Influence of nursing staff attitudes and characteristics on the use of coercive measures in acute mental health services—A systematic review

Accessible summary What is known on the subject? Aggressive behaviour is a major problem in clinical practice of mental health care and can result in the use of coercive measures. Coercive measures are dangerous for psychiatric patients and international mental healthcare works on the elimination of these interventions. There is no previous review that summarizes the attitude of nursing staff towards coercive measures and the influence of nursing staff characteristics on attitude towards and the use of coercive measures. What the paper adds to existing knowledge? The attitude of nurses shifted from a therapeutic paradigm (coercive measures have positive effects on patients) to a safety paradigm (coercive measures are undesirable, but necessary for the wards’ safety). Nurses express the need for less coercive interventions to prevent seclusion and restraint, but their perception of intrusiveness is influenced by how often they use specific coercive measures. The knowledge from scientific literature on the influence of nursing staff on coercive measures is highly inconclusive, although the feeling of safety of nurses might prove to be promising for further research. What are the implications for practice? There is need for increased attention specifically for the feeling of safety of nurses, to better equip nurses for their difficult work on acute mental health wards. Abstract Introduction The use of coercive measures generally has negative effects on patients. To help prevent its use, professionals need insight into what nurses believe about coercion and which staff determinants may influence its application. There is need for an integrated review on both attitude and influence of nurses on the use of coercion. Aim To summarize literature concerning attitude of nurses towards coercive measures and the influence of staff characteristics on the use of coercive measures. Method Systematic review. Results The attitude of nurses changed during the last two decades from a therapeutic to a safety paradigm. Nurses currently view coercive measures as undesirable, but necessary to deal with aggression. Nurses express the need for less intrusive interventions, although familiarity probably influences its perceived intrusiveness. Literature on the relation between staff characteristics and coercive measures is inconclusive. Discussion Nurses perceive coercive measures as unwanted but still necessary to maintain safety on psychiatric wards. Focussing on the determinants of perception of safety might be a promising direction for future research. Implications for practice Mental health care could improve the focus on the constructs of perceived safety and familiarity with alternative interventions to protect patients from unnecessary use of coercive interventions.


| INTRODUC TI ON
Aggressive behaviour is a broad behavioural construct that includes the concept of violence and causes safety issues in mental health care (Gaynes et al., 2017;Liu, 2004). The definition of violence is an act including physical force such as slapping, punching, kicking and biting; use of an object as a weapon; aggressive behaviour such as spitting, scratching and pinching; or a verbal threat involving no physical contact (Nolan, Soares, Dallender, Thomsen, & Arnetz, 2001). The prevalence of physical violence of patients during psychiatric admission differs in Western countries between 7.5% and 15% (Cornaggia, Beghi, Pavone, & Barale, 2011). To protect patients and staff on psychiatric wards from harm caused by violence, professionals use coercive measures, such as seclusion, restraint and compulsory medication (Cowman, Bjorkdahl, Clarke, Gethin, & Maguire, 2017). In Europe, some countries use seclusion as a "preferred" intervention of last resort in case of dangerous situations, while others resort to physical or mechanical restraint (Bak & Aggernaes, 2012). Coercive measures have no therapeutic value and can result in post-traumatic stress and severe physical injuries for patients (Frueh et al., 2005;Nath & Marcus, 2006;Rakhmatullina, Taub, & Jacob, 2013;Sailas & Fenton, 2000;Steinert, Birk, Flammer, & Bergk, 2013). Consequently, prevention of coercive measures has become a priority of care professionals, researchers and policymakers in mental health services. The international mental health community developed several quality improvement projects in the last few years to diminish its use (Bierbooms, Lorenz-Artz, Pols, & Bongers, 2017;Bowers, 2014;Duxbury et al., 2019;Lombardo et al., 2018).
To help prevent the use of coercive measures, it is important to know about variables that are predictive for its use. In their systematic review on patient and staff characteristics associated with higher use of restraint, Beghi, Peroni, Gabola, Rossetti, and Cornaggia (2013) reported that male gender, young age, foreign ethnicity, involuntary admission, diagnosis of schizophrenia and presence of male staff were variables associated with more use of restraint. Laiho et al. (2013) described the influence of the previous experience of nurses with coercion on the decision to use coercive measures. The attitude of nurses towards coercive measures is also important. In their systematic review on nurses' attitudes towards coercion, Happell and Harrow (2010) found a contradiction between practice of seclusion and attitudes and beliefs of nurses about its efficacy and appropriateness.
Nurses acknowledged that seclusion had a negative impact on service users, but inpatient violence justified its use (Happell & Harrow, 2010). This is in line with other review studies, such as Riahi, Thomson, and Duxbury (2016) and Laukkanen, Vehviläinen-Julkunen, Louheranta, and Kuosmanen (2019) who concluded that coercive measures are still seen as necessary measure of "last resort," although the attitude of nurses is turning increasingly negative. Furthermore, Riahi et al. (2016) suggest that staff composition and nurses' perception are important themes in the decision-making process towards the use of coercive measures. Happell and Harrow (2010) suggest that future research needs to consider staff characteristics together with attitude towards seclusion. Currently, a systematic review that evaluates both the Aim: To summarize literature concerning attitude of nurses towards coercive measures and the influence of staff characteristics on the use of coercive measures.

