Background
From nursing homes to rehabilitation hospitals, long-term care facilities (LTCFs) worldwide are hotspots for outbreaks of coronavirus disease 2019 (COVID-19) [
1]. LTCF patients (or residents) require continuing care, live in close proximity to one another, and are typically elderly and multimorbid, placing them at elevated risk of both acquiring severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2, the virus) and suffering severe outcomes from COVID-19 (the disease) [
2‐
4]. Healthcare workers (HCWs) are also susceptible to infection and, amidst imperfect hygiene and infection prevention measures, potentially transmit the virus through necessary daily interactions with both patients and staff [
1,
5]. Although the full extent of the ongoing pandemic is unclear and ever-evolving, LTCFs have and continue to bear a disproportionate burden of SARS-CoV-2 infection and COVID-19 mortality [
3,
6,
7]. Across Europe, for instance, LTCFs have accounted for an estimated 30–60% of all COVID-19 deaths as of June 2020 [
8].
Effective COVID-19 surveillance is essential for timely outbreak detection and implementation of necessary public health interventions to limit transmission, including case isolation, contact tracing and enhanced infection prevention [
9‐
11]. The current gold-standard diagnostic test for active SARS-CoV-2 infection is reverse transcriptase polymerase chain reaction (RT-PCR), typically performed on clinical specimens from nasopharyngeal swabs [
12]. Though sensitive and highly specific, RT-PCR is relatively resource intensive, must be outsourced for institutions lacking on-site infrastructure, and is widely subject to shortages and specific usage guidelines. For instance, a common practice in LTCFs in France, the Netherlands, the UK, the USA, and elsewhere has been to restrict testing to individuals presenting with characteristic COVID-19 symptoms [
4,
13‐
15]. Yet symptomatic infections represent just the tip of the iceberg: many infections cause no or only mild symptoms, produce high quantities of virus in the absence of symptoms, and experience relatively long delays until symptom onset [
16‐
19]. Silent transmission from asymptomatic and pre-symptomatic infections is a known driver of COVID-19 outbreaks [
20,
21], with non-symptomatic cases acting as Trojan Horses, unknowingly introducing the virus into healthcare institutions and triggering nosocomial spread [
8,
22,
23].
Insufficient surveillance systems, including those lacking testing capacity or relying only on symptoms as indications for testing, have been identified as aggravating factors for COVID-19 outbreaks in LTCFs [
8,
16,
24‐
27]. Various surveillance strategies have been proposed to optimize testing while accounting for the particular transmission dynamics of SARS-CoV-2, including randomly testing HCWs, testing all patients upon admission, and universal or serial testing [
28‐
30]. Yet COVID-19 surveillance is limited in practice by available testing capacity and health-economic resources, particularly for institutions in low- and middle-income settings [
31,
32]. In light of testing shortages, group testing (sample pooling, combining clinical specimens from multiple individuals into a single biological sample for a single RT-PCR test) has garnered attention as a diagnostically sensitive and resource-efficient alternative to individual-based testing [
33‐
38].
In order to mitigate and prevent future nosocomial outbreaks, there is an urgent need to optimize COVID-19 surveillance in long-term care settings, taking into account both the unique epidemiological characteristics of SARS-CoV-2 and limited availability of testing resources [
1]. Here, we investigated the efficacy, timeliness, and resource efficiency of a range of COVID-19 surveillance strategies using simulations from a dynamic SARS-CoV-2 transmission model that uses detailed inter-individual contact data to describe interactions between patients and staff in long-term care.
Discussion
The ongoing COVID-19 pandemic continues to devastate LTCFs worldwide, with high rates of mortality among particularly frail and elderly patients, and high rates of infection among patients and staff alike [
3,
5,
6,
8]. This motivates a need for timely and efficient surveillance strategies that optimize limited testing resources to detect emerging outbreaks as quickly as possible. We used an individual-based transmission model to simulate COVID-19 outbreaks in LTCF settings and evaluated a range of testing strategies for their efficacy and efficiency in detecting these outbreaks across various epidemiological assumptions and scenarios.
