Systematic knowledge regarding the "state of mind" of those who attempt suicide remains elusive. Ratings of thoughts and attitudes around the time of a suicidal act have some power to predict future suicidal behavior +(1–+3), but little is known about the most basic aspects of cognitive processing in suicidal individuals, even though impaired cognitive functioning is common in psychiatric disorders for which suicide risk is elevated +(4, +5).

Suicide attempters have been characterized as "cognitively rigid" on the basis of self-ratings and performance on mental flexibility tasks +(6–+13). From case studies +(14), Rourke et al. +(15) suggested that a specific nonverbal learning disability may predispose individuals to suicidal behavior. Bartfai et al. +(16), using standard neuropsychological measures, found poorer performance on measures of fluency (verbal as well as nonverbal) and reasoning in a small sample of recent suicide attempters compared to chronic pain patients and nonpatients. Ellis et al. +(17), on the other hand, found no neuropsychological differences between diagnostically heterogenous groups of suicidal and nonsuicidal patients.

In this study, neuropsychological performance was assessed in a group of well-characterized, unmedicated subjects who met DSM-III-R criteria for a major depressive episode with or without a history of suicidal behavior. Past attempters were divided into two groups on the basis of the severity of their attempts. Prior research has suggested that those making more severe attempts are a distinct subgroup +(18–+20) who are demographically and biologically more similar to suicide completers +(21–+24). The depressed patient groups were also compared to a nonpatient group.

On the basis of prior studies of neuropsychological functioning in patients with depression +(25–+27), all depressed subjects were expected to perform poorly, relative to nonpatients, on measures of general intellectual functioning (because of deficits on performance IQ-type tasks), motor functioning, attention, and memory. On the basis of these studies and on potential links between suicidal behavior and prefrontal cortical dysfunction +(28, +29), past suicide attempters were expected to perform relatively more poorly on executive functioning measures.

The study group consisted of 50 patients who met DSM-III-R criteria for a current major depressive episode (either part of a major depressive disorder [N=38] or the depressed phase of a bipolar disorder [N=12]) and 22 nonpatients. All patients had a Hamilton Depression Rating Scale (24-item) score ≥16. All nonpatients were free of any current or past axis I or axis II disorder. All subjects were free of neurological disease as determined by clinical history and examination.

A detailed history of past suicidal behavior was obtained for all subjects by using a standard, structured interview +(30). All past suicide attempts were rated for severity based on Beck’s medical damage scale +(31), which quantifies the degree of physical injury resulting from an attempt. Scores range from 0 (no physical damage) to 8 (death). Low-lethality attempts were defined as attempts rated between 0 and 4, and high-lethality attempts were those with ratings of 5–7 (involving, at minimum, coma and intensive care). This definition of high lethality is more stringent than in prior research +(30, +31) but divided the sample into two equivalent-sized groups around the median score.

Diagnoses were determined in patients by using the Structured Clinical Interview for DSM-III-R (SCID) (for axis I and axis II) and were ruled out in nonpatients by using the nonpatient version of the SCID +(32). Other clinical ratings and methods to ascertain reliability have been described previously +(5).

The neuropsychological battery included tests that assessed five general domains: general intellectual functioning, motor functioning, attention, memory, and executive functioning. General intellectual functioning was measured with the Mini-Mental State +(33) and an abbreviated version (i.e., minus the Comprehension and Object Assembly subtests) of the Wechsler Adult Intelligence Scale, Revised (WAIS-R) +(34). Motor functioning was assessed with the Finger Tapping Test +(35) and a computerized choice reaction time task +(36). Attention was assessed with the computerized Continuous Performance Test, Identical Pairs version +(37), a computerized, single-item Stroop task +(38, +39), and the Attention/Concentration subtest of the Wechsler Memory Scale, Revised (WMS-R) +(40). Memory was measured by performance on the WMS-R Verbal and Visual Memory subtests +(40), the Buschke Selective Reminding Test +(41), and the Rey-Osterrieth Complex Figure +(42). Last, executive functioning was assessed with the FAS tests of letter fluency and animal category fluency +(42, +43), the Trail Making Test +(35), Wisconsin Card Sorting Test +(44), and a computerized A not B reaction time task +(45) that measures logical reasoning and working memory.

Patient subjects were recruited by referral from local clinicians. Nonpatients were recruited by advertisement. After providing informed consent, patient subjects underwent a medication washout, generally as inpatients. Suicide risk was a determinant of the need for hospitalization, so past attempters were more likely to be studied as inpatients. Subjects had been free of psychotropic medication for at least 2 weeks before testing. This drug-free period was extended to 4 weeks for those receiving oral neuroleptics and 5 weeks for those receiving fluoxetine. Clinical ratings were conducted at the end of the washout just before neuropsychological testing. Neuropsychological tests were administered in one to three sessions over the course of 1–2 days. All neuropsychological testing was conducted blind to clinical ratings and suicide history.

Groups were first compared on demographic and clinical variables by using analyses of variance (ANOVAs), chi-square analyses, and t tests.

