This is the first study to explore risk-sensitive decision-making using a cognitive probe in elderly depressed suicide attempters. On the Cambridge Gamble Task, we found that older suicide attempters with major depression were less likely to predict the probable outcome. We included a group of non-suicidal depressed cases, and a group of depressed ideators with active suicidal intent but no history of attempt as well as a comparison group of non-depressed controls. The attempters had significantly impaired Quality of Decision-Making in direct comparisons to the non-suicidal depressed and the non-depressed controls. The direct comparison to the ideators was not significant, potentially as a result of low power (n=13 in the ideator group, p=.081) but yielded a moderate effect size (Cohen’s d=0.66), suggesting that replication with larger groups would be likely to show an effect. While there was a significant effect of group on the time to make these probability decisions (Decision Latency), the between-group comparisons did not reveal any significant differences. There were no group differences in Betting, or in overall performance in terms of bankruptcy and net points won. Impaired Quality of Decision-Making was associated with impulsive/careless problem-solving on the SPSI, a self-report measure of perceived social problem-solving ability.
These findings support the relevance of decision processes to our understanding of suicidal behavior in the context of depressive illnesses. Attempted suicide has been previously associated with impaired performance on the Iowa Gambling Task, in adolescents (Oldershaw, et al., 2009
) and younger adults (Jollant, et al., 2005
; Malloy-Diniz, et al., 2009
). Our findings extend these observations to elderly depressed subjects, who are known to be most at risk of death by suicide. Our study used an alternative task (Cambridge Gamble Task), which was not confounded by information processing demands that can be impaired in the elderly, to probe risk-sensitive decision-making. This task also enables us to dissect several components of decision-making cognition. The Iowa Gambling Task assesses decision-making based on the prior experience of rewarding and punishing outcomes with the four card decks. The subject is given no explicit information about the chances of winning and losing on each deck. Thus, the Iowa Gambling Task measures decision-making under uncertainty and risk,
in a learning context. Using the Cambridge Gamble Task, we are able to show an impairment in decision-making in elderly attempters on a task with minimal learning requirements, where the outcome probabilities are provided explicitly by the ratio of red to blue boxes in the array (i.e. decision-making under risk
only). In fact, each of the four groups, including the attempters, used the box ratio information about outcome probability to adjust their color decision and bet (the main effects of Ratio), and there were no significant differences in these adjustment slopes across groups (the Group by Ratio interaction terms). Thus, the ability to derive decision weights from probabilities appears unimpaired in the attempters, while their deficit was in the ability to consistently apply this knowledge of probability in selecting the probable outcome.
Why did suicide attempters ignore outcome probabilities, even when they are explicitly presented and correctly understood? One can formulate a few testable hypotheses about biases in their decision-making. Suicide attempters’ behavior could be explained by a belief in spurious contingencies, such as the gambler’s fallacy (e.g. red is more likely to win after several successive blue outcomes). Probability matching is a related phenomenon: instead of always selecting the highest-probability option, humans often match the estimated probability (e.g. 80:20) by selecting the lower-probability option on some trials. Assessment of these facets of judgment and numerical reasoning in suicidal individuals represents one line of future enquiry. Also, if the subject believed that the computer provided inaccurate or misleading probability information, he or she may approach the task as an uncertain environment and engage in sampling from both colors to learn the underlying probability distribution. In light of our previous findings using a Probabilistic Reversal Learning task (Dombrovski, et al., 2010
), which showed that suicide attempters neglected to integrate prior experiences into their decisions, our current findings indicate that elderly suicide attempters appear to also neglect basic probability information, thus ignoring the broader context in which decisions are made. This tendency can clearly undermine deterrents to suicide in someone who fails to consider prior experiences and the probable consequences of their choices.
