|Home | About | Journals | Submit | Contact Us | Français|
Research into personality factors related to suicidality suggests substantial variability among suicide attempters. A potentially useful approach that accounts for this complexity is personality subtyping. As part of a large sample looking at personality pathology, this study used Q-factor analysis to identify subtypes of 311 adult suicide attempters using SWAP-II personality profiles. Identified subtypes included Internalizing, Emotionally Dysregulated, Dependent, Hostile-Isolated, Psychopathic, and Anxious-Somatizing. Subtypes differed in hypothesized ways on criterion variables that address their construct validity, including adaptive functioning, Axis I and II comorbidity, and etiology-related variables (e.g., history of abuse). Furthermore, dimensional ratings of the subtypes predicted adaptive functioning above DSM-based diagnoses and symptoms.
The World Health Organization recognizes suicide as one of the world’s leading causes of death (DeLeo et al., 2002). In the United States, data from the National Comorbidity Survey indicate 4.6% of individuals have a history of suicide attempts (Nock and Kessler, 2006). Factors associated with suicide are numerous. The Centers for Disease Control and Prevention (CDC, 2006) found that approximately half of individuals who died by suicide in 2004 had at least one diagnosed mental illness, the most common being major depression (85.2% of cases), bipolar disorder (7.4%), and schizophrenia (3.3%). Additionally, males are over four times as likely to complete suicide (CDC, 2006) whereas females are more likely to attempt suicide (Nock and Kessler, 2006). Questions regarding lethality of method and what actually constitutes a suicide attempt further complicate the challenge to identify prevalence and risk for suicide, although recently there have been efforts to standardize the nomenclature to aid such research (Silverman et al., 2007a; Silverman et al., 2007b). To date, the majority of studies on risk for suicide have focused on Axis I diagnoses and demographics with relatively few focusing on Axis II disorders and personality. Thus, understanding personality as it relates to suicidality will contribute to our knowledge and treatment of suicidal behavior. Specially, this paper focuses on the potential contribution of personality subtyping of suicidal individuals. Before describing the goals of the current study, we briefly review extant relevant personality studies.
In terms of ‘normal’ personality, personality traits associated with suicidality include a positive association with Neuroticism (especially its depressive and anxious facets) and a negative association with Extraversion (especially its Assertiveness and Positive Emotions facets) (e.g., Chioqueta and Stiles, 2005; Lester, 1987; Velting, 1999). A recent study, not using traditional Five-Factor Model measures, also found relationships between social introversion, irritable temperament, and suicide risk (Pompili et al., 2008). Personality disorders, particularly Cluster B (emotional/erratic) disorders (e.g., Pompili et al., 2004), though, are often more clearly associated with risk for suicide attempts and completion (i.e., general suicidality) because of their more extreme presentations. In particular, some studies suggest hostility, a common trait of Cluster B disorders, is positively related to suicide attempts (Brittlebank et al., 1989; Farmer and Creed, 1989; Weissman et al., 1973). Other studies find small positive correlations exist between antisocial behaviors of psychopathy and suicide (Douglas et al., 2006).
The personality disorder (PD) most highly associated with suicidality, as well as impulsiveness, aggression, depression, and affective dysregulation, is Borderline PD (e.g., Critchfield et al., 2004; Soloff et al., 1994). Because suicidal behavior is one DSM-IV criterion of Borderline PD (APA, 2000), that 60–70% of borderline patients have attempted suicide at least once is not surprising (Gunderson, 2001), although these suicidal gestures are often viewed as attempts to elicit responses from others rather than genuine attempts (Kernberg, 2001). Still not all those with BPD attempt suicide, and within the diagnosis of BPD many varying symptoms patterns exist (Paris, 2007), some of which may theoretically be more related to suicidal behaviors (Oldham, 2006). Taken together, these equivocal findings relating suicidality to depressive, impulsive, psychopathic, and borderline personality characteristics suggest substantive diversity in suicidal populations.
To the extent that empirically derived, validated, and clinically meaningful subgroups of persons with suicidal behaviors can be identified, we may be able to develop a better understanding of risk factors and interventions based on different subtypes. Past research has utilized cluster analyses on a variety of characteristics including self-reported symptoms (Steer et al., 1993), seriousness of attempts (Kurz et al., 1987), and self-reported reasons for attempts (Colson, 1973). These studies illuminate advantages of subtyping suicidal patients but may benefit from supplementation with alternative subtyping strategies.
