When can we stop wearing masks? Agent-based modeling to identify when vaccine coverage makes nonpharmaceutical interventions for reducing SARS-CoV-2 infections redundant in indoor gatherings

As vaccination efforts to combat the COVID-19 pandemic are ramping up worldwide, there are rising concerns that individuals will begin to eschew nonpharmaceutical interventions for preventing SARS-CoV-2 transmission and attempt to return to pre-pandemic normalcy before vaccine coverage levels effectively mitigate transmission risk. In the U.S.A., some governing bodies have already weakened or repealed guidelines for nonpharmaceutical intervention use, despite a recent spike in national COVID-19 cases and majority population of unvaccinated individuals. Recent modeling suggests that repealing nonpharmaceutical intervention guidelines too early into vaccine rollouts will lead to localized increases in COVID-19 cases, but the magnitude of nonpharmaceutical intervention effects on individual-level SARS-CoV-2 infection risk in fully- and partially-vaccinated populations is unclear. We use a previously-published agent-based model to simulate SARS-CoV-2 transmission in indoor gatherings of varying durations, population densities, and vaccination coverage levels. By simulating nonpharmaceutical interventions in some gatherings but not others, we were able to quantify the difference in SARS-CoV-2 infection risk when nonpharmaceutical interventions were used, relative to scenarios with no nonpharmaceutical interventions. We found that nonpharmaceutical interventions will often reduce secondary attack rates, especially during brief interactions, and therefore there is no definitive vaccination coverage level that makes nonpharmaceutical interventions completely redundant. However, the reduction effect on absolute SARS-CoV-2 infection risk conferred by nonpharmaceutical interventions is likely proportional to COVID-19 prevalence. Therefore, if COVID-19 prevalence decreases in the future, nonpharmaceutical interventions will likely still confer protective effects but potential benefits may be small enough to remain within “effectively negligible” risk thresholds.

