Impact of self-imposed prevention measures and short-term government intervention on mitigating and delaying a COVID-19 epidemic

Background: With new cases of COVID-19 surging around the world, many countries have to prepare for moving beyond the containment phase. Prediction of the effectiveness of non-case-based interventions for mitigating, delaying or preventing the epidemic is urgent, especially for countries affected by the ongoing seasonal influenza activity. Methods: We developed a transmission model to evaluate the impact of self-imposed prevention measures (handwashing, mask-wearing, and social distancing) due to the spread of COVID-19 awareness and of short-term government-imposed social distancing on the peak number of diagnoses, attack rate and time until the peak number of diagnoses. Findings: For fast awareness spread in the population, self-imposed measures can significantly reduce the attack rate, diminish and postpone the peak number of diagnoses. A large epidemic can be prevented if the efficacy of these measures exceeds 50%. For slow awareness spread, self-imposed measures reduce the peak number of diagnoses and attack rate but do not affect the timing of the peak. Early implementation of short-term government interventions can only delay the peak (by at most 7 months for a 3-month intervention). Interpretation: Handwashing, mask-wearing and social distancing as a reaction to information dissemination about COVID-19 can be effective strategies to mitigate and delay the epidemic. We stress the importance of rapidly spreading awareness on the use of these self-imposed prevention measures in the population. Early-initiated short-term government-imposed social distancing can buy time for healthcare systems to prepare for an increasing COVID-19 burden. Keywords: SARS-CoV-2, COVID-19, mathematical model, prevention measures, mitigation, epidemic control, disease awareness, social distancing, handwashing, mask-wearing

Introduction attempt to stop its spread. 22 Moreover, if a COVID-19 epidemic cannot be prevented, it is important to know 88 how the epidemic peak can be diminished and postponed to give healthcare professionals more time to prepare 89 and react effectively to the increasing health care burden. For affected areas like Europe, where the outbreak runs 90 concurrently with the influenza season, the importance to identify such interventions is profound.

92
Using a transmission model we evaluate the impact of self-imposed measures (handwashing, mask-wearing, and 93 social distancing) due to awareness about COVID-19 and of a short-term government-imposed social distancing 94 intervention on the peak number of diagnoses, attack rate, and time until the peak number of diagnoses since 95 the first case. We provide a head-to-head comparison of these interventions and assess for which efficacy of these 96 interventions a large COVID-19 outbreak can be prevented. 5 . CC-BY-NC-ND 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)  . Individuals leave the E compartment at rate α. A proportion p of the latently infected individuals (E) will go to the I M compartment, and the proportion (1 − p) of E individuals will go to the I S compartment. Infectious individuals with mild symptoms (I M ) recover undiagnosed (R M ) at rate γ M . Individuals with severe symptoms (I S ) are diagnosed and kept in isolation (I D ) at rate ν until they recover (R S ) at rate γ S or die at rate η. Individuals who are diagnosed (I D ) will be isolated and individuals recovered from a severe infection (R S ) know 125 that they cannot contract the disease again. Hence we assume their behaviour in the contact process is identical 126 to disease-unaware individuals. is slower for infectious individuals with severe symptoms (I S ) than for the remaining disease-aware population.

135
Estimates of epidemiological parameters were obtained from most recent literature (Table 1). We used contact 136 rates for the Netherlands, but the model is appropriate for other Western countries with similar contact rates. A 137 6 . CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint detailed mathematical description of the model with and without awareness can be found in the Appendix. Infectious individuals with severe symptoms who are disease-aware (I a S ) get diagnosed and are kept in isolation (I a D ) at rate ν a , recover at rate γ a S and die from disease at rate η a . (B) shows awareness dynamics. Infectious individuals with severe symptoms (I S ) acquire disease-awareness (I a S ) at rate λ aware proportional to the rate of awareness spread and to the current number of diagnosed individuals (I D and I a D ) in the population. As awareness fades, these individuals return to the unaware state at rate µ S . The acquisition rate of awareness (kλ aware ) and the rate of awareness fading (µ) rates are the same for individuals of type S, E, I M , and R M , where k is the reduction in susceptibility to the awareness acquisition compared to I S individuals. Table 1 provides the description and values of all parameters.

139
We considered short-term government intervention aimed at fostering social distancing in the population and a 140 suite of measures self-imposed by disease-aware individuals, i.e., mask-wearing, hand washing, and self-imposed 141 social distancing. Since infectious individuals may transmit the virus to others without direct physical contact, we assume that hand-150 washing only reduces one's susceptibility. The efficacy of handwashing is described by the reduction in susceptibility 151 (i.e., probability of transmission per single contact) of susceptible disease-aware individuals (S a ) which ranges from 152 0% (zero efficacy) to 100% (full efficacy).

