There are more than 30,000 suicide victims each year in the United States (R15602BABDIADH). Over 90% of suicide victims have a psychiatric disorder at the time of death (R15602BABBDCGD–R15602BABBEEBE). However, most psychiatric patients do not commit suicide (R15602BABBGGDJ–R15602BABCJCHC. Therefore, a psychiatric disorder is generally a necessary but insufficient condition for suicide. To identify suicide risk factors, it is necessary to look beyond the presence of a major psychiatric syndrome.

A previous suicide attempt is the best predictor of a future suicide or suicide attempt (R15602BABDIBEH). However, since only 20%–30% of those who commit suicide have made a previous suicide attempt, it is necessary to identify other indicators of risk for either a suicide attempt or suicide completion.

Most studies of suicide or suicide attempts are restricted to one domain of possible risk factors, e.g., social (R15602BABCAGJG, R15602BABBEJFD–R15602BABBJADC), psychiatric (R15602BABCBBJI, R15602BABBEEBE, R15602BABDGFDF–R15602BABDICGI), psychological (R15602BABCABDG–R15602BABBGGIC), or familial (R1560241– R1560242). Such studies are too narrowly focused to estimate the relative importance of different types of risk factors or their interrelationship by using multivariate techniques.

A related difficulty is that many studies have ignored the diagnosis of those who commit suicide (R15602BABBJGAB, R15602BABCCCGB, R15602BABDCDGD) or studied only one diagnostic group, such as those with major depression (R15602BABBEEBER15602BABCFFJE, R15602BABDGHIA, R15602BABEAGAB, R15602BABCJCJE–R15602BABDEFJC), schizophrenia (R15602BABCDIHE, R15602BABDJJJJ), borderline personality disorder (R15602BABCBGCH), mania (R15602BABCCACJ), panic disorder (R15602BABBGBCG), or alcoholism (R15602BABEAGIC, R15602BABBJDAA). Such a design cannot determine whether risk factors are specific to a single diagnosis or allow generalizations across diagnostic boundaries. This distinction is critical in the effort to construct a general model of suicidal behavior.

Despite the importance of a psychiatric diagnosis, the quality of diagnostic methods in studies of suicide attempts varies greatly. Studies have used hospital charts, general clinical interviews (R15602BABCCCGB, R15602BABCFFJE, R15602BABDJJJJ–R15602BABEAGAB, R15602BABBIBAG, R15602BABDBEHB, R15602BABCABDG, R15602BABCDIEE), or a structured clinical interview for an axis I diagnosis (R15602BABBEEBE, R15602BABDGFDF, R15602BABCCACJ, R15602BABBGBCG, R15602BABBJDAA, R15602BABCFACJ, R15602BABBIBFG). Few studies included axis II diagnoses (R15602BABBGBCG, R15602BABCBGCH), mostly generated from a clinical interview in combination with research diagnostic criteria such as DSM-III or DSM-III-R (R15602BABCEFIJ, R15602BABDGHIA, R15602BABCGHAF, R15602BABBIBAGR15602BABDBEHB). Studies of completed suicides have similar diagnostic deficiencies (R15602BABBGGIC, R15602BABBCDCE–R15602BABCCHBD).

Accurate diagnosis is important because, like completed suicides, suicide attempts usually occur in the context of a psychiatric illness (R15602BABCCCGB, R15602BABCEFIJ, R15602BABCJFEC, R15602BABBIBAG, R15602BABDBEHB). Moreover, 82% of suicide attempters have comorbid psychiatric diagnoses (R15602BABBEFDI), and this is probably an underestimate because the investigators did not systematically diagnose axis II disorders.

Personality traits, particularly cluster B traits and disorders (R15602BABDBCGF, R15602BABCAAGB), correlate with extent of suicidal behavior (age at first attempt, number of lifetime attempts) (R15602BABDBCGF). Despite the need for evaluation of both axis I and axis II disorders, to our knowledge, no published study of suicide attempts has prospectively employed DSM-III-R or DSM-IV criteria and used structured clinical interview methods for both axis I and axis II disorders in multiple diagnostic groups. Nor have studies combined such assessments with measurement of relevant personality traits, family history, demographics, and psychosocial factors.