The magnitude of nonpharmaceutical intervention effects on individual-level SARS-51 CoV-2 infection risk in fully-and partially-vaccinated populations is unclear. This information is 52 crucial for identifying vaccination levels at which it would be appropriate to scale-back 53 guidelines for nonpharmaceutical interventions, as it would allow governing bodies to base 54 policies on concrete risk estimates. The United States Centers for Disease Control and 55 Prevention (CDC) has updated guidelines on safe gathering protocols, recommending that groups 56 of fully-vaccinated people can now safely interact amongst themselves, or with small groups of 57 unvaccinated people at low risk for developing severe COVID-19, without utilizing any 58 nonpharmaceutical Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) 59 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 through vaccination efforts alone (Gozzi et al. 2021;Moore et al. 2021). Considering that most 83 people also have yet to be fully vaccinated, guidelines that advocate phasing out 84 nonpharmaceutical interventions during interpersonal interactions may be premature at this time. 85 In Farthing & Lanzas (2021), we described an agent-based model (ABM) for simulating 86 indoor respiratory pathogen transmission. We previously used this model to quantify effects of 87 nonpharmaceutical interventions on reducing SARS-CoV-2 transmission risk during an indoor 88 superspreading event (Farthing & Lanzas 2021). Here, we use it to simulate SARS-CoV-2 89 transmission in indoor gatherings of varying durations, population densities, and proportional 90 vaccination coverage. By simulating nonpharmaceutical interventions in some gatherings but not 91 others, we were able to quantify the difference in SARS-CoV-2 infection risk when 92 nonpharmaceutical interventions were used in conjunction with vaccination efforts, relative to 93 scenarios with no nonpharmaceutical interventions. Using these data, we demonstrate how 94 interested parties can easily estimate the potential reduction in SARS-CoV-2 infection risk 95 attributable to nonpharmaceutical interventions, and try to answer the question: "at what point 96 during vaccine rollout are gatherings without non-pharmaceutical measures safe?" 97 98 Methods 99 We used the ABM we first described in Farthing & Lanzas (2021) to simulate the effect 100 of increasing vaccination coverage and nonpharmaceutical interventions on SARS-CoV-2 101 transmission risk during indoor gatherings. The simulation input levels and parameter values we 102 used are given in Table 1. We made the assumptions that any infectious individuals at gatherings 103 would be asymptomatic because symptomatic people would consciously decide to stay away, 104 and that no one with partial immunity exists within the group of attendees. Vaccinated people 105 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 had a fixed probability of becoming completely immune to SARS-CoV-2 infection (Table 1), 106 and those that did not become immune remained susceptible to infection (i.e, 'all-or-nothing' 107 vaccine). Finally, we only simulated use of cloth face coverings, rather than notably more-108 effective masks like N95s, because we make the assumption that the majority of Americans have 109 ready access to, and are more-likely to use cloth masks. 110 All simulations were carried out within the open-source modeling software, NetLogo 111 (Ver. 6. 1. 1 -Wilensky 1999). We executed a factorial simulation run in the NetLogo 112 BehaviorSpace using our specified input levels, and ran 200 simulations replicates of each 113 parameter set combination when the nonpharmaceutical interventions were included and when 114 they were not. We ran these factorial combination sets separately in order to save computation 115 time as there were two inputs (i.e., mask efficacy, attempted social distance) that only changed 116 when nonpharmaceutical interventions were simulated. We ultimately produced 1,612,800 117 simulations without nonpharmaceutical interventions, and 9,676,800 including them (i.e., 118 11,289,600 total simulations). We recorded the number of susceptible individuals infected in 119 each simulation, and aggregated this information into a single data set prior to analysis. 120 We reported the mean probability of observing ≥ 1 successful infection event(s) and 121 mean secondary attack rates in indoor gatherings when an asymptomatic person was also in 122 attendance across factorial combinations of "between-group comparison" variables (Table 1). 123 Secondary attack rates here were calculated by dividing the number of people that were infected 124 at the gathering by the number of "healthy" people at the start of the gathering, and can also be 125 considered to be the individual-level probability of a previously healthy attendee being infected 126 at the gathering. To assess the difference between protection conferred by the simultaneous 127 deployment of pharmaceutical and nonpharmaceutical interventions, versus use of only 128 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 nonpharmaceutical interventions, we first smoothed the observed mean secondary attack rates 129 ( ) by fitting them to a beta regression model with a fixed unknown precision parameter, using 130 a logit link function to map (0,1) values (Ferrari & Cribari-Neto 2004). The specific model is 131