153
Self-imposed social distancing 154 Disease-aware individuals may also practice social distancing, i.e., maintaining distance to others and avoid con-155 gregate settings. 27 As a consequence, this measure leads to a change in mixing patterns in the population. The 156 efficacy of social distancing of disease-aware individuals is described by the reduction in their contact rate which is 157 varied from 0% (no social distancing or zero efficacy) to 100% (full self-isolation or full efficacy).

158
8 . CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint Governments may decide to promote social distancing policies through interventions such as school and workplace 160 closures or by issuing a ban on large gatherings. Such policies will cause a community-wide contact rate reduction, 161 regardless of the awareness status. Here, government intervention is initiated if the number of diagnosed individuals 162 exceeds a certain threshold (10-1000 persons) and terminates after a fixed period of time (1-3 months). As such, 163 we assume that the intervention is implemented early in the epidemic. The efficacy of government-imposed social 164 distancing is described by the reduction of the average contact rate in the population which ranges from 0% (no 165 distancing) to 100% (complete quarantine of the population). (government-imposed social distancing). We refer to these quantities as the efficacy of a prevention measure and 174 vary it from 0% (zero efficacy) to 100% (full efficacy) ( Table 1). The main analyses were performed for two values of 175 the rate of awareness spread that corresponded to scenarios of slow and fast spread of awareness in the population 176 (Table 1). For these scenarios, the proportion of the aware population at the peak of the epidemic was 40% and 177 90%, respectively. In the main analyses, government-imposed social distancing was initiated when 10 individuals 178 got diagnosed and was lifted after 3 months. Sensitivity analyses for parameters indicated in Table 1   The funders of the study had no role in study design, data collection, data analysis, data interpretation, writing of 183 the manuscript, or the decision to submit for publication. All authors had full access to all the data in the study 184 and were responsible for the decision to submit the manuscript for publication.

186
Our analyses show that disease awareness has a significant effect on the model predictions. We first consider the 187 epidemic dynamics in a disease-aware population where handwashing is promoted, as an example of self-imposed 188 9 . CC-BY-NC-ND 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 March 16, 2020.  be significant, namely we predict a 65% reduction in the peak number of diagnoses, a 29% decrease in the attack 203 rate, and a delay in peak timing of 2.7 months (Figure 3 A and B).
CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint The effect of awareness on the disease dynamics can also be observed in the probability of infection during the 206 course of the epidemic. In the model with awareness and no measures, the probability of infection is reduced by 207 4% for all individuals. Handwashing with an efficacy of 30% reduces the respective probability by 14% for unaware 208 individuals and by 29% for aware individuals. Note that the probability of infection is highly dependent on the 209 type of prevention measure. The detailed analysis is given in the Appendix. show the relative reduction in the peak number of diagnoses, the attack rate (proportion of the population that recovered or died after severe infection) and the time until the peak number of diagnoses. The efficacy of prevention measures was varied between 0% and 100%. In the context of this study, the efficacy of social distancing denotes the reduction in the contact rate. The efficacy of handwashing and mask-wearing are given by the reduction in susceptibility and infectivity, respectively. The simulations were started with one case. Government-imposed social distancing was initiated after 10 diagnoses and lifted after 3 months. For parameter values, see Table 1. 11 . CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint Figure 5. Impact of prevention measures on the epidemic for a fast rate of awareness spread. Same description as in Figure 4 but for a fast rate of awareness spread. For parameter values, see Table 1.
A comparison of prevention measures 211 Figure 4 shows the impact of all considered self-imposed measures as well as of the government-imposed social 212 distancing on the peak number of diagnoses, attack rate, and the time to the peak for slow rate of awareness 213 spread. In this scenario, the model predicts progressively larger reductions in the peak number of diagnoses and in 214 the attack rate as the efficacy of the self-imposed measures increases. In the limit of 100% efficacy, the reduction 215 in the peak number of diagnoses is 23% to 30% (Figure 4 A) and the attack rate decreases from 16% to 12-13% 216 (Figure 4 B). The efficacy of the self-imposed measures has very little impact on the peak timing when compared 217 to the baseline, i.e., no awareness in the population (Figure 4 C). Since the proportion of aware individuals who 218 change their behavior is too small to make a significant impact on transmission, self-imposed measures can only 219 12 . CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint mitigate but not prevent an epidemic. When awareness spreads at a slow rate, a 3-month government intervention 220 has a contrasting impact. The time to the peak number of diagnoses is longer for more stringent contact rate 221 reductions. For example, at 100% efficacy (full quarantine) the government can postpone the peak by almost 7 222 months but its magnitude and attack rate are unaffected. Similar predictions are expected, as long as government-223 imposed social distancing starts early (e.g, after tens to hundreds cases) and is lifted few weeks to few months 224 later (Appendix). This type of intervention halts the epidemic for the duration of intervention, but, because of a 225 large pool of susceptible individuals, epidemic resurgence is expected as soon as social distancing measures are lifted. 226

