Introduction

Homelessness in Canada is primarily experienced by men, yet data show that over one-third of the homeless people in this country are women and close to 90% of the single parents who are homeless are women [1–3]. Mental health and substance use disproportionately affect women experiencing homelessness. Between 20–50% experience major depressive disorder and major depressive episodes as demonstrated in systematic reviews [4], between 30–40% experience post-traumatic stress disorder [5, 6], as many as one in five report moderate-to-high levels of suicidality [6], and a large proportion experience exceptional vulnerability to an array of health and social risks [6–11], including premature mortality [12]. A recent meta-analysis of observational studies in high-income countries reports that three-quarter of homeless individuals have a mental illness diagnosis with no differences by gender [13].

Women who are homeless have different health and social vulnerabilities compared to homeless men. They have higher rates of distress including suicidal ideation [14], greater experiences of trauma prior to and during episodes of homelessness including more adverse childhood events (ACEs) such as physical and sexual abuse and trauma-related disorders [6, 15–22], few options for effective woman-centred services and treatments for these traumatic events and other ailments, all of which predispose them to homelessness [14, 23–27]. They are also substantially more likely than homeless men to engage in survival sex work and are thus more vulnerable to exploitation, violence, infectious disease, gynecological problems and unwanted pregnancies [6, 14]. Homeless women with children express considerable traumatic distress because of frequent child apprehensions [28]. Because women who are homeless often seek shelter in informal settings such as couch surfing or sleeping rough, they are less visible and have less access to services [14, 29, 30] and have received less attention in research and policy and practice.

Program planners and policy makers operate in a context of too little gender-specific data on programs and interventions and weak studies addressing the experiences of women [3, 4, 31]. As such, there have been calls for more research that moves away from gender-neutral approaches (i.e., examining men’s and women’s experiences with the same lens, as if there are no gender differences) [14, 31–33] towards more gender-sensitive frameworks, including women-only analyses that capture the variability around the unique challenges women who are homeless face and accelerating evidence-informed practice with this population [3, 29, 34, 35]. For interventions, there is a need for studies with large enough samples to examine whether and how programs work for women [4, 36–41].

Currently, with some exceptions [42], what we do know about housing interventions for women experiencing homelessness comes from small subsamples in longitudinal observational studies [34, 43], randomized trial pilot [44], or qualitative evaluations [32, 45]. A large multi-site supportive housing demonstration project for chronically homeless adults in 11 US communities, with 173 women as part of the sample, reported substantial improvements in housing stability but no gender differences in this and other service utilization or clinical outcomes over two years of follow-up [46]. In an observational study of an integrated service system initiative in 18 sites across the US, women (N = 2727) demonstrated greater improvements in family relationships and social support compared to men (N = 4502) over 18-months of follow-up [42].

Housing First, a supportive housing approach with demonstrated success in transitioning individuals experiencing chronic homelessness and mental illness into permanent housing, has negligible evidence on outcomes for women [47–50]. The Housing First model is based on harm reduction principles of choice and self-determination, and provides immediate access to scattered-site housing through the provision of rent supplements and flexible recovery-oriented supports [49, 51]. Participation in Housing First is not contingent on participation in services or requirements of ‘housing readiness’ (e.g., sobriety or medication compliance) [49].

The At Home/Chez Soi randomized trial conducted in five cities across Canada provides an unprecedented opportunity to examine the impact of a housing first initiative among women for several reasons. First, it is the largest randomized trial of a Housing First initiative and includes an unprecedented sample of several hundred women. This large sample size facilitates a woman only analysis to focus on variability within this subsample and factors that are relevant for women living with homelessness such as sources of trauma or parental roles. A woman only analysis, too rarely seen in studies of homelessness, ensures that the focus remains on women within the sample as gender analyses largely focus on comparing experiences with those of men. With 24 months of follow-up and novel risk variables not previously examined in a Housing First analysis we can provide unique point-in-time and longitudinal evidence not previously considered because of prior sample size and variable availability limitations.

The present study employs the largest longitudinal sample of women experiencing homelessness and mental illness in a Housing First randomized controlled trial (RCT) and extends the follow up duration typically seen to 24 months. Our objectives were: (i) To compare change at the end of the study in health and social outcomes between women who received and did not receive the Housing First intervention; ii) To examine whether the Housing First intervention leads to better housing stability over 24 months, and health and social and outcomes between 6- to 24-months post enrolment; and, (iii) To explore baseline demographic, social, health measures, and ACEs as predictors of key study outcomes during the study period.

