A review and meta-analysis of the patient factors associated with psychiatric in-patient aggression

Abstract

Objective

To combine the results of earlier comparison studies of in-patient aggression to quantitatively assess the strength of the association between patient factors and i) aggressive behaviour,ii) repetitive aggressive behaviour.

Method

A systematic review and meta-analysis of empirical articles and reports of comparison studies of aggression and non-aggression within adult psychiatric in-patient settings.

Results

Factors that were significantly associated with in-patient aggression included being younger, male, involuntary admissions, not being married, a diagnosis of schizophrenia, a greater number of previous admissions, a history of violence, a history of self-destructive behaviour and a history of substance abuse. The only factors associated with repeated in-patient aggression were not being male, a history of violence and a history of substance abuse.

Conclusion

By comparing aggressive with non-aggressive patients, important differences between the two populations may be highlighted. These differences may help staff improve predictions of which patients might become aggressive and enable steps to be taken to reduce an aggressive incident occurring using actuarial judgements. However, the associations found between these actuarial factors and aggression were small. It is therefore important for staff to consider dynamic factors such as a patient's current state and the context to reduce in-patient aggression.

Summations

• Psychiatric in-patients who are younger, male, admitted involuntarily, not married, have a diagnosis of schizophrenia, have a greater number of previous admissions, a history of violence, a history of self-destructive behaviour and a history of substance abuse were more likely to be aggressive than non-aggressive during their stay.
• Psychiatric in-patients who are female and have a history of substance abuse or a history of violence were more likely to be repetitively aggressive than aggressive once during their stay.

Considerations

• The associations between patient characteristics and aggression were small suggesting that other factors may be helpful in predicting aggression.
• There were significantly high levels of heterogeneity across the articles entered into most of the meta-analyses.
• A relatively small number of comparison studies were found relative to the number of publications on in-patient aggression suggesting that this is an underused study design.

Introduction

One of the main objectives of acute in-patient wards is to ensure that patients and staff are safe from harm [1]. However, patient's behaviour can be aggressive, in some cases violent and occasionally suicidal. This can make treatment and supervision a very difficult task for the psychiatric staff involved in their care as well as threatening the safety of both patients and staff [2, 3].

Estimates of the percentage of patients who are aggressive during their stay on acute psychiatric wards are extremely variable, with figures between 8% and 44% cited in the literature [4-6]. A third of in-patients report having experienced violent or threatening behaviour while in psychiatric care [7]. Figures for staff were somewhat higher with 41% for clinical staff and nearly 80% of nursing staff working in in-patient units reporting they had experienced aggressive behaviour. These incidents of aggression may also have a negative physical and psychological impact on patients and staff [8-11]. It is important, therefore, to know the strength of association between risk factors for in-patient aggression and the extent to which these disruptive and distressing events can be predicted and prevented.

A recent systematic review of aggression in psychiatric wards [12] examined the variables that were most frequently associated with aggression or violence. They found a history of previous aggressive incidents, a longer period of hospitalization, involuntary admission, the presence of impulsiveness and hostility and if the aggressor and victim were of the same gender the factors most frequently involved were incidents of aggression. Weaker evidence also indicated that alcohol/drug misuse, a diagnosis of psychosis, a younger age and the risk of suicide were also related factors to aggressive events. However, this study did not perform a meta-analysis combining the results. This approach might suffer from Type 1 errors owing to chance findings of significant results in studies that examined a large number of factors. It may also lead to Type 2 errors caused by the lack of statistical power characteristic when examining a series of studies that each report a small number of events.

There are no previous meta-analyses of aggression by in-patients. Using a quantitative, meta-analytic approach had a number of advantages. It combines several studies allowing an estimation of the strength, variation and generalizability of associations across studies, reducing the likelihood of Type 1 and Type 2 errors. In an effort to better understand the factors associated with in-patient aggression we conducted a systematic review and meta-analysis of comparison studies.

Aims of the study

The aim of the study was to combine the results of earlier comparison studies of aggressive and non-aggressive psychiatric in-patients to quantitatively assess the strength of the association between patient factors and aggressive behaviour. Secondary aims were to identify differences between patients who were repetitively aggressive and those who were only aggressive once during their admission, and to explore differences by setting (acute vs. forensic wards).

Material and methods

Studies

Search methods

A search was carried out using the following databases: MEDLINE, PsychInfo, Cochrane Clinical Trials, EMBASE Psychiatry, CINAHL and DARE, and the following keywords: (psychiat* or mental*) and (hospital or ward or inpatient or in-patient) and (aggressi* or violen*). No attempt was made to search for unpublished results. As the literature accumulated, further references were obtained by following up citations.

Inclusion and exclusion criteria

The publications included were those that were peer reviewed (journal articles, book chapters or reports with primary empirical data), published in English between 1960 and 2009 and examined in-patient aggression (acute, forensic, rehabilitation units, veterans and psychiatric intensive care unit settings). Excluded publications included secondary data, non-empirical data and in-patient data from child.

Search outcome

The final number of identified empirical studies was 428. Of these, 75 studies made some sort of comparison between groups of patients as can be seen in (Table 1).

