The Association of Residential Instability and Hospitalizations among Homeless and Vulnerably Housed Individuals: Results from a Prospective Cohort Study

Abstract

The present study examined the association of residential instability with hospitalizations among homeless and vulnerably housed individuals over a 4-year time period. Survey data were linked to administrative records on hospitalizations. Specifically, we used data from the Health and Housing in Transition study, a prospective cohort study that tracked the health and housing status of homeless and vulnerably housed individuals in Canada. Responses from Vancouver-based participants (n = 378) from baseline and 3 follow-ups were linked to their administrative health records on hospitalizations (Discharge Abstract Database – Hospital Separation Files; 2008–2012). A generalized estimating equations model was used to examine associations between the number of residential moves and any hospitalizations during each year (none versus ≥ 1 hospitalizations). Analyses included demographic and health variables. Survey data were collected via structured interviews. Hospitalizations were derived from provincial administrative health records. A higher number of residential moves were associated with hospitalization over the study period (adjusted odds ratio: 1.14; 95% confidence interval: 1.01, 1.28). Transgender, female gender, perceived social support, better self-reported mental health, and having ≥ 3 chronic health conditions also predicted having been hospitalized over the study period, whereas high school/higher education was negatively associated with hospitalizations. Our results indicate that residential instability is associated with increased risk of hospitalization, illustrating the importance of addressing housing as a social determinant of health.

It has been estimated that, globally, 1.6 billion people lack adequate housing [1]. Homelessness is a major issue across North America. Each year, homelessness is experienced by over 1.5 million individuals in the United States [2]. In Canada, almost one-quarter of a million individuals experience homelessness annually, while an estimated 35,000 individuals are homeless on a given night [3].

Housing is a key social determinant of health. Homelessness and vulnerable housing are related to markedly worse health, social, and educational outcomes. A large proportion of individuals who are homeless experience mental health problems, poor health conditions, and substance use disorders [4, 5]. Moreover, population-based data show homelessness and vulnerable housing are major risk factors for mortality [6]. Homelessness is an important public health issue, not only through its impact on the health of individuals who are experiencing homelessness but also through the extensive impact on healthcare and social service usage. Several studies have examined healthcare use based on administrative health records over time among individuals experiencing homelessness or vulnerable housing (e.g., [7–11]). For example, comparing administrative records from homeless and vulnerably housed persons in Toronto to age- and gender-matched income controls, Hwang et al. [9] showed that the healthcare utilization of individuals who are homeless was substantially higher over the study period (mean follow-up duration: 3.9 years). This pattern was especially pronounced for emergency department and inpatient hospital usage for psychiatric conditions (with rate ratios of 8.5 and 9.3, respectively). Another Toronto-based study [7] found persons who were homeless to be high users of healthcare – an average of two emergency department visits per person-year while 10% of individuals had over 12 visits per person-year. In that study, females, Canadian-born individuals, current smokers, those with worse self-rated physical health (SF-12: physical component summary scores), those with recent drug problems, and those who reported an unmet need for mental healthcare had higher odds of emergency department usage over the observation period.

These studies did not include detailed information on participants’ self-reported housing history and residential instability, which would be important to further elucidate the complex association between housing and health. A recent study [10] based on the Health and Housing in Transition (HHiT) project used self-reported survey data from over 1000 homeless and vulnerably housed individuals in three cities (Vancouver, Toronto, Ottawa) to examine the association between residential stability and self-reported emergency department utilization. Residential stability (operationalized as being housed and living in the same location for at least 6 months) was found to be associated with lower emergency department utilization across a 4-year follow-up period. It remains unknown how self-reported housing history and, in particular, residential moves may relate to healthcare utilization based on administrative records. Additionally, a recent report by the National Academy of Sciences explicitly called for further evidence concerning how housing relates to health by employing prospective studies with more than 2 years of follow-up [12]. To address these gaps, and building on the study by Jaworsky et al. [10], we linked the longitudinal HHiT survey data to administrative health data to estimate how residential instability may impact hospitalizations.

