We applied a quasi-experimental design where individuals discharged from one mental health center (MHC) were offered the SAFE intervention and compared to a propensity-score-matched sample of individuals discharged from other MHCs in the same region. The study may be considered as a natural experiment because individuals were included into the study if they lived in the catchment areas of the studied MHCs. The catchment area consisted of eight MHCs within the Capital Region of Denmark. Each center served as a defined catchment area. The SAFE intervention was implemented at MHC Copenhagen, while four adjacent MHCs, MHC Ballerup, MHC Glostrup, MHC Amager, and MHC North Zealand, were used for the comparison. The three other MHCs in the Capital Region of Denmark were excluded from the study; MHC Bornholm, which was situated on an island two hours from Copenhagen city with a different population and level of urbanicity; Psychotherapeutic Center Stolpegård, which exclusively provided outpatient treatment; and MHC Sct. Hans, which was a specialized center for treatment of forensic patients and patients with dual diagnosis. The SAFE project was implemented from March 1, 2018 to March 31, 2020. Information on individuals discharged from the included psychiatric hospitals was obtained from national hospital registers, in which all in-, out-, and emergency department patients seen at psychiatric hospitals in Denmark since 1995, have been recorded. Individual-level data on discharges prior to 2. February 2019 were obtained from the Danish Psychiatric Research Register [16]. After this date, data on all discharges were derived from the National Patient Register. Each hospital in Denmark has an id number, and this id number was used to identify inpatients discharged from each of the five MHCs [17]. Psychiatric diagnoses were classified according to the International Classification of Diseases, version 10 (ICD-10) [18].

We obtained data on all patients, aged 18 or above, discharged from one of the five MHCs. Each discharge was considered as a separate unit of follow-up, i.e., if a patient was admitted to psychiatric hospital several times during the follow-up, each discharge would be analyzed separately in the study. Patients were excluded if they had been recorded with primary diagnoses of one of the following disorders: organic psychiatric disorders (ICD-10: F00–F09), eating disorders (ICD-10: F50–F59), mental retardation (ICD-10: F70–F79), pervasive and specific developmental disorders (ICD-10: F80–F89), and behavioral and emotional disorders with onset usually occurring in childhood and adolescence (ICD-10: F90–F98). Individuals with these disorders (n = 361) were excluded because they attended specialized outpatient treatment.

The SAFE intervention consisted of the following three components, which were systematically implemented: (1) Face-to-face introduction to SAFE prior discharge where eligible inpatients at the intervention site were invited to participate in the project during a personal meeting with a health care provider from the outpatient clinic, while still being in admission. During this first meeting, the health care provider from the outpatient clinic interviewed the patient regarding their home environment and together they complied a safety plan as well as scheduled a face-to-face meeting for first week after discharge. The health care providers professional backgrounds were social workers, nurses, occupational therapist or psychologists.; (2) A face-to-face meeting during the first week after discharge was held between patient and, ideally, the same health care provider, which they met at the hospital ward. The meeting was preferably conducted as a home visit, but patients could also request to meet in different venue, for instance, the outpatient clinic or at café. The home visit allowed the clinician to assess the patients’ post-discharge situation, i.e., identify possible stressors, monitor the plans for follow-up, review the safety plan, and reassess the suicide risk as well as inform the patient about suicide warning signs; and (3) Involvement of relatives where patients were encouraged to involve relatives in their treatment course as relatives can provide emotional and practical support to the patient in a vulnerable situation. If involved, the health care provider would inform the relatives on suicide warning signs, as this might foster faster help-seeking if the patient’s condition should deteriorate.

The SAFE intervention was compared with treatment-as-usual offered at the comparison sites, which depending on severity of condition and disorder consisted of reference to a variety of services. These services could be referral to a private psychiatrist, substance abuse treatment, psychotherapy or follow-up care in secondary health care and the availability of appointments to these services are often several weeks due to waitlists. Also, some were referred at discharge to contact their general practitioner when at home. The overall difference to the intervention was that in SAFE, everyone, no matter of condition or disorder, was systematically offered to receive a face-to-face follow-up in the week after discharge where the risk of suicide is really high whereas in treatment-as-usual many did not have a follow-up due to waitlist problems.

