key: cord-0892448-48ypkp9t
authors: Lucas, T. C. D.; Davis, E. L.; Ayabina, D.; Borlase, A.; Crellen, T.; Pi, L.; Medley, G.; Yardley, L.; Klepac, P.; Gog, J.; Hollingsworth, T. D.
title: Engagement and adherence trade-offs for SARS-CoV-2 contact tracing
date: 2020-08-22
journal: nan
DOI: 10.1101/2020.08.20.20178558
sha: 301e93885aa7aefba77c5aff12540b45851c9c57
doc_id: 892448
cord_uid: 48ypkp9t

Contact tracing is an important tool for allowing countries to ease lockdown policies introduced to combat SARS-CoV-2. For contact tracing to be effective, those with symptoms must self-report themselves while their contacts must self-isolate when asked. However, policies such as legal enforcement of self-isolation can create trade-offs by dissuading individuals from self-reporting. We use an existing branching process model to examine which aspects of contact tracing adherence should be prioritised. We consider an inverse relationship between self-isolation adherence and self-reporting engagement, assuming that increasingly strict self-isolation policies will result in fewer individuals self-reporting to the programme. We find that policies that increase the verage duration of self-isolation, or that increase the probability that people self-isolate at all, at the expense of reduced self-reporting rate, will not decrease the risk of a large outbreak and may increase the risk, depending on the strength of the trade-off. These results suggest that policies to increase self-isolation adherence should be implemented carefully. Policies that increase self-isolation adherence at the cost of self-reporting rates should be avoided.

functional effects of different levels of compliance, it is even more difficult to quantify the strengths of the trade-offs. Legal enforcement might have a weak 68 effect on improving self-isolation adherence 22 but a strong deterrent effect 69 on self-reporting. Alternatively, perhaps legal mandation has a strong effect 70 on self-isolation adherence without being a strong deterrent to self-reporting 71 rates Furthermore, the shapes of these trade-offs are likely to differ in different 72 countries and social groups based on culture, trust in the government and 73 other factors. Careful quantitative and qualitative studies will need to be 74 conducted to quantify these effects. 75 Here we use a previously published branching process model 11,6 to exam-76 ine the effects of these trade-offs on the risk of a large outbreak of SARS-CoV-77 2. We examine trade-offs between self-isolation duration and self-isolation 78 probability with self-reporting rates, contact information reporting probabil-79 ities and sensitivity of home swab tests. It is important to note however that 80 we do not consider the societal costs 28 of legal enforcement of self-isolation; 81 we aim to quantify the benefits of these policies without considering the costs 82 noting that the costs are not easy to directly compare to the benefits. 83 

In this paper we extend a previous model of SARS-CoV-2 transmission 11 . 85 An overview of the model is given in Figure S1 while parameter values and 86 references are given in Table 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.

The copyright holder for this preprint this version posted August 22, 2020. .

Refs Self-isolation probability 30%-90% 32 Self-reporting probability 30%-90% Test sensitivity 35%-65% Figure 1 : Overview of adherence in test and trace. An untraced individual must selfreport and give the name and details of close contacts. The contact tracing team must then manage to contact the close contacts. The close contacts must self-isolate when asked and remain in self-isolation for the full isolation period (14 days in the UK). In some systems, the isolated individual is given a self-administered swab test which must be administered correctly. There is imperfect adherence or performance at each of these stages. In this paper we focus on trade-offs between self-report rate (stage 1) and self-isolation adherence (stages 4 and 5). We combine stages 2 and 3 into one parameter, which we call control effectiveness.

