key: cord-0258863-0gn3b98n
authors: van Zandvoort, K.; Favas, C.; Checchi, F.
title: Shielding individuals at high risk of COVID-19: a micro-simulation study
date: 2022-01-03
journal: nan
DOI: 10.1101/2022.01.03.22268675
sha: bbce9bd1e76b136f1a779ecbc85cc503e69497e5
doc_id: 258863
cord_uid: 0gn3b98n

Background One of the proposed interventions for mitigating COVID-19 epidemics, particularly in low-income and crisis-affected settings, is to physically isolate individuals known to be at high risk of severe disease and death due to age or co-morbidities. This intervention, known as 'shielding', could be implemented in various ways. If shielded people are grouped together in residences and isolation is imperfect, any introduction of infections within the shielding group could cause substantial mortality and thus negate the intervention's benefits. We explored the effectiveness of shielding under various modalities of implementation and considered mitigation measures to reduce its possible harms. Methods We used an individual-based mathematical model to simulate the evolution of a COVID-19 epidemic in a population of which a fraction above a given age cut-off are relocated to shielding residences, in which they have variable levels of contacts with their original household, the outside world and fellow shielding residents. We set our simulation with the context of an internally displaced persons' camp in Somaliland, for which we had recently collected data on household demographics and social mixing patterns. We compared an unmitigated epidemic with a shielding intervention accompanied by various measures to reduce the risk of virus introduction and spread within the shielding residences. We did sensitivity analyses to explore parameters such as residence size, reduction in contacts, basic reproduction number, and prior immunity in the population. Results Shielded residences are likely to be breached with infection during the outbreak. Nonetheless, shielding can be effective in preventing COVID-19 infections in the shielded population. The effectiveness of shielding is mostly affected by the size of the shielded residence, and by the degree by which contacts between shielded and unshielded individuals are reduced. Reductions in contacts between shielded individuals could further increase the effectiveness of shielding, but is only effective in larger shielded residences. Large shielded residences increase the risk of infection, unless very large reductions in contacts can be achieved. In epidemics with a lower reproduction number, the effectiveness of shielding could be negative effectiveness. Discussion Shielding could be an effective method to protect the most at-risk individuals. It should be considered where other measures cannot easily be implemented, but with attention to the epidemiological situation. Shielding should only be implemented through small to medium-sized shielding residences, with appropriate mitigation measures such as reduced contact intensity between shielded individuals and self-isolation of cases to prevent subsequent spread.

Introduction COVID-19 epidemics may prove particularly difficult to manage in low-income and crisis-affected settings of the world, where resources to scale up case management are insufficient to meet demand, insufficient public health and laboratory capacity precludes effective use of testing and contact tracing, and socio-economic circumstances (e.g. overcrowding, inadequate water and sanitation, imperative to generate an income) make it difficult for populations to adopt and sustain physical distancing 1, 2 . Over the next years, waning immunity, novel variants, and insufficient vaccine access could expose populations to renewed COVID-19 waves. This suggests a need to identify options for mitigating epidemics affecting these populations that do not require socially and economically harmful lockdown measures. One such option, known as 'shielding', consists of physically isolating individuals known to be at high risk of developing severe disease and dying if infected with SARS-CoV-2 2 . This intervention could reduce severe disease and mortality while herd immunity builds up in the low-risk population groups 3 ; it would also lessen health service pressure and enable societies and economies to remain functional.

While shielding could best be viewed as a community-led and -designed intervention with no pre-set modalities, we have previously suggested that likely arrangements could include grouping high-risk individuals together into 'shielding residences', particularly where individual shielding within households is impracticable 4 . The number of people shielded together, as well as various other characteristics of the intervention (e.g. its timing of introduction; arrangements for infection prevention and control) could also vary.

While compartmental dynamic models indicate that shielding has a substantial potential to reduce mortality and health service pressure 3 , such models do not fully capture the individual-level dynamics of the intervention: in particular, what remains unexplored is the potential harm of inadvertently introducing infection into shielded residences. This harm might or might not be outweighed by the benefit of shielding, compared to no mitigation, and could itself be mitigated through a number of measures, such as not grouping too many high-risk people together; only shielding people who are not symptomatic; and supporting people to isolate if they develop symptoms while in a shielding residence.

