Social Psychiatry and Psychiatric Epidemiology

, Volume 48, Issue 8, pp 1245–1259

Housing First improves subjective quality of life among homeless adults with mental illness: 12-month findings from a randomized controlled trial in Vancouver, British Columbia


    • Faculty of Health SciencesSimon Fraser University
  • Akm Moniruzzaman
    • Faculty of Health SciencesSimon Fraser University
  • Anita Palepu
    • Division of General Internal Medicine, Centre for Health Evaluation and Outcome StudiesUniversity of British Columbia
  • Denise Zabkiewicz
    • Faculty of Health SciencesSimon Fraser University
  • Charles J. Frankish
    • School of Population and Public HealthUniversity of British Columbia
  • Michael Krausz
    • Department of Psychiatry, Centre for Health Evaluation and Outcome StudiesUniversity of British Columbia
  • Julian M. Somers
    • Faculty of Health SciencesSimon Fraser University
Original Paper

DOI: 10.1007/s00127-013-0719-6

Cite this article as:
Patterson, M., Moniruzzaman, A., Palepu, A. et al. Soc Psychiatry Psychiatr Epidemiol (2013) 48: 1245. doi:10.1007/s00127-013-0719-6



This study used an experimental design to examine longitudinal changes in subjective quality of life (QoL) among homeless adults with mental illness after assignment to different types of supported housing or to treatment as usual (TAU, no housing or supports through the study). We hypothesized that subjective QoL would improve over time among participants assigned to supported housing as compared to TAU, regardless of the type of supported housing received or participants’ level of need.


Participants (n = 497) were stratified by level of need (“high” or “moderate”) and randomly assigned to Housing First (HF) in scattered-site apartments, HF in a congregate setting (high needs only), or TAU. Linear mixed-effects regression was used to model the association between study arm and self-reported QoL at baseline and at 6 and 12 months post-baseline by need level.


Based on the adjusted overall score on the QoL measure, participants randomized to HF reported significantly greater overall QoL as compared to TAU, regardless of need level or type of supported housing at both 6 and 12 months post-baseline. Scores on the safety and living situation subscales were significantly greater for both high and moderate need participants assigned to supported housing regardless of type at both 6 and 12 months post-baseline as compared to TAU.


Despite multiple health and social challenges faced by homeless individuals with mental illness, HF in both scattered-site and congregate models results in significantly greater perceived QoL as compared to individuals who do not receive HF even after a relatively short period of time.


Housing firstHomelessnessMental illnessQuality of lifeRandomized controlled trial


Over the past 40 years, the closure of large psychiatric facilities has resulted in the majority of adults with serious mental illness (SMI) living in community settings. The goal of deinstitutionalization was to integrate people with SMI into residential communities and to improve their quality of life (QoL). However, in both Canada and the United States, many individuals were discharged from institutions into the community without appropriate housing and support [13]. People with SMI are over-represented among homeless populations and often experience a “revolving door” of homelessness, criminal justice involvement, and acute healthcare system contacts [35]. Affordable, supported housing that is of good quality and located in safe neighborhoods is integral to the provision of care for people with SMI and their ability to participate in community life [46].

Housing for people with SMI and histories of homelessness often involves congregate settings in inner-city neighborhoods with minimal support available [7, 8]. Programs typically require people to transition from more to less restrictive settings based on their treatment progress, including adherence to medication and sobriety. This continuum approach to housing often provides little choice and fails to engage those with the most severe needs. Alternatively, Housing First (HF) provides homeless individuals with SMI immediate access to permanent housing as well as services and supports that are flexible and consumer-driven [9]. Research on HF has documented improved residential stability [7, 10], community integration [11, 12], and high levels of client satisfaction [13].

Given the chronic and persistent nature of SMI, it is widely agreed that a primary focus on long-term outcomes related to distress and psychopathology is inadequate [5, 6]. Researchers and clinicians have argued that, among homeless people with SMI, subjective QoL is perhaps the most appropriate outcome variable when assessing the effectiveness of various interventions including housing [14]. Subjective QoL refers to the psychological processes involved in evaluating one’s wellbeing or satisfaction across a variety of domains such as family relationships, leisure activities, finances and housing. This approach assumes that the individuals are the best judges of their QoL and that different individuals may perceive and interpret the same event very differently based on divergent values, expectations and perspectives [15]. Furthermore, objective assessments of QoL such as health, housing and income often do not reflect an individual’s experience and ability to respond to positive and negative events in their lives [14].

