Preventing COVID-19 spread in closed facilities by regular testing of employees—An efficient intervention in long-term care facilities and prisons?

Background Different levels of control measures were introduced to contain the global COVID-19 pandemic, many of which have been controversial, particularly the comprehensive use of diagnostic tests. Regular testing of high-risk individuals (pre-existing conditions, older than 60 years of age) has been suggested by public health authorities. The WHO suggested the use of routine screening of residents, employees, and visitors of long-term care facilities (LTCF) to protect the resident risk group. Similar suggestions have been made by the WHO for other closed facilities including incarceration facilities (e.g., prisons or jails), wherein parts of the U.S., accelerated release of approved inmates is taken as a measure to mitigate COVID-19. Methods and findings Here, the simulation model underlying the pandemic preparedness tool CovidSim 1.1 (http://covidsim.eu/) is extended to investigate the effect of regularly testing of employees to protect immobile resident risk groups in closed facilities. The reduction in the number of infections and deaths within the risk group is investigated. Our simulations are adjusted to reflect the situation of LTCFs in Germany, and incarceration facilities in the U.S. COVID-19 spreads in closed facilities due to contact with infected employees even under strict confinement of visitors in a pandemic scenario without targeted protective measures. Testing is only effective in conjunction with targeted contact reduction between the closed facility and the outside world—and will be most inefficient under strategies aiming for herd immunity. The frequency of testing, the quality of tests, and the waiting time for obtaining test results have noticeable effects. The exact reduction in the number of cases depends on disease prevalence in the population and the levels of contact reductions. Testing every 5 days with a good quality test and a processing time of 24 hours can lead up to a 40% reduction in the number of infections. However, the effects of testing vary substantially among types of closed facilities and can even be counterproductive in U.S. IFs. Conclusions The introduction of COVID-19 in closed facilities is unavoidable without a thorough screening of persons that can introduce the disease into the facility. Regular testing of employees in closed facilities can contribute to reducing the number of infections there, but is only meaningful as an accompanying measure, whose economic benefit needs to be assessed carefully.

The global COVID-19 pandemic that emerged in Wuhan, China in December 2019 was 2 declared a Public Health Emergency of International Concern by the WHO 3 Director-General in late January 2020 and drastically changed the way of living across 4 the globe [1]. SARS-CoV-2 is an extremely contagious virus affecting the respiratory 5 system [2]. While most infections are asymptomatic and mild, severe infections are 6 life-threatening [3,4,5,6]. If the virus affects the lung it can result in diffuse 7 pneumonia, requiring oxygen supply, hospital, or even ICU treatment [7,8,9,10,11]. 8 With no effective treatment against the virus, severe episodes can result in death by 9 multiple organ failure [12]. Moreover, severe (and even mild) infections can cause 10 substantial long-term effects, potentially imposing long-term burdens on healthcare 11 systems [13,14]. From the beginning of the pandemic, older adults and individuals with 12 underlying medical conditions, particularly lung or heart disease, diabetes, obesity, etc. 13 are associated with an increased risk of developing serious complications from 14 SARS-CoV-2 infections [15]. The Centers for Disease Control and Prevention (CDC) 15 identified people aged 65 years and older and people living in a long-term care facility 16 (LTCF) as high-risk groups. Indeed, every second COVID-19 related death in the 17 Federal Republic of Germany occurred within LTCFs [16]. Likewise the rapidly growing 18 elderly population in U.S. prisons [17] is at high risk due to the exceedingly high 19 numbers of infections in such facilities [18]. 20 Draconic control measures were implemented by governments across the globe to 21 prevent the spread of the pandemic, including social distancing (cancellation of mass 22 crowdings and events, enforced physical distance, etc.), curfews, quarantine, and home 23 isolation measures, mandatory use of face masks, accompanied by massive deployment of 24 disinfectants, supply of contact tracking mobile-device applications, and diagnostic tests 25 [19,20,21]. Most commonly used are PCR tests that detect the virus in nasopharyngeal 26 swabs, diluted gargle samples, or peripheral blood. As PCR tests amplify virus-specific 27 RNA, they are characterized by very high specificity. The sensitivity of such tests varies 28 across different products on the market. Moreover, PCR tests are easy to perform. 29 Alternatives to quantitative PCR tests are CRISPR-based [22,23], which are rapidly 30 performed, and have high specificity and sensitivity. Other tests are antibody or antigen 31 based. Such tests are less specific and do not necessarily detect active infections, since 32 antibodies and antigens are present in the blood serum after the infection is cured. 33 The WHO established guidelines -including regular testing of employees and 34 residents -to protect individuals in LTCFs [24,25] due to high case fatality rates [26]. 35 Residents of LTCFs constitute a substantial group in high-income countries such as the 36 Federal Republic of Germany. With a population of 82.79 million, the number of people 37 depending on nursing either in LTCFs or at home in Germany increased from 2.5 38 million in 2011 to 3.41 million in 2017 (over 66% of them being over 90 years old) [27]. 39 The capacity of LTCFs in Germany was 952 367 beds (full stationary capacity: 885 488) 40 in 2017, with 743 120 beds (723 451 full stationary) filled (623 182 beds in 2011, 612 183 41 being full stationary) [27]. These are sustained by 764 648 employees, 64% of which are 42 care and support personnel [27] [17].