The present study involved 347 patients with major affective disorders, psychoses (schizophrenia, schizoaffective disorder, or schizophreniform psychosis), or personality disorders. Our goal was to determine the generalizability and relative importance of risk factors for suicidal acts across diagnostic boundaries. Structured clinical interviews generated axis I and axis II diagnoses, and we measured key personality characteristics such as aggression and impulsivity, recent life events, and a variety of clinical and demographic variables. The goal of this study was to develop a hypothetical, explanatory, and predictive model of suicidal behavior that can subsequently be tested in a prospective study.

Following admission to a university psychiatric hospital for evaluation and psychiatric treatment, 347 consecutive patients were recruited. These patients were 14–72 years old, 51% were male, 68% were Caucasian, and they all had IQs greater than 80. All patients gave written informed consent for the protocol as approved by our institutional review board.

Clinical Assessment and Psychiatric Diagnosis

DSM-III-R axis I psychiatric disorders were diagnosed by using the Structured Clinical Interview for DSM-III-R (SCID) (R15602BABDCDGD). Axis II diagnoses were obtained by using the structured Personality Disorder Examination (R15602BABCBEHJ). Psychiatric symptoms were assessed with the Brief Psychiatric Rating Scale (BPRS) (R15602BABCAGAA), the Scale for the Assessment of Positive Symptoms (SAPS) (R15602BABBJDHJ), the Scale for the Assessment of Negative Symptoms (SANS) (R15602BABDCGGE), the 24-item Hamilton Depression Rating Scale (R15602BABCHBJI), the Beck Depression Inventory (R15602BABDJCHF), and the Beck Hopelessness Scale (R15602BABCIFCD). Lifetime aggression and impulsivity history were rated with the Brown-Goodwin Aggression Inventory (R15602BABCIHGB), Buss-Durkee Hostility Inventory (R15602BABEBBDG), and Barratt Impulsivity Scale (R15602BABDDIFD). Life events were assessed with the St. Paul-Ramsey Scale, and potential protective factors were assessed with the Reasons for Living Inventory (R15602BABBHGJF).

A suicide attempt was defined as a self-destructive act that was sufficiently serious to require medical evaluation and carried out with the intent to end one’s life. The Scale for Suicide Ideation (R15602BABCIFDC) measured suicidal ideation during the week preceding hospitalization, and suicide intent was assessed with the Suicide Intent Scale (R15602BABCJBJE).

Raters were nurses or social workers with a master’s degree or Ph.D. Interrater agreement kappa scores were >0.80 for axis I and >0.65 for axis II diagnoses (R15602BABEACFA). Intraclass coefficients (ICC) were greater than 0.70 for scales such as the Hamilton depression scale (ICC=0.93), BPRS (ICC=0.80), SANS (ICC=0.87), SAPS (ICC=0.71), and Scale for Suicide Ideation (ICC=0.97).

A history of childhood physical or sexual abuse or a head injury with loss of consciousness were recorded as present or absent. Assessment of past alcoholism or substance abuse was based on the SCID. Cigarette smokers were defined by asking if patients had ever smoked cigarettes regularly.

One hundred eighty-four (53%) of the 347 patients had made a previous suicide attempt; 118 (64%) of these patients had made multiple suicide attempts. Attempters and nonattempters were comparable in terms of axis I diagnoses: 176 (51%) of the 347 patients had a major depressive episode (93 attempters and 83 nonattempters); 126 (36%) had schizophrenia, schizoaffective disorder, or schizophreniform disorder (65 attempters and 61 nonattempters); and 45 (13%) had another diagnosis (26 attempters; 19 nonattempters). Seventy-four patients (21%) had comorbid borderline personality disorders. The mean Hamilton depression scale score of the 347 patients was 25.7 (SD=10.2), mean Beck Depression Inventory score was 21.2 (SD=13.1), and mean BPRS score was 43.5 (SD=12.1).