Lehman and colleagues [1618] developed a widely-used measure of QoL for people with SMI and found a response gradient. People with SMI consistently report lower QoL than the general population, and those who are homeless report lower levels than those who are housed [19]. Several studies [20, 21] have examined QoL among adults with SMI in various residential settings and found that people housed in more restrictive settings such as institutions reported lower QoL than their counterparts living in less restrictive community settings, particularly in the domains of living situation, safety, and health. Studies have also examined QoL among adults with SMI in different types of supported housing such as private, public, and voluntary housing [22] and small congregate residences versus independent housing with supports [21]; to date, no differences have been found in subjective QoL among various community residential settings.

A growing body of research has documented the impact of stable housing on QoL among homeless adults with SMI over time. Using a randomized controlled trial design, Schutt et al. [23] examined QoL among participants who were assigned to independent apartments or small group homes. Participants assigned to independent apartments reported greater satisfaction with housing than those living in group homes, however, there was no difference in overall life satisfaction between the two groups. More recently, Gilmer et al. [7] compared self-reported QoL using a short version of Lehman’s [16] QoL interview among homeless adults with SMI who received HF and Assertive Community Treatment (ACT) as compared to a group of homeless adults who received public mental health services. Participants who received HF reported greater QoL across all domains, particularly ‘living situation,’ ‘daily activities,’ and ‘legal and safety issues.’ However, these findings were based on an observational rather than experimental design, therefore, it is possible that the HF and comparison groups were not equivalent.

In summary, past research suggests lower subjective QoL among homeless as compared to housed persons with SMI but indicates no consistent effects of housing type or other objective differences in life circumstances. Very few longitudinal studies using experimental designs have explored the relationship between housing and subjective QoL using validated measures. The current study uses a randomized controlled trial design and examines longitudinal changes in subjective QoL among homeless adults with mental illness after random assignment to different types of supported housing (scattered site and congregate) or to treatment as usual (TAU, no housing or supports through the study). We hypothesized that subjective QoL would be greater among participants assigned to supported housing as compared to TAU, regardless of the type of supported housing received and participants’ level of need.


Participants and sampling

The Vancouver At Home Study includes two randomized controlled trials involving homeless adults with mental disorders in Vancouver, British Columbia (ISRCTN57595077 and ISRCTN66721740). Differences between the two trials concentrated on participants’ level of need (see below). Study design and sample size were determined by the At Home/Chez Soi National Research Team, which coordinated activities across five Canadian study centers. Each site followed a common design and protocol but also had distinctive features. The Vancouver study includes a higher proportion of participants who met criteria for substance dependence, and is the only site that implemented HF in a congregate setting as well as in independent apartments. Details related to the study protocol, including measures and intervention arms have been reported elsewhere [24, 25].

Eligibility criteria for the two trials included legal adult status (19 years of age or over), presence of a current mental disorder on the Mini International Neuropsychiatric Interview (MINI) [26], and being absolutely homeless or precariously housed. Absolute homelessness was defined as living on the streets or in a shelter for at least seven consecutive nights with little likelihood of obtaining housing in the upcoming month or being discharged from an institution with no fixed address. Precariously housed was defined as living in a rooming house, hotel or other form of transitional housing with at least two episodes of absolute homelessness in the past year.

Participants were recruited through referral from over 40 agencies representing approximately 13 different types of services available to homeless adults in Vancouver. The majority was recruited from homeless shelters, drop-in centers, homeless outreach teams, hospitals, community mental health teams, and criminal justice programs. Referral was initiated by service providers in the community, and general eligibility criteria were assessed using a brief telephone screen with the referral agent. If the potential participant appeared to meet the study criteria, a face-to-face interview was scheduled to more formally assess eligibility. All interviews were conducted by trained research interviewers who explained procedures and obtained informed consent. Participants received a cash honorarium of $35 after the baseline interview and $30 for each of the 6- and 12-month follow-up interviews. Institutional Research Ethics Board approval was obtained through Simon Fraser University and the University of British Columbia.