Method: Systematic review.
Results: The attitude of nurses changed during the last two decades from a therapeutic to a safety paradigm. Nurses currently view coercive measures as undesirable, but necessary to deal with aggression. Nurses express the need for less intrusive interventions, although familiarity probably influences its perceived intrusiveness.
Literature on the relation between staff characteristics and coercive measures is inconclusive.
Discussion: Nurses perceive coercive measures as unwanted but still necessary to maintain safety on psychiatric wards. Focussing on the determinants of perception of safety might be a promising direction for future research.
Implications for practice: Mental health care could improve the focus on the constructs of perceived safety and familiarity with alternative interventions to protect patients from unnecessary use of coercive interventions.

K E Y W O R D S
coercion, nurse role, safety and security, seclusion and restraint, systematic literature review attitude of nurses and the influence of nursing staff characteristics related to coercive measures is lacking.

| AIMS
The aim of this paper is to summarize scientific literature con-

| Design
We performed a systematic review and used the PRISMA statement to guide our reporting (Moher, Liberati, Tetzlaff, & Altman, 2009). We defined attitude towards coercive measures according to Bowers et al. (2007) p.358 as "the pattern of beliefs, judgements and feelings about coercive measures." We divided nursing staff characteristics into individual characteristics (e.g., gender, age, personality traits), professional characteristics (e.g., education, work experience) and organizational characteristics (e.g., staff-patient ratio).

| Study selection
We performed the first selection based on title and abstract. We subsequently retrieved the full text of the included studies for the final assessment of eligibility. Two reviewers (PD and JV) performed the selection independently and settled disagreements through discussion. In case of disagreement, the reviewers consulted a third reviewer (CL).
We selected studies based on inclusion and exclusion criteria.
Inclusion criteria concerning study design were cohort studies, case-control studies, case series, cross-sectional studies, surveys and qualitative studies on the attitude of nursing staff towards coercive measures and/or the influence of nursing staff characteristics on the use of one or more coercive measures (seclusion, mechanical restraint, physical restraint and compulsory medication).
We included studies performed in acute mental health inpatient services or psychiatric facilities in general or academic hospitals that cared for psychiatric patients with primary diagnosis of axis I or II of the DSM-IV-TR (American Psychiatric Association, 2000), except addiction disorders and learning disabilities or their equivalent in the DSM-5 (American Psychiatric Association, 2013).
Studies that included also other professionals (such as physicians) and other settings (such as forensic wards) were included if the majority (>50%) of the staff members or settings met our inclusion criteria. We excluded studies performed solely in forensic, child, adolescent and geriatric psychiatry, in general hospital wards, emergency departments, nursing homes or with an outpatient patient population. We excluded studies that addressed aggressive behaviour as outcome measure. We also excluded reviews, case reports, theses, conference abstracts and non-empirical publications, such as editorials.

| Assessment of the risk of bias
We used the Quality in Prognostic Studies (QUIPS) tool (Hayden, van der Windt, Cartwright, Cote, & Bombardier, 2013) for cohort studies, the Newcastle-Ottawa Scale (NOS)  for case-control studies and the Consolidated criteria for reporting qualitative research (COREQ) (Tong, Sainsbury, & Craig, 2007) for qualitative research.

