The first review on coercive interventions in 1994 stated that seclusion and mechanical restraint “can have deleterious physical and (more often) psychological effects on patients” [1]. Politics and clinical guidelines claim that coercive interventions have to be considered as interventions of last resort [2], though this claim is based on moral considerations rather than on empirical evidence regarding clinical and societal outcomes. Nevertheless, the variations in the use of seclusion or mechanical restraint point to powerful local effects which have been replicated over the years [1-9].

Manifold actions have been undertaken to reduce the use of coercive measures in the last 20 years. On a political level, claims have been pushed [10-15]. Several psychiatric associations have established guidelines [16, 17] and hospitals have trained staff in de-escalation [18, 19] and forced institutional changes (e.g. user involvement [20, 21] and crowding [22]). This progress was accompanied by mental health research that provided the basis for data assessment, benchmarking [7, 8], and evaluation of instruments (for overview, see [23, 24]).

For further reduction in the use of coercive interventions, it will be necessary to implement these guidelines into routine clinical care. In addition, it will be meaningful to identify patients at risk. However, up until now only few predictors of violent behaviour or self-directed aggression, which are mostly antecedent causes of coercive measures, and no predictors of coercive interventions themselves could consistently be found.

In research papers on predictors in this field, it is often discussed that studies are difficult to compare, legal premises [25] and even the definitions of coercive interventions differ significantly, different methodological approaches were undertaken, sample sizes were too small [9], and a reference population to the aggressive inpatients is difficult to define without being a source of bias itself [26].

Structural factors of wards, staff-related, patient-related, and treatment-related factors are considered to be associated with coercive measures [27]. Furthermore, environmental and interactional variables may be as important as the patient variables [26].

Whittington & Richter [28] argue that rather than focusing on patient variables, interactional effects between patient and staff behaviour should be considered in predicting violence in mental health settings. The same might be true for coercive interventions in the same settings.

In a recent study, using a multi-level approach and taking both, ward and patient characteristics into account, no predictors of coercive measures other than overactivity and aggressiveness could be identified, although substantial between-ward variance in the use of coercive measures was observed [9].

The aim of this study was to explore the impact of basic patient characteristics at admission, history within 24 hours before admission, and living conditions on the risk of experiencing coercive measures, controlling for ward characteristics in a multi-level approach.

In the sample, 362 (10.7%) patients experienced at least one coercive measure. Two hundred and fifty-eight patients (7.6%) experienced mechanical restraint, 195 patients (5.8%) experienced seclusion, and 55 patients experienced involuntary medication (1.6%). Both mechanical restraint and seclusion were experienced by 143 patients (4.2%), and 27 patients were subjected to the combination of mechanical restraint, seclusion, and involuntary medication. In addition, 12.5% of the male patients and 9.3% of the female patients were subjected to at least one coercive measure, and 10.7% of the patients with German citizenship were subjected to coercion. In patients with any other citizenship, 10.4% were subjected to coercive measures. Of those patients who experienced coercive measures, 150 (41.4%) had an F0.xx diagnosis, 15 (4.1%) had an F1.xx diagnosis, 129 (35.6%) had an F2.xx diagnosis, 32 (8.8%) had an F3.xx diagnosis, 22 (6.1%) had an F4.xx diagnosis, and 14 (3.9%) had an F6.xx diagnosis. Of those 1113 patients with recommended anti-psychotic medication (i.e. patients with an F2.xx or an F31.xx diagnosis), 102 patients (9.0%) had stopped anti-psychotic medication prior to admission without medical advice. Of those patients who had stopped anti-psychotic medication without medical advice, 24 patients (20.6%) experienced coercive measures. The type of coercive measure and mean number of coercive measures in each diagnostic category is displayed in Table 2.

This study supports the discussion on prevention of coercive interventions with empirical data concerning the predictors. Predictors of coercive interventions could be analysed regarding ward characteristics, patient characteristics at admission, history within 24 hours before admission, and living conditions.

We found a high rate of coercive measures in the organic mental disorder (F0) and the schizophrenia (F2) diagnostic group, and lower rates in the other diagnostic groups.

Although substance abuse and mania are generally considered to be two of the most important risk factors of aggressiveness and violence, these data are well in line with previous studies, including the hospitals investigated in this study [31]. At first glance, the low prevalence of coercive measures among people with substance use disorders (F1) and affective disorders including mania (F3) may seem striking. However, under conditions of psychiatric hospital treatment in Germany, a considerable proportion of admissions of patients with addictive disorders are planned on a voluntary basis and most of them are not intoxicated at admission. Thus, the rate of involuntary admissions and coercive measures is low in relation to the total number of admissions. Patients with mania according to ICD-10 are subsumed in the category "affective disorders", together with depressive disorders. The latter are exposed to coercive measures to only a small extent. Actually, patients with mania had the highest risk of receiving any type of coercive measure (Table 4). Many patients with suicidal behaviour, frequently with a comorbid diagnosis of a personality disorder, receive a primary diagnosis of adjustment disorder (F4) and some of them require coercive measures due to their behaviour.

Aggressive behaviour in the 24 hours prior to admission, in combination with an involuntary admission, turned out to be the best predictor for the use of coercion, regardless of the type of coercive measure applied. Compared to non-aggressive patients, aggressive patients had a higher risk of being exposed to coercive measures. A clear effect was also found for the voluntariness of the hospital stay. Voluntariness of stay appeared to be the best protective factor against coercive measures. Severity of illness increased the risk of coercion markedly. Taken together, these results are in accord with the findings of Husum et al. [9], as aggressive behaviour prior to admission and severity of illness seemed to influence both the risk of coercive measures as well as the rate of coercive measures. No conclusive evidence was found for the impact of living conditions.

