Association between behavioral parameters of men in psychiatric emergency department video recording and subsequent violence in the psychiatric ward

Abstract

Violence risk is a major challenge among acute psychiatric inpatients. The study aimed to predict violent behavior risk in an acute psychiatric ward using video recordings from the emergency department. 69 videos of the emergency department recording the first ten minutes following patients’ arrivals were included. Psychiatrists watched the videos, completed relevant Brief Psychiatric Rating Scale items and answered intuitive questions about each patient’s risk of violence. Demographic and clinical data were also collected. Motoric mannerisms as rated in the BPRS significantly differed between violent and non-violent patients (p < 0.05). Additionally, we found a significant correlation between intuitive prediction of violence and actual violence (p = 0.008). Violent behavior was predicted in 42.1% of the cases by the intuitive evaluation compared to 11.5% mistakenly evaluated patients. Logistic regression revealed that the intuitive question and the BPRS items regarding tension and motoric mannerism created a successful model for predicting violence with 88.2% sensitivity and 72.5% specificity. We sought to define the factors that most accurately predict violence in the acute psychiatric ward, based solely on behavior in the emergency department. Intuitive impressions of clinicians and motoric mannerisms should be considered when evaluating patients for potential violent behavior.

Introduction

Exposure to violence in psychiatric wards affects the victims, mentally and emotionally and could result in physical impairment or disability [1]. Additional consequences may include stigma of mental health patients and difficulty in hiring staff for psychiatric hospitals. There may also be financial implications for the staff resulting from injuries inflicted by patients, sick leave, and potential shuttling back and forth to court [2]. The prevalence of violence in acute psychiatric wards is relatively high and can reach 18% [3]. Thus, short-term prediction of violence is a necessary focus of clinical care [4]. Indeed, the high rate of violence points toward the need for improvement in the identification and prevention of violent behavior in psychiatric wards in order to provide safer environments.

Several assessments tools are currently used to evaluate and manage violent incidents at the workplace, especially in psychiatric hospitals. For example, The Staff Observation Scale - Revised (SOAS-R) questionnaire [5]. Assessments tools had been developed to increase the accuracy of predicting imminent violence, such as the Brøset Violence Checklist (BVC) and the Dynamic Appraisal of Situational Aggression (DASA-IV) [6]. Brief psychiatric Rating Scale (BPRS) was also investigated for its correlation with aggression. It was found that generally, subscales of hostile/suspiciousness were found higher in aggressive patients while subscales of depression/anxiety were found lower in aggressive patients, compared to non-aggressive patients [7]. Unstructured judgments of psychiatrists using their clinical experience and judgment is also a common practice, while it remains unclear whether risk assessment tools outperform unstructured judgment [8].

However, despite their importance, all these tools specifically predict short-term risk, and the need for daily administration of these scales limits their usefulness [2]. Additionally, those assessments require staff involvement, which is costly and not always available. Therefore, there is a growing need for other assessments, as computerized automatic assessments of patients.

Others methods of violence assessment have recently been developed, for example identification of violence using computerized analysis of video recordings [9]. This method was successful in identifying violence in real time, enabling the optimization of interventions.

The present study assessed the association between video recordings in the emergency department and violent behavior in the psychiatric ward in the three days following admission. We propose a model for the prediction of violent behavior, based on video analysis of patients’ behavior in the emergency department, in order to clarify which risk factors are associated with subsequent violent behavior in the psychiatric ward.

Materials and methods

Study population

Patient files of men assessed in the emergency department who were then consecutively admitted, voluntary or involuntary to a closed inpatient department from December 2020 to February 2021 were included in the study. Inclusion criteria were male gender, age above 18, and hospitalization in a closed psychiatric ward. The study included only men in order to have sufficient sample size as violence is more prevalent in men. Exclusion criteria included intellectual disability, head trauma and unstable physical illness, display in the videos of any type of physical violence (to another patient, staff or property) or administration of medication/injections, as well as duration of hospitalization less than 3 days. For the purpose of the study 93 patient files were reviewed. Twenty-four of the patients did not meet the inclusion criteria and were excluded from the analyses (short duration of hospitalization). Three videos were removed due to repeated hospitalization and bias possibility. None of the videos included physical violence or administration of medication/injections.

Assessments

The videos recorded in the emergency department prior to admission included the 10 first minutes of the patients’ arrival sand were retrieved from the security cameras, without sound. The emergency room is equipped with 4 K high resolution IP security cameras. The videos typically included the entry of the patient at the emergency department, primary conversation with the receptionist and nurse and waiting time. Data regarding the patient’s accompanying person in the emergency department were retrieved from the emergency department records.

