.7–75.7)
North Africa and Middle East	2	71.2 (35.7–106.7)
South Asia	2	6.7 (6.1–7.4)
Southeast Asia, East Asia, and Oceaniaa	-	-
Sub-Saharan Africa	1	602.5 (602.5–602.5)

aMatching cumulative incidence data not available for the seroprevalence study periods.

Discussion

This systematic review and meta-analysis provides an overview of global SARS-CoV-2 seroprevalence based on data from 9,329,185 participants in 968 serosurveys from 605 reports. Overall, in the first year of the COVID-19 pandemic, estimates of population-wide seroprevalence were low (median 4.5%, IQR 2.4–8.4%), however, population-specific estimates of seroprevalence varied widely from a low of 0.6% (perinatal) to a high of 59% (persons in assisted living and long-term care facilities).

Seroprevalence varied considerably between GBD regions after correcting for study characteristics and test sensitivity and specificity. Given the limited evidence for altitude or climate effects on SARS-CoV-2 transmission [36, 37] variations in seroprevalence likely reflect differences in community transmission based on behaviour, public health responses, local resources, and the built environment. Stakeholders should carefully review the infection control measures implemented in Southeast Asia, East Asia, and Oceania as they appear to have been effective at limiting SARS-CoV-2 transmission [38, 39].

Our results suggest clear population differences in SARS-CoV-2 infection, with marginalized and high-risk groups disproportionately affected. Differences in infection risk based on race might be attributed to crowding, higher-risk occupation roles (e.g., front-line service jobs) and other systemic inequities [40-43]. Some of these groups (Black, Asian, and other minority racial and ethnic groups) are also known to have higher infection fatality rates [44]. Such differences may inform policy on vaccine distribution, workforce protections, and other public health measures designed to protect marginalized persons.

Our review found that health care workers who had close contact with confirmed COVID-19 cases had a higher risk of seropositivity, consistent with previous reports [45]. Results in this study regarding contact with a COVID-19 case among non-health care workers warrant further investigation. Our meta-analysis of seroprevalence in persons with and without contact in studies reporting both subgroups found no significant difference, despite the fact that studies of persons with exposure to COVID-19 reported much higher seroprevalence estimates compared to population-wide studies (31.5% vs. 4.5%). These results align with other evidence synthesis examining persons with and without COVID-19 exposure however, they conflict with studies of high-risk exposure, including health care workers [9, 46]. It is possible that contact exposure in a clinical setting may be more narrowly defined and carefully measured, whereas definitions of exposure in non-clinical studies may be more heterogenous or prone to potential misclassification due to asymptomatic infection. Future analysis should explore the association of different definitions and measurement of contact status with seroprevalence estimates.

Few studies (23%) have been conducted in low- and middle-income countries. Results from the ongoing WHO Unity studies will help to bridge this knowledge gap and contribute to a more comprehensive understanding of the spread and impact of COVID-19 globally [15]. Use of the standardized Unity protocols will also help to increase the pool of robust, comparable seroprevalence data.

Approximately half of studies reporting population-wide SARS-CoV-2 seroprevalence estimates used blood from donors and residual sera as a proxy for the community. Our results showed that these studies report seroprevalence estimates that are similar to studies of household and community-based samples. It has previously been shown that these groups contain disproportionate numbers of people that are young, White, college graduates, employed, physically active, and never-smokers [47, 48]. However, the results of our study suggest that investigators may use these proxy sampling frames to obtain fairly representative estimates of seroprevalence if studies use large sample sizes with adequate coverage of important subgroups (e.g., age, sex, race/ethnicity) to permit standardization to population characteristics, tests with high sensitivity and specificity, and statistical corrections for imperfect sensitivity and specificity.

Our results suggest that studies at moderate, high, or unclear risk of bias may generate lower seroprevalence estimates relative to studies at low risk of bias. There are many possible explanations for this somewhat counterintuitive finding. Common reasons for unclear or elevated risk of bias were absent reporting of test information, use of tests with low sensitivity and specificity, no statistical correction for demographics or test sensitivity and specificity, use of non-probability sampling, and use of non-representative sample frames. Therefore, selection bias that favoured healthier, affluent, non-racialised groups at lower risk of infection paired with no adjustment for sample characteristics may have contributed to lower estimates of seroprevalence. It is also possible that the false negative rate was higher for studies in which authors used low sensitivity tests, particularly when authors did not statistically correct estimates for imperfect test performance or used inflated estimates of test sensitivity, as are often reported by manufactures, to conduct such corrections.

Systematic reviews of SARS-CoV-2 serological test accuracy have found that many tests have poor sensitivity and specificity [24, 25]. Of the studies included in this review, only 298 (31%) corrected for test sensitivity and specificity, and 118 (12%) failed to report identifying information on the test used altogether. Our study corrected seroprevalence estimates for test sensitivity and specificity in an additional 290 (30%) studies. The median absolute difference between corrected and uncorrected estimates was 1.1%—a substantial change, given that the median corrected seroprevalence in studies reporting population-wide estimates was 4.5%. This difference emphasizes the importance of conducting such corrections to minimize bias in serosurvey data. Furthermore, improved reporting of serological testing information in serosurveys is needed to maximize the amount of robust and comparable data for evidence synthesis.

Seroprevalence estimates were 18.1 times higher than the corresponding cumulative incidence of COVID-19 infections, with large variations between the Global Burden of Disease Regions (seroprevalence estimates ranging from 6 to 602 times higher than cumulative incidence). This level of under-ascertainment suggests that confirmed SARS-CoV-2 infections are a poor indicator of the extent of infection spread, even in high-income countries where testing has been more widely available. The broad range of ratios mirrors estimates from other published evidence on case under-ascertainment, which suggests a range of 0.56 to 717 [49, 50].

Seroprevalence to cumulative case ratios can provide a rough roadmap for public health authorities by identifying areas that may be receiving potentially insufficient levels of testing and by providing an indication of the number of undetected asymptomatic infections.

