Background
The coronavirus disease 2019 (COVID-19) pandemic caused by infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) poses a great risk to people experiencing homelessness. Across the United States (US), the estimated 568,000 people who experience homelessness nightly [
1] are likely to suffer a disproportionate disease burden and need for hospitalization [
2,
3]. People experiencing homelessness are on average older and have a high prevalence of comorbidities that are risk factors for severe COVID-19 [
2]. Multiple outbreaks in homeless shelters have occurred in several cities, including San Francisco, Boston, Seattle, Atlanta, and Los Angeles, with attack rates of up to 67% [
4‐
8]. Homeless shelters have remained open in many cities despite the high incidence of infection in the community, concern about the risk of further outbreaks, and uncertainty over the effectiveness of different infection control strategies. There is an immediate need to identify the best infection control strategy to reduce the risk of outbreaks and assess the safety of continuing to operate congregate shelters where transmission in the community is high.
The role of shelters and associated infection control practices in the transmission of COVID-19 among people experiencing homelessness is still poorly understood. Given the current understanding that the SARS-CoV-2 virus is transmitted predominantly through respiratory droplets, with some airborne transmission [
9], there is a need to consider policies to limit transmission within high-density congregate living environments. Different infection control strategies are currently recommended based on the level of transmission in the external community [
10]. These include routine symptom screening, polymerase chain reaction (PCR) testing, universal mask wearing, and relocation of individuals at high risk of severe disease to non-congregate settings [
11]. There is limited evidence on the effectiveness of strategies to reduce transmission in congregate settings, and thus, further research is urgently needed to guide city-level policy across the US.
The goal of this study is to identify the most effective infection control strategy to slow the spread of COVID-19 among people experiencing homelessness who reside in shelters. We address this pressing question by estimating the comparative health outcomes of key infection control strategies using a simulation model calibrated to data on homeless shelter outbreaks.
Discussion
Several outbreaks of COVID-19 with high attack rates have occurred in homeless shelters across the US, and there remains uncertainty over the best infection control strategies to reduce outbreak risk in shelters. In this study, we applied a simulation analysis to identify infection control strategies to prevent future outbreaks. We found that in high-risk shelters that are unable to maximize basic infection control practices that sufficiently reduce the transmissibility of SARS-CoV-2 (e.g., social distancing, reduced living density), no additional infection control strategy is likely to prevent outbreaks. Similarly, in cities with high community incidence, no infection control practices are likely to prevent an outbreak. In contrast, in lower-risk shelters with low background community incidence, the implementation of strategies such as symptom screening, routine PCR testing, and masking would help reduce outbreak risk.
We found a wide range of transmissibility of SARS-CoV-2 based on observed outbreaks in homeless shelters, which greatly affects intervention impact. We estimated basic reproduction numbers (
R0) of 2.9–6.2 from aggregate PCR test data from outbreaks in five shelters in Seattle, Boston, and San Francisco between March and April 2020. This range of
R0 values is at the high end of estimates reported in the literature [
52,
63,
83,
84] and likely reflects a high degree of heterogeneity in infectiousness between individuals [
62‐
66] and a highly conducive environment for transmission within these shelters early in the pandemic due to lack of existing infection control practices and high living density at the time of the outbreaks. For these
R0 values and representative background infection rates, we found that the infection control strategies considered are unlikely to prevent outbreaks (probability < 50%), even when combined. Nevertheless, they do reduce the incidence of infection and clinical disease and slow the growth of the outbreak (Fig.
2). Our
R0 estimates are likely not entirely representative of general transmission potential in shelters now given that the outbreaks occurred early during the pandemic when control measures were limited and that non-outbreaks and smaller outbreaks may go undetected or unreported. Control measures such as rehousing of individuals to single hotel rooms appear to have been successful, and incidence has in general been lower in the homeless population than anticipated [
85]. However, there have been subsequent large outbreaks in homeless shelters despite reduced shelter density and stringent control efforts [
86‐
88]. This supports our finding that outbreaks in congregate shelters remain likely even with fairly intensive infection control practices.
In lower transmissibility settings, e.g., with R0 = 1.5, which may be more representative of typical shelters now due to improved social distancing and basic infection control practices, the intervention strategies we have considered are more likely to prevent outbreaks (probability up to nearly 75% under combined interventions, for a moderate background infection rate of approximately 120/1,000,000/day).