Given the importance of personality as a context for understanding psychopathology (Krueger and Tackett, 2006; Westen et al., 2006a), subtyping patients based on their personalities may be a particularly useful strategy for studying patients exhibiting suicidality. Personality subtyping of Axis I psychopathology has led to useful conceptualizations of a variety of Axis I disorders (e.g., alcoholism, Dush and Keen, 1995; adolescent and adult eating disorders, Thompson-Brenner et al., 2008; Thompson-Brenner and Westen, 2005). One of the most helpful frameworks to understand personality subtypes is the distinction by Krueger and colleagues (Blonigen et al., 2005; Kramer et al., 2008; Krueger and Tackett, 2006) between internalizing and externalizing disorders and presentations. Suicidal individuals clearly span the gamut of internalizing-externalizing presentations, but whether this theoretical model provides enough specificity for subtyping is yet to be tested.
Using trait vulnerabilities to psychopathology as measured by the Karolinska Scales of Personality (KSP; Schalling et al., 1987), Engström et al. (1996) identified six clusters in their Swedish sample of hospitalized suicide attempters. Two of the clusters were “close to normal,” but others included “neurotic and introverted,” suspicious and guilty neurotic, impulsive with high psychoticism, and the extreme group with high anxiety, aggressiveness, suspiciousness, and impulsiveness (p. 690). Engström and colleagues (1997) reanalyzed the same data, reduced the number of clusters from six to four, and compared them on biological measures (e.g., cortisol, MAO in platelets). The revised subtypes included two close-to-normal subtypes, neurotic and introverted, and the extreme group. Higher serotonin levels were present in one close-to-normal group after a dexamethasone test, which the authors interpreted as indicating a biologically based dysfunction, not a temperamental one.
Another pair of studies (Ellis et al., 1996; Rudd et al., 2000) used the Millon Clinical Multiaxial Inventory (MCMI, MCMI-II) to cluster analyze suicidal behaviors. Ellis and colleagues used a sample of young army enlistees who had attempted suicide (60.6%) or had suicide ideation (39.4%) and the MCMI to identify four clusters that correspond roughly to Schizoid, Antisocial, Dependent, and Histrionic-Narcissistic subtypes. The Schizoid and Dependent subgroups had more comorbid diagnoses, while the Dependent subtype had the greatest suicide risk, highest hopelessness, largest problem-solving deficits, and greatest borderline features. Subsequently, Rudd and colleagues, using MCMI-II data from psychiatric patients hospitalized after a suicide attempt, found three clusters: Negativistic-Avoidant with schizoid and borderline features, Dependent/Self-defeating, and Antisocial with borderline features. Again, the Dependent group exhibited the most severe symptoms, even though their borderline features were subclinical. These studies suggest personality subtypes of suicidal patients may exist in a variety of samples; however, they are clinically non-intuitive, and replicability is unclear, as has been the case with cluster-analytic studies of many disorders and psychiatric phenomena such as suicide (Blashfield and Morey, 1979).
The present study seeks to expand past research in personality subtyping by identifying subtypes of individuals with a history of attempting suicide. Two primary methodological factors distinguish this study from prior studies in this area. First, we are using data derived from quantified clinician psychometric judgments about patients’ personalities. Prior studies have focused exclusively on self-report measures, which may or may not detect important personality features in patients who have limited insight or awareness of their own behaviors and personality patterns. Secondly, this study presents results based on Q-factor analysis, in place of cluster analysis, to identify the existence of subtypes. Q-factor analysis is mathematically identical to exploratory factor analysis except it groups together similar individual personality profiles, not traits or items (Block, 1978). The advantage is that it does not assume mutually exclusive taxa. Like conventional factor analysis, it identifies dimensions (in this case constellational dimensions1; see Westen et al., 2008c). After identifying a potentially meaningful typology, we used correlational analyses to compare subtypes on criterion variables such as Axis I and II comorbidity, etiology-related variables (e.g., history of abuse), and attempt lethality to assess construct validity.