In order to support the design of appropriate, safe shielding interventions, we used mathematical modelling to generate quantitative predictions of the potential benefit and harms of shielding people together under different scenarios of density of shielded residents, timing of shielding implementation with respect to local epidemic onset, compliance with physical isolation and interventions to mitigate the risk of outbreaks within shielding residences. Harm here is defined as infection: it is assumed that high-risk individuals, once infected, would have a high probability of severe disease and death, as described elsewhere 3, 5 .

We used an individual-based, probabilistic mathematical model (IBM) to study the research question. We set our model within the real-life setting of Digaale internally displaced persons' (IDP) camp near Hargeisa, Somaliland, for which data on demographics and other relevant parameters was recently collected. 6 While most IBMs attempt to answer population-level questions (e.g. the potential impact of an intervention), they do so by simulating the trajectories of each individual across different infection and disease states, as a function of various individual-level characteristics. They are particularly useful to simulate small population units (e.g. households) in which chance is very influential.

In the model, individuals fall within . . CC-BY-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 January 3, 2022. 

We assume that unshielded individuals only make effective contact with shielded residents from their own household. and return to their households (we assume this is a more likely prospect than hospitalisation in . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. Parameter values and data sources Table 1 lists model parameters and their input values.

Data on social contacts were taken from a cross-sectional survey conducted among 501 individuals living in Digaale IDP camp, Somaliland. 6 The IDP camp is a permanent settlement established in 2014, has an area of approximately 150 hectares, and is situated 4km from Hargeisa, the capital city.

There are an estimated 715 inhabited shelters in Digaale, with an average household size of 4.4

individuals. The median age of residents is 15 years (interquartile range 7 to 34). 7.5% of the population is aged 60 years or older.

We simulated a total of 715 households using the Digaale data, where we randomly sampled households with replacement from the set of all households and household members in the empirical data. The population structure and contact matrix for non-household contacts was resampled in every iteration of the model. To ensure comparability between scenarios, the same population structure and random seed were used for all modelled scenarios within the same iteration of the model. . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. 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 January 3, 2022. . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. If relative contact intensity between shielded individuals were to increase rather than decrease, the benefit of shielding in shielding residences of moderate sizes decreases, though this extent is mitigated by residence size.

Self-isolation of symptomatic cases who are shielded could further increase the effectiveness of shielding, though it is less effective than an overall reduction in contact intensity within the shielded . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. Sensitivity to ܴ and pre-existing immunity Our main analysis has assumed an ܴ of 2.5, but we assessed the sensitivity of this assumption to our results in scenarios where ܴ was 1.5, 3.5, or 5. In addition, whereas our main analysis assumed that all individuals in the population were susceptible, we assessed the sensitivity of our results to scenarios where 25% or 50% of the population was immune to SARS-CoV-2 infection. Table 2 lists the median attack rates in unmitigated outbreaks under each scenario. Supplemental Figure 1 shows While the maximum effectiveness of shielding compared to unmitigated scenarios decreases as the proportion of the population that is already immune increases, more leaky implementations of shielding (a lower reduction in contacts between shielded and unshielded individuals) increase in their effectiveness as the proportion that is already immune increases. For instance, when ܴ is 5, shielding in residences of size 2 with 100% reduction in contacts between shielded and unshielded individuals prevents 60% (45 -75) of cases with 0% pre-existing immunity, 60% (44 -74) with 25% pre-existing immunity, and 46% (25 -66) with 50% pre-existing immunity. Shielding in residences of size 2 with only 40% reduction in contacts prevents 7% (3 -12) of cases with 0% pre-existing immunity, 14% (7 -23) with 25% pre-existing immunity, and 21% (8 -33) with 50% pre-existing immunity.

. CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. 

The potentially devastating impact of COVID-19 on humanitarian settings has been highlighted many times [15] [16] [17] [18] , but few empirical and modelling studies have focussed on these settings due to a lack of available data 19 . To our knowledge, our study is the first modelling study to specifically look at shielding as a COVID-19 mitigation option for IDPs or camp-like settings. Our results could be extended to other low resource settings with low vaccine coverage, where large scaled lockdowns cannot be sustained long term.