distribution with mean 1.43 days and sd of 0.66 39 . All individuals, whether symptomatic or asymptomatic are given a symptom onset time as the expo- 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 August 22, 2020. . https://doi.org/10.1101/2020.08.20.20178558 doi: medRxiv preprint distributions are shown in Figure S2 . First, we assume that without policies to encourage self-isolation most 155 people attempt some self-isolation but the lack of adherence is with respect 156 to the duration of self-isolation that decreases. We keep the probability 157 of self-isolation constant at 70%. We assume that each person that does 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 August 22, 2020. . between a minimum and maximum value. For the maximum values we use either the full 14 days currently recommended in the UK or a shorter seven 161 day maximum isolation. We vary the minimum duration of self-isolation 162 from 1 day to being equal to the maximum duration. 163 Second, we examine the trade-off between self-report probability and self-164 isolation probability. We expect that policies that increase self-isolation prob-165 ability will reduce self-report probability. We use values of self-isolation from 166 10% -70% in increments of 20% and examine all combinations with self-167 report probabilities from 10% -70% also in increments of 20%. The upper 168 bound for self-isolation here is certainly above the rate of self-isolation cur- Finally, we assume that policies that increase self-isolation probability will 180 decrease test sensitivity. This scenario applies to the case of home adminis-181 tered tests. With strong incentives to test negative, people will be less likely 182 to perform swabs correctly. We therefore examine a range of test sensitivities 183 from a baseline of 65% down to 35% in increments of 10%. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 22, 2020. Figure 2: Trade-off between self isolation time (columns) and self report rate (rows) with error bars denoting 95% confidence intervals. Individuals self isolate for a randomly selected duration between min isolation and 14 days. Untraced, symptomatic individuals self-report with a probability that varies across the rows. The proportion of close contacts that are divulged and asked to self-isolate varies across the x-axis of each subplot. The y-axis shows the risk of a large outbreak (greater than 2,000 cases) over 15,000 simulations. The probability that an individual self-isolates at all is fixed at 70%. If we assume we are currently near the top left we expect that introducing legal ramifications for breaking self isolation to move us down and right. This generally increases risk.

We find that increasing the duration of self-isolation increases the risk of . 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.

The copyright holder for this preprint this version posted August 22, 2020. . between 1 and 14 days), policies that move us down and right generally 206 increase the risk of a large outbreak. For example, if we consider a control 207 effectiveness of 60%, with a self-isolation duration of between 1 and 14 days 208 and a self-report rate of 70% the risk of a large outbreak is 1%. If we increase 209 the self-isolation duration to always be 14 days but reduce the self-report rate 210 to 10%, the probability of a large outbreak increases from 1% to 6%. If the 211 trade-off is very weak, such that increasing self-isolation duration to always 212 be 14 days only decreases self-report rates to 50%, we see no change in the 213 probability of an outbreak.

If we assume a more pessimistic starting scenario of a self-isolation du-215 ration of between 1 and 14 days and self-reporting rates of 10% and given a 216 control effectiveness of 60% we have a 6% risk of a large outbreak. We find 217 that increasing self-report rates gives a larger reduction in risk. Increasing 218 self-report rates from 10% to 70% reduces risk from 6% to 1%. In contrast, 219 increasing the duration of isolation to always being 14 days does not change 220 the risk of a large outbreak. We find that reducing the maximum isolation 3.2. Trade-off between self-isolation probability against self-report probability 224 We find that increasing self-isolation probability while decreasing self re-225 port probability does not strongly alter the probability of a large outbreak.

The probability of a large outbreak for all combinations of self-isolation rates 227 and self-report rates are shown in Figure 3 . If we assume that we are cur-228 rently in the top left panel (high self report rates but low self-isolation rates), 229 policies that increase self-isolation rates but decrease self-report rates would 230 move us down and right. However, whether this decreases the risk of an 231 outbreak depends on the strength of the trade-off. For example, if we con-232 sider a control effectiveness of 60%, with a self-isolation rate of 10% and a 233 self-report rate of 70% the risk of a large outbreak is 6%. If we increase the 234 self-isolation rate to 70% and equivalently reduce the self-report rate to 10%, 235 the probability of a large outbreak is still 6%. If the trade-off is weak, such 236 that increasing self-isolation from 10% to 70% only incurs a reduction in self-237 report rate to 50%, the reduction in risk of a large outbreak is substantial, 238 reducing from 6% to 1.5%. However, if the trade-off is strong, such that 239 increasing self-isolation from 10% to 30% causes a reduction in self reporting 240 rate from 70% to 10%, the risk of an outbreak instead marginally increases 241 from 6% to 7%.