Table 1. Types of comparison studies found

Type of comparison	N	%
Aggressive vs. Non-aggressive Random	29	38.67
Aggressive vs. Non-aggressive Controls	7	9.33
Repeater/High vs. Single/Low	6	8
Repeater/High vs. Single/Low vs. Non-aggressive Random	6	8
Repeater/High vs. Single/Low vs. Non-aggressive Control	4	5.33
Other	23	30.67

Aggressive vs. non-aggressive comparison studies

Inclusion criteria

From the 29 aggressive vs. non-aggressive random group, studies were excluded from the analysis if there were any missing data [13-15], if it was unclear whether the aggression had occurred on the ward when defining the groups [16, 17] and if the demographic data were reported in terms of the number of aggressive incidents rather than the number of aggressive patients [18]. From the aggressive vs. non-aggressive control studies, four studies were excluded [19-22] as they were comparing factors other than the patient demographics we were interested in. Ten studies reported more than one comparison. Where aggressive patients were split into those who had been physically aggressive vs. verbally aggressive or aggressive against objects [5, 23, 24], the data were combined to produce one aggressive group for comparison. This was also the procedure for studies that compared persistently and transiently aggressive patient over the study period [25, 26]. Some studies made further comparisons using a subset of patients from either the aggressive group [27, 28] or the non-aggressive group [29]. These extra comparisons were not included in the analysis. Two studies were excluded from the analysis [30, 31] as the non-aggressive group was matched to the aggressive group.

A total of 34 studies (see Table 2) remained and were analysed in the present series of meta-analyses. Where available the following information was collected from each study: patient demographics (for gender, ethnicity, marital status, diagnosis, history of violence, suicide, drug abuse and admission type – this was split into binary data, e.g. the proportions of patients in the aggressive and non-aggressive group that were male or female; for age and years of education continuous data were collected, e.g. means and standard deviations for each group), size of sample, type of ward, violence definition (verbal, physical against others, physical against objects, physical against self) and country.