While there is interest in using these seroprevalence to cumulative case ratios in identifying inadequate testing and estimating case ascertainment, caution is required in the quantitative interpretation of these ratios. Our study found a median ratio of 18.1, which aligns with other published analysis [50]. This would imply that 2.9 billion people globally have been infected with SARS-CoV-2 rather than the 160 million reported as of May 15, 2021 [2]. This is not likely, and this estimate conflicts with the evidence that seroprevalence remains low in the general population. If applying this global ratio to countries with high cumulative incidence, such as the United States (32 million by May 15, 2021), then the total number of infections would exceed the population.

There are several possible reasons for these discrepancies. Firstly, these ratios clearly vary by geographic region and regional health policy, with higher diagnostic testing rates likely to correspond to lower seroprevalence to case ratios. Country-specific ratios, or region-specific ratios if available, should be used to inform planning wherever possible. Second, diagnostic testing-based estimates of cumulative incidence vary by assay; for example, lower RT-PCR cycle thresholds or the use of less sensitive rapid antigen tests would lead to lower estimates of cumulative cases. Finally, our analysis compares seroprevalence to cumulative case ratios at different point in time. As diagnostic testing measures expanded, these ratios may have declined over time, complicating the process of applying a single fixed ratio to a cumulative incidence number. As such, there is a need for more nuanced analysis of case under-ascertainment and caution should be exercised if utilizing them in public health planning.

This study has limitations. Firstly, some asymptomatic individuals may not seroconvert, some individuals may have been tested prior to seroconversion, and others may have antibodies that have waned by the time of blood collection, so the data in this study may underestimate the number of SARS-CoV-2 infections [51]. To ameliorate this, we prioritized estimates that tested for anti-spike IgG antibodies, which show better persistence in serum compared to non-IgG and anti-nucleocapsid IgG antibodies [20-25]. Secondly, to account for measurement error in seroprevalence estimates resulting from poorly performing tests, it was necessary to use sensitivity and specificity information from multiple sources of varying quality. While we prioritized independent evaluations, these were not available for all tests. Furthermore, lab-to-lab variation may undermine the generalizability and comparability of the test evaluation data we utilized. Going forward, investigators should conduct evaluations of their assays using a standard international reference panel, such as the panel created by the WHO [52], and report their results in international units referenced against the World Antibody Titres Standard to increase comparability of serosurvey results. Where this is not feasible, investigators should at least report the test name, manufacturer, and sensitivity and specificity values to improve data comparability [53]. Thirdly, some of the summary results may have been driven by the large volume of data from high-income countries, which primarily reported lower seroprevalence estimates. While we frequently stratified by or adjusted for GBD region, caution is required when interpreting some of the summary estimates. Fourthly, the residual heterogeneity in our meta-regression indicates that not all relevant explanatory variables have been accounted for. Many factors may contribute to the spread of infection. Even if all important factors were known, it would be difficult to account for the variation in seroprevalence due to limited availability of data with sufficient granularity and changing health policy and individual behavior.

This systematic review is the largest synthesis of SARS-CoV-2 serosurveillance data to date. Our search was rigorous and comprehensive: we included non-English articles, government reports, unpublished data, and serosurveillance reports obtained via expert recommendations and the SeroTracker website. This comprehensive search is important because many serosurveys—especially in LMICs—have not been published or released as preprints. A strength of this review was the use of corrected prevalence estimates for analysis, revealing that imperfect sensitivity and specificity have major effects on seroprevalence findings. To our knowledge, this is the largest systematic comparison of seroprevalence estimates from blood donors, residual sera, and household and community-based general population samples. Finally, this study is part of a regularly-updated systematic review, and summary results will continue to be disseminated throughout the pandemic on a publicly available website (SeroTracker.com) [12].

Serosurveillance efforts so far have mostly taken the form of formal studies led by academic institutions. This approach makes sense when serosurveys are used as a tool to periodically monitor the spread of infection and identify high-risk groups. However, given the rise of more infectious SARS-CoV-2 variants, continued uncertainty about the global prevalence of infection, and variably quality of serosurvey design and reporting, more coordinated, standardized, and routine serosurveillance may be needed. Furthermore, as vaccines are deployed, there may be additional value derived from serosurveys, specifically in evaluating vaccine effectiveness in the real world, monitoring aggregate immunity arising from infection and vaccination, and measuring population antibody titres as a correlate of protection and as an indicator for vaccine boosters. Therefore, going forward, serosurveillance efforts may better serve end-users if they take the form of real-time monitoring programs housed in public health units, using standardized serosurvey protocols and reporting. Leaders who can compare studies in their regions over time and pair vaccine distribution data with live serosurveys will be well-equipped to track the pandemic, understand the impact of variants, and monitor outcomes of vaccination efforts in their communities in real time.

Conclusion

Our review shows that SARS-CoV-2 seroprevalence remains low in the general population, indicating the importance of remaining vigilant until vaccine-derived herd immunity is achieved. There are clear geographic and population differences in SARS-CoV-2 infection prevalence, with certain groups disproportionately affected. Policy and decision makers need to better protect these groups to reduce inequity in the impact of COVID-19.

As the COVID-19 pandemic progresses and serology data accumulate, ongoing evidence synthesis is needed to inform public health policy. We will continue to update our systematic review and seroprevalence dashboard to help address this need.

(DOCX)

Click here for additional data file.

Transfer Alert

This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present.

1 Feb 2021

PONE-D-20-40466

Global seroprevalence of SARS-CoV-2 antibodies: a systematic review and meta-analysis

PLOS ONE

Dear Dr. Bobrovitz,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Mar 15 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Yury E Khudyakov, PhD

Academic Editor

PLOS ONE

Additional Editor Comments:

Your manuscript was reviewed by 2 experts in the field. Both reviewers provided suggestions to improve your paper. Please review the attached comments and provide your responses.

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1.) Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2.) While we note that methods and results have been described in the Supplementary file, we would suggest that at least some of this information (for example, the search strategy, the list of studies included and the results of the risk of bias assessment) is reported in the main text.

