BACKGROUND AND AIMS: This study evaluated a three-year residential rehabilitation programme, which aimed to support recovery from problem substance use via peer support and social enterprise activities. The aims were to clarify programme mechanisms and identify contextual factors associated with variation in outcomes.

METHODS: The study took place within River Garden, a residential rehabilitation for problem substance use, based in South Ayrshire, Scotland. A mixed-methods realist-informed process evaluation was undertaken, using participant observation, repeated qualitative interviews and routinely collected admissions data. Fieldwork was conducted with residents, staff and trustees between April 2019 and November 2020. Nine (of ten) residents were recruited into the study. All residents were male, aged 20-47 years (median 35 years) and were White Scottish or English. Data collection and analysis was guided by Medical Research Council guidance on process evaluation and informed by selected principles from realist evaluation.

RESULTS: Three key contextual factors and six key mechanisms were associated with variation in resident outcomes. The severity of residents' substance use problems, their physical and mental health and their socioeconomic backgrounds shaped whether they responded to the programme's instrumental and relational resources with trust, respect or motivation (constituting six mechanisms, e.g. instrumental-respect, relational-trust). The programme was most beneficial for residents for whom intended outcomes were less constrained by contextual moderators.

CONCLUSIONS: In residential rehabilitation for substance use disorders, residents with higher problem severity, worse physical and mental health and greater socioeconomic disadvantage appear to be less likely to respond to rehabilitation resources with trust, respect or motivation compared with the other residents. These findings may support the development of strategies to improve outcomes for residents with greater contextual barriers to change.