Abstract
Suicide attempts by adolescents are not rare. A survey of 3764 high school students in the US found that 5.9% reported suicide attempts not requiring medical care and 1.6% reported attempts requiring medical care [1]. In a review of epidemiological findings from a community sample (the Oregon Adolescent Depression Project) a self-reported lifetime prevalence rate of 7.1% for attempts was found (girls 10.1% and boys 3.8%) [2]. Attempts before puberty were uncommon and rates rose with each older adolescent age group then declined in the young adult ages [2]. In an Australian study of 1699 school attending adolescents aged 15–16 years (the Victorian Adolescent Health Cohort Study), 12 month deliberate self-harm was reported by 5.1% (girls 6.4% and boys 4.0%) of the subjects but ‘true suicide attempt’ rate, as classified by the authors, was only 0.2% (all girls) [3]. It is likely that different definitions of suicide attempt or deliberate self-harm, sample characteristics and methodologies used in different studies have contributed to variations in the reported rates. Nonetheless, a recent review calculated medians of 7.1% lifetime and 5.1% one-year suicide attempts by adolescents, supporting the notion that suicide attempts by young people are not uncommon [4]. Several studies have shown drug ingestion and cutting are the most common method of suicide attempt by community adolescents [2], and deliberate self-poisoning (DSP) by hospital-treated adolescents [5–8], though there are some gender differences.
Risk factors for adolescent non-fatal suicide attempts, deliberate self-harm (DSH) and DSP have been frequently reported. These ‘risk factors’ have usually been derived from cross-sectional or case-control studies and more rarely from longitudinal cohort studies, although the general convention of reporting the correlates identified as risk factors or protective factors is widespread. These risk factors have usually been derived from clinical populations [9–13], school based samples [3, 14–16] or community studies [17–21]. However, risk factors derived from a comparison of both clinical and community-based samples have been reported less often [22, 23]. This is probably because of the practical and methodological difficulties of studying such groups contemporaneously.
There has been considerable similarity in the risk factors for ‘suicidal behaviour’ identified across these different populations. In an extensive review of risk factors for suicide and attempted suicide in young people (15–24 years), which included studies from clinical and community samples, using case-control or psychological autopsy methods and meeting other inclusion criteria, it was concluded that the risk factors for ‘suicidal behaviour’ included: social and educational disadvantage, childhood and family adversity, psychopathology, individual vulnerabilities, exposure to stressful life events and social, cultural and contextual factors [24]. The importance of the cumulative effect of multiple risk factors was also reported. The strongest risk factors were mental disorders (affective disorders, substance abuse disorders and antisocial behaviours) and a history of psychopathology; the author concluded that interventions focused on youth with mental disorders would be a priority for reducing suicidal behaviours [24].
Nevertheless, there remains disagreement and uncertainty as to which risk factors to target and whether it would be more advantageous to target interventions in community populations or in clinical populations or both [25–29]. Recently in the US, a comprehensive national plan for attempting to reduce suicide by targeting multiple, modifiable risk factors across all levels of clinical and community populations has been proposed and this approach may be implemented in the future [30]. Additionally, it would also be useful to know the magnitude of the risk for each risk factor (ideally the population attributable risk) in order to help prioritize which risk factors to address in intervention. Arguing for the need for studies to account for the comorbidity of psychiatric disorder in order to better interpret the magnitude of risk and for studies to account for age and gender differences, Gould says; ‘a monolithic diagnostic risk profile for suicidality, ignoring gender- and age-specific risks, is inadequate’ [21]. It has been suggested that there is a continuum of risk factors for suicidal behaviour, from ideation through attempt to completed suicide. One example of this has been a dimensional model of suicidal behaviours in which those attempting suicide are distinguished from those reporting suicidal ideation alone by having a greater burden of the same risk factors [18]. Furthermore, it has been suggested that there may be a continuum of risk factors for suicidal behaviour across different samples. That is to say the same risk factors may apply in clinical and community samples and that these risk factors are more powerful in the hospitaltreated samples [23].
A Norwegian hospital-treated sample of suicide attempters (n = 91, 13–19 years) was compared with an age-matched general population sample (n = 141). The multivariate risk factors for the clinical sample were: depression, disruptive disorders, low self-worth, infrequent support from parents or peers, parents excessive drinking and low socioeconomic status. For community adolescents the risk factors were: depression and loneliness [23]. The Norwegian study suggested targeting the same risk factors at the community level for nonhospitalized adolescents and at an individual level for hospitalized adolescents, namely depression, low selfesteem and family communication. An English study compared 33 adolescent hospital-treated overdose patients (12–18 years) with two groups; 30 psychiatric controls and 30 community controls, matched for age and sex [22]. However, the community controls did not have psychiatric diagnoses made and the psychiatric control group had history of overdose or the presence of a psychotic disorder as exclusion criteria. Depression and impulsivity distinguished the overdose group.