3.) Please amend your list of authors on the manuscript to ensure that each author is linked to an affiliation. Authors’ affiliations should reflect the institution where the work was done (if authors moved subsequently, you can also list the new affiliation stating “current affiliation:….” as necessary).

4.) We note that Figure 2 and Appendix figure 1 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright.

We require you to either (a) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (b) remove the figures from your submission:

a.) You may seek permission from the original copyright holder of Figure 2 and Appendix figure 1 to publish the content specifically under the CC BY 4.0 license.

We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text:

“I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.”

Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.

In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].”

b.) If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only.

The following resources for replacing copyrighted map figures may be helpful:

USGS National Map Viewer (public domain): http://viewer.nationalmap.gov/viewer/

The Gateway to Astronaut Photography of Earth (public domain): http://eol.jsc.nasa.gov/sseop/clickmap/

Maps at the CIA (public domain): https://www.cia.gov/library/publications/the-world-factbook/index.html and https://www.cia.gov/library/publications/cia-maps-publications/index.html

NASA Earth Observatory (public domain): http://earthobservatory.nasa.gov/

Landsat: http://landsat.visibleearth.nasa.gov/

USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): http://eros.usgs.gov/#

Natural Earth (public domain): http://www.naturalearthdata.com/

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The manuscript summarises the serology studies available for SARS-CoV-2 through a systematic review and meta-analysis of the resulting data. In addition to describing the overall global seroprevalence, they characterise seroprevalence in various demographic groups e.g. age, gender, country, ethnicity, type of patient/sample etc. Since the approach taken corrects for bias such as lack of sensitivity/sensitivity or demographic adjustments in the original publications the results here should provide an increased accuracy in the estimates.

The manuscript is interesting and from what I understood from the methods, technically sound. However, it would benefit from some clarifications that I detail in the attached document.

Reviewer #2: This is a clear and well written report of a systematic review/meta-analysis of the literature on sera-prevalence of SARS-CoV-2 antibodies published worldwide. The authors have a clear understanding of the pitfalls associated with both study design and laboratory evaluation of population based seroprevalence and have brought together the world literature up to August 28th 2020 in an accessible way with appropriate corrections.

As this is such a rapidly evolving field, the only concern is whether this data adequately reflects the current situation. With a cut off date for analysis of late August 2020, most of the completed studies will represent seroprevalence estimates relatively early in the pandemic. If an updated analysis to the end of December 2020 could be incorporated into this manuscript that would be ideal and would add value as the authors could estimate seroprevalence in relation to time when the relevant population was sampled and that in turn could be evaluated in the context of time since the onset of the pandemic.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: David Goldblatt

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: Review SARS2_SeroReview_PlosOne.pdf

Click here for additional data file.

16 May 2021

1. Responses to editor comments

1.1) Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Thank you very much for reviewing our manuscript. We have responded to the editorial and reviewer comments below and resubmitted a revised version of the manuscript for your consideration.

1.2) Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at XX.

Thank you. We have revised the style of the manuscript to meet PLOS ONE’s style requirements.

1.3) While we note that methods and results have been described in the Supplementary file, we would suggest that at least some of this information (for example, the search strategy, the list of studies included and the results of the risk of bias assessment) is reported in the main text.

Thank you for this suggestion. We have moved the following information into the main text: search strategy, the summary risk of bias assessment figure, and Table 4 Summary of seroprevalence data for general and special population sub-groups (formerly appendix Table 4).

1.4) Please amend your list of authors on the manuscript to ensure that each author is linked to an affiliation. Authors’ affiliations should reflect the institution where the work was done (if authors moved subsequently, you can also list the new affiliation stating “current affiliation:….” as necessary).

We have amended the list of authors to ensure that each is linked to an affiliation.

1.5) We note that Figure 2 and Appendix figure 1 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright.

We require you to either (a) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (b) remove the figures from your submission:

a.) You may seek permission from the original copyright holder of Figure 2 and Appendix figure 1 to publish the content specifically under the CC BY 4.0 license.

We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text:

“I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.”

Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.

In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].”

The revised version of this manuscript contains map figures that were generated using map data from Natural Earth (specific dataset at this link). Natural Earth provides map data that is in the public domain.

From their About Page: “All versions of Natural Earth raster + vector map data found on this website are in the public domain. You may use the maps in any manner, including modifying the content and design, electronic dissemination, and offset printing. The primary authors, Tom Patterson and Nathaniel Vaughn Kelso, and all other contributors renounce all financial claim to the maps and invites you to use them for personal, educational, and commercial purposes. No permission is needed to use Natural Earth. Crediting the authors is unnecessary.”

In the caption of all map figures, we have added the following phrase: “Map data reprinted from Natural Earth under a CC BY license, with permission from Natural Earth, original copyright 2009.” If this phrase is unnecessary for public domain maps, we would also be glad for this to be removed.

b.) If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only.

The following resources for replacing copyrighted map figures may be helpful:

USGS National Map Viewer (public domain): http://viewer.nationalmap.gov/viewer/

The Gateway to Astronaut Photography of Earth (public domain): http://eol.jsc.nasa.gov/sseop/clickmap/

Maps at the CIA (public domain): https://www.cia.gov/library/publications/the-world-factbook/index.html and https://www.cia.gov/library/publications/cia-maps-publications/index.html

NASA Earth Observatory (public domain): http://earthobservatory.nasa.gov/

Landsat: http://landsat.visibleearth.nasa.gov/

USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): http://eros.usgs.gov/

Natural Earth (public domain): http://www.naturalearthdata.com/

1.6) Additional comments to the editor

Note for the editor regarding Figure 1 and number of screened full texts: we re-classified our definition of a full text screen for Google News articles such that fewer articles were counted as full text in this updated review. While the overall number of full text articles screened has increased, a number of previous full text exclusions have been re-classified as abstract exclusions. The updated data are provided in the revised Figure 1.