Methods

Sample

Data were drawn from the Health and Housing in Transition (HHiT) study [13], a prospective multi-site cohort study that tracked the health and housing status of a representative sample of individuals who were homeless or vulnerably housed in three Canadian cities from 2009 to 2013. For this paper, we focused on data from 394 participants in Vancouver, British Columbia, from baseline and three follow-up interviews at approximately 1 year apart. Of these, 378 participants could be linked to administrative health records for the province of British Columbia (2008–2012, covering the 12 months prior to the baseline interview as well as the 3-year follow-up period, thus a 4-year study period). (Please see linkage section for further information below.)

Homelessness was defined as having lived in a shelter, public space, vehicle, abandoned building, or someone else’s home and not having one’s own place for which rent was paid within the last 7 days. These individuals were recruited at shelters and meal programs. Vulnerably housed was defined as having lived in one’s own room or apartment but having been homeless and/or having had two or more moves over the preceding 12 months. These individuals were recruited at single-room occupancy hotels, meal programs, drop-in centers, and community health centers. Further detail on the sampling method, sampling frame, and recruitment strategy are provided in Hwang et al. [13]. For follow-up purposes, participants provided personal contact information (including aliases and handles they used when accessing services) and were asked to also provide contact information for friends, relatives, service providers, and/or caseworkers as alternative ways to contact them at each interview. Participants were also asked to give consent for hospitals, homeless shelters, municipal social services departments, prisons, and treatment centers to disclose their updated contact information to the research team (further information on retention strategies is provided in Gerlitz et al. [14]). The response rate at the follow-up 3 interviews (approximately 3 years after entry into the study) was 86%. Attrition was due to participant deaths, participant refusal, inability to access participants (e.g., if in jail), and inability to locate participants. All participants provided informed consent and were remunerated ($20 CDN) for their participation following each of the interviews. Ethics approval was obtained from the Research Ethics Board at the University of British Columbia.

Survey Instrument

At each time point, trained research assistants administered the survey via structured interviews that lasted between 60 and 90 minutes. Details on the survey instruments can be found in Hwang et al. [13]. The interview focused on a range of variables using scales and items with previous validity evidence supporting their use with vulnerable populations. For the present analyses, we focused on the following sociodemographics, health- and health service-related variables, as well as perceived social support reported at baseline. Sociodemographic characteristics included gender (male, female, and transgender), age, born in Canada versus elsewhere, self-reported ethnicity (white versus other), and level of education (no high school completion versus high school completion or higher level of education). Participants also reported their lifetime duration of homelessness (months prior to baseline) and age at first time homeless.

Participants were also asked if they had ever been diagnosed with a mental health problem as well as whether a health professional had diagnosed them with chronic health conditions that had persisted or were expected to persist for at least 6 months. Chronic health conditions listed in the survey tool were adapted from the Canadian Community Health Survey [15] and included hypertension, heart disease, obstructive lung disease, cirrhosis, chronic diarrhea, viral hepatitis, peptic or duodenal ulcers, urinary incontinence, inflammatory bowel disease, arthritis, physical disabilities limiting mobility, human immunodeficiency virus, tuberculosis, epilepsy, fetal alcohol syndrome, migraines, traumatic brain injury, stroke, glaucoma, cataracts, hearing impairment, cancer, diabetes, anemia, and dermatologic conditions. The number of identified chronic health conditions was categorized (< 3 versus ≥ 3 conditions). Self-reported physical and mental health status was assessed using the SF-12-item Health Survey Version 1 (SF-12) [16]. The SF12 is a shortened version of the SF-36 designed to reproduce the Physical Component Summary (PCS) and Mental Component Summary (MCS) scales [17]. Items on the SF-12 are binary or ordinally rated with the number of response options ranging from 2 to 6. Scores on these scales are expressed as T-scores with a mean of 50 and a standard deviation of 10. Higher scores reflect better functioning.