54
The use of routine screening of residents, employees, and visitors before entering an 55 LTCF by diagnostic tests was mentioned in guidelines by public health authorities 56 [24,25,34] and also suggested for incarceration facilities [35]. The impact of such 57 control measures can be studied through the use of mathematical models.

58
Here, a mathematical model, based on the freely available CovidSIM simulation tool, 59 is adapted to estimate the benefit and economic gain of routine screening for COVID-19 60 infections of employees in LTCFs and retention facilities by PCR tests. In particular, we 61 study the impact of I. the frequency at which employees are tested, (ii) the processing 62 time to obtain test results, and (iii) the quality of the PCR test in terms of sensitivity. 63 While the model is described verbally in the main text, a concise mathematical 64 description can be found in the S1 Appendix. The model is exemplified by parameters 65 that reflect the situation in the Federal Republic of Germany.

67
We study the impact of testing employees in LTCFs or incarceration facilities to protect 68 immobile risk groups from COVID-19 infections using an extended SEIR model, i.e., by 69 a deterministic compartmental model of ordinary differential equations. In particular, 70 the model is an extension of that underlying the pandemic preparedness tool CovidSIM 71 [36]. The flow chart of the model is presented in Fig 1. The model is described verbally 72 with the concise mathematical description found in S1 Appendix. In the description, we 73 focus on LTCFs, although the model equally applies to prisons.

74
A population of size N is divided into three groups, I. the immobile risk group (Ri), 75 i.e., residents of LTCFs, (ii) the employees (staff) working in LTCFs (St), who are in 76 close contact with the risk group, (iii) and the general population (Ge), i.e., the rest of 77 the population.

78
Each group (Ge, St, Ri), is further subdivided into susceptible, infected, recovered, 79 or dead individuals. Infected individuals pass through I. a latency period (not yet 80 infective), (ii) a prodromal period (already partly infective but not yet exhibiting 81 characteristic symptoms), (iii) a fully contagious period (symptoms ranging from 82 non-existent or mild to severe), and (iv) a late infective period (no longer fully 83 contagious). All individuals either recover from COVID-19 and obtain full permanent  ignored, as we assume a pandemic in a large population in a relatively short time period. 89 In classical SEIR models, individuals in the latent, prodromal, infected, and late 90 infected classes simply proceed from one stage to the next at a rate directly related to 91 the residence time in each stage. This simplistic flow implicitly assumes that the times 92 individuals spend in the various compartments are exponentially-distributed, and hence 93 have a large variance. In particular, a proportion of individuals progresses too fast, 94 whereas others progress much too slow.

95
To resolve this, we subdivide the latent, prodromal, fully contagious, and late 96 October 5, 2020 3/14 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

97
This ultimately leads to more realistic durations and hence dynamics.

98
The characteristics of the population subgroups (Ge, St, Ri) are incorporated within 99 the contact behavior. Namely, the risk group has mainly contacts with other individuals 100 in the risk group and the LTCF employees, whereas their contacts with the general 101 population are limited. The employees (St) have contacts among themselves, with the 102 risk group and the general population. However, the general population has mainly 103 contacts among themselves. Given a contact within or between certain sub-populations, 104 the contact occurs at random. The contact behavior is captured by the contact matrix 105 (see S1 Appendix section "The basic reproduction number and the next generation 106 matrix").

107
Susceptible individuals acquire infections through contacts with individuals in the 108 prodromal, the fully contagious, or the late infectious periods at rates β P , β I , β L , 109 respectively, which are identical for each subgroup.

110
The basic reproduction number R 0 is the average number of infections caused by an 111 average infected individual in a completely susceptible population during the infectious 112 period. In a subdivided population (here Ge, St, Ri), the definition of R 0 is not 113 straightforward and has to be derived from the next-generation matrix [37] (see S1

114
Appendix section "The basic reproduction number and the next generation matrix").

115
Importantly, R 0 fluctuates seasonally with a yearly average basic reproduction number 116 ofR 0 .

117
First, infected individuals are latent carriers, before they enter the prodromal period, 118 in which they become partly infective. From there, they enter the fully contagious 119 period, at the beginning of which, it is determined whether the infection proceeds as 120 symptomatic or asymptomatic. The likelihood to suffer from a symptomatic episode is 121 elevated in the risk group (Ri).