2) Reviewer 1 major comments:

2.1) The manuscript summarises the serology studies available for SARS-CoV-2 through a systematic review and meta-analysis of the resulting data. In addition to describing the overall global seroprevalence, they characterise seroprevalence in various demographic groups e.g. age, gender, country, ethnicity, type of patient/sample etc. Since the approach taken corrects for bias such as lack of sensitivity/sensitivity or demographic adjustments in the original publications the results here should provide an increased accuracy in the estimate. The manuscript is interesting and from what I understood from the methods, technically sound. However, it would benefit from some clarifications that I detail below.

Thank you for reviewing our article. We have addressed your requests for clarification below.

2.2) The rationale for needing sero-surveys and a review of sero-surveys is not clear. For example, but not exclusively, in the introduction (3rd paragraph), the authors mention that is increasingly important to measure baseline prevalence of antibodies in the vaccine era. While I agree with importance of understanding serological patterns, sero tests are not being done prior to vaccination nor is being taken into account for the number of doses distributed since there is limited supply, and is also not being used for prioritizing groups. So why is it important that we know this?

We have clarified the rationale for study as follows:

Page 5, Line 105: Serological assays identify SARS-CoV-2 antibodies, indicating previous infection in unvaccinated persons.7 Population-based serological testing provides better estimates of the cumulative incidence of infection by complementing diagnostic testing of acute infection and helping to inform the public health response to COVID-19. Furthermore, as the world moves through the vaccine and variant era, synthesizing seroepidemiology findings is increasingly important to track the spread of infection, identify disproportionately affected groups, and measure progress towards herd immunity.

SARS-CoV-2 seroprevalence estimates are reported not only in published articles and preprints, but also in government and health institute reports, and media.8 Consequently, few studies have comprehensively synthesized seroprevalence findings that include all of these sources.9,10 Describing and evaluating the characteristics of seroprevalence studies conducted over the first year of the pandemic may provide valuable guidance for serosurvey investigators moving forward.

2.3) Correction of seroprevalence estimates: This needs a bit more explanation and clarification throughout text and tables. For corrected seroprevalences, did you correct all studies or do you use a mixture of published corrections and corrections made in this study? Either way how do you ensure unbiased/equivalent corrections?

We have provided more details about the correction of seroprevalence estimates in the methods section. This text clarifies that for corrected seroprevalences, we corrected all those studies where we had sufficient information to do so, even where published corrections were available; where we did not have sufficient informations, we used author-corrected seroprevalence estimates.

Page 9-10, Line 200-208: To account for imperfect test sensitivity and specificity, seroprevalence estimates were corrected using Bayesian measurement error models, with binomial sensitivity and specificity distributions.26 The sensitivity and specificity values for correction were derived, in order of preference, from: (i) the FindDx -McGill database of independent evaluations of serological tests27; (ii) independent test evaluations conducted by serosurvey investigators and reported alongside serosurvey findings; (iii) manufacturer-reported sensitivity and specificity (including author evaluated in-house assays); (iv) published pooled sensitivity and specificity by immunoassay type.25 If uncorrected estimates were not available, we used author-reported corrected seroprevalence estimates. Details of these evaluations are located in S4 File.

For the studies included in this meta-analysis, there was no comprehensive source of perfectly equivalent correction data available, where tests were evaluated against the same panel of samples and with the same antibody titre standard. These efforts are still underway by the WHO and other groups. For this reason, we have used data from several sources which use similar methods to evaluate these tests. While this method cannot completely remove bias related to test performance, it minimizes bias as compared to not correcting for test sensitivity and specificity altogether.

We have provided more details in the results section about the sources of correction that were available for the final data set.

Page 14-15, Line 277-300: In order to improve comparability between data and correct for misclassification error, we corrected seroprevalence values for imperfect sensitivity and specificity. To do so, we sourced additional evaluation data as described in the methods. Overall, there were 795 studies (82%) for which test sensitivity and specificity values were reported or located (S5 Table). Authors reported sensitivity and specificity data in 229 studies, with reported sensitivity values ranging from 35-100% and specificity between 87-100%.

Independent evaluation data from the FindDx initiative were available for 359 studies (37%), manufacturer evaluations were available for 182 studies (19%), and published pooled sensitivity and specificity results for ELISAs, LFIAs, and CLIAs, based on the test type known to have been used, and using the definitions for these test types provided by Bastos et al.25, were available for 101 studies (10%). Between FindDx, manufacturer evaluations, and published pooled results, test sensitivity ranged from 9-100% and specificity from 0-100%.

Estimates from 587 studies (61%) were corrected for imperfect sensitivity and specificity. We corrected seroprevalence estimates from 290 studies (30%), while author-corrected estimates were used in 297 (31%) studies as uncorrected estimates were not available for our analysis. The median absolute difference between corrected and uncorrected seroprevalence estimates was 1.1% (IQR 0.6-2.3%).

Of the 381 studies for which estimates were not corrected, data were insufficient to inform the correction analysis in 118 studies (12%). Corrected seroprevalence estimates could not be determined for 261 studies (27%), most of which were population-specific studies using small sample sizes and low test sensitivity and specificity. In these studies, the model used to correct for test sensitivity and specificity often failed to converge to a reasonable adjusted prevalence value.

2.4) What sort of independent evaluations did you base your corrections, are these for sensitivity/specificity values provided by commercial kits?

We have provided more details of the source of independent evaluations in the main text and highlighted the additional information on these sources in the appendix. Most of these independent evaluations were for sensitivity and specificity in commercial kits.