Statistical Methods

Differences between attempters and nonattempters on state and trait variables were tested by the two-sample t test (with equal or unequal variances as appropriate) or the Wilcoxon test for quantitative variables. For two-by-two tables, we used the chi-square statistic with Yates’s correction for continuity. The odds ratio was also determined to show the strength of the relationship between attempter status and dichotomous (yes/no) variables. Interpretation of an odds ratio with a quantitative scale is not possible. For the purpose of data reduction, we performed two principal component factor analyses with no rotation, one for state-dependent rating scales and one for measures of aggression and impulsivity. Factors of the correlation matrix with eigenvalues greater than 1 were retained as significant. Logistic regression models were used to study the multivariate relationships of potential predictors with attempter status and their relative importance. Predictors in these models were selected a priori on the basis of previous research.

Patients who had attempted suicide recently or in the past did not significantly differ from nonattempters in age (mean=32.0, SD=9.5, versus mean=32.7, SD=11.1) (t=–0.65, df=321.1, p=0.52), percentage of males (51% versus 52%), percentage married (14% versus 17%) (χ²=0.40, df=1, p=0.53), percentage Caucasian (66% versus 71%) (χ²=0.71, df=1, p=0.40), height (mean=67.3 cm, SD=4.2, versus mean=67.2 cm, SD=4.1), or number of children (mean=1.2, SD=1.5, versus mean=1.4, SD=1.7) (t=–0.67, df=264, p=0.50). Attempters had significantly fewer years of education (mean=12.7, SD=2.6, versus mean=14.0, SD=3.0) (t=–3.93, df=317, p=0.0001). However, this difference was not substantial enough to be clinically significant. The median personal income in the previous year was low in both groups, but for attempters it was 25% lower than for nonattempters ($6,000 versus $8,000) (χ²=8.03, df=1, p=0.005). Fewer attempters were Catholic (27% versus 41%) (χ²=6.6, df=1, p=0.01).

T1 shows that attempters and nonattempters did not differ significantly on objective severity measures of acute psychopathology, such as major depression (Hamilton depression scale), psychosis (SAPS and SANS), and general psychopathology (BPRS), In contrast, subjective ratings of depression (Beck Depression Inventory), hopelessness (Beck Hopelessness Scale), and severity of suicidal ideation were all significantly greater in suicide attempters. Duration of current episode, another measure of severity of depression, was longer in nonattempters, indicating that attempters had shorter exposure to depression during the current episode. Attempters had more lifetime episodes of major depression or psychosis (median=3 in attempters versus 2 in nonattempters) (χ²=4.18, df=1, p=0.04) and a younger age at onset of illness (median=23 years in attempters versus 27 years in nonattempters) (χ²=13.8, df=1, p=0.0002). Although the level of stress associated with life events (St. Paul-Ramsey Scale) did not appear to differ significantly between the two groups, the suicide attempters reported significantly fewer reasons for living (T1).

T2 addresses traits or stable characteristics and chronic conditions that have been implicated in suicidal behavior. In contrast to the findings in T1, many differences were found between attempters and nonattempters. Attempters had significantly higher scores for lifetime aggression and impulsivity as well as higher rates of comorbid cluster B personality disorders, comorbid past alcoholism or substance abuse or dependence, smoking, head injury, and a family history of a first-degree relative who had attempted or completed suicide. Most of these differences were highly significant, even when Bonferroni’s method for multiple comparisons was applied (T2).

Detailed results of the factor analyses of the state-dependent and trait-dependent psychopathology rating scale scores, including scoring coefficients that indicate how the factor scores were computed from the rating scales, are available on request. Two state factors (total variance explained: 75%) and one trait factor (total variance explained: 64%) were generated. We called the state factors the psychosis factor (derived from BPRS, SAPS, and SANS) and the depression factor (derived from the Hamilton depression scale, Beck Depression Inventory, and Beck Hopelessness Scale) because of correlations with state-dependent measures of these components of psychopathology. We called the trait factor the aggression/impulsivity factor (derived from Brown-Goodwin, Buss-Durkee, and Barratt scales).

A logistic regression model in which suicide attempter status was the dependent variable and the three factors were the independent variables indicated that only the aggression/impulsivity factor was strongly associated with lifetime suicide attempt (odds ratio=3.30, 95% confidence interval=1.99–5.47, p=0.0001). The psychosis factor (odds ratio=1.17, 95% confidence interval=0.78–1.75, p=0.45) and the depression factor (odds ratio=0.98, 95% confidence interval=0.66–1.46, p=0.94) were not significant predictors.