Given the theoretical background and past research described above, we hypothesized three broad subtypes: internalizing (i.e., depressive and anxious neurotic), externalizing (i.e., antisocial and substance abusers), and emotionally dysregulated (i.e., borderline psychopathology). Nevertheless, we considered this study exploratory and remained open to the possibility of additional or more specific subtypes.
We contacted a national sample (unstratified) of psychiatrists and psychologists with at least 5 years experience post-training from the membership registers of the American Psychiatric and American Psychological Associations. Participating clinicians received a $200 consulting fee.
We asked clinicians to describe “an adult patient you are currently treating or evaluating who has enduring patterns of thoughts, feeling, motivation or behavior—that is, personality patterns—that cause distress or dysfunction.” To obtain a broad range of personality pathology, we emphasized that patients need not have a DSM-IV PD diagnosis. Patients had to meet the following additional inclusion criteria: ≥ 18 years of age, not currently psychotic, and known well by the clinician (using the guideline of ≥ 6 clinical contact hours but ≤ 2 years). To ensure random selection of patient from clinicians’ practiced, we instructed clinicians to consult their calendars to select the last patient they saw during the previous week who met study criteria. In a subsequent follow-up, over 95% of clinicians reported following the procedures as instructed. Each clinician contributed data on one patient.
From the larger group of 1201, we used the item “Has the patient ever attempted suicide?” from the Clinical Data Form (see below) of the larger questionnaire packet to select only those individuals who had attempted suicide. The clinicians who provided data on 311 patients (25.9% of the larger sample) with a history of suicide attempt(s) were 54.7% female and had an average of 17.7 years of clinical experience post-formal training (SD = 7.8). The patients with a history of suicide attempt had a mean age of 40.4 (SD = 11.8) and were 66.9% female. Caucasians made up 80.1% of the patient sample, followed by African Americans (8.4%), Hispanics (7.1%), and Asians (1.9%). Socioeconomic status of the patient’s family of origin showed some skew with 43.5% being in the poor and working class, 36.5% middle class, and 20.0% upper middle or upper class; accordingly, 64.4% had less than a college degree.
The CDF is a clinician-report form that gathers information on a wide range of demographic, diagnostic, and etiological variables (Westen and Shedler, 1999a; Westen et al., 2003). We used principal components analyses to aggregate variables (i.e., global adaptive functioning, externalizing behavior, victimization, employment, childhood trauma, and childhood psychopathy) for maximum reliability and non-redundant analyses. For the global adaptive functioning variable, for example, we standardized and averaged together each patients’ DSM-IV Global Adaptive Functioning score, level of personality disturbance (1 = severe personality disorder to 5 = high functioning), employment stability (1 = unable to keep job to 5 = working to potential), quality of romantic relationships and friendships (both scored from 1 = very poor to 5 = close and/or loving), and number of confidants (0 = none to 4 = many). Also within the CDF, suicide attempt lethality was rated on a 5-point scale with anchors at 1 (symbolic gesture), 3 (medical attention was required), and 5 (life-threatening). Other important criterion variables derived from the CDF included Externalizing (history of arrest, violent crime, and abuse perpetration), Victimization (frequency of rape and abuse in adulthood), Employment (work stability, job loss due to interpersonal reasons), and Childhood Trauma (physical and sexual abuse frequency and severity) composites.
The SWAP-II is the latest version of the SWAP instrument (Shedler and Westen, 2004a, 2004b, 2007; Westen and Shedler, 1999a, 1999b). The SWAP-II consists of 200 personality-descriptive statements, each of which may describe a given patient well, somewhat, or not at all. Clinicians sort the statements into eight categories, from least descriptive of the patient (assigned a value of 0) to most descriptive (assigned a value of 7), according to a fixed distribution (Block, 1978). A web-based version of the instrument can be viewed at www.SWAPassessment.org. An increasing body of research supports the validity and reliability of the adult SWAP in predicting a wide range of external criteria, such as suicide attempts, history of psychiatric hospitalizations, ratings of adaptive functioning, interview diagnoses, and developmental and family history variables (e.g., Westen and Muderrisoglu, 2003; Westen and Shedler, 1999a; Westen and Weinberger, 2004; Westen et al., 2003).
