Introduction

Housing is a key determinant of health and is modifiable by public health and other policy interventions [1]. Despite such policy opportunity, regions across the United States (US) and Canada have faced a worsening housing crisis in recent years. While there is no universal definition of homelessness, the United Nations and other international and national organizations tend to conceptualize it along a continuum of situations characterized by a lack of stable, safe and adequate housing, ranging from living in public spaces or in places not intended for human habitation (i.e., unsheltered homelessness) to residing in emergency shelters, other temporary accommodations or housing that fails to meet public health and safety standards (i.e., sheltered homelessness) [2,3]. Between 2023 and 2024, the US had an 18% increase in the number of people experiencing homelessness (including sheltered and unsheltered) [4]. In the Greater Vancouver region in Canada, the number of individuals experiencing homelessness increased by 9% between 2023 and 2025 [5]. The city of Vancouver within this region experienced a slightly higher increase (12%) during this period, with 2,715 individuals identified as homeless in 2025 [5]. These numbers were generated using a point-in-time count method, which is known to result in undercounts of actual homelessness [5,6].

As a strategy to manage the growth of survival homeless encampments resulting from the housing crisis, many jurisdictions across the US and Canada increasingly forcibly displace unhoused people from public spaces [7–10]. Importantly, personal belongings, including those that are integral for survival (e.g., clothes, food, the structures of their homes, identification cards, and medications), are routinely confiscated during these processes [7,11,12]. Previous research has identified a number of severe health and safety harms associated with these practices (often called “forced/involuntary displacement” and “encampment/street sweeps”), including acute injuries, loss of survival items, intensified substance use and overdose, violent victimization, worsening mental health, and disconnection from health and social services [7,11–18]. As rates of substance use are known to be disproportionately high among unhoused people [19], a previous simulation modeling study involving 23 US cities over a 10-year period estimated that continual involuntary displacement might contribute to 16%–24% additional deaths among unhoused people who inject drugs [20]. Despite these immense public health harms, encampment sweeps continue to be endorsed and employed by multiple levels of government [21,22].

Previous qualitative research has indicated that encampment sweeps may undermine pathways to housing among unhoused people [8,9,11,13], yet few quantitative studies have investigated this issue at a population level. A previous audit report of the New York City (NYC) Comptroller documented that only 90 out of 2,308 unhoused people who were forcibly displaced by the NYC Department of Homeless Services in 2022 stayed in shelter for more than 1 day, and only three of them secured permanent housing thereafter [23]. However, this report also pointed to the inadequate tracking of placement referrals as being a significant limitation to their findings [23]. Therefore, we sought to leverage our longitudinal data generated from ongoing prospective cohort studies of community-recruited people who use drugs in Vancouver to examine if: (1) having one’s personal belongings confiscated; and (2) experiencing forced displacement; are associated with transitioning from homelessness during study follow-up among unhoused participants.

Methods

Statistical analysis

We first examined the baseline sample characteristics stratified by experiencing confiscation of belongings, using the Kruskal–Wallis test (for age) and the exact chi-squared test (for all other variables). We generated a Kaplan–Meier curve for each sample for the analyses of the primary and secondary explanatory variables to visualize the probability of the event (no longer reporting homelessness) over time. While we did not collect data on the length of homelessness or the number of times participants experienced homelessness, among sub-samples of participants who completed a semi-annual interview prior to the baseline assessment for the current study, we counted the number of participants who reported incident homelessness in the past 6 months at baseline (i.e., not reporting homelessness in the past 6 months in a semi-annual interview preceding to the study baseline), the number and percentage of participants who experienced the event, and the median length of time to event.

Next, for each of the primary and secondary explanatory variables, we fit a proportional hazard marginal structural model (MSM) with inverse probability of treatment weighting and inverse probability of censoring weighting for each semi-annual follow-up period, hypothesizing that time-dependent confounding variables might have been affected by past exposure to confiscation of belongings and forced displacement [28]. Figure 1 is a directed acyclic graph depicting the hypothesized relationships between the time-varying exposure (confiscation of belongings or forced displacement), the time-invariant and time-varying confounders, and the event (no longer reporting homelessness in the past 6 months during study follow-up). The proportional hazard MSM used the same covariates in the inverse probability of censoring weighting as in the inverse probability of treatment weighting as per the standard approach [28]. Event times were assumed to equal the midpoint of the last visit reporting a nonevent and the first visit reporting that an event had occurred. Visit timing was typically quite variable within each 6-month follow-up period, and occasionally visits were missed (13%), but models accounted for the number of days between visits as time-to-event models typically do. We analyzed descriptive statistics from sub-questions related to the types of housing participants were placed in, and whether they were offered housing at the time of forced displacement.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Directed acyclic graph depicting the hypothesized relationships among the exposure, the confounders, and the event.

