2.1 Protocol, Registration and Reporting Structure

The meta-review was registered with the PROSPERO international prospective register of systematic reviews in October 2021: CRD42021260221. It followed the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA; Page et al. 2021) and reporting guidelines for systematic reviews and meta-analyses.

2.2 Scoping Search

Before the meta-review, a scoping search was conducted using Google Scholar to map out the extent and nature of previous studies. Studies were initially examined for key themes without quality checking or full descriptions of their results. The results of the scoping search were used to determine whether a systematic review, meta-analysis or meta-review would be the most appropriate format for the main review, prior to designing search and eligibility criteria.

2.3 Search Strategies

Guided by the results of the scoping search, a search was conducted of the academic databases PsycInfo, MEDLINE, EMBASE and CINAHL from inception to October 2022 for reviews and meta-analyses examining factors that influence staff perceptions of inpatient psychiatric hospitals. The literature search was confined to a search of academic databases because a broader search would have involved contacting multiple local health districts, which was not feasible and would limit generalisability to an international context. Search strategies were tailored to each database using a combination of keywords and subject headings, such as ‘Health Personnel’, ‘Psychiatric Hospital’, ‘Perception’ and ‘Systematic Review’. Individual database search strategies are listed in Appendix A.

2.4 Eligibility Criteria

2.5 Review Selection

Following deduplication, titles and abstracts were independently screened by two authors (K.A.T. and M.M.) using the eligibility criteria and cross-checked for agreement before full-text articles were obtained. Full-text reviews were then independently assessed for inclusion in the meta-review (K.A.T. and M.M.). Any disagreement was resolved through discussion between raters.

2.6 Data Extraction

The data extraction form was created using Excel (version 2209). This form was pretested by author K.A.T. using data collected from three included reviews and used to correct any problems/issues with the form. The same author then used the revised form to extract data from all reviews, including citation details, review type (qualitative, quantitative or mixed methods), quality rating, key aims, relevant findings, extracted factors/themes and the primary studies related to each extracted factor (Table 2). Only factors that staff reported affected their perception of inpatient psychiatric hospitals were extracted. Subsequently, a second author (M.M.) conducted a quality check of ~10% of the extracted data. Any disagreements were resolved through consensus.

When conducting a meta-review, it is important to consider the degree of potential duplication of extracted data where multiple studies of the included reviews have analysed data from the same primary studies (Petrovskaya, Lau, and Antonio 2019). Therefore, to minimise bias introduced by overlapping studies, the data extraction table listed the primary studies referenced in each included review for each extracted factor, allowing for the subsequent removal of duplicate studies. This approach aimed to neither overestimate nor underestimate the degree of overlap between studies, that is neither counting evidence twice and thereby magnifying its relevance nor excluding evidence from reviews that included the same primary studies but that extracted nonoverlapping data from those primary studies (Petrovskaya, Lau, and Antonio 2019). Where it was unclear whether the data were overlapping, a conservative approach was used in which it was assumed that there was overlap, and the duplicate study was removed.

2.7 Quality and Risk of Bias Assessments

The quality of all included reviews was independently evaluated using the AMSTAR 2 (Shea et al. 2017) appraisal tool (K.A.T. and M.M.). The AMSTAR 2 is a validated tool for assessing the quality of systematic reviews, including randomised and nonrandomised studies of healthcare interventions. As many of the included reviews analysed qualitative data, a few changes were made to the AMSTAR 2 items using peer-reviewed guidelines from the Joanna Briggs Institute (2017) critical appraisal checklist for systematic reviews and research syntheses. It is noted that the AMSTAR 2 is regarded as validly modifiable as appropriate (Shea et al. 2017). Disagreements between evaluators were resolved through discussion.

For this meta-review, Items 1 and 7 were amended, as described in Modini, Burton, and Abbott (2021). Item 7 asks if ‘the review authors provide a list of excluded studies and justify exclusion’. It was amended to ‘Did the authors provide a tally, with explanations given, of full-text studies excluded?’. This was to align with the PRISMA guidelines used in this review, which do not specify the need for a list of excluded studies. Changes were also made to Item 8. Item 8 in its original form focuses largely on intervention outcomes. Therefore, Item 8 was amended to reflect the largely qualitative nature of the reviews in this meta-review: It was amended to include criteria for assessing whether included qualitative noninterventive studies were described in adequate detail; that is, a determination of ‘partial yes’ considered or at least described the risk of bias in the method, results or discussion; and a ‘yes’ also assessed risk of bias using a validated tool or, in the case of older studies, reviewed findings with peers and/or field experts.