Demographic and clinical data were collected from the inpatients’ electronic medical records and included age, family status, primary psychiatric diagnosis, physical diagnosis, medications (dose and type), duration of the psychiatric disorder and number of previous admissions.

The following data were retrieved from the first three days of hospitalization: administration of antipsychotics (type and dosage), antipsychotic injections (yes/no), benzodiazepines (yes/no), benzodiazepine injections (yes/no), mood stabilizers (yes/no), antidepressants (yes/no) and anticholinergics (yes/no).

Four senior psychiatrists, with more than ten years’ experience, watched the video recordings. The researchers verified that the psychiatrists had no contact with the patients prior to the study, neither exposed to the patients’ name. The psychiatrists were blind to outcome: violence or not in the department. They assessed the following measures:

1. Eight parameters from the Brief Psychiatric Rating Scale (BPRS) [10], were included in the assessments: tension, anxiety, mannerisms and posturing, hostility, suspiciousness, motor retardation, uncooperativeness and excitement. These parameters were chosen because they can be quantified in video recordings.

2. Intuitive question: “do you think this patient will be violent in the closed ward?“(Yes/no).

Patients were evaluated as violent in the ward if one or more of the following occurred in the first three days of hospitalization: restriction due to violence, seclusion due to violence, physical harm inflicted on another patient, staff, or property and/or verbal threats.

Statistical analysis

Study participants were categorized into two groups: inpatients who exhibited violence in the ward, and non-violent inpatients. Between group comparisons were conducted for all demographic and clinical data using t-test, Mann-Whitney U test or chi-square, as appropriate. Finally, a logistic regression model was built in order to predict violence in the ward according to the maximum likelihood method, so that with the help of a minimum number of independent variables the model will be explained in an optimal way. Therefore, the model included all the variables that were found to be significant in univariate tests, minus variables that were found to be over correlated with each other, and variables that increased the likelihood of the model were added.

Statistical analyses were conducted using IBM SPSS statistics version 28 (IBM, Chicago, IL). P < 0.05 was considered statistically significant.

Ethics

Admission videos from the emergency department were kept in a locked and secured place. Videos of patients were kept separately from the data collected about the patients. Anonymity was maintained throughout all study stages, from data collection to analysis. The study was approved by the IRB Committee of Lev Hasharon Mental Health Center on 9th December 2020 (Trial number LH13/2020). Due to the retrospective design of the study, the IRB Committee of Lev Hasharon Mental Health Center waived the participants from signature on informed consent. The research was performed in accordance with the Declaration of Helsinki.

Results

Characteristics of the study participants

Of 69 inpatients who were admitted following an examination in the emergency department, 17 patients exhibited violent behavior in the ward (24.6%).

Inpatients who received intramuscular (I.M) benzodiazepines (p < 0.001; O.R = 11.43, 95% C.I: [3.71, 11.43]) or I.M anti-psychotics (p < 0.001; O.R = 10, 95% C.I: [2.64, 41.7]) were more likely to exhibit violence (Table 1).

Table 1.

Differences in demographic and clinical variables according to violent behavior in the department

Violence			Patients (N = 69)		p
			No (n = 52)	Yes (n = 17)
Normally distributed variables: Mean (SD)
Age			39.2 (13.5)	36.6 (13.5)	0.48
Abnormally distributed variables: Median (Q1, Q3)
Prior hospitalizations			3 (0, 9)	3 (0, 23)	0.74
Illness duration (years)			8 (0.6, 18)	8 (0.75, 18)	0.82
Categorical variables: n (%)
Benzodiazepines P.O			42 (80.8%)	15 (88.2%)	0.7
Benzodiazepines I.M			6 (11.5%)	10 (58.8%)	< 0.001**
Antipsychotics I.M			16 (30.8%)	14 (82.4%)	< 0.001**
Antidepressants			5 (9.6%)	0	0.32
Mood Stabilizers			8 (15.4%)	4 (23.5%)	0.47
Anticholinergics			34 (65.4%)	9 (52.9%)	0.35
Family status	Married		5 (9.6%)	2 (11.8%)	0.8
	Single, Divorced, Widow		47 (90.4%)	15 (88.2%)
Primary diagnosis	Schizophrenia		39 (75%)	7 (41.2%)	0.22
	Bipolar Disorder		3 (5.8%)	5 (29.4%)
	Acute psychotic disorder		4 (7.7%)	4 (23.5%)
	Other		6 (11.5%)	1 (5.9%)
Accompanying person to emergency department	Alone		9 (17.3%)	1 (5.9%)	0.046*
	Family member, Doctor, Social worker		42 (8.8%)	13 (76.5%)
	Security		1 (1.9%)	3 (17.6%)
Intuitive prediction of violence	Yes		6 (11.5%)	8 (42.1%)	0.008**
	No		46 (88.5%)	9 (57.9%)

* p < 0.05 ** p < 0.01

Additionally, the results show an association between violent behavior in the department and type of person that accompanied the patient to the psychiatric emergency department (p < 0.05). Inpatients who came to the emergency department accompanied by a security guard were found to be more violent in the department compared to those who came alone or with family/doctor/social worker. Cramer’s V effect size was found strong between violent behavior in the department and type of accompanying person (V = 0.31). Other factors, such as age, prior hospitalizations, illness duration, family status, primary diagnosis, treatment with oral benzodiazepines, antidepressants, mood stabilizers and anticholinergic medications did not differ between the groups (Table 1).