If the individual met all study criteria, they were enrolled as a participant and the baseline interview commenced. Details pertaining to participant flow through the Vancouver study for enrollment and allocation to study arm (Fig. 1) and follow-up and analysis (Table 1) are illustrated below. At baseline, enrolled participants completed a series of interviewer-administered questionnaires that included questions on socio-demographic characteristics, symptoms of mental illness, substance use, physical health, service use, community integration, and QoL. After completion of the baseline interview, participants were assessed as having either “High Needs” (HN) or “Moderate Needs” (MN) on the basis of a computerized algorithm. Participants were classified as HN if they scored 62 or less on the Multnomah Community Ability Scale [27], met criteria on the MINI for current psychotic disorder or (hypo) manic episode, and met criteria for at least one of the following: (a) current alcohol or substance abuse or dependence on the MINI, (b) recent arrest or incarceration, or (c) more than one psychiatric hospitalization in any one of the past 5 years. Participants who did not meet the above criteria were classified as MN.
Fig. 1

Participant flow through enrolment, need level assessment, and allocation to study arm

Table 1

Follow-up and analysis of participants after 12 months

Study arm

No follow-up data collecteda

Analyzed sample



No contact


Withdrew consent

High needs




















Moderate needs

















aNo follow-up data collected at either 6 or 12 months

bTotal number of deaths over 12 months was as follows: ACT-3, CONG-1, HNTAU-5, ICM-2, MNTAU-4. In some cases, follow-up at 6 months was completed

Participants assessed as HN were randomly assigned to one of three study arms: (1) HF with ACT1 in which participants could choose from up to three market lease apartments in a variety of neighborhoods and services were provided by a trans-disciplinary outreach team, (2) Congregate Housing2 with on-site support (CONG) in which participants had their own room and bathroom but shared amenity space with 100 other program participants and received three meals per day as well as activity programming and various health and social services on site; and (3) TAU (HNTAU) which provided no additional housing or support services beyond what was already available in the community. Participants assessed as MN were assigned to one of two study arms: (1) HF with Intensive Case Management (ICM) in which participants could choose from up to three market lease apartments in a variety of neighborhoods and services were provided by a team of case managers who connected participants to existing services; and (2) MNTAU as described above.

This analysis is based upon data from the baseline questionnaires from 297 HN participants and 200 MN participants recruited from the study’s inception in October 2009 to June 2011, and follow-up data at 6 and 12 months after the baseline interview (see Table 1).

Variables of interest

Demographic variables including age, ethnicity, education, and duration of homelessness, as well as other variables, were assessed by self-report. Mental disorders, including alcohol and substance dependence, were assessed at baseline using the MINI. ‘Less severe cluster’ included at least one of major depressive episode, panic disorder, and post-traumatic stress disorder; ‘severe cluster’ included at least one of psychotic disorder, mood disorder with psychotic features, and manic/hypomanic episode. Daily drug use during the previous month was assessed using the Maudsley Addiction Profile [29]. Medical conditions were assessed using a self-report checklist of 28 chronic health conditions (lasting longer than 6 months). Infectious disease included positive self-report diagnosis of HIV, hepatitis B or hepatitis C. Arrest was assessed by asking participants how many times the police had arrested them in the past 6 months. At baseline, no significant differences were observed across a variety of variables except for the MN group, where participants in ICM reported a longer duration of lifetime homelessness than those in MNTAU (p = 0.037) and participants in MNTAU were more likely to be absolutely homeless than those in ICM (p = 0.041; see Appendix Table A1). Based on baseline characteristics, no differences were observed between the full sample (n = 497) and participants who were subsequently contacted for follow-up (n = 471; see Table 2).
Table 2

Socio-demographic and mental disorder characteristics for full samples and follow-up samples at baseline among high need (HN) and moderate need (MN) participants


HN participants

MN participants

Full sample (n = 297), n (%)

At least one follow-up visit (n = 286), n (%)

Full sample (n = 200), n (%)

At least one follow-up visit (n = 185), n (%)

Gender (male)

218 (74)

208 (74)

141 (71)

129 (70)

Age at randomization, mean years (SD)

39.7 (11.2)

39.8 (11.0)

42.6 (10.5)

42.2 (10.6)



44 (15)

42 (15)

33 (16)

30 (16)


170 (57)

163 (57)

110 (55)

104 (56)


83 (28)

81 (28)

57 (29)

51 (28)

Incomplete high school

179 (61)

171 (60)

101 (51)

95 (51)

Single/never married

214 (73)

205 (72)

129 (65)

121 (65)

Lifetime duration of homelessness, mean months (SD)

62.0 (67.0)

60.8 (64.4)

57.4 (74.9)