The question remains for older adolescents and young adults, (whether in the community or presenting at hospital for treatment) ‘What are the potentially modifiable risk factors that could be targeted for future intervention in order to reduce suicidal behaviour?’
Our study seeks to address the issue of risk factors for suicidal behaviour in a clinical sample and a community sample in comparison with non-suicidal community controls. It uses representative samples, a standardized diagnostic instrument and multivariate techniques to examine the contribution of gender, marital status, education level, mental illness and disability. It does this for a clinical and a community population 18–24 years and so complements the findings from other studies using clinical and community samples for younger adolescents [22, 23]. It also provides comparison with New Zealand studies for mental illness (affective, anxiety and substance related disorders) and suicide attempt [31, 32]. There are no Australasian controlled studies examining educational level, personality disorder or disability levels in relation to suicide attempts and so the current study may offer new information for these variables.
Objective
The aim of this study was to compare correlates for suicidal behaviour in older adolescents and young adults who had hospital treatment for an episode of deliberate self-poisoning (DSP) with a community sample reporting attempted suicide (AS) during their lifetime.
Specifically, we were interested in two questions:
1 Do the same correlates apply to suicidal behaviour in the community and in clinical populations?
2 Do these correlates show a stronger association among hospital-treated attempters than among community attempters?
Method
Design
Three study designs were used; case–case, case–control and crosssectional population studies.
Subjects
Three groups of subjects were used in the present analysis
1 Clinical sample of deliberate self-poisoning (DSP) patients. These were recruited from the Hunter Area Toxicology Service in Newcastle, Australia, a regional service for the treatment of poisoned patients. The model for the treatment of self-poisoning patients has been previously described [33]. This service provided for a population of around 550 000 at the time of this study. Consenting patients were usually assessed within two weeks and up to a maximum of eight weeks after an episode of DSP, from April 1998 to June 2000. Patients aged 18–24 (n = 51) were selected from the total sample aged 18–82 years (n = 219) who had agreed to the detailed interview process at a response rate of 50%.
2 Community sample of attempted suicide (AS). The Australian National Survey of Mental Health and Wellbeing (NSMHWB) surveyed a random sample of adults (n = 10 641) in 1997, with a response rate of 78%. The methodology and principal findings of this survey [34] and the suicide attempt questions have been reported elsewhere [35]. Subjects 18–24 years (n = 962) were selected from this sample and of these (n = 31) reported lifetime ‘attempted suicide’. These respondents were used as the community AS sample.
3 Community control (CC) sample. Of the NSMHWB sample 842 subjects, 18–24 years, reported no suicidal thoughts and no AS and this sample was used as the control group. The remainder of the (n = 962) had reported lifetime suicidal ideation but no attempts and they were excluded from the analysis (n = 89).
Measures
All subjects were interviewed with the instrument used in the NSMHWB. This is a computerized interview, administered by trained and accredited interviewers. The interview used the Composite International Diagnostic Interview (CIDI-A) [36] for assessment of DSM-IV anxiety, affective and substance use disorders, a screener for psychosis, neurasthenia (modified version) [37] and personality disorders (modified International Personality Disorder Examination Questionnaire) [38] and the Short-Form 12 (SF-12) [39] for disability related to mental and physical health. Higher scores on the SF-12 indicate lower levels of disability. The survey asked about the presence of 12 chronic physical conditions (asthma, chronic bronchitis, anaemia, high blood pressure, heart trouble, arthritis, kidney disease, diabetes, cancer, stomach ulcer, chronic gallbladder or liver trouble and hernia or rupture). Demographic information and past history of AS (lifetime and past 12 months) were also collected.
Statistical analyses
Standard statistical techniques as described below were used for all analyses. However, due to the weighting procedures and complex sampling used in the NSMHWB, special software was required for analyses of the community AS and CC samples. These analyses were conducted using SUDAAN software [40] which is specifically designed for use with complex survey samples. No weighting was used for the clinical DSP group.
Case–case study
A comparison of the clinical DSP group and community AS group was made on sex, marital status, education level, mental health disability, physical disability and mental disorder. Mental disorders using DSM-IV diagnoses were: affective, anxiety, substance-use, personality disorder and any mental disorder (which included neurasthenia and psychosis in addition to those above). T-tests were used for continuous variables and χ2 tests for categorical variables. Weighted values were used for the categorical variables from the community sample and unweighted means were used for continuous variables. Bonferroni adjusted significance values (p < 0.008) were used to account for the number of tests.
Cross-sectional population study
Comparison of the community AS group with the CC group. Bivariate logistic regression analyses were conducted for the same variables as in the case–case study above. Significant variables (p < 0.05) were then included in a backward step-wise logistic regression to develop a multivariate model to estimate the adjusted odds ratios and confidence intervals (CI 95%) for the correlates of AS. The model controlled for the presence of any physical disorder.