. 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.

The copyright holder for this preprint this version posted August 22, 2020. Figure 3 : Trade-off between self-isolation probability (columns) and self-report probability (rows) with error bars denoting 95% confidence intervals. The y-axis shows the risk of a large outbreak (greater than 2,000 cases) over 15,000 simulations. If we assume we are currently near the top left we expect that introducing legal ramifications for breaking self isolation to move us down and right. Whether this decreases risk depends on the strength of the trade-off. If the trade-off is weak, such that as we move from the top left to isolation probability of 70% and self report probability of 50%, risk is reduced. In contrast, if increasing isolation probability from 10% to 30% incurs a drop in self reporting from 70% to 10%, risk does not change.

We could instead assume a more pessimistic starting scenario of self-243 isolation rates of 10% and self-reporting rates of 10%. Given a control ef-244 fectiveness of 60% we have a 7% risk of a large outbreak. However, from 245 this scenario we can consider whether it is better to increase self-isolation 246 or to increase self-reporting. Increasing self isolation probability to 70% re-247 duces risk to 6% and increasing self-report probability to 70% also reduces 248 risk to 6%. Increasing both to 30% reduces risk to 5%. Overall, these two 249 parameters are relatively evenly balanced.

. 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.

The copyright holder for this preprint this version posted August 22, 2020. Figure 4 : Trade-off between self isolation probability (columns) and test sensitivity (rows) with error bars denoting 95% confidence intervals. Untraced, symptomatic individuals selfreport with a probability that varies across the rows. The proportion of close contacts that are divulged and asked to self-isolate varies across the x-axis of each subplot. If we assume we are currently near the top left, introducing legal ramifications for breaking self isolation might move us down and right. This generally decreases risk unless the trade off is very strong such that a small increase in isolation probability incurs a large decrease in test sensitivity.

To model a decrease in careful administration of home swab tests, we 252 vary the test sensitivity and isolation adherence. We find that increasing is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 22, 2020. . https://doi.org/10.1101/2020.08.20.20178558 doi: medRxiv preprint we increase the self-isolation rate to 70% while reducing the test sensitivity 263 to 35%, the probability of a large outbreak reduces from 6% to 3%. 264 13 . 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.

The copyright holder for this preprint this version posted August 22, 2020. .

Overall we have found that policies that increase self-isolation rates at 266 the expense of self-report rates are unlikely to improve the effectiveness of 267 contact tracing systems. If the primary trade-off is between the duration 268 of self-isolation and the probability of self-reporting, we find that policies 269 that increase self-isolation and reduce self-report rates will cause either an Policies that improve self-report rates or self-isolation rates without an 282 associated trade-off will also improve contact tracing efficacy. For example, 283 economic support and employment protection for individuals that self-isolate 284 would be expected to improve self-isolation rates 14,18,25 without decreasing 285 self-report rates. Similarly, efforts to communicate the reasons why people 286 should self-report and self-isolate may improve both of these rates simulta-287 neously 18,25 .

One of the core assumptions to this work is that legal consequences for 289 breaking self-isolation would improve self-isolation rates. However, the evi-290 dence for this is not strong and there is evidence that feelings of shame do not 291 promote adherence 21,25 . In contrast there is good evidence that other factors 292 such as income and boredom 41 do affect self-isolation rates. How effectively 293 legal consequences for breaking self-isolation can increase self-isolation rates 294 is a complex question that will depend on cultural norms, perceived enfor-295 cability, and the strength of economic and psychological consequences for 296 self-isolation. An important consequence of this is that self-isolation rates 297 and the effectiveness of policies aimed to improve these rates will be strongly 298 correlated such that individuals that are most likely to infect each other are 299 also likely to have similar self-isolation rates. This is not included in our 300 model but has the potential to strongly reduce contact tracing efficacy in 301 14 . 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.