59.0 (75.4)

Age of first homelessness, mean years (SD)

28.7 (12.5)

28.8 (12.5)

32.6 (14.1)

31.9 (14.2)

Mental disorder (less severe cluster)

133 (45)

128 (49)

131 (66)

122 (66)

Mental disorder (severe cluster)

272 (92)

261 (91)

91 (45)

83 (45)

Multiple mental disorders (2 or more)

148 (49)

142 (50)

92 (46)

85 (46)

Substance dependence (past month)

183 (61)

179 (63)

105 (53)

98 (53)

Chronic medical conditions (2 or more)

231 (78)

221 (77)

171 (86)

159 (86)

Infectious disease (HIV, hep B/C)

87 (30)

85 (30)

70 (35)

67 (36)

Daily illicit drug use (past month)a

82 (28)

80 (28)

44 (22)

42 (23)

Arrest (past 6 months)

128 (44)

124 (45)

45 (23)

41 (23)

aDoes not include alcohol

QoL was assessed using the Quality of Life Interview-20 (QOLI-20) [18], which was developed from the original 35-item interview [16] using item-response theory resulting in 20 subjective items that retain the internal consistency of the original scale. The scale consists of six subscales: family, finances, leisure, living situation, safety and social. In addition, a final item (global) assesses the individual’s overall satisfaction with life. Each item is rated on a seven-point scale, anchored by “terrible” and “delighted.” “Do not know” or “declined” were also possible responses.


Continuous variables are reported in terms of means and standard deviations (SD) and categorical variables are reported as proportions. Bivariate comparisons of continuous variables were conducted using t tests or one-way ANOVA where appropriate. Comparisons of categorical variables between groups were conducted using Pearson’s Chi-square test.

In order to evaluate the effect of study arm on QOLI-20 scores over time, an intention-to-treat analysis was conducted. Due to the longitudinal nature of the data (repeated measures for each participant), linear mixed effects regression was used to model the association between study arm and a normally distributed outcome variable (i.e., QOLI-20 total score). In the regression model, we also included time (as a repeated factor) and the interaction terms between time and study arm. Effects for HN and MN study arms (compared to the relevant TAU group) were estimated separately. In addition to study arm, time, and their interactions, we included additional baseline covariates in the model building process based on significant differences between study arms (p ≤ 0.05; see Appendix Table A1 in the “Appendix”). Participants who did not complete both the 6- and 12-month follow-up visits (HN: n = 11, MN: n = 15) were not included in the regression models (see Table 1).

In addition, each QOLI-20 subscale score was analyzed separately using similar linear mixed-effects modeling techniques and controlling for the same covariates. Estimated coefficients with robust standard errors (SE) are reported as a measure of association. All reported p values are two-sided.

In addition to skipped items, the responses ‘do not know’ and ‘declined’ on the QOLI-20 were considered missing values. Mean substitution for individual missing items was used to obtain the total and subscale scores as long as no more than half of the items were missing (recommended by Lehman, personal communication, 3 November 2011). No values were missing for the controlling covariates included in the MN model. We conducted sensitivity analyses using multiple imputation techniques to evaluate the effect of mean substitution used to obtain QOLI-20 scores. We used the chained equation method [30] to impute individual missing items from the QOLI-20 for HN and MN groups separately. We imputed the missing values ten times and then averaged over imputations. We re-ran the linear mixed-effects regression analyses (separately for HN and MN groups) using the imputed datasets for all participants as well as for participants with at least one follow-up visit. Findings from these sensitivity analyses are included in the “Appendix” (see Appendix Tables A2, A3). Based on these different techniques for managing missing data (mean substitution and multiple imputation), no meaningful differences were observed in the results. Therefore, results are presented using mean substitution of missing values. All analyses were performed using STATA-12 [34].

Finally, responsiveness of the QOLI-20 total score to the interventions was assessed for each study arm by examining effect sizes from baseline to 6 months and from baseline to 12 months using the observed total score. Several methods and benchmarks have been recommended for assessing the responsiveness of self-report QoL measures, including effect size (0.8 is considered large) and SD (0.5 SD is considered moderate) [3133]. We used the “sizefx” module in STATA-12 to calculate effect sizes and the pooled SD.


