We searched the following databases: MEDLINE, Embase, Cochrane Library, CINAHL, PsycINFO, and Psyndex. Documents published between 1998 and May 31, 2018 were included in our search strategy. Database-specific searches included the following index terms: (“bridging” OR “transitional care” OR “patient discharge” OR “discharge planning”) AND (“Psychiatric Hospital” OR “Mental Institution” OR “Mental Hospital”) AND (“intervention” OR “programme” OR “preparation”) NOT (child* OR dement*). Additional searches were conducted in Google and Google Scholar to find relevant grey literature. Hand searching of the references of key papers (e.g., existing systematic reviews and included articles) complemented the search. The detailed search strategy is available upon reasonable request.

To be included, studies had to meet the following criteria: published studies or study protocols written in German or English, participants aged 18-65 years, participants had a psychiatric diagnosis and were discharged from a psychiatric inpatient unit. Included interventions were those that aim to improve discharge from psychiatric inpatient care to home with a combination of predischarge and postdischarge components (e.g., need assessment or development of discharge plan predischarge and home visits or telephone contacts postdischarge). Moreover, all components of the intervention must have emanated from the inpatient setting. Finally, although randomization is desirable to minimize selection bias, it may not be feasible in mental health care. We therefore included nonrandomized or one-group study designs into our qualitative synthesis.

The following exclusion criteria were defined: interventions or programs on a structural or organizational level; psychotherapeutic treatment programs with a specific focus (e.g., therapy for substance dependence or medication adherence); interventions specifically targeting homeless persons; participants with physical or mental handicap; and discharge from forensic settings.

Titles and abstracts of studies retrieved from database searches and additional sources were independently screened by two authors (AH and BK) to identify those that met inclusion criteria. The full texts were then retrieved and independently assessed for eligibility by the same two authors. Disagreements over the eligibility of any studies were resolved through discussion.

Data extraction from the included studies was carried out by one team member (AH) and checked for accuracy by another (BK). Extracted information included: study design, setting, and population; details of the intervention and control conditions; recruitment and study completion rates; outcomes and times of measurement; and information for assessment of the risk of bias. Missing or nonreported data on the primary outcome (readmission rates) was requested from study authors.

To classify the intensity of the interventions, we adapted the intervention intensity score developed by Holzinger et al. (22). The score (low/moderate/high intensity) considers the length of intervention, number of intervention components (e.g., psychoeducation, cognitive behavioral therapy [CBT], case management, and peer support), and number of contacts with the patient. Details of our adaptations and ratings are available from AH.

Two authors (AH and BK) independently assessed the risk of bias in the included studies by using the Revised Cochrane Risk-of-Bias for randomized trials (26) and the Mixed Methods Appraisal Tool for all other study designs (27). Discrepancies were solved by discussion within the team.

Studies were included in the present meta-analysis if data were available on readmission rates in absolute numbers or percentages. The random effects meta-analysis was conducted with the package ‘meta' (version 4.9-5) in R statistical software (version 3.6.1; R Core Team, Vienna, Austria). Odds ratios were calculated for effect sizes. Study heterogeneity was assessed by I², and 95% confidence intervals (CIs) were used to assess uncertainty. Forest and funnel plots were used for the graphical display of effect sizes and analyzing publication bias, respectively. Subgroup comparisons were conducted for length of intervention, risk of bias, and intervention intensity.

Meta-analyses of other outcomes were not possible due to the low number of studies and heterogeneous assessment instruments.

The search strategy yielded 2,673 publications, which after screening resulted in 16 included studies. There were 10 RCTs, three quasi-experimental studies, and three cohort studies. From this, nine RCTs were included in the meta-analysis ( Figure 1 ). In the RCTs, the percentage of participants with a diagnosis of psychosis or schizophrenia varied between 8% and 73% and between 26% and 75% for participants with depressive or anxiety disorders. In the cohort studies, a strong focus on patients with psychosis or schizophrenia was apparent (range 87%-100%). Study characteristics are shown in Table 1 .

The included studies tested 15 different interventions ( Table 1 ). All interventions included multiple components and were conducted by mental health workers, nurses (a portion with a master's degree), case or care managers, social workers, or peer support workers.

Predischarge interventions included components from case management, such as needs assessments (28, 29, 32, 34, 35, 40–43), discharge or care planning (e.g., crisis planning) (28, 31, 35), scheduling or preparing for follow-up appointments (39, 42), and family or carer involvement (28, 44); psychoeducational components, such as individualized psychoeducation (33) and medication reconciliation elements (29, 44); and elements of CBT, such as skills training (29, 40, 41, 44) and peer support (38, 43). Interventions were delivered in one-to-one sessions, except for the study conducted by Khaleghparast et al. (41), where a family member was present and in Noda et al. (44), where skills training was delivered in groups.

