SARS-Cov-2 prevalence, transmission, health-related outcomes and control strategies in homeless shelters: systematic review and meta-analysis

Search strategy and selection criteria

The review protocol was registered in PROSPERO(PROSPERO registration 2020 CRD42020187033)¹⁰. The recommendations by the taskforce on guidelines for systematic reviews in health promotion and public health¹¹ were followed. A systematic search of scientific databases was conducted in EMBASE, the WHO Covid19 database, and Web of Science Core Collection. The research questions are presented in Box 1. The search strategy details are presented in supplementary material 1.

To cover unpublished research results, we searched pre-print repositories (medRxiv, bioRxiv), websites of 6 relevant institutions (see supplementary material 1) and the live map of COVID-19 evidence by the Norwegian Institute for Public Health.

A PICO framework was used to help define the inclusion/exclusion criteria: population was defined as PEH fitting the FEANTSA-ETHOS definitions; empirical quantitative, qualitative and mixed-methods studies, documenting exposure to SARS-Cov-2 or any policy measure or specific IPC strategies for PEH during the Pandemic were included; the general population was considered the standard for comparison, although a comparison group was not a criteria for inclusion; outcomes could be any health-related effect of SARS-Cov-2, including prevalence, incidence, evidence on transmission, development of disease, risk of severe course of disease, hospitalisation, ICU utilization, ventilation, mortality, etc. Effects of lockdown and other NPIs on the health status of PEH were also considered.

Empirical studies in English, indexed from December 1^st 2019 onwards, were included, and the searches were conducted on June 5^th and 9^th 2020. The table of contents of 4 relevant journals were also searched (Clinical Infectious Diseases, European Journal of Homelessness, The Lancet, Lancet Psychiatry).

Data analysis

Titles and abstracts were screened by two independent researchers applying pre-defined inclusion/exclusion criteria. After screening 20%, the review team refined the criteria for the remaining articles. Full texts were obtained for all included studies and for those where no agreement could be established based on title/abstracts alone. For all included full text articles, backward and forward citation search was conducted (PRISMA Flowchart figure 1).

• Download figure
• Open in new tab

Figure 1: PRISMA 2009 Flow Diagrams

We extracted relevant parameters from cross-sectional quantitative studies to synthesise a pooled measure of SARS-Cov-2 prevalence.

The extracted data was stratified based on the presence of an outbreak situation during data collection, either marked as such by authors or based on a prevalence of >=5%. Risk of bias and publication bias were assessed through funnel plots and regression-based statistical testing by Egger ^12,13.

Stata V16¹⁴ with commands metaprop¹⁵ and meta¹⁶ were used in analysis.

Role of the funding source

There was no funding source for this study

Prevalence of SARS-Cov2 in homeless shelters

We found 9 quantitative articles reporting PCR-based SARS-Cov-2 prevalence estimates in PEH and 6 studies reporting prevalence estimates for shelter staff members. To adjust for the varying settings and contexts, we pooled the measures in a random-effects meta-analysis model for PEH (n=7107) and staff (n=1375). To consider the difference in nature of investigating an outbreak^{18,19,22,23,25,27} and testing for baseline estimation^{18,20,21,26,27} in our meta-analysis, we sub-pooled the measures accordingly.

For PEH, the pooled SARS-Cov-2 prevalence estimate was 29·49% (95% Confidence Interval, 95%CI=16·44%-42·55%, I²=97·79%, p=0·00) in the context of outbreak situations. As a sensitivity analysis, we removed an outlier that could reflect a super-spreading event (Imbert et al.,²² 67·33%, 95%CI=59·48%-74·32%), resulting in a pooled prevalence of 23·32% (95%CI=12·12%-34·53%, I²=96·79%, p=0·00). For baseline estimation, the pooled prevalence was 2·14% (95%CI=1·02%-3·27%, I²=83·80%, p=0·00) (figure 2).