| Data extraction and analysis
Two independent reviewers (PD & JV) performed the data extraction with a standardized form. Studies that described the attitude of nurses were mostly qualitative or survey studies, and the results were not suitable for statistical pooling. We carefully read the studies and extracted important themes from these studies independently. Thereafter, we discussed the interpretation of the qualitative findings. Subsequently, we extracted descriptive themes from the analysis of the qualitative studies based on consensus between the reviewers and combined these with the results from the surveys.
We observed that literature on nursing staff characteristics had high levels of heterogeneity, which made it unlikely that performing a meta-analysis would be appropriate. We summarized the most important results of the included studies. We extracted data on the research question, design, sample size, population, setting and outcome measures from the included studies.

| Search results and quality assessment
The initial search resulted in 7,517 references. After the selection process, we included 84 publications ( Figure 1). Among these were papers written in English (78), Dutch (2), German (2) and French (2).
Sixty of these papers were reported on the attitudes of nurses and 31 papers reported on the influence of nursing staff characteristics.
The data of a large cross-sectional study from the United Kingdom, named City-128, accounted for seven publications (Bowers, 2009;Bowers & Crowder, 2012;Bowers, Nijman, Simpson, & Jones, 2011;Bowers, Stewart, Papadopoulos, & Iennaco, 2013;Bowers et al., 2010;Bowers, Van Der Merwe, Paterson, & Stewart, 2012;Whittington, Bowers, Nolan, Simpson, & Neil, 2009). A cross-sectional study from Norway accounted for two publications (Husum, Bjorngaard, Finset, & Ruud, 2010, and a survey from Australia accounted for two publications (Happell & Koehn, 2010. These papers were not duplicates, but described different analyses based on a single, large data set. Therefore, we included 76 unique studies in our review, of which four were prospective cohort studies, five were retrospective cohort studies, four were case-control studies, one was a mixedmethod study, nine were cross-sectional studies, 31 were surveys and 22 were qualitative studies. These studies originated from 25 different countries. We provide an overview of the included studies in Data S2. The quantitative studies showed large clinical and methodological heterogeneity. Most of the studies were cross-sectional studies or surveys based on questionnaires. Several of these studies used self-developed questionnaires of which the psychometric properties were unknown. Others used validated questionnaires, mostly the Attitudes Toward Seclusion Survey (Heyman, 1987) and the Attitudes to Containment Measures Questionnaire (Bowers, Alexander, Simpson, Ryan, & Carr-Walker, 2004). Sample size varied from very small (e.g., questionnaire administered with n = 13 nurses (Tooke & Brown, 1992)) to very large (e.g., cross-sectional study with n = 11,128 admissions over 136 psychiatric wards (Bowers, 2009)). The available cohort studies and case-control studies often had methodological limitations, such as small sample sizes, retrospective design, limited information on the sampling procedure and data collection on a single ward or hospital.
Most of the studies from the eighties and early nineties presented no comprehensive description of the method, statistics and results. The majority of the qualitative studies were of moderate quality. The comprehensiveness of reporting of qualitative studies showed substantial improvement in the last decades, especially in methodological rigour.

| Attitudes towards coercive measures
In our study of the included literature on the attitudes of nurses towards coercive measures, we observed two major themes: (a) the

| Treatment paradigm versus safety paradigm
We observed a paradigm shift in the attitude towards coercive measures from a treatment paradigm to a safety paradigm. The belief that patients experience therapeutic benefits from the use of coercive measures characterizes the treatment paradigm. Distinctive for the safety paradigm is the belief that the patient undergoing coercive measures experiences negative consequences, but coercive measures are necessary to maintain safety for patients and staff members. Tooke and Brown (1992) were the first to report attitudes of nurses from the therapeutic paradigm and found that nurses believed seclusion was a calming, therapeutic experience. Coercive measures were seen as effective interventions to protect patients' dignity (Palazzolo, Favre, Halim, & Bougerol, 2000). Nurses consid- An early example of the safety paradigm was DiFabio (1981), who reported that although nurses had numerous emotional and negative experiences with restraint, its use was necessary to control patients' behaviour in case of dangerous situations. Lendemeijer (1997) stated that the safety of psychiatric wards prevailed over the individual patient's interest and therefore seclusion was required.
The necessity of using seclusion and other coercive measures in case of aggressive behaviour, despite doubts on the therapeutic effect, was also reported by several other authors during the nineties (De Cangas, 1993;Holzworth & Wills, 1999;Muir-Cochrane, 1996;Olofsson, Gilje, Jacobsson, & Norberg, 1998 Wynaden et al., 2001;Wynn, 2003). Bigwood and Crowe (2008) stated that physical restraint was undesirable but unavoidable: "it's part of the job, but spoils the job." Lemonidou et al. (2002) found that nurses had "positive" attitudes towards seclusion, but mainly because they viewed seclusion as necessary, not desirable. Nurses viewed seclusion as effective for controlling "difficult situations," but also expressed their concerns about negative consequences for patients (Lee et al., 2003). From 2010, the paradigm shifted more and more towards coercive measures being a "necessary evil," rather than a therapeutic tool (Wilson, Rouse, Rae, & Kar Ray, 2017).
Numerous studies reported that nurses considered coercive measures unwanted and harmful, but necessary to regain safety in the case of aggressive behaviour (Fereidooni Moghadam et  In sum, the necessity of coercive measures for dealing with danger due to aggressive behaviour of patients seems a key element of the current attitude of nurses.