In line with its author's recommendations for the use of the AMSTAR 2 (Shea et al. 2017), an overall confidence rating was made for each included review: high (no or one noncritical weakness); moderate (one or more noncritical weakness); low (one critical flaw with or without noncritical weaknesses); critically low (more than one critical flaw with or without noncritical weaknesses). Reviews determined to be ‘critically low’ were excluded from the meta-review to enhance confidence in the findings of the meta-review (Abdalrahim 2013; Al-Maraira and Hayajneh 2019; Bosman and van Meijel 2008; Jansen, Dassen, and Groot Jebbink 2005; Johnson 2004; Lelliott and Quirk 2004; Loukidou, Ioannidi, and Kalokerinou-Anagnostopoulou 2010; Moreno-Poyato et al. 2016, 2021; Morse et al. 2012; Odes et al. 2021; Papoulias et al. 2014; Paris and Hoge 2010; Sailas and Wahlbeck 2005; Stevenson and Taylor 2020; Stewart et al. 2009; van der Merwe et al. 2009, 2013; Wynn 2018). There was total agreement between evaluators on the reviews to be included in the meta-review.

Additionally, following Petrovskaya, Lau, and Antonio's (2019) umbrella review protocol, modified Grading of Recommendations Assessment, Development and Evaluation—Confidence in the Evidence from Reviews of Qualitative research (GRADE-CERQual; Lewin et al. 2015) criteria were applied at the level of individual findings to examine the confidence in the evidence in each extracted theme for qualitative data (Colvin et al. 2018; Lewin, Bohren, et al. 2018; Lewin, Booth, et al. 2018; Munthe-Kaas et al. 2018; Glenton et al. 2018; Noyes et al. 2018). Confidence was assessed with respect to this meta-review's aims rather than the included reviews' aims. A ‘middle ground’ decision rule was decided upon by the authors before the analysis (i.e., after reviewing the confidence ratings given to each factor, the final confidence rating represented an overall judgement of confidence rather than just the lowest or highest rating given to any single factor). This decision rule was used to make overall judgements about the confidence level of each theme in line with suggestions made in the CERQual guidelines (Lewin, Bohren, et al. 2018). Confidence was judged with regards to the three CERQual criteria thus far validated for use with meta-reviews: coherence (inappropriate oversimplification of data), adequacy (data richness) and methodological limitations (appropriate assessment of the risk of bias) (Lewin, Bohren, et al. 2018). The results of this appraisal are outlined in Table 1.

2.8 Data Synthesis

There were insufficient data to conduct statistical meta-analyses. Therefore, quantitative findings from the included reviews were descriptively synthesised and integrated into the results of the thematic synthesis. One review (Gee et al. 2017) also included theory building following data extraction. In keeping with the logic of aggregation used in this meta-review, data were extracted only from the presynthesised summary of findings reported by Gee et al. (2017), which could be analysed at the level of primary studies.

3.1 Staff and Patient Safety

Eleven reviews examined the role of ward safety in influencing staff perceptions of inpatient hospitals (Alexander and Bowers 2004; Delaney and Johnson 2014; Doedens et al. 2020; Fletcher et al. 2021; Hallett, Huber, and Dickens 2014; Jackson, Baker, and Berzins 2018; Laukkanen et al. 2019; Muir-Cochrane and Oster 2021; Scholes, Price, and Berry 2021; Wilson et al. 2021; Wyder et al. 2017). Overall confidence in this evidence was moderate (Table 1).

