Abstract
Depressed individuals commonly admit to preoccupations about self-harm, and while determinants of suicide have been extensively pursued over a lengthy period, studies of deliberate self-harm (DSH) have been fewer. This may reflect the view that DSH overlaps with suicide attempts, which is hardly surprising when DSH and recent discharge from psychiatric hospital have been identified as the two key risk factors to suicide [1], and when causes of parasuicidal acts have also been extensively studied. However, the recent introduction of the acronym DSH suggests a growing focus on a phenomenon increasing in many Western regions. For instance, an Oxford study [2] quantified increased rates over a 10-year period (1990–2000). Its high frequency is evidenced in an Australian study [3] which quantified 25 120 hospital admissions in a 12-month period (1997–1998) – a DSH population rate of 137.5/100 000. It may or may not equate with parasuicial or suicidal behaviour. The latter may reflect a wide variety of determinants (e.g. the consequence of an episode of psychotic depression, internalizing and externalizing personality styles, proximal stressors), while – as considered later – DSH appears to reflect an externalizing response in an isolated individual who has commonly been exposed to earlier deprivational experiences.
In this paper we consider possible contributors to DSH in depressed patients. In a recent review paper [2], Hawton suggested that DSH was most closely identified with a range of sociodemographic variables (e.g. being single, divorced or of lower social class) and with misuse of illicit substances. Another Oxford study [4] quantified rates of psychiatric disorders associated with DSH. An affective disorder (72%) was most common, followed by personality disorder (46%), substance use (alcohol=27%; drugs=9%) and anxiety disorder (23%). Our study seeks to determine distal as well as proximal antecedents and correlates of DSH in a sample of depressed subjects. By analysing data from three samples, the likely replicability of findings is considered and an overall pattern analysis allowed.
Method
Study design
We studied patients referred to our tertiary-referral Mood Disorders Unit Clinic. Over the 20 years of its existence, patients have been consistently referred for treatment-resistant and/or severe depressive disorders. There has again been consistency in patients completing a set of self-report questionnaires, being interviewed by a research psychologist and by a research psychiatrist, and with structured and semistructured probe sheets allowing information to be recorded in standardized ways. For the current study we included subjects from our last three databases, comprising two studies focusing on assessing the contributions of personality, anxiety and life events to the onset of depressive disorders (i.e. PAL-I and PAL-II, assessing personality, anxiety and life event stressors) and a study weighted to assessing causes of treatment-resistant depression (CORD or causes of resistant depression study), with recruitment to these three studies occurring over a 10-year period. Across the studies, some data (e.g. sociodemographic, diagnostic groupings) were measured in consistent ways, other domains (e.g. assessing developmental risk factors such as anomalous parenting) were consistently studied but with variable measures or approaches, while some domains (e.g. assessing factors contributing to treatment resistant depression) were unique to each of the individual studies. As the individual and total datasets are extremely large, study variables are not detailed here and are, instead, specified in the relevant section of the Results.
In all three studies, DSH was assessed by having the psychiatrist determine whether the patient had a lifetime history of ‘self-injury’ (e.g. wrist-cutting, head-bashing) and, if so, the age of initiation. Self-harm did not include suicide attempts (as identified by questioning the patient) and the motive for any particular DSH act was not pursued.
Analyses
Percentages for categorical data and odds ratios (ORs) for risk were obtained from cross-tabulation analysis of the two groups and significance determined by the McNemar test in non-parametric analysis of ‘2-related samples’. Continuous data were compared by paired sample t-tests. In considering a large number of univariate analyses of data from the three studies, we regard an OR of 2.0 or more and a formally significant t-test (i.e. p<0.05), as indicating a variable potentially differentiating self-harmers and controls, but provide confidence interval (CI) estimates. We undertook multivariate logistic regression analyses to determine the sets of variables distinguishing DSH from control subjects.
Results
The respective numbers of those who admitted to self-harming acts (DSH subjects) across the three studies were 56 (PAL-I), 49 (PALII) and 36 (CORD), giving respective percentages of 20.7%, 23.1% and 16.7%. Table 1 shows a distinct female preponderance in all three samples. We then derived sets of control subjects within each sample to match the DSH subjects by age, gender and by diagnostic subtype (i.e. psychotic, melancholic or non-melancholic), using DSM (DSM-III-R or DSM-IV) and clinical diagnostic decisions.
