Introduction

Homelessness and housing insecurity are major public health issues in Canada, with 3% of Canadians experiencing homelessness in their lifetime [1]. Temporary emergency housing for this population is often shared, crowded and subject to high turnover [2], all of which create environments where it is difficult or impossible to physically distance. As a result, people experiencing homelessness have been found to be at heightened risk of contracting severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in most settings [3–11], and to have increased risk for adverse outcomes following SARS-CoV-2 infection [10].

Yet, few reliable estimates of longitudinally-ascertained infection incidence exist for this population [12–13]. Instead, most research falls into one of two categories: cross-sectional analyses in specific settings (e.g., shelters) using study-administered PCR or serologic assay testing [3–5,7,14–17], or larger-scale analyses reliant on administrative registries of community-based PCR testing [10,11,18,19]. The former yielded good, but highly variable estimates, emphasizing the importance of local context and study timing. The latter group of studies had broader representativeness, but were sensitive to testing policies [20–22] and willingness of individuals to be tested [23], leading to potential underestimation of infection or unrepresentative findings [24]. Our interim report [12] on incidence of first SARS-CoV-2 infection among people experiencing homelessness showed this population had a high rate of incident infection, but also confirmed that timing does strongly influence incidence estimates and most infections were only identified through study-administered testing.

Our interim analysis did not cover the full observation period, and did not account for re-infections, which are now recognized to be extremely common in this population [25]. Re-infections are also associated with increased risk for acute and chronic complications that substantially reduce quality of life, such as post-COVID condition, and are therefore important to quantify [26–28].

In the present study, we report the incidence of SARS-CoV-2 infection and re-infections in a representative cohort of longitudinally-followed people experiencing homelessness in Toronto, Canada. Using a combination of community-based and study-administered testing, we provide the most comprehensive and contemporary estimate of infection incidence for this population, and also assessed individual and housing-related characteristics associated with new infections.

Methods

Results

Of the 1102 individuals that were randomly approached during recruitment efforts, 354 declined to participate and 12 were unable to consent (Fig 1). Of the overall 736 participants who were recruited into the study, 44 individuals refused the additional consent for linkage to ICES, and 52 were not successfully linked, resulting in a final sample of 640 participants for this analysis, or 86.9% of the overall Ku-gaa-gii pimitizi-win cohort. Supplement C, table 1 in S1 File compares characteristics of participants who were successfully linked to ICES compared to the cohort overall: individuals successfully linked were very similar to the overall cohort, except for a somewhat lower rate of refugees and persons with temporary or other legal status. A majority of participants were 30–49 years old (44.3%) or 50–69 years old (35.7%), self-reported being male (67.4%), White (48.8%), and Canadian citizens (76.6%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Ku-gaa-gii pimitizi-win recruitment, reasons for non-participation or exclusion from analysis.

https://doi.org/10.1371/journal.pone.0319296.g001

Among participants, 81 (16.7%) were lost to follow-up after baseline; 71 (11.1%) provided one follow-up interval of data; 75 (11.7%) provided two follow-up intervals of data; 112 (17.5%) provided three follow-up intervals; and 301 (47.0%) provided all four follow-up interviews. Overall, participants contributed a total of 2,401 interviews, representing 883.9 person-years of observation time pre-Omicron and 309.8 person-years of observation time post-Omicron.

SARS-CoV-2 infection and associated outcome rates pre- and post-Omicron are presented in Table 1. A total of 613 infections were detected across 2,401 intervals, of which 312 were identified during interviews before Omicron became prevalent and 301 identified during interviews after Omicron became prevalent. This represents an infection incidence rate of 35.3 (95% CI 31.6–39.4) infection events/100 person-years pre-Omicron and 97.2 (95% CI 86.8–108.8) infection events/100 person-years post-Omicron. 29.7% of infection events were re-infections, with a majority of re-infections (n = 127) occurring post-Omicron. Overall, by the end of the observation period, 419 (65.5%) participants overall had evidence of at least one SARS-CoV-2 infection; 25% of these had at least two infection events.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. SARS-CoV-2 infections among Ku-gaa-gii pimitizi-win participants, by period.