Table 2. Characteristics of studies for meta-analysis

Author	Country	Setting	Data	Status	Number	Violence definition	Measurement	Duration	Type of sample
agg, aggressive; non-agg, non-aggressive; V, verbal aggression/threat; P, physical aggression against others; O, physical aggression against objects; S, aggression against self; SIR, Standard Incident Reports; OAS, Overt Aggression Scale [68]; SOAS, Staff Observation Aggression Scale [69]; MOAS, Modified Overt Aggression Scale [70]; NOSIE, Nurses' Observation Scale for In-patient Evaluation [71]; REFA, Report Form for Aggressive episodes [72]; VS, the Violence Scale [73]; C & C, Conflict and Containment.
Barlow et al. [27]	Australia	Acute	Gender, Diagnosis, No. previous admission	agg/non-agg	174/1096	VPOS	SIR	18 months	Random
Blomhoff et al. [37]	Norway	Acute	Gender, Age, History of C & C	agg/non-agg	25/34	P	SIR	12 months	Random
Coldwell et al. [48]	UK	Forensic	Age	agg/non-agg	31/20	PO	Patient notes	12 months	Random
Daffern et al. 2005 [28]	Australia	Forensic	Age, Gender, Diagnosis	agg/non-agg	105/127	VPO	OAS	12 months	Random
Dietz et al. 1982 [38]	USA	Forensic	Age, Ethnicity, Years of education	agg/non-agg	64/147	P lead to seclusion	SIR	12 months	Random
Dolan et al. 2008 [39]	UK	Forensic	Age, Gender, Ethnicity, Marital status, Diagnosis	agg/non-agg	79/68	VP	SIR	Missing	Random
Doyle et al. 2002 [40]	UK	Forensic	Age, Gender, Ethnicity, Marital status, Diagnosis	agg/non-agg	45/52	VP	Patient notes	3 months	Random
Edwards et al. 1988 [57]	UK	Mixed	Marital status, Diagnosis, Admission type, History of C & C	agg/non-agg	25/25	P	Interviews	12 months	Matched for Age and Gender
Fullam et al. 2008 [41]	UK	Forensic	Age, Yrs in education, medication	agg/non-agg	33/49	PO had to be instigated	SIR	Missing	Men/schizophrenics
Grassi et al. 2001 [4]	Italy	Acute	Age, Gender, Marital status, Diagnosis	agg/non-agg	116/1418	VPO	SOAS	60 months	Random
Harris et al. 1983 [49]	USA	Forensic	Age, Yrs in education, Diagnosis, No. previous admissions	agg/non-agg	45/45	P	SIR	60 months	Random
Hillbrand et al. 1996 [47]	USA	Forensic	Age, History of C & C	agg/non-agg	79/79	P lead to injury	Patient notes	36 months	Random
Hoptman et al. 1999 [50]	USA	Forensic	Age, Ethnicity, Yrs. in education	agg/non-agg	60/123	P	NOSIE	3 months	Men
James et al. 1990 [51]	UK	Acute	Age, Gender, Ethnicity, Diagnosis, Admission type	agg/non-agg	64/216	POS	SIR	15 months	Random
Karson et al. 1987 [52]	USA	Research	Age, Gender, Diagnosis, No. previous admissions, History of C & C	agg/non-agg	45/95	P	Patient notes	135 months	Aggressive group had not responded well to neuroleptic treatment
Kennedy et al. 1995 [55]	UK	Forensic	Gender, Ethnicity, Diagnosis, History of C & C	agg/non-agg	27/54	PO 10 +  incidents	SIR	48 months	Random
Ketelsen et al. 2007 [29]	Germany	Mixed	Age, Gender, Marital status, Diagnosis, Admission type, No. previous admission	agg/non-agg	171/2039	VPO	SOAS	12 months	Random
Krakowski et al. 1989 [25]	USA	Mixed	Diagnosis & History of C & C	agg/non-agg	77/40	VPO	Patient notes	2 months	Matched for age, gender, race and chronicity of illness
Krakowski et al. 1997 [26]	USA	Mixed	History of C & C, Admission type	agg/non-agg	75/62	P	MOAS and Patient notes	26 months	Matched for Age, Gender, Ethnicity, Diagnosis, Length of stay
Lam et al. 2000 [42]	USA	Acute	Age, Gender, Ethnicity, Diagnosis, Admission type, History of C & C	agg/non-agg	76/314	P lead to injury	SIR	129 months	Random
Lanza et al. 1994 [22]	USA	Veteran hospital	Ethnicity, Marital Status, Diagnosis	agg/non-agg	36/36	VP	Patient notes	Missing	Matched for Age and Gender
McKenzie et al. 2005 [33]	UK	Forensic	Age, Gender	agg/non-agg	70/24	PO	SIR	2 weeks	Random
McNiel et al. 1988 [23]	USA	Acute	Age, Gender, Ethnicity, Marital status, Diagnosis, History of C & C	agg/non-agg	138/100	VPO	Patient notes	3 days	Involuntary patients
Mellesdal et al. 2003 [43]	Norway	Acute	Age, Gender, Diagnosis, Admission type	agg/non-agg	98/836	VP	REFA	36 months	Some day patients, numbers not specified
Nijman et al. 1997 [3]	Netherlands	Acute	Age, Gender, Diagnosis, Admission type	agg/non-agg	31/31	VPO	SOAS	6 months	Random
Nijman et al. 2002 [44]	Netherlands	Acute	Age, Gender, Diagnosis, Admission type	agg/non-agg	31/58	VPOS	SOAS	9 months	Random
Oulis et al. 1996 [45]	Greece	Acute	Age, Gender, Diagnosis	agg/non-agg	32/104	VPOS	MOAS	5 days	Random
Raja et al. 2005 [5]	Italy	PICU	Age, Gender, Marital status, Yrs in education, Diagnosis, Medication, Admission type, History of C & C	agg/non-agg	70/1322	P	VS	72 months	Random
Raja et al. 1997 [24]	Italy	PICU	Age, Gender	agg/non-agg	22/256	P	Patient notes	13.5 months	Random
Soliman et al. 2001 [53]	UK	Acute	Age, Gender, Diagnosis, Medication, Admission type, History of C & C	agg/non-agg	49/280	PO	SOAS	12 months	Random
Tardiff et al. 1982 [54]	USA	Chronic	Age, Gender, Diagnosis	agg/non-agg	384/4780	P	NOSIE	3 months	Random
Troisi et al. 2003 [46]	Italy	Acute	Age, Admission type	agg/non-agg	20/20	VPOS	MOAS	6 months	Men
Walker et al. 1994 [56]	UK	PICU	Gender, Ethnicity, Diagnosis, History of C & C	agg/non-agg	16/32	P	SIR	6 months	Random

Once only vs. repeated aggression comparison studies

Inclusion criteria

From the six repeater vs. once aggressive group studies one was excluded from the analysis as a large number of patients included were on out-patient wards [32]. Ten studies reported more than one comparison. Of these half [25-27, 30, 31] made some sort of comparison between persistently aggressive and transiently aggressive patients. mckenzie et al. [33] was also included as patients within the aggressive group were analysed further based on their number of aggressive incidents. This left a total of 11 studies (see Table 3) to be analysed in the present series of meta-analyses. Where available similar information as the aggressive vs. non-aggressive comparison studies was collected from each study.

Table 3. Characteristics of studies for meta-analysis among high- and low-aggressive psychiatric in-patients

