Sample

The sample comprised 168 consecutive patients (aged 16-64 years) with a psychotic illness making first contact with psychiatric services in Nottingham (UK) between 1 June 1992 and 31 May 1994 (from a population of 397 148). All subjects met the criteria in the Screening Schedule for Psychosis (Reference Jablensky, Sartorius and ErnbergJablensky et al, 1992). Descriptions of case ascertainment and demographic characteristics of this cohort (including determination of ethnic origin) have been reported previously (Reference Brewin, Cantwell and DalkinBrewin et al, 1997; Reference Harrison, Glazebrook and BrewinHarrison et al, 1997).

Follow-up procedure

Follow-up assessments took place between June 1995 and May 1997, as close as possible to 3 years after first contact. One hundred and forty-three direct interviews (86%) were completed, with the remainder relying on other clinical material. We obtained informed consent to interview subjects, to examine their records and to interview their informal carers.

Assessment variables

Data analysis

Demographic and clinical characteristics were reported using proportions, means or medians according to their measurement level (discrete or continuous) and distribution (normal or skewed).

Stepwise logistic regression analyses were used to identify clinical and demographic characteristics associated with aggression. Separate analyses were conducted for all acts of aggression and for serious aggression. For each model, odds ratios with 95% confidence limits were reported for variables meeting the entry criteria. The 11 independent variables (chosen for their previously reported or hypothesised association with aggression and mental disorder) were age, gender, employment, history of violent convictions, MINI score, diagnosis at first contact, comorbid substance misuse diagnosis, total symptom score, psychotic symptom score, behaviour symptom score and TCO score.

Sample

Two of the 168 patients who entered the study were excluded: one had an organic psychosis and one withdrew consent for the study. Detailed analysis is therefore presented for 166 subjects. Socio-demographic characteristics and ICD diagnoses have been described for onset (Reference Brewin, Cantwell and DalkinBrewin et al, 1997; Reference Harrison, Glazebrook and BrewinHarrison et al, 1997) and follow-up (Reference Amin, Singh and BrewinAmin et al, 1999).

Details of aggressive acts

Details of timing and severity of all aggressive acts are presented in Fig. 2 and Table 1. A description of victims of all assaults committed ‘in episode’ is shown in Fig. 3.

Fig. 2

Subjects exhibiting aggression per time period (number and % within groups).

Fig. 3

Victims of all assaults by subjects in episode.

Table 1

Number of subjects exhibiting aggressive behaviour by type, timing and severity (n=166)

Type of aggressive act1	Period A (pre-illness)		Period B (onset to contact)		Period C (post-contact)		Total no. of subjects
	No. of subjects	No. of episodes	No. of subjects	No. of episodes	No. of subjects	No. of episodes
Total	16	20	28	42	35	104	61
Serious	10	14	8	15	8	10	21
Less serious	6	6	20	27	27	94	40

Fifty-five (33.1%) subjects were responsible for 166 aggressive acts towards others during periods B and C, some subjects being responsible for several incidents within one time period (e.g. two subjects were responsible for 44 of the aggressive incidents during period C). Of these, 16 subjects (9.6% of the total sample or 29.1% of the aggressive group) demonstrated at least one act of serious aggression in periods B and/or C and 4 (2.4%) exhibited aggressive behaviour in all time periods. All but one of the 35 subjects who were aggressive during period C were judged by investigators to be ill at the time of such incidents.

For 20 subjects (71.4% of the subjects exhibiting aggression in period B) aggression at presentation led directly to psychiatric contact, and in 18 cases the police were involved (10.8% of the total sample).

Demographic characteristics of subjects exhibiting any aggression towards others while in episode compared with non-aggressive subjects are shown in Table 2 and their clinical attributes in Table 3. Other features characteristic of subjects who exhibited aggression after service contact were high rates of absconding from in-patient care (34.3%, OR=16.6, 95% CI 4.9-55.9) and increased compulsory admissions over the following 3 years (mean number of sections=2.2 and s.d.=1.5 v. 0.11 and 0.32 for non-aggressive subjects; t=-2.4, P=0.017).