Page 9-10, Line 200-208: To account for imperfect test sensitivity and specificity, seroprevalence estimates were corrected using Bayesian measurement error models, with binomial sensitivity and specificity distributions.26 The sensitivity and specificity values for correction were derived, in order of preference, from: (i) the FindDx -McGill database of independent evaluations of serological tests27; (ii) independent test evaluations conducted by serosurvey investigators and reported alongside serosurvey findings; (iii) manufacturer-reported sensitivity and specificity (including author evaluated in-house assays); (iv) published pooled sensitivity and specificity by immunoassay type.25

S4 file, page 11, line 171-206: The sensitivity and specificity values for correction were derived, in order of preference, from: (i) the FINDDx-McGill database of independent evaluations of serological tests9; (ii) independent test evaluations conducted by serosurvey investigators and reported alongside serosurvey findings; (iii) manufacturer-reported sensitivity and specificity; (iv) published pooled sensitivity and specificity by immunoassay type.10 If uncorrected estimates were unavailable, we used author-reported corrected seroprevalence estimates in lieu of performing our own correction. When none of the above corrections were possible, we excluded estimates from further analysis. Details of this order of priority follow:

1. The FINDDx-McGill database of independent evaluations of serological tests.9 We only considered evaluations reporting both sensitivity and specificity for test performance across all sickness days (as opposed to day 1, day 5, etc). Where multiple evaluations were available, we prioritized in the following order:

a. The evaluation needed to match the test name, manufacturer, and target isotype used in the study

b. The evaluation needed to match the sample specimen type used in the study

i. For sample types that were not reported in either the test evaluation or the serosurvey study we assumed whole blood was used for LFIA tests and serum/plasma was used for non-LFIA tests

ii. Plasma/Serum were used interchangeable when no direct match for index sample type was available

c. Prioritized reference specimen type which yield the most virus according to a systematic review and meta-analysis published in July 2020 that compared RT-PCR positivity of different specimens11

i. It was assumed that “respiratory specimen” was referring to upper respiratory specimen as a conservative assumption as these viral loads are lower than bronchoalveolar lavage or sputum. We ranked it along side throat swab.

ii. “Lower respiratory specimen” was ranked with with bronchoalveolar lavage fluid

iii. “Upper respiratory specimen” was ranked with throat swab

iv. If a mixed reference sample was used in the independent evaluation then an even distribution of sample types was assumed; the average % yield of the viral load was calculated and the sample type was ranked accordingly

d. Largest sample size

2. Serological test evaluations conducted by study authors, where those authors were at arms-length from the design of the study in questions

3. Manufacturer-reported sensitivity and specificity, which includes evaluations of in-house serological tests published by the research group that developed the same test

4. Published pooled sensitivity and specificity results for ELISAs, LFIAs, and CLIAs, based on the test type known to have been used, and using the definitions for these test types provided in the cited article.10

2.5) What about lab-to-lab variations and in-house assays?

Excellent question. Lab to lab variation is a major barrier to comparisons of assay evaluations and serological study findings. We have added a statement about this to the discussion section and highlight the need for use of international standards such as the WHO standard reference panel and international antibody titre standard described in January 2021. Data using this standard will begin to become more broadly available later in 2021.

Page 27-28, line 484-494: Secondly, to account for measurement error in seroprevalence estimates resulting from poorly performing tests, it was necessary to use sensitivity and specificity information from multiple sources of varying quality. While we prioritized independent evaluations, these were not available for all tests. Furthermore, lab-to-lab variation may undermine the generalizability and comparability of the test evaluation data we utilized. Going forward, investigators should conduct evaluations of their assays using a standard international reference panel, such as the panel created by the WHO52, and report their results in international units referenced against the World Antibody Titres Standard to increase comparability of serosurvey results. Where this is not feasible, investigators should at least report the test name, manufacturer, and sensitivity and specificity values to improve data comparability.53

If in-house assays designed by the authors of the seroprevalence study were used, and the authors reported sensitivity and specificity values from in-house validation, these test evaluations were considered to be manufacturer values (as they were done by the same group developing the assay) and used to independently correct seroprevalence estimates. We have added a statement to the methods and appendix to clarify this.

Page 9-10, Line 200-208: To account for imperfect test sensitivity and specificity, seroprevalence estimates were corrected using Bayesian measurement error models, with binomial sensitivity and specificity distributions.26 The sensitivity and specificity values for correction were derived, in order of preference, from: (i) the FindDx -McGill database of independent evaluations of serological tests27; (ii) independent test evaluations conducted by serosurvey investigators and reported alongside serosurvey findings; (iii) manufacturer-reported sensitivity and specificity (including author evaluated in-house assays); (iv) published pooled sensitivity and specificity by immunoassay type.25

Page 12, line 203-204: Manufacturer-reported sensitivity and specificity, which includes evaluations of in-house serological tests published by the research group that developed the same test

2.6) What type of sensitivity analysis was conducted on uncorrected data and for what?

We reported the uncorrected data for all major analyses in the main text and/or the appendix so that readers can directly compare the results. The following data displays include uncorrected data:

Table 2. Summary of seroprevalence data from studies reporting population-wide estimates by global burden of disease region, geographic scope, and risk of bias

Table 3. Summary of seroprevalence data by study sampling frame

Table 4. Summary of seroprevalence data from studies reporting population-specific estimates by global burden of disease region, geographic scope, and risk of bias

S4 Table. Summary of unadjusted meta-analysis results

S6 Table. Summary of meta-regression results

In the results section, we also present the median absolute difference between corrected and uncorrected seroprevalence estimates for population-wide studies.

Results, page X, Line X-X: The median absolute difference between corrected and uncorrected seroprevalence estimates was 1.1% (IQR 0.6-2.3%).

2.7) How do you correct for power of the studies? The sample size of studies would have been planned taking into account the population size and demographic of the region, hence providing a powered measure of seroprevalence, but many not.

Our Risk of Bias assessments contain an item which accounts for study power. Item 3 in the Risk of Bias tool evaluates study sample size. Our scoping work on this topic revealed that very few studies (< 10%) reported sample size calculations. For this reason, we carried out our own calculations to determine a threshold sample size: n = 599, which is sufficient to have 80% power in detecting a 2.5% seroprevalence to a precision of 1.5%. The full rationale for this and details have been added to the S3 file.