The outcomes of interest were suicide and suicide attempt, respectively. Suicide deaths were identified in the Cause of Death Register as one of the following ICD-10 codes: X60–X84 or Y87.0 [19].

Suicide attempt was defined as a presentation to either psychiatric or somatic hospitals, including emergency departments where a suicide attempt had been recorded using the following ICD-10 codes X60–X84 or when reason for contact was indicated as suicide attempt in both The National Patient Register and The Danish Psychiatric Research Register. In addition, the following combinations of ICD diagnoses were included as hospital contacts due to suicide attempt: a main diagnosis of a mental disorder (ICD-10: F00–F99) together with one of the following sub-diagnoses: S51, S55, S59, S61, S65, S69 (cutting by sharp objects), T36–T50 (poisoning by pharmaceuticals), T52–T60 (poisoning by non-pharmaceuticals) as well as all admissions with a main diagnosis of T39, T40 (poisoning by mild analgesics; except T40.1), T42, T43, and T58 (poisoning by opioids, psychotropics, and by carbon monoxide).

Participants were followed from date of discharge until episode onset of the studied outcome, new psychiatric admission, death, emigration or six months post-discharge, which ever occurred first.

Based on estimates on previous population attributable risk calculations [20], approximately 40 suicidal acts in the first week after discharge and approximately 200 in the first 6 months after discharge were expected. With approximately 6000 discharges, our power calculation (based on a statistical significance level of 5 percent and a power of 90 percent) shows that we have the power to find a difference in suicidal acts of 7% versus 5% between the control and SAFE sites, corresponding to a 28 percent reduction in suicidal behavior.

Individuals discharged from the SAFE intervention site were matched to individuals from the comparison group using exact matching and a propensity score. The unit for the matching was discharges not individuals, hence, if an individual was discharged more than once from the intervention site then a matching was conducted for each discharge. The propensity score was derived from following socio-demographics and health-related variables from nationwide registers: sex (male, female); age group (18–29 years, 30–44 years, 45–64 years, 65 + years); socio-economic status (working, unemployed/social or sickness benefit, retired, missing); highest achieved educational level (elementary school, vocational training, high school, bachelor degree or higher, missing); civil status (married, divorced, never married, missing or other); placed in foster care; primary diagnoses of substance misuse (ICD-10: F10–F19), schizophrenia spectrum disorders (ICD-10: F20–F29), affective disorders (ICD-10: F30–F39), anxiety or stress-related disorders (ICD-10: F40–F49) or personality disorders (ICD-10: F60–F69); admission lasted more than 7 days; ever diagnosed with substance misuse; ever diagnosed with schizophrenia spectrum diagnosis; ever diagnosed with affective disorders; previous suicide attempt; prescribed antipsychotic medication within 12 months of being discharged; prescribed antidepressant medication within 12 months of being discharged; parents with history of psychiatric disorder; and parents with history of a suicide attempt or suicide. Date of discharge was used as time point for the matching and all matching factors were measured on this date. Matching factors were selected because they were considered as predictors of suicidal behavior or factors, which might have described differences between the intervention and comparison groups. An exact matching was prioritized for previous suicide attempt and prescribed antidepressant medication within 12 months of being discharged and a propensity score was calculated for the remaining variables. After having identified a matched comparison group with a 1:1 ratio, odds ratios (OR) of suicide and suicide attempt were calculated in separate models with their 95% confidence intervals at 2 weeks, 1 months and 6 months after discharge. We accounted for multiple discharges of the same individual through adjustments for repeated subjects in the statistical models. Following sensitivity analyses were conducted: Firstly, the first months of the SAFE intervention (from March 1, 2018 to August 31, 2018) were omitted, implying that only discharges after September 1, 2018 onwards were included, to ensure a higher level of adherence to the SAFE intervention, as it might have taken time to implement new procedures into clinical practice. Second, we excluded all patients discharged with a main diagnosis of depression or bipolar disorder, as members of this group were enrolled in a congruent research project during half of the follow-up and, thus, precluded from participating in the SAFE intervention. Third, individuals were censored after a first recorded suicide attempt to test that the risk estimate for suicide attempt in the primary analyses would not be biased by individuals with multiple suicide attempts.

All analyses were conducted using SAS version 9.4 and the cut-off for statistical significance level was set at p = 0.05.