[Skip to Navigation]
Sign In
COVID-19: Beyond Tomorrow
May 15, 2020

Reopening Society and the Need for Real-Time Assessment of COVID-19 at the Community Level

Author Affiliations
  • 1Medical Development and Scientific/Clinical Affairs, Pfizer Vaccines, Collegeville, Pennsylvania
  • 2Center for Observational and Real-World Evidence, Merck & Co Inc, Kenilworth, New Jersey
JAMA. 2020;323(22):2247-2248. doi:10.1001/jama.2020.7872

Through May 13, 2020, approximately 1.39 million cases of coronavirus disease 2019 (COVID-19) have been reported in the United States (by the Centers for Disease Control and Prevention) and more than 4.3 million cases of COVID-19 have been reported from 188 countries.1 There is an urgent need for COVID-19 data, including community-level incidence, spectrum of disease, diagnostic test penetration, and proportion of the community with protective immunity to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (herd immunity). These data are vital to understanding where communities are on the continuum of COVID-19 cumulative incidence and prevalence and how nonpharmaceutical interventions can be titrated to reopen business and society. Real-time incidence and seroepidemiologic data are also essential to plan scenarios for the development of COVID-19 vaccines and therapeutics. Cross-sectional community surveys combined with seroepidemiology can help inform the present and help navigate the path forward.

Determining Where a Community Is Now

Rapid community surveys have been widely used by the World Health Organization and other groups to rapidly assess health conditions such as immunization coverage or to conduct demographic surveys.2 Although other sampling methods have been used, 2-stage cluster sampling design is commonly used in rapid community surveys. Balancing precision and speed, these surveys frequently include 7 households randomly selected from each of 30 clusters, with the clusters selected from the community using “probability proportionate to size” sampling. In a rapid community COVID-19 survey, selected individuals would be queried for a designated recall period and asked if they or their household members have or had COVID-19 symptoms; have sought care, been tested, been hospitalized, been hospitalized in intensive care, or undergone ventilation; and if household members have died.

If a blood specimen and an upper respiratory tract specimen are collected at the time of the survey, the proportion of the community with antibodies to SARS-CoV-2 and the proportion of the community with symptomatic past and current infections can be estimated. These data are fundamental for determining the effectiveness of previous community mitigation strategies and the direction of future such strategies. Simultaneous rapid surveys in multiple adjacent communities can provide generalizable information across geographic areas to allow synchronization and titration of public health interventions and to allow more seamless societal interaction.

A hypothetical example can be used to illustrate how surveillance data and rapid community and seroepidemiologic surveys can be used to estimate where a community is in the continuum of pandemic disease transmission. In this hypothetical scenario, assume that laboratory surveillance data demonstrate that 3000 laboratory-confirmed COVID-19 cases have occurred in a community of 500 000 persons. Also assume that the community survey reveals that only 10% of respondents with symptoms report that they have been tested for COVID-19; therefore, only one-tenth of patients with suspected COVID-19 have been detected. Extrapolating these hypothetical findings to the larger community would suggest that an estimated 30 000 persons (3000 × 10) in the community likely had symptomatic illness.

The proportion of persons with symptoms from the rapid survey could be compared with serologic results to estimate the proportion of persons that have asymptomatic infection with SARS-CoV-2. With the additional assumption that the serosurvey in this community demonstrates that 20% of the population has antibodies to SARS-CoV-2, extrapolation of these hypothetical data would suggest that an estimated 100 000 people (0.2 × 500 000) in the community have been infected. If the reproduction number, R0 (the average number of cases of transmission caused by an infected person), for COVID-19 is 2.2,3 then the estimated proportion of the population who would need to be immune to SARS-CoV-2 to achieve herd immunity (estimated as 1 − 1/R0) would be 55%.4

The results of community and seroepidemiology surveys (indicating that 20% of the population has antibodies to SARS-CoV-2) would suggest that herd immunity has not been achieved and that an additional 35% of the population would need to be infected before there is herd immunity. The results of the seroepidemiologic survey also would suggest that this community could expect to have continued SARS-CoV-2 transmission and should implement public health measures accordingly.

A multiplier model is a complimentary approach that can be used to assess the current status of community-level disease in the pandemic. Multiplier models use extant surveillance data that are adjusted by survey and other data for underascertainment and underreporting. Surveillance data need adjustment because not all symptomatic persons seek care from physicians or other health care clinicians, not all clinicians or health care centers order tests for COVID-19, test kits and supplies may not be available, and test sensitivity and specificity are less than 100%. A multiplier modeling approach was used early in the 2009 influenza A(H1N1) pandemic to estimate the potential number of cases that had occurred at a time when only laboratory-confirmed cases were being reported in the United States. The model used multipliers based on estimates from general population and community surveys that determined the proportion of persons with influenza-like symptoms who sought medical care, the proportion of those who sought care who had a diagnostic specimen collected, the sensitivity of the tests, and the proportion of laboratory-confirmed cases that were reported to public health. According to this modeling approach, although only 43 677 laboratory-confirmed influenza A(H1N1) cases were reported, an estimated 1.8 million to 5.7 million cases may have occurred during the first 4 months of the pandemic in the United States.5