To screen for problematic substance use, the ten-item Drug Abuse Screening Test (DAST-10) was used, with a DAST-10 score of ≥ 6 classified as problematic drug use [18]. The ten-item Alcohol Use Disorders Identification Test (AUDIT) was used to screen for alcohol use disorder, with a score of ≥ 20 indicating problematic alcohol use [19]. Thus, problematic substance use was defined as having a DAST-10 score of ≥ 6 and/or an AUDIT score of ≥ 20 in the past 12 months [20]. Unmet healthcare needs (physical or mental) were defined as answering “yes” to either of the following: “During the past 12 months, was there ever a time when you felt that you needed healthcare, but you didn’t receive it?” and “Have you needed mental health care in the past 12 months but were not able to get help?”. These questions were based on other national surveys, such as the Joint Canada/US Survey of Health [21] and the National Population Health Survey [22], and have been used to assess unmet healthcare needs among homeless adults in previous research [23]. Participants also reported whether they had had a primary care provider in the past 12 months.

Perceived available social support was measured with the first item of the Social Support Network Inventory [24] (“Are there any people with whom you feel at ease and can talk to about personal issues?”; yes/no.) Toro’s Housing Quality Instrument [25] was used to measure the comfort, safety, spaciousness, privacy, friendliness, and overall quality of respondents’ current housing or living conditions as perceived by them. Responses are provided on six items with a 7-point scale from “very bad” to “very good.” Items are combined into a total summed score ranging from 6 to 42, with higher scores reflecting higher levels of housing quality.

Main Explanatory Variable. To obtain information on participants’ housing, the Residential Time-Line Follow-Back Inventory was utilized [26]. Housing history information was classified based on methods adapted from Tsemberis et al. Each residence in a participant’s housing history was classified into one of 19 types of residence, which were then categorized into two residence categories: housed and homeless. Types of residence for the housed category were own house or apartment, stay with friends/family (paying rent), rooming house, boarding home, group home, single-room occupancy (SRO), motel or hotel, trailer, supportive housing, student housing, alternative housing, transitional housing, recovery house/second stage housing, domiciliary hostel, and employer-provided housing. Types of residence for the homeless category were staying with friends/family (not paying rent), homeless shelter, homeless on streets, and campground.

We used the number of residential moves that the participant experienced during the year prior to each of the survey interviews to assess housing stability. Residential moves were defined as the number of moves between distinct types of primary residence (e.g., apartment to jail, jail to SRO would be counted as two residential moves). In this paper, hospital stays were not counted as a residential move.

Outcome Variable. Hospitalizations were derived from the Discharge Abstract Database Hospital Separation File housed at Population Data BC [27]. We used a dichotomized variable – whether or not participants had experienced any hospitalization during the 12 months prior to the baseline interview or the interval period between follow-up interviews.

Data Linkage

The individual linkage between survey data and administrative health data was completed by Population Data BC using Personal Health Numbers and probabilistic linkage using identifiers, such as names and date of birth. Population Data BC creates linkages pursuant with privacy legislation (Freedom of Information and Protection of Privacy Act), and all researchers accessing de-identified data undergo comprehensive privacy training.

Statistical Analyses

Descriptive statistics were used to summarize quantitative variables at baseline. Comparisons between participants with versus without any hospitalizations in the 12 months prior to baseline were conducted using Student t-tests for normally distributed continuous variables, Wilcoxon test for continuous variables that were not normally distributed, and χ² test or the Fisher’s exact test for categorical variables.

A generalized estimated equations (GEE) logistic regression model was used to examine the association between the number of residential moves during the interval (i.e., a time-varying count variable) and hospitalizations during the interval (dichotomized to no or at least one hospitalization in 12 months preceding the date of the interview). As a sensitivity analysis, we reran the multivariable model using negative binomial and Poisson models, with counts for the hospitalization outcome variable. The adjusted odds ratio estimates were consistent with those obtained from the GEE results reported here. (Further details are available upon request.) In the preliminary model, we adjusted for the following 14 variables (measured at baseline) that have been shown to be associated with healthcare use: age, gender, whether born in Canada, ethnicity, level of education, physical and mental health functioning, number of chronic health conditions (< 3 versus ≥ 3 conditions), problematic substance use, lifetime history of mental health diagnosis, unmet healthcare needs, having a primary care provider, perceived social support, and housing quality.

