We conducted a retrospective mirror-image study with anonymised medical record data of ISH service users. The individual start date of ISH was defined as index. Number of events defined below (see outcomes variables) were compared between post-index (during intervention) and pre-index (before intervention) periods of equal length.
ISH is a community-based outreach housing rehabilitation service provided since 2012 by the Centre for Psychiatric Rehabilitation of the University Hospital of Psychiatry in Bern, Switzerland . It follows the principles of the “Housing First” approach  as it is independent of service users’ therapy and care and is not transitional, but permanent and without time limitation. ISH addresses adult people with SMI and provides its users with psychosocial support in their independently rented accommodation. The main goals of the ISH intervention is the social inclusion of its users, including fostering their autonomy and personal recovery. According to the Simple Taxonomy for Supported Accommodation (STAX-SA; ) the intervention classifies as a Type 4 service as service users live in individual accommodations with no staff on-site and the intervention provides low to moderate support with no time-limitation.
Support services are provided up to 8 h per week by non-medical staff with nursing or social work training. An offsite residential coach supports the service users according to their needs in all aspects related to finding and keeping ones’ accommodations. This may include contacts with the landlord, social environment, administration, and cooperation with mental health services. Service users also have the option to consult an ISH psychiatrist. The psychiatric, psychotherapeutic and medical treatment of service users takes place outside the ISH service by appropriate specialists.
Model fidelity of the ISH intervention was assessed in 2019 as part of another study  using the MSSW model fidelity scale (Modelltreue-Skala Selbstbestimmtes Wohnen [Independent Supported Housing Fidelity Scale]; ). Fidelity was high with 141 out of a possible 155 total sum score (m = 4.5) and subscale scores of m = 4.67 (housing conditions), m = 4.5 (staff/team), m = 4.4 (support conditions), and m = 4.83 (inclusion orientation) out of a possible mean score of five.
From the medical records, we extracted every ISH utilisation period with start date between July 2nd 2016 and February 28th 2019. The start date was chosen because the patient medical records were retrospectively introduced into the medical records system on July 1st 2016 and therefore lack some information about ISH utilisation before this implementation. The end date was chosen because an augmented Home Treatment/Crisis Resolution Treatment program was implemented in the ISH programme in March 2019. We included all ISH utilisation periods within this time window, if it was the service user’s first utilisation. If a service user had multiple ISH utilisations within this time window, we included the first utilisation period and excluded the latter. In addition, ISH utilisation periods of less than 30 days (all of them had a start date after January 2019) were excluded to increase the probability that included service users actually received support within the observation period. The included ISH utilisation periods are either limited by withdrawal from the program (case finalisation for any reasons) or by censoring on February 28th 2019 in case of ongoing ISH use.
The cantonal ethics committee of Berne, Switzerland reviewed the study and confirmed that approval of an institutional review board was not required (Req-2021–00042, January 2021).
Measures and source of information
Outcome data was retrieved from patient medical records of the psychiatric hospital. Outcome variables were extracted for each pre- and post-index mirror-image period defined below (see statistical methods) and include the number of inpatient psychiatric admissions and the length of inpatient psychiatric hospital stays, defined as the number of person-days hospitalised (including censored stays, e.g., with admission before a mirror-image period’s start).
The medical records were used to retrieve sample characteristics and consist of service users’ demographic information: age (in years), sex (female, male), nationality (Swiss vs. non-Swiss), and civil status (single vs. married, divorced, widowed). Clinical information of the main psychiatric diagnosis category according to the ICD-10 classification of mental and behavioural disorders  was also obtained.
Sample characteristics and hospitalisation patterns are reported descriptively.
The primary analysis was a mirror-image analysis of the outcome measures defined above. Post-index outcomes were compared with pre-index outcomes in each mirror-image period. The maximum period length for each service user was defined as the individual ISH utilisation period as described above (see sample). In addition, we defined mirror-image periods of 90, 180, 270, and 365 days to assess the possible influence of the different utilisation period lengths. Service users could be included in several mirror-image periods if their utilisation period covered the entire period. The change of psychiatric hospitalisations from pre- to post-index was analysed by computing incidence rate ratios (IRR) with 95% confidence intervals (95%-CI). In line with other self-controlled studies [24, 25], a simple analysis was conducted that did not account for the fact that users were observed under two conditions (before and after ISH initiation). This approach was adopted to circumvent the issue of zero events i.e. in psychiatric admissions. This is a conservative approach that exaggerates the magnitude of standard errors .
We further conducted sensitivity analyses to address the potential of a regression towards the mean [14, 15, 18], which could have led to an overestimation of ISH effects. Since this bias is assumed to be more strongly affected by pre-index outcomes occurring close to the index rather than by long-term outcomes , we reanalysed every mirror-image analysis after excluding all ISH users who had a psychiatric admission within 90 days before index.
All statistical analyses were performed with R version 4.0.3  and the fmsb package for computing IRR . Statistical significance was set at p < 0.05 for all analyses.