| Need for less intrusive alternative interventions
Our second theme observed in the studies about nursing staff's attitude was the need for alternative interventions to maintain the safety of patients and staff on psychiatric wards.
The shift from the treatment to the safety paradigm is a key factor in the need for alternatives. Despite the negative consequences and feelings, nurses feared elimination of coercive measures as a tool for dealing with aggressive behaviour and expressed concerns that society will blame them in the future for using coercion and for the negative consequences of not using coercion (Muir-Cochrane et al., 2018). Because of the perceived necessity of using coercive measures, alternative interventions are vital to align with the ambition to diminish their use from mental health care. Specifically, nurses seem to perceive the severity of coercive interventions as something that needs attention.
Nurses expressed the desire for more "gentle" interventions to manage patients' behaviour (Olofsson et al., 1998 Wilson et al., 2017;Wynaden et al., 2002;Wynn, Kvalvik, & Hynnekleiv, 2011). However, the concept of "last resort" is unclear and some staff members viewed the point that an intervention is "of last resort" earlier than others did (Happell et al., 2012;Wilson et al., 2017). Seclusion and restraint have major impact on the patient, and nurses were generally concerned about their well-being when applying these interventions (Lee et al., 2003;Wynn, 2003).
Although seclusion and restraint are both seen as highly intrusive, several authors reported that nurses viewed seclusion and forced medication as less intrusive and, thus, favourable compared to mechanical restraint ( (2019) suggested that nurses were supportive towards the elimination of mechanical restraint use because they are less frequent than other coercive measures. Dahan et al. (2018) reported that participants who were present during mechanical restraint practices had more positive attitudes than participants who were never present. Pettit et al. (2017) found that availability of a coercive measure was associated with approval of the use of the coercive measure. For example, access to a seclusion room was associated with greater acceptability of seclusion as a method of containment (Pettit et al., 2017).
In sum, nurses consider seclusion and restraint generally as most intrusive interventions and express the need for less intrusive alternatives to diminish their use. The attitude of nurses towards specific coercive measures seems more positive for interventions used more frequently in practice.