The analyses of neuropsychological test data were carried out in a hierarchical fashion. First, all test scores were converted to z scores normatively corrected according to external normative values (culled from test manuals [+34, +40, +44], a compendium of normative tables +[42], or performance standards from an external normative study group [N=60, age range=20–75]). Domain scores were then computed by averaging the z scores of the primary measure for each test within each domain (global intellectual functioning: full-scale IQ and Mini-Mental State total score; motor functioning: finger tapping [averaged across hands] and choice reaction time; attention: performance on the WMS-R Attention/Concentration subtest, Continuous Performance Test, and Stroop interference task; memory: scores on the WMS-R Verbal and Visual Memory subtests, Buschke Selective Reminding Test total recall, and Rey-Osterrieth Complex Figure recall; executive functioning: average of letter and category fluency totals, Wisconsin Card Sorting Test categories, time score on part B of the Trail Making Test, and A not B reaction time).

Domain scores were entered into a multivariate analysis of variance (MANOVA) comparing four groups. Given a significant main effect of group in the omnibus MANOVA, the groups were compared with univariate ANOVAs and post hoc comparisons (Tukey B procedure).

Domain score analyses were followed by univariate analyses of individual test scores (including measures from each test that were not included in domain computations). To characterize the underlying dimensions that accounted for group differences, scores with significant ANOVAs (one from each task) were submitted to a descriptive discriminant function analysis +(46). This analysis was used to derive a structure matrix that grouped individual tests by their correlations with the empirically derived discriminant functions.

The hierarchical nature of the data analysis provided some protection for multiple comparisons, since univariate analyses were used to characterize multivariate effects or composite score differences. Significance level was set at p<0.05.

The four subject groups were comparable in age, education, and occupational level (+Table 1). The patient groups, in turn, were comparable in diagnostic composition. There were higher proportions of inpatient subjects in the suicide attempter groups, but measures of depression severity (Hamilton depression scale, Beck Depression Inventory) and functional impairment (Global Assessment of Functioning excluding the suicide item) were similar across patient groups.

Within the attempter groups, the primary method of attempt was overdose; violent attempts occurred almost exclusively in the low-lethality group and usually involved superficial cutting. Suicide Intent Scale scores were not different between the attempter groups, although the higher intent scores in patients with high-lethality previous attempts approached significance. Suicide risk measures (e.g., hopelessness, suicide ideation) differed across the groups but were not highest in the group who made the most lethal attempts. Hopelessness scores were higher in nonattempters and low-lethality attempters than in high-lethality attempters, and suicide ideation scores were highest in low-lethality attempters. There tended to be more young (<35 years) patients with concomitant borderline personality disorder in the low-lethality attempter group (35.7% [N=5] versus 9.5% [N=2] in nonattempters and 6.7% [N=1] in high-lethality attempters) (χ²=5.67, df=2, p=0.06). When their presence/absence was included as a factor in analyses, group differences in hopelessness (F=0.57, df=2, 43, p=0.57) and suicidal ideation (F=2.16, df=2, 44, p=0.13) became nonsignificant.

Multivariate analysis of the five neuropsychological domain scores revealed a significant difference among the groups (Wilks’s lambda=0.667, approximate F=1.87, df=15, 177, p<0.03). A plot of the mean domain scores for each group is presented in +Figure 1.

Univariate comparisons of domain scores revealed significant differences among various groups in all domains except motor functioning (+Table 2). Post hoc comparisons revealed that high-lethality attempters performed more poorly than all other groups on tasks of executive functioning and were the only patient group to perform significantly more poorly than nonpatients on tasks of general intellectual functioning, attention, and memory.

Mean scores on individual test measures within domains are presented in +Table 2. Most individual test score differences were found within the memory and executive functioning domains, where high-lethality attempters typically were most impaired.

To characterize the underlying dimensions that differentiated groups, scores with significant ANOVAs (one from each test) were entered into a descriptive discriminant function analysis. Two orthogonal dimensions were extracted. The first explained 53% of the total variance and discriminated high-lethality attempters from all other groups (F=16.7, df=3, 66, p<0.001). The second explained 36% of the variance and discriminated all depressed patient groups from nonpatients (F=11.3, df=3, 66, p<0.001). The tests that correlated with each of these functions are presented in the structure matrix in +Table 3. Tests are grouped by the strength of their correlation with each discriminant function. The function that discriminated high-lethality attempters from all other groups was defined by performance variance on the tasks of FAS letter fluency, Buschke Selective Reminding Test total recall, Wisconsin Card Sorting Test failure to maintain, and A not B reaction time—primarily executive functioning tests. The function that discriminated nonpatients from depressed patients, regardless of suicide history, was defined by performance variance on the Stroop interference task, WMS-R Verbal and Visual Memory scores, Rey-Osterrieth Complex Figure recall, and the WAIS-R Information subtest—a set of primarily attention and memory measures.

