Controlling COVID-19 via test-trace-quarantine

Within 12 months of the emergence of COVID-19, diagnosed cases exceeded 60 million, with the true infection count likely higher. Early public health responses transitioned from border controls to broad non-pharmaceutical interventions (NPIs) such as lockdowns and physical distancing, which, while effective, carried substantial societal and economic costs. Governments have increasingly considered targeted 'test-and-trace' strategies, requiring only probable cases and their contacts to isolate or quarantine to achieve control with lower costs. Previous modeling and empirical studies indicate that success depends on the proportion of symptomatic cases identified, speed and completeness of contact tracing, and adherence to isolation and quarantine. Given R0 estimates of 2.4–5.6, achieving epidemic control generally requires a 60–80% reduction in effective contacts. There are global precedents (e.g., China, South Korea, Singapore, Taiwan) demonstrating feasibility, but concerns remain about resurgence after relaxation. This study asks what levels of testing, tracing, and quarantine effectiveness are required for a high-income, urbanized setting (Seattle/King County) to transition from mobility restrictions to a test-trace-quarantine (TTQ) approach prior to widespread vaccination, using Covasim calibrated to local data.

Multiple studies have assessed test-and-trace containment strategies, finding their effectiveness hinges on timely contact tracing, the fraction of symptomatic cases, and adherence to isolation/quarantine. Empirical successes include China’s robust tracing and mandatory quarantine, followed by South Korea, Singapore, and Taiwan. Observed resurgences upon relaxing measures underscore the need for sustained, high-performance implementation. Modeling work suggests that backward contact tracing and comprehensive tracing can mitigate spread, but that high transmissibility and potential immune escape variants complicate control. The literature collectively supports TTQ as a viable alternative to broad lockdowns if implemented with high coverage, speed, and compliance, potentially complemented by mask use and other NPIs.

The authors used Covasim, an open-source agent-based model incorporating intra-host viral dynamics and detailed demography/contact structure, to model SARS-CoV-2 transmission in Seattle/King County. Calibration covered January–June 2020 using age-stratified diagnosed cases and deaths from Washington State Department of Health, with additional context and data from Public Health–Seattle & King County (PHSKC). Mobility data from SafeGraph informed changes in workplace and community contacts. The model included multiple contact layers: households, schools, workplaces, community, and long-term care facilities (LTCFs). LTCF residents and staff were explicitly modeled using Washington state LTCF demographics and facility size/staffing data, with reduced external contacts for residents and structured facility contact subsets. Synthetic populations (approximately 225,000 agents representing 10% of King County, scaled to 2.25 million) were generated using Syntheths/SynPhoS approaches with ACS-based demographics and contact structures; contact weights by layer were set (household 100%, schools/workplaces 70%, community 10%, retirees 40%) to reflect time spent and known contact patterns. Model calibration used Optuna (Tree-structured Parzen Estimator) to fit key parameters (e.g., per-contact transmission probability, relative transmissibility changes over time, LTCF transmission reduction, symptomatic testing rates/odds ratios) to minimize error against observed time series of cases/deaths and testing metrics (volume, positivity, symptom-to-swab delays). Calibrations were performed with and without explicit mobility inputs to separate effects of mobility vs other NPIs (distancing, hygiene, masks). The model estimated initial growth (pre-intervention R0), attack rate, diagnosis rate, and effective reproduction number over time. Scenario analysis comprised: (1) idealized TTQ in a hypothetical population (full tracing, universal quarantine of contacts for 14 days regardless of test result, minimal delays shorter than infectious period) to illustrate self-limiting dynamics; (2) realistic TTQ in Seattle with three reopening levels (60%, 80%, 100% workplace/community mobility) under continued masks/NPIs, varying six TTQ parameters: isolation/quarantine effectiveness, routine testing rate (tests per 1000 per day), tracing probability (household/workplace), testing of traced contacts, swab-to-result delay, and tracing delay. Baseline high-mobility scenario assumed ~6000 routine tests/day, 70% of household/workplace contacts traced within 2 days, and continued mask/other NPI reducing transmissibility by ~10–15%. Sensitivity analyses examined the marginal impact of each parameter on attack rate and infections averted per unit change. Model validation compared projections from 1 June 2020 against observed June–August 2020 testing, diagnoses, tracing, and inferred infections, including a counterfactual high testing/tracing scenario.