Methods

Methods and measures for this analysis have been extensively described elsewhere [48, 52–54]. Briefly, the Canadian At Home/Chez Soi five city RCT randomized Housing First among adults experiencing homelessness and serious mental illness [53]. Participants randomized to the Housing First (HF) intervention received scattered-site housing and either moderate or high needs-based services to support their recovery and integration into the community (i.e., Intensive Case Management or Assertive Community Treatment). Participants randomized to treatment-as-usual (TAU) had access to housing and services through other community programs. Each site had additional focus on a unique population (e.g., ethno-racial populations in Toronto and Indigenous populations in Winnipeg) and had specialized services to address their needs. Fidelity assessments ensured strong adherence to program theory of change [55]. Information about the study, such as the randomization process and a description of the intervention, was provided to all study participants prior to enrolment. An informed consent process was created to ensure that participants had the time and information to understand the complexities of the trial and follow-up activities. All participants provided written informed consent prior to enrolment. All enrolment and consent documents were harmonized across sites, however, each site had separate approval from relevant university/site Research Ethics Boards.

Over 24 months, participants were interviewed every 3 months to assess housing stability and every 6 months to evaluate health and social outcomes. Previous results from 2,148 participants (including 683 women) demonstrated that those in the HF arm showed positive improvements in housing stability, quality of life, and community functioning compared to participants in usual care [46, 48, 52]. This paper reports on an exploratory analysis of 653 participants who self-identified as women and had at least one follow-up after their baseline assessment (96%), including 588 followed to 24 months. Across all sites, 374 participants were from the HF group and 279 from the TAU group (S1 Fig in S1 File).

Results

We took advantage of the rich data collected in this study and present a broad array of participant demographic, social, and health characteristics at enrolment (Table 1). Fifteen percent of the women were born outside of Canada (median time in Canada: TAU = 25yrs; HF = 21yrs). Three percent of women were employed at baseline and 90% received social assistance of some type as a source of income. Around 40% (n = 274) reported having minor children, and of these, 22% (n = 60) were providing full or partial financial support to the child(ren). The most prevalent mental illnesses were depressive episodes, PTSD, and alcohol or substance dependence. Approximately 40% had moderate-to-high suicidality, around 40% had post-traumatic stress disorder, about 60% reported ≥5 ACEs, and the majority had three or more co-morbid medical conditions.

Download:

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

Table 1. Baseline characteristics of women participants in the Canadian At Home/Chez Soi study by randomization group (N = 653).

https://doi.org/10.1371/journal.pone.0277074.t001

Table 2 displays results from models examining changes from baseline to 24 months for those in HF and TAU arms for each of our social and well-being outcomes, and comparing groups with respect to housing stability over the 24-month period. Over the 24-month follow-up period, substantial improvements were observed for all the outcomes for both HF and TAU participants. Of note, mean total QoLI-20 scores improved 16.0 points (95%CI = 13.6–18.4) and 13.4 points (95%CI = 10.6–16.2) in the HF and TAU groups, respectively, and number of ED visits decreased by approximately 60% in both groups.

Download:

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

Table 2. Post-imputation pooled change in health and social measures from baseline to 24 months among women living with mental illness and homelessness, adjusted for study city, and need level (ACT or ICM).

https://doi.org/10.1371/journal.pone.0277074.t002

These improvements over the 24-month period are more clearly illustrated in Fig 1 for the HF intervention and TAU groups for all social and health unadjusted outcomes at each observation point from baseline to 24-months. For all outcomes, except CIS-Physical where scores remained flat throughout the follow-up, both groups experienced similar significant change from baseline to 24-months (Fig 1).

Download:

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

Fig 1. Health and social outcomes among women who are homeless and mentally ill, unadjusted means and 95% confidence intervals over 24 months of follow-up by study time point and treatment group.