134
( 1 )  135 where "Intervention level" is a categorical variable containing the following mutually-exclusive 136 levels: "cloth face masks & vaccination," "cloth face masks & 2-m social distancing & 137 vaccination," and "vaccination only." Additionally, "Vaccine efficacy" here refers to the ability 138 of vaccines to induce complete immunity to infection. "Vaccine coverage" and "Vaccine 139 efficacy" are given in terms of decimal percent, not percentage points (e.g., 0.1, not 10%). 140 Because beta regression models assume all dependent variable values fall between 0 and 1, we 141 used the data transformation procedure described by (Cribari-Neto & Zeiles, 2010) to reconstruct 142 our proportion data without these extremities prior to model fitting. We used the pseudo-R 2 143 calculation procedure given by Ferrari & Cribari-Neto (2004) to assess the goodness of fit for 144 our regression model. 145 After fitting our data, we used the regression model to predict the mean secondary attack 146 rates during a 60-minute gathering with a single asymptomatic person in attendance across the 147 complete factorial combination of covariate inputs described in Table 2. We report the difference 148 between predicted values when all interventions (i.e., cloth face masks & 2-m social distancing 149 & vaccination) are utilized, and predicted values assuming vaccinations are the only 150 interventions. All analyses and plotting were carried out using functions from the "betareg" 151 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) We found that the probability of ≥ 1 successful transmission event generally increased 157 with population density (Fig. 1). This is unsurprising, as SARS-CoV-2 transmission in this ABM 158 is highly sensitive to within-room population density (Farthing & Lanzas 2021). We observed 159 that at low population densities and/or short-duration gatherings, the use of nonpharmaceutical 160 interventions can significantly reduce the probability of successful transmission. Furthermore, it 161 is clear that at low population densities, 2-m social distancing confers additional protective 162 effects when used in conjunction with cloth face coverings, even during relatively-long duration 163 gatherings. This is consistent with what we observed when we used the same ABM to directly 164 compare the effectiveness of varied nonpharmaceutical interventions to prevent SARS-CoV-2 165 transmission during a superspreading event (Farthing & Lanzas 2021). We found that cloth face 166 masks alone conferred few protective effects in long-duration gatherings. 167 The probability of transmission events occurring was unlikely to reach ≈ 0% outside of 168 scenarios with low population density and multiple nonpharmaceutical interventions, or ≥ 95% 169 vaccine coverage and vaccines that were 100% effective at preventing infections. Given that 1) 170 current estimates place SARS-CoV-2 vaccine efficacies against infection between 60-90% (Hall 171 et al. 2021;Lipsitch & Kahn 2021;Yellen et al. 2021), 2) historical precedence suggesting adult 172 populations will fall well short of these high vaccination levels (Applewhite et al. 2020;CDC 173 2020), and 3) the difficulty government institutions have had enforcing nonpharmaceutical 174 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 intervention policies (Jacobs & Ohinmaa 2020; Pedersen & Favero 2020), it is unlikely that these 175 scenarios will be representative of average real-world gatherings. Moreover, in 60-min gathering 176 scenarios, the probability of ≥ 1 successful transmission event occurring is relatively high even 177 when gathering attendees utilize nonpharmaceutical interventions and most are vaccinated. 178 The probability that ≥1 SARS-CoV-2-positive individual is in attendance at a gathering 179 can be calculated as 180 where is the local COVID-19 prevalence, and is the number of people at the gathering 182 (Chande et al. 2020). The prevalence of infectious cases (p) can be highly uncertain because of 183 the variable testing effort across time and space, but it can be estimated by assuming that any 184 SARS-CoV-2-positive individuals are infectious at time of testing and will remain infectious for 185 a given period of time. Additionally, ascertainment bias can be factored in. The probability that a 186 given individual will be infected at a gathering is then 187 where " is the probability that individual will be infected given exposure to an asymptomatic 189 individual at the gathering. Effectively, what we report in Fig. 2 are estimates of " under 190 different circumstances. Our findings suggest that cloth-based mask use, with or without 2-m 191 social distancing, often does not confer significant protective effects during long-duration 192 gatherings ( Fig. 2), we have also shown that implementing these nonpharmaceutical 193 interventions can reduce overall transmission probability ( Fig. 1) and secondary attack rates (Fig. 194 2, Table 3) during brief interactions or gatherings with relatively-few people (e.g., fewer than 10 195 people, the limit for indoor and/or outdoor social gatherings enforced by some U.S. states 196 (MultiState 2021)). This effectively means that strict guidelines for continued nonpharmaceutical 197 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