• Download figure
• Open in new tab

Figure 2: Forest plot of SARS-Cov-2 prevalence pooled by outbreak situation and in total for PEH

For staff members, the random-effect meta-analysis resulted in a pooled prevalence of 15·18% (95%CI=8·95%-21·42%, I²=65·76%, p=0·02) during an outbreak situation and 1·72% (95%CI=0·31%-3·12%)) for baseline estimation (figure 3).

• Download figure
• Open in new tab

Figure 3: Forest plot of SARS-Cov-2 prevalence pooled by outbreak situation and in total for staff

Transmission among PEH

While all studies tested PEH in homeless shelters, one study by Yoon et al.²¹ conducted additional testing of clients living unsheltered during homeless outreach service events (e.g. meal services). This resulted in a prevalence of 0·5% (3 out of 636) in the unsheltered PEH.

Clinical and health-related outcomes among PEH

Five articles^{19,21,24,25,27} presented data on clinical outcomes of PEH, ranging from respiratory symptoms to previous mental health diagnosis and other somatic issues (e.g. chronic kidney or liver disease), and one article²² provided data on emergency department visits, hospitalization and mortality. Four articles^19,21,22,25 provided data specifically for PEH testing positive: specific symptoms (cough, shortness of breath, fever) were either uncommon or did not differ significantly from the overall tested population of PEH. Neither did comorbidities²⁵. Imbert et al.²² used a city-wide administrative database and reported 12% (n=12) of those tested positive to have had a treatment and release emergency department visit, 8% (n=8) a hospitalization and 1% (n=1) died.

Comparing to RT-PCR, Yoon et al. calculated sensitivity and specificity of symptom screening, stating that fever, cough, or shortness of breath was 14% sensitive and 89% specific for identifying COVID-19 cases when reported in the previous day, changing to 24% and 85% when reported in the previous week. An expansion to any symptom reported resulted in no improvement of sensitivity, but a loss in specificity (last week: 81%; last day: 87%).

Two studies modelled different health-related outcomes among PEH with one modelling conducted for England²⁸ and one for the USA²⁹.

Lewer et al. (2020)²⁸ aimed to estimate the impact of COVID-19 on the population of PEH of England and estimate the reduction in mortality and healthcare use that could result from a targeted intervention compared with a “do nothing” scenario.The authors used a Markov chain model with a modified susceptible-exposed-infectious-recovered (SEIR) model, assuming a basic reproductive rate of 1·4 in the baseline scenario, a latency, incubation and infectiousness period of 2·5 days, 5 days, and 6·5 days respectively, and infection fatality ratio for patients of 4.8% for those homeless defined as vulnerable (37%) and 0·6% for the remaining population. For the latter scenario, the authors estimated that 15448 cases of COVID-19 could occur among the 35817 homeless and 10748 people sleeping rough in England, resulting in a cumulative incidence of 34%.

This incidence could be reduced to 21% (i.e. 9934 cases) with a targeted intervention by the housing and health authorities in England. The authors also estimated a total of 364 death if nothing was implemented for PEH in England but a reduction to 200 deaths with the intervention.

The report by Culhane et al (2020)²⁹ aimed to estimate the mortality, hospitalisation costs and system capacity to manage the impact of COVID-19 pandemic on the existing population of PEH in the USA. Based on data on the age distribution of PEH from two large cities (Los Angeles and New York) and on studies that older homeless populations bear health risks comparable to those of individuals 15-20 years old, the authors estimated that there were 4·3% PEH across the USA (corresponding to 21295 people, ranging from 2·4% to 10·3%) that could require hospitalization assuming an infection rate reaching 40%. They also estimated that critical care needs could range from 0·6% to 4·2% for the homeless and fatality rates could range from 0·3% to 1·9% (central estimate 0·7% or 3454 deaths).

