Examining SARS-CoV-2 Interventions in Residential Colleges Using an Empirical Network

Abstract

Universities have turned to SARS-CoV-2 models to examine campus reopening strategies^1-9. While these studies have explored a variety of modeling techniques, all have relied on simulated data. Here, we use an empirical proximity network of college freshmen¹⁰, ascertained using smartphone Bluetooth, to simulate the spread of the virus. We investigate the role of testing, isolation, mask wearing, and social distancing in the presence of implementation challenges and imperfect compliance. Here we show that while frequent testing can drastically reduce spread if mask wearing and social distancing are not widely adopted, testing has limited impact if they are ubiquitous. Furthermore, even moderate levels of immunity can significantly reduce new infections, especially when combined with other interventions. Our findings suggest that while testing and isolation are powerful tools, they have limited benefit if other interventions are widely adopted. If universities can attain high levels of masking and social distancing, they may be able to relax testing frequency to once every two to four weeks.

Keywords: Bluetooth; COVID-19; Copenhagen Network Study; Proximity network; Repeat testing; SARS-CoV-2.

Figure 1:

Number of proximity events per user per day for 675 students in the Copenhagen Network Study proximity data. Only close proximity Bluetooth connections (RSSI ≥ −75, corresponding to physical proximity of approximately one meter) with participating devices are shown, corresponding to 8.0% of all detected proximity events. Number of active users indicates users with at least one Bluetooth ping with a fellow participants. All other users only had empty scans or pings with non-participating devices.

Figure 2:. Modified SEIR model for the spread of SARS-CoV-2.

β is a pair-specific transition probability per 5-minute close proximity interaction. All other transition probabilities are constant at each time step and do not incorporate the underlying contact network. For specific parameter values, see Table 1.

Figure 3:

SARS-CoV-2 test sensitivity based on nasopharyngeal swab data adapted from Wikramaratna et al. Test sensitivity outside the range shown here was assumed to be zero.

Figure 4:. Proximity networks by day for 675 students.

Each node represents a study participant and each edge represents the presence of one or more Bluetooth pings on that day. Only close proximity Bluetooth connections (RSSI ≥ −75, corresponding to physical proximity of approximately one meter) with participating devices are shown. For each day, the largest connected component is shown at left, with the remaining connected components displayed at right ordered from largest to smallest.

Figure 5:. Number of students infected over the course of a simulated 16 week semester.

Rows show different epidemic parameters with R₀ ≈ 1.5, R₀ ≈ 3, and R₀ ≈ 4.5, respectively. Columns show scenarios where scheduled testing was done every 3, 7, 14, and 28 days, respectively, and where no testing was done. Grey lines show individual simulations, while blue lines indicate the point-wise average trajectory over all 100 replicates. Vertical red lines and text indicate the average time to reach 20% of the population infected, computed by identifying the time to 20% infected for each realization and averaging those times.

Figure 6:. Number of students who tested positive over the course of a simulated 16 week semester.

Rows show transmission probabilities of β = 0.003078 (R₀ ≈ 1.5), β = 0.006157 (R₀ ≈ 3), and β = 0.009235 (R₀ ≈ 4.5), respectively. Columns show scenarios where scheduled testing was done every 3, 7, 14, and 28 days, respectively. Grey lines show individual simulations, while blue lines indicate the average trajectory over all 100 replicates. Vertical red lines and text indicate the average time to reach 20% infected.

Figure 7:. Number of students isolated over the course of a simulated 16 week semester.

Rows show transmission probabilities of β = 0.003078 (R₀ ≈ β 1.5), β = 0.006157 (R₀ ≈ 3), and β = 0.009235 (R₀ ≈ 4.5), respectively. Columns show scenarios where scheduled testing was done every 3, 7, 14, and 28 days, respectively. Grey lines show individual simulations, while blue lines indicate the average trajectory over all 100 replicates. Vertical red lines and text indicate the average time to reach 20% infected.

Figure 8:. Relative cumulative incidence for various proportions of the population social distancing and/or wearing masks.

Each cell displays the cumulative incidence with testing divided by the cumulative incidence without resting under comparable levels of social distancing and mask wearing. Rows one and two show the setting where precautions were randomly assigned to participants and where mask wearing was clustered on the network, respectively. Rows three and four show random assignment of mask wearing and social distancing, but with 20% and 40% of the population immune to SARS-CoV-2 at the outset, respectively. Columns show scenarios where testing was done every 3, 7, 14, and 28 days. In each panel, we consider the moderate transmission scenario with R₀ ≈ 3. A value of 1 indicates that testing offered no reduction in cumulative incidence while a value of 0.5 indicates that testing reduced the cumulative incidence by 50% relative to no testing. All comparisons are for comparable levels of mask wearing and social distancing.

Figure 9:. Relative cumulative incidence for various proportions of the population social distancing and/or wearing face coverings with reduced efficacy of masking and distancing.

Values are relative to a scenario with comparable proportions of social distancing and mask wearing, but with no testing. Columns show scenarios where testing was done every 3, 7, 14, and 28 days, respectively, with R₀ ≈ 3.