Note. N = 653. CSI = Colorado Symptom Index; MCAS = Multnomah Community Ability Scale; Lehman Quality of Life Interview 20 = QoLI-20; CIS = Community Integration Scale; GAIN-SS = Global Assessment of Individual Needs Short Screener.

https://doi.org/10.1371/journal.pone.0277074.g001

The mean percentage of days spent stably housed during follow-up for women receiving HF was 74.8% (95%CI = 71.7%–77.8%) compared with 37.9% (95%CI = 34.4%–41.3%) TAU, a difference of 36.9% (95%CI = 32.4%–41.4%, p<0.001). In addition, the odds of stable housing during follow-up was over 5 times higher among those in the HF group than in the TAU group (OR = 5.09, 95% CI = 4.08–6.35, p<0.001). Table 3 displays changes over time in greater detail through results of the models to assessing whether social and health outcomes were better in HF vs. TAU groups at each follow-up point. For condition-specific quality of life, significant improvements of 7.1 and 6.1 points were observed in the HF group at the 6- and 12-month follow-ups, respectively. The HF group also showed significant improvements in psychological community integration by 0.7 points at the 6-month follow-up. However, in the last year of the trial, the ED visits fell faster for the TAU group (0.6 visits) compared to the HF group (1.1 visits) at 18 months, with a rate ratio of 1.81 (95%CI = 1.26–2.60).

Download:

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

Table 3. Post-imputation pooled comparisons between housing first and treatment as usual at 6, 12, 18 and 24 months for women in the At Home/Chez Soi trial adjusted for study city, need level (ICM or ACT) and baseline outcome values^f.

https://doi.org/10.1371/journal.pone.0277074.t003

Table 4 no longer focuses on differences between HF and TAU as results presented here identify predictors of our outcomes. Moderate to high suicidality had the most consistent association to poorer health and service utilization outcomes at 24 months follow-up. Compared to those with no or low suicidality at baseline, moderate-to-high suicidality at baseline was associated with nearly three points on the CSI indicating greater mental health symptom severity (beta = 2.85, 95%CI = 1.59–4.11, p<0.001), QoLI-20 mean scores were almost four points lower indicating reduced quality of life (beta = -3.99, 95%CI = -6.49 to -1.49, p = 0.002), and almost 1.5 more ED visits (rate ratio = 1.44, 95%CI = 1.10–1.87, p = 0.007). Low education was also associated with several outcomes. Compared to those with higher education, women with less than a high school level of education at baseline predicted lower community functioning (MCAS) scores (beta = -1.32, 95%CI = -2.27 to -0.37, p = 0.006), 15% fewer social activities (CIS) in the previous month (rate ratio = 0.85, 95%CI = 0.77–0.93, p<0.001), and more severe substance-related problems at 24 months follow-up (rate ratio = 1.27, 95%CI = 1.06–1.52, p = 0.009). Having been homeless for ≥3 years at baseline was associated with lower percentage days stably housed at 24-months (beta = -6.1%, 95%CI = -11.0% to -1.3%, p = 0.013; OR = 0.74, 95% CI 0.58–0.94, p = 0.0132). An ACE score of 5 or more was associated with a 1.5 point increase in mental illness symptom severity (95%CI = 0.25–2.76, p = 0.019). Finally, we explored and found no evidence that baseline levels of ACEs and suicidality, separately, were effect modifiers for any of the above analyses.

Download:

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

Table 4. Post-imputation pooled coefficient estimates (95% confidence interval) for predictors in linear mixed models for repeated measures for continuous outcomes and rate ratios (95% confidence interval) for generalized estimating equations models for count outcomes for the sample of women in the At Home/Chez Soi study (N = 653).

https://doi.org/10.1371/journal.pone.0277074.t004

Discussion

This is the first analysis of women’s outcomes in an RCT of Housing First. Prior studies and trials on housing interventions for people who experience homelessness often included very small numbers of women or few variables reflecting women’s unique needs and experiences that have precluded documentation of the heterogeneity of experiences and in-depth understanding of this population.

In this Canadian multi-site study, the largest longitudinal sample of women experiencing homelessness to date, the intervention group experienced housing stability for 75% of the 24-month follow-up, as compared to 38% for women in the TAU group, a finding similar to previously published reports of At Home/Chez Soi participants, the majority of whom were men [48, 52]. Even if the large proportion of time spent stably housed for women who received Housing First compared to the TAU group did not translate into greater improvements in health and social outcomes by the end of the study, there were some early differences favoring the treatment group in the first year for quality of life and psychological community integration. Immediate placement in housing and early housing stability for HF compared to TAU [70] participants likely contributed to faster improvements in these outcomes.

