Abstract
Adolescent patients in psychiatric settings often perform self-destructive acts or become aggressive towards peers, staff or property [1, 2]. These events, which we will call critical incidents, threaten the safety of patients and staff, and create therapeutic and management challenges [2–4]. However, little is known about the frequency, precursors and correlates of critical incidents. These seem to vary widely in type and frequency according to factors including the setting, the age and the type of patients treated, making it difficult to generalize findings [1, 2, 4].
Aggressive acts in adolescent inpatient units are particularly common among males [1, 4]. Individuals involved in aggressive incidents often have a past history of aggression [4], are more likely to have a diagnosis of conduct disorder [5] and respond poorly to treatment [6]. On the contrary, patients who harm themselves are more likely to be female [7], to have a history of self-harm [5], to have experienced physical or sexual abuse and to have been diagnosed with borderline personality disorder [2, 8]. Critical incidents have also been associated with transition times in the routine of units, increasing in frequency during periods in which little supervision or structure is available [2, 9]. In addition, self-harm occurs more frequently during the evening, while there are few critical incidents between 23.00 hours and 07.00 hours [2]. Aggression often targets clinical staff. For example, Garrison et al. [2] reported that most assaults were directed towards staff rather than towards other patients.
This study examined retrospectively all critical incidents reported at an adolescent unit over a period of 30 months. The aims were to: (i) identify patterns in the timing and nature of the incidents; (ii) determine differences among patients who engage in critical incidents and those who do not; and (iii) ascertain whether there was an association between incidents and outcome of treatment. Hypotheses were that involvement in critical incidents would result in worse outcomes and that associations would be similar to those reported in the literature. Findings may be useful for a better understanding of the factors associated with critical incidents and to find ways to reduce their frequency and severity.
Method
The setting
The unit is a mental health facility that provides residential, day patient and outpatient treatment programmes for children and adolescents in Sydney, Australia. During the time of the study (January 1993 to June 1995), inpatients comprised residential and day patients. Residential patients attended the unit from Monday to Friday and resided with families on weekends. Day patients attended the unit from 08.45 hours to mid afternoon. All patients attended a school on the premises from 09.00 hours to 15.00 hours.
Critical incidents
Data on critical incidents were collected from hospital incident reports completed by nursing staff on a standard form. Incident reports were filed when acts of aggression or self-harm occurred, or when patients were absent without leave (AWOL). The types of incidents included in the study are listed in Table 1. Information collected for each event included the type of incident (aggression, self-harm, AWOL), severity, location, time of day and whether the incident was carried out individually or involved two or more patients (group incident).
Frequency of the various critical incidents according to gender
Patient data
Individual data was gathered by reviewing the medical file of every patient admitted to the unit from January 1993 to June 1995. Information gathered included the length of stay, program attended (residential or day patient), gender, age at admission, clinical diagnoses at admission and discharge (up to three concurrent diagnoses at each time) and outcome at discharge.
Outcome measure
Outcome ratings were made by GB according to information in the file as: (i) became worse; (ii) same or no change, admission goals not achieved; (iii) improved, admission goals partially achieved; and (iv) markedly improved, admission goals achieved.
Reliability
The policy of the unit requires that all patient-related events resulting in (or having the potential for) injury, ill health or damage be reported. However, there are no data about the reliability of these reports or about whether all incidents were reported. To assess the interrater reliability of outcome ratings, GB and ED independently rated the outcome of 20 randomly selected cases. Agreement was satisfactory (intraclass r = 0.81). To measure test-retest reliability, GB rated the same 20 cases twice, approximately one year apart. Correlation between outcome ratings on the two occasions was also satisfactory (r = 0.84). The diagnostic process and reliability of clinical diagnoses in the unit (moderate, κ = 0.59) have been reported elsewhere in detail [10]. No specific assessment of the reliability of clinical diagnosis was performed for this study.
Statistical analysis
Outcome at discharge was collapsed into a dichotomous variable of either ‘improved’ (outcome ratings = 3, 4) or ‘not improved’ (ratings = 1, 2). Related diagnoses were also grouped to reduce their number. For example, separation anxiety disorder and other anxiety disorders were consolidated as ‘anxiety disorders’, and oppositional defiant disorder and conduct disorder as ‘oppositional defiant/conduct disorder’. The borderline personality disorder group also included adolescents described as having ‘borderline personality traits’. Stability of diagnosis between admission and discharge was acceptable (average, κ = 0.78), and ranged from 42% (κ = 0.56) for borderline personality disorder to 85% (κ = 0.91) for obsessive compulsive disorder. Analyses were performed using diagnoses made at assessment due to the reasonably good stability and because they were not influenced by events that may have occurred during the admission itself.
Categorical data were analysed using χ2. Odds ratios (OR) are reported when associations were significant (p < 0.05) to indicate the strength of association. Continuous variables were compared using independent groups t-tests (two-tailed). Multiple logistic regression analyses were performed to determine associations between variables when controlling for the effects of confounders. Percentages are rounded to the nearest unit.
Results
There were 243 patients admitted to the unit from January 1993 to June 1995. Their ages ranged from 11 to 19 years (mean = 14.4, SD = 1.6). Sixty-one per cent were male (n = 148). One hundred patients (41%) were involved in critical incidents. Characteristics of patients according to their involvement in critical incidents are presented in Table 2. Overall, there were few significant differences. Individuals involved in critical incidents were more likely to be residential patients, to have a specific learning disorder diagnosis and to have a worse outcome.