Sociodemographic and diagnostic variables for deliberate self-harming (DSH) and control (Con) subsets in each of three samples
Table 1 reports sociodemographic and diagnostic variables. Matching of the DSH and control groups ensured no difference in terms of the three principal depressive subtypes, gender rates were equal, although the subjects were somewhat younger than the controls in two of the three samples. There was a general trend for the DSH subjects to have never married and to be less likely to be currently in full-time employment, while in two of the samples DSH subjects tended to be more likely to be receiving social services. In terms of illness course, there was also a general trend (significant in two samples) for DSH subjects to be more likely to have bipolar disorder.
Lifetime anxiety disorders were assessed by the Composite International Diagnostic Interview (CIDI) [6] in each sample and with rates compared in Table 2. The only significant difference was for PAL-II DSH subjects to be more likely to meet criteria for generalized anxiety disorder, with a similar trend evident in the other two samples. Examining ORs shows no consistent over-representation of any other lifetime anxiety disorder. A past history of illicit drug use was examined in each study (Table 3). The only significant, but not consistent, illicit drug over-represented in DSH subjects was marijuana, being significant in the PAL-II and CORD studies. Current use of any illicit drug was consistently over-represented across the three studies in DSH subjects but was non-significant (despite large ORs), reflecting the small sub sample numbers. A high intake of alcohol was similarly non-significantly over-represented in DSH subjects across the three samples.
Comparing deliberate self-harm (DSH) subjects with matched controls (Con) in lifetime anxiety disorder rates
Comparing deliberate self-harm (DSH) subjects with matched controls (Con) on drug and alcohol use
In each study, some 30–40 depressive symptoms were examined. 7] or self-report Beck Depression Inventory [5] scores. Historically, the DSH subjects were distinctly more likely (all p<0.001) to report a history of suicide attempts (OR=7.49, CI=3.1–18.3 in PAL-I; OR=2.73, CI=1.2–6.1 in PAL-II; and OR=10.8, CI=3.5–32.4) in the CORD study. During the current episode of depression, the DSH subjects were also more likely (p<0.001) to report suicidal attempts in the PAL-I (OR=4.53, CI=1.9–11.0) and CORD (OR=8.98, CI=2.3–34.9) studies but not in the PAL-II sample (30.6% vs 20.5%, OR=1.22).
Comparing deliberate self-harm (DSH) subjects with matched controls (Con) on depression symptoms
Comparing deliberate self-harm (DSH) subjects with matched controls (Con) on formal depression, stressor severity and disordered personality measures
In each of the studies, parental care and overprotection variables were examined by use of either the parental bonding instrument (PBI) [8] or the measure of parental style (MOPS) [9] measures. Subjects and controls did not differ on either measure. The impact of dysfunctional parenting, alcoholism in parents, abuse between parents, physical violence between parents and other measures of dysfunctional parenting were examined (Table 6) in each study. A depressed mother was more likely to be reported by DSH subjects in the PAL-I study (42.9% vs 27.3%, OR=2.00) and in the PAL-II group (49.0% vs 29.8%, OR=2.6) but not in the CORD study. Maternal problems with alcohol were also more likely to be reported by the DSH subjects in the PAL-I (13.2% vs 6.5%, OR=2.18) and CORD (11.4% vs 5.6%, OR=2.19) studies, but not in the PAL-II study. Paternal depression, paternal alcoholism, verbal abuse between parents, dysfunctional parenting as experienced by the recipient and dysfunctional relationships between the parents did not appear to be over-represented in the DSH groups in any sample, although a higher rate of paternal alcoholism was reported by DSH groups in the CORD study (28.6% vs 16.7%; OR=2.0).
Comparing deliberate self-harm (DSH) subjects with matched controls (Con) on childhood environment variables
In addition to these consistently assessed variables, each study has had a number of more study-specific early environmental variables assessed, such as parental depression and alcoholism, parental conflict, multiple parenting figures, bullying and abuse. Table 6 data indicate few consistent differences. However, and while not formally significant, there was a consistent trend for the DSH subjects to be more likely to have experienced sexual abuse before the age of 16 years – both from a parent and from a non-parent. As the numbers of DSH males exposed to sexual abuse were relatively few, we specify data for females. In each study, female DSH subjects were more likely to report childhood sexual abuse (CSA) (PAL-I, 44.1% vs 23.5%; PAL-II, 46.3% vs 22.0%; CORD, 41.4% vs 13.8%).
Personality assessment was formally undertaken only in the PAL-II study, using the Temperament and Character Inventory [10]. There were no group differences on the four temperament (i.e. novelty seeking, harm avoidance, reward dependence and persistence) or character (self-directiveness, cooperativeness and self-transcendence) dimensions.