| Influence of nursing staff characteristics
Next, we summarize the results of the quantitative studies on the influence of nursing staff characteristics (individual, professional and organisational) on the use of and attitude towards coercive measures.
Some authors reported that young age was associated with more positive attitudes towards seclusion (Happell & Koehn, 2010;Wynn, 2003) or coercive measures in general (Husum et al., 2011), although an opposite effect was found for physical restraint (Wynn, 2003).
The City-128 study investigated ethnicity of the nurse and found that the proportion of white staff members in a team was associated with more use of coercive measures, compared with African and other ethnicities (Bowers, 2009 A creative personality, measured on Gough's Adjective Checklist (Gough, 1960), and high leadership scores, measured on Kolb's Organizational Climate Questionnaire (Kolb, Rubin, & McIntyre, 1971), were found to be associated with less initiation of coercion (Pawlowski & Baranowski, 2017). High scores on transactional leadership, measured as a subscale of the Multifactor Leadership Questionnaire (Bass & Avolio, 1995), were also found to be associated with less use of coercive measures (Bowers, 2009). Staff members with high empathy scores (scored on a scale of one (below-average empathy) to five (above average empathy)) were less prone to use seclusion and restraint (Yang, Hargreaves, & Bostrom, 2014 (Maslach & Jackson, 1981)) and low scores of therapeutic optimism (nurses' optimism related to treatment outcomes for patients, measured with the Elsom Therapeutic Optimism Scale (Elsom & McCauley-Elsom, 2008)). There was no association between anger of nurses and the incidence of seclusion and restraint (Jalil, Huber, Sixsmith, & Dickens, 2017). Bowers (2009) did not find an association between score on the MBI and the use of coercive measures.
Feelings of safety of nurses were likely to be associated with the use of coercive measures, although definition and measurement is complicated. Moreover, direction of causality is mostly unknown.
Higher subjective feeling of safety of nurses was associated with less seclusion (Vollema et al., 2012). These authors measured the feeling of safety at the end of each shift. Therefore, an aggressive incident that led to seclusion during the shift may have caused a lower feeling of safety. The feeling of safety was negatively influenced by physical environment (e.g., lack of safety equipment), organisational factors (e.g., low staff-patient ratio), lack of communication with hospital security, patient characteristics and trust within teams, while aggression management training, work experience and information about patients contributed to the feeling of safety (Goulet & Larue, 2017). Goulet and Larue (2017) also described that being a victim or witness of patient assault made nurses feel less safe and may even induced hypervigilance. Gray and Diers (1992) suggested that a decrease in staff stress and increase in feelings of control by staff was associated with an increase in the use of coercive measures, while referring to the "reverse hypothesis" (patient will not act out when staff members are upset). These authors measured staff stress and coercive measures before and after a major organisational change, making it likely that the organizational change caused confounding.
Nurses that were assaulted and injured by patients decided to use restraint later in the course of an aggressive incident than nurses that were never injured by patients (Moylan & Cullinan, 2011). A positive attitude towards patients with personality disorders was associated with less seclusion, but not with other forms of coercion (Bowers, 2009;Bowers et al., 2010

Several authors investigated the educational level of nurses in rela-
tion to the use of coercive measures. The City-128 study divided staff members into qualified and non-qualified staff. Wards with more qualified staff were associated with more use of seclusion (Bowers et al., 2010). This seemed also to be the case for mechanical restraint . Khalil et al. (2017)  ward was also associated with more mechanical restraint . However, most studies that incorporated educational level of nurses in their model found no association with the use of coercive measures (Bornstein, 1985;De Benedictis et al., 2011;Doedens et al., 2017;Janssen et al., 2007;Kodal et al., 2018).
Several authors reported no association between the work experience of nurses and the frequency of use of coercive measures (De Benedictis et al., 2011;Doedens et al., 2017;Janssen et al., 2007;Khalil et al., 2017;Kodal et al., 2018;O'Malley et al., 2007). Janssen et al. (2007) found an association between more variability in the nursing team of a shift and less frequent use of seclusion. Morrison and Lehane (1995) suggested that more experienced nurses ("charge nurses") might be associated with less use of seclusion, although they did not perform any statistical testing. Some authors suggested that experienced nurses tended to have less supportive attitudes towards the use of coercive measures (Gelkopf et al., 2009;Happell & Koehn, 2010;Korkeila et al., 2016). However, Gandhi et al. (2018) and Bregar et al. (2018) reported more positive attitudes for restraint of nurses with more work experience. Mann-Poll et al. (2015) found that experienced nurses rated the use of seclusion equally appropriate and necessary, while less experienced nurses showed more ambivalence in necessity and appropriateness.
There is no evidence for an association between the amount of fulltime nurses in a team (De Benedictis et al., 2011;Doedens et al., 2017), the length of time that nurses are working at the ward (Doedens et al., 2017) or their training in aggression management (De Benedictis et al., 2011;Khalil et al., 2017) and the frequency of use of coercive measures.