Nurses were reported to agree that establishing ward safety was a core aspect of psychiatric nursing (Alexander and Bowers 2004; Wyder et al. 2017) and a duty/responsibility (Muir-Cochrane and Oster 2021; Jackson, Baker, and Berzins 2018) that took precedence over their caring role when faced with patient violence (Fletcher et al. 2021). However, there was reported heterogeneity within nursing staffs' use of violence prevention strategies, with some relying solely on restraint (Laukkanen et al. 2019) and others endorsing debriefing, training, enhanced staff–patient and staff–staff interaction, increased staffing and addressing lack of patient privacy and overcrowding (Fletcher et al. 2021).

Nurses were reported to think they needed to remain vigilant of risk on the ward (Delaney and Johnson 2014; Fletcher et al. 2021) and felt overloaded by a hostile work environment (Fletcher et al. 2021). Nurses' perceptions of ward safety were reportedly positively influenced by trauma-informed practices (Wilson et al. 2021), limit setting (Alexander and Bowers 2004), increased autonomy in decision-making, consultation with colleagues, effective crisis management (Delaney and Johnson 2014), training, knowledge of patient history, pro-active de-escalation techniques (Hallett, Huber, and Dickens 2014) and patient engagement (Wyder et al. 2017). Factors that were noted to decrease staff perceptions of safety included under-resourcing (Scholes, Price, and Berry 2021), relocating from a ‘mental institution’ to a general hospital setting (Alexander and Bowers 2004), being the victim/witness of patient assault (Doedens et al. 2020) and the relatively greater severity of self-harming behaviours and likelihood of assault in women's-only wards (Scholes, Price, and Berry 2021).

3.2 Views on Inpatients' Experiences

Three reviews described staff perceptions of inpatients' experiences on the ward (Delaney and Johnson 2014; Muir-Cochrane and Oster 2021; Laukkanen et al. 2019). Overall confidence in this evidence was low (Table 1).

Nurses were reported to believe that the therapeutic relationship positively impacted patient and staff perceptions of ward life and increased patient empowerment (Delaney and Johnson 2014). There was some heterogeneity in staff perceptions of inpatients' experiences of coercive measures, though staff largely thought patients were negatively impacted, particularly emotionally (Muir-Cochrane and Oster 2021; Laukkanen et al. 2019).

3.3 Relationships on the Ward

Ten reviews highlighted the importance of relationships on the ward for staff in inpatient hospitals (Alexander and Bowers 2004; Delaney and Johnson 2014; Fletcher et al. 2021; Hallett, Huber, and Dickens 2014; Jackson, Baker, and Berzins 2018; Muir-Cochrane and Oster 2021; Scholes, Price, and Berry 2021; Westwood and Baker 2010; Wilson et al. 2021; Wyder et al. 2017). Overall confidence in this evidence was low (Table 1).

3.4 Ward Rules

Ten reviews discussed the impact of ward rules on staff perceptions of inpatient wards (Alexander and Bowers 2004; Delaney and Johnson 2014; Fletcher et al. 2021; Hallett, Huber, and Dickens 2014; Laukkanen et al. 2019; Muir-Cochrane and Oster 2021; Nelstrop et al. 2006; Scholes, Price, and Berry 2021; Wilson et al. 2021; Wyder et al. 2017). Overall confidence in this evidence was low (Table 1).

All the above reviews, except Fletcher et al. (2021), highlighted the idea that staff enforced ward rules in contrast to their values, beliefs or principles of trauma-informed care. Nurses were reported to perceive that they had little control over their nursing practice, yet, at the same time, they were aware they had substantial control over the day-to-day functioning of inpatient wards (Wyder et al. 2017). Similarly, physicians were described as perceiving they lacked choice about using chemical restraint, given the need to maintain the safety of service users and staff and the pressure to meet the needs and expectations of nursing staff (Muir-Cochrane and Oster 2021). Nurses were described as simultaneously believing they provided ‘questionable’ care, given the circumstances, yet having pride in their ability to meet patients' needs while working in a flawed system (Delaney and Johnson 2014). Nurses were also reported to perceive a mismatch between restrictive hospital policies and ‘unreasonable’ organisational expectations to use a trauma-informed, person-centred approach that was simultaneously quantifiable, time-limited and task-focused (Wilson et al. 2021; Wyder et al. 2017).

