Modeling serological testing to inform relaxation of social distancing for COVID-19 control

The study investigates whether integrating serological testing into COVID-19 control could enable safe relaxation of social distancing by identifying individuals with antibodies who are likely to have partial immunity. Against the backdrop of multiple epidemic waves in 2020–2021 and variable adherence to social distancing, the authors propose an immune shielding strategy: deploying seropositive (and later, vaccinated) individuals into higher-contact roles to reduce exposure risk for susceptibles while maintaining essential societal functions. The introduction reviews that seroprevalence exceeded documented cases, most infected individuals seroconvert with antibodies persisting for months, and reinfections were relatively rare at the time. The key challenge is test performance—especially specificity—since false positives could increase risk if misclassified individuals resume normal contacts. The research question is how serology-informed shielding, test frequency, and assay performance would affect transmission, mortality, and the proportion of the population that could be safely released from distancing across three distinct U.S. metropolitan areas.

The paper summarizes evidence that seroprevalence substantially exceeded reported cases, indicating many undetected infections. It notes that most infected individuals seroconvert and maintain detectable antibodies for months; reinfections had been documented but were uncommon. Over 50 serological assays had FDA EUA with wide performance variability; for policy, specificity is more critical than sensitivity due to risks of false positives. Prior serosurveys (e.g., Havers et al.) provided cross-sectional seroprevalence benchmarks used for model fitting. The authors also reference concerns regarding variants potentially affecting immunity and emphasize that while antibody levels may wane, protection can persist. This context motivates evaluating serology-informed shielding while accounting for test specificity and operational realities in mass testing.

The authors developed a deterministic, age-structured compartmental transmission model (susceptible, exposed, infectious, symptomatic/asymptomatic, hospitalized, recovered/removed) with additional layers for test status (test-positive vs not tested/test-negative) to reflect serological screening and potential misclassification. Contacts are modeled across home, work, school, and other settings for three age groups (0–20, 20–64, 65+). The model incorporates serological test sensitivity (for detecting prior infection) and specificity (to avoid misclassifying non-immune individuals). False positives/negatives are explicitly represented as flows dependent on specificity/sensitivity. Social distancing policies alter contact matrices by location and time, including stay-at-home orders in March–June 2020 and partial reopening thereafter; schools are assumed to reopen at 50% capacity on Sept 1, 2020 (South Florida) and Oct 1–2, 2020 (Washington, New York). Serology-informed shielding is implemented by preferentially assigning test-positive individuals to higher-contact roles, increasing the probability that contacts occur with presumed immune individuals; a 5:1 fixed shielding factor is examined (i.e., contacts at work/other settings are five times more likely with test-positive persons than expected by their population frequency). Testing strategies vary by frequency (monthly to annual) and by assay performance (specificity 99.8% representing high-performance tests; 90% representing suboptimal mass-scale rollout; a 50% scenario to illustrate an unreliable correlate of immunity). Model fitting: A Markov Chain Monte Carlo (MCMC) approach is used to fit to time series of reported deaths and cross-sectional seroprevalence estimates for New York City, South Florida, and Washington Puget Sound through mid-2020. Key parameters estimated include infection probability per contact, fractions symptomatic, and social distancing parameters. Credible intervals are obtained from converged MCMC chains. Forward simulations project outcomes to June 1, 2021 under counterfactual testing/shielding scenarios. Sensitivity analyses: Latin Hypercube Sampling and partial rank correlation coefficients evaluate robustness to uncertainties in latent period, relative transmissibility of asymptomatic infections, hospital length of stay, and durations of symptomatic/asymptomatic infection. Assumptions include homogeneous mixing within strata, immediate detectability of antibodies post-resolution (acknowledging real-world ~11–14 day seroconversion), retained immunity over the model horizon, and antibodies as a correlate of protection.