Relation between BPRS in the emergency department and actual violence in the department

According to an independent t-test, the total BPRS scores did not differ significantly between the violent and non-violent inpatients (p = 0.44). However, when analyzing each item of the BPRS scale separately, a Mann-Whitney U test for abnormally distributed means revealed that mannerisms and posturing as rated in the emergency department differed significantly between violent and non-violent inpatients (p < 0.05) (Table 2).

Table 2.

BPRS clinical parameters with potential correlation to aggression

Violence	Patients (N = 69)		P
	No (n = 52)	Yes (n = 17)
Normally distributed variables: Mean (SD)
Total score BPRS	18.6 (7.7)	19 (7.4)	0.44
Abnormally distributed variables: Median (Q1, Q3)
BPRS tension	3.5 (2.5, 4.5)	3.5 (2, 4.25)	0.45
BPRS mannerisms and posturing	1.5 (1, 3)	2.5 (2.5, 4)	0.03*
BPRS hostility	2 (1, 3.5)	2.5 (1, 3.5)	0.9
BPRS suspiciousness	2.5 (1.5, 4)	3 (1.5, 4.25)	0.9
BPRS motor retardation	1.5 (1, 2.5)	1.5 (1, 3.5)	0.26
BPRS lack of cooperation	1.5 (1, 3.5)	1 (1, 3.25)	0.15
BPRS excitement	3.5 (1.5, 4.25)	3.5 (1.5, 4.25)	0.7

* p < 0.05

Relation between intuitive prediction of violence in the department and actual violence in the department

According to Fisher’s exact test for linear trend, we found a significant correlation between intuitive prediction of violence and actual violence in the department (O.R = 5.58, 95% C.I: [1.6, 19.4]; p = 0.008). Eight out of 17 violent inpatients in the department were rated violent in the intuitive prediction (42.1%), versus six out 52 non-violent inpatients in the department (11.5%) (Table 1).

Logistic regression for prediction of violent patients

The general model included three variables (intuitive assessment, BPRS tension and BPRS mannerisms and posturing ≥ 3). The model was found significant (p < 0.001), with R² = 0.395, suggesting that these the variables explain almost 40% of the likelihood of a patient’s violent behavior.

Patients who were rated with a positive intuitive assessment of violence were found 18.5 times more likely to be violent, compared with patients with a negative assessment [O.R = 18.59, 95% C.I: (2.6, 131.1), p < 0.01[. Additionally, BPRS mannerisms and posturing ≥ 3 increased patient’s probability of violence 10.8 times [O.R = 10.82, 95% C.I: (2.3, 51.5), p < 0.01[, while every BPRS tension score decreased patient’s probability of violent behavior by 54% [O.R = 0.46, 95% C.I: (0.25, 0.85), p < 0.05)(Table 3).

Table 3.

Logistic regression for prediction of violent patients

	Exp (B)	95% C.I	p
Positive intuitive assessment	18.59	(2.64, 131.1)	< 0.001**
BPRS mannerism and posturing> 3	10.82	(2.27, 51.48)	< 0.001**
BPRS tension	0.46	(0.25, 0.85)	< 0.05*

Chi square = 10.56, R² = 0.395, p < 0.001

* p < 0.05; ** p < 0.001

Using a cut-off of ≥ 0.2 to define violent behavior in a patient yielded a sensitivity of 88.2% for detecting violence in patients and a specificity of 72.5% for ruling out violent behavior.

Discussion

Detection of patients at high risk of violence is crucial for effective treatment and functioning in psychiatric wards. Hence, mental health professionals are required to perform risk assessments on a regular basis as part of hospital policy. Risk assessment is an estimate of the likelihood that a certain person will become violent, and it contributes to the development of future intervention policies for preventing or decreasing such behaviors. In our study, we sought to define which factors are the most accurate for predicting violence in the acute psychiatric ward, based on behavior in the emergency department, with no previous knowledge about the patient. An assessment of this type, watching security cameras, is innovative and might be a platform for new technologies aimed at predicting violence. This study searched for predictors other than routine clinical evaluation.