- Calibration and early epidemic dynamics (Seattle/King County, Jan–Jun 2020): - Estimated cumulative infections by 6 June 2020: ~100,000 (95% CI: 80,000–115,000), attack rate 3.5–5.1% in a population of 2.5 million. - Diagnosed cases by 9 June: 8,548; overall diagnosis rate ~9% (95% CI: 7–11%). - Pre-intervention R0 estimated at 2.3 (95% CI: 2.0–2.6); Re dropped below 1 during shelter-in-place. - Including mobility trends, relative transmissibility decreased by 12 ± 5% versus initial value (reflecting additional NPIs such as masks, distancing, hygiene). Without mobility data, a 71 ± 3% drop in transmissibility was required, separating mobility from other NPIs. - LTCF transmission reductions estimated at 80–92%. - Symptomatic testing odds ratio high (12–74) with routine testing yield declining from 10–15% (March) to 1.5–2.5% (early June) as incidence fell. - Transmission heterogeneity and settings: - Workplace/community contact layers accounted for ~58% of infections early, then ~52% post-distancing, indicating persistent non-household transmission. - Highly overdispersed transmission: ~53% of primary infections caused zero secondary infections; 10% of primary infections caused 50% of secondary infections; these 10% infected on average ~6.3 others. - Estimated 54% of transmissions from symptomatic individuals. - Traced contact positivity was far higher than routine testing: ~34% positivity for traced contacts versus ~1.5% for routine testing in early June. - Idealized TTQ dynamics: - With complete tracing, universal quarantine of contacts, and delays shorter than infectious periods, TTQ yields self-limiting dynamics capable of control even at high Re and regardless of start time. - Realistic TTQ requirements for high mobility: - Returning to 100% workplace/community mobility without other measures drives Re well above 1 and a second wave. - High testing and high tracing are both necessary to maintain control at full mobility; a fourfold increase in testing alone is insufficient without increased tracing. - Marginal impacts (Fig. 4a): - Isolation/quarantine effectiveness: 1.20 ± 0.04 infections averted per person fully isolated. - Routine testing probability: 0.21 ± 0.01 infections averted per routine test (including negatives). - Swab-to-result delay: +3.94 ± 0.14 additional infections per day of delay per positive test. - Contact tracing probability: 3.75 ± 0.15 infections averted per index case. - Quarantine testing probability: 0.40 ± 0.16 infections averted per index case whose contacts are tested. - Contact tracing delay: +1.95 ± 0.13 additional infections per day of delay per index case. - Validation (June–Aug 2020): - The model calibrated to data through 1 June reproduced observed trends over the subsequent 3 months, including increased transmission among younger age groups post-reopening and the impact of scaled-up testing/tracing. - A counterfactual high testing/tracing scenario would have produced a faster decline in active infections, with transiently higher daily tests (up to ~3500 vs ~500 observed) and contacts traced (~900 vs ~200), but by 31 August would require similar or fewer tests/traced contacts due to lower incidence.

The study addresses whether test-trace-quarantine can replace broad mobility restrictions in a high-income, urban context. Findings show that while reduced mobility initially achieved control in Seattle, sustained control under high mobility is feasible with strong TTQ performance: high routine testing focused on symptomatic individuals, high probabilities of contact tracing (especially household/workplace), prompt test result turnaround and rapid tracing, and high adherence to isolation and quarantine. Transmission is highly overdispersed and dominated by non-household settings, implying that rapid identification and isolation of clusters is crucial; TTQ leverages positive feedback—fewer cases improve TTQ effectiveness, further reducing transmission. International comparisons indicate that robust TTQ combined with mask use and other NPIs can maintain control without lockdowns, but shortfalls in any TTQ component risk resurgence. When cases rise quickly and strain TTQ capacity, mobility restrictions may again be necessary absent high vaccination coverage. The model’s accurate reproduction of observed trends during June–August 2020 strengthens the evidence that TTQ can work in practice if implemented at sufficient scale and speed.

Agent-based modeling with Covasim accurately fit detailed epidemic time trends and age distributions and produced validated short-term forecasts for Seattle/King County. Idealized TTQ can theoretically achieve epidemic control even at high transmission rates via self-limiting dynamics, while realistic scenarios show that high testing and tracing rates, rapid turnaround, and high quarantine compliance can maintain control even with a return to full mobility and low vaccine coverage. Each TTQ component contributes meaningfully; deficits in one area must be offset by superior performance elsewhere (e.g., masking or mobility reduction). The Seattle experience provides empirical support for TTQ as a feasible control strategy. Future research should refine estimates of age-specific susceptibility/transmissibility, seasonality, and asymptomatic/presymptomatic transmission; incorporate geographic clustering and day-of-week mobility patterns for hotspot targeting; and evaluate capacity constraints and equity considerations within TTQ systems.

Key limitations include: (1) no explicit modeling of geographic clustering or day-of-week mobility changes, limiting hotspot/outbreak resolution; (2) uncertainty in age- and comorbidity-specific susceptibility and transmissibility parameters; (3) uncertainty in SARS-CoV-2 characteristics (seasonality, proportions of asymptomatic/presymptomatic transmission) and intervention impacts (e.g., mask efficacy). While extensive calibration and uncertainty propagation were employed, additional empirical data would improve parameterization and pathway-specific understanding. Capacity constraints and behavioral adherence, while discussed, are challenging to model precisely and may affect generalizability.