In each study subjects rated the extent to which they engaged in 21 differing behaviours when under stress [11], measuring items assessing both ‘acting in’ and ‘acting out’ coping responses. Table 5 data indicate that DSH subjects returned significantly higher scores on the subset of violent and physically aggressive ‘acting out’ behaviours in all three samples. However, DSH subjects did not differ on the ‘acting in’ scale (assessing internalizing responses to stressors) in any study.
In each study we assessed the probability of disordered personality functioning by having a psychiatrist rate the patient's functioning across a range of constructs. Our ‘domain’ measure required rating the extent to which intimate, family, peer and work relationships – as well as work maintenance –were ‘functional’, ‘probably dysfunctional’ or ‘definitely dysfunctional’ over time, with scale scores creating a total ‘domain’ score. Similarly, a total ‘parameter’ score was created by assessment of eight expressions of disordered personality (e.g. inflexible or defective response styles, personality style causing significant discomfort for the individual and inability to function effectively and efficiently). Table 5 data show that in all three samples DSH subjects returned significantly higher ‘parameter’ scores, while there was a trend for DSH subjects to score higher on the ‘domain’ measure – both indicative of greater dysfunction.
Levels of reported psychosocial stressors were assessed in each study in differing ways. In the PAL-1 study, severity of stressors both in the 12 months before depression onset and during the episode were assessed but were non-differential. Enduring stress emerging from a range of situations (including relationships with partner, family, children, employers and work mates), as well as work itself and related factors, was also non-differential. In the PAL-II study, the DSH subjects did report exposure to higher levels of stress over the previous 12 months (3.7 vs 3.1, t=2.43, p<0.025), while the two groups did not differ on the same variable in the CORD study. In that last study the psychiatrist was required to make a number of ratings about long-standing and recent psychosocial stresses. These DSH subjects were more likely to be rated as having experienced deprivational experiences in childhood, such as abuse or neglect (68.8% vs 33.3%, OR=4.40, p<0.025), but did not differ in terms of judged stress levels in adulthood. The psychiatrists also rated them as having high levels of disordered personality functioning (84% vs 48.5%, OR=5.74, p<0.025). The two groups did not differ in terms of psychiatrist ratings of levels of either acute or ongoing adversity and stress.
We undertook a number of logistic regression analyses, assembling variables identified as having consistency across samples and entering them into analyses that sought to examine a priori likely DSH: (i) antecedents; and (ii) correlates, with analyses undertaken for all subjects as well as for male and female subjects separately. Our a priori ‘antecedent’ variables were: (i) depressed mother; (ii) alcoholic mother; (iii) exposure to physical abuse in childhood; (iv) exposure to sexual abuse in childhood; and (v) having a bipolar disorder. In such analyses, no significant predictors emerged in the PAL-I study. In the PAL-II study, maternal depression was the only significant variable in the whole sample (Wald=4.22, OR=2.40, p<0.05), no variable was identified for the male subset, while exposure to sexual abuse in childhood (Wald=5.24, OR=3.07, p<0.025) emerged in the female subset. In the CORD study, CSA emerged as a significant predictor in the whole sample (Wald=7.22, OR=5.99, p<0.01) and in the female subset (Wald=5.13, OR=5.14, p<0.025), but not in the male subset.
Our a priori ‘correlates of self-harming variables’ were current: (i) unemployment; (ii) illicit drug use; (iii) appetite loss; (iv) suicidal ideation; and (v) poor concentration, as well as lifetime (vi) cigarette use; (vii) marijuana use; and (viii) suicide attempt, as well as (ix) high ‘acting out’; (x) high domain; and (xi) parameter scores. In the PAL-I study, appetite loss (Wald=4.69, OR=2.62), previous suicide attempt (Wald=13.1, OR=5.59) and current suicidal ideation (Wald=4.08, OR=3.73) were significant predictors of DSH for the whole sample, while a previous suicide attempt was significant in analyses of separate male and female subsets and appetite loss significant in the female subset. In the PAL-II subset, high ‘acting out’ scores (Wald=11.1, OR=1.34) and current suicidal ideation (Wald=4.75, OR=4.71) were significant predictors of DSH in the whole sample (as they were in the female subset) while a lack of predictors in the male subset was likely to reflect the fewer males (i.e. eight self-harmers and eight controls). In the CORD study, there were four significant predictors in the whole set: current illicit drug use (Wald=3.88, OR=35.93), previous suicide attempt (Wald=10.56, OR=13.83), current suicidal ideation (Wald=6.46, OR=33.14) and higher ‘parameter’ scores (Wald=7.89, OR=1.16). In this last sample, both current suicidal ideation and past suicide attempts were significant predictors for the female subset and higher ‘acting out’ scores were predictive for the male subset.