t0: Cohort enrollment; t1: First interview during the study period (baseline); t2: First follow-up interview during the study period; E: confiscation of belongings or forced displacement in the past 6 months; C: time-varying confounders; Y: no longer reporting homelessness in the past 6 months during study follow-up.

https://doi.org/10.1371/journal.pmed.1005033.g001

In addition, we conducted several sensitivity analyses. First, we fit a multivariable extended Cox regression model for each of the primary and secondary explanatory variables, including all potential confounders as covariates. We conducted this extended Cox regression analysis to address potentially high variance in the results of proportional hazard MSM due to relatively small sample sizes and potentially unstable weights. Second, to address the possibility of reverse causation, we lagged the primary and secondary explanatory variables to one immediately prior 6-month study follow-up cycle and re-ran the proportional hazard MSM. Third, we evaluated the potential effect modification by gender or ethnicity/ancestry by creating and adding interaction terms between gender or ethnicity/ancestry and the primary or secondary explanatory variables to the Cox regression models. Fourth, we examined the potential impacts of seasonality by adding a seasonality variable based on calendar quarters of interview dates to the Cox regression models. All p values were two-sided. All analyses were performed using SAS 9.4 (Cary, North Carolina, USA).

We applied an a priori-defined statistical protocol involving the proportional hazard MSM and Cox regression models to estimate the independent effects of the primary and secondary explanatory variables on the event. In response to the journal’s editor’s and peer reviewers’ suggestions, we have added or updated the following components to our analyses: (1) adding history of incarceration as a covariate to the analyses; (2) adding Kaplan-Meier curves; (3) adding available descriptive data on incident homelessness; (4) examining the potential effect modification by gender or ethnicity/ancestry; and (5) examining the potential impacts of seasonality.

Results

Figure 2 illustrates how we derived our analytic samples. As shown, 235 participants were eligible for the analysis of the primary explanatory variable (confiscation of belongings) and followed for 1.2 years on average (standard deviation [SD]: 0.8, interquartile range [IQR]: 0.5, 1.5), contributing 230.3 person-years of observation. A Kaplan–Meier curve and the number of participants at risk at 6-month intervals are shown in Fig 3. At baseline, the median age was 43 (interquartile range: 37–53) years, and 130 (55.3%) self-identified as men, 98 (41.7%) as women, and 7 (3.0%) as transgender or nonbinary. Furthermore, 121 (51.5%) participants self-identified as white, 103 (43.8%) as Indigenous, and 11 (4.7%) as other persons of color. The majority (n = 221, 94.4%) resided in the city of Vancouver in the past 6 months; of those, 190 (86.0% or 81.2% of the entire sample) lived in DTES. At baseline, 59 (25.1%) reported having their belongings confiscated in the past 6 months. Other baseline sample characteristics are presented in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Baseline sample characteristics stratified by experiencing municipal government-enforced confiscation of belongings among VIDUS/ACCESS participants who reported homelessness in the past 6 months at baseline in Vancouver, Canada, 2021−25 (n = 235).

https://doi.org/10.1371/journal.pmed.1005033.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Flowchart for the determination of analytic samples.

ACCESS: The AIDS Care Cohort to evaluate Exposure to Survival Services Study. VIDUS: The Vancouver Injection Drug Users Study.

https://doi.org/10.1371/journal.pmed.1005033.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Kaplan–Meier curves for the analyses for the primary and secondary explanatory variables.

https://doi.org/10.1371/journal.pmed.1005033.g003

During follow-up, 128 (54.5%) participants stopped reporting homelessness in the past 6 months, corresponding to an incidence rate of 55.6 (95% confidence interval [CI]: 46.4, 66.1) cases per 100 person-years. Among these 128 participants, 127 (99.2%) described the types of housing they were living in at the time of interview when they stopped reporting homelessness. The most commonly reported type of housing was SROs (n = 89, 70.1%), followed by apartment or house (n = 30, 23.6%), modular housing (n = 6, 4.7%), and treatment/recovery house (n = 1, 0.8%). One participant (0.8%) self-reported supportive housing as a type of housing that they resided in. Among the 235 participants, 111 (47.2%) individuals completed a semi-annual interview prior to the study’s baseline assessment. Among them, 87 (78.4%) individuals were identified as incident homelessness in the past 6 months at baseline. Among the 87 participants, 65 (74.7%) stopped reporting homelessness in the past 6 months during study follow-up, after a median of 0.6 year (IQR: 0.5, 1.1).