Within the CDF, we assessed Axis I psychopathology by a simple checklist of possible disorders coded as present/absent according to DSM-IV criteria. Because the focus of the larger study was to assess Axis II disorders, we collected more detailed data on PD diagnoses. Clinicians received a randomly ordered checklist of the criteria for all Axis II disorders and checked which criteria the patient met. To generate DSM-IV diagnoses, we applied the DSM-IV diagnostic decision rules. To generate DSM-IV dimensional diagnoses that mirror those widely used in the PD literature, we summed the number of criteria judged present for each disorder. This method tends to produce results that mirror those of structured interviews (Blais and Norman, 1997; Morey, 1988; Westen et al., 2003; Zittel and Westen, 2005).
Based on descriptive coherence, we retained six Q-factors from the seven-factor Unweighted Least Squares extraction that accounted for roughly 44% of the total variance (see Table 1). Because these Q-factors are meant as prototypes that patients can match to varying degrees, we chose an oblique rotation, but to minimize intercorrelations among the factors, we used a Promax rotation with the Kappa set at 2, which imposes limits on how oblique the solutions are likely to be (rendering the prototypes similar in some respects to ideal types). Importantly, the interpretation of the factors remained quite similar no matter the rotation chosen. Table 1 displays the top 10 SWAP-II items most descriptive for each subtype. Factor scores are in terms of standard deviations above the average descriptive capacity of other SWAP items for that subtype. We derived the hypothesized subtypes of Internalizing (20.4% initial explained variance; 43.6% explained after rotation), Emotionally Dysregulated (9.2%; 39.4%), and Psychopathic (2.2%; 12.1%) as well as three additional subtypes, Dependent (3.7%; 16.6%), Hostile-Isolated (2.5%; 13.0%), and Anxious-Somatizing (1.8%; 10.2%). Correlations among the latent Q-factors ranged from −.19 to .14.
To validate subtypes, we compared the groups on several external criteria based on variables that should distinguish a valid taxonomic distinction (Livesley and Jackson, 1992; Robins and Guze, 1970; Westen et al., 2006b). Table 2 displays the correlations among the degree of match to each subtype (i.e., Q-correlations) and important criterion variables.
As would be expected based on prior research on related Axis II disorders, the Internalizing and Dependent subtypes were more associated with women whereas the Hostile-Isolated and Psychopathic subtypes were more likely men.
As expected, depression was associated more with Internalizing and less with Psychopathic; GAD was correlated positively with Anxious-Somatizing; and substance use correlated positively with Emotionally Dysregulated, Dependent, and Psychopathic. Correlations were strong and in expected directions for Axis II symptoms, providing strong support for interpretation of the subtype descriptions.
Looking at both a global adaptive functioning variable (using aggregation to increase reliability) as well as more specific adaptive functioning variables (e.g., employment, victimization), the overall correlation pattern suggests the Emotionally Dysregulated and Psychopathic groups were most associated with lower levels of adaptive functioning.
Because traumatic childhood experiences are often suggested as risk factors for development of suicidal behavior, we compared subtypes on a global indicator of childhood trauma, which included the frequency and severity of both physical and sexual abuse. The childhood abuse composite variable was associated negatively with the Hostile-Isolated, Dependent, and Anxious-Somatizing groups and positively with the Emotionally Dysregulated group. Lastly, we looked at the relationship between our aggregated variable reflecting childhood psychopathic behaviors. As expected from the subtype description, the Psychopathic subtype was highly correlated with this variable.
Only the negative correlation between Anxious-Somatizing and lethality was significant. Self-mutilation was correlated positively with Internalizing and Emotionally Dysregulated groups, but the severity of the mutilation was only positively correlated with the Emotionally Dysregulated subtype.
To examine the question of the utility of personality subtyping as compared to related DSM-based Axis I diagnoses and Axis II symptoms, we conducted hierarchical multiple regressions with dimensional measures of the identified subtypes (i.e., Q-correlations). Because it is essential for diagnosis and classification to predict patients’ overall functioning, we chose the aggregated adaptive functioning composite as the primary criterion variable. To limit the number of predictors, we included only Axis I disorders most often related to suicidality: Major Depressive Disorder, Bipolar I/II, and Substance Abuse Disorder. For Axis II dimensional diagnoses, we included disorders that correspond closest to the subtypes (i.e., Antisocial, Borderline, Dependent) plus the remaining Cluster B disorders because past research has shown the emotional, dramatic cluster to be most associated with suicide (e.g., Pompili et al., 2004).