Case–control study
Comparison of the clinical DSP group with the CC group. Bivariate logistic regression analyses were conducted for the same variables as in the case–case study above. Significant variables (p < 0.05) were then included in a backward step-wise logistic regression to develop a multivariate model to estimate the adjusted odds ratios and confidence intervals (CI 95%) for the correlates of DSP. The model controlled for the presence of any physical disorder.
Results
The results of the case–case study can be seen in Table 1. Clinical DSP subjects were significantly more likely than community AS subjects to be female (p = 0.007), have lower educational attainment (p = 0.023), to suffer specific psychiatric disorders (affective, p = 0.001; anxiety, p = 0.004; substance use, p = 0.007; and personality disorders, p = 0.014), any mental disorder (p = 0.007), and to have greater levels of mental health related disability (p < 0.0001).
Characteristics of community attempted suicide (AS) versus clinical deliberate self-poisoning (DSP) group
The results of the multivariate analyses for the cross-sectional population and case–control studies can be seen in Table 2. Community AS were not different to CC on demographic factors nor any personality disorder but were more likely to be suffering an anxiety disorder (OR = 9.4, 95% CI = 1.7–52.8), any substance use disorder (OR = 3.0, 95% CI = 1.1–8.7), any mental disorder (OR = 7.9, 95% CI = 2.4–26.0) and to have greater mental health related disability (OR = 0.5, 95% CI = 0.4–0.7 for a 1 SD decrease in SF-12 score). Affective disorder just failed to significantly distinguish the AS group from CC (p = 0.052). The presence of a physical disorder was not a significant correlate.
Multivariate correlates for self-harm among community suicide attempters (AS) and hospital-treated deliberate self-poisoning (DSP) patients compared with community controls (CC)
The clinical DSP group was significantly distinguished from the CC group by; female gender (OR = 5.7, 95% CI = 1.7–19.4), any affective disorder (OR = 23.0, 95% CI = 6.9–76.5), any anxiety disorder (OR = 7.4, 95% CI = 2.2–25.1), any substance use disorder (OR = 19.2, 95% CI = 5.6–65.4) and mental health related disability (OR = 0.5, 95% CI = 0.3–0.7). Any mental disorder was not included in the regression equation because 96% of the DSP group had at least one mental disorder. Personality disorder and physical disorders were not significant correlates.
Discussion
The findings support a continuum model of suicidality. Psychopathology and mental health disability were greater in clinical DSP than the community AS group. Moreover, a similar pattern of correlates was observed for both DSP and AS in comparison to the community controls. Mental health related disability, any mental disorder, anxiety and substance use disorders were all associated with suicidal behaviour in both DSP and AS groups. The association between affective disorder and AS was close to significant and this may have been because the small sample size limited the statistical power of the analysis. Affective and substance-related disorders showed a much stronger association with the DSP group than for the community AS group compared to the community controls. Conversely, the strength of association for anxiety disorder and mental health related disability was similar for both groups. Moreover, direct comparison of the DSP and AS groups suggested that the DSP group had higher proportions of educational disadvantage, mental disorder, and mental health related disability than the community AS group.
These findings are somewhat different from those of Groholt et al. whose study examined a younger sample using different diagnostic instruments, a wider range of social and family factors and no physical illness factors or disability factors [23]. Although the findings for depression were similar, anxiety disorder and excessive use of alcohol or illegal drugs was not found to be a significant multivariate correlate in the Norwegian study. Educational level was not independently associated with suicidal behaviour in either study. Anxiety disorder remained significant in our study even when adjusting for the presence of affective disorder and other variables. This was similar to the finding by Gould et al. [21] but in contrast to other studies [10, 12] where anxiety disorder was not a correlate. There is considerable similarity of correlates in these Australian samples with those reported in younger community adolescents (9–17 years) in the US [21] and in a longitudinal cohort of New Zealand adolescents (15–21 years) [41]. Overall, the independent contribution of mental illnesses; depression, substance related disorders and any mental disorder is similar to that reported in comprehensive reviews arising from predominantly US and New Zealand original studies [24]. The current study helps to confirm the relevance of these factors for Australian community and hospital-treated populations.
This pattern of any mental illness and these specific mental illnesses as correlates for AS or DSP is also similar to the well-established findings for completed suicide [42–44] and for suicidal ideation [21, 41]. This is consistent with another continuum notion. Where the mental illness risk factors for ideation, attempts and completed suicide are similar, there may be a spectrum of suicidal behaviour for which a system of interventions can be targeted at the same potentially modifiable risk factors.