For the analysis of the secondary explanatory variable (forced displacement), the sample consisted of 154 participants, who were followed for 0.9 years on average (SD: 0.4, IQR: 0.5, 1.3), contributing 123.5 person-years. A Kaplan–Meier curve and the number of participants at risk at 6-month intervals are shown in Fig 3. During follow-up, 62 (40.3%) stopped reporting homelessness in the past 6 months, corresponding to an incidence rate of 50.2 (95% CI: 38.5, 64.3) cases per 100 person-years. Among the 154 participants, 79 (51.3%) individuals completed a semi-annual interview prior to the study’s baseline assessment. Among them, 40 (50.6%) individuals were identified as incident homelessness in the past 6 months at baseline, Among the 40 participants, 34 (85.0%) stopped reporting homelessness in the past 6 months during study follow-up, after a median of 1.0 year (IQR: 0.5, 1.7).

At baseline, 50 (32.5%) reported forced displacement in the past 6 months. During the study period, there were 121 reports/observations of forced displacement in the past 6 months. Among them, 119 (98.4%) reported that participants were not offered housing at the time of displacement, while two (1.7%) did receive an offer of housing. Among these two, one was able to successfully move into the housing offered but the other was not because “there was a line-up/no space,” which was a pre-listed response option for the question.

In the proportional hazard MSM, both confiscation of belongings and forced displacement remained negatively associated with no longer reporting homelessness in the past 6 months: adjusted hazard ratio (AHR) of 0.35 (95% CI [0.13, 0.94]; p = 0.036) and AHR of 0.07 (95% CI [0.02, 0.20]; p < 0.001), respectively. In the multivariable extended Cox regression analyses, the results were largely unchanged (see Table 2). In the sensitivity analyses where the primary and secondary explanatory variables were lagged to one immediately prior 6-month study follow-up cycle, both variables lost the statistical significance; however, the direction of the associations remained the same as in the primary analyses: AHR of 0.73 (95% CI [0.29, 1.83]; p = 0.497) for confiscation of belongings and AHR of 0.50 (95% CI [0.14, 1.70]; p = 0.264) for forced displacement. None of the interaction terms between gender or ethnicity/ancestry and the primary or secondary explanatory variables were statistically significant (all p > 0.3). The seasonality variable was not statistically significant (all p > 0.3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Bivariate and multivariable extended Cox regression analyses of securing housing among VIDUS/ACCESS participants who reported homelessness in the past 6 months at baseline in Vancouver, Canada, 2021−25.

https://doi.org/10.1371/journal.pmed.1005033.t002

Discussion

In our cohort of unhoused people who use drugs in Vancouver, one in four reported having their belongings confiscated by city engineering workers, park rangers and/or police officers while one in three experienced municipal government-enforced forced displacement in the past 6 months at baseline. After adjusting for potential confounders, these experiences were associated with significantly lower rates of transitioning from homelessness. Among those who experienced forced displacement, the vast majority (98%) were not offered housing at the time of displacement.

Our findings corroborate previous reports based on qualitative evidence, indicating that encampment sweeps interfere with broad efforts for housing placements [8,9,11,13]. These reports suggested that interference with housing placements could occur in multiple ways, including destruction or loss of vital personal documents (e.g., identification cards) that may be required to sign a lease or join a waitlist, disrupting connections with community and services, worsening mental health, and undermining trust with public service providers, including housing service workers, as forced displacement is implemented by public workers [8,9,11,13]. Indeed, in our previous study on municipal government-enforced confiscation of belongings among unstably housed people who use drugs in the same two cohorts employed by the current study, we found that 36% of participants who had their belongings confiscated by city workers in the past 6 months reported that they tried but were unable to access housing services and 16% reported being unable to access substance use treatment in the same 6-month period [7]. These findings were mirrored in another recent cross-sectional study conducted in Massachusetts, the United States [18] and collectively demonstrate that displacement can impede access to these essential services, which in turn exacerbates the housing outcomes. Furthermore, we found that municipal government workers who forcibly removed unhoused people from public spaces almost never offered housing services to those who were displaced. These findings are consistent with the NYC Comptroller’s audit report that the number of people who were eventually housed after encampment sweeps in 2022 was only three out of 2,308 (0.1%) [23]. Taken together with previous research on the negative health impacts of street sweeps [7,12,14,15,20], our findings indicate that encampment sweeps not only increase health and safety harms, but also appear to perpetuate homelessness.

