Suicide is a major public health concern in most Western countries. Unfortunately, current prevention strategies are based on screening for numerous risk factors, none of which has been shown to have sufficient predictive power (1). A stress-diathesis model was recently constructed (2) to model and expand knowledge concerning the pathophysiology of suicidal behavior. According to this model, only persons with a susceptibility (diathesis) to suicidal behavior are at risk of attempting to take their own lives after exposure to stress.
Many studies have provided evidence that the serotonergic system is involved in this susceptibility. Indeed, several studies of suicide attempters or completers have identified specific serotonergic impairments, predominantly in the orbitofrontal cortex and the brainstem (for review, see reference 3). Thus, alterations in the activity of the serotonergic projections to the orbitofrontal cortex appear to be an important component of susceptibility to suicidal behavior. Furthermore, molecular genetic studies have reported associations of suicidal behavior with serotonin-related genes (4–6). Nonreplication (7) may be accounted for by the heterogeneity and multiple causes of suicidal behavior. Neuropsychological traits could be used to increase the homogeneity of suicidal behavior groups in studies.
A few studies have investigated neuropsychological dysfunction in suicide attempters. Keilp et al. (8) found that depressed suicide attempters presented a deficit in executive functions, which are known to be linked to the prefrontal cortex, independently of the severity of depression. The cognitive function of decision making has been shown to be linked to the orbitofrontal cortex by lesion (9–11) and functional imaging studies (12–15). Damage to the orbitofrontal cortex leads to high-risk, disadvantageous decisions in real life (16, 17). If a lesion is restricted to the orbitofrontal cortex, all classic neuropsychological test results are normal, except those assessing decision making, such as the Iowa Gambling Task (18) and others (10). Moreover, previous studies reported the possible modulation of decision-making functions by the serotonergic system (10, 19).
Assuming that the orbitofrontal cortex and the serotonergic system are involved both in this cognitive function and in susceptibility to suicidal behavior, we tested the hypothesis that decision making is impaired in suicide attempters. We also investigated whether Iowa Gambling Task performance was correlated with personality traits that are thought to be linked to the prefrontal cortex, suicidal behavior, and serotonergic dysfunction.
+
Subjects and Experimental Design
Sixty-nine suicide attempters ages 19 to 70 years were included in the study. This study was part of a larger multicenter project on the genetics of suicidal behavior, which was approved by the local research ethics committees. Subjects had to be between 18 and 75 years old, with all four biological grandparents originating from countries in Western Europe. After complete description of the study to the subjects, written informed consent was obtained.
A suicide attempt was defined as a self-damaging act carried out with some intent to die and distinguished from other self-destructive types of behavior, such as self-mutilation, noncompliance with medical treatment in severely ill individuals, and the use of substances such as alcohol and tobacco (3). Patients who exhibited only suicidal ideation or who threatened to commit suicide without actually taking action were not included.
We characterized the suicide attempts by using the French versions of the Risk Rescue Rating Scale (20), a 10-item scale evaluating the risk (and more precisely, the lethality) and the possibilities of rescue regarding the act, and the Suicidal Intent Scale (21), a 15-item scale assessing the circumstances and expectations relating to the suicide attempt. Lifetime axis I psychiatric diagnoses were made according to the DSM-IV. Diagnostic assessment had been done by experienced psychiatrists (F.B., B.A., S.T., and P.C.) with the French version of the Mini-International Neuropsychiatric Interview (22), medical records, and any other information. All participants were also assessed with the 21-item Hamilton Depression Rating Scale (23) to ensure that they were normothymic at the time of evaluation (score below 7). We assessed intellectual ability using the French version of the National Adult Reading Test (24).
We assigned suicide attempters to two groups of similar size that were defined on the basis of the means used in the most serious suicide attempt (25): violent (N=32) or nonviolent (N=37). Violent suicide attempts involved deep cutting (N=8), jumping from heights (N=6), hanging (N=6), electrocution (N=3), jumping in front of a car (N=2), gunshot wounds (N=2), crashing a scooter or a car (N=2), asphyxia with a bag and gas (N=1), immolation by fire (N=1), and drowning (N=1). Nonviolent suicide attempts were drug overdoses (N=33), superficial wrist cutting (N=1), and a combination of drug overdose and superficial wrist cutting (N=1). Two patients who drank detergent were classified as nonviolent attempters. Thirteen of the 34 patients who took drug overdoses required hospitalization in an intensive care unit after their suicide attempts.