The selection of the most parsimonious GEE model requires completing two steps. The first step is to decide on a suitable working correlation structure for the full model based on the quasi-likelihood under the independence model criterion (QIC; smaller values representing better models) [28]. We considered the autoregressive (1) (AR (1)), exchangeable, and independent correlation structures. In the second step, we selected covariates using the best subset regression approach based on the QICu criteria (designed to evaluate which combination of predictors best explain the outcome [29, 30]). Furthermore, the analyses also included the natural log of person-years as an offset, and time points were specified to indicate the time ordering of repeated measurements on the same subject. Person-years were included to account for differential time at risk between the intervals. Person-years for each interval were calculated as the difference between the ending and starting interval dates of survey data, and the administrative health records were aggregated during this interval.

Of the 378 individuals, 45 had missing data on one or more covariates at baseline to be included in the multivariable model. Analyses were run excluding these individuals (“complete case analysis”) as well as with the full sample of 378 using the multiple imputation (MI) by chained equation [31] approach for missing variables. This approach utilizes separate conditional predictive models for each adjustment variable, controlling for the rest of the variables. We repeated the process of imputation m = 5 times [32] using the R package mice: multivariate imputation by chained equations [31]. We then implemented our GEE approach in each of these 5 datasets and obtained 5 different parameter estimates (corresponding to the explanatory variable and each covariate). At the final stage, following Rubin’s rules [33], we pooled these parameter estimates to obtain average and variability measures to reflect the uncertainty associated with the missing-data imputation.

Statistical tests were 2-tailed, and p ≤ 0.05 was considered statistically significant for the descriptive statistics and multivariable model. All statistical analyses were performed using SAS statistical software version 9.4 and R 3.5.0.

Results

Baseline Characteristics for Those Hospitalized or Not Prior to Baseline

Baseline characteristics for the whole sample and stratified by whether participants had ever been hospitalized in the 12 months prior to baseline are provided in Table 1. The mean age was 42.0 (standard deviation [SD] = 10.2) years, and the majority of participants were white, were male, had completed high school or higher level of education, and were born in Canada. The median number of residential moves in the 12 months prior to baseline was 2 (interquartile range (IQR): 2, 3), 54.8 % of the participants reported that they had ever received a mental health diagnosis, and 56.9% reported 3 or more chronic health conditions. The mean SF-12 PCS score was 43.8 (SD = 11.4), and the mean SF-12 MCS score was 38.6 (SD = 12.8), indicating that, in terms of both physical and mental health status, the mean scores of participants were well below average compared to US population norms (M = 50, SD = 10). Furthermore, 44.3% participants reported problematic substance use. Close to half (48.9%) reported unmet physical or mental healthcare needs, and 65.5% reported having a primary care provider. Regarding perceived social support, the majority of participants (81.9%) reported having someone to talk to about personal issues.

Table 1.

Baseline characteristics of the sample by hospitalization

Variables	Total sample (N = 378)	Hospitalization 12 months prior to baseline		Pvalue
		Yes (n = 83)	No (n = 295)
Age based on DOB (truncated)a	42.0 (10.2)	42.3 (9.6)	41.9 (10.3)	0.86
Gender				< 0.01
Female	132 (34.9)	41 (49.4)	91 (30.8)
Transgender	8 (2.1)	d.s.	d.s.
Male	238 (63.0)	38 (45.8)	200 (67.8)
White ethnicity	216 (59.0)	42 (51.9)	174 (61.1)	0.14
Canadian-born	339 (91.1)	75 (91.5)	264 (91.0)	0.90
High school or higher education level	200 (53.6)	35 (42.7)	165 (56.7)	0.03
Partnered	61 (16.4)	11 (13.3)	50 (17.2)	0.39
Lifetime duration of homelessness (yrs) a	5.4 (6.2)	5.3 (5.8)	5.5 (6.3)	0.98
Age at first homeless b	24 (15, 39)	24 (15, 39)	23.5 (16, 39)	0.94
Total income (last 12 months; CAD $)a	1644 (1856)	1795 (2125)	1602 (1775)	0.35
Currently working	48 (12.8)	9 (10.8)	39 (13.4)	0.54
SF-12 PCSa	43.8 (11.4)	39.8 (10.6)	44.9 (11.4)	< 0.01
SF-12 MCSa	38.6 (12.8)	40.9 (13.3)	37.9 (12.6)	0.09
≥ 3 chronic conditions	215 (56.9)	57 (68.7)	158 (53.6)	0.01
Problematic substance use	167 (44.3)	40 (48.8)	127 (43.1)	0.36
History of mental health problem	204 (54.8)	49 (59.8)	155 (53.4)	0.31
Unmet health care needs (mental and physical)	185 (48.9)	40 (48.2)	145 (49.2)	0.88
Have a primary care provider	247 (65.5)	63 (76.8)	184 (62.4)	0.02
Housing qualitya	27.5 (8.0)	28.2 (8.3)	27.3 (7.9)	0.32
Social support (at least one person)	303 (81.9)	71 (86.6)	232 (80.6)	0.21
Residential moves b	2 (2, 3)	2 (2, 3)	3 (2, 3)	0.63