( 5 ) 214
(Ferrari & Cribari-Neto 2004). Our regression model had a pseudo-R 2 of 0.37. Given the number 215 of stochastic processes in our ABM and the variability purposely introduced into simulations 216 (Table 1), we believe the explanatory power of the model is acceptable for our purposes here. 217 Assuming mean population-level vaccine efficacies of 60% and 80%, which we believe are 218 conservative estimates for U.S.-approved vaccine efficacies, our regression model consistently 219 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ; https://doi.org/10. 1101/2021 predicts that secondary attack rates decrease by 55-58% when attendees utilize cloth masks and 220 2-m social distancing, regardless of gathering duration (Fig. 3). However, it is important to 221 reiterate that here we estimate the probability or infection given contact with an infectious 222 individual at a gathering ( " ) and comment on the relative risk difference attributable to 223 intervention use. This should not be confused with the absolute risk of becoming infected at a 224 gathering (see Equation 3). We demonstrate the difference in Figure 4, which is a simplistic 225 example intended to show that even at relatively high COVID-19 prevalence levels, 20 people 226 gathering indoors for 60 minutes have a substantially-lower individual-level risk of SARS-CoV-227 2 infections than is suggested by " alone. Though predicting intervention effects on community-228 level COVID-19 prevalence and infection-related events (e.g., symptom-onset, mortality, or 229 hospitalization) is outside the scope of our model, our simulations do suggest that secondary 230 attack rates are negatively correlated with vaccine coverage. Given that we expect local COVID-231 19 prevalence to eventually follow similar trends (Gozzi et al. 2021), the relative impact of 232 nonpharmaceutical interventions on infection risk reduction will likely decrease over time as 233 vaccine rollouts continue. 234 In addition to being unable to comment on community-level infection metrics, there are a 235 few other limitations associated with our results that we must acknowledge. Aside from the 236 ABM design limitations outlined in Farthing et al. (2021), we make a number of assumptions in 237 our simulations. Most of these assumptions are directly tied to our parameter space detailed in 238 Table 1, and include such things as: in simulated gatherings only one asymptomatic individual 239 was in attendance, no individuals wear masks with exposure-reduction efficacies > 50% and 240 therefore we are not simulating the use of N95 or similar masks, and there is no simulated 241 forced-air ventilation or infectious individuals that produce superspreader-level of contaminated 242 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. within a relatively large room. In short, our results must be viewed through the lens of simulated 250 world parameters and behaviors, and likely will not wholly reflect all variability that may exist in 251 real-world transmission events. This is very common for ABM-based studies however, and we 252 feel that our model is sufficiently accurate to highlight general trends in indoor SARS-CoV-2 253 transmission and infection risk. 254 255

Conclusions 256
We found that nonpharmaceutical interventions will often reduce secondary attack rates, 257 especially during brief interactions, and therefore there is no definitive vaccination coverage 258 level that makes nonpharmaceutical interventions completely redundant. However, the beneficial 259 effect on absolute SARS-CoV-2 infection risk reduction conferred by nonpharmaceutical 260 interventions used during indoor gatherings is likely proportional to COVID-19 prevalence. 261 Therefore, if U.S. COVID-19 prevalence decreases in the future, nonpharmaceutical 262 interventions will likely still confer protective effects, but any potential benefits may be small 263 enough to remain within "effectively negligible" risk thresholds. 264 265 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Scenario environment and individual behavior inputs
Area ( Fixed value 2 -

Number of asymptomatic infectious individuals (people)
Fixed value 1 - . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ; Table 1. Model parameter and scenario-specific input descriptions for transmission simulations. 398 * All simulated worlds were square-shaped. ¶ The Purpose column describes why the parameter or 399 input was included as it relates to analyses. Specifically, "Fixed value" indicates that values are 400 unchanged across all simulations, and are thus irrelevant for analyses. "Between-group 401 comparison" indicates that levels were used in factorial combinations for data aggregation and 402 reporting. "Within-group variation" indicates that different levels were included to increase the 403 variation in simulation results, and by doing so increase model realism. m/s to be 0.55 m. They also found that the majority of 100-μm droplets will fall 0.55-2.35 m 409 away from the expelling individual, depending on initial velocity, but droplets may settle up to 410 3.2 m away very rarely. A random draw of 10,000,000 samples from a log-normal distribution 411 parameterized using 1.7-m and 0.2095-m droplet spread distance mean and standard deviation 412 values, respectively, generated a distribution in line with this finding. The standard deviation we 413 use in simulations for non-coughing expectoration is proportionate to the one used in this random 414 draw. ** Instead of specifying a fixed number of individuals in simulations, we scaled the 415 simulated population with world size. 416

Vaccine coverage (%)
Between-group comparison 0:100 by 5 Population density (people/m 2 ) ** Between-group comparison 0.17, 0.33, 0.67, 1 -. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 gatherings of varied sizes. Estimates were obtained by plugging Figure 3

predictions into 440
Equation 3 with fixed COVID-19 prevalence and n values. a) Absolute risk of SARS-CoV-2 441 transmission given that 10 people attend the gathering. b) Absolute risk of SARS-CoV-2 442 transmission given that 20 people attend the gathering. 443 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 27, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021