Our findings suggest that LTCFs can detect emerging COVID-19 outbreaks most quickly by using testing cascades, provided that they have substantial daily testing capacity (on the order of at least 1 test/10 beds/day). The most effective cascades considered multiple indications, including both COVID-like symptoms and patient admission, and detected outbreaks days ahead of traditional symptom-based screening and prior to the accumulation of additional infections. By extension, cascades had the greatest probability of identifying non-symptomatic cases, a known challenge for COVID-19 surveillance in real LTCF settings [
1]. These findings held in sensitivity analyses considering outbreaks in a smaller, 30-bed geriatric LTCF (Additional File
2: Fig. S8), as well as when halving or doubling SARS-CoV-2 transmissibility (Additional File
2: Figs. S9, S10). Although only a select few indications were considered in the present study, LTCFs may consider a wider range of known risk factors for SARS-CoV-2 acquisition in their own cascades to maximize the probability of detecting emerging outbreaks before widespread transmission.
COVID-19 surveillance was less effective in resource-limited settings because of an inability to regularly test large numbers of patients and staff. In our analysis, group testing was the most effective means of COVID-19 surveillance under limited testing capacity, and across all epidemiological scenarios and capacities was the most resource-efficient means to improve surveillance with respect to a “bare minimum” reference of only testing individuals with severe COVID-like symptoms. Even when assuming strict diagnostic cut-offs in sensitivity analysis, group testing strategies remained effective up to a maximum of 32 swabs per test (Additional File
2: Fig. S5). This broadly agrees with modelling results suggesting that group testing could be cost-effective for screening in large populations, as well as empirical evidence for the efficiency of group testing for COVID-19 surveillance in nursing homes [
33,
71]. As with cascades, LTCFs that conduct group testing may consider a wider range of indications than was possible to include in this study, in order to maximize the probability of including potentially infected patients and staff in routine group tests. These findings reinforce current guidance from the World Health Organization, endorsing sample pooling to increase COVID-19 diagnostic capacity when testing demand outstrips supply, but cautioning against its use for contact tracing or in high-prevalence settings [
63]. This is consistent with its implementation in the present study, as a means of surveillance in resource-limited long-term care settings without known active cases, but which are nonetheless susceptible to outbreaks.
Our analysis was limited to classical two-stage group testing, initially proposed by Dorfman in 1943 for syphilis screening among World War II soldiers [
67], in which all individuals in a positive group test are individually re-tested to determine who is infected. This is regarded as the most straightforward approach [
35], and we conservatively assumed re-swabbing in addition to re-testing of all individuals in a positive group test to account for potential logistical challenges of storing and maintaining large numbers of swabs for re-testing. Various alternative group testing strategies have been proposed and implemented elsewhere, including the use of simultaneous multi-pool samples, non-adaptive pooling schemes, and others [
35,
37,
65,
66]. These have the advantage of not requiring separate re-testing of all individuals in a positive group test and are hence more efficient in terms of the number of tests required for case identification. However, these strategies may also require additional testing infrastructure and expertise, which may be cost-prohibitive for the resource-limited settings that may benefit most from group testing in the first place. Decision-makers must consider trade-offs between the various costs and benefits of different group testing technologies, including how many individuals to include per test, how many stages of testing to conduct, and other potential logistical challenges [
63].
We predicted that silent introductions of SARS-CoV-2 led to large outbreaks in the absence of specific control strategies. This is consistent with large COVID-19 outbreaks observed in LTCFs worldwide [
6,
8,
10,
16], including an infamous outbreak in early 2020 in King County, Washington that resulted in 167 confirmed infections within 3 weeks of the first reported case [
5]. We further predicted that larger proportions of patients became infected than staff, consistent with emerging evidence of higher SARS-CoV-2 incidence in patients than staff across LTCF settings worldwide [
3,
8]. We also predicted larger and more rapid outbreaks when SARS-CoV-2 was introduced through admission of an infected patient, rather than through a member of staff infected in the community, with important implications for surveillance efficacy (Additional File
2: Fig. S11). These findings are likely due to the nature of human interactions in the LTCF upon which we based our model, in which patient-patient contacts were particularly long and numerous [
41]. Overall, these findings reinforce both (i) a need to screen incoming patients potentially exposed to or infected with SARS-CoV-2 [
72] and (ii) the importance of interventions to limit contact between patients (e.g. social distancing among retirement home residents), as already widely recommended for affected facilities in the current pandemic context [
4].