Sample description

Of the 497 participants interviewed at baseline, 457 (92 %) were re-interviewed at 6 months and 444 (89 %) at 12 months follow-up. Overall, 471 (95 %) participants completed either the 6 or the 12 months follow-up interview and 430 (87 %) completed both interviews; there were no significant differences in attrition between the HN and MN participants. As expected, there was greater loss to follow-up among the TAU groups than among the HF groups (see Table 1). However, sensitivity analyses using multiple imputation did not reveal large differences in results based on the full sample as compared to participants who completed at least one follow-up interview (see Appendix Tables A2, A3).

QOLI-20 total score

Table 3 presents the unadjusted QOLI-20 total score and subscale scores by study arm at each time point. As expected, there were no significant differences in total score at baseline across study arms. For the HN group, the housing intervention arms (ACT and CONG) reported significantly higher total scores as compared to HNTAU at both 6 and 12 months (p ≤ 0.05). For the MN group, a significant trend toward higher total scores was observed in the housing intervention arm (ICM) as compared to MNTAU at both 6 and 12 months (p ≤ 0.10).
Table 3

QOLI-20 total and subscale scores at baseline and at 6- and 12-month follow-up by study arm


Study arm


6 monthsa

12 monthsa

Mean ± SD

p valueb

Mean ± SD

p value

Mean ± SD

p value


HN Arms



72.6 ± 21.7


88.4 ± 23.5


89.7 ± 24.5




76.2 ± 21.3


90.3 ± 21.0


92.7 ± 21.4



74.7 ± 21.4


80.3 ± 25.6


83.2 ± 19.4


MN arms



72.2 ± 21.6


88.3 ± 20.5


91.3 ± 20.6




72.8 ± 23.3


82.2 ± 23.2


85.7 ± 23.2



HN arms



14.2 ± 6.1


16.4 ± 6.8


17.3 ± 7.9




15.7 ± 6.2


17.0 ± 6.8


16.9 ± 7.0



14.6 ± 6.0


16.2 ± 7.7


16.6 ± 6.1


MN arms



14.8 ± 6.1


16.3 ± 6.6


16.2 ± 6.4




14.2 ± 6.3


15.1 ± 6.8


16.5 ± 6.8



HN arms



5.8 ± 3.4


6.8 ± 3.5


7.3 ± 3.8




5.8 ± 3.1


6.5 ± 3.3


6.5 ± 3.4



5.7 ± 3.0


5.8 ± 3.2


6.2 ± 3.4


MN arms



5.0 ± 2.9


5.6 ± 3.4


6.8 ± 3.2




5.0 ± 3.3


5.4 ± 3.0


6.3 ± 3.5



HN arms



19.0 ± 6.6


21.7 ± 7.2


21.6 ± 7.4




19.6 ± 7.1


21.9 ± 6.7


23.2 ± 6.2



19.4 ± 6.8


19.8 ± 7.6


20.6 ± 6.3


MN arms



18.7 ± 6.8


21.6 ± 6.9


22.6 ± 6.7




18.6 ± 7.3


21.7 ± 6.9


22.1 ± 7.2


Living situation

HN arms



2.4 ± 1.6


4.7 ± 1.9


4.4 ± 2.0




2.9 ± 1.8


5.1 ± 1.9


5.1 ± 1.9



2.9 ± 1.9


3.9 ± 2.1


3.6 ± 1.9


MN arms



2.4 ± 1.6


5.2 ± 1.9


5.1 ± 1.9




2.5 ± 1.8


3.4 ± 2.1


3.6 ± 2.0



HN arms



15.4 ± 6.6


20.7 ± 6.5


21.0 ± 7.2




15.9 ± 6.1


21.5 ± 5.6


22.2 ± 5.3



16.0 ± 6.1


18.0 ± 6.7


19.1 ± 5.5


MN arms



15.6 ± 5.5


21.8 ± 5.4


22.3 ± 5.0




16.3 ± 6.3


19.2 ± 6.1


19.7 ± 5.6



HN arms



12.0 ± 3.9


13.4 ± 4.7


12.5 ± 5.7




12.2 ± 3.9


13.5 ± 4.1


13.6 ± 4.2



12.2 ± 3.8


12.4 ± 4.4


13.5 ± 3.7


MN arms



12.2 ± 3.9


13.2 ± 4.5


13.6 ± 4.3




12.7 ± 4.5


13.0 ± 4.4


13.3 ± 4.5


Global item

HN arms



3.8 ± 1.9


4.8 ± 1.6


4.8 ± 1.6




4.1 ± 1.7


4.7 ± 1.5


4.8 ± 1.5



4.0 ± 1.7


4.2 ± 1.8


4.4 ± 1.4


MN arms



3.6 ± 1.7


4.7 ± 1.5


4.7 ± 1.6




3.6 ± 1.8


4.5 ± 1.6


4.3 ± 1.6


p ≤ 0.05

** p ≤ 0.01

*** p ≤ 0.001

aNumber of participants across study arms who completed the QOLI-20: baseline (CONG-107, ACT-89, HNTAU-100, ICM-99, MNTAU-99), 6 months (CONG-100, ACT-87, HNTAU-89, ICM-94, MNTAU-83), 12 months (CONG-100, ACT-85, HNTAU-84, ICM-93, MNTAU-84)