We found that shielding can be an effective measure to protect individuals at high risk of morbidity and mortality due to COVID-19 using demographic and contact data from Digaale IDP camp, though there is a risk of harm in specific scenarios. The impact of shielding largely depends on how effectively it is implemented, requiring stringent isolation of shielded people, and high uptake (coverage) for substantial population impact.

Specifically, the effectiveness of shielding mostly depends on two factors, i. the total number of individuals who are shielded together, and ii. the reduction in contact between shielded and unshielded individuals. This is not surprising, as the effectiveness of public health and social measures largely depends on changing existing social contact structures, which are impacted by these two factors.

Generally, our model projects that smaller shielding residences would be considerably more effective than larger ones. They are less likely to be breached, and there is only limited opportunity for onward transmission when breached. These small-scale implementations of shielding may also be more acceptable to implement: shielding was seen as an acceptable method to protect the most vulnerable in Sudan, though extra-household shielding arrangements were generally viewed to be socially unacceptable 20 . Shielding within or close to the old household may result in more intense care by lowrisk individuals to occur, where contact rates with unshielded individuals in general cannot be reduced by high levels.

Large shielded residences should be avoided, as they were not effective in all scenarios unless a very stringent form of shielding is implemented, with none or very little contact between shielded and unshielded individuals and among shielded individuals within the same shielding residence. This would be a potentially daunting standard to achieve in real-life application, as has been shown in the control of COVID-19 in nursing homes. 21 . CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. In settings where it is not feasible to implement small-group shielding strategies, shielding in medium sized shielding residences could be an alternative. However, shielding in medium sized shielding residences would highly benefit from further mitigation measures, such as a reduction in contact intensity between shielded individuals living in the same shielding residence, and self-isolation or quarantine of shielded cases once any resident becomes symptomatic. Contact intensity may well be lower when individuals from different households are housed together, compared to when individuals who are shielded together only belong from the same household.

If self-isolation in a given shielding residence is unfeasible, syndromic screening may be an effective alternative where symptomatic individuals exit the shielding residence as soon as possible after symptom onset. However, as COVID-19 symptoms are highly non-specific 22 , this would ideally be combined with testing strategies, and symptomatic individuals would ideally be quarantined.

Our model predicts little effect of other mitigation measures considered, such as testing individuals before they are shielded, or dismantling the shielding residence once a case arises. Both measures are most likely to detect current infections, but would miss any individuals already infected but not yet infectious. Continuous testing may be an effective alternative 23 , but was not considered in our analysis, as (asymptomatic) testing capacity may be limited in humanitarian settings.

ܴ is low, either as a result of a low initial ܴ , or pre-existing immunity in the population through natural immunity or vaccination, shielding individuals in medium-to large groups could increase their infection risk and result in harm. The epidemiological situation should be considered to assess the appropriateness of shielding in any setting.

There are several limitations to this study. Although we were able to use baseline empirical contactdata, these contact patterns could be further altered if shielding would be implemented and contact dynamics change.

We only considered Digaale IDP camp, as to our knowledge it is the only humanitarian setting for which detailed contact and demographic data is available. Social contacts are context specific 24 , and

Digaale is a relatively small-scale peri-urban settlement, which may not be representative of other lowresource or crisis-affected settings. Our model did not include an additional force of infection from outside the camp, beyond the first seeding event. Fewer than 2% of all contacts were reported to be made outside of the camp in the contact data, so the relative impact of additional transmission from outside the camp in the unshielded population is expected to be small once transmission has already been established.

. CC-BY-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.

The copyright holder for this preprint this version posted January 3, 2022. We did not estimate the number of cases and deaths that would be expected after infection, though these are proportional to the number of infections. We focussed our analysis on the impact of shielding high-risk individuals, but only assumed individuals aged 60+ years old to be at high-risk.

Although the age-risk profile may well be different in low-resource settings compared to stable settings 25 , empirical estimates are missing, and similar levels of effectiveness could be expected with broader definitions of high-risk groups.

We did not consider simultaneous implementations of other measures, such as physical distancing in the general population. In reality, shielding would most likely be implemented in addition to other nonpharmaceutical interventions. We previously found that a combination of shielding, self-isolation, and moderate social distancing may be an effective and feasible strategy for low-income countries 3 .