Simulated outbreaks were further characterized by delays between silent introduction of SARS-CoV-2 and first onset of COVID-19 symptoms, during which time new infections not (yet) showing symptoms accumulated. This is consistent with reported transmission dynamics of SARS-CoV-2; for instance, modelling studies have estimated that 30–57% of secondary infections among identified transmission pairs resulted from pre-symptomatic transmission [
49], and that, early on in COVID-19 outbreaks in Singapore and Tianjin, pre-symptomatic transmission accounted for at least 65% of all transmission events [
19]. Findings are also consistent with high proportions of asymptomatic infection, and important roles for pre-symptomatic and asymptomatic transmission reported in various LTCF outbreaks [
5,
16,
21,
26‐
29]. The often silent nature of SARS-CoV-2 transmission highlights epidemiological challenges associated with screening for emerging outbreaks using symptoms alone. In addition to the strategies highlighted above, we found that testing patients and healthcare workers with any and not only severe COVID-like symptoms can substantially improve outbreak detection, supporting recommendations to expand testing criteria in LTCFs to include individuals with atypical signs and symptoms of COVID-19, such as muscle aches, sore throat, and chest pain [
72].
A strength of the present study is the use of highly detailed inter-individual contact data to inform our individual-based transmission model. This allowed for recreation of life-like interaction dynamics among and between LTCF patients and staff and facilitated simulation of more realistic SARS-CoV-2 dissemination than a traditional mass-action transmission process. We are aware of no other studies using detailed individual-level contact networks to simulate SARS-CoV-2 transmission in healthcare settings, nor of studies using transmission modelling to evaluate COVID-19 surveillance strategies for emerging outbreaks.
Previous studies of COVID-19 surveillance have largely focused on the ability of testing strategies to mitigate ongoing SARS-CoV-2 transmission in active outbreak settings [
73]. In particular, contact tracing has been identified as a highly effective form of surveillance, by targeting testing and isolation interventions to individuals at high risk of infection [
74,
75]. However, these findings have limited relevance for emerging outbreaks, where active SARS-CoV-2 infection and ongoing transmission are not yet known. For healthcare facilities vulnerable to SARS-CoV-2 introductions, specific surveillance strategies are required for initial outbreak detection, in order to alert healthcare professionals and decision-makers to the presence of the virus in their institutions. Only then can proven measures like contact tracing and case isolation be implemented. By targeting this important epidemiological context, our findings complement an existing evidence base that has until now largely focused on how to control outbreaks that are already detected and well underway.
This work has several limitations. First, substantial uncertainties remain regarding epidemiological characteristics of COVID-19. It is well established that various COVID-19 outcomes vary with age, comorbidity, and frailty [
76‐
78], but quantitative descriptions of these associations are incomplete, and it was not possible to reliably integrate such individual-level variation into our model. For instance, owing to individual-level risk factors, higher rates of symptomatic infection may be expected among LTCF residents than staff. Yet an outbreak investigation across six London care homes experiencing COVID-19 outbreaks estimated similar rates of asymptomatic infection in patients and staff and found no association with age [
79], while the meta-analysis used to inform asymptomatic infection in our work highlighted poor reporting of age in included studies, precluding quantification of its relationship to COVID-19 symptom risk [
51]. Nonetheless, calibrating model parameters to individual-level risk factors would facilitate more realistic simulations, and accounting for potentially higher rates of severe infection among older and frailer individuals could result in improved performance of symptom-based surveillance, including corresponding cascades and group testing strategies. This distinction may be particularly relevant for hospices, nursing homes, and other LTCFs with particularly frail populations; however, patients in the present rehabilitation hospital population were relatively young (median 58 years, IQR 47–72), limiting potential impacts of age-stratified disease progression in this study.
Other epidemiological uncertainties that we were unable to account for include temporal variability in SARS-CoV-2 transmissibility over the infectious period, individual-level variation in transmissibility, and a potential role for environmental acquisition [
80,
81], although recent evidence suggests the former may be of limited relevance [
82]. Further, most LTCFs have already implemented control measures, such as interruption of social activities and provisioning of personal protective equipment, that should act to reduce transmission from baseline. We conducted sensitivity analyses to consider unusually high and low transmission rates to reflect these uncertainties. Although SARS-CoV-2 spread more or less quickly, the relative efficacies of surveillance strategies were largely unchanged in these analyses, resulting in the same conclusions for optimizing use of limited testing resources to detect COVID-19 outbreaks (Additional File
2: Figs. S9, S10, S12, S13).