bTo compare groups, one-way ANOVA was used for HN groups and independent sample t tests were used for MN groups

Table 4 presents the linear mixed-effect regression analyses that were conducted to estimate the effects of study arm, time and their interaction terms on the QOLI-20 total score with the appropriate TAU group as the reference. For the HN group, study arm × time interactions were significant for ACT and CONG as compared to HNTAU at both 6- and 12-month follow-up periods (p ≤ 0.05). However, for the MN group, interactions were significant for ICM at 6 months (p ≤ 0.05) and a significant trend was observed at 12 months (p = 0.061) as compared to MNTAU. These results are illustrated in Fig. 2, which plots the adjusted mean total score on the QOLI-20 over time for HN and MN study arms, respectively. Figure 2 shows that adjusted total scores between study arms were significantly different at both 6 months (HN: p ≤ 0.001; MN: p ≤ 0.001) and 12 months (HN: p ≤ 0.001; MN: p = 0.006).
Table 4

Linear mixed-effects regression analysis to estimate the effects of study arm on QOLI-20 total score among high need (HN) and moderate need (MN) participants


HN participants (n = 286)

MN participants (n = 185)a

Adjusted coefficient (SEb)

p value

Adjusted coefficient (SE)

p value





 6-month visit

6.46 (2.57)


9.55 (2.08)


 12-month visit

10.07 (2.20)


12.94 (2.70)


HN study arms


−1.38 (3.05)




1.50 (3.09)






Time × HN study arms

 CONG × 6 months

9.17 (3.59)



 CONG × 12 months

6.62 (3.39)



 ACT × 6 months

7.90 (3.34)



 ACT × 12 months

6.77 (3.39)



 HNTAU × baseline



MN study arms



−1.08 (3.28)





Time × MN study arms

 ICM × 6 months


7.38 (3.04)


 ICM × 12 months


6.48 (3.46)


 MNTAU × baseline



Two multivariable models were examined (HN and MN). Includes participants with at least one follow-up visit

p ≤ 0.05

** p ≤ 0.01

*** p ≤ 0.001

aThe multivariable model for MN participants was controlled for housing status at baseline (absolutely homeless vs. precariously housed) and lifetime duration of homelessness (continuous)

bRobust standard errors
Fig. 2

Adjusted QOLI-20 total score (mean and 95 % CI) by study arms

Effect sizes (Cohen’s d) were moderate-to-large for all HF study arms at 6 months (ACT: d = 0.69; CONG: d = 0.70; ICM: d = 0.83) and 12 months (ACT: d = 0.80; CONG: d = 0.70; ICM: d = 0.93) as compared to small-to-moderate effect sizes observed in TAU at 6 months (HNTAU: d = 0.29; MNTAU: d = 0.39) and 12 months (HNTAU: d = 0.56; MNTAU: d = 0.54). It should be noted that, unlike the regression models, effect sizes do not take into account any variation (e.g., heterogeneity within individuals) other than the observed total score and thus provide a cross-sectional view of change [32].

QOLI-20 subscale scores

With regards to subscales, scores on the safety and living situation domains were significantly greater at both follow-up periods across all housing intervention arms as compared to TAU (see Table 3). In addition, participants in the HN housing intervention arms (ACT and CONG) reported a significantly higher score on the leisure subscale at 12 months, and on the global life satisfaction item at 6 months as compared to HNTAU (see Table 3). Further, in the domain of Finances a significant positive trend was observed at both 6 and 12 months among the HN housing intervention arms as compared to HNTAU (see Table 3).