Author	Country	Setting	Data	Status (repeaters vs. single aggressors)	Number	Violence definition	Measurement	Duration	Type of sample
V, verbal aggression/threat; P, physical aggression against others; O, physical aggression against others; S, aggression against self; SIR, standard incident reports; SOAS, staff observation aggression scale [69]; MOAS, modified overt aggression scale [70]; VS, the Violence Scale [73]; ASAP, Assaulted Staff Action Program report forms [74]; C & C, Conflict and Containment.
Barlow et al. 2000 [27]	Australia	Acute	Age	Multiple/Single	70/104	VPOS	SIR	18 months	Random
Convit et al. 1990 [62]	USA	Psychiatric Hospital	Age, Gender, Diagnosis	3+/1–2	70/243	P	SIR	6 months	Random
Flannery et al. 2002 [58]	USA	Other	Age, Gender, Diagnosis, History of C & C	3+/1	61/566	VPO Sexual	ASAP	120 months	Random
Grassi et al. 2006 [59]	Italy	Other	Age, Gender, Marital Status, Education, Diagnosis, No of previous admissions, History of C & C	2+/1	65/95	VPOS	SOAS	84 months	Random
Krakowski et al. 1989 [25]	USA	Special Unit	Age, Gender, Ethnicity, Diagnosis, History of C & C	Persistent/Transient	38/39	VPO	SIR	28 days	Consecutive admissions to special unit designed to manage assaultive behaviour
Krakowski et al. 1989 [30]	USA	Special Unit	Age, Gender, Ethnicity, Yrs of educations, History of C & C	2 + /0-1	28/27	VPO	SIR	Missing	Schizophrenic admitted to special unit designed to manage assaultive behaviour
Krakowski et al. 1997 [26]	USA	Admission wards	Age, Gender, Ethnicity, Diagnosis, Medication, Admission Type	Persistent/Transient	34/43	P	MOAS	28 days	Random
Krakowski et al. 1999 [31]	USA	Admission wards	Age, Gender, Ethnicity, Diagnosis, Medication, History of C & C	Persistent/Transient	44/52	VP	MOAS	28 days	Schizophrenic
McKenzie et al. 2005 [33]	UK	Forensic	Age, Gender	10 + /1-5	17/40	PO	SIR	2 weeks	Random
Owen et al. 1998 [60]	Australia	Mixed	Age, Gender, Marital Status, Diagnosis, Medication, Type of admission, Previous admissions	20 + /1	20/22	VPS	VS	7 months	Random
Rutter et al. 2004 [61]	UK	Forensic	Age, Gender, Admission type, Diagnosis, Ethnicity	25 + /<25	17/217	Unclear	SIR	192 months	Random

Data analysis

The combined risk ratio (RR) and standardized mean difference (SMD) effects and their 95% confidence intervals were calculated using random-effects models which facilitated external generalizability and protected against sample heterogeneity [34, 35]. The STATA v.11 (StataCorp, College Station, TX, USA) function ‘metan’ was used to run each meta-analysis. For binary data (i.e. gender, ethnicity, marital status, diagnosis, type of admission and past history of conflict), the common effect measure was the ratio of proportions of aggressive (or repeatedly aggressive) and non-aggressive (or single aggressor) patients (RR) related to each patient factor. For continuous data (i.e. age, years of education and number of previous admissions), the common effect was the standardized differences in means between the aggressive and non-aggressive group's patient factors. Subgroup analyses of psychiatric setting (acute, picu, forensic) were conducted using the ‘by()’ option. Heterogeneity was assessed using a Q-value and I² for each factor [36].

Results

Aggressive vs. non-aggressive comparison studies

Description of included studies

The 34 studies differed in the settings in which they were conducted and in their sampling methods (see Table 2). The majority of studies collected demographic data from an in-patient acute (n = 12) or forensic setting (n = 11). Other settings included mixed (n = 5), psychiatric intensive care units (PICU) (n 3), research, chronic and a veterans in-patient setting (all n = 1). The majority of studies were from the USA (n = 11) or the UK (n = 11). Other countries included were Italy (n = 4), the Netherlands, Australia, Norway (all n = 2), Greece and Germany (both n = 1).

The majority of aggressive and non-aggressive participants in the studies were randomly sampled (n = 23) based on the systematically collected data from standard incident forms (n = 11), patient notes (n = 10), the Staff Observation Aggression Scale (soas; n = 5) and various other scales or interviews (see Table 2). These various methods used different definitions to categorize aggressive patients. In the majority of studies, an aggressive patient was defined as being reported by staff as having an incident of just physical aggression against others (P; n = 14). The other studies also included incidents of verbal aggression (V), physical aggression against objects (O) and aggression against self (S) as well as physical aggression against others.

Meta-analysis of possible factors associated with in-patient aggression

Fourteen factors were reported in two or more studies (see Table 4). The results of the meta-analysis for each factor are reported below.

Table 4. Meta-analysis of factors associated with in-patient aggression

	Total no. aggressive subjects/Total no. non-aggressive subjects [% of aggressive subjects to total subjects (agg + non-agg)]	No. aggressive subjects with factor/Total no. of aggressive subjects (%)	Number of studies	I-square	SMD	Relative ratio	Lower limit	Upper limit	Z	P
^a Total aggressive subjects only.
Age	1186/8398 [12]	1186a	18	51.10	-0.32	N/A	-0.39	0.25	9.30	<0.001
Male sex	2997/13312[18]	1666/2997 (56 male)	21	48.00	N/A	1.10	1.03	1.17	2.88	<0.01
Majority ethnicity	523/1008 [34]	300/523 (57 majority)	7	56.40	N/A	0.92	0.83	1.00	1.82	>0.05
Married	1581/4989 [24]	255/1581 (16 married)	6	2.30	N/A	0.72	0.63	0.83	4.70	<0.001
Years of Education	269/1686 [14]	269a	5	48.90	-0.13	N/A	-0.29	0.03	1.58	>0.05
Affective	1399/11575 [11]	201/1399 (14 affective)	11	44.10	N/A	0.94	0.82	1.08	0.87	>0.05
Schizophrenia	1470/11703 [11]	809/1470 (55 schizophrenic)	13	92.50	N/A	1.16	1.10	1.22	5.52	<0.001
Involuntary admission	1539/4981 [24]	753/1539 (49 involuntary admission)	8	98.50	N/A	2.17	2.01	2.34	20.37	<0.001
Number of Previous Admissions	216/2084 [9]	216a	2	98.50	1.2	N/A	1.05	1.35	15.60	<0.001
History of Violence	332/681 [33]	197/332 (59 history of violence)	5	75.80	N/A	2.27	1.90	2.69	9.24	<0.001
History of Self-destructive Behaviour	266/301 [50]	127/266 (48 history of self-destructive behaviour)	3	95.30	N/A	1.24	1.03	1.50	2.26	<0.05
History of Substance abuse	353/760 [32]	180/353 (51 history of substance abuse)	6	76.40	N/A	1.15	1.00	1.31	2.02	<0.05
History of illicit substance abuse	90/206 [30]	38/90 (42 of illicit drug abuse)	3	7.80	N/A	2.09	1.46	3.00	4.03	<0.01
History of Violent Convictions	185/187 [50]	82/185 (44 of violent convictions)	4	87.90	N/A	0.80	0.65	0.98	2.18	<0.05