Table 4 displays results of the hierarchical regression with Axis I in the first step, Axis II in the second, and the subtypes last. Although both Axis I and II disorders significantly predicted 14.8% of the variability in adaptive functioning, including the personality subtypes accounted for an additional 18.5%. At the final step, only degree of match with Internalizing (β = −.51), Emotionally Dysregulated (β = −.26), and Psychopathic (β = −.48) subtypes uniquely predicted variance in adaptive functioning.
As a follow-up analysis to this we also conducted a hierarchical regression with subtypes added first, Axis I second, and Axis II last. Results indicated that subtypes accounted for 31.1% of the variance in adaptive functioning, and neither Axis II (Fchange (5, 297) = 1.14, pchange > .05,ΔR2 = .013) nor Axis I (Fchange (3, 294) = .24, pchange > .05, ΔR2 = .002) contributed significant additional variance. This finding indicates these personality subtypes contribute not only unique variance but also account for the variance contributed by the current DSM.
A final test is to examine the extent to which degree of match to each subtype (dimensional scores) can predict whether an individual has had at least one past self-reported suicide attempt. We used a logistic regression to predict negative or positive suicide history by degree of match to all six personality subtypes for the entire sample (N = 1201). Together the subtypes significantly explained the equivalent of 32% of the variance in suicide attempt history (see Table 4; for discussion of pseudo-R2 indices, see Cohen et al., 2003). The strongest predictors of a suicide attempt history were the Emotionally Dysregulated subtype dimension (OR = 73.87) followed by the Internalizing (OR = 35.99). The Hostile-Isolated (OR = .02) and Anxious-Somatizing (OR = .13) subtypes were negatively predictive of past suicide attempt(s) after holding constant the other dimensions. These results show that the two largest subtypes are predictive of a suicide attempt history, but the less common subtypes do not add substantially to prediction, with two important exceptions. The Hostile-Isolated and Anxious-Somatizing subtypes seem to be more commonly negatively associated with suicide attempts.
Because of variability in risk factors and personality characteristics associated with suicide, subtyping can provide greater specificity and organization of findings than simple comparisons of suicidal to non-suicidal individuals. In this study, Q-factor analysis identified six personality subtypes of individuals with a suicide attempt history. As hypothesized, we identified subtypes reflecting the continuum of internalizing to externalizing personality styles found in many other classification systems for both Axis I symptoms and personality variables (Krueger and Tackett, 2006). First, we found an Internalizing subtype (the largest group in our sample) that reflected chronically depressed, anhedonic, and passive personality features. Externalizers were well represented in the Psychopathic subtype, displaying substance abuse problems, criminality, imperviousness to consequences, and lack of remorse. In addition, difficulty with emotion regulation emerged central to one of the subtypes. The Emotionally Dysregulated subtype, seemingly a close relative of Borderline PD, mixes elements of internalizing spectrum symptoms with externalizing characteristics.
Three additional subtypes, albeit not entirely without precedence in past suicide research, described individuals who do not easily fit into the internalizing-externalizing distinction. First, the Dependent subtype contained individuals who fantasize about ideal love, often become attached too quickly, and fear rejection. Conversely, a Hostile-Isolated subtype contained social outsiders who lack relationships, are critical towards others, and are competitive. Finally, an Anxious-Somatizing subtype captures a hypochondriacal patient who uses their mental and medical problems to avoid responsibility. Unsurprisingly, they often catastrophize and develop somatic complaints when under stress. These last two subtypes shared the distinction of being more commonly associated with less likelihood of having a suicide attempt history.
Comparing current findings to past efforts at personality subtyping provides clues to how replicable these groups are. The Internalizing subtype embodies general negative affect present in many subtypes in previous studies (e.g., guilty neurotic, neurotic introverts; Engström et al., 1996). Likewise, the Emotionally Dysregulated subtype has appeared most often in the form of Borderline PD symptoms (Ellis et al., 1996; Engström et al., 1996; Rudd et al., 2000). The Dependent subtype also had clear counterparts in both Ellis et al. and Rudd et al. Additionally, the Psychopathic subtype overlaps with Ellis et al.’s Antisocial subtype, although our Psychopathic group appeared less impulsive and aggressive and more classically nonchalant and uncaring towards harming others.