Characteristics of patients according to their involvement in critical incidents
Critical incidents
The 100 patients involved in critical incidents generated 214 reports. About half (56%) were involved in a single incident, while two patients (2%) were involved in 10 incidents. The proportion of incidents according to type is shown in Table 1. Male adolescents primarily engaged in aggressive acts (74% vs 20%), while female adolescents engaged in acts of self-harm (51% vs 7%).
Most acts of aggression (85%) were rated as mild or moderate in severity, took place in the residential wing (60%) and occurred throughout the day. By contrast, incidents of self-harm were severe in nature one-third of the time, took place mainly in the residential wing (75%) and mostly during the evening, from 18.00 hours to 24.00 hours (79%) (χ2 = 15.9, df = 2, p < 0.001). There was only one self-harm incident between 24.00 hours and 08.00 hours. Incidents of AWOL were mild in severity and occurred throughout the day. Aggressive and selfharm incidents were rarely group events, while running away occurred more often in a group context (44%; χ2 = 16.9, df = 1, p < 0.001).
Correlates of critical incidents
Involvement in critical incidents may be associated with, or influenced by, other variables (e.g. gender, diagnosis). Therefore, results of bivariate analyses can be misleading. Multivariate analyses (logistic regression) that took into account the effect of potential confounders were carried out. The dichotomous dependent variables were any incident and the incident type (aggression, self-harm, AWOL, each present or absent). Predictor variables were gender, age, diagnoses, programme and length of admission.
Patients who performed aggressive acts were more likely to have received a diagnosis of oppositional/conduct disorder at admission (adjusted odds = 2.9) or have specific developmental disorders (adjusted odds = 3.9). Self-harm incidents were more likely to occur among females (adjusted odds = 3.9), particularly those with a diagnosis of borderline personality (adjusted odds = 7.7) and those who had longer admissions.
Critical incidents and outcome
Logistic regression analysis indicated that patients with a good outcome were more likely to have been in the residential programme, to have had a longer admission and not to have been involved in critical incidents. After controlling for confounders, involvement in critical incidents increased the likelihood of a poorer outcome three-fold (95% CI = 1.7–5.2).
Longer admissions were associated with better outcomes. Patients in the residential programme who improved had an average length of stay of 10.8 weeks compared with 7.9 weeks for those who did not improve (t = 3.05, p = 0.003). This is in spite of the fact that those involved in critical incidents tended to stay longer (11.3 vs 8.9 weeks; t = 1.79, p = 0.076).
Discussion
This is a naturalistic study. As such, it has strengths and limitations. The main weakness is possible inconsistencies between staff when reporting incidents. Also, we did not consider other potentially relevant clinical, developmental and family factors, which have been found to be associated with critical incidents in other studies (e.g. the use of medication [9]). Strengths include the reasonably large sample, that the rater was blind to the number of incidents, the reliability of ratings of outcome and, being retrospective, that reporting practices could not have been influenced by the study.
Overall, we found a lower number of incidents than have been previously reported. The majority of patients (69%) were not involved in critical incidents in this study and half of these were involved only once. As a comparison, Garrison et al. [2] reported on critical incidents by 99 patients admitted to the child psychiatry service of a general hospital. The majority of children (76%) were involved in critical incidents nine times on average. These figures are much higher than those reported here. Apart from differences in setting and programmes, the patients in our study were older. However, in spite of these differences, there are also similar findings. Aggressive acts were more common in boys and self-harm incidents more common in girls and during the evening. In both studies, there were few incidents reported between 24.00 hours and 08.00 hours.
There were differences in the pattern of aggressive and self-destructive events. Acts of aggression were quite constant throughout the day, while acts of self-harm were more common between 18.00 hours and 24.00 hours. Increased prevalence of self-harm in the evening and night may be due to several factors. For example, structured activities tend to decrease at that time, this may result in less supervision, more time alone and increased access to the means to self-harm. Alternatively, nighttime may be a period of particular stress for patients prone to self-harm [11]. It is also unclear whether residential care contributes to such behaviour. Further research in these areas is required.
It is not surprising that adolescents involved in aggressive incidents were more likely to have a diagnosis of oppositional/conduct disorder, as history of aggression is a criterion for diagnosis. Similarly, a diagnosis of borderline personality disorder was a strong predictor of self-harm incidents, but a history of self-harm is also a criterion for diagnosis. This emphasizes the fact that previous conduct is a predictor of future behaviour. Longer hospitalization was associated with increased likelihood of self-harm events and those who self-harmed were less likely to show improvement. This suggests that individuals who self-harm may be kept in hospital longer (perhaps as a result of families finding it hard to cope with such behaviour at home or difficulties with placement), although these patients do not seem to draw much benefit from the extra time in hospital. Early identification of the former group would allow workers to deal with those difficulties and facilitate timely discharge.
We found an unexpected association between specific developmental disorders and aggressive acts, which was not accounted for by age, gender or comorbidity. This requires further research. It was also of interest that a longer admission was associated with a positive outcome. This incidental finding calls into question the trend for increasingly shorter admissions, and requires further study.
Safety of patients and staff is becoming an increasingly important issue in mental health care. Given the rates of aggression towards staff in these settings (16% of all incidents in this study, Table 1) it is necessary to provide specific training to all personnel involved in the supervision of young people in mental health facilities. This should include creating an environment that reduces the likelihood of aggressive acts, promoting selfcontainment among adolescents and giving particular attention to the management of those with oppositional/ conduct disorder and with borderline personality disorder or traits. Structured activities in the evening, group work focusing on improving the problem-solving skills of adolescents and the teaching of how to relate to peers appropriately may go some way to achieving these goals. However, implementing such programmes has staffing and cultural implications. It will probably require increased staff numbers in the evening and a culture that relies on generating skills in patients rather than on safe custody.