- Without serological testing: Models predict second epidemic peaks in fall–winter 2020–2021 across all three sites. If fall 2020 distancing levels persisted without further interventions, by June 2021 an estimated 45–55% of the population across the sites would have been infected (e.g., 55% in New York, 95% CrI: 27–69%; 46% in Washington, 95% CrI: 2–55%), with location-specific cumulative deaths since the start of the pandemic of approximately 43,000 (New York City; 95% CrI: 21,000–64,000), 10,000 (South Florida; 95% CrI: 6,000–17,000), and 10,000 (Washington; 95% CrI: 100–32,000). - With serological shielding (high specificity test, shielding factor ~1.5 to 5:1; implementation beginning Nov 20, 2020): Widespread serological testing combined with moderate shielding could have reduced cumulative deaths by June 2021 by 22% across the three sites. The strongest relative reduction is in Washington (59%; 95% CrI for deaths averted: 0–17,000), with smaller relative impacts in New York City (8%; 95% CrI: 300–600 deaths averted) and South Florida (14%; 95% CrI: 900–1300 deaths averted). - Abstract-reported averted deaths under widespread testing starting in late 2020: approximately 3,300 (New York City), 1,400 (South Florida), and 11,000 (Washington) by June 2021; all sites showed blunted subsequent waves. - Testing frequency matters: More frequent testing increases both deaths averted and the fraction of the population safely released from social distancing when using a highly specific test. • New York City: Monthly testing could release 51% (95% CrI: 27–70%) from distancing by June 1, 2021 and reduce deaths by ~300 (95% CrI for total deaths: 20,000–63,000). Annual testing results in higher mortality and fewer released (e.g., ~21% released; 95% CrI: 13–37%). • Washington: Monthly testing could release 21% (95% CrI: 3–33%) and avert ~11,000 deaths (95% CrI for total deaths: 1,000–150,000), versus annual testing releasing 14% (95% CrI: 1–22%) and averting ~400 deaths (95% CrI for total deaths: 1,000–26,000). • South Florida: Monthly testing could release 41% (95% CrI: 24–61%) and avert ~1,500 deaths (95% CrI for deaths: 5,000–15,000), versus yearly testing releasing 21% (95% CrI: 12–32%) and averting ~500 deaths. - Test specificity is critical: With 90% specificity, cumulative deaths across the three locations (~66,000) are lower than no-testing (~72,000) but higher than with a high-performance test (~56,000). Monthly testing with a suboptimal assay (90% specificity) without shielding could release 97–99% of the population from distancing, potentially increasing risk; if antibodies are not a reliable correlate of protection (e.g., very low effective specificity), serological testing could lead to more deaths than no testing. - Sensitivity analyses: Greater transmissibility of asymptomatic cases increases the benefit of shielding; faster recovery reduces the impact of shielding in settings with milder initial waves (South Florida, Washington). Changes in latent period and hospitalization duration had minimal effects.

The modeling shows that in the pre-vaccine era, frequent serological testing coupled with strategic shielding could have substantially reduced mortality while enabling a sizeable portion of the population to relax social distancing, thereby mitigating the social and economic burdens of stringent measures. The approach is most effective with highly specific assays and when implemented with preferential placement of immune individuals in higher-contact roles. However, the strategy can backfire—especially with low specificity tests, absent shielding, or if antibodies are not a reliable correlate of protection—potentially increasing deaths. Even in the vaccine era, serology-informed shielding can complement vaccination to identify individuals with immunity (from infection or vaccination) who can safely resume higher-contact activities, particularly in high-risk settings like healthcare and long-term care facilities, and amid concerns about variant emergence and heterogeneous vaccine uptake.

Serologic testing integrated into a shielding strategy could have meaningfully blunted COVID-19 waves in late 2020–2021, averting deaths and allowing many individuals to safely relax social distancing. High-frequency testing with highly specific assays, combined with preferential placement of immune individuals into high-contact roles, is key to maximizing benefits and minimizing risks. In ongoing and future responses—including potential variant-driven resurgences—serology can complement vaccination status to guide safer relaxation of distancing, particularly in targeted high-risk settings. Future work should refine behavioral contact estimates, optimize targeting of testing to high-yield groups, evaluate integration with vaccine certificates, and assess performance against emerging variants and waning immunity.

- Behavioral uncertainty: Considerable uncertainty in how social distancing and contact patterns evolved through mid-2021; limited empirical data on mixing patterns constrain parameter identifiability. - Random test allocation: The model assumes random serology testing; targeted strategies (e.g., healthcare workers, high-contact occupations) could increase efficiency and reduce false positive burden. - Immunological assumptions: Assumes immediate detectability of antibodies post-infection resolution, durable immunity over the model horizon, and antibodies as a correlate of protection; waning and variant immune escape could alter outcomes. - Data limitations: Reliance on available seroprevalence data (e.g., Havers et al.) and assumption of constant age-specific case fatality rates may bias death estimates. - Model structure: Homogeneous mixing within strata and simplifying compartmental assumptions may not capture heterogeneity in networks or settings. - Ethical/legal considerations: Potential inequities in access to testing, risks of perverse incentives for deliberate infection, and concerns about immunity certification; these risks may be mitigated in the vaccine era but remain relevant. - Test performance and implementation: Population-level benefits hinge on high specificity; mass rollout challenges could degrade accuracy. If antibodies are not a reliable correlate, testing could be harmful.