Finally, ignoring whether predictor variables might be antecedents or correlates of DSH we examined the predictability of an overall refined set (i.e. maternal depression; exposure to sexual abuse in childhood; previous suicide attempt; current suicidal ideation; current illicit drug use; high ‘acting out’ scores; high ‘parameter’ scores) in a set of discriminant function analyses. In the PAL-I sample, two significant variables were identified: previous suicide attempt (Wald=12.9, OR=21.8) and high ‘acting out’ scores (Wald=4.82, OR=1.32), generating an overall prediction of 85.1% (87.5% sensitivity and 82.6% specificity). In the PAL-II study, current suicidal ideation (Wald=4.75, OR=4.71) and high ‘acting out’ scores (Wald=11.10, OR=1.34) were the two predictors (69.4% overall, 71.4% sensitivity and 67.3% specificity). In the CORD study, there were four predictors identified: previous suicide attempt (Wald=10.56, OR=13.8); current suicidal ideation (Wald=6.46, OR=33.1); current illicit drug use (Wald=3.88, OR=35.9); and high ‘parameter’ scores (Wald=7.89, OR=1.2), with an overall prediction of 76.8% (82.4% sensitivity, 71.4% specificity). Exposure to sexual abuse in childhood was therefore not identified in any of the three samples and, when similar analyses were undertaken in female and male subsets, it was only identified as a significant variable in the female subset of the PAL-II study.
Discussion
Our samples were recruited from a tertiary referral unit assessing those with severe and/or treatment-resistant depressive disorders. Thus, our DSH rates may be inflated, whether DSH is a determinant of such disorder characteristics or whether as failure to respond to treatment. Again, the determinants of DSH may vary considerably in those attending such a tertiary referral centre and those assessed in other contexts. Our prevalence estimates (overall, suggesting that some 20% of the depressed subjects had such a history and a strong female preponderance) may therefore be idiosyncratic to our sample. The extent to which sample characteristics influenced our pattern analyses and our attempts to identify possible DSH determinants and correlates is more problematic, and we concede that findings may not extrapolate to other depressive and non-clinical samples. Study advantages include close age, gender and diagnostic matching and the use of multiple samples to allow consistency of findings to be considered. Disadvantages included the relatively few DSH subjects in each sample (i.e. 36–56) which, when assessing the relevance of a low prevalence variable (e.g. sexual abuse) may have prevented meaningful risk variables from achieving statistical significance.
Our study considered an extremely wide set of constructs, with univariate analyses assisting identification of candidate constructs either contributing to or being associated with DSH. We found support for sociodemographic variables identified in previous studies (e.g. more likely to be unemployed or unmarried), but not for the common finding that DSH subjects were more likely to report, or be rated as, experiencing greater acute or chronic stress. The last finding may reflect the nature of our control group. That is, all our subjects were depressed and thus likely to be experiencing high rates of stress. Thus, associations between DSH and high levels of concurrent stress in community studies may merely reflect the DSH subjects' depression status rather than their DSH status.
Clinical observation suggests that many who self-harm (and particularly those who cut themselves) have high levels of trait or state anxiety, describing such acts as often lowering their anxiety. However, across our samples we found no clear evidence for DSH subjects to differ in terms of lifetime anxiety disorder rates and no consistency in individual anxiety disorder representation. Thus, anxiety is unlikely to be a major determinant or correlate of DSH and is in line with a recent report [12].
We confirmed previous research identifying the relevance of illicit drug use (with both past and current rates increased in our DSH subjects) and the suggestion of high alcohol use. Such use may act so as to disinhibit individuals (state model) and promote DSH behaviours, or be a concomitant of the greater chance of social and personality functioning variables identified in the study.
We found no evidence to suggest a link between depression severity and DSH, but did find a consistent trend (significant in two of the samples) for DSH subjects to be more likely to have bipolar disorder. If valid, it could reflect the common pattern of forme fruste presentations of bipolar disorder before formal onset (which might include DSH), the disinhibiting nature of manic states, shared impulsivity, or the commonly more severe episodes of depression in those with bipolar (as against unipolar) disorder.
Subjects with definite or possible bipolar disorder were somewhat more likely to have a suicide attempt history in the PAL-I (56% vs 34%) and PAL-II studies (69% vs 42%) but no more likely (40% vs 44%) in the CORD study with such results disallowing clarification of any one possible explanation.
The DSH subjects were distinctly and consistently (across samples) more likely to have made a suicide attempt, and to report current suicidal ideation, as expected when the self-injurious nature of each action is considered. However, suicide attempts also occurred to significant extents in our control groups, suggesting that, while DSH and suicide attempts may overlap, there is wisdom in examining for their shared and independent determinants.