Whereas large-scaled lockdowns could temporarily be effective in delaying epidemic peaks 3,26 , these cannot sustainably be implemented in most humanitarian responses and low-income countries. . CC-BY-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 January 3, 2022. ܴ (y-axis). Points represent median estimates over 1000 model runs, while bars represent their 95% uncertainty intervals. In all scenarios shown, 0% of the population were immune at the start of the simulation.

. CC-BY-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 January 3, 2022. ܴ (in facet rows). Panel A shows scenarios where 20% of the high-risk population was shielded, while panel B shows scenarios where 80% of the high-risk population was shielded. In each scenario, lines represent the median attack rate across 1000 model runs at each timepoint, whereas corresponding 95% uncertainty intervals are shown by shaded areas of the same colour. Estimates for (unshielded) low-risk individuals are shown in dark-green and a dashed line. Unshielded high-risk individuals are shown in light-green, while shielded high-risk individuals are shown in blue. In all scenarios shown, 0% of the population were immune at the start of the simulation, and contacts between shielded and unshielded individuals are reduced by 80%.

. CC-BY-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 January 3, 2022. ܴ (in facet columns), and the proportion immune at the start of the simulation (in facet rows). Points represent median estimates over 1000 model runs, while bars represent their 95% uncertainty intervals. In all scenarios shown, contacts between shielded individuals within the same shielded residence remain unchanged, symptomatic individuals who are shielded do not reduce their effective contacts, and no other mitigation measures are implemented. Contacts between shielded and unshielded individuals are reduced by the levels on the yaxis. Estimates to the left of the thick vertical grey line correspond to an increase in infection risk after shielding compared to no shielding. In all scenarios, 100% of high-risk individuals were shielded. There were not enough model iterations with large enough outbreaks to implement shielding in scenarios with an ܴ of 1.5 or 2.5 and 50% prior immunity, so estimates are not available for these scenarios.

. CC-BY-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 January 3, 2022. ; https://doi.org/10.1101/2022.01.03.22268675 doi: medRxiv preprint

COVID-19 threatens health systems in sub-Saharan Africa: the eye of the crocodile

COVID-19 control in low-income settings and displaced populations: what can realistically be done? | Conflict and Health | Full Text

Response strategies for COVID-19 epidemics in African settings: a mathematical modelling study

Guidance for the prevention of COVID-19 infections among high-risk

Estimating the infection and case fatality ratio for coronavirus disease (COVID-19) using age-adjusted data from the outbreak on the Diamond Princess cruise ship

Social contacts and other risk factors for respiratory infections among internally displaced people in Somaliland

Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus-Infected Pneumonia

Epidemiology and Transmission of COVID-19 in Shenzhen China: Analysis of 391 cases and 1,286 of their close contacts. Infectious Diseases (except HIV/AIDS)

Serial interval of novel coronavirus (2019-nCoV) infections | medRxiv

The timing of COVID-19 transmission

Age-dependent effects in the transmission and control of COVID-19 epidemics

Covid-19: Temporal variation in transmission during the COVID-19 outbreak

reproductive number of the Delta variant of SARS-CoV-2 is far higher compared to the ancestral SARS-CoV-2 virus

Diagnostic accuracy of rapid antigen tests in asymptomatic and presymptomatic close contacts of individuals with confirmed SARS-CoV-2 infection: cross sectional study

Humanitarian crises in a global pandemic

Refugee camps and COVID-19: Can we prevent a humanitarian crisis? -Hanne Dahl Vonen

Responding to the COVID-19 pandemic in complex humanitarian crises

Addressing COVID-19 in humanitarian settings: a call to action

A COVID-19 Model to Inform Humanitarian Operations -Model Methodology

Acceptability and feasibility of strategies to shield the vulnerable during the COVID-19 outbreak: a qualitative study in six Sudanese communities

Impact of coronavirus in care homes in England -Office for National Statistics

COVID-19) -Symptoms

The impact of population-wide rapid antigen testing on SARS-CoV-2 prevalence in Slovakia

A Systematic Review of Social Contact Surveys to Inform Transmission Models of Close-contact Infections

Pneumococcal conjugate vaccine use during humanitarian crises

The impact of COVID-19 control measures on social contacts and transmission in Kenyan informal settlements