Unless otherwise stated, values represent n (proportion); ^a represents mean (standard deviation);

^brepresents median (interquartile range). d.s. = data suppressed due to privacy concerns

p ≤ 0.05 are bolded

Participants who had been hospitalized in the 12 months prior to baseline (n = 83) were more likely to report being female, lower than high school education, having lower self-reported physical functioning, having a primary care provider, and having 3 or more chronic health conditions compared to participants who had not been hospitalized (n = 295). With regard to the length of hospital stay prior to baseline, the median number of days per hospital stay was 4 (IQR: 2, 8).

Predictors of Hospitalizations across 4 Years

Among the 378 participants, there were a total of 519 hospital admissions over the 4-year study period, with 50 participants having 4 or more hospitalizations. The most responsible diagnosis (defined as the diagnosis that accounted for the greatest length of stay) categories for hospital admission across the 4-year study period were mental and behavioral disorders due to psychoactive substance use; schizophrenia and related conditions; human immunodeficiency virus disease; mood disorders; chronic lower respiratory disease; disorders of the gallbladder, biliary tract, and pancreas; infections of the skin and subcutaneous tissue; influenza and pneumonia; and other bacterial diseases.

As described above, the QIC was used to select the correlation structure for the full model. The QIC values were the same under different correlation structures (to the third decimal), and we chose the autoregressive (1) or AR (1) correlation structure (QIC = 1, 074). Variable selection was then conducted by examining the QICu values. Out of a total of 16,384 (2¹⁴) all possible combinations of main effects models under the AR [1] correlation structure, we evaluated the top 20 models with the lowest QICu values. We chose the model with the lowest QICu (1098.73), which included variables selected in more than 50% of these models. Table 2 presents the GEE results of the final multivariable model excluding missing data (i.e., complete cases). The pooled estimates based on imputed data analyses were consistent with those from the complete case analysis reported here (results available on request). In the complete case GEE analysis, factors that were significantly associated with being hospitalized over the 3-year follow-up period were the number of residential moves (adjusted odds ratio (AOR) = 1.14; 95% confidence interval (95% CI) = 1.01–1.28), transgender (AOR = 3.56; 95% CI = 1.35–9.61), female gender (AOR = 2.05; 95% CI = 1.40–2.99), self-reported mental health status (AOR = 1.02; 95% CI = 1.00–1.03), having 3 or more chronic health conditions (AOR = 1.87; 95% CI = 1.24–2.82), and perceived social support (AOR = 1.87; 95% CI = 1.10–3.17), whereas a high school or higher level of education was negatively associated with hospitalizations (AOR = 0.67; 95% CI = 0.47–0.97)

Table 2.

Generalized estimating equation logistic regression model to estimate the independent associations of predictor variables with hospitalizations across the 4-year study period

	Multivariable model (complete case; n = 333)
Variables	Adjusted odds ratio (95% CI)
Residential movesa	1.14 (1.01, 1.28)
Gender
(Ref: male)
Transgender	3.56 (1.35, 9.61)
Female	2.05 (1.40, 2.99)
High school or higher education level (Ref: less than high school)	0.67 (0.47, 0.97)
SF-12 physical component score	0.99 (0.97, 1.00)
SF-12 mental component score	1.02 (1.00, 1.03)
Has ≥ three chronic conditions (Ref: has fewer than three chronic conditions)	1.87 (1.24, 2.82)
Has a history of mental health problem (Ref: no history of a mental health problem)	1.33 (0.91, 1.96)
Social support (≥ one person; ref: no persons)	1.87 (1.10, 3.17)

Significant ORs (p ≤ 0.05) are bolded. ^a time-varying variable; all others were fixed at baseline. The analyses also included the natural log of person-years as an offset, and time points were specified to indicate the time ordering of repeated measurements on the same subject.