Our study examined mental health related disability, which has not usually been done in other studies. Mental health related disability was significantly worse in the DSP sample compared with the AS group and mental health disability was an independent correlate (controlling for physical and mental illness variables) in the multivariate analyses for both DSP and AS compared with CC. This suggests that greater mental health related disability is associated with DSP and AS that is not fully accounted for by common psychopathology (during the previous 12 months). This may be analogous to the finding that hopelessness, a potent risk factor for suicidal behaviour, may persist even after the symptoms of depression or other mental illness have remitted [30]. This finding suggests that mental health related disability might be used to identify high risk groups and that treatments aimed at reducing disability could be explored as potentially useful interventions. There are situations in the community and clinical services where this could be of practical advantage. There are circumstances where the diagnosis of mental disorder by clinical assessment or diagnostic instrument is not readily available or perhaps not part of accepted practice but where disability is a more familiar construct. In community settings this might include Departments of Social Security or Social Welfare, Juvenile Justice, School counsellors or support groups. In clinical or hospital settings (particularly those without ready access to psychiatric services) this may include Social Work or Occupational Therapy services. Moreover, general practice and speciality physician or paediatric practices in the public sector may find the systematic measurement of disability by a brief standardized instrument a possibility for their patients.
Our study also controlled for chronic physical illness in the two logistic regression models. Of the 12 chronic illnesses that were examined, asthma and diabetes are two of the more common illness affecting older adolescents and young adults. Few other studies have included measures of physical illness in the assessment of correlates for suicidal behaviour.
Strengths and limitations
The methodological strengths of this study include the sampling of the DSP and community groups in ways that would tend to minimize ascertainment bias. The DSP sample is representative of one regional geographical area and the community sample representative of the Australian population. The study designs are acceptable and the same validated assessment instruments, conducted by trained interviewers, were used in all three groups. The samples were studied at a similar date. The age range addresses older adolescents and young adults which complements the similar study of clinical and community younger adolescents in the literature [23].
However, there are some methodological problems. The time since an episode of AS/DSP to assessment is substantially different and the meaning of ‘attempted suicide’ in the question asked of the AS and CC groups is open to interpretation by the subjects. In our study AS was determined by subject self-report and hospital clinical staff determined DSP. Moreover, DSP and AS are not the same construct and generally DSP is thought to include subjects attempting suicide with subjects who have other intentions. The symptom report of the DSP subjects may be greater because of the recency of the self-poisoning episode and thus an increased likelihood that any illnesses or symptoms will be reported. Nevertheless, the findings are generally in keeping with many studies from other countries and offer some basis for intervention activities for Australian youth.
Clinical and public health implications
A recent review tried to define possible interventions to optimally target risk factors based on qualities of the putative risk factors and a classification of interventions: Clinical, Indicated, Selective and Universal [45]. They identified several areas of shortfall where putative risk factors were known. In particular, these authors suggested that affective psychopathology was the most important risk factor for suicide and suicidal behaviour in community settings but that the few available intervention studies had serious methodological limitations. They also concluded that there are no adequately trialled interventions targeting the treatment or prevention of suicidal behaviour in adolescent and young adult groups. The results of our study suggest that the modifiable risk factors to be targeted by future interventions would include at least: any mental illness; affective, anxiety, substance use related disorders and mental health related disability. Whether these interventions might be best utilized in Clinical, Indicated, Selective and Universal approaches is yet to be determined. However, at the hospital level, especially in the emergency room, it would seem reasonable to offer deliberate self-harm patients a psychiatric assessment which among other things specifically tried to detect depression, anxiety and substance related disorders as well as measure of mental health related disability. There are guidelines available that suggest that such patients should have a psychiatric assessment [46]. In Western Australia this was reportedly achieved for 70% of patients [47] and in the New South Wales unit where this study was conducted 97% of DSP patients had psychiatric assessment [33]. Detection would identify a group worthy of specific intervention, (i.e. those with extant disability, potentially treatable disorders and/or potentially reversible risk factors for suicidal behaviour and future completed suicide).
Intervention in the community, on the other hand, is more complex [48]. Public health and primary care strategies that screened for the relevant disorders and for related disability would be one relevant response. Environmental and individual approaches that decrease the prevalence of modifiable risk factors like mental disorders and disability would also be potentially worthwhile. Of course there are other public health measures which go beyond the findings of this study, such as teaching coping skills, improvement in relationship problems, reduction of social isolation and reducing child maltreatment, physical abuse and sexual abuse [30]. Such interventions need to be part of a comprehensive and coordinated response to the problem of youth suicide attempts in Australia.
Footnotes
Acknowledgements
Thanks to Rachel Garrett, Ketrina Sly and Neta Moses for data collection. Supported in part by a contract from the Australian Department of Health and Aged Services to the World Health Organization Collaborating Centre for Mental Health to support a survey data analysis.