| Organisational characteristics
Staff-patient ratio has received extensive attention in scientific research in the last 30 years. Several authors reported an association between a lower staff-patient ratio (i.e., less staff members for each patient) and an increase in the use of coercive measures (Convertino et al., 1980;Donat, 2002;Morrison & Lehane, 1995;O'Malley et al., 2007). On the contrary, Bowers and Crowder (2012) found that more qualified staff members in the shifts and in the shifts prior to the incident were associated with more frequent use of coercive measures. Fukasawa, Miyake, Suzuki, Fukuda, and Yamanouchi (2018) found a small association between higher staff-patient ratio and an increase in the use of seclusion and restraint. Other authors found no association for staff-patient ratio and the use of coercive measures (Bowers, 2009;Bowers et al., 2010Husum, Bjorngaard, Finset, & Ruud, 2010;Janssen et al., 2007;Khalil et al., 2017;Kodal et al., 2018;Sercan & Bilici, 2009;Vollema et al., 2012;Yang et al., 2014) or reported no outcome measurement despite the fact that they mentioned measuring this variable in the method section (Betemps, Somoza, & Buncher, 1993). Klimitz, Uhlemann, and Fahndrich (1998) reported no association between the use of restraint and shortage of nursing staff. The staff-patient ratio varied in most studies of different shifts (day, evening and night). According to Klimitz et al. (1998) and Morrison and Lehane (1995), the night shift has the least use of coercive measures compared to the other shifts. However, other studies found that the night shift has most use of coercive measures compared with other shifts (Convertino et al., 1980;O'Malley et al., 2007). Several authors claim that most coercive measure occurred during the evening shift (Klimitz et al., 1998;Kodal et al., 2018;Reitan, Helvik, & Iversen, 2018). Yang et al. (2014)  De Benedictis et al. (2011) found that seclusion and restraint occurred more at psychiatric emergency departments or intensive care units than at regular psychiatric wards, but less frequent in non-teaching hospitals compared to teaching hospitals. The availability of (and compliance to) aggression management protocols was not associated with the use of seclusion and restraint (De Benedictis et al., 2011). Changing a 20-bed unit into two ten-bed units (while holding the staff-patient ratio stable) seemed to decrease the use of seclusion, suggesting that deviant patient behaviour can be managed better at small wards (O'Malley et al., 2007).
A higher score on the subscale programme clarity of the Ward Atmosphere Scale (Moos, 1974), indicating an effective structure on the ward, was associated with less use of coercive measures (Bowers, 2009;. Other authors found no association between ward atmosphere and frequency of use of coercive measures (De Cangas, 1993;Klimitz et al., 1998). Bowers (2009) found no association between team climate and the use of coercive measures, contrary to De Benedictis et al. (2011) who reported an association of the subscale anger and aggression of the Group Environment Scale (Moos, Shelton, & Petty, 1973) and the use of seclusion and restraint.

| D ISCUSS I ON
This systematic review aimed to summarize the scientific literature on attitudes of nurses towards coercive measures and on the association between nursing staff characteristics and the use of coercive measures and the attitude of nurses towards coercive measures in acute mental health services.
With respect to the first aim, we observed two major themes in the attitude of nurses towards use of coercive measures. Firstly, the abandonment of a treatment paradigm towards a safety paradigm. In the therapeutic paradigm nurses considered coercive measures as harsh, but helpful, for example calming the agitated patient and protecting patients' dignity (Lendemeijer, 1997;Palazzolo et al., 2000;Tooke & Brown, 1992). The support for the therapeutic paradigm in the attitude of nurses decreased substantially in the last decades and shifted to the safety paradigm. In the safety paradigm, staff members consider coercive measures a measure of last resort and there is a preference for the least intrusive intervention. This This could explain the association between a positive attitude and the frequency of use of a specific coercive measure (Özcan et al., 2015;Whittington et al., 2009). This theory is in line with Laiho et al. (2013), who stated that the threshold to use coercion gets lower  (Bowers et al., 2010;De Cangas, 1993;Khalil et al., 2017;Kodal et al., 2018;Morrison & Lehane, 1995), three of them concluded that female nurses were more prone to use coercion (Bornstein, 1985;Convertino et al., 1980;Janssen et al., 2007) and four of them found no effect in multivariable analysis (Bowers et al., 2011;De Benedictis et al., 2011;Doedens et al., 2017;Vollema et al., 2012).
Findings on the influence of the attitude towards coercive measures showed similar pattern; male gender was associated with more positive attitudes by six studies (Bregar et al., 2018;Husum et al., 2011;Khalil et al., 2017;Lind et al., 2004;Mohammed, 2015;Whittington et al., 2009) and also associated by four studies with more negative attitudes (Gandhi et al., 2018;Hasan & Abulattifah, 2018;Jonker et al., 2008;Wynn, 2003). Beghi et al. (2013) concluded in their review that male staff were associated with more restraint; our findings show that this conclusion might have been too firm. We found no conclusive evidence for an association of age, religion or the physique of the nurse and the use of coercion (Bowers, 2009;De Benedictis et al., 2011;Doedens et al., 2017;Kodal et al., 2018).
Some authors reported an association between personality factors and use and attitude of coercive measures, but the current studies are too small and inconsistent in methodology to draw conclusions.
When combining the findings of the perceived necessity of coercive measures for safety reasons and the inconsistency in the influence of nursing staff characteristics, we want to stipulate the possible importance of the feeling of safety of nurses. Despite the troubles of measuring this trait, some authors suggest that the feeling of safety of nurses may be associated with less use of coercive measures (Goulet & Larue, 2017;Vollema et al., 2012). This is in line with the findings of the nurses' attitude towards coercion.
Nurses that feel unsafe may very well view a coercive measure as necessary to restore safety, while nurses that feel safe may settle for alternative (less coercive) interventions. This is in line with the findings of Cusack, McAndrew, Cusack, and Warne (2016) that staffs' fear motivates for the use of coercion. Happell et al. (2012) and Wilson et al. (2017) reported nurses that were concerned that some nurses considered the necessity of a "last resort intervention" earlier than others. Feelings of safety or danger are not objective constructs, so interpersonal differences in perception and perspec- limitations of the studies on nursing staff characteristics in associations with coercive measures made it impossible to perform a meta-analysis. Another limitation is that the concept of attitude is not well defined and that several authors use other words to describe attitude. In our search, we also evaluated studies on perspectives, experiences and views of nurses to find additional studies on this matter. There were also specific limitations applicable to individual studies. The studies were of moderate to low methodological quality, which hinders the validity of the results of this review. Another limitation is that authors tend to report only significant associations or large effect sizes. Because of that, we cannot rule out the underreporting of some characteristics due to publication bias. We extracted the data from manuscripts as thoroughly as possible to summarize all reported (non-significant) results in our study.