S3 file, page 8, line 99: To calculate the required sample size we used an assumed prevalence of 2.5%, which was the global average estimated by the WHO in April, 2020.3 Based on guidance by the Joanna Briggs Institute and published medical statistical recommendations we selected a precision value that was half the assumed prevalence (1.25%)4,5 We calculated a minimum sample size of 599 using these inputs:

Sample size calculation:

Where n = sample size;

Z = Z statistic for level of confidence (95%);

P = expected prevalence (2.5% WHO global estimate);

d = precision (1.25%)

In cases where the sample size calculation was provided, this item was marked as yes — even if the required sample for 80% power was below the n = 599 threshold.

2.8) How did the meta-analysis account for the level of risk of bias identified? And how can we interpret this risk of bias?

Both our meta-regression and meta-analysis of seroprevalence ratios accounted for study risk of bias.

First, we conducted a multivariable linear meta-regression which included risk of bias as a categorical covariate. This analysis is described in the methods.

Page 10, line 214-219: To examine study-level factors affecting population-wide seroprevalence estimates, we constructed a multivariable linear meta-regression model. The outcome variable was the natural logarithm of corrected seroprevalence. Independent predictors were defined a priori. Categorical covariates were encoded as indicator variables, and included: study risk of bias (reference: low risk of bias), GBD region (reference: high-income); geographic scope (reference: national); and population sampled (reference: household and community samples).

Second, we conducted a meta-analysis to identify differences between sub-groups within studies. This analysis is described in the methods. In this analysis, the seroprevalence ratio between groups (e.g., the ratio between the seroprevalence in males and the seroprevalence in females) was first calculated within each study, so the risk of bias was controlled for inherently. The ratios were then pooled across studies.

Page 11, line 227-232: To quantify population differences in SARS-CoV-2 seroprevalence, we identified subgroup estimates within population-wide studies that stratified by sex/gender, race/ethnicity, contact with individuals with COVID-19, occupation, and age groups. We calculated the ratio in prevalence between groups within each study (e.g., prevalence in males vs. females) then aggregated the ratios across studies using inverse variance-weighted random-effects meta-analysis (S4 File). Heterogeneity was quantified using the I² statistic.35

We provide an interpretation of the risk of bias in the Supplement. The definition centres on the degree to which there is systematic error in an estimate that would result in its deviation away from the “true” value.

S3 file, page 10, line 120:

Item 10: Risk of bias

Low The estimates are very likely correct for the target population. To obtain a low risk of bias classification, all criteria must be met or departures from the criteria must be minimal and unlikely to impact on the validity and reliability of the prevalence estimate. These include sample sizes that are just below the threshold when all other criteria are met, reporting only some of characteristics of the sample, test characteristics below the threshold but corrections for the test performance, and response rates that are just below the threshold in the context of probability based sampling of an appropriate sampling frame with population weighted seroprevalence estimates.

Moderate The estimates are likely correct for the target population. To obtain a moderate risk of bias classification, most criteria must be met and departures from the criteria are likely to have only a small impact on the validity and reliability of the prevalence estimates.

High The estimates are not likely correct for the target population. To obtain a high risk of bias, many criteria must not be met or departures from criteria are likely to have a major impact on the validity and reliability of the prevalence estimates.

Unclear There was insufficient information to assess the risk of bias.

Reviewer 1 minor comments:

2.9) Introduction: 2nd paragraph: ‘previous infection’ – infection or exposure?

Previous infection was intended here, as individuals who were exposed may not be infected and may not mount an immune response.

Page 5, line 105: Serological assays identify SARS-CoV-2 antibodies, indicating previous infection in unvaccinated persons.7

2.10) Introduction: 4th paragraph: what gap? There was no clear gap identified up to here.

We have removed this language.

Page 6, line 118: We conducted a systematic review and meta-analysis of SARS-CoV-2 seroprevalence studies published in 2020

2.11) Introduction- what is the start and end dates for the lit review?

We have added the date

Page 6, line 118: We conducted a systematic review and meta-analysis of SARS-CoV-2 seroprevalence studies published in 2020.

These details are also provided in the methods.

Page 7, Line 149: Our search dates were from January 1, 2020 to December 31, 2020.

2.12) Introduction- ‘true burden’ – I wonder if this is the best term (which is mentioned throughout the manuscript). Doesn’t burden refer to mortality and morbidity? Or at least something that incurs some sort of cost. Many, if not most, seropositives will have been asymptomatic.

Throughout the manuscript we have replaced the term “burden” with “spread”, “infection”, or “prevalence”.

2.13) Data sources: Is there a reason to exclude PubMed?

Our health sciences librarian, Diane Lorenzetti, advised us that 98% of articles in PubMed are captured in the MEDLINE database. Given the large scope of our search strategy (4 databases, 4 public health agency websites, Google News search, serotracker platform submissions, expert recommendations) and ongoing nature of the living review we have tried to balance comprehensiveness with feasibility.

2.14) Data sources: Who is the librarian? At least add the affiliation.

Our librarian is Diane Lorenzetti. We have acknowledged her in the manuscript and have now added her affiliation to this acknowledgment.

Page 30, line 539-540: We would like to thank Dr. Diane Lorenzetti, a health science librarian at the University of Calgary, for her assistance in developing the search strategies.

2.15) Data sources: key eligibility criteria/ search words should be specified in the main text.

We have added details on key eligibility criteria to the methods section of the main text.

Page 7-8, line 154-168: We included SARS-CoV-2 serosurveys in humans. We defined a single serosurvey as the serological testing of a defined population over a specified time period to estimate the prevalence of SARS-CoV-2 antibodies.14,15 To be included, studies had to report a sample size, sampling date, geographic location of sampling, and prevalence estimate. Articles not in English or French were included if they could be fully extracted using machine translation.16 Articles that provided information on two or more distinct cohorts (different sample frames or different samples at different time points) without a pooled estimate were considered to be multiple studies.

If multiple articles provided unique information about a study, both were included. Articles reporting identical information to previously included articles were excluded as duplicates – this rule extended to pre-print articles that were subsequently published are peer-reviewed journals. In these cases, the peer-reviewed articles were considered the definitive version.

