Optimizing COVID-19 control with asymptomatic surveillance testing in a university environment

Abstract

The high proportion of transmission events derived from asymptomatic or presymptomatic infections make SARS-CoV-2, the causative agent in COVID-19, difficult to control through the traditional non-pharmaceutical interventions (NPIs) of symptom-based isolation and contact tracing. As a consequence, many US universities developed asymptomatic surveillance testing labs, to augment NPIs and control outbreaks on campus throughout the 2020-2021 academic year (AY); several of those labs continue to support asymptomatic surveillance efforts on campus in AY2021-2022. At the height of the pandemic, we built a stochastic branching process model of COVID-19 dynamics at UC Berkeley to advise optimal control strategies in a university environment. Our model combines behavioral interventions in the form of group size limits to deter superspreading, symptom-based isolation, and contact tracing, with asymptomatic surveillance testing. We found that behavioral interventions offer a cost-effective means of epidemic control: group size limits of six or fewer greatly reduce superspreading, and rapid isolation of symptomatic infections can halt rising epidemics, depending on the frequency of asymptomatic transmission in the population. Surveillance testing can overcome uncertainty surrounding asymptomatic infections, with the most effective approaches prioritizing frequent testing with rapid turnaround time to isolation over test sensitivity. Importantly, contact tracing amplifies population-level impacts of all infection isolations, making even delayed interventions effective. Combination of behavior-based NPIs and asymptomatic surveillance also reduces variation in daily case counts to produce more predictable epidemics. Furthermore, targeted, intensive testing of a minority of high transmission risk individuals can effectively control the COVID-19 epidemic for the surrounding population. Even in some highly vaccinated university settings in AY2021-2022, asymptomatic surveillance testing offers an effective means of identifying breakthrough infections, halting onward transmission, and reducing total caseload. We offer this blueprint and easy-to-implement modeling tool to other academic or professional communities navigating optimal return-to-work strategies.

Keywords: Asymptomatic surveillance testing; Branching process model; COVID-19; University control.

Graphical abstract

Fig. 1

Conceptual schematic of branching process model of SARS-CoV-2 dynamics., Person A is isolated through testing after exposing Person B and Person C. Person B is then isolated through contact tracing, while Person C is not traced but is nonetheless ultimately isolated through symptomatic surveillance. A viral titer trajectory (right) is derived from a within-host viral kinetics model (Text S2)—independent trajectories from 20,000 randomly-selected individuals are shown here to highlight the range of possible variation. The 25th and 75th titer threshold percentile for the onset of symptoms are depicted in pink, such that 32% of individuals modeled in our simulations did not present symptoms.

Fig. 2

Effects of group size limits on COVID-19 dynamics., A. Negative binomial R_E distribution with mean = 1.05 and dispersion parameter (k) = 0.10. The colored vertical dashes indicate group size limits that ‘chop the tail’ on the R_E distribution; for 90% of the population, coincident cases allocated to the same transmission event were truncated at the corresponding threshold for each intervention. B. Daily new cases and, C. Cumulative cases, across a 50-day time series with 95% confidence intervals by standard error depicted under corresponding, color-coded group size limits.

Fig. 3

Impacts of NPIs on COVID-19 control., A. Mean reduction in R_E * and B. cumulative cases saved across 50-day simulated epidemics under assumptions of differing non-pharmacological interventions (NPIs). NPIs are color-coded by threshold number of persons for group-size limits, lag-time for symptom-based isolations, and mean turnaround time from test positivity to isolation of infectious individuals for testing isolations. For testing isolations, shading hue corresponds to test limit of detection with the darkest colors indicating the most sensitive tests with a limit of detection of 10¹ virus copies/μl of RNA. Progressively lighter shading corresponds to limits of detection = 10³, 10⁵, and 10⁷ cp/μl. *Note: R_Ereduction (panel A) is calculated as the difference in mean R_Ein the absence vs. presence of a given NPI. The upper confidence limit (uci) in R_Ereduction is calculated as the difference in uci R_Ein the absence vs. presence of NPI. In our model, mean R_Ein the absence of NPI equals 1.05 and uci R_Ein the absence of NPI equals 8.6.

Fig. 4

Combining behavioral and asymptomatic surveillance testing NPIs for COVID-19 control., A. Mean reduction in R_E * , B. cumulative cases saved, and C. daily case counts for the first 50 days of the epidemic, across regimes of differing testing frequency and a combination of asymptomatic surveillance testing, contact tracing, symptomatic isolation, and group size limit interventions. All scenarios depicted here assumed test turnaround time, symptomatic isolation lags, and contact tracing lags drawn from a log-normal distribution with mean=one day. Limit of detection was fixed at 10¹ and group size limits at 12. Dynamics shown here are from simulations in which testing was limited to two test days per week., *Note: R_Ereduction (panel A) is calculated as the difference in mean R_Ein the absence vs. presence of a given NPI. The upper confidence limit (uci) in R_Ereduction is calculated as the difference in uci R_Ein the absence vs. presence of NPI. In our model, mean R_Ein the absence of NPI equals 1.05 and uci R_Ein the absence of NPI equals 8.6.

Fig. 5

Targeted testing of high transmission risk cohorts in a heterogenous population., A. Schematic of transmission risk group cohorts in the heterogenous model. The population is divided into 5000 “high transmission risk” and 15000 “low transmission risk” individuals, for which, 90% and 40% of the proportion of transmission events take place within the UC Berkeley community, respectively. Of those transmission events within the Berkeley community, the majority (80%) are restricted within the same transmission risk group as the infector, while 20% are sourced to the opposing risk group. Half of each cohort is assumed to be enrolled in asymptomatic surveillance testing and subjected to the differing test frequency regimes depicted in panels B. through D. Panel B. shows the progression of cumulative cases across 730 days of simulation for each testing regime, while panel C. and D. give, respectively, the reduction in R_E * and the total cases saved achieved by each test regime vs. a no intervention baseline., *Note: R_Ereduction (panel A) is calculated as the difference in mean R_Ein the absence vs. presence of a given NPI. The upper confidence limit (uci) in R_Ereduction is calculated as the difference in uci R_Ein the absence vs. presence of NPI. In our model, mean R_Ein the absence of NPI equals 1.05 and uci R_Ein the absence of NPI equals 8.6.