Non-pharmaceutical interventions (NPI)

We could not identify articles quantifying the effects of NPIs on PEH. However, one qualitative study³⁰ conducted unstructured interviews through telephone or social media conversations with 35 participants from a US cohort study of criminal justice-involved women to explore how they were faring during this period. Thirteen of the interviewees reported some form of homelessness, and for some of the women their unstable living situations prevented them from being able to follow the social distancing recommended.

Infection protection and control (IPC) strategies in homeless shelters

We identified no study quantifying the effectiveness of strategies for IPC among PEH. Five papers^{23,25,26,28,29} discussed strategies for IPC more generally, and one study²¹ actively investigated these at 9 shelters. Yoon et al.²¹ describe different pandemic preparation measures, e.g., 55·6% of shelters (5 in 9) discontinued taking in new clients, 77·8% (7 in 9) created separate isolation areas for PEH with suspected infection and 55·6% (5 in 9) increased the spacing between individual beds.

Across all studies, the major strategies discussed as important and necessary were comprehensive and rapid testing, expansion of non-congregate supportive housing for PEH and, specific strategies for IPC to be taken in congregate settings, e.g. increased spacing between beds and/or limited staff rotation. O’Shea et al. further discussed potential benefits in collaborating with regional laboratories for quick turnaround of test results which enable the isolation of the infected.

Furthermore, Henwood et al²⁴ reported that 39% (207 of 532) of the tenants of supportive housing in Los Angeles considered themselves to be at serious risk for COVID-19 due to a pre-existing condition, 59% (316 of 532) had a registered mental health diagnosis and 55·3% (294 of 532) believed they can shelter in place if needed.

Funnel plots of studies reporting SARS-Cov-2 prevalence and corresponding standard errors are presented in figure 4. The Egger test suggests absence of significant publication bias in all strata except for the overall prevalence of PEH, which disappears after stratification by outbreak situation (supplementary material 3), supporting differentiation between baseline estimation and outbreak investigation.

Risk of bias assessment is presented in table 2.

Quality of Evidence

Studies providing prevalence estimates were based on PCR testing and some included testing of staff for a complete picture of SARS-Cov-2 infection in shelters. Overall, the quality of evidence was low. Most articles provide little detail on how and why shelters were sampled. Studies may have missed infected patients who do not have enough viral particles yet to be detected in PCR. Only two articles conducted follow-up testing^21,23. The recruitment of participants was based on accommodation in homeless shelters with no systematic surveying of housing situation and residence type in shelter. This may be of special importance considering the diverse background of PEH and their everyday life, because not all shelters offer accommodation during the day. Finally, all identified studies were based in high-income countries raising questions of generalizability especially to middle- and lower-income countries that differ in population demographics, cultural and social dimensions. Further details about the risk of bias are provided as supplementary material 3.

Strengths and limitations

Besides major databases, we searched pre-print repositories and websites of relevant institutions for unpublished manuscripts and consulted the Covid-19 evidence live map of the Norwegian Institute of Public Health. Further, we have used forward and backward reference screening on all identified articles and screened the TOCs of journals deemed relevant for potential publications during the screening, extraction, and analysis phases. All steps of the review have been conducted by two reviewers for quality assurance. Finally, we tested for publication bias, both visually through funnel plotting as well as statistically based on Egger^12,13. Additionally, danger of publication bias might be low as we are investigating prevalence (compared to measures of effect) during the onset of a pandemic of global relevance.

We have used English search terms only, thus might have missed relevant studies in other languages. This could explain why only one study from a non-English language country was identified. Our findings may not be generalizable to other settings than the ones where the studies were conducted, as droplet or potentially airborne transmission may be different in other contexts of social and cultural communication and interaction. Third, even though we searched several databases and repositories, we might not have detected all relevant studies considering the amount of evidence being generated during the current pandemic. Additionally, we cannot rule out the possibility of publication bias, particularly regarding the baseline prevalence among staff since estimates for staff were not part of these studies’ initial objectives.