We have added details on the search to the methods section of the main text. The search strategies themselves are extensive; for this reason, we have left them in the supplement.

Page 6, line 129-135: We searched Medline, EMBASE, Web of Science, and Europe PMC, using a search strategy developed in consultation with a health sciences librarian (DL). The strategies for MEDLINE and EMBASE were an expanded version of the published COVID-19 search strategies created by OVID librarians for these databases.13 Search terms related to serologic testing were identified by infectious disease specialists (MC, CY, and JP)7 and expanded using Medical Subject Heading (MeSH) or Emtree thesauri. These searches were adapted for the other databases. The full search strategy can be found in S2 File.

2.16) Study selection: ‘SARS-CoV-2 infection’ –do you mean studies that included only previously PCR positives?

We have revised this exclusion criteria statement to offer more clarity.

Page 8, line 165-168: We excluded studies conducted only in people previously diagnosed with COVID-19 using PCR, antigen testing, clinical assessment, or self-assessment; dashboards that were not associated with a defined serology study; and case reports, case-control studies, randomized controlled trials, and reviews.

2.17) Study selection: Associated factors: there are far more studies for high-income countries, how do you take study effort into account for global or even large regional scales?

We have stratified the results by Global Burden of Disease region so that readers are aware of the proportion of data coming from high-income countries. In the meta-regression, we included study Global Burden of Disease region as a categorical covariate. We highlight in the discussion that the majority of data comes from high-income countries and that some of the estimates may therefore be driven by these data. We recommend that more studies be conducted in low and middle income countries.

Page 28, line 494-497: Thirdly, some of the summary results may have been driven by the large volume of data from high-income countries, which primarily reported lower seroprevalence estimates. While we frequently stratified by or adjusted for GBD region, caution is required when interpreting some of the summary estimates.

2.18) Results: what is considered general and special populations?

For clarity, we have changed this terminology to studies providing either population-wide or population-specific estimates. We have provided definitions for these groups in the methods.

Page 9, line 188-192: Seroprevalence studies were grouped as providing either population-wide or population-specific estimates. Population-wide studies included those using household or community sampling frames as well as convenience samples from blood donors or residual sera used for monitoring other conditions in the population. Population-specific studies were those sampling from well-defined population sub-groups, such as health care workers or long-term care residents.

2.19) Results: blood donors seem to be considered as general population, given they are typically young and healthier/fiter than average, are they not a special population?

Given that public health agencies often use blood donor samples as a practical strategy to measure seroprevalence in general population, we have opted to categorize them as studies providing population-wide seroprevalence estimates. We acknowledge the demographic and behavior differences between blood donors and the broader community and cite this in our discussion. Our meta-regression also quantifies the difference between seroprevalence in household/community samples, blood donor samples, and residual sera samples. After adjusting for confounding factors, the results show no statistically significant difference. This is a useful seroepidemiological finding that we have added to the discussion.

Page 15, line 302-306: In studies reporting population-wide seroprevalence estimates, median corrected seroprevalence was 4.5% (IQR 2.4-8.4%, Table 2). These studies included household and community samples (n=125), residual sera (n=248), and blood donors (n=54), with median corrected seroprevalence of 6.0% (IQR 2.8-15.1%), 4.0% (IQR 2.4-6.8%), and 4.7% (IQR 1.4-6.8%), respectively (Table 3).

Page 24-25, line 415-425: Approximately half of studies reporting population-wide SARS-CoV-2 seroprevalence estimates used blood from donors and residual sera as a proxy for the community. Our results showed that these studies report seroprevalence estimates that are similar to studies of household and community-based samples. It has previously been shown that these groups contain disproportionate numbers of people that are young, White, college graduates, employed, physically active, and never-smokers.47,48 However, the results of our study suggest that investigators may use these proxy sampling frames to obtain fairly representative estimates of seroprevalence if studies use large sample sizes with adequate coverage of important subgroups (e.g., age, sex, race/ethnicity) to permit standardization to population characteristics, tests with high sensitivity and specificity, and statistical corrections for imperfect sensitivity and specificity.

2.20) Results: the time window for these estimates need to be stated at the start of the results. I would imagine that now, seroprevalence is considerably higher in many regions/groups.

We have added this date range to the start of the results.

Page 12, line 250-251: Study sampling dates ranged from September 1, 2019 to December 31, 2020.

2.21) Table 4: Could remove rows for reference as this information is already in columns. The risk seems higher for children than adults? This seem to contradict many studies no?

Thank you for this suggestion. We have removed the reference rows from Table 5.

Using updated data, the results show that the risk for adults and children are not significantly different.

2.22) I wonder if some of the tables can be transformed into plots for an easier visualization?

We have included two additional figures in the main text to help with visualization (Figure 2, Figure 3).

2.23) Conclusion: 2nd paragraph: Or baseline health…. The sentence starting ‘Given’ is important and should be expanded. How does Community transmission impact SARS-CoV-2 transmission? It currently read transmission impacts transmission which seems a bit circular and empty. Is community transmission a proxy or behaviour?

Thank you for pointing this out. We have revised this statement in the conclusion.

Page 23, line 384-387: Given the limited evidence for altitude or climate effects on SARS-CoV-2 transmission36,37 variations in seroprevalence likely reflect differences in community transmission based on behaviour, public health responses, local resources, and the built environment.

2.24) Conclusion: what are the units of (24.0 local vs 11.9 national vs 15.7 regional)?

These were ratios between seroprevalence and cumulative incidence. We have clarified this metric in the conclusion.

Page 26, line 449-451: Seroprevalence estimates were 18.1 times higher than the corresponding cumulative incidence of COVID-19 infections, with large variations between the Global Burden of Disease Regions (seroprevalence estimates ranging from 6 to 602 times higher than cumulative incidence).

2.25) Conclusion: the 11.9 ratio values is without applying spatial heterogeneity in under-ascertains both between countries and within a country - and is biased by the countries that had capacity to perform a serological test. How would these estimates change if these heterogeneities were included?

