The Supporting Transitions and Recovery Learning Centre (STAR) was launched in April 2014 at St. Michael's Hospital, an urban academic hospital in Toronto, Canada. Toronto is home to over 7,000 homeless people each night, with over 27,000 individuals using homeless shelters each year (21, 22). Toronto is also a service rich setting, in which physician, hospital services and community support services, including mental health services, are offered at no cost to individuals. In addition to health services for those experiencing homelessness, drop-in centres, social services and clubhouses offer recreational and vocational supports to this population.

This study used a pre-post quasi-experimental design. Although a randomised controlled trial (RCT) was initially considered, it was thought to compromise the choice options of a disadvantaged population and therefore not pursued at this time, given the centrality of choice in REC philosophy. A total of 92 individuals with mental health challenges and recent histories of homelessness were recruited to each of the intervention and control groups from January 2017 to July 2018 and followed for 12 months. Control group participants were frequency-matched to REC participants on age and self-identified gender (14). Data were collected at the time of study enrolment (baseline) and 6 and 12 months post baseline. To be eligible, intervention participants had to have registered in the REC during the recruitment period and met REC eligibility criteria: (i) experienced challenges with maintaining housing within the past 2 years and (ii) being over 16 years of age (14). Control group participants were individuals with no prior or current participation in the REC and were recruited from community agencies that provide non-clinical, low barrier social, recreational, and vocational supports for this population (e.g., drop-in centres, clubhouses, and community centres). Given the inclusive nature of Recovery Education Centres, diagnostic information about mental health or substance use challenges was not required. A detailed description of the study protocol has been published (14).

A sample with at least 92 participants in each of the intervention and control groups provided 80% power to detect a medium effect size (Cohen d = 0.5) for the primary outcome, perceived empowerment, at 12 months, using a 2-sided t-test, assuming an attrition rate of 30% over the follow-up period. This was calculated using estimates from a quasi-experimental evaluation of a REC in Boston (13) that observed a difference in mean changes of 0.08 between the intervention and comparison groups from baseline to 12 months with an effect size of 0.41.

The study received Research Ethics Board approval by St. Michael's Hospital and the Centre for Addiction and Mental Health. Participation in the study was entirely voluntary. Prior to their baseline interview, individuals were asked for written informed consent. Capacity to consent was assessed by trained research staff who had experience working with this population. Specifically, together staff and participants reviewed a document with information about the study for participants to sign. When reviewing the consent form, interviewers were trained to regularly inquire if participants had questions. When a staff was uncertain about a participant's comprehension of the consent form, a consent checklist was used to confirm this comprehension. In addition, if a participant could not read or was visually impaired, another staff member was brought in to witness the consent portion of the interview and the participant's signature of the declaration of assistance.

Study participants received a $30 honorarium for attending each interview.

STAR is among few RECs worldwide with a mandate to support people transitioning out of homelessness, and to our knowledge, unique in North America. Individuals could self-refer to STAR or be referred from hospitals, primary care clinics, as well as local shelters, community organisations and other settings serving this population. They were required to register and develop individualised learning plans upon enrolment.

STAR classes and workshops were taught by peers with lived experience of housing instability and mental health challenges, as well as social and health service professionals. Classes focus on life skills related to transitioning from homelessness to housing and other topics such as health and wellness, vocational skills, leadership and community engagement, hobbies, and interests. While operated by a hospital, the central classroom and staff offices were located at a community centre and additional participation in learning opportunities is offered in partnership with other organisations (e.g., Employment Services Centre, Libraries).

The primary outcome for this study was the change in self-reported personal empowerment from baseline to 12 months. Empowerment is a key dimension of recovery (23) and has been positively associated with perceived quality of life, self-esteem, and social support, and negatively associated with the severity of psychiatric symptoms (24). It was assessed using the Rogers Making Decisions Empowerment Scale (25). This measure has been used in other studies of individuals with mental health and addiction challenges (13), and has adequate psychometric properties, including internal consistency (alpha = 0.86) and factorial validity (26).