The addition of other independent variables, such as comorbid borderline personality disorder, past head injury, abuse in childhood, and a positive family history for suicidal behavior, did not alter the finding that the two state factors did not appear to predict attempter status. The addition of the score for severity of suicidal ideation to the logistic regression with the three factors did not alter the finding that the aggression/impulsivity factor is an important predictor of attempter status (odds ratio=3.10, 95% confidence interval=1.73–5.57, p=0.0001 for the factor versus odds ratio=1.10, 95% confidence interval=1.05–1.16, p=0.0001 for suicidal ideation).

Because about 41% of attempters had attempted suicide within 30 days of hospitalization, we also ran the logistic regression model including only these recent attempters to address the possibility that, since the psychosis and depression factors reflected current acute state, they were more likely to be associated with suicidal behavior that occurred more recently. The results were unchanged. The aggression/impulsivity factor was a significant predictor of recent attempter status. Suicidal ideation is also an important predictor (according to the logistic regression described earlier) and is correlated with the aggression/impulsivity factor (r=0.28, N=137, p=0.001).

Missing data did not explain the findings of the logistic regression model with the three factors (complete data were required for all items of all nine rating scales for each subject included). We compared subjects included in the multivariate analysis (N=126) with those who were dropped (N=221). No significant differences were found in any demographic or clinical variable except for slightly higher personal income, higher percentage Caucasian (78% versus 62%), and more lethal lifetime suicide attempts (3.8 versus 3.1) (t=2.44, df=181, p=0.02) in the group included in the logistic regression models with the three factors.

A trait factor, aggression/impulsivity, assessing lifetime externally directed aggression and impulsivity, was highly significant in distinguishing past suicide attempters from nonattempters. Individuals with a past history of attempting suicide exhibited greater lifetime aggression and impulsivity than nonattempters with the same psychiatric illness. Hostility in association with major depression and other psychiatric conditions has been reported to be linked to suicidal behavior (R15602BABEAGAB, R15602BABBGEAE) but not defined on the basis of lifetime behavior or stable trait-like features. A more pronounced impulsive-aggressive trait characterizes individuals at risk for suicide attempts regardless of psychiatric diagnosis. Most previous studies could not specifically address this point because they either examined one diagnostic group (R15602BABBEEBE, R15602BABEAGIC, R15602BABCCACJR15602BABCBGCH, R15602BABDGHIA, R15602BABEAGAB, R15602BABCJCJER15602BABDEFJC, R15602BABCGCAA, R15602BABBCHGA), had too few cases to separate diagnostic groups (R15602BABBGEAE), or did not specify diagnoses (R15602BABBJGAB, R15602BABCCCGB, R15602BABBICEA).

Externally directed aggression, suicidal acts, and other manifestations of impulsivity are all highly related (R15602BABBGEAE, R15602BABBECEB, R15602BABBBBCJ). Our scales cannot separate impulsivity from its manifestations, such as aggression or borderline personality disorder. For example, the aggression/impulsivity factor comprises scores from the aggression and impulsivity scales, features also found in borderline personality disorder. A diagnosis of borderline personality disorder and the aggression/impulsivity factor were both robust predictors of attempter status but were strongly interdependent. Patients with a diagnosis of borderline personality disorder scored much higher on the aggression/impulsivity factor than those without borderline personality disorder (0.54 versus –0.16) (t=3.09, df=83, p=0.003). The component of borderline personality disorder that best correlated with suicidal behavior was impulsivity, highlighting the importance of impulsivity in determining risk of suicidal acts (R15602BABDCDGD). Greater impulsivity or impaired decision making may underlie a generalized propensity to suicidal and aggressive acts. Impulsivity is measurable by neuropsychological tests that ultimately may prove to be more sensitive than a clinical history.