Two groups of comparison subjects were included after a clinical examination by the experienced psychiatrists we mentioned with the Mini-International Neuropsychiatric Interview: 82 subjects with no personal history of suicide attempt or current pharmacological treatment (healthy comparison subjects) and 25 subjects with no personal history of suicide attempt but a personal history of mood disorder (affective control subjects). Inclusion of the affective control group made it possible to compare suicide attempters with other individuals susceptible to mood and anxiety disorders who were taking medication. All control subjects were between 19 and 69 years old. None of the participants had a neurological disorder.
We tested decision-making skills using the computerized version of the Iowa Gambling Task (26). The procedure was identical for all participants and has been described elsewhere (26). Briefly, the subject has to make a choice between four decks of cards (60 cards each). When he or she clicks on a deck, he or she may win or both win and lose money. The goal of the game is to win as much money as possible. The subject does not know that he or she will pick up a total of 100 cards. The subject also does not know that there are two advantageous decks for which little money is won but even less is lost (resulting in a net gain) and two disadvantageous decks for which a lot of money is won but even more is lost (resulting in a net loss). The subject does not know when he or she will lose money, which creates a feeling of uncertainty. The recording of the choices is made by a computer.
Personality traits were assessed by means of self-administered questionnaires: the Barratt Impulsiveness Scale (27), the Buss-Durkee Hostility Inventory (28), and the Spielberger State-Trait Anger Expression Inventory (29). We used two self-report scales to assess emotional changes: the Affective Intensity Scale (30), which scores the intensity of emotional response, and the Affective Lability Scales (31), which scores various types of affective variation (depression/normal, elation/normal, bipolar [i.e., elation/depression], anxiety/normal, anxiety/depression, and anger/normal). In our analyses, a low score on the scale means high affective lability. We also used the Brown-Goodwin Life History of Aggression (32), a questionnaire concerning lifetime aggression.
Choices were classified as advantageous (decks C and D) or disadvantageous (decks A and B). The score is the difference between the number of advantageous and disadvantageous choices. Five intermediate scores were calculated for each subject (blocks of 20 choices), indicating changes in the pattern of choices during the game. A net score (sum of the intermediate scores) was then calculated.
Statistical analysis was carried out with SAS software (SAS Institute, Cary, N.C.). Since the Shapiro-Wilks test showed nonnormal distribution for most of the scores, we used nonparametric tests. Significance was set at p<0.05. Changes in intermediate scores during the course of the test were analyzed for each group with Friedman’s test. Quantitative values were compared by Kruskal-Wallis test, followed by Bonferroni correction (if more than three-group comparisons) or Wilcoxon’s test (if two-group comparisons). A chi-square test was used to compare qualitative values. Spearman’s correlation coefficients were calculated for correlation analyses.
+
Demographic and Clinical Variables
The four groups (healthy comparison subjects, affective control subjects, nonviolent suicide attempters, and violent suicide attempters) were statistically similar in terms of age, level of education, and National Adult Reading Test score (t1). They differed significantly in terms of sex ratio, with fewer women among the healthy comparison subjects and the violent suicide attempters (26.8% and 34.4%, respectively) than among the affective control subjects and the nonviolent suicide attempters (64.0% and 67.6%, respectively) (p=3.10–5).
The two groups of suicide attempters did not differ in terms of the number of previous suicide attempts and the age at first suicide attempt. Violent suicide attempters had a higher mean risk score (Risk Rescue Rating Scale) than the nonviolent suicide attempters (11.1 versus 8.7) (p=0.01).
For axis I disorders, only substance abuse differed between the groups since it was more frequent in the violent and nonviolent suicide attempters than in the affective control subjects (36.7%, 33.3%, and 4.0%, respectively) (p=6.10–3). The three groups of patients differed in terms of medication use on examination: violent and nonviolent suicide attempters were more likely than the affective control subjects to be taking benzodiazepines (41.9%, 67.6%, and 4.0%, respectively) (p=10–6). The proportion of patients not taking medication and using antidepressants was similar in all three patient groups.
+
Iowa Gambling Task Performance
Within each group, intermediate scores significantly changed during the course of the test for the healthy comparison subjects (Friedman’s test=50.9, df=4, p<10–5), the nonviolent suicide attempters (Friedman’s test=8.1, df=4, p=0.05), and the violent suicide attempters (Friedman’s test=11.6, df=4, p=0.04) but not for the affective control subjects (Friedman’s test=4.6, df=4, p=0.30) (F1, F2, t1).
Comparison of scores among the four groups showed that for the fifth score (p=5.10–5) and the net score (p<10–5), the healthy comparison subjects differed significantly from both groups of suicide attempters, and the violent suicide attempters differed significantly from the affective control subjects. In contrast, no significant differences were observed between the nonviolent and violent attempters, between the healthy comparison and affective control subjects, and between the nonviolent suicide attempters and the affective control subjects.
