There is increasing emphasis on reducing the use and improving the safety of mechanical restraint (MR) in psychiatric settings, and on improving the quality of evidence for outcomes. To date, however, a systematic appraisal of evidence has been lacking.

We included studies of adults (aged 18–65) admitted to inpatient psychiatric settings. We included primary randomised or observational studies from 1990 onwards that reported patterns of MR and/or outcomes associated with MR, and qualitative studies referring to an index admission or MR episode. We presented prevalence data only for studies from 2010 onwards. The risk of bias was assessed using an adapted checklist for randomised/observational studies and the Newcastle-Ottawa scale for interventional studies.

We included 83 articles on 73 studies from 1990–2022, from 22 countries. Twenty-six studies, from 11 countries, 2010 onwards, presented data from on proportions of patients/admissions affected by MR. There was wide variation in prevalence (<1–51%). This appeared to be mostly due to variations in standard protocols between countries and regions, which dictated use compared to other restrictive practices such as seclusion. Indications for MR were typically broad (violence/aggression, danger to self or property). The most consistently associated factors were the early phase of admission, male sex, and younger age. Ward and staff factors were inconsistently examined. There was limited reporting of patient experience or positive effects.

MR remains widely practiced in psychiatric settings internationally, with considerable variation in rates, but few high-quality studies of outcomes. There was a notable lack of studies investigating different types of restraint, indications, clinical factors associated with use, the impact of ethnicity and language, and evidence for outcomes. Studies examining these factors are crucial areas for future research. In limiting the use of MR, some ward-level interventions show promise, however, wider contextual factors are often overlooked.