Note. Several variables in the multivariable model were not significant individually. However, dropping these variables from the model would have resulted in an increased QICu value. Therefore, based on the information criterion, these variables were retained in the final model for their collective contribution in better explaining the hospitalization outcome.

As a sensitivity analysis, we re-ran the multivariable model using multiple imputation for missing variables. The aOR estimates were consistent with those obtained from the complete case model reported here.

Discussion

We found a consistent significant association between our main explanatory variable, a higher number of residential moves (i.e., residential instability), and hospitalizations among homeless and vulnerably housed persons over the study period. Other variables that were significant predictors of hospitalization were transgender and female gender, education level, self-reported mental health status, having ≥ 3 chronic health conditions, and perceived social support.

Residential Instability

Prior work has found residential stability to be associated with lower levels of emergency department utilization among persons who are homeless, including results from a three-city Canadian study [10] as well as a US-based study [34]. Moreover, residential stability has been shown to attenuate homeless persons’ reported unmet healthcare needs [10]. A lack of residential stability may undermine one’s health through a sense of uncertainty/anxiety about one’s residence, difficulty in developing or sustaining meaningful social connections, and/or changes in one’s proximity, familiarity, and ultimately access with nonemergency healthcare services. The association between the number of residential moves and hospitalization is congruent with the concept of competing priorities; the concern of meeting basic needs for shelter may come at the expense of caring for one’s health and healthcare needs [35, 36]. The stress and efforts associated with finding shelter/moving and the time spent on residential transitions, as well as concerns about stability on a regular basis, consume energy that could have been otherwise dedicated to accessing primary care and supporting their health [35]. Accordingly, homeless individuals with higher subsistence difficulties appear to engage less in preventative health activities (e.g., having a regular health provider [35] or meeting medical appointments [37] or healthcare in nascent stages of health conditions), which would lead to higher levels of hospitalizations when such conditions more fully emerge. O’Toole et al. [38] suggested that residential stability among homeless and vulnerably housed persons may represent an enabling context that allows them to access healthcare via nonemergency healthcare providers rather than emergency services for acute needs. Indeed, analysis of healthcare expenditures among homeless persons in Oregon yielded evidence that lower emergency department and inpatient care occurred following movement into supportive housing [39]. Our results provide additional evidence of the importance of stable housing and – by extension – underscore the importance of supportive housing initiatives [40] as a means to potentially lower hospitalization among homeless and vulnerably housed persons. They also highlight the importance of making healthcare more readily available to individuals who are homeless or precariously housed.

Female Gender and Transgender

Females have been shown to be more likely to use healthcare services than males in previous research with homeless persons [41]. In analyses specific to emergency department use, Chambers et al. [7] observed females as more likely than males to have utilized such services over a 4-year period. In the general population also, females are more likely than males to use emergency services [42]. In line with this, our study also found that women were more likely to be hospitalized. Our results also indicate that participants identifying as transgender were more likely to be hospitalized although this finding needs to be interpreted with caution given the small number in our sample and the large confidence intervals in our analyses.

Homeless transgender individuals are particularly vulnerable and may have difficulty in accessing shelters due to their gender nonconformity, leaving them potentially exposed to risks on the street [43] or to transphobia and violence in the shelter system [44]. Previous research indicates that transgender homeless persons experience even higher rates of sexual or physical assault compared to homeless women and men [45]. Furthermore, transgender individuals experience high rates of HIV, depression, and suicide attempts [46], as well as drug use [47]. Despite transgender people facing barriers in accessing healthcare services in general [48, 49], this combination of factors may explain why transgender homeless and vulnerably housed individuals were more likely to be hospitalized.