The net score for the Iowa Gambling Task did not differ significantly according to sex when we considered the entire population and within each of the four groups of subjects. There was no difference according to a history of psychiatric disorders when we considered the three patients groups as a whole. Concerning the particular case of substance abuse disorders, we compared all suicide attempters with and without a substance abuse disorder, and we found no significant difference for the mean net scores. Moreover, after removing all of the substance abusers, comparisons of the net scores among the four groups (violent suicide attempters, nonviolent suicide attempters, affective control subjects, and healthy comparison subjects) yielded no qualitative change compared to what was found previously when we did not exclude the substance abusers. Thus, we think that our results may reasonably be linked to suicidal behavior and not to substance abuse.
Furthermore, the net score for the Iowa Gambling Task was not significantly correlated with age, level of education, National Adult Reading Test score, age at first suicide attempt, the number of previous suicide attempts, or the various scores for the Suicidal Intent Scale and the Risk Rescue Rating Scale. Finally, the net scores were not statistically different within each group between centers.
To exclude the effect of medication, we compared all patients with (N=79) and without (N=15) any medication and found that their performances were similar. In our groups, benzodiazepines were overrepresented in both groups of suicide attempters, and these drugs may have affected the attention of these patients. Therefore, we compared patients with (N=39) and without (N=55) benzodiazepines and found no significant difference in net score for the Iowa Gambling Task (Wilcoxon’s test=1709.5, df=1, p=0.30). Patients with (N=67) and without (N=26) antidepressants also performed similarly (Wilcoxon’s test=3007.5, df=1, p=0.20).
It is possible that minimal brain lesions resulting from the suicidal act may account for the differences in performance between groups. We investigated whether this was the case by comparing the Iowa Gambling Task performances of patients who had (N=24) and had not (N=61) experienced a coma or a head trauma (data missing for nine subjects) and found no significant difference between these two groups (Wilcoxon’s test=959.5, df=1, p=0.50). The distribution of these possible sources of brain lesions was similar in violent and nonviolent suicide attempters (39.1 and 40.5%) (χ2=0.01, df=1, p=1.00).
+
Psychometric Measures and Iowa Gambling Task Performance
We evaluated the correlations among all the scores of the various scales assessing personality dimensions and net score on the Iowa Gambling Task for violent suicide attempters, nonviolent suicide attempters, and all suicide attempters together. For all suicide attempters together, net score on the Gambling Task was significantly correlated with the bipolar score of the Affective Lability Scales (r=–0.32, p=0.05). For nonviolent suicide attempters, net score on the Iowa Gambling Task was significantly correlated with the anger-out score on the Spielberger State-Trait Anger Expression Inventory (r=0.45, p=0.02) and the resentment score on the Buss-Durkee Hostility Inventory (r=0.32, p=0.04). For violent suicide attempters, no correlation was significant.
To our knowledge, this study is the first to report a specific impairment in decision making in suicide attempters. We found that violent and nonviolent suicide attempters achieved lower scores in a decision-making task, the Iowa Gambling Task, than did healthy comparison subjects. This impairment in decision making seems to be associated with susceptibility rather than with a state because all patients were evaluated during a period in which they did not have an axis I disorder. No difference was observed between healthy and affective control subjects, which is consistent with a previous study in euthymic bipolar patients (33). In contrast, there was a significant difference between violent suicide attempters and affective control subjects, suggesting that the decision-making impairment identified was associated with susceptibility to suicidal behavior rather than susceptibility to affective disorders. Furthermore, psychiatric history had no effect on performance. It has to be outlined that in our group, a diagnosis of substance abuse disorder does not impair decision-making performance, although in a previous study, the two-thirds of substance abusers were impaired on the Iowa Gambling Task (34). This raises the question of a common trait that would be shared by suicide attempters and substance abusers. Thus, in our groups of suicide attempters, susceptibility to psychiatric disorders did not seem to affect decision making as much as susceptibility to suicidal behavior did. Finally, we found no significant difference between the two groups of suicide attempters.
The Iowa Gambling Task has been shown to involve various cerebral regions, especially the orbitofrontal cortex and the amygdala (26, 35). Our results seem to be consistent with a recent imaging study (36) reporting an association between lower levels of ventromedial activity and both higher suicidal intent and higher lethality of the act. First, we observed that suicide attempters performed poorly on the Iowa Gambling Task. Violent, but not nonviolent, suicide attempters showed a gradual change in their pattern of choices during the Iowa Gambling Task that was very similar to that observed in patients with an orbitofrontal cortex lesion. Violent suicide attempters begin by sampling all decks until the 60th card, as do subjects from other groups; then they make more and more disadvantageous choices (35). Second, the impairment was more marked in violent suicide attempters who also had a higher Risk Rescue Rating Scale risk score (i.e., a higher lethality). Thus, impaired decision making in suicide attempters may reflect an orbitofrontal cortex dysfunction. Functional imaging studies, with a decision-making paradigm, would be useful in identifying precise regional dysfunctions in suicide attempters.