There was no evidence that the variable "patient stopped anti-psychotic medication without medical advice" significantly increased the risk of coercion. Those patients even had a lower number of coercive incidents on average, indicating that re-establishing the previous treatment may limit crisis situations rather quickly. An indirect effect mediated by aggressive behaviour before admission seems rather unlikely. Among those patients who had stopped their anti-psychotic medication, 31% had shown aggressive behaviour within 24 hours prior to admission. Furthermore, of those patients who both had stopped medication and had shown aggressive behaviour, only 41% experienced a coercive measure. However, it should be recognised that we only assessed this item dichotomously. As no data on dose-reduction had been available, our results may be positively biased and must be interpreted with caution.

No effect on the risk of coercion was found for a comorbid diagnosis of substance abuse or substance abuse 24 hours prior to admission. The latter even reduced the risk for coercive interventions by number. This is well in accordance with clinical experience: intoxicated patients may frequently show aggressive behaviour requiring coercive interventions, but after some hours of detoxification they calm down and there is no need to repeat further coercive measures during an inpatient stay.

While the proportion of patients who were subjected to a coercive measure did not differ with respect to citizenship, after controlling for the other variables, non-German patients had a higher risk of coercion. This also applied for the psychotic subgroup. This may be due to cultural strangeness and language difficulties. A lack of oral proficiency can be assumed for many of the non-German patients.

Comparing this study with the study of Husum et al. [9], the data presented in this article are comparable regarding the sample size. However Husum et al. included only the involuntarily admitted patients in the analyses of the predictors of coercive interventions, which may lead to bias. In our sample, 34% of the involuntarily admitted patients experienced at least one coercive measure in contrast to only 6% of the voluntarily admitted patients.

As voluntariness of admission reduces the risk of coercive interventions, further efforts to reduce the threshold of psychiatric inpatient-treatment should be undertaken. Actions could comprehend a so-called trialogue (i.e. communication between patients, relatives, and health care professionals), a close cooperation between out and inpatient treatment, and establishing a continuous treatment forging close personal relationships with the staff to provide confidence.

Initially, the reduced risk of coercion for patients who threatened with suicide or attempted suicide within 24 hours prior to admission seems counterintuitive. However this might be explained by the fact that the use of coercive measures in suicidal or self-harming patients is considered widely inappropriate. Exceptions occur if the behaviour cannot be stopped by staff interventions or continuous surveillance.

The sample size was a strong point of this study. A large sample size allows precise statistical analyses, such as the presented multi-level approach. Further strengths of the study lie in the inclusion of different settings of psychiatric care which represent the reality of mental health care of a whole region. Furthermore, while many studies and meta-analyses are available for the predictors of violence, so far only a scare amount of literature exists concerning the predictors of coercion. In addition, we could investigate a comprehensive set of predictors for different forms of coercion. The basic patient documentation which had been established five years before provided a structured and consistent evaluation of each single stay. Inter-rater effects were minimised. In an independent evaluation [30], the data were more than sufficiently valid and reliable.

Unfortunately, detailed information on the type of aggression (assault, threatening behaviour, damage to property, self-harm) prior to admission was not available. Quantifying type and extent of aggression would have been useful, but such an item is not yet included in the basic documentation and therefore could not be considered as an independent variable. A differentiated rating of the psychopathological symptoms of each patient by trained interviewers would have been preferable from a methodological point of view, but is not feasible for several thousand patients under routine conditions. A shortcoming of the study is that mainly all diagnoses were analysed together. A possible difference between diagnostic subgroups was only tested with respect to patients with schizophrenia vs. all other diagnoses and vs. bipolar disorder. Only a subgroup of patients with a F2 or F3.1 diagnosis (schizophrenia or bipolar disorder) was analysed separately in order to assess the effect of stopping recommended medication without consultation. This approach may blur differences between diagnostic groups. In our study, ward characteristics were limited to only two variables. This was due to the use of basic documentation data without any further data acquisition. As the basic documentation does not contain any items on ward characteristics, only information about the total number of admissions within a year and the total number of incidents of aggressive behaviour could be obtained. Possibly crucial ward characteristics, such as staff attitudes towards the use of coercion and excess of bed occupancy in wards, were not assessed. While Virtanen et al. [31] found an association between overcrowding and assaults of patients towards staff, Husum et al. [9] found no substantial association between staff attitudes towards the use of coercion and the use of coercion itself. In our study, a substantial proportion of variation between wards remained unexplained by the included variables. The complexity of clinical situations could not be recognised. Due to this shortcoming, further analyses should take into account not only structural characteristics of wards, but should also capture features of the situation imminent to a coercive measure.

The results of this study show that aggressive behaviour before admission and involuntary commitment are meaningful risk factors for the use of coercive measures. Reducing coercive measures may be more successful with early identification of aggressive behaviour and with efforts to lower the threshold of admission.

To identify patients at risk, it is most important to intensively monitor patients with aggressive behaviour prior to admission and with a greater severity of psychopathological symptoms. Further efforts should be undertaken to reduce the threshold for psychiatric inpatient treatment and to increase the proportion of voluntarily admitted patients.

The authors declare that they have no competing interests.

EF: Statistical analyses, discussion

TS: Introduction, discussion

FE: Data acquisition, discussion

CU: Discussion

JB: Introduction, discussion

All authors read and approved the final manuscript.

The authors confirm that this article content has no conflicts of interest.