The multiplier approach was also used at the end of the influenza A(H1N1) pandemic to estimate 60.8 million clinical influenza cases in the United States (range, 43.3 million–89.3 million) through the first year.6 These data, combined with serologic data, estimated that approximately 40% of influenza A(H1N1)–infected persons were asymptomatic7 and suggested that an estimated 102 million persons in the United States could have been infected with influenza A(H1N1) in 2009-2010, equal to approximately 33% of the US population. Given that the estimated proportion necessary for herd immunity was 41% (an R0 of influenza A[H1N1] of approximately 1.7),8 these data suggested that the level of herd immunity in the United States approached the level needed to disrupt sustained transmission of influenza A(H1N1).

Determining Where a Community Will Be in the Coming Months

Projections of likely future COVID-19 scenarios for communities are needed for policy making, public health planning, and intervention development. Communities will need to project future scenarios to adjust intensity of nonpharmaceutical interventions, assuming which specific scenarios and how well they work is known, and to decide when to reopen businesses and schools. In addition, information about baseline levels of immunity are important for planning clinical trials for vaccines and therapeutic agents. Community and seroepidemiologic surveys combined with modeling approaches are useful tools for making projections of future scenarios.

National and local COVID-19 scenario projection models have been developed by a number of groups. Inputs for these models include the number of laboratory-confirmed COVID-19 cases and deaths; underlying assumptions may include the sensitivity of case and death reporting, the number of symptomatic persons tested for COVID-19, and the sensitivity of diagnostic tests. Other model inputs include estimates of the effects of community mitigation efforts and achieving herd immunity on projected declines in COVID-19 incidence. Although these models have been controversial, community and seroepidemiologic surveys can capture data that reduce the uncertainty around underlying assumptions. Repeated seroepidemiologic surveys can provide cross-sectional estimates of the degree of success of nonpharmaceutical interventions and progress toward herd immunity. These data provide iterative inputs for the model to focus model projections and can assist in model validation.

Several groups have initiated ongoing community and seroepidemiologic surveys. A preliminary report of a community serosurvey that was conducted April 3-4, 2020, at 3 drive-through test locations and included 3330 persons invited to participate to be representative of Santa Clara County, California (population, 1.9 million), found a weighted prevalence of 2.5% to 4.2% immunity; with 956 laboratory-confirmed COVID-19 cases on April 1, these data suggest that there are 50 to 85 persons with antibodies for each laboratory-confirmed case in Santa Clara County.9 However, because the specificity of the serologic assay was less than 100%, there is the potential for false-positives in this low-prevalence population and therefore the multiplier may be lower than estimated. A preliminary report of a community serosurvey conducted in mid-April at 40 locations in 19 counties in New York State (population, 19.4 million) among 3000 persons who were shopping found 13.9% immunity; with 213 779 laboratory-confirmed cases on April 15, these data suggest that there are approximately 13 persons with antibodies for each laboratory-confirmed case.10

Accurate assessment of COVID-19 at the community level requires reliable evidence about the proportion of symptomatic persons tested for COVID-19, the proportion of cases that are asymptomatic, COVID-19 cumulative incidence, the proportion of persons who are hospitalized, and the proportion who die. Serial measurement of antibody levels and population immunity is needed to titrate nonpharmaceutical interventions and for development of COVID-19 vaccines and therapeutics. Integrated community and seroepidemiologic surveys combined with modeling and multiplier approaches to estimate the incidence and prevalence of COVID-19 at the community level are essential to assess the current status of the pandemic, to plan ongoing and future interventions, and to inform decision-making for gradually returning to normal activities within communities.

Back to top
Article Information

Corresponding Author: David L. Swerdlow, MD, Pfizer Vaccines, 500 Arcola Rd, Collegeville, PA 19426 (david.swerdlow@pfizer.com).

Published Online: May 15, 2020. doi:10.1001/jama.2020.7872

Conflict of Interest Disclosures: Dr Angulo reported being employed by Pfizer Vaccines and owning stock and stock options in Pfizer. Dr Finelli reported being employed by Merck & Co Inc and may own stock in the company. Dr Swerdlow reported being employed by Pfizer Vaccines and owning stock and stock options in Pfizer, as well as providing overviews of SARS and Middle East respiratory syndrome epidemiology to a consulting firm for a minimal honorarium.

Funding/Support: This work was supported by Pfizer.

Role of the Funder/Sponsor: Pfizer Inc reviewed the manuscript and approved the decision to submit the manuscript for publication.