Secondary outcomes captured changes from baseline to 12 months in other domains inherent in the multidimensional view of recovery (see Table 1). These measures include:

For all scales, higher scores indicate greater recovery outcomes. Demographic variables that were collected at baseline were age, self-identified gender, ethnicity, employment, education, duration of longest period of homelessness, and housing stability over the past year, defined as the percentage of days stably housed. Housing stability was ascertained using the Modified Residential Time-Line Follow-Back Inventory that has high test–retest reliability and good concurrent validity in homeless populations (33).

Sociodemographic characteristics were compared for intervention and control group participants using the two-sample t-test or Fisher's exact test. To assess change in scores from baseline to 12 months, as per the protocol analysis (14), linear mixed models with random intercepts were applied to perform repeated measures analysis of perceived empowerment and secondary outcomes (all continuous). An exchangeable correlation structure was used. The model included the main fixed effects of group (intervention vs. comparison), time (12 months vs. baseline), and the intervention × month interaction, adjusting for participant's age at baseline, self-identified gender, ethnicity, education, longest period of homelessness, and housing stability over the past year (a priori analysis). For each outcome, model-estimated mean changes from baseline to 12 months follow up and 95% CIs were examined. Differences in mean changes, 95% CIs and p-values were calculated to compare the groups (34).

We conducted an “as-treated” post-hoc analysis to examine the effect of participation, as has been done in other studies on recovery services (35). This involved examining the effect of the number of hours of REC participation, rather than group membership. There is no precedent for the minimum “dose” of REC participation (i.e., number of sessions attended, or amount of time spent attending) needed to support recovery in RECs. As such, we consulted a review of 23 studies on similar interventions (i.e., peer led group sessions for participants with serious mental illness) and with similar designs to the present evaluation (longitudinal design, recovery outcomes of interest) (36). Of 23 studies, seven reported the mean number of sessions that participants attended [mean = 6.5 sessions or 13 h (6.5 × 2 h)].

To examine the effect of varying levels of participation on changes in primary and secondary outcomes from baseline to 12 months, we created three intervention subgroups based on hours of REC participation: REC registrants with no hours of participation, and therefore no exposure to the intervention (n = 18); (ii) REC registrants with 1–13 h of REC participation (n = 37), and (iii) REC registrants with 14 or more REC hours of participation during the 12 month follow up period (n = 37). The mixed effects model described above was applied again for this analysis, with the main fixed effect of group replaced with the hours of participation variable (0 h, 1–13 h, 14 or more hours, vs. control group).

Most scales had little missing data, or none (see Table 1). All statistical tests were two-sided, and a value of 0.05 or less indicated statistical significance. R software (version 3.4.1) was used for all analyses (37).

Although a randomised controlled trial (RCT) could have yielded stronger findings, pragmatic considerations advised against an RCT design at this time (13, 14). While the post-hoc analysis showed that REC participation for 14+ h was associated with improvements in perceived empowerment and mastery, it is notable that the intervention subgroup with 14+ h of participation in the REC was older, more likely to be white, to have attended University or post-secondary education, and less likely to be employed than the control group. Although multivariable models were adjusted for participant age, self-identified gender, ethnicity, education, longest period of homelessness, and housing stability over the past year, it is possible that greater participation in RECs could be associated with unmeasured individual characteristics that result in greater increases in self-reported empowerment. Additionally, the dose threshold of 14 h for recovery education was determined by a review of studies that examined the published the number of sessions needed to support recovery. However, the mean number rather than median number of sessions was used so it is possible that the dose determination may have been overly skewed. An additional limitation is that the study, similar to other intervention studies (18, 42) did not capture additional services used by the intervention or control groups. This lack of information is particularly salient, as both the intervention and control groups experienced improvements over time. Another limitation is that in the field of patient-reported outcomes, such as empowerment, there are not standards established as to how to measure minimally important differences in outcomes (43).

Despite these limitations, this study has notable strengths compared to previous studies of REC outcomes, plagued by small sample sizes, and non-experimental designs. Importantly, the present study used rigorous quasi-experimental methods, focusing on a disadvantaged population experiencing multiple barriers to recovery, and included a control group of participants engaged with other community services. Another strength is the 12 month longitudinal follow-up period. In the review (36) of 23 longitudinal studies with similar outcomes cited earlier, only four had follow up periods of a year or more. Finally, grouping REC participants by hours of participation allowed for a more nuanced understanding of the potential impact of hours of participation on recovery outcomes.