Age

From the 34 comparison studies 26 studies included information comparing the age of aggressive and non-aggressive patients. Of these half reported no significant difference in age [23, 28, 33, 37-46], whereas the other half found that aggressive patients were significantly younger than non-aggressive patients [3-5, 24, 29, 47-54]. Eight studies were excluded from the meta-analysis as the ages were recorded as categorical data [23, 45, 51, 54], means were not reported [37, 46] or because standard deviations were not reported [33, 38].

The findings show that aggressive patients were significantly younger by −0.32 years, however, this result was statistically heterogeneous. The meta-analysis was also run by setting. Aggressive patients remained significantly younger than non-aggressive patients on the six studies from acute wards (SMD = −0.24 years, 95% CI = −0.35 to −0.13, z = 4.13, P < 0.001) and on the eight studies from forensic wards (SMD = −0.33 years, 95% CI = −0.46 to −0.21, z = 5.17, P < 0.001). However, the result within acute wards was statistically heterogeneous (Q = 13.87, P < 0.05, I² = 60.9%).

Gender

Twenty-two studies included information comparing the gender of aggressive and non-aggressive patients. Twenty-one of these reported no significant difference in the gender of aggressive to non-aggressive patients [3-5, 24, 27-29, 33, 37, 39, 40, 42-45, 51-56]. Blomhoff et al. [37] was excluded from the meta-analysis as data were missing.

The findings show that there is a significant effect. Males have a higher probability of being in the aggressive group compared with the non-aggressive group. The amount of variability owing to heterogeneity was not significant. The meta-analysis was also run by setting. Within acute wards male patients were more likely to be in the aggressive group than the non-aggressive group (combined RR = 1.14, 95% CI = 1.03–1.27, z = 2.55, P < 0.01, test for heterogeneity: I² = 48.00%, Q = 38.48, P > 0.05). However, the opposite was found within forensic wards where male patients were more like to be in the non-aggressive group than the aggressive group (combined RR = 0.80, 95% CI = 0.66–0.95, z = 254, P < 0.01, test for heterogeneity: I² = 0.0%, Q = 0.78, P > 0.10).

Ethnicity

Eleven studies included information comparing the ethnicity of aggressive and non-aggressive patients. Nine of these reported no significant difference between aggressive and non-aggressive patients [23, 29, 40, 42, 47, 50, 51, 54, 56]. One study [38] found a significant association between being non-white and committing an assault, whereas another study [39] found that Caucasian patients were more likely to be aggressive than non-Caucasian patients. Two studies were excluded from the meta-analysis as data were missing [47, 54] and two studies [29, 56] were also excluded as both studies compared very specific ethnicities with all others (German nationality in the former and Afro-Caribbean in the latter).

The findings show no effect of ethnicity. A patient of an ethnic majority (in this case Caucasian) is no more likely to be in the non-aggressive group than the aggressive group. However, this result was statistically heterogeneous. The analysis was also re-run by setting. No effect was found within acute or forensic patients.

Marital status

Seven studies included information comparing the marital status of aggressive and non-aggressive patients. Two of these reported no significant difference between aggressive and non-aggressive patients [38, 40] and one [39] that did not report whether the differences between groups were significant. Four studies [4, 5, 23, 29] found that aggressive patients were significantly overrepresented as single compared with non-aggressive patients. One study [38] was excluded from the meta-analysis as data were missing.

The findings show a significant effect and the results were not statistically heterogeneous. Married patients are more likely to be in the non-aggressive group than the aggressive group.

Years of education

Seven studies included information comparing the number of years in education for aggressive and non-aggressive patients. Six of these reported no significant difference between aggressive and non-aggressive patients [5, 24, 38, 41, 46, 50]. One study [49] found that aggressive patients had significantly fewer years of education than non-aggressive patients. Two studies were excluded from the meta-analysis as data were missing [24, 46].

The findings show that there was no significant difference in the mean number of years in education between aggressive patients and non-aggressive patients.

Diagnosis

Nineteen studies included information comparing the diagnoses of aggressive and non-aggressive patients. Nine of these reported no significant difference between aggressive and non-aggressive patients [3, 24, 39, 40, 44, 45, 47, 52, 56]. Seven studies found that schizophrenia was more prevalent among the aggressive group [4, 5, 27, 29, 43, 49, 54]. Diagnoses that were found to be less prevalent in the aggressive group were bipolar and adjustment disorder [27], depression [51] and substance abuse and affective disorders [29].