A key remaining issue given the limited availability of alternative housing for people experiencing homelessness is identifying the characteristics that distinguish low-risk shelters (those similar to the
R0 = 1.5 scenario considered here) that can be operated with low outbreak risk with implementation of infection control strategies. Data are limited, but available evidence suggests that social distancing and reductions in super-spreading are likely to be key factors [
62,
63,
89‐
91]. Strategies that may achieve these goals include reducing living density, spacing bedding, reducing communal activities, and adopting staffing models that limit social contacts.
The fact that intervention impact and the probability of averting an outbreak decrease significantly with increasing background infection rate in the community (Fig.
1) suggests a need for alternative housing arrangements for people experiencing homelessness in locations in which community incidence is moderate to high—100–500 infections/1,000,000/day, equivalent to 25–125 confirmed cases/1,000,000/day assuming fourfold underreporting (see Additional file
1). In lower background incidence settings, combined daily symptom-based screening, twice-weekly PCR testing, universal masking, and relocation of high-risk individuals to non-congregate settings would reduce outbreak risk and limit the incidence of infection and severe disease if outbreaks do occur.
Our findings broadly agree with those of two other modeling studies of interventions against COVID-19 in homeless shelters: one in the US [
92] and the other in England [
93]. The former found that a combination of daily symptom screening with PCR testing of symptom-positive individuals, universal PCR testing every 2 weeks, and alternative care sites for those with mild/moderate COVID-19 would significantly reduce infections, while remaining cost-effective, but unlike our analysis did not consider variation in the effectiveness of interventions with community incidence. The latter study supports our results on the high risk of outbreaks in congregate homeless shelters, as it found that outbreaks in homeless shelters are likely even when incidence in the general population is low and estimated that closure of congregate shelters during the first pandemic wave in England averted over 90% of infections that would have otherwise occurred in the homeless population.
Each infection control strategy is limited in some aspect [
17,
18,
69,
94‐
96]. Symptom-based screening has very low sensitivity to detect infections early in the clinical course (when people are most infectious) and has poor specificity [
72,
76,
77,
97]. The impact of routine PCR testing is limited by imperfect PCR sensitivity (~ 75%), especially early in the infection course [
28], as well as the need for frequent testing and missing onset of infectiousness between testing periods. Other analyses support our finding that testing less than once or twice weekly leaves a high risk of outbreaks (e.g., testing once every 2 weeks gives a 30% lower probability of averting an outbreak than twice-weekly testing, Fig.
3) [
73‐
75]. However, once- or twice-weekly testing may be financially and logistically infeasible. Similarly, relocation of high-risk persons to independent housing is resource-intensive. Frequent testing and universal masking also suffer issues with adherence and may not be possible for all individuals at all times in homeless shelters.
This study has a number of limitations. Due to limited data availability, we only calibrated the model to a small number of shelter outbreaks, the
R0 estimates for which are likely to be higher than for the average shelter since they occurred early in the pandemic and larger outbreaks are more likely to be reported. The cross-sectional aggregate nature of the majority of the data also led to wide uncertainty intervals around the fitted parameters, without independent identifiability between them (Additional file
1: Figure S10). Our results suggest that universal masking would significantly reduce the risk of outbreaks in homeless shelters, even with 60% compliance. However, the impact of masking is highly sensitive to the assumed masking effectiveness and compliance, estimates for which still vary considerably despite accumulating evidence that masks reduce infection risk [
36,
37,
39,
98,
99]. Many uncertainties in the biology of SARS-CoV-2 transmission remain, particularly regarding differential infectiousness over time and by the severity of illness, and the relationship of PCR positivity and infectiousness [
17,
19,
67]. Our assumption of equal infectiousness for different individuals means that our model is unlikely to fully reproduce super-spreading events [
62,
63]. We made several simplifying assumptions in modeling transmission within the shelter and from the surrounding community, namely, homogenous mixing within the shelter population, no entry of new people, a stable background infection rate over time, and full immunity upon recovery from infection given the short duration of the simulation. Our assumption that individuals who are isolated within homeless shelters while awaiting test results are unable to transmit or become infected may have led to a slight overestimation of the impact of testing, since in reality isolation is not perfect. We assumed homogeneous mixing due to a lack of contact data for the shelter outbreaks, which meant that we were not able to consider cohorting and contact tracing as interventions.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.