Clinicians intuitively rely on objective measures of severity of psychiatric illness as a guide to the risk for suicidal acts. Our findings and those of others indicate that objective severity of illness does not distinguish patients with a history of suicide attempts (R15602BABCDIEER15602BABBBBCJR15602BABCDBAG). Given the cumulative evidence that a history of suicide attempt elevates the risk for future suicidal acts (R15602BABDIBEH), objective severity of illness is also unlikely to predict future suicide attempts. This observation emphasizes the importance of the diathesis or trait-like predisposition, relative to severity of the illness, in predicting suicidal acts.

Patients with recurrent major depression or schizophrenia, who have a history of suicide attempt(s), make the attempt(s) relatively early in the course of illness (R15602BABCDIHE, R15602BABBBBCJ). In our study, the attempters had made an average of 2.5 lifetime suicide attempts, evidence that suicide attempters have a predisposition to suicidal acts. The predisposition to suicidal acts appears to be part of a more fundamental predisposition to both externally and self-directed aggression.

The model shown in F1 summarizes our results. Subjective depression, hopelessness, and suicidal ideation were greater in suicide attempters than in nonattempters, despite comparable rates of objective severity for depression or psychosis. Conversely, suicide attempters scored lower on the Reasons for Living Inventory—a scale that has been considered to measure the protective effect of having more reasons for living (R15602BABBHGJF). The Reasons for Living Inventory score correlated negatively with hopelessness, providing further support for the notion that the Reasons for Living Inventory is an index of deterrents to suicide. There was no significant difference in life events between attempters and nonattempters; therefore, life events do not appear to explain the perception of fewer reasons for living. The more pronounced adverse response of suicide attempters to depression or psychosis implies some degree of state-trait interaction (F1). Hopelessness is greater in suicide attempters than in nonattempters during an acute depression (R15602BABBBBCJ), after successful treatment (R15602BABDGICE, and intermorbidly (R15602BABCIDHG). Hopelessness can predict future suicide (R15602BABCDIEE, R15602BABCIDHG, R15602BABCCABD). These observations suggest that the degree of hopelessness is determined by both state and trait, and it may have predictive properties.

Genetic or familial factors contribute to suicide risk (R15602BABBEDBG, R15602BABDBGDJ, R15602BABCFEGD). Aggression, impulsivity, and borderline personality disorder may also be results of genetic factors (R15602BABCEDEH–R15602BABBHGCJ) or early life experiences, including a history of physical or sexual abuse (R15602BABCDIJE, R15602BABCJEFD). A common underlying genetic or familial factor, therefore, may explain the association of suicidal behavior with the aggression/impulsivity factor or borderline personality disorder. An indicator of genetic or familial risk is a history of suicidal behavior in a first-degree relative (R15602BABBEDBG). As evidence of this association, a higher rate of a positive family history of a suicide attempt or completion was associated with the presence of comorbid cluster B personality disorder (χ²=5.08, df=1, p=0.02). This hypothesis of familial (perhaps genetic) transmission of a propensity for externally directed aggression and suicidal behavior, independent of transmission of major depression or psychosis, is consistent with evidence that suicidal behavior is transmitted in families independently of psychopathology but not independently of impulsive aggression(R15602BABBJHJB); it is also consistent with adoption studies reporting that transmission of a genetic risk factor for suicide that is independent of psychopathology (R15602BABBBCJG).

Suicide risk was also related to past head injury, as previously reported by others (R15602BABBIBCJ–R15602BABBGCHB), and to a history of abuse in childhood (R15602BABCDIJE, R15602BABCJEFD) (T2). Both head injury and abuse in childhood were independent predictors of suicide status, and both were associated with the aggression/impulsivity factor. Higher scores on the aggression/impulsivity factor were found in patients with a past head injury (0.28 versus –0.19) (t=2.64, df=131, p=0.009) and in those with a history of abuse in childhood (0.33 versus –0.28) (t=3.20, df=106, p=0.002). In terms of causality, aggressive, impulsive children and adults are more likely to sustain a head injury, and head injuries can cause disinhibition and aggressive behavior (see reference R15602BABDHDDC for review). Alternatively, mothers of abused children have an elevated suicide attempt rate and may transmit the risk for suicidal behavior both genetically and by upbringing. Aggressive, impulsive children may provoke child abuse in vulnerable parents and thereby create a tendency for self-injurious behavior in adult life (R15602BABCDIJE, R15602BABCJEFD).