Nurses were noted to find it challenging to balance restrictive practices with patients' rights and to sometimes disagree with aspects of involuntary treatment, perceiving that the necessity of enforcing such policies interfered with their capacity to provide authentic nursing care (Alexander and Bowers 2004) and that inpatients directed their anger and frustration about involuntary treatment at staff (Wyder et al. 2017). Furthermore, nurses were reported to believe there was a ‘subtle mismatch’ between the type of knowledge required to work within the medical model of psychiatry and the knowledge required to understand a patient's suffering (Delaney and Johnson 2014).

Additionally, ward routines and rules were reported to be important to staff perceptions of aggressive incidents (Fletcher et al. 2021) and violence prevention (Hallett, Huber, and Dickens 2014) and sometimes pose barriers to patient care if not applied flexibly (Delaney and Johnson 2014). Physician, nurse, occupational therapist and social worker perceptions of having low controllability of patients' aggression were linked to negative emotions towards patients, whereas low perceived patient controllability over their own emotions was linked with positive staff emotions because it denoted staff acceptance of the patient as a suffering individual rather than a deliberate perpetrator (Fletcher et al. 2021). It is perhaps not surprising then that clear ward routines and guidelines were reported to facilitate staff's sense of controllability over patient aggression (Fletcher et al. 2021) and that enhancing ward rule clarity was reported to boost staff–patient communication and staff morale (Alexander and Bowers 2004).

3.5 Knowledge and Experience

Nine reviews examined the impact of staff's knowledge and experience on their perceptions of inpatient wards (Alexander and Bowers 2004; Delaney and Johnson 2014; Doedens et al. 2020; Gee et al. 2017; Jackson, Baker, and Berzins 2018; Laukkanen et al. 2019; Muir-Cochrane and Oster 2021; Wilson et al. 2021; Wyder et al. 2017). Overall confidence in this evidence was low (Table 1).

Recovery-oriented training for all staff of adult and forensic units was found to significantly increase staff perceptions that ‘what [patients] say makes a difference in their treatment’, ‘staff spend time talking to and doing things with’ patients and that patients ‘would get out of the hospital and not come back’ as well as staff's perceived autonomy (Gee et al. 2017).

Nurses were reported to perceive that their experience and expert knowledge were important in supporting treatment adherence for pharmacotherapy (Wyder et al. 2017). Similarly, nurses were noted to link their level of experience with decision-making about coercive practices like seclusion and restraint (Alexander and Bowers 2004; Delaney and Johnson 2014; Doedens et al. 2020; Jackson, Baker, and Berzins 2018; Laukkanen et al. 2019). However, reviews varied in their reports of the direction of this effect. That is, whether greater seniority increased or decreased nurses' approval of coercive measures, which Doedens et al. (2020) suggested may be indicative of a cohort effect with nursing staff's attitudes reportedly becoming more critical of coercive practices since about 2010.

3.6 Service Delivery Issues

Six reviews examined the role of service delivery issues in influencing staff perceptions of inpatient psychiatric hospitals (Delaney and Johnson 2014; Doedens et al. 2020; Hallett, Huber, and Dickens 2014; Jackson, Baker, and Berzins 2018; Richards et al. 2006; Wyder et al. 2017). Overall confidence in this evidence was low (Table 1).

Staff were described as facing service delivery issues, including clinical (intense and low visibility work), administrative (responsibility for non-nursing tasks) and organisational (lack of support) issues (Delaney and Johnson 2014) as well as perceiving that poor staffing and training levels interfered with violence prevention (Hallett, Huber, and Dickens 2014). Sociopolitical factors were also identified as influencing nurses' attitudes towards coercive measures (Doedens et al. 2020; Jackson, Baker, and Berzins 2018), with nurses perceiving they were in a double bind where they might be blamed by (future) society for using coercion and for the detrimental effects of not using coercion (Doedens et al. 2020).

Nurses were reported to perceive that they experienced high levels of burnout, occupational stress and psychological distress (Delaney and Johnson 2014; Wyder et al. 2017) due, in part, to service delivery issues, though this was not supported by Richards et al.'s (2006) quantitative analysis of these issues. However, Richards et al.'s analysis was hampered by the heterogeneity and low quality of available data on these issues.