Subjects with a history of high-lethality suicide attempts exhibited deficits in executive functioning that were independent of deficits associated with depression alone. These deficits could not be explained by differences in prior education, occupational attainment, or apparent intellectual capacity (e.g., WAIS-R vocabulary scores), which were comparable across groups. Nor were these deficits accounted for by depression severity. In supplementary analyses, findings were maintained when analyses were restricted to nonpatients and inpatient depressed subjects and were not affected by gender distribution. If scores on the Hamilton depression scale or Beck Depression Inventory were covaried in analyses of domain scores, differences in general intellectual functioning, attention, and memory became nonsignificant, but differences in executive functioning remained (with Hamilton depression scale score as covariate: F=4.11, df=1, 67, p=0.01; with Beck Depression Inventory score as covariate: F=3.63, df=1, 67, p=0.02). The independence of these deficits from depression severity was evident in the results of the discriminant analysis as well. The first discriminant function, which was primarily associated with performance on executive functioning tasks, was uncorrelated with depression severity measures (Hamilton depression scale: r_s=0.06, p=0.61; Beck Depression Inventory: r_s=0.07, p=0.55). However, the second discriminant function, associated with performance on attention and memory tasks, was significantly correlated with both (r_s=–0.45, p<0.001, and r_s=–0.41, p<0.001, respectively). Discriminant function analysis maximizes group differences by using available data, but the dimensions extracted, and their association with high-lethality attempt status versus depression alone, are consistent with previous research. Impairments of fluency and logical reasoning have been found in +(16) or suggested by +(6–+13) prior studies of suicide attempters, while impairments of attention and memory are common in depression +(25–+27). The Buschke Selective Reminding Test is a memory task that discriminated high-lethality attempters, but it has stronger encoding/learning demands than other memory measures +(41, +42). Deficits in high-lethality attempters, then, might best be characterized in terms of these subjects’ difficulties on tasks requiring organization and focused effort, a specific realm of executive functioning +(47).

Neuropsychologically, deficits found in high-lethality attempters were consistent with prefrontal cortical dysfunction, even though these subjects were not more impaired on a number of measures commonly used to characterize prefrontal cortical dysfunction in psychiatric disease (e.g., number of categories or perseverative error scores on the Wisconsin Card Sorting Test, Trail Making Test performance). These commonly used measures tend to be more sensitive to dysfunction in superior prefrontal regions +(48), whereas the deficits found in high-lethality attempters—in language fluency, verbally mediated logical reasoning, and verbal learning—are consistent with dysfunction in inferior prefrontal subregions +(49–+51). Even on the Wisconsin Card Sorting Test, the failure to maintain score, which discriminated high-lethality attempters, loads on a separate factor from other Wisconsin Card Sorting Test scores +(52, +53) and assesses a function that is most commonly associated with inferior prefrontal damage +(54). The potential association between high-lethality attempters’ deficits and inferior prefrontal cortical dysfunction is consistent with autoradiographic postmortem studies of suicide completers that have found specific receptor abnormalities in the inferior but not superior prefrontal cortex +(55).

It is possible that brain damage resulting from high-lethality suicide attempts produced additional impairments. High-lethality past attempters in this study showed a modest deficit in general intellectual performance and impairment on a number of tests of mental efficiency and complex mental operations similar to that resulting from diffuse brain damage +(56). However, high-lethality attempters did not perform more poorly than other depressed groups on measures that are typically most sensitive to diffuse brain dysfunction +(42, +56), especially that produced by hypoxia +(57). These include measures of motor speed (finger tapping and choice reaction time) and psychomotor functioning (WAIS-R Digit Symbol subtest and Trail Making Test). Although brain damage effects cannot be ruled out completely, high-lethality attempters’ deficits are not typical of the type of damage resulting from nonspecific cerebral insult.

Given the study group sizes here, statistical power was limited for comparisons of four groups across five domains of functioning. The data enabled us to identify impairment dimensions related to high-lethality attempter status and depression, but larger samples are needed to validate and refine these neuropsychological profiles.

In spite of these limitations, these data suggest that neuropsychological dysfunction may play a role in determining risk for highly lethal suicidal acts. Although one recent study of geriatric patients found that mental status examination scores alone did not predict lethality of suicide attempts +(58), studies of adolescent samples have found a higher incidence of premorbid attention disorders +(59) in those making more lethal attempts as well as reduced communicativeness +(60). Neuropsychological dysfunction may contribute to risk by impairing decision making during acute psychiatric crises or may reflect neurobiological abnormalities associated with risk +(54). At least one study of cognitive performance in remitted past attempters suggests that deficits may resolve with treatment +(61), but to our knowledge there are no studies that used a comprehensive battery of tests to determine if relative patterns of strengths and weaknesses can still be detected following successful treatment of depression. Prospective longitudinal studies are needed to better assess the contribution of neuropsychological dysfunction to suicide risk as well as its interaction with other risk indicators.

Received May 11, 2000; revision received Nov. 20, 2000; accepted Nov. 27, 2000. From the Department of Neuroscience, New York State Psychiatric Institute; and the Department of Psychiatry, Columbia University College of Physicians and Surgeons, New York. Address reprint requests to Dr. Keilp, Box 42, New York State Psychiatric Institute, 1051 Riverside Dr., New York, NY 10032. Supported by a Young Investigator Award from the National Alliance for Research on Schizophrenia and Depression and NIMH grant MH-46745.