Johns Hopkins University Coronavirus Resource Center. Accessed April 24, 2020. https://coronavirus.jhu.edu/map.html
Bennett  S, Woods  T, Liyanage  WM, Smith  DL.  A simplified general method for cluster-sample surveys of health in developing countries.   World Health Stat Q. 1991;44(3):98-106.PubMedGoogle Scholar
Flaxman  S, Mishra  S, Gandy  A,  et al. Report 13: Estimating the Number of Infections and the Impact of Non-pharmaceutical Interventions on COVID-19 in 11 European Countries. Imperial College London; March 30, 2020. doi:10.25561/77731
Fine  P, Eames  K, Heymann  DL.  “Herd immunity”: a rough guide.   Clin Infect Dis. 2011;52(7):911-916. doi:10.1093/cid/cir007PubMedGoogle ScholarCrossref
Reed  C, Angulo  FJ, Swerdlow  DL,  et al.  Estimates of the prevalence of pandemic (H1N1) 2009, United States, April-July 2009.   Emerg Infect Dis. 2009;15(12):2004-2007. doi:10.3201/eid1512.091413PubMedGoogle ScholarCrossref
Shrestha  SS, Swerdlow  DL, Borse  RH,  et al.  Estimating the burden of 2009 pandemic influenza A (H1N1) in the United States (April 2009–April 2010).   Clin Infect Dis. 2011;52(52)(suppl 1):S75-S82. doi:10.1093/cid/ciq012PubMedGoogle ScholarCrossref
Cowling  BJ, Chan  KH, Fang  VJ,  et al.  Comparative epidemiology of pandemic and seasonal influenza A in households.   N Engl J Med. 2010;362(23):2175-2184. doi:10.1056/NEJMoa0911530PubMedGoogle ScholarCrossref
World Health Organization.  Mathematical modelling of the pandemic H1N1 2009.   Wkly Epidemiol Rec. 2009;84(34):341-348.PubMedGoogle Scholar
Bendavid  E, Mulaney  B, Sood  N,  et al. COVID-19 antibody seroprevalence in Santa Clara County, California. medRxiv. Preprint posted April 30, 2020. doi:10.1101/2020.04.14.20062463
LaVito  A, Brown  KV, Clukey  K. New York finds virus marker in 13.9%, suggesting wide spread. Bloomberg website. Published April 23, 2020. Accessed May 11, 2020. https://www.bloomberg.com/news/articles/2020-04-23/new-york-finds-virus-marker-in-13-9-suggesting-wide-spread
1 Comment for this article
An example of real time analysis of Covid 19 data at the level of the community
Al Pfadt, PhD | New York State Institute for Basic Research, retired research scientist
The article by Angelo, Finelli, and Swerdlow published in today’s issue of JAMA calls attention to a critical lack of real time analyses of Covid 19 data at the level of local communities. Most studies reported in the literature use complex mathematical models that use historical data to make predictions regarding certain parameters of interest to policy makers (such as anticipated hospital bed utilization) using certain assumptions that are hard-wired into the model. Accordingly, we have a plethora of articles anticipating when a country (like the United States) or state (like New York) is likely to bend its response curve regarding cumulative frequencies of reported incidents or deaths attributed to the virus.

However, these data gathered at that level of analysis offer little guidance for local communities who need real time data analysis to inform decisions they must make today. Neither do they offer a way to use objective outcome indicators to evaluate the effects of those decisions in real time.

Angelo et al provide a refreshing antidote to this problem that illustrates how cross- sectional data provided by community surveys can be used in conjunction with estimates of epidemiological parameters (the number of people in the community who have been exposed to the virus) to “ inform the present and help navigate a path forward“. p E1).

My colleagues Drs. Donald Wheeler and Kathyrn Whyte and I published an article in Quality Digest ( April 6th, 2020) that provides an example of how to track Covid 19 data in real time to evaluate outcomes associated with mitigation strategies implemented at the community level. (1). In this case, our level of analysis was the municipality of New Rochelle, located in Westchester County, within commuting distance of NY City.
On March 2nd, 2020, the first case of Covid 19 was reported in the New York City metropolitan area. This patient had been sick since February 27th but Covid 19 was not diagnosed until March 2nd, when contact tracing was immediately initiated. Our article presents information about the spread of the disease between then and March 31st. During that interval the number of reported cases increased exponentially to the point where 378 people had been diagnosed and 25 deaths had been reported. Of particular interest for readers of this article is our graphic display of the number of new cases reported in New Rochelle during that 29 day interval. While it is not possible to establish a causal relationship between variables based of data gathered under such real world conditions, we were able to correlate the initiation and modification of certain critical containment efforts with dramatic changes in the distribution of new cases reported each day. Closing the synagogue where the initial spread of the disease started and progressively stricter social isolation practices (culminating in the establishment of a clear isolation zone within the city) were associated with significant reductions in the number of cases reported each day. Likewise, orders that relaxed this level of social isolation was followed by a return to previous levels of case reporting and in fact the highest daily total reported during that period occurred on March 31st, the last day we studied.

Together with more precise epidemiological data reported by Angelo et al, real time investigations of the spread of Covid 19 within local communities can help accumulate an evidence base that truly helps inform contemporaneous decisions and evaluate the extent to which they accomplish their stated objectives.


1. https://www.qualitydigest.com/inside/healthcare-article/tracking-covid-19-033120.html