For the meta-analysis, studies that included data about patient's diagnoses were collated into three categories: schizophrenic (including schizoaffective, etc.), affective (depression, mania, etc.) and other (personality disorder, organic brain syndrome, etc.). Analyses were then made comparing the ratio of affective diagnoses compared with all other diagnoses in the aggressive and non-aggressive group as well as comparing the ratio of schizophrenic diagnoses with all other diagnoses in both groups. Five studies were excluded from the analyses because of missing data [3, 24, 47], inaccurate data [5] or because comorbid diagnoses were included [53].

Affective vs. all other diagnoses

The findings showed no significant effect and results were not significantly heterogeneous. There is no difference in the probability that patients with an affective disorder will be aggressive or non-aggressive. The analysis was also re-run by setting. There was also no significant effect within forensic wards or acute wards found.

Schizophrenia vs. all other diagnoses

The findings showed a significant effect. Patients with a diagnosis of schizophrenia are more likely to be in the aggressive group than the non-aggressive group. However, this result was statistically heterogeneous and remained so when some outlier studies were removed. However, when the analysis was re-run by setting the effect remained significant within acute wards (combined RR = 1.32, 95% CI = 1.21–1.44, z = 6.16, P < 0.001), but the test for heterogeneity was no longer significant (Q = 9.30, P > 0.1, I² = 35.5%).

Type of admission

Ten studies included information comparing the type of admission (involuntary vs. voluntary) between aggressive and non-aggressive patients. Seven of these reported that there were significantly higher numbers of aggressive patients who were admitted involuntarily compared with non-aggressive patients [3, 5, 29, 43, 46, 51, 53]. Two studies found no significant differences between the admission type of the two groups [44, 56]. Two studies were excluded from the meta-analysis as data were missing [3, 56].

The findings showed a significant effect. Across studies involuntary patients were more likely to be in the aggressive group. However, this result was statistically heterogeneous.

Number of previous admissions

Nine studies included information comparing the number of previous admissions of aggressive and non-aggressive patients. Five of these reported that aggressive patients had significantly more previous admissions than non-aggressive patients [3, 27, 29, 43, 44]. One study [49] reported the opposite finding and three studies found no significant differences between the number of previous admissions between the two groups [47, 52, 56]. Studies were excluded if they had missing data [3, 27, 43, 47, 52] or if they reported categorical data [44, 56].

The findings showed aggressive patient had significantly more previous admissions than non-aggressive patients. However, it may not be appropriate to perform a meta-analysis on two studies [29, 49] and as the heterogeneity was also very high this finding should be interpreted with caution.

Patient past history of conflict

Previous history of violence

Six studies included information about patient's previous history of violence for both the aggressive and non-aggressive groups. All of these reported that aggressive patients were significantly more likely to have a history of previous violence [37, 42, 45, 52, 53] or a significant association between violent behaviour in the community 2 weeks before admission and aggressive patient behaviour on the ward [23].

The findings showed a significant effect. Across studies patients with a history of violence were more likely to be in the aggressive group. However, this result was statistically heterogeneous. The meta-analysis was re-run on studies set in acute in-patient care. The effect remained significant [combined RR = 2.37; 95% confidence intervals (CI), 1.97– 2.86, z = 9.04, P < 0.001] and heterogeneity was no longer significant (I² = 80.3%, Q = 15.23, P < 0.01).

History of self-destructive behaviour (suicidal behaviour, suicidal risk, self-harm, suicide attempts)

Three studies included information about patient's previous history of self-destructive behaviour for both the aggressive and non-aggressive groups [23, 47, 53].

The findings showed a significant effect. Across studies patients with a history of self-destructive behaviour were more likely to be in the aggressive group. However, this result was statistically heterogeneous.

History of substance use

Six studies included information about patient's previous history of substance abuse for both the aggressive and non-aggressive groups [23, 28, 37, 42, 53, 56]. The findings showed a significant effect. Patients with a history of substance abuse were more likely to be in the aggressive group. However, this result was statistically heterogeneous. Findings suggest that when alcohol use is included there is little difference between the two groups, whereas the use of illicit drugs seems to be more likely within the aggressive group. The analysis was therefore re-run on just the studies that looked at a patient's history of previous illicit drug use. The results were no longer statistically heterogeneous (Q = 2.17, P > 0.1, I² = 7.8%) and the significant effect remained (combined RR = 2.09, 95% CI = 1.46–3.00, z = 4.03, P < 0.01).

History of previous arrests or convictions for violent crime

Eight studies [28, 39-41, 49, 50, 56, 57] included information about patient's criminal records in each group. Four studies were excluded from the analysis because of missing data [28, 41, 50, 57]. The findings showed a significant effect. Patients with a history of violent convictions were less likely to be in the aggressive group. However, this result was statistically heterogeneous.

Once only vs. repeated aggression comparison studies

Description of included studies

The 11 studies differed in the settings in which they were conducted and in their sampling methods (see Table 3 for further details).