https://doi.org/10.1371/journal.pone.0319296.t001

Table 2 shows the results of unadjusted models assessing factors potentially associated with new infection. Timing of interviews was strongly associated with infection, with interviews during the post-Omicron period were more than three times more likely to reveal a new infection than earlier interviews (unadjusted rate ratio [uRR] 3.40 [95% CI 3.00–3.84]). A number of other factors were also positively associated with infection, including: Black (uRR 1.28 [95% CI 1.10–1.49]) or other racialized (uRR 1.20 [95% CI 1.01–1.43]) self-reported race category; Immigration to Canada (>10 yrs ago uRR 1.18 [95% CI 1.02–1.36]); Cancer diagnosis in the past ten years (uRR 1.38 [95% CI 1.07–1.80]); Obsessive compulsive or other personality disorder (uRR 1.55 [95% CI 1.20–2.00]); Number of SARS-CoV-2 infections at baseline (1 infection uRR 2.37 [95% CI 2.12–2.65]; 2 + infections uRR 2.00 [95% CI 1.47–2.73]); and a number of environmental factors, such as proportion of interval residing in congregate shelters (per 10% increase in time, uRR 1.04 [95% CI 1.02–1.05]) and spending their largest share of time residing in high-exposure settings (uRR 1.50 [95% CI 1.22–1.86]). COVID-19 vaccination status at baseline was not associated with infection risk.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Unadjusted modified Poisson regression with generalized estimating equations assessing each factor potentially associated with SARS-CoV-2 infection by the end of an interval (2,401 intervals overall).

https://doi.org/10.1371/journal.pone.0319296.t002

Correlation results (not shown) demonstrated that immigration was significantly correlated with race category, and most of the significant environmental factors were strongly collinear. Therefore, we selected self-reported race category and level of housing-related exposure risk to include in the multivariable model.

Table 3 presents the joint associations between selected factors and new infection at an interview. Few covariate associations were attenuated after adjustment. Interview after Omicron remained strongly associated with new infection (aRR 3.54 [95% CI 3.12–4.02]), as was history of COVID-19 infection at baseline (1 infection aRR 2.55 [95% CI 2.29–2.83]; 2 + infections aRR 2.28 [95% CI 1.80-2.89]) and spending their largest share of time residing in high- or moderate-exposure housing (High-exposure housing aRR 1.74 [95% CI 1.43–2.11]; Moderate-exposure housing 1.39 [95% CI 1.15–1.68]). Additional covariates significantly associated with new infection included self-reported race category (Black aRR 1.37 [95% CI 1.20–1.55]; Other racialized aRR 1.14 [95% CI 1.00–1.31]); Cancer diagnosis (aRR 1.58 [95% CI 1.39–1.78]); and obsessive compulsive or other personality disorder (aRR 1.69 [95% CI 1.32–2.18]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multivariable modified Poisson regression with generalized estimating equations assessing factors associated with SARS-CoV-2 incident infection by the end of an interval (2,401 intervals overall).

https://doi.org/10.1371/journal.pone.0319296.t003

Discussion

In this representative cohort of people experiencing homelessness in Toronto, Canada, 65.5% of participants had at least one SARS-CoV-2 infection by the end of observation, and more than 25% had two or more infection events. Overall, 35.3 infection events/100 person-years occurred prior to Omicron, and 97.2 events/100 person-years occurred post-Omicron. New infection risk was higher among individuals who self-identified as Black, as non-Indigenous visible minorities or as multiracial, and among people residing in housing deemed to be high- or moderate- exposure risk by virtue of it being crowded and/or congregate.

To our knowledge, only two longitudinal studies have measured SARS-CoV-2 incidence in this population using study-administered testing. One was our interim report [12], which assessed incidence of first SARS-CoV-2 infections in the first 6-months of the Ku-gaa-gii pimitizi-win study, and the other was a French cohort study [13] that determined cumulative incidence through seroprevalence measured twice, three months apart. Both of these studies did not measure incidence including re-infections, and the French study was conducted prior to the emergence of Omicron variants.

Other literature about COVID-19 infection rates among people experiencing homelessness have mostly been cross-sectional in nature [3–5,7,14–17] or leveraged administrative registries of community-based PCR testing [10,11,18,19]. The cross-sectional analyses were mostly completed in the early phases of the pandemic, were quite sensitive to study timing and local context, and those leveraging serology were, by design, unable to report on re-infections. The studies based on administrative data were, by contrast, highly sensitive to local testing policies and generally only captured a small fraction of infection events [12], which could also result in biased estimates wherever local testing was limited by policies [20–22] or individual testing-related behaviours [23]. For example, a study in our region and approximate time frame using administrative data [11] estimated a maximum of 25 infections/100 person-years in this population (during the earliest Omicron wave), which is nearly four times less than our observed Omicron-era incidence rate of 97.2 events/100 person-years. Thus, alternative methodologies have generally presented an incomplete or drastically underestimated picture of SARS-CoV-2 incidence in this population, and our findings provide the most robust and comprehensive estimate of SARS-CoV-2 incidence encompassing re-infections in a cohort of people experiencing homelessness.