Violence is a complex phenomenon, related to a variety of biological, social and psychopathological factors [11]. Previous studies have shown that the best predictor of violent behavior for an inpatient is probably previous violent episodes [12], however this information is not always available. Our study relies on behavioral parameters available in the emergency department without the need to examine medical or behavioral history.

We found that the intuitive question regarding the likelihood of violent behavior (yes/no) was related to actual violence in the department. Psychiatrists more accurately rated probable violent behavior in patients in line with their response to the intuitive question. Responding to an intuitive yes/no question could be related to quick thinking. The psychiatrists probably did not engage in slow and analytical thinking when answering the question. In such cases, cognitive system number one (quick review) might be activated [13]. Psychiatrists subconsciously weighed all relevant parameters for predicting violence and succeeded in predicting violence without being able to tell which parameters they used. We should mention that all the psychiatrists who participated in our study had over ten years of experience and probably subconsciously developed techniques for identifying potential violence among patients. Human intuition involves many factors, some unconscious. Previous literature also shows that clinical methods (staff judgment based on knowing the patient’s history, their expertise, experience, and intuition) are still preferred in some cases [14]. Still, in our research, the ability of psychiatrists to correctly assess risk prediction stands at about 40%. Previous research found that the ability of psychiatrists to assess the risk of violence without knowing any background about the patient, compared to risk assessment tools, is still controversial [8, 15]. Misclassification of risk assessment (false positives) can result in unnecessary treatment, seclusion and stigma, and the failure to anticipate violence (false negatives) can prevent much needed treatments [16].

Our study found that the total BPRS score did not differ between violent and nonviolent individuals. However, sub-scales such as low BPRS tension and high BPRS mannerism and posturing score, together with the intuitive question, contributed significantly to the regression model of violence prediction. Interestingly, previous research revealed other findings, such as the BPRS sub-scale hostility that was previously found to be helpful in predicting violence [17]. Motoric mannerisms might express a more severe state of psychosis and hallucinatory behavior, as higher rates of primary and motor coordination were found to be associated with psychosis [18]. Additionally, psychosis and hallucinatory behavior are strongly associated with severe violence [19]. We propose that in our study, the relationship between motoric mannerisms and violence was mediated by psychotic state and hallucinations. Our results also showed that low tension was a predictor of violence in the department. Previous studies have shown that violence is related to impulsivity [20], which is the inability to inhibit behavioral impulses, is generally not forethought and cannot be predicted. In general, psychotic patients act impulsively from internal stimulations therefore tension might be lower in violent patients compared to non-violent patients.

In the present study, patients with violent behavior received more intramuscular benzodiazepines and antipsychotics during the first three days of hospitalization than non-violent patients. This was expected, as those medications are generally administered intramuscularly for the immediate treatment of acute agitation or violent behavior [21, 22].

Finally, we found in our study that patients who came to the emergency room accompanied by security guards were found to exhibit more violent behavior. This was expected as security escorts are generally required when patients are at risk for violent behavior or referral to a psychiatric emergency department following a violent altercation.

The main limitation of our study is the lack of sound in the videos. Some variables, such as loud voices or screaming, could significantly contribute to predicting violence. Additional limitations are the relatively small sample size, the lack of information about voluntary or involuntary admission and lack of information about the patients’ ethnicity. Also, in our study, only the first ten minutes in the emergency room could be analyzed. Finally, our prediction is only short-term (first days of hospitalization), and longer and repetitive watching is required for long-term prediction. Watching patients for a repetitive and longer period in the emergency room or the department could provide more information about the risk for violence.

To conclude, intuitive impressions of clinicians and motoric mannerisms should be cautiously considered when evaluating potential violent behaviors in patients. Replication studies among additional patient populations (such as women) are warranted. Future studies may be able to employ artificial intelligence / machine learning to evaluate similar parameters on a larger patient population for longer durations of observation to more accurately predict the likelihood of violent behavior among psychiatric inpatients. For example, training algorithms to learn recognizing body language patterns associated with violent behavior in psychiatric care could identify such actions in security cameras and alert the staff [23].

Author contributions

Realization of the study: R.R, S.H, D.P, O.S, M.M, A.P, A.S. Conception: R.R, S.H, S.W, E.B.B, A.S. Statistical analyses: E.B.B. Main manuscript writing: R.R, S.W, E.B.B, A.S. Review of the manuscript: N.B, R.S. All authors reviewed the final manuscript.

Funding

This study received no funding.

Data availability

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the IRB Committee of Lev Hasharon Mental Health Center on 9th December 2020 (Trial number LH13/2020). Due to the retrospective design of the study, the IRB Committee of Lev Hasharon Mental Health Center waived the participants from signature on informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.