Table 2

Demographic and diagnostic characteristics of subjects aggressive during period B (pre-contact) and period C (post-contact) compared with non-aggressive subject group

Variable	Non-aggressive group (n=111)	Aggression during illness group (some subjects are included during both periods B and C)
	% of 166 subjects	Pre-contact (period B) (n=28)		Post-contact (period C) (n=35)		Total n=55 (aggression v. non-aggression) OR (95% CI)
		%	OR (95% CI)	%	(95% CI)
Gender
Male	53.2	78.6	3.2* (1.2-8.6)	68.6	1.9 (0.9-4.3)	2.1* (1.1-4.3)
Ethnicity
White	74.8	67.9	0.7 (0.3-1.8)	68.6	0.7 (0.3-1.7)	0.8 (0.4-1.5)
African—Caribbean	17.1	21.4	1.3 (0.5-3.7)	28.6	1.9 (0.8-4.7)	1.6 (0.8-3.6)
Other	7.2	10.7	1.5 (0.4-6.2)	2.9	0.4 (0.5-3.1)	0.7 (0.2-2.9)
Unemployed	46.4	75.0	3.5** (1.4-8.8)	80.0	4.6** (1.9-11.5)	3.7** (1.8-7.7)
Single	56.8	78.6	2.8* (1.1-7.4)	74.3	2.2 (0.95-5.1)	2.2* (1.1-4.6)
Living alone	28.9	21.4	0.6 (0.3-1.8)	31.4	1.1 (0.5-2.6)	0.8 (0.4-1.6)
Age (mean, s.d.)1	32.9 (12.3)		26.5 (9.1)**		29.0 (5.9)**	26.6 (7.9)**
Diagnosis
Schizophrenia	34.2	35.7	1.2 (0.4-2.5)	37.1	1.1 (0.5-2.5)	1.2 (0.6-2.3)
Other psychosis	65.8	64.3	0.9 (0.4-2.2)	62.9	0.8 (0.4-1.9)	0.8 (0.4-1.6)
Any compulsory admissions	32.4	67.9	4.4** (1.8-10.7)	85.7	12.5** (4.5-35.0)	6.7** (3.2-14.1)
Any convictions	9.0	32.1	4.8** (1.7-13.3)	30.3	4.3** (1.6-11.8)	4.4** (1.8-10.5)
Previous violent convictions	4.5	25.0	7.1** (2.0-24.4)	22.9	6.8** (2.0-23.5)	5.6** (1.8-16.9)
Any aggression	5.4	25.0	5.8** (1.8-19.1)	26.5	6.3** (2.1-19.3)	5.0** (1.8-14.2)
Prior imprisonment	5.4	21.4	4.8* (1.4-16.2)	22.9	5.2** (1.7-16.2)	4.4** (1.5-12.6)

Table 3

Clinical variables for subjects aggressive during period B (pre-contact) and period C (post-contact) individually compared with non-aggressive subject group

Variable	Non-aggressive group (n=111) %	Aggression during illness group (some subjects were aggressive during both periods B and C)
		Pre-contact (period B) (n=28)		Post-contact (period C) (n=35)		Total (n=55) (non-aggressive v. aggressive) OR (95% CI)
		%	OR (95% CI)	%	OR (95% CI)
Illness variables
Mode of illness onset
Insidious (> 6 months)	27.0	42.9	2.0 (0.9-4.8)	22.9	0.8 (0.3-2.0)	1.2 (0.6-2.5)
Single illness episode	36.8	28.6	0.7 (0.3-1.7)	17.1	0.4* (0.1-0.9)	0.6 (0.3-1.2)
Duration of untreated illness (median, days)1	75	136		68		85
Substance misuse variables
ICD-10 drug-related psychosis at onset (Flx), n=13	9.0	7.4	0.8 (0.2-4.0)	6.1	0.7 (0.1-3.2)	0.8 (0.2-2.9)
Flx additional ICD-10 diagnosis (‘dual diagnosis’), n=15	4.0	11.1	3.0 (0.6-14.3)	27.3	9.0** (2.6-31.7)	6.4** (1.9-21.4)
Symptom variables (from the IGC)
High IG total symptom score	34.2	57.1	2.6* (1.1-6.0)	45.7	1.6 (0.8-3.5)	1.9 (0.96-3.6)
High IG behaviour score	54.1	85.7	5.1** (1.7-15.7)	82.9	4.1** (1.6-10.7)	3.6** (1.8-8.3)
High IG psychotic score	54.1	32.1	0.4 (0.2-1.0)	48.6	0.8 (0.4-1.7)	0.7 (0.3-1.3)
High IG TCO score	63.1	32.1	0.3** (0.1-0.7)	54.3	0.7 (0.3-1.5)	0.5* (0.3-0.9)

For the 29.1% of aggressive subjects who exhibited serious aggression while ill, a comparison is shown in Table 4 with subjects exhibiting less serious aggression (only positive statistical or relevant associations are shown).