Postdischarge components aimed to support patients during a transition period and were most frequently delivered through phone calls, home visits, or letters. Components associated with case management included: efforts to ensure timely follow-up with outpatient care providers (28, 32, 34, 39, 42), treatment coordination (28, 31, 39), timely communication between inpatient staff and outpatient care or community service provider after discharge (29, 34), monitoring of health status or implementation of postdischarge plan (31, 33, 35), and activation of resources in the social network (32). Elements of CBT consisted of therapeutic meetings with staff (30, 37) and skills training (33, 40, 41, 44). In addition, interventions included psychoeducation and counseling (31) or peer support (30, 37, 38, 43). Peer support consisted of facilitating access to local communities, promoting friendship, providing basic necessities, understanding, and encouragement. Similar to the predischarge components, postdischarge contact took place in sessions between patients and mental health workers, occasionally with explicit inclusion of important members of the patient's social network (32).

Interventions lasted between 1 week (40) and 2 years (33), or until a therapeutic relationship was established between the patient and outpatient care provider (30, 37); however, the majority of the interventions ended 3 months after discharge (29, 31, 32, 35, 39, 41).

Control groups received treatment as usual, which included aftercare or treatment planning (39, 40), referral to outpatient treatment (28, 29, 33, 37, 39), arrangements from community mental health services (38), and case and medication management (31). Treatment as usual in the study by Puschner et al. (35) was delivered without a manualized or structured discharge plan. Assistance from inpatient units ended after the patient was discharged, except in the study by Cuffel et al. (39). Finally, three studies did not specify treatment as usual (30, 34, 41) and an additional three studies did not apply a control group design (42–44).

Details on the quality assessment of the included studies are displayed in Tables 2 and 3 . From the 10 RCTs included, only the study by Hengartner et al. (32) proved to have a low risk of bias. Five RCTs were rated to have some concerns (28, 29, 31, 33, 35) and four studies (30, 34, 37, 38) were rated with high risk of bias. Khaleghparast et al. (41) did not describe a randomization process and was therefore included and rated as a study using a quasi-experimental design. Risk of bias was evident primarily in non-RCTs and studies with a one-group design due to incomplete outcome data (i.e., high drop-out rates) and potential confounders that were not accounted for.

The present review demonstrates clear limitations associated with many of the included publications. Most studies had small sample sizes and were likely underpowered to detect clinically relevant effect sizes. The overall risk of bias was high in four of the 10 RCTs. In addition, the high heterogeneity of intervention components limited their comparability and our ability to generate practice and policy implications. More detailed classification systems are needed (e.g., including characteristics of health/social care system, risk of readmission in certain populations, and quality of inpatient care) to better understand the holistic value of transition interventions.

We limited the included populations by excluding studies where we expected considerable differences during the inpatient stay or unique discharge situations (e.g., long inpatient stays in forensic settings (23, 50) or discharge into homelessness). Nevertheless, the comparability of the included studies was limited by a broad range of interventions and a high number of diverse outcomes and outcome measurements. In addition, we observed diverse definitions of “treatment as usual”, which may limit the significance of the results.

We included studies with participants across various psychiatric diagnostic groups. This did not considerably add to the heterogeneity of the review; however, a quantitative subgroup analysis was not possible. Therefore, future research is recommended to assess the effects of transitional interventions on various psychiatric disorders.

The high risk of bias in the selected RCTs further underlines the limitations of the present study design. Therefore, the synthesis of our present systematic review offers an overview of their efficacy, while not confirming it.

Through rigorous search methods in different databases and hand search of bibliographies, we attempted to identify all possible eligible studies. The funnel plot did not indicate possible limitations regarding publication bias. By including nonrandomized and cohort studies, our present review highlights the value of pragmatic study designs in this setting.

We observed that transitional interventions with bridging components were not more effective in reducing readmission compared to treatment as usual; however, the results are based on limited evidence. Therefore, we currently cannot make a final recommendation for or against the use of transitional interventions. Nevertheless, transitional interventions with bridging components are preferred by service users and should be considered as alternatives or supplements to regular or no discharge strategies.

Our review highlights that the identification and future design of effective interventions requires higher quality studies with resulting comprehensive publications. First, researchers need to explore proper outcome measures and reach a consensus on measured outcomes (e.g., core outcomes set). This would allow for the conduction of systematic reviews and meta-analyses for health-related and social outcomes, adding to the limited existing evidence. Second, future research on transitional interventions needs more profound classifications of interventions and health care systems. In addition, we need to understand what “treatment as usual” means across various settings to identify effective intervention components and their correlations. Third, studies are needed to assess the effects of transitional interventions or specific intervention components on various disorders. These aspects would allow for the conduction of high-quality research and generate evidence for the best practices.