3.7 Coercive Measures

Eight reviews reported on staff perceptions of coercive measures (seclusion, physical restraint and/or chemical restraint) (Alexander and Bowers 2004; Delaney and Johnson 2014; Doedens et al. 2020; Jackson, Baker, and Berzins 2018; Laukkanen et al. 2019; Muir-Cochrane and Oster 2021; Nelstrop et al. 2006; Wyder et al. 2017). Overall confidence in this evidence was very low (Table 1).

There was considerable heterogeneity in nursing staff's reported attitudes towards the use of coercive measures, with variation according to the restrictiveness of the measure (Doedens et al. 2020; Laukkanen et al. 2019; Muir-Cochrane and Oster 2021; Nelstrop et al. 2006), country (Alexander and Bowers 2004; Laukkanen et al. 2019), gender of staff and patients, staff age and experience level (Doedens et al. 2020). Another review stated that other allied health staff believe coercive measures violate patient integrity and damage the therapeutic alliance (Laukkanen et al. 2019).

4.1 Strengths and Limitations

Much of the literature reviewed here focused on staff perceptions of inpatient psychiatric hospitals in the last 20–30 years. Only one included review directly considered whether staff views had changed over time. Most perceptions were drawn from staff working in inpatient psychiatric hospitals after the major shift away from ‘asylums’ towards greater community care of patients. Furthermore, the majority of included reviews did not differentiate between the age or experience levels of staff in reporting their perceptions of inpatient settings. Therefore, we likely did not see as much of a ‘cohort effect’ as we might have had the included reviews gone back further in time or delineated staff perceptions by age and experience levels. Future primary research should consider exploring this further.

Similarly, much of the literature in the included reviews focused on the perceptions of nursing staff rather than staff from other disciplines. although the views of nursing staff are important, with nurses making up the largest proportion of the mental health workforce internationally (AIHW 2022; Jones 2006; Søvold et al. 2021), their views are not the only ones worthy of consideration. Most notably, the perceptions of psychiatrists were absent. Psychiatrists tend to have leadership roles in MDTs, managing inpatient teams for long periods. They may therefore provide valuable insight not only into their role but into changes in staff perceptions over time. Consequently, future research into the perceptions of other staff should be prioritised.

Strengths of this meta-review include using a preexisting protocol for conducting meta-reviews of qualitative literature to maximise the systematic nature of the process and using the AMSTAR 2, a validated tool to assess the methodological quality of healthcare reviews, to assess for bias. Although a number of potential reviews were excluded as part of this latter process, this ensured that only the highest quality reviews were summarised and synthesised, enhancing confidence in the results of this meta-review. Similarly, a particular strength of this meta-review was the additional assessment of the quality of evidence at the level of each extracted theme, including the deduplication of fully overlapping data from original primary studies. However, this does not negate the fact that recall or selection bias in the original primary studies in the included reviews may carry over to this meta-review.

Overall, the highest quality evidence included in this meta-review came from the examination of themes of safety and staff views on inpatients' experiences, offering a reasonable basis for confidence in conclusions drawn from these findings. Other themes (relationships on the ward, ward rules, knowledge and experience, ward rules and service delivery issues) were rated low in confidence or very low in the case of coercive measures. The two most common issues affecting the confidence in the evidence underlying the identified themes were methodological quality (inappropriate/lack of examination of the risk of bias) and coherence. Issues with coherence affected all themes, indicating that the included reviews tended to inappropriately smooth out data, leaving out lower frequency cases (e.g., perspectives of non-nursing staff) or merging data across settings (acute and other wards) and time. To some extent, this smoothing over is inevitable in the review process, which typically aims to identify overarching patterns/themes. However, it suggests more focused research is warranted into these ‘smoothed over’ areas. It is also important to recognise that this meta-review used thematic synthesis to explore staff perceptions, which incorporates a degree of subjectivity, and over which there is debate as to the degree of bias this may introduce (Barnett-Page and Thomas 2009). Efforts were made to minimise subjective bias by using a data saturation approach to maximise the thoroughness of data extraction and involving a second independent reviewer to assess that this had been achieved. Additionally, there were insufficient quantitative data to perform meta-analyses to judge the overall levels of staff morale, job satisfaction or burnout. This is another avenue for future research.