Table 4

Comparison between serious and less serious aggression groups while ill

Variable	Serious aggression to others (n=16)	Less serious aggression to others (n=39)	Odds ratio
Demographic and aggression variables
Age (mean, s.d.)1	28.9(11.1)	28.8(6.0)
Any pre-illness aggression, % (n=10)	31.3	12.8	3.1 (0.8-12.7)
Any serious pre-illness aggression, % (n=7)	25.0	7.7	4.0 (0.8-20.5)
Any pre-illness convictions, % (n=16/53)	60.0	18.4	6.6** (1.8-24.8)
Any pre-illness violent convictions, % (n=11/53)	46.7	10.5	7.4** (1.7-31.7)
Clinical variables
Mode of illness onset: insidious	56.3	20.5	5.0* (1.4-17.5)
(> 6 months), % (n=17)
Duration of untreated illness, days
Mean (s.d.)2	613(954)	227(620)
Median	226	76
High IG symptom score, % (n=27)	56.3	46.2	1.5 (0.5-4.8)
High IG behaviour score, % (n=45)	68.8	87.2	0.3 (0.1-1.3)
High IG psychotic score, % (n=24)	25.0	51.3	0.3 (0.1-1.2)
High IG TCO score, % (n=25)	25.0	53.8	0.3 (0.1-1.0)

Other than the characteristics shown in Table 4, demographic or clinical features failed to discriminate between lesser and serious aggression. Although more subjects committing serious aggression while ill but before contact had a slow mode of illness onset compared with those committing less serious aggression during the same period (6/8 v. 6/20, OR=7.0, 95% CI 1.1-45.2, P<0.05), their DUI was not significantly longer (median 554 days v. 128 days, Mann-Whitney U=48.0, P=0.10).

Multivariate analyses

Two multiple logistic regression models were used to assess the risk of subjects exhibiting either any aggression or serious aggression while in episode. Of the 11 independent variables, age, gender, diagnosis at first contact, history of violent convictions, MINI score, total symptom severity score and psychotic symptom score did not meet the entry criteria. A high behaviour symptom group score (OR=6.9, 95% CI 2.7-17.8), comorbid substance misuse diagnosis (OR=3.1, 95% CI 1.1-8.8) and being unemployed (OR=3.6, 95% CI 1.6-8.0) all had an independent effect on the risk of exhibiting any aggression. A high TCO score had an independent effect on reduced risk of any aggression (OR=2.8, 95% CI 1.3-6.3).

For the model of serious aggression while in episode, a history of convictions for violence (OR=12.6, 95% CI 3.5-45.1) was the only independent variable to meet the entry criteria other than high TCO score, which was negatively associated with risk (OR=4.1, 95% CI 1.2-14.7).

Methodological strengths and limitations

Criticisms of previous studies examining the relationship between violence and mental disorder have been described in the MacArthur Violence Risk Assessment (MVRA) Study (Reference Steadman, Monahan, Appelbaum, Monahan and SteadmanSteadman et al, 1994).

Briefly, there are three types of problem: use of ‘static’ predictor variables (usually socio-demographic variables) rather than dynamic (symptom profile or timing of violent behaviour) variables; narrowly defined patient samples — in-patients (Reference Raja, Azzoni and LubichRaja et al, 1997), ‘high risk’ discharges from secure psychiatric or prison facilities (Reference Rice and HarrisRice & Harris, 1995) or single diagnostic groups (Reference Humphreys, Johnstone and MacMillanHumphreys et al, 1992) and data-sets restricted to individuals arrested or re-admitted to hospital (Reference Wessely, Castle and DouglasWessely et al, 1994; Reference Tiihonen, Isohanni and RasanenTiihonen et al, 1997). This study therefore sought to use a representative first-episode community-based sample, a broad diagnostic range of psychotic disorders and multiple data sources.