Our findings add detailed prospective evidence to recent calls for alternative policy approaches to encampment sweeps, centering on access to permanent, dignified, and affordable housing. These calls have been made by health and human rights authorities and experts in the US and Canada [9,21,29], including the American Public Health Association [8] and the Canadian Federal Housing Advocate [22], as well as those directly impacted by street sweeps in Vancouver [11]. Furthermore, amid large increases in police budgets [30], the continuation of costly forced displacement can come with an opportunity cost in the form of cuts to crucial social services and/or the acquisition or development of adequate, dignified, and affordable housing options [31].

In the province of British Columbia (B.C.), where our study setting of Vancouver is located, most encampment sweeps are carried out by municipal governments, which have legal authority over public spaces and local bylaws. The provincial government’s official stance has been to accept encampment sweeps when alternative shelter space is made reasonably available to people living in encampment [32], though it is widely known that such shelter space has never been sufficiently available [33,34]. Our findings that alternative shelter space that is reasonably available has never been offered to our study participants who were forcibly displaced demonstrate the highly problematic nature of ongoing encampment sweeps tolerated by the provincial government. Moreover, while Canada’s stock of nonmarket housing is much lower compared to the average in other countries in the Organisation for Economic Co-operation and Development (OECD) (3.5% versus 7%), Vancouver is one of the settings within Canada that suffer the most acutely from the shortages of affordable housing [35]. A report published in September 2025 estimated the need for $10.1-19.3 billion in investments to address the unmet housing needs (i.e., 29,250−54,500 affordable rental housing units) in the Metro Vancouver region over the next 5 years while only $1.2 billion was invested to create 12,500−19,500 affordable housing units in the last 5 years [35]. In addition to the severe shortages of affordable housing, since 2023 the province and the city of Vancouver have introduced several key policy changes and decisions that have limited the growth of government-subsidized, shelter-rate housing supply in Vancouver [36,37] and removed some key tenancy rights in supportive housing in B.C. [38–41], with the latter potentially increasing eviction rates. Given the ongoing encampment sweeps, our findings raise concern that these policy decisions might collectively exacerbate the structural conditions that lead to poor housing outcomes among those experiencing forced encampment sweeps.

A crucial limitation to our study is that we did not capture or assess the quality of housing that participants were placed in. In fact, the majority (70%) of participants reported ending up in SROs, which are known to have poor conditions and high rates of unlawful evictions [25,42,43]. The conditions within SROs and other government-subsidized housing have been worsening amid a dwindling supply of shelter spaces or otherwise affordable housing options across Vancouver [37]. The 2025 point-in-time homeless count report in the Greater Vancouver region also documented that 42% of respondents attributed their current state of homelessness to residential evictions [5]. Therefore, while rates of transitioning from homelessness may appear high in our study, participants continue to be at high risk of returning to homelessness after being allocated an SRO room. Future research should establish more ideal endpoints, including permanent, secure, and dignified housing.

This study has additional limitations. First, a nonrandom sample limits the generalizability of our findings. In particular, unhoused people who were the most severely impacted by encampment sweeps might have also experienced challenges in participating in our research. Therefore, the rates of transitioning from homelessness to being housed in our study might have been overestimated. Nevertheless, a cohort study of community-recruited unhoused people that ascertains exposure to and outcomes of encampment sweeps in a longitudinal manner like ours is rare. Second, self-reported data may be influenced by reporting bias, although reasonable validity of self-reports has been demonstrated in a similar drug-using population [44]. Third, as with any observational research, our findings might have been influenced by unmeasured confounding. Thus, causal inference is limited though we employed proportional hazard MSM to minimize the impacts of time-dependent confounding. In addition, we cannot exclude the possibility of reverse causation as the results of sensitivity analyses where the primary and secondary explanatory variables were lagged to one immediately prior 6-month study follow-up cycle lost the statistical significance. However, as the direction of the associations was consistent with those in the primary analyses, it might be that negative impacts of encampment sweeps on housing outcomes are more acute during the initial few months after experiencing encampment sweeps compared to at least 6 months later.

In sum, we found a high prevalence of experiencing municipal government-enforced confiscation of belongings and forced displacement among our cohort of unhoused people who use drugs in Vancouver. These experiences were associated with significantly lower rates of transitioning from homelessness. Our findings underscore the need for policy approaches that prioritize access to permanent, dignified, and affordable housing rather than practices that perpetuate homelessness.

Supporting information

Acknowledgments

The VIDUS and ACCESS cohorts take place on the unceded (i.e., never legally ceded or given up to the Crown) territory of the Coast Salish Peoples, including the territories of the xwməθkwəy̓əm (Musqueam), Skwxwú7mesh (Squamish), Stó:lō and Səl̓ílwətaʔ/Selilwitulh (Tsleil-Waututh) Nations. The authors thank the study participants for their contribution to the research, as well as current and past researchers and staff.