Another comparison group is available using the general population, through Canadian population-based seroprevalence studies summarized by the Canadian COVID-19 Immunity Task Force [46]. By the end of October 2022 (when our study observation ended), Ontario seroprevalence estimates of infection-induced antibodies averaged around 64% (95% CI 60.9-67.0%) [46], approximating rates of prevalence of any infection in our cohort. Thus, in line with our interim report [12], differences in history of any SARS-CoV-2 infection have continued to narrow in the post-Omicron era. However, this comparison does not account for re-infections [26], which by definition cannot be measured in cross-sectional seroprevalence studies. Unfortunately, there is no appropriate comparison for our re-infection estimate, as we uniquely incorporate evidence from repeated serology testing [25]. However, our previous study [25] did estimate that 9% of the cohort had at least one re-infection identified without serology, which was nearly triple that of pooled estimates from Omicron-era re-infection studies in the general population [47]. Therefore, although people experiencing homelessness may now have similar levels of overall history of any infection, the total number of infection events strongly suggests that important disparities remain in the rates of re-infection and thus overall COVID-19 infection burden.

Our finding that infection risk was higher among individuals who self-identified as Black, as non-Indigenous visible minorities or as multiracial is consistent with existing literature [48,49], which links this disparity to structural racism, including higher exposure to the virus through work settings and/or overcrowded living environments [48], or decreased access to clinical care [49]. In our study, living environment and engaging in paid or unpaid work did not explain heightened risk in these groups. However, it is possible that the type of work-related exposure [50] (not measured in our study) may have contributed to observed differences.

Finally, the significant association between crowded and congregate housing and new infection is also consistent with previous literature [4,6,8,12,51]. While the city of Toronto made significant efforts to enhance emergency shelter system infection prevention and control measures during the pandemic [33] we still observed evidence of very high infection transmission. This suggests that such measures on their own are insufficient to mitigate against risks inherent in congregate and crowded settings, which is consistent with recent simulation and intervention work [52,53]. What did appear to reduce risk was providing lower-risk housing settings, such as physical distancing hotels (which remain “moderate-risk”, but were a significant improvement over the congregate shelters from which individuals were moved), or exiting homelessness itself (to mostly ‘low-risk’ settings like permanent or supportive housing).

Housing is clearly a modifiable risk factor, and our findings support the conclusion that reductions in crowding and separate rooms in shelters may be the most effective approach to protect individuals experiencing homelessness from infection. Unfortunately, the pandemic-related funding for hotels and similar distancing programs in Toronto were a one-time, time-limited initiative, and funding ran out shortly after the end of this study’s observation period. These programs have now mostly ceased operations [54,55] and returned to offering pre-pandemic forms of emergency shelter [55] which, despite continuation of enhanced infection prevention and control efforts, is likely to result in increased infection transmission moving forward. Given the known risks of downstream acute and chronic issues associated with SARS-CoV-2 infection and re-infection [26–28], this will continue to be a major health equity issue. All three levels of government in Canada should recognize their shared responsibility in providing safe shelter that reflects the ongoing presence and endemicity of SARS-CoV-2 (as well as other infectious diseases). In particular, policymakers should pivot funding towards the provision of specifically non-congregate shelter services, ideally with minimum space requirements to avoid overcrowding.

Conclusions

People experiencing homelessness in Toronto in 2021 and 2022 had very high SARS-CoV-2 infection incidence, and particularly high re-infection rates. Findings suggest that existing literature significantly underestimates the burden of COVID-19 infection in this population. Findings also support the conclusion that upstream structural risks and reliance on congregate shelter settings render certain unhoused individuals particularly vulnerable, with less crowded and private shelter options associated with lower infection risk.

Future efforts to prevent COVID-19 infection as well as other infectious disease transmission in this population should focus on prioritizing non-congregate and less crowded shelter systems.

Supporting information

Acknowledgments

The authors thank the Ku-gaa-gii pimitizi-win study’s research participants, without whom this study would not have been possible and who participated even while navigating numerous other priorities. We also wish to thank Survey Research Unit staff, including Olivia Spandier, Alexandra Carasco, Annika Khan, and the entire SRU interview team for undertaking recruitment and data collection; shelter and COVID-19 physical distancing shelter hotel staff for facilitating visits; Dr. Suzanne Stewart from the Waakebiness-Bryce Institute for Indigenous Health, Joe Hester from Anishnawbe Health Toronto and members of MAP’s Community Expert Group for participating in the review of this project; and finally, staff at the MAP Centre for Urban Health Solutions who contributed in numerous ways to the study.

This document used data adapted from the Statistics Canada Postal CodeOM Conversion File, which is based on data licensed from Canada Post Corporation, and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from Canada Post Corporation and Statistics Canada. Parts of this material are based on data and information compiled and provided by the Ontario Ministry of Health, Ministry of Long-term Care and Canadian Institute for Health Information. The analyses, conclusions, opinions, and statements expressed herein are those of the authors and do not reflect those of the data providers; no endorsement is intended or should be inferred.

The Ku-gaa-gii pimitizi-win study was conducted on Anishnawbe and Haudenosaunee territories. The authors acknowledge and pay respects to the Elders and the communities of past, present and future.