| CON CLUS ION
The attitude of nurses towards coercive measures has changed over the years from a therapeutic paradigm to a safety paradigm. The current attitude towards use of coercive measures is not to treat patients, but to protect patients and staff from violence. Nurses consider coercive measures as necessary interventions and express the need for less intrusive alternatives. Although nurses recognize the negative consequences for patients, the frequent use of a specific coercive measure may decrease the value that nurses give to the negative consequences associated with that measure. The research on the influence of nursing staff characteristics is highly inconclusive. However, the feeling of safety of nurses may be a key concept in the prevention of coercive measures.

| IMPLI C ATI ON S FOR PR AC TI CE
We propose that mental health care could improve the focus on the constructs of safety and danger to protect patients from unnecessary use of coercive interventions. Lack of attention to the feeling of safety of nurses working at psychiatric wards can threaten further reduction in the use of coercive measures. Using coercive measures has been common practice in mental health care for centuries, as well as the debate on reducing them (Yellowlees, 1872).
It is part of our culture and, "culture eats strategy for breakfast" (Muir-Cochrane, 2018). It is important to invest in the feeling of safety of nurses to help them cope with changing the policy on using coercive measures. Evidence-based intervention programmes such as Safewards (Bowers, 2014) and Six Core Strategies (LeBel et al., 2014) can help nurses gain confidence in doing their job. To develop specific strategies to improve these feelings could be an interesting topic for researchers in the mental health field. Improvement of patient safety relies on qualified nurses that feel safe and are equipped for the difficult task they are facing when working in acute clinical psychiatry.

| RELE VAN CE S TATEMENT
The use of coercion is associated with adverse events. Nurses have influence on the decision to use coercive measures. Attitude of nurses towards coercion and nursing staff characteristics influence these decisions. This review summarizes the literature on the influence of attitude of nurses and nursing characteristics on the use of coercive measures. Our findings indicate, based on the attitude towards coercive measures and some evidence on perception of safety, the importance of the feeling of safety of nurses by clinicians, researchers and policymakers. This might be a more relevant road towards better quality of care than focus on nursing characteristics.

ACK N OWLED G M ENTS
The authors thank Joost Daams PhD (clinical librarian) for his extensive assistance in performing the electronic search, Lotta Raijmakers MSc for assisting in the title and abstract selection and Emma Verhoeven RN and Harald Jorstad MD PhD for their assistance in comprehending the articles in (respectively) French and Norwegian in the full-text selection and data extraction.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R CO NTR I B UTI O N S
All authors listed meet the authorship criteria according to the latest guidelines of the International Committee of Medical Journal Editors. All authors agree with the manuscript.