While ED visits declined in a similar fashion for HF and TAU over the 24-month follow-up, an unanticipated finding was a significant reduction in ED visits of 0.5 visits at 18 months for the TAU group over the HF group. For this rare count outcome, the dip in ED visits for TAU at 18 months could be a chance improvement as both groups had similar levels of ED visits by 24 months. Analyses of the full At Home/Chez Soi trial data did not show differences for this outcome [48]. One possibility is that uncertainty around continued housing as the trial was ending may have created high levels of stress in the Housing First group, but if this were the case, we would have likely seen similar changes in other study outcomes or among men in the analyses of the full sample.

Since improvements in outcomes over the 24-month period were similar for both the HF and TAU groups, we moved beyond the intention-to-treat analysis and examined baseline predictors for our outcomes across the full sample. We focused on baseline suicidality, ACEs, and parental responsibilities as predictors- variables rarely examined in past studies of supported housing or Housing First [34, 42–44, 71, 72]. As might be expected, women with ≥5 ACEs or moderate-to-high levels of suicidality at baseline showed poorer improvement by 24 months compared to those with low ACE or suicidality scores. Moderate-to-high suicidality was a strong predictor of poorer outcomes at 24 months, including mental health, quality of life, community integration, and ED visits. Past cross-sectional studies of combined samples of homelessness men and women have demonstrated similar links between suicidality, childhood and adult trauma, substance abuse, and mental health [20, 21, 73]. However, while cross-sectional associations between suicidality and substance among women who are homeless have been reported in the literature [19, 74] our study is the first to examine these associations longitudinally.

Contrary to what we might have anticipated on the importance of ACEs in predicting multiple study outcomes given past research [75], we found that high levels of ACEs only predicted significantly worse scores on mental illness severity (CSI). This is consistent with a recent study of the full sample of At Home/Chez Soi where ACEs total scores predicted several mental health conditions at 24-months [50]. Our choice of a cutoff representing high numbers of ACEs, while informed by existing literature, may have limited our ability to detect an association. Alternatively, ACEs may be limited in capturing relevant adverse conditions. A recent study has suggested that an expanded ACEs tool might be used to capture important domains currently missing such as severe poverty or family loss and separation [76], which might be particularly relevant in capturing sources of trauma and adversity for women experiencing homelessness. Our findings also suggested that low educational attainment at baseline was a significant predictor of worse mental health, community integration, and substance use outcomes independently of the other variables in the model. Low educational attainment may be a proxy for other factors such as low health literacy which might impact management of health conditions and requires further investigation [77]. Providing economic support for dependent children was not related to any of our study outcomes.

Although the At Home/Chez Soi support services were important for intervention participants to remain stably housed and make gains in health and social standing, these supports and services were not designed in this Housing First program to be tailored to women’s unique needs. Psychological housing stability which includes considerations of safety, security and feelings of home, is particularly important for women while the At Home/Chez Soi focused predominantly on material housing stability [78]. Women who are unstably housed or homeless need services that address trauma and promote informal peer support networks [32, 79]. Although staff were trained to be responsive to individual needs, the At Home/Chez Soi intervention was not specifically designed to provide the latter two resources. In addition, more than 40% of our participants were homeless for more than three years at enrolment. Two years of stable housing with mental health supports may not have been sufficient to resolve chronic distress from being homeless with co-morbid health, trauma and substance use problems, particularly given the high rates of reported PTSD and ACEs [14].

The lack of greater improvements in the HF arm in this analysis might have been due to the intensive follow-up schedule that could have led to a Hawthorne effect among the TAU group, and higher loss to follow-up in the TAU group might have contributed to smaller differences between groups at 24 months. Despite our sample of homeless women being relatively large, it may not have been sufficient to detect small effects. The At Home/Chez Soi study did not capture gender-sensitive outcomes specific to women experiencing homelessness and therefore we may have missed important changes, such as those associated with positive social support, increased safety from crime and victimization, or strengthening of relationships with children, family or friends [3, 80].

Our study adds to the growing yet sparse literature on the characteristics and needs of women experiencing homelessness and the effectiveness of interventions to eliminate homelessness. Given that women make up more than one third of the homeless population, and their needs differ from that of men [3, 14, 27, 31], it is important that their experiences with homelessness and recovery receive ample focus to better inform the design and implementation of interventions that meet their unique needs. Moreover, the homelessness epidemic has only worsened in the context of the COVID pandemic so the issues remain relevant and these new analyses can provide insights [81].

Supporting information