An Educational Attainment Less than High School

Participants with an educational attainment of high school completion or higher were less likely to have been hospitalized. Univariable analyses of homeless participants in Toronto found that those with some post-high school education had a lower probability of ever using emergency department services over a 4-year period relative to those without high school education [7]. Among a US sample of homeless adults, adjusted analyses found higher educational attainment (years of education) related to lower levels of hospitalization [50]. These findings and those from the present study suggest that lower educational attainment may be associated with higher levels of hospitalization, perhaps in part through persons with lower educational attainment experiencing greater barriers to accessing care from non-hospitalization means. For example, these participants may have lower health literacy related to lower educational attainment, which has been shown to be associated with hospitalization [51].

Mental Health Status

In this sample, individuals with better self-reported mental health based on the SF-12 measure had higher odds of hospitalizations over the follow-up period, whereas lifetime mental health diagnosis was not significantly associated with hospitalizations. These findings are in contrast to previous research indicating that a self-reported history of mental conditions was associated with self-reported acute care utilization among homeless or vulnerably housed individuals [36] or self-reported hospitalizations among street-involved youth [52]. Similarly, a study based on administrative health records of homeless individuals found a significant association between mental health conditions and hospitalization/acute care use [53].

Multiple Chronic Health Conditions

Participants who reported having multiple chronic health conditions had higher odds of hospitalizations, as such persons likely had a higher number or greater severity/complexity of health concerns requiring utilization of hospital services, chronic and/or acute. O’Toole at al. [38] also found evidence of comorbidity related to healthcare use among homeless individuals in the United States, while a Toronto-based study found higher odds of hospitalization among homeless persons with multiple chronic health conditions [11].

Social Support

Participants who indicated that they have a person with whom they feel at ease and can talk to about personal issues had higher odds of hospitalizations. Greater social support has been related to higher healthcare utilization among homeless persons in North America in previous studies [38, 41, 54]. Individuals who are homeless or vulnerably housed may be more inclined to seek healthcare if they have a friend or supportive individual in their lives who may encourage them to utilize the services whether through emotional support, through practical advice/suggestions, or through accompanying them to the service.

Strengths and Limitations

Limitations of this study include that hospitalization data may be incomplete. Participants may have moved and received healthcare in other provinces; thus, the number of hospitalizations could be underestimated. Furthermore, the chronic health conditions and prevalence of lifetime mental health conditions were based on self-report, resulting in a possible underestimation of these conditions. A hospital stay could also potentially lead to change in residence, as a person may lose access to housing, e.g., to an apartment or an SRO, because of a hospital stay. However, given that the median number of days per hospital stay was relatively short (4; IQR 2, 8), it can be assumed that this was less likely to disrupt a housing situation. The strengths of this study include the use of a unique longitudinal dataset with linked self-reported survey and administrative health data, including data on granular housing history (i.e., capturing the number of moves between distinct residences per interview period). Furthermore, despite this being a challenging population to follow, we had a very high retention rate (attrition was only 14% over the 3-year follow up period).

Conclusion and Implications

Our findings highlight the importance of a more granular understanding of residential instability and its effect on health status as manifested in needing hospitalization, illustrating the benefits of addressing housing as a social determinant of health. Our findings also underscore the need for supportive housing to improve residential stability, thereby minimizing residential moves that have a negative effect on the health of individuals who are vulnerably housed and homeless. There have been demonstrable successes with Housing First models [55–57] and other related housing approaches [58].

As part of successful interventions for this group, one may examine how integration or collaboration between primary health services and housing [39, 59, 60] may help manage multiple chronic diseases. Supportive housing that addresses the unique needs of women and transgendered individuals should also be explored. The facilitation of access to healthcare services for people who are homeless or precariously housed by outreaching to emergency shelters would also merit consideration.

Future research is needed to examine the association of housing and residential instability with other healthcare use outcomes (e.g., with a specific focus on emergency department versus ambulatory care use) and associated cost estimates to further understand the complex association between housing and healthcare use.

Funding Information

This project was supported by an operating grant (MOP-86765) and an Interdisciplinary Capacity Enhancement Grant on Homelessness, Housing and Health (HOA-80066) from the Canadian Institutes of Health Research.