Our findings have several limitations. First, and of greatest hindrance to conclusions about prevalence, is the absence of a non-psychosis comparison group representative of the population from which our subjects were taken.

Second, the overall study was not designed primarily to obtain data on aggression or offending. Because difficulties with confidentiality and consent were encountered in accessing statutory criminological data, an incomplete criminological data-set probably reduced the prevalence of reported offending. Although the use of multiple data sources (e.g. informant interviews or psychiatric records) has been shown to increase significantly the reported prevalence of violence (Reference Steadman, Mulvey and MonahanSteadman et al, 1998), our findings still have limitations. For example, even in prospective studies such as the MVRA Study, subject attrition can be high, leading to restrictions in collection and analysis of violence-related data. Sources such as psychiatric case notes have been shown to underestimate convictions for violence among mentally disordered offenders by almost 50% in a UK sample of homicide perpetrators (Reference Shaw, Appleby and AmosShaw et al, 1999). Additionally, use of ‘untoward incident’ records may bias the reporting of aggressive incidents after service contact and in in-patient settings: either overreporting due to individuals' increased exposure to the scrutiny of health care staff or underreporting because aggression may be accepted by staff as symptomatic of mental disorder.

Third, modifications made to the IGC to allow comparison of symptoms across the whole cohort have yet to be validated as a means of assessing symptom clusters. However, the SCAN interview is widely used and good reliability between raters was demonstrated for this study (Reference Brewin, Cantwell and DalkinBrewin et al, 1997). We chose particular item groups and then adopted sub-scales based upon established empirical evidence for their association with violence. This approach has face validity although it could be criticised for focusing on particular target symptoms, thus biasing findings towards certain symptoms at the expense of other previously unreported but potentially important symptom areas.

Lastly, because the study was not conceived to ascertain clinical and contextual variables at the precise time of all aggressive acts, it was not always possible to elicit retrospectively (either from patients or informants) the precise motives after illness onset, although most attributed aggression after service contact to illness. Also, only those patients whose pre-contact aggression led to immediate admission received a SCAN interview temporally related to the incident. For the remaining subjects, we drew conclusions from SCAN interviews shortly after service contact about the relationship between psychopathology at service contact and at times of aggression. It could be argued also that the differentiation of illness episode into specific time periods is arbitrary, particularly when change in behaviour (rather than emergence of first psychotic symptom) was the criterion used. This could be relevant if different risk factors for aggression prevail at difference time periods of illness.

Type and frequency of aggression towards others

Approximately one-third of subjects exhibited aggressive behaviour during periods B and C and one-sixth after the onset of symptoms but prior to service contact. Comparisons with other UK samples are limited by diagnostic heterogeneity. However, the proportion of subjects with schizophrenia exhibiting aggression prior to contact (10/54 or 18.5%) is similar to another UK study (21%) (Reference Humphreys, Johnstone and MacMillanHumphreys et al, 1992) but less than that described for developed countries (31.5%) in the World Health Organization's study on the Determinants of Outcome of Severe Mental Disorders (DOSMD; Reference Volavka, Laska and BakerVolavka et al, 1997).

Aggressive acts were more common after service contact. Despite this, only four (2.4%) subjects demonstrated aggression in all the time periods studied (i.e. premorbidly and after illness). Serious aggression was more prevalent before onset of illness and diminished in frequency in proportion to less serious aggression after onset and contact with secondary services. However, the overall number of subjects committing lesser acts of aggression increased. Although this may relate in part to reporting bias after service contact, further analysis of the timing of violent events relative to symptoms and illness course is required.

Our findings regarding the victims of aggression by subjects with psychosis are in broad agreement with the MVRA Study (Reference Steadman, Mulvey and MonahanSteadman et al, 1998), where family and acquaintances form the majority of victims. Further study of the location, social context and social networks of subjects with psychosis in relation to aggression is needed.

Diagnosis, symptoms and violence

Excluding substance misuse (ICD—10 F1x) comorbidity, diagnostic group alone did not confer any special risk of aggression, regardless of aggression type or severity. This contrasts with other studies of serious offending, which report an increased risk for schizophrenia of both homicide (Reference Eronen, Tiihonen and HakolaEronen et al, 1996) and violent conviction (Reference Tiihonen, Isohanni and RasanenTiihonen et al, 1997).

