Assessment of Simulated Surveillance Testing and Quarantine in a SARS-CoV-2–Vaccinated Population of Students on a University Campus

Key Points Question Are surveillance testing and quarantine still important for limiting virus transmission on college campuses where the student population is completely vaccinated against SARS-CoV-2? Findings In this decision analytical model study of 5000 simulated undergraduates, if 100% were vaccinated with 90% vaccine effectiveness, surveillance testing and quarantine were not associated with a substantial reduction in infections. However, if vaccine effectiveness was reduced to 75%, weekly surveillance testing was associated with a substantial reduction in the number of infections; at 50% vaccine effectiveness, surveillance testing and quarantine were associated with a marked reduction in the estimated number of infections. Meaning The study results suggest that surveillance testing and isolation of positive cases may remain important mitigation strategies on university campuses with a vaccinated student body with only modest erosion of vaccine effectiveness, while quarantining contacts offers limited added benefit over surveillance testing alone and may be effectively replaced by an increased testing cadence of reported contacts.

eFigure 1. Estimated daily infection prevalences, with and without surveillance testing (pessimistic initial infections and exposures) eFigure 2. Estimated daily infection prevalences, quarantining vs. testing contacts (pessimistic initial infections and exposures) eFigure 3. Overview of model architecture eTable 1. Estimated cumulative infection prevalences across conditions and mitigations (pessimistic initial infections and exposures) eTable 2. Estimated maximum fraction of isolated and quarantined individuals (pessimistic initial infections and exposures) eTable 3. Model parameters This supplemental material has been provided by the authors to give readers additional information about their work. eFigure 1. Estimated daily infection prevalence, with and without surveillance testing. The medians (dark lines) and 2.75%-97.5% quantile ranges (shaded bands) of the daily infection prevalence in the modelled population of 5000 individuals over 100 simulations assuming an interaction multiplier of 10. The outside community infection prevalence was fixed at 0.1% (top) or 1.0% (bottom), with vaccine effectiveness chosen to be 90% (left), 75% (middle) or 50% (right). Simulations were initialized with 0.5% (25) expected initial exposures and 0.5% (25) expected initial infections. All other model parameters were identical across simulations. eFigure 3. Overview of model architecture. (Top) Flowchart describing each of the four mitigation strategies considered in this study: 1) no testing, 2) surveillance testing only, 3) surveillance with quarantining of contacts, and 4) surveillance with follow-up testing of contacts every 2 days. Dashed arrows indicate optional pathways if the specified mitigation is being modelled. Individuals in red boxes may have interactions with others in the modelled population, while individuals with blue attributes are isolated from the modelled population and outside community. (Bottom) Flowchart of SEIR model and agent health state progression. Arrows indicate transitions between states and daily update rules.

Population Parameters Mitigation Strategies, % (IQR)
Vaccine Effectiveness 14.5% ± 1.0% 28.2% ± 2.6% 10.2% ± 1.0% eTable 2. Estimated maximum fraction of isolated and quarantined individuals. The medians and interquartile ranges of the daily maximum fraction of agents in isolation or quarantine in a modelled population of 5000 individuals for 100 simulations of each choice of population parameters and mitigation strategies. For mitigation strategies involving contact tracing, the contact tracing efficacy was fixed at 15%. Simulations were initialized with 0.5% (25) expected initial exposures and 0.5% (25) expected initial infections.

Variable Description Value Disease Progression Duration of exposed
The number of days agents stay in the exposed state.

(*) Duration of infected
The number of days agents stay in the infected state. (**)

Size
The number of individuals in the modelled population. 5000 Initial exposed The average fraction of individuals initially in the exposed state. 0.1% / 0.5% Initial infected The average fraction of individuals initially in the infected state. 0.1% / 0.5%

Baseline daily interactions
The baseline number of unique daily interactions between members of the modelled population on average.

( †)
Baseline interaction exposure probability The baseline probability that a susceptible-infected interaction will result in an exposure.

Baseline outside daily interactions
The baseline number of unique daily interactions individuals will have with the outside community on average.

( †)
Baseline outside interaction exposure probability The baseline probability that an interaction between a susceptible member of the modelled population and the outside community will result in an exposure.

Outside prevalence
The current fraction that is infected in the outside community. 0.1% / 1.0% Interaction multiplier Factor by which the baseline number of interactions is multiplied. 1 / 10 / 20 Vaccine effectiveness Percent reduction in the baseline interaction exposure probability. 90% / 75% / 50%

Surveillance and Individual Testing
Days between surveillance tests The number of days between tests of individual agents, and the number of days required to test the entire population exactly once. The number of days a positive case spends in isolation. 10

Contact Tracing Parameters
Contacts reported The number of contacts an agent will report if they test positive (***) Tracing effectiveness The maximum fraction of reported contacts who are exposed or infected at the time of the contact trace.

15%
Quarantine duration The number of days a reported contact of a positive test spends in quarantine (or undergoes additional testing) 10 eTable 3. Model parameters. The set of model parameters used in this study. Disease progression parameters and reported contacts determined by parameterized distributions: (*) lognormal(mean=4.6, stdev=4.8, minimum=3) or lognormal(mean=4.5, stdev=1.5, minimum=1), (**) 1.1% drawn from lognormal(mean=14.0, stdev=2.4, minimum=1) and 98.9% drawn from lognormal(mean=8.0, stdev=2.0, minimum=1) or lognormal(mean=8.0, stdev=2.0, minimum=1), and (***) the empirical distribution from the Duke University 2020-21 academic year with an average of 2.54 contacts per positive test. The function lognormal(mean, stdev, minimum) returns a sample of a log normal distribution, with specified mean and standard deviation, and with values rounded to the nearest integer, with values less than the specified minimum replaced by that minimum. ( †) The parameters determining daily interactions and outside interactions were found, among 500 samples, to minimize the total squared error between the model estimated daily positive cases (averaged over 100 simulations) and the measured positive cases at Duke University during the Spring 2021 semester. The 500 samples were drawn from the two-dimensional parameter space [0,20]x[0,20] using Bayesian optimization starting with the uniform prior.