Second, LTCFs represent a diverse range of healthcare institutions, each with unique specializations, patient populations and living conditions, and the generalizability of our findings across these settings is not clear. In a sensitivity analysis, we restricted simulations to the 30-bed geriatric ward to approximate a smaller LTCF geared towards elder care, with an average 8.0 daily patient-patient contacts and 8.3 daily patient-staff contacts. This compares to patterns observed in a nursing home in Paris (5.0 daily patient-patient contacts, 6.3 daily patient-staff contacts) [
83], and corresponding results may better reflect a nursing home environment than the baseline analysis. In this much smaller facility, high testing and swabbing capacities approximated universal testing strategies, in which large proportions of individuals were routinely tested. This explains why randomly testing among all individuals was among the most effective strategies at highest testing capacity (Additional File
2: Fig. S8), and why pooling even relatively small numbers of randomly selected individuals was a particularly efficient strategy in this setting (Additional File
2: Fig. S14). Otherwise, overall conclusions for surveillance were similar to the baseline LTCF.
Finally, the testing landscape for COVID-19 is due to shift quickly, with increased testing capacity and alternative testing technologies, such as rapid diagnostic tests, likely to become increasingly available in the coming months and years. However, uptake of new technologies is certain to be heterogeneous, and testing resources may remain limited for the foreseeable future, particularly in low- and middle-income settings [
31,
32]. To reflect a scenario with more effective testing technology, in a sensitivity analysis, we assumed higher and more stable RT-PCR sensitivity over the course of infection. In this analysis, qualitative surveillance conclusions were again unchanged from the main analysis, although testing patients upon LTCF admission was notably more effective than in the main analysis (Additional File
2: Fig. S15). Although we explicitly modelled standard RT-PCR testing throughout this study, our findings may be broadly generalizable to other COVID-19 testing technologies with limited capacity. Findings for group testing, however, necessarily assume that pooling clinical samples is both logistically feasible and retains sufficient diagnostic sensitivity, as demonstrated for RT-PCR and SARS-CoV-2. Further, even in settings with abundant testing capacity, limiting the number of tests necessary to detect an outbreak will remain a priority given health-economic concerns.
Acknowledgements
This work resulted from the MOD-COV Project (Modelling of the hOspital Dissemination of SARS-CoV-2), a collaboration between the Institut Pasteur, the Conservatoire Nationale des Arts et Métiers, and the AVIESAN/REACTing working group “Modelling SARS-CoV-2 dissemination in healthcare settings”, whose members we thank (Niccolo Buetti, Christian Brun-Buisson, Sylvie Burban, Simon Cauchemez, Guillaume Chelius, Anthony Cousien, Pascal Crepey, Vittoria Colizza, Christel Daniel, Aurélien Dinh, Pierre Frange, Eric Fleury, Antoine Fraboulet, Marie-Paule Gustin, Lidia Kardas-Sloma, Elsa Kermorvant, Jean Christophe Lucet, Chiara Poletto, Rodolphe Thiebaut, Sylvie van der Werf, Philippe Vanhems, Linda Wittkop, Jean-Ralph Zahar).
We also thank the i-Bird Study Group (Anne Sophie Alvarez, Audrey Baraffe, Mariano Beiró, Inga Bertucci, Pierre-Yves Boëlle, Camille Cyncynatus, Florence Dannet, Marie Laure Delaby, Pierre Denys, Matthieu Domenech de Cellès, Eric Fleury, Antoine Fraboulet, Jean-Louis Gaillard, Boris Labrador, Jennifer Lasley, Christine Lawrence, Judith Legrand, Odile Le Minor, Caroline Ligier, Lucie Martinet, Karine Mignon, Catherine Sacleux, Jérôme Salomon, Thomas Obadia, Marie Perard, Laure Petit, Laeticia Remy, Anne Thiebaut, Damien Thomas, Philippe Tronchet, Isabelle Villain), as well as Jacob Barrett for helpful discussion.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.