Alcoholism and substance abuse are associated with suicidal acts and may be causal factors by increasing the probability of a head injury due to acute intoxication. Fifty percent of head injuries, which result in disinhibition and a greater probability of suicidal behavior (R15602BABCCBAJ, R15602BABBFHHJ), are sustained while using alcohol (R15602BABECFAH). Alcoholism and substance abuse are related to the aggression/impulsivity factor and comorbid borderline personality disorder (which, in turn, can increase the risk for head injury as well as be aggravated by a head injury). Comorbid past alcoholism or substance abuse are associated with head injury (χ²=11.96, df=1, p=0.001), with borderline personality disorder (χ²=7.57, df=1, p=0.006), and with higher scores on the aggression/impulsivity factor (0.26 versus –0.44) (t=4.42, df=141, p<0.0001). This constellation of aggression, alcoholism, substance abuse, and impulsivity resembles a syndrome called "disinhibitory" psychopathology described by Gorenstein and Newman (R15602BABDBDDC) in a review of relevant clinical and animal studies of brain injury written almost 20 years ago.

Given the evidence linking low serotonergic activity separately to suicidal behavior (R15602BABCGGJA–R15602BABDGJJI), aggression, and alcoholism (see reference R15602BABBECEB for a review), it is conceivable that low serotonergic activity may, to some degree, underlie all three problems and that low serotonergic activity may mediate genetic and developmental effects on suicide, aggression, and alcoholism (R15602BABBFAJD). Reports of increased aggression, impulsivity, and cocaine and alcohol consumption in mutant mice lacking the 5-HT_1B receptor (R15602BABDIIGE, R15602BABDBGJC) indicate that a genetically mediated serotonin abnormality can result in increased impulsive aggression as well as alcoholism and substance abuse. Suicidal behavior is never observed in a rodent model, but the other elements of "disinhibitory" psychopathology are observed in this 5-HT_1B knockout mouse model.

Potentially related to the higher rate of substance abuse and alcoholism in the suicide attempters is our observation that cigarette smoking is more common in suicide attempters. Smoking is associated with elevated rates of suicide (R15602BABBDIAH–R15602BABDGBDD), and our study demonstrates that the association of smoking and suicidal acts is independent of any association of psychiatric illness with smoking. While this manuscript was under review, another study was published reporting an association of smoking and suicide attempts in patients with a variety of psychiatric diagnoses (R15602BABDCDDC).

The coexistence of a greater propensity for suicidal ideation and for impulsive behaviors in suicide attempters can partly explain why suicide attempters are prone to attempt suicide. They feel more suicidal and are more likely to act on feelings. These results suggest that clinicians should carefully assess severity of suicidal ideation and lifetime impulsivity or aggression in order to estimate the risk for future suicidal acts. This study, like most previous studies, evaluated associations with past suicidal acts. A prospective study is underway to determine correlates of future suicidal acts in the same patient population as part of an effort to develop a predictive model of suicidal acts.

Received Sept. 16, 1997; revisions received March 31 and May 26, 1998; accepted July 14, 1998. From the Mental Health Clinical Research Center for the Study of Suicidal Behavior, Department of Neuroscience, New York State Psychiatric Institute; and the Department of Psychiatry, Western Psychiatric Institute and Clinic, University of Pittsburgh Medical Center, Pittsburgh. Address reprint requests to Dr. Mann, Department of Neuroscience, New York State Psychiatric Institute, 1051 Riverside Dr., Box 42, New York, NY 10032; jjm@columbia.edu (e-mail). Supported by NIMH grants MH-46745, MH-48514, and MH-48492 and by the Audrey Wallace Otto Fund of the Saint Louis Community Foundation and the Diane Goldberg Foundation. Patient assessments were performed by Donna Abbondanza, Tom Kelly, Diane Dolata, and Elizabeth Radomsky. Statistical help and advice was provided by John A. Sweeney, Steven Ellis, and Xunhua Liu. Nancy Geibel expertly typed the manuscript.