Distal stressors and possible determinants were pursued by examining the impact of many deprivational and abusive experiences in early childhood. Importantly, we found no suggestion that DSH subjects differed from controls on broad measures (i.e. PBI and MOPS) of parental care and control. Although we have undertaken numerous studies [8] suggesting that those with non-melancholic depression are more likely to report low parental care and parental overprotection, their current lack of differentiation suggests that such parental characteristics are more likely to influence the vulnerability to depression per se and not directly to DSH. The DSH subjects were more likely, however, to report their mothers as having experienced depression and to have had problems with alcohol, so that DSH may relate more to such nuances of dysfunctional parenting. Childhood sexual abuse was consistently more likely to be reported by DSH subjects, with such exposure at least twice as likely to be reported by female DSH subjects. Whether such abuse and dysfunctional early environments might lead directly to DSH or indirectly via the impact on personality development, impulse control and other factors remains to be clarified. In an earlier paper [13], analyses indicated that ‘depression’ was unlikely to be a direct consequence of CSA. Instead, as CSA was associated with a greater chance of having experienced a dysfunctional childhood home environment and a borderline personality style, we argued that the greater chance of depression was more likely to reflect those intermediate factors. The same model may hold for the link between CSA and DSH, although our several analyses failed to identify distinctly overrepresented personality styles in the DSH subjects. To the extent that the TPQ measures personality style – as against disordered personality functioning – the lack of differentiation on the TPQ measure would suggest that personality style per se is unlikely to be a contributory factor. We did, however, find the DSH subjects scored higher on our ‘acting out’ and ‘parameter’ measures. The former comprises items focusing on more violent externalizing responses to ‘stress’, while the latter assesses attitudes and behaviours likely to reflect disordered personality functioning (e.g. inability to adjust to the environment, vicious or self-defeating cycles). Thus, such differences suggest that DSH is likely to reflect a level of disordered personality functioning manifested in poor impulse control and the externalization of stress rather than by ‘internalizing’ behaviours such as crying, becoming asocial and keeping to oneself.
As our data were collected cross-sectionally and as many of our measures relied on indirect clinical assessment, we elected not to undertake path analyses or employ multivariate strategies that might attempt to sequence the possible determinants of DSH. Together with the study being ‘underpowered’, we therefore regard our multivariate logistic regression analyses as providing modest interpretative power and focus on the final analyses (which did not distinguish between a priori antecedent and correlate variables). Such analyses consistently identified previous suicide attempts and high ‘acting out’ scores as predictors of DSH, while, in one sample, current illicit drug use and high ‘parameter’ scores made an additional significant prediction. Those results could be seen as circular, but we suggest that they can also be interpreted as indicating that the roots of DSH are shared with parasuicidal and externalizing stress responses, again suggesting poor impulse control as being a central variable. Determinants of such a process are likely to reflect multiple genetic and environmental factors – with our univariate analyses indicating the possible contribution of a number of deprivational and abusive experiences in childhood.
How do our results tally with the many more general studies of DSH? The body of literature is remarkably consistent in suggesting that individuals engaging in such behaviours were distinctly more likely to have been exposed to deprivational (particularly neglectful and abusive) experiences in childhood, and that DSH reflects an interaction of predisposing (constitutional and developmental) and precipitating stressful events. However, as most studies have been cross-sectional, and as many people experiencing such deprivation do not show such behaviours, more longitudinal models are required. Jones and Daniels [14] reviewed a large literature in proposing an ethological approach (in both man and animals) for understanding, contributing and underpinning processes. In essence, that model assumes a number of component proximate steps (i.e. the potential for aggression, an isolating or abusive early environment) that lead to frustration and subsequent threat and arousal. In a socially integrated individual, a socially aggressive response might be expected but, for those who are ‘isolated’ (perceived or real), self-injurious and self-mutilating behaviours are more likely in response to frustration and rage.
Our results suggest that DSH in depressed individuals is unlikely to be ‘caused’ by proximal or immediate stressors per se, and is more likely to reflect a diathesis effected by distal developmental factors in conjunction with an externalizing stress-response style. Deliberate self-harm acts might thus be expected to reduce in consequence of addressing the individual's response to stress. By studying DSH in a depressed sample, our study has provided a pattern analysis of those depressed patients who are at risk of DSH, and likely contributing factors that may benefit from consideration in determining a richer treatment plan.
Footnotes
Acknowledgements
We thank the NHMRC (Program grant 222308), the NSW Department of Health, our many consultants and research assistants who assisted data collection and Kerrie Eyers, Chris Boyd and Yvonne Foy for manuscript preparation.