227
Since the government intervention reduces the contact rate of all individuals irrespective of their awareness status, 228 it has a comparable impact on transmission for scenarios with fast and slow rate of awareness spread (compare 229 Figure 4 and Figure 5). However, the impact of self-imposed measures is drastically different. When awareness 230 spreads fast, all self-imposed measures are more effective than short-term government intervention. These measures 231 not only reduce the attack rate (Figure 5 B), diminish and postpone the peak number of diagnoses ( Figure 5 A 232 and C), but they can also prevent a large epidemic altogether when their efficacy is sufficiently high (about 50%). 233 Note that when the rate of awareness is fast, as the number of diagnoses grows, the population becomes almost 234 homogeneous, with most individuals being disease-aware. It can be shown that in such populations prevention 235 measures yield comparable results if they have the same efficacy.

237
For many countries around the world, the focus of public health officers on the COVID-19 epidemic has shifted 238 from containment to mitigation and delay. Our study provides new evidence for designing effective outbreak control 239 strategies. We show that hand-washing, mask-wearing, and social distancing adopted by disease-aware individuals 240 are all viable strategies for delaying the epidemic peak, flattening the epidemic curve and reducing the attack rate. 241 We show that the rate at which disease awareness spreads has a strong impact on how self-imposed measures affect 242 the epidemic. For a slow rate of awareness spread, self-imposed measures have little impact on transmission, as not 243 many individuals adopt them. However, for a fast rate of awareness spread, their impact on the magnitude and 244 timing of the peak increases with increasing efficacy of the respective measure. For all measures, a large epidemic 245 can be prevented when the efficacy exceeds 50%. In practical terms, it means that SARS-CoV-2 will not cause a 246 large outbreak in a country where 90% of the population adopt handwashing that is 50% efficacious (i.e., reduces 247 susceptibility by 50%). 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint soap or using alcohol-based sanitizers may remove the virus completely leading to 100% efficacy. 28, 15 For surgical 252 masks, their filtration efficiency has a wide range (0%-84%) and thus their actual efficacy is difficult to quantify. 17 253 For this reason, the promotion of handwashing might become preferable. Thus, for a fair comparison between 254 measures, realistic efficacy values of a specific measure should be taken into consideration.

256
We contrast self-imposed measures stimulated by disease awareness with mandated social distancing. Our analyses 257 show that short-term government-imposed social distancing that is implemented early into the epidemic, can delay 258 the epidemic peak but does not affect its magnitude nor the attack rate. For example, a 3-month government 259 intervention imposing community-wide contact rate reduction that starts after tens to hundreds diagnoses in the 260 country can postpone the peak by about 7 months. Such an intervention is desirable, when a vaccine is being 261 developed or when healthcare systems require more time to treat cases or increase capacity.

262
Since the COVID-19 epidemic is still in its early stages, government-imposed social distancing was modeled   We show that spreading disease awareness such that highly efficacious preventive measures are quickly adopted by  Our model has several limitations. It does not account for stochasticity, demographics, heterogeneities in contact 281 patterns, spatial effects, inhomogeneous mixing and imperfect isolation. Our conclusions can, therefore, be 282 drawn on a qualitative level. Detailed models will have to be developed to design and tailor effective strategies 283 in particular settings. To take into account the uncertainty in SARS-CoV-19 epidemiological parameters, we 284 performed sensitivity analyses to test the robustness of the model predictions. As more data become available, 285 14 . CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint our model can be easily updated. In addition, our study assumes that individuals become disease-aware with a 286 rate of awareness acquisition proportional to the number of currently diagnosed individuals. Other forms for the 287 awareness acquisition rate that incorporate, e.g., the saturation of awareness, may be more realistic and would be 288 interesting to explore in future studies.

290
In conclusion, we provide the first empirical basis of how stimulating the uptake of effective prevention measures, 291 such as handwashing, can be pivotal to achieve control over a COVID-19 epidemic. While information on the rising 292 number of COVID-19 diagnoses reported by the media may fuel anxiety in the population, wide and intensive 293 promotion of self-imposed measures with proven efficacy by governments or public health institutions may be a key 294 ingredient to tackle COVID-19.  Declaration of interests 302 We declare that we have no conflicts of interest.

303
. CC-BY-NC-ND 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 March 16, 2020.

18
. CC-BY-NC-ND 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 March 16, 2020. . https://doi.org/10.1101/2020.03.12.20034827 doi: medRxiv preprint