Meta-analysis of possible factors associated with repetitive in-patient aggression

Eight factors were reported in two or more studies (see Table 5). The results of the meta-analysis for each factor are reported below.

Table 5. Meta-analysis of factors associated with repetitive in-patient aggression

	Total no. repeatedly aggressive/total no. single aggressors [% of repeatedly aggressive to total subjects (repeat + single)]	Repeatedly aggressive subjects with factor/Total repeatedly aggressive subjects (%)	Number of studies	I-square	SMD	Relative ratio	Lower limit	Upper limit	Z	P
^a Total repeatedly aggressive subjects only.
Age	252/935 [21]	252a	6	23.5	-0.08	N/A	-0.23	0.07	1.06	>0.1
Male sex	374/1320 [22]	211/374 (56 male)	9	56.7	N/A	0.83	0.75	0.93	3.3	<0.01
Majority ethnicity	144/159 [48]	31/144 (22 majority)	4	48.4	N/A	0.79	0.52	1.18	1.15	>0.1
Affective	226/817 [22]	23/226 (10 affective)	4	65.5	N/A	0.7	0.43	1.13	1.46	>0.1
Schizophrenia	260/958 [21]	184/260 (71 schizophrenic)	5	26.3	N/A	1.04	0.95	1.14	0.81	>0.1
History of Violence	51/652 [7]	50/51 (98 history of violence)	2	99.1	N/A	1.58	1.45	1.73	10.26	<0.01
History of Substance abuse	113/589 [16]	65/113 (58 history of substance abuse)	3	11.3	N/A	1.28	1.04	1.59	2.3	<0.05
History of Violent Convictions	83/283 [23]	28/83 (34 history of violent convictions)	3	79.6	N/A	0.78	0.5	1.21	1.12	>0.1

Age

From the 11 comparison studies all included information comparing the age of once and repeated aggressors [25-27, 30, 31, 33, 58-62]. Five studies were excluded from the meta-analysis as the standard deviations or mean value was not reported [25, 27, 33, 61, 62]. The findings showed no significant difference between the two groups mean age. The result was not statistically heterogeneous.

Gender

Ten of the studies included information comparing the gender of repeatedly aggressive and once only aggressive patients [25, 26, 30, 31, 33, 58-62]. One study [60] was excluded from the meta-analysis as data were missing. The findings showed that males were significantly less likely to be in the repeated aggression group than the once only group. However, the test for heterogeneity was significant.

Ethnicity

Four studies included information comparing the ethnicity of repeatedly aggressive and once only aggressive patients [25, 26, 30, 31]. The findings show no significant effect of ethnicity and the result was not statistically heterogeneous.

Diagnosis

Eight studies [25, 26, 30, 31, 58-60, 62] included information comparing the diagnoses of repeatedly aggressive and once only patients. For the meta-analysis, studies that included data about patient's diagnoses were collated into three categories: schizophrenic (including schizoaffective, etc.), affective (depression, mania, etc.) and other (personality disorder, organic brain syndrome, etc.). Analyses were made comparing the ratio of affective diagnoses compared with all other diagnoses in the repeatedly aggressive and once only aggressive group as well as comparing the ratio of schizophrenic diagnoses with all other diagnoses in both groups. Three studies were excluded from the analyses because of missing data [60], or because all patients were either diagnosed with schizophrenia or schizoaffective disorder [30, 31].

Schizophrenia vs. all other diagnoses

The findings showed no significant effect.

Affective vs. all other diagnoses

The findings showed no significant effect.

Patient past history of conflict

Previous history of violence

Three studies included information about patient's previous history of violence for both the repeatedly aggressive and once only aggressive groups [58-60]. The findings showed a significant effect of a previous history of violence. Patients with a history of violence were more likely to be in the repeatedly aggressive than the once only aggressive group.

Previous history of a violent conviction

Three studies [25, 30, 61] included information about patient's previous history of violent convictions for both the repeatedly aggressive and once only aggressive groups. The findings showed no significant effect of a history of violent convictions.

Previous history of substance use

Four studies [25, 30, 31, 58] included information about patient's previous history of substance use for both the repeatedly aggressive and once only aggressive groups. One study [30] was excluded from the analysis because of missing data. The findings showed a significant effect of a history of substance abuse. Patients with a history of substance use were more likely to be in the repeatedly aggressive group than the once only group. The results were not statistically heterogeneous.

Discussion

In total, a relatively small number of comparison studies were found relative to the number of publications on in-patient aggression (17.5%). The majority of publications instead tended to focus on the rate of aggressive incidents within wards, or the antecedents and consequences of aggression. The small number of comparison studies suggests that future research would benefit from focusing on this type of research design.

All but seven of the studies included had fewer than 100 patients in both the aggressive and non-aggressive groups and the demographics were often not the focus of the research. More prospective comparison studies are needed which should be designed on the basis of power analyses to calculate the minimum sample size required to detect an effect or a reliable difference between aggressive and non-aggressive patients. Despite these limitations, a number of demographic and historical characteristics appear to be associated with an increased likelihood of in-patient aggression. These included being younger, male, involuntarily admitted, not being married, a diagnosis of schizophrenia, a greater number of previous admissions, a history of violence, a history of self-destructive behaviour and a history of substance abuse. Interestingly, a history of previous violent convictions was associated with a decreased likelihood of in-patient aggression. Separate analyses of the comparison studies that looked at repeatedly aggressive patients and less aggressive patients found very few factors that distinguished the two groups. Repeatedly aggressive patients are responsible for multiple episodes of aggression so it is important to understand the triggers for this subset of patients or the variables that may be associated with this repetitive behaviour. The only factors associated with an increased likelihood of repeated aggression were a history of substance abuse or a history of violence. Interestingly, in contrast to the aggression vs. non-aggressive studies, being male was associated with a decreased likelihood.