122
Covid-19 confirmed individuals are subject to case isolation. Specifically, a fraction 123 of symptomatic individuals will be detected and isolated in quarantine wards (perfect 124 isolation preventing all contacts). If the wards are full, infected individuals are sent into 125 home isolation (imperfect isolation, preventing only a fraction of contacts). Regarding 126 this, there are differences in the subgroups: each symptomatic individual in the risk 127 group will be detected and isolated in quarantine inside the LTCF (perfect isolation), 128 whereas only a fraction of the general population and the LTCF employees will be 129 isolated. Infected individuals further progress to the late infective stage, during which 130 they will stay quarantined as determined during the fully contagious stage. Importantly, 131 LTCF employees will be tested for COVID-19 on a regular basis. We assume that the 132 test is 100% specific, i.e., there are no false-positive test results, reflecting PCR-or 133 CRISPR-based tests [22,23]. If tested positive, they will be isolated either in quarantine 134 wards or at home, in which case all contacts with the risk group are prevented. Staff can 135 be tested positive during any of the infected stages (latent, prodromal, fully contagious, 136 late infective), however with different sensitivity depending on the characteristic of the 137 COVID-19 test being used, irrespective of symptoms. In particular, there is a possibility 138 (depending on the sensitivity of the COVID-19 test) that employees are isolated already 139 during the latent period before they are infectious. Test results are not obtained 140 instantaneously, but with a time delay. Infected staff can still infect others during this 141 time. The waiting time for the test results depends on the available infrastructure.

142
Finally, late infected individuals, either recover or die. Only symptomatic infections 143 can cause death. The fraction of lethal infections is higher in the risk group.

144
All case isolation mechanisms are not initially present, but implemented with a time 145 delay after the initial occurrence of the disease. In addition to case isolation, general 146 contact reducing interventions, e.g., social distancing, curfews, etc. will be sustained for 147 a specific time interval during the beginning of the epidemic. During the time interval 148 in which case isolation measures are sustained, contacts between the risk group and the 149 general population are reduced, reflecting preventative measure. Furthermore, visitors 150 have to provide a negative test result before entering the LTCF. To obtain conservative 151 estimations the latter intervention is ignored in the simulations.

157
The effect of protecting an immobile risk group (LTCF residents) by regularly 158 PCR-testing LTCF employees, who are the most likely to import the disease into the 159 facility, is studied. Model parameters are adjusted to roughly reflect the situation in the 160 Federal Republic of Germany, one of the countries that has successfully intervened in 161 the COVID-19 epidemic. The model itself is applicable to any industrial nation with an 162 aging population. The aim is to investigate the effects of protecting the risk group and 163 to estimate the demand for PCR tests. Some testing scenarios are not feasible in terms 164 of logistics and available testing capacities, and just serve as a comparison. 165 The parameters used are listed in Tables 1, and S1 Table-S6 Table. Germany has a 166 population of roughly N = 83 million. We assume 700 000 elderly individuals live in 167 LTCFs in which control interventions by PCR testing can be implemented. All 168 employees of LTCFs amount to approximately 760 000. This number however includes 169 employees in the administration and external services, who are not regularly working in 170 these facilities. Hence, a number of 500 000 employees was assumed to work regularly in 171 the LTCFs. The first COVID-19 cases were introduced in the middle of February 2020, 172 corresponding to time t = 0. A basic reproduction number of R 0 = 3.5 was assumed.

173
When studying the effect of seasonal variation, R 0 was assumed to fluctuate 43% over 174 the year, with an annual averageR 0 = 3.5 and a peak roughly in late December 175 (t R0 max = 200). The average duration of the latent, prodromal, fully contagious and late 176 infective states were assumed to last on average D E = 3.7, D P = 1, D I = 7, D L = 7 177 days, respectively. In the prodromal and late invective states, individuals were assumed 178 to be half as infective as in the fully contagious state. Individuals in the risk group were 179 more likely to develop severe symptoms (f Sick = 58% vs. f  In essence, the model without testing interventions is equivalent to the one 184 underlying CovidSim 1.0 or 1.1 [cf. 36, http://covidsim.eu]. Hence, we used a 185 combination of general contact reduction and case isolation as proposed in [36]. 186 Clearly, the available capacities of tests, the infrastructure to perform tests, the  Tables 1, and S1 Table -S6 Table. The dashed grey line in panels (E-H) corresponds to the value of R 0 (right y-axis).  In the presence of seasonal fluctuations in R 0 , qualitatively the same picture emerges 211 (see Fig 3E-H, S4 Fig). However, the differences between the testing rates are more 212 pronounced, particularly between testing every 7 vs. 14 days.  Tables 1, and S1 Table -S6 Table. The dashed grey line in panels (E-H) corresponds to the value of R 0 (right y-axis).  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Test-processing time
The copyright holder for this preprint this version posted October 14, 2020. .  Tables 1, and S1 Table -S6 Table. The dashed grey line in panels (E-H) corresponds to the value of R 0 (right y-axis).
testing with a processing time of 48 hours was assumed in the absence of seasonal    Tables 1, and S1 Table - While this suggests, that the intervention is cost-efficient, the actual gain is likely to 247 be underestimated. Namely, long-term effects of infections and additional costs are not 248 accounted. Furthermore, costs for testing can presumably be reduced. In particular, the 249 testing intervention results in a 10-fold reduction of the number of infected individuals 250 even under the most conservative setup. Hence, the follow-up costs of infections reduce 251 10-fold.