In agreement with recent findings from the MVRA Study (Reference Appelbaum, Robbins and MonahamAppelbaum et al, 2000), we did not confirm previous reports linking specific psychotic symptoms and aggression. We found that although delusions, hallucinations and passivity phenomena were common, they did not specifically distinguish subjects who were aggressive, either in isolation or as the TCO syndrome. We also confirmed Appelbaum et al's (Reference Appelbaum, Robbins and Monaham2000) finding that higher TCO scores at baseline assessment were inversely related to the risk of further aggression, serious or otherwise. Such findings, relating to an inception rather than a discharge cohort, may add weight to Appelbaum et al's conclusions that “presumed risk of violence associated with delusions per se does not justify hospitalisation of a patient, in the absence of other indicators of violence risk or of other reasons for inpatient treatment”.

Symptoms that were associated with aggression and present at service contact were apparent in two areas. First, behavioural symptoms, such as overactivity and emotional turmoil, had large and independent effects on risk of aggression. This has some precedence in the symptoms of excitement found in acute-onset disorders during the DOSMD study (Reference Volavka, Laska and BakerVolavka et al, 1997). Second, symptoms related to aspects of negative symptoms and deficits in communication appeared to be characteristic of subjects, mainly younger men with insidious illness onset, who exhibited serious aggression towards others and specifically after illness onset and before service contact. However, higher global ratings of negative symptoms were not characteristic of these subjects.

In addition to insidious illness onset and past history of convictions for violence, subjects with psychosis who demonstrated serious aggression also had longer periods of untreated illness. Duration of untreated illness has attracted attention as an important factor associated with outcome across a number of domains (Reference McGorry, Krstev and HarriganMcGorry et al, 2000). Although the association with aggression could provide opportunities for reducing the DUI in order to diminish indirectly the risk of aggression, the relationship may be an artefact. Because the length of period B varied considerably (from a few days to over 10 years), it is possible that the connection between the DUI and aggression represents a longer period at risk.

Only a small number of subjects received any Axis II diagnosis. Our findings are likely to represent an underestimation of personality disorder, owing to the absence of interview- or informant-based personality data for all subjects.

Substance misuse and violence

Previous research has identified a relationship between substance misuse and violence, either alone or in combination with major mental disorder (Reference Swanson, Holzer and GanjuSwanson et al, 1990; Reference Tiihonen, Isohanni and RasanenTiihonen et al, 1997; Reference Steadman, Mulvey and MonahanSteadman et al, 1998). In a UK sample, Scott et al (Reference Scott, Johnson and Menezes1998) reported that individuals with comorbid psychosis and substance misuse are more likely to exhibit hostile behaviour and assaults than those with psychosis alone.

We confirmed that ICD—10 F1x diagnostic comorbidity, but not primary drug-related psychosis, at onset had an independent effect on the risk of subjects exhibiting aggression during a psychotic episode. Individuals with drug-related comorbidity were nine times more likely to exhibit aggression after service contact. Whether this represents reporting bias and how such individuals' substance misuse covaries with their psychotic disorder over time requires further study. Comorbid alcohol misuse was not a significant association for violence, as reported in the DOSMD study (Reference Volavka, Laska and BakerVolavka et al, 1997).

CLINICAL IMPLICATIONS

• ▪ Within a community-based first-episode psychosis cohort 33.1% of the subjects exhibited aggression, predominantly minor acts, towards family members.

• ▪ Comorbid substance misuse and overexcited behaviour (but not psychotic symptoms) were independently associated with subjects perpetrating aggressive acts.

• ▪ A history of convictions for violence was characteristic of the 9.6% of subjects committing acts of serious aggression.

LIMITATIONS

• ▪ The absence of a comparison group of individuals without psychosis with similar socio-demographic characteristics limits the empirical utility of our findings.

• ▪ Underreporting of aggressive or offending behaviour may have occurred, owing to incomplete statutory criminological data.

• ▪ The use of symptom subgroups has yet to be validated as a means of assessing symptom clusters in relation to aggressive acts.

Scores of Symptom Severity

The following 17 items (including all psychotic symptom items) are taken from the first 41 item groups of the Item Group Checklist (World Health Organization, 1994) as being the most relevant to violent behaviour (based upon previous reported research).