Figures 3 and 4 present the adjusted mean scores from the linear mixed-effects regression analyses for HN and MN study arms on the safety and living situation subscales, respectively. These figures illustrate a significant difference between the subscale scores for both HN and MN housing intervention arms as compared to their respective TAU groups. In other words, all housing intervention arms scored significantly greater than the respective TAU group at all follow-up periods regardless of level of need.
Fig. 3

Adjusted safety subscale score (mean and 95 % CI) by study arms
Fig. 4

Adjusted living situation subscale score (mean and 95 % CI) by study arms


After 6 and 12 months of participation in the study, participants who were randomized to HF reported significantly greater overall QoL as compared to TAU regardless of the type of housing and support they received (i.e., independent apartments with outreach team support or congregate housing with on-site supports) and regardless of their level of need (high or moderate). As compared to TAU, these differences were most notable in the subscale domains of safety and living situation, wherein all HF intervention arms scored significantly greater than the respective TAU group at all follow-up periods regardless of level of need. That is, participants assigned to HF reported stronger feelings of safety in their neighborhoods, and comfort and satisfaction with their living arrangements at both 6 and 12 months post-baseline than did participants assigned to TAU. There was no significant effect of housing type (i.e., scattered site or congregate) on overall or subscale scores. Our findings demonstrate that improvement in subjective QoL is seen within the first 6 months of receiving HF and continues to increase (albeit not as much) from 6 to 12 months.

Notably, the TAU study arms (no housing or support provided through the study) also reported a sizeable increase in QoL over the 12-month study period. Since the 2010 Winter Olympics in Vancouver, there has been a collective effort to improve housing and services for chronically homeless individuals. After 12 months, participants in both TAU groups spent approximately 20 % of the period in stable housing (i.e., residences expected to last longer than 6 months where the individual had tenancy rights and no current plans to move) and 80 % in various other settings (i.e., unstable housing, institutions, shelters, street places). Participants in the HF groups (ACT, CONG, ICM) spent approximately 80 % of the same period in stable housing and 20 % of the time was spent in other settings. However, there were no significant correlations between participants in HNTAU or MNTAU who received stable housing and higher QOLI-20 total score. It is also known that repeated administration of cognitive tasks such as questionnaires can result in improvement due to a familiarity effect or perhaps due to a demand characteristic inherent in the interviewer-participant relationship [35].

No guidelines around minimal clinical differences have been published for the QOLI-20 or any of its earlier versions. However, responsiveness on self-report QoL scales is typically calculated using effect size (Cohen’s d) [3133]. It is widely recognized that, in the absence of more specific information, 0.5 SD represents minimally important change in self-reported QoL among individuals with chronic disease [31]. Effect sizes for the QOLI-20 observed total score were all moderate-to-large for the HF study arms and low-to-moderate for the TAU study arms. These indices further support our findings that the QOLI-20 is sensitive to changes in QoL that result from receiving good quality housing and support. However, the clinical significance of these findings in terms of functioning and engagement across a variety of domains requires further research.

Our findings corroborate those of Lehman et al. [20] who found no difference in subjective QoL between participants living in group homes and supervised apartments, as well as findings by Oliver and Mohamad [22] who found no difference in subjective QoL in three different community residential settings. This pattern of results suggests that the differences between alternative types of supported housing are not as important to QoL as having safe, good quality housing. Past research suggests that some individuals may be predisposed to react differently to group versus independent living [23], suggesting that individual expectations and attitudes play a key role in shaping general dimensions of satisfaction, apart from the type of residence itself. However, it may also be that the QOLI-20 is not sensitive enough to discern differences among various community residential settings.

The fact that the greatest improvement in QoL occurred within the first 6 months following randomization to HF supports previous research that suggests that the first few months of community living are crucial for adjustment [36]. The housing provided through the Vancouver At Home Study was of good quality and was accompanied by substantially higher levels of support than are available to most other housing programs in the local area. The ACT team had a client-to-staff ratio of approximately 9:1 while the ICM team had a ratio of approximately 16:1. To date, ACT is not widely available in Vancouver and existing case management services often have client-to-staff ratios exceeding 25:1. Therefore, the level and quality of outreach support available to participants in our study likely contributed to the observed improvements in QoL among HF participants. Furthermore, in all HF intervention arms, participants were offered considerable choice. Those in scattered site apartments (ACT and ICM) could choose from housing in several different neighborhoods throughout the city. Participants assigned to congregate housing (CONG) were offered some choice around what room they were assigned and the floor they lived on. A perceived sense of control over one’s environment and life circumstances is often considered an important aspect of the QoL concept [14]. This may be particularly important to people who are homeless and have mental disorders as they typically have few choices in their lives and the opportunities and choices they do have are often greatly constrained by structural factors such as poverty and inequity [37].