The copyright holder for this preprint this version posted August 22, 2020. . https://doi.org/10.1101/2020.08.20.20178558 doi: medRxiv preprint certain groups and locations.

With regards to test sensitivity, our results are relevant only to self-303 administered swab-tests. Swab-tests may be replaced with reliable paper-304 based tests. Given that we found that optimising self-isolation rates over 305 test-sensitivity minimises risk, other considerations such as test timing and 306 access are probably more important. Furthermore, currently in the UK, 307 traced contacts are not allowed out of quarantine after a negative test so the 308 system is more robust to low test sensitivity than in our simulations.

Here we have focused solely on the probability of a large outbreak as a 310 consequence of policy change. However, there are other costs and benefits to 311 changing values of self-report rates and self-isolation rates. High self-report 312 rates not only improves contact tracing efficacy directly, it also creates a more 313 effective system for measuring the incidence of SARS-CoV-2 in the commu-314 nity. This gives better early warning for when an outbreak is beginning in 315 an area or group and allows for health care resources to be deployed more 316 efficiently. In contrast, self-isolation comes with many economic and social 317 costs both for the individual and the community. Avoiding strict penalties 318 for breaking self-isolation allows those most affected by these costs to self-319 isolate less and may increase buy-in to the system as a whole. Furthermore, 320 enforcement of self-isolation policies are an infringement on a basic liberty.

While we have not tried to compare these costs to the epidemiological bene-322 fits, they must always be taken into account when implementing policy.

. 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.

The copyright holder for this preprint this version posted August 22, 2020. 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.

The copyright holder for this preprint this version posted August 22, 2020. 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.

The copyright holder for this preprint this version posted August 22, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint 

. 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.

The copyright holder for this preprint this version posted August 22, 2020. Figure S3: Trade-off between self isolation time (columns) and self report rate (rows) with error bars denoting 95% confidence intervals. Individuals self isolate for a randomly selected duration between min isolation4 and 14 days. Untraced, symptomatic individuals self-report with a probability that varies across the rows. The proportion of close contacts that are divulged and asked to self-isolate varies across the x-axis of each subplot. Self isolation probability is fixed at 70%. If we assume we are currently near the top left we expect that introducing legal ramifications for breaking self isolation to move us down and right. This generally increases risk.

. 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 August 22, 2020. Figure S4 : Trade-off between self isolation probability (columns) and self report (rows) with error bars denoting 95% confidence intervals. If we assume we are currently near the top left we expect that introducing legal ramifications for breaking self isolation to move us down and right. Whether this decreases risk depends on the strength of the trade-off.

. 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 August 22, 2020. Figure S5: Trade-off between self isolation time (columns) and self report rate (rows) with error bars denoting 95% confidence intervals. Individuals self isolate for a randomly selected duration between min isolation4 and 14 days. Untraced, symptomatic individuals self-report with a probability that varies across the rows. The proportion of close contacts that are divulged and asked to self-isolate varies across the x-axis of each subplot. Self isolation probability is fixed at 70%. If we assume we are currently near the top left we expect that introducing legal ramifications for breaking self isolation to move us down and right. This generally increases risk.

. 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 August 22, 2020. . https://doi.org/10.1101/2020.08.20.20178558 doi: medRxiv preprint

South China Morning Post, Coronavirus: China's first confirmed 360

Real-time tracking of self-reported symptoms to predict potential 487 COVID-19

Comparison of seven commercial RT-PCR diagnostic kits for COVID-491 19

Estimating the 493 asymptomatic proportion of coronavirus disease 2019 (COVID-19) cases 494 on board the Diamond Princess cruise ship

Suppression of COVID-19 outbreak in the municipality of Vo

Temporal dynamics in viral shedding 502 and transmissibility of COVID-19