This is a very important point. Thank you for raising it. We have provided a stratified analysis showing how the ratio between seroprevalence and cumulative incidence varies by Global Burden of Disease region. This means that separate estimates are now provided for high-income countries globally and for the low- and middle-income countries in each World Health Organization region. We now comment on these issues in the discussion and highlight the limitations of this data. We agree that bias in the overall estimate is introduced based on the disproportionate amount of data coming from high income countries and caution readers about this.

Page 26-27, line 449-478: Seroprevalence estimates were 18.1 times higher than the corresponding cumulative incidence of COVID-19 infections, with large variations between the Global Burden of Disease Regions (seroprevalence estimates ranging from 6 to 602 times higher than cumulative incidence). This level of under-ascertainment suggests that confirmed SARS-CoV-2 infections are a poor indicator of the extent of infection spread, even in high-income countries where testing has been more widely available. The broad range of ratios mirrors estimates from other published evidence on case under-ascertainment, which suggests a range of 0.56 to 717.49,50

Seroprevalence to cumulative case ratios can provide a rough roadmap for public health authorities by identifying areas that may be receiving potentially insufficient levels of testing and by providing an indication of the number of undetected asymptomatic infections.

While there is interest in using these seroprevalence to cumulative case ratios in identifying inadequate testing and estimating case ascertainment, caution is required in the quantitative interpretation of these ratios. Our study found a median ratio of 18.1, which aligns with other published analysis.50 This would imply that 2.9 billion people globally have been infected with SARS-CoV-2 rather than the 160 million reported as of May 15, 2021.2 This is not likely, and this estimate conflicts with the evidence that seroprevalence remains low in the general population. If applying this global ratio to countries with high cumulative incidence, such as the United States (32 million by May 15, 2021), then the total number of infections would exceed the population.

There are several possible reasons for these discrepancies. Firstly, these ratios clearly vary by geographic region and regional health policy, with higher diagnostic testing rates likely to correspond to lower seroprevalence to case ratios. Country-specific ratios, or region-specific ratios if available, should be used to inform planning wherever possible. Second, diagnostic testing-based estimates of cumulative incidence vary by assay; for example, lower RT-PCR cycle thresholds or the use of less sensitive rapid antigen tests would lead to lower estimates of cumulative cases. Finally, our analysis compares seroprevalence to cumulative case ratios at different point in time. As diagnostic testing measures expanded, these ratios may have declined over time, complicating the process of applying a single fixed ratio to a cumulative incidence number. As such, there is a need for more nuanced analysis of case under-ascertainment and caution should be exercised if utilizing them in public health planning.

2.26) Conclusion: P17 1st parag: ‘may not seroconvert’ - or antibodies could have wained by the time of blood collection...

Thank you for highlighting this. We have added that statement to the conclusion section.

Page 27, line 479-482: Firstly, some asymptomatic individuals may not seroconvert, some individuals may have been tested prior to seroconversion, and others may have antibodies that have waned by the time of blood collection, so the data in this study may underestimate the number of SARS-CoV-2 infections.51

2.27) Conclusion: Many studies have repeated patients. Was this considered?

This was considered. Part of our review process includes identifying studies with overlapping participants and linking the studies in our database. As such, these participants are not double counted during analysis.

2.27) Conclusion: P17 2nd parag: ‘there may be other factors…’ such as what?

We have provided more information about this limitation and added examples of potential confounding factors.

Conclusion, page X, line X-X: Fourthly, the residual heterogeneity in our meta-regression indicates that not all relevant explanatory variables have been accounted for. Many factors may contribute to the spread of infection. Even if all important factors were known, it would be difficult to account for the variation in seroprevalence due to limited availability of data with sufficient granularity and changing health policy and individual behavior.

2.28) Conclusion: P17 3rd parag: given the different level of scrutiny of these types of articles, do you think the results are comparable?

We agree that these different articles are subject to varying levels of scrutiny. We have relied on the risk of bias checklist and meta-regression to increase the comparability of these articles.

3) Reviewer 2 comments:

3.1) Reviewer #2: This is a clear and well written report of a systematic review/meta-analysis of the literature on sera-prevalence of SARS-CoV-2 antibodies published worldwide. The authors have a clear understanding of the pitfalls associated with both study design and laboratory evaluation of population based seroprevalence and have brought together the world literature up to August 28th 2020 in an accessible way with appropriate corrections.

Thank you for reviewing our manuscript. We have responded to your request below.

3.2) As this is such a rapidly evolving field, the only concern is whether this data adequately reflects the current situation. With a cut off date for analysis of late August 2020, most of the completed studies will represent seroprevalence estimates relatively early in the pandemic. If an updated analysis to the end of December 2020 could be incorporated into this manuscript that would be ideal and would add value as the authors could estimate seroprevalence in relation to time when the relevant population was sampled and that in turn could be evaluated in the context of time since the onset of the pandemic.

Thank you for suggesting this. We agree that data is rapidly emerging. As such, our team of reviewers have updated the search to include literature from January 1, 2020 to December 31, 2020. The review is now triple the size of the original draft. It grew from 338 studies reported in 221 articles to 968 studies reported in 605 articles. We have positioned this review as a summary of seroprevalence studies in 2020.

As the pandemic developed at different rates in different locations we have included a variable in the analysis to account for cumulative incidence of cases and, therefore, time when the relevant population was sampled relative to the onset of the pandemic in each country.

Submitted filename: 2021.05.15 Responses to reviewers.docx

Click here for additional data file.

19 May 2021

Global seroprevalence of SARS-CoV-2 antibodies: a systematic review and meta-analysis

PONE-D-20-40466R1

Dear Dr. Bobrovitz,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Yury E Khudyakov, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

15 Jun 2021

PONE-D-20-40466R1

Global seroprevalence of SARS-CoV-2 antibodies: a systematic review and meta-analysis

Dear Dr. Bobrovitz:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Yury E Khudyakov

Academic Editor

PLOS ONE