It is not surprising that Safety and Living Situation were the subscales that were significantly greater than TAU across all HF intervention groups and time periods. Good quality housing clearly impacts one’s sense of safety and living arrangements more directly than the domains covered by other subscales such as relationships with family, social ties, and leisure time. This finding supports past research that suggests that obtaining good quality housing affects housing-related QoL specifically rather than QoL more generally [14]. This is an important reminder that, while good quality housing is critical to the stabilization and wellbeing of homeless people with mental illness, it is by no means sufficient to promote recovery in all areas of life. Domains such as family and social relationships, leisure activities, and work are likely determined by housing quality, and complex interactions among individual, community and societal factors [6]. For people who have been homeless and who experience mental illness, recovery in these social domains likely requires more time than is necessary to accomplish meaningful change in one’s housing arrangements. However, significant differences between the HN housing intervention arms and HNTAU in the domain of Leisure (at 12 months) and in global satisfaction (at 6 months) as well as a trend toward significant differences in the domain of Finances at both 6 and 12 months suggest that other QoL domains respond to improved housing quality and stability in the relatively short term.

Despite the strengths of our sample (e.g., randomized controlled trial, high retention rates, structured diagnostic interviews, validated QoL measure), several limitations bear on the interpretation of our results. First, our HF interventions included a combination of good quality housing and support services, therefore, we could not dismantle the contributing effects of housing vs. support services. However, very few studies have randomly assigned homeless individuals with SMI to supported housing versus TAU. Given that group assignment cannot be blind, it is possible that participant responses to the QOLI-20 at follow-up may have been influenced by this factor. Furthermore, it is not clear whether our findings generalize to congregate residences in geographic locations different to the one in our study. Participants were assigned to one congregate residence with embedded multi-disciplinary supports so we were not able to compare the effects of different locations or the impact of choice of location, different service models, or tenant mix on QoL for this model of housing. Finally, interventions that produce objective improvements such as good quality housing may lead to a transient decrease in life satisfaction as individuals become aware of the discrepancy between their current circumstances and how things could be [20, 38].

Future research could explore how subjective QoL relates to behavioral observations, service use, societal factors such as stigma and discrimination based on class, and associated costs. Further work could also examine objective assessments of housing quality and their relationship to subjective QoL. Researchers associated with the current study have developed such an objective assessment (Adair et al., unpublished manuscript), which will be implemented at multiple sites. While subjective QoL measures often do not correspond to objective life circumstances, they are important in their own right because they reflect the perspectives and interpretations of affected individuals, which are likely to influence the success of any attempted change [14]. For service systems and individual agencies, objective indicators such as housing quality and stability, access to treatment, improvements in health and changes in social role functioning may be the most important domains to measure. However, in order to understand why some individuals do not accept housing or supports that are offered or later drop out of services, subjective measures are critical. The combination of subjective and objective measures can lead more directly to service and public health improvements that are sensitive to peoples’ needs and may help clarify the mechanisms by which the components of complex service interventions have both positive and negative effects on people’s lives.


Support services were available to participants assigned to Housing First (ACT, CONG, and ICM) but were not mandatory. The only requirement for housing was compliance with the terms of the rental lease and weekly visits with a case manager to ensure safety and well-being [28].


The CONG residence was located in downtown Vancouver in a neighborhood consisting of primarily businesses, including an inner-city hospital and a number of affluent condominiums.



Special thanks to the Vancouver At Home service teams who are working with participants to increase their QoL, the Vancouver field research team, and the participants who gave their time to answer our questionnaires. The authors also thank the At Home/Chez Soi Project collaborative at national and local levels; National project team: J. Barker, PhD (2008–2011) and C. Keller, National Project Leads; P. Goering, RN, PhD, Research Lead; approximately 40 investigators from across Canada and the US; five site coordinators; numerous service and housing providers; and persons with lived experience.

Conflict of interest

This research was funded by Health Canada and the Mental Health Commission of Canada. The views expressed herein solely represent the authors. The authors declare no conflicts of interest.

Copyright information

© Springer-Verlag Berlin Heidelberg 2013