Decision making has been found to be impaired in various psychiatric disorders involving discrete association of aggressive impulsivity, increased risk of suicidal behavior, and orbitofrontal cortex and/or serotonergic dysfunctions, such as substance abuse (34, 37, 38), affective disorders (33, 39), conduct disorder (38), impulsive aggressive disorder (40), and psychopathic (41) and borderline personality disorders (42). Moreover, both the orbitofrontal cortex and the serotonergic system have been linked to the control of impulse (43, 44), and suicide attempters tend to be more impulsive (2). In this study, we found no correlation between impulsivity (motor or cognitive), as measured by the Barratt Impulsiveness Scale, and Iowa Gambling Task performance. This is consistent with the results of a recent study carried out in borderline patients (42). As discussed by Bechara et al. (35), the link between decision making and a specific component of impulsivity remains to be clarified. Furthermore, it has been suggested that questionnaire-based and laboratory-based assessments of impulsivity may differ (45).
We found that in suicide attempters, Iowa Gambling Task performance was better as affective lability was higher. Furthermore, anger dyscontrol was positively correlated with the Iowa Gambling Task only in the subgroup of nonviolent suicide attempters. The orbitofrontal cortex is involved in the emotional processing thought to be necessary for normal decision making (35). Thus, the decision-making deficit reported here may reflect a more global emotional dysfunction in some suicide attempters, even when they are not suffering from a current affective disorder. This also raises the question of the heterogeneity of suicidal behavior (violence, lethality) and of its link with different patterns of personality traits.
This study has several limitations. First, although the number of participants was large for a neuropsychological study, the size of the groups combined with the use of nonparametric tests—which are more robust but less sensitive than parametric tests—limited the statistical power of some comparisons. Replication in larger groups is required to validate these results.
Smoking status has not been taken into account in the present study. However, in a recent work published during the review process for this article, Rotheram-Fuller et al. (46) reported an effect of smoking in methadone-maintained patients but not in comparison subjects. Thus, smoking status needs to be controlled in further studies.
Moreover, most of our patients were taking medication, and we found it difficult in this preliminary study to stop them. Therefore, we chose to compare the suicide attempters with a group of comparison subjects who were also taking medication. This affective control group did not differ significantly from the healthy comparison group, whereas the group of violent suicide attempters did. Furthermore, as shown in a previous study using different decision-making tests in medicated depressed and manic patients (39), we found no effect of this variable on the performance of the Iowa Gambling Task.
It is also possible that minimal brain lesions resulting from the suicidal act may account for the differences in performance among groups. Again, comparison of performance between subjects with or without head trauma/coma yielded no difference. Nevertheless, there remains a small risk that subclinical brain damage may have biased certain results.
Finally, it remains necessary to investigate nondepressed suicide attempters by using other neuropsychological functions related to prefrontal activity, such as working memory, error monitoring, and conflict resolution.
In conclusion, we report here a decision-making defect in noncurrently depressed suicide attempters that may be correlated with emotional dysfunction. Decision-making impairment may therefore represent a vulnerability factor for suicidal behavior.
Received June 5, 2003; revisions received July 24, 2003, and Feb. 9, 2004; accepted April 9, 2004. From the Department of Psychological Medicine and Psychiatry, Lapeyronie Hospital and Institut National de la Santé et de la Researche Médicale (INSERM) E 0361; the Department of Psychiatry, Albert Chenevier and Henri Mondor Hospitals and INSERM U513, Créteil, France; the Department of Biostatistics, Arnaud de Villeneuve Hospital, Montpellier, France; and the Division of Neuropsychiatry Genetics, Department of Psychiatry, Geneva University Hospitals, Geneva, Switzerland. Address correspondence and reprint requests to Dr. Courtet, Service de Psychologie Médicale et Psychiatrie, Hôpital Lapeyronie, 371 avenue du Doyen G. Giraud, 34295 Montpellier cedex 5, France; p-courtet@chu-montpellier.fr (e-mail). The authors thank Caroline Alter and Maria Boucherie for data collection.Supported by the CHU of Montpellier (PHRC UF 7653). Mr. Jollant received a grant from the Fondation pour la Recherche Médicale (Action Dynamique en Psychiatrie 2001).