Two subtypes matched less clearly with previous findings. Nevertheless, some of their characteristics are evident in past research. First, the Hostile-Isolated subtype had some overlap with Ellis et al.’s Schizoid group in their social isolation and angry/resentful attitude and with Rudd et al.’s Antisocial subtype in their competitiveness, criticality, and self-reliance. Further research may clarify distinctions, if any, among these groups. The Anxious-Somatizing subtype only shared similarities with the anxiety of the neurotic and introverted subtype (Engström et al., 1997; Engström et al., 1996). Somatization itself, though, has not been found in past subtyping efforts, possibly due to differences in measures or samples. Of the subtypes found previously, only one did not have a counterpart in the present study: the Histrionic-Narcissistic subtype (Ellis et al., 1996). Overall, though, these subtypes show relatively strong convergence with past personality subtypes of suicidal adults.
Subtypes differed in meaningful and predictable ways on multiple measures not used in their derivation, including Axis I and II diagnoses, adaptive functioning, and childhood experiences. Furthermore, hierarchical regressions showed their dimensional form’s ability to predict adaptive functioning outperforms categorical Axis I and dimensional Axis II disorders, an impressive feature for these empirically derived personality subtypes. Thus, these subtypes gain the most utility when viewed as a set of dimensional prototypes to which any individual can match to varying degrees, resulting in an MMPI-like profile of elevations.
This study has several methodological limitations. First, this study was cross-sectional of suicide attempters. Second, details about suicide attempts were quite limited in that no information about attempt frequency and time of the last attempt was available. Third, patients included in this study were in treatment and may not represent the entire population of suicide attempters. Nevertheless, because this sample has a wide variety of clinical settings represented, the sample should include most kinds of suicidal patients seen in treatment settings. Fourth, one informant provided all data. Ideally, patients would have been diagnosed by one observer and assessed using the SWAP-II by another. However, most published PD studies also rely on a single informant—the patient (e.g., via self-report questionnaires or responses to structured interview questions). The quantified observations of experienced psychiatrists and clinical psychologists (with an average of 17.7 years practice experience) are no less credible than self-descriptions of patients with personality pathology. SWAP data are also unrelated to theoretical orientation or other aspects of professional training (Betan et al., 2005; Bradley et al., 2005). If clinician biases exist, they appear to account for little variance in SWAP ratings. Finally, absence of psychosis and problems with personality were criteria for inclusion in the overarching study and may have affected the heterogeneity of the sample. The possibility of a psychotic subtype and/or a relatively healthy subtype exists. Future research in this area may illuminate the actual effects of these limitations.
This study of personality and suicidality has opened up several avenues of future research. One important area of future investigation is the exploration of additional, sex-specific subtypes, considering recent research has found different risk factors for suicidal acts in males and females (e.g., Oquendo et al., 2007). Another possibility is that different personality subtypes exist in diverse ethnic groups. Although our sample had diverse ethnic representation, more diversity could lead to additional findings. Finding similarities and differences by sex and ethnicity will increase the much-needed specificity in identifying individuals at risk for suicide.
In sum, this study showed that personality subtypes of suicidal individuals do exist and are somewhat replicable of previous research. Subtypes differed on applicable variables and provided more predictive capabilities of adaptive functioning than DSM-based diagnoses. Two subtypes appeared especially related to suicide attempt history in clinical populations. Key clinical features for the Internalizing subtype included greater likelihood of being female, depressed, and avoidant whereas the Emotionally Dysregulated subtype was associated more with externalizing behaviors, less overall adaptive functioning, a traumatic childhood, and Borderline PD symptoms. Importantly, both subtypes were associated with a history of self-mutilation. Future research and experience will demonstrate the clinical utility in attending to personality subtypes in the assessment of suicidality, but if these findings are any indication, one may gain much over a simple list of one-size-fits-all risk factors.
Research presented in this article was funded by NIMH grant MH-62377 to Drew Westen and an American Suicide Prevention Foundation Young Investigator Grant to Rebekah Bradley
1Westen D, Waller N, Blagov P, Shedler J, Bradley R (2008) Measuring pathological personality traits by clinician-reports using the SWAP-II: Factor structure, validity, and retest reliability. Unpublished manuscript, submitted for publication.