253
Elderly citizens and particularly residents of long-term care facilities (LTCF) were 254 identified early as a vulnerable risk group that deserves particular protection, as 255 reflected by the WHO guidelines in March 2020 [24]. Regular testing of LTCF employees 256 and residents for COVID-19 was explicitly mentioned by the John Hopkins University 257 in their Guidance on Protecting Individuals Residing in Long-Term Care Facilities [25]. 258 Furthermore, such recommendations can also be found in the WHO policy brief on 259 preventing and managing COVID-19 across long-term care services from July 2020 [34]. 260 To evaluate the effectiveness of testing interventions to protect resident risk groups 261 in LTCFs we extended the model underlying the pandemic preparedness tool CovidSim 262 [36, http://covidsim.eu]. In particular, the deterministic model formulated as systems of 263 differential equations was extended to separate the risk group of LTCF residents and  In our investigations, the model was adjusted to reflect the situation in the Federal 274 Republic of Germany. However, the model is not restricted to one particular country 275 but will be applicable to any other industrialised nation with a similar age structure.

276
The results clearly indicate that regular COVID-19-screening of LTCF employees by 277 testing successfully reduces the number of cases and deaths in the resident risk group. 278 Even with relative conservative assumptions a 10-fold reduction is achieved. Our results 279 indicate that the frequency at which employees are tested has a strong effect. Testing 280 once every 7 to 14 days is sufficient and seems to be a realistic. Although more frequent 281 testing further improves the intervention, the gain is insignificant. Importantly, the 282 waiting time for the return of test results (ranging from 12 to 96 hours) has no 283 noticeable effect. A waiting time of 48-72 hours seems to be realistic when compared 284 with the requirements for international air traveling since summer 2020, requiring 285 passengers to provide proof of a negative COVID-19 test, taken no longer than 72 hours 286 before departure. The quality of the test in terms of sensitivity has a clear impact on 287 the outcome. Here, PCR tests were assumed to be relatively conservative, considering 288 the fact that these tests are constantly improved. Our simple rough estimates of the 289 economic gain of the proposed intervention, comparing the potential costs of COVID-19 290 treatments with the costs for the testing intervention, is substantial. These estimates 291 are conservative as they do not account for health care costs for long term effects of the 292 infection and capacity shortages in the LTCFs, e.g., due to isolation measures of 293 infected residents. Notably, testing a population of 500 000 LTCF employees every two 294 weeks requires a total of 11.5 million tests per year (assuming 23 tests per person per 295 year), or 221 000 tests per week, which is a realistic number in Germany, having a 296 capacity of approximately 1.4 million tests per week in September 2020 [39]. 297 Notably, similar results can be obtained for serological tests. However, these tests 298 October 5, 2020 8/14 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) 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 October 14, 2020. . https://doi.org/10.1101/2020.10.12.20211573 doi: medRxiv preprint typically have lower specificity, so that false-positive results can no longer be ignored. per test. Hence, the economic gain would be further amplified due to the cost efficiency. 307 The proposed intervention considers regular testing only of LTCFs employees (staff) 308 not of residents or the general population upon entry. The reason is that we wanted to 309 study the impact of minimal-invasive control measure. Namely, the risk group is twice 310 as large as the target population being tested. Hence, also testing the risk group would 311 result in the requirement of unrealistically many tests.  Tables 1, and S1 Table -S6 Table,  The parameters used are listed in Tables 1, and S1 Table -S6 Table. Seasonal (which was not certified by peer review) 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 October 14, 2020.  Tables 1, and S1 Table -S6 Table,  Tables 1, and S1 Table - Tables 1, and S1 Table -S6 Table, Table 1. The parameters used are listed in Tables S1 Table -S6 Table, Tables 1, and S1 Table -S6   385   Table. 386 S1 Appendix. Mathematical Description 387 S2 Sick Dead g=f f Sick Dead All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

H.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) 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 October 14, 2020.

H.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) 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 October 14, 2020.

H.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) 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 October 14, 2020.

H.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.