We can be fairly confident that these results represent true effects as they have been estimated by combining the results of several studies. The effects are also more powerful than those reported in individual studies whose findings are sometimes mixed. They also support another recent review on factors associated with in-patient aggression [12]. However, care needs to be taken when interpreting the associations between demographic and historical factors and aggression as they were fairly small and for some factors (involuntary admission, number of previous admissions, history of violence, history of self-destructive behaviour and history of previous convictions) the heterogeneity was high (I² was greater than 50%), suggesting that combining the studies for these factors in particular may lead to less generalizable estimates. The high levels of heterogeneity found may be because psychiatric services can vary a great deal in terms of setting, routines, ward rules and atmosphere. This was shown as when subgroup analyses of psychiatric setting were run, the heterogeneity was significantly lowered for some of the factors. The studies also varied in the measurements that were used to measure aggression incidents and the definition used for violence. The majority of studies used patient notes or standard incident reports, but some used scales specifically designed to measure aggression. This could be an additional variable attributing to the high levels of heterogeneity observed.

A further limitation may have been the selection procedure (choosing to analyse comparison studies only) of the studies included in the review. This meant that important factors such as length of stay and other studies analysing predictions and frequencies were not taken into account in the analyses. It is probable that levels of aggression are influenced more by some of these factors than patient demography and history. For example, these may include a patient's current presentation, e.g. whether or not they appear under the influence of alcohol or drugs; the symptoms they are currently displaying such as: fear, agitation, anger, confusion, excitement, suspiciousness or irritability; whether patients are having delusions and/or hallucinations and their current attitude towards treatment and management. Cornaggia et al. [12] found that length of stay and the presence of impulsivity/hostility were associated with aggression. Contextual factors may also have a substantial impact upon the levels of aggression, for example whether the patient has a weapon available; the ward environment (i.e. levels of surveillance/visibility, ward door-locking policies, ward rules); the relationship and proximity between a victim and aggressor and the extent of social support both within the ward with staff and patients and outside the ward with family and friends [see [63] for a review of the antecedents of aggression on in-patient wards].

Although we have identified a number of static factors that are associated with in-patient violence, the utility of these for an actuarial-based risk assessment tool is questionable. The generally small effects found coupled with the heterogeneity between studies, suggest that any such instrument would be too inaccurate to be useful. Approaches based on short-term prediction may prove to be more practically useful [64, 65]. There is a significant debate in the literature about the link between schizophrenia, substance misuse and violence in the community, and this review is consistent with research which has found associations between these variables [66]. The mechanism of the link between substance use, schizophrenia and aggression is uncertain as they share a number of confounding risk factors such as male gender, younger age, increased suicide rate, non-adherence to treatment, higher levels of social deprivation. Common factor models suggest that the links are the result of shared risk factors such as genetics, antisocial personality disorder, socioeconomic status an impaired cognitive functioning. Secondary substance-use models posit that there are certain reasons (self-medication, alleviation of dysphoria) why having a diagnosis of schizophrenia increases the risk of substance misuse. Secondary psychiatric disorder models put forward the opposite argument that substance misuse leads to a diagnosis of schizophrenia in individuals who would not have developed the disorder had they not taken illicit substances. In addition, there are bidirectional models that propose that either variable (schizophrenia, substance misuse) can increase the likelihood of the other co-occurring [for a review of the evidence for and against each model, see Mueser et al. [67]].

Future research is needed that follows patients longitudinally to shed more light on the direction and relationships between aggression and the significant demographic and historical factors identified here (age, gender, history of previous violence, history of substance misuse, type of admission, diagnosis and marital status). This in turn may provide useful information about which of these factors reliably predicts an aggressive or repeatedly aggressive patient. It may be possible to then start thinking about management strategies for these patients. What is it about a young, single, male admitted involuntarily with a diagnosis of schizophrenia and a history of previous violence, self-destructive behaviour and substance misuse that makes an aggressive incident more likely? Perhaps the way in which staff makes requests of a patient could be a potential antecedent of an aggressive incident [63]. Requests may be perceived as demands and feel a lack of control over their environment or their actions. This may be attenuated by a lack of social support, and symptomology such as delusions and irritability. It is important to note that these combined characteristics have also been found to be relevant in the prediction of out-patient violence suggesting that there may be common mechanisms involved for a general increased violence risk. Social interactions in other contexts and the aggressor's perspective of these exchanges could be an important trigger of aggressive episodes within the community. Future research should focus more on patient's and out-patient's perception of aggressive incidents rather than just staff perceptions, and how providing patients and out-patients with more choices might impact on these events.

Acknowledgements

This paper presents independent research commissioned by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research scheme (RP-PG-0707-10081). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.