The observed changes on a neuropsychological probe of decision-making are broadly consistent with dysfunction in the ventromedial PFC in suicide attempters (J. J. Mann, 2003
; J. J. Mann et al., 2000
). Quality of Decision-Making on the Cambridge Gamble Task was previously reported to be impaired in patients with orbitofrontal cortex damage, who were also slowed in their decision latencies (Rogers, et al., 1999
). However, subsequent studies have shown that ventral PFC pathology preferentially changes the Betting parameter (Clark, et al., 2008
; Mavaddat, Kirkpatrick, Rogers, & Sahakian, 2000
; Rahman, Sahakian, Hodges, Rogers, & Robbins, 1999
). It is, of course, possible that these measures are related. In groups where the basic probability judgment is intact, Betting may represent the more sensitive marker of risky decision-making. However, in cases where basic probability processing is disrupted, the amount bet may be less meaningful. Thus, we are cautious about inferring strong evidence for ventromedial PFC dysregulation. It should also be noted that deficits on the Quality of Decision-Making variable are not unique to suicidality: previous studies have described deficits on this parameter in patients with bipolar mania (Murphy, et al., 2001
), chronic schizophrenia (Hutton et al., 2002
), and as a predictor of treatment outcomes in opiate users (Passetti, Clark, Mehta, Joyce, & King, 2008
). Some previous studies of cognition in suicide have tested heterogeneous groups with a variety of primary diagnoses (Jollant, et al., 2005
; Raust et al., 2007
). A strength of the present study is that we focused on major depression, as the single most common antecedent of elderly suicide (Conwell, et al., 1996
). Individuals with primary diagnoses of schizophrenia or bipolar disorder were excluded. The three mood disorder groups did not differ on lifetime anxiety co-morbidity, and whilst the attempters were more likely to have lifetime substance use problems, those attempters with substance use co-morbidity did not display worse decision-making than those without, so this seems unlikely to confound our effects. Inclusion of these co-morbidities increases the generalizability of these findings.
In adolescents and young adults, suicide is often regarded as an impulsive act (McGirr et al., 2008
), and neuropsychological correlates of impulsivity have been reported in young adult samples (Horesh, 2001
; Raust, et al., 2007
; Swann et al., 2005
). Our clinical experience suggests that suicide attempts in older adults are more premeditated, and indeed, this may account for the higher lethality of attempts in the elderly (Dombrovski, Szanto, et al., 2008
). To what extent can a deficit in risk-sensitive decision-making be related to impulsivity? Prior work has shown that impulsivity did not predict the degree of impairment in suicide attempters on the Iowa Gambling Task (Jollant, et al., 2005
) or on a test of decision-making under risk, similar to the Cambridge Gamble Task, in healthy volunteers (Franken, van Strien, Nijs, & Muris, 2008
). In the Cambridge Gamble Task, the discrepancy between the level of betting in the Ascend and Descend conditions provides a direct index of delay aversion and impulsivity, and this parameter was not associated with attempted suicide in the present study. Rather, having processed the probability information, the suicide attempters made a decision that neglected knowledge of probability. We observed no significant differences between groups on the Decision Latency measure, which mitigates against an explanation in terms of speed-accuracy trade-offs or ‘reflection impulsivity’ (see Clark, Robbins, Ersche, & Sahakian, 2006
). Poor Quality of Decision-Making was significantly associated with the impulsivity/carelessness subscale of the SPSI, and scores on this subscale were significantly higher in the attempters compared to the other three groups (see also Gibbs, et al., 2009
). However, other facets of the SPSI were also linked to suicidality (Total score, and Negative Problem Orientation and Avoidance subscales), and although the only significant correlation was with the impulsivity/ carelessness subscale against the decision-making parameter, this association may not be selective. We infer that the profile observed in suicidal depressed elders is not readily explained in terms of impulsivity. Specifying any age-related change in impulsive versus deliberative, premeditated decisions in suicidal behavior would seem to represent an important target for future research, comparing younger and older groups.
The similar performance of the non-suicidal depressed and non-depressed controls on the Cambridge Gamble Task is also notable. Previous studies have reported some decision-making abnormalities in mid-life depressed populations. For example, on the Iowa Gambling Task, patients with major depression made fewer selections from the safe decks (Must et al., 2006
), and on the Cambridge Gamble Task, one study reported slower deliberation times and some attenuation of risk adjustment (the variation in betting across box ratios) (Murphy, et al., 2001
) while a later study reported no effects (Taylor Tavares, et al., 2007
). The history of suicidal behavior and ideation was not described in these earlier reports, and it is possible that the depressed individuals whose decision-making was impaired had prior suicide attempts or ideation. If our findings are replicated, decision-making impairments may differentiate those who contemplate suicide from those who act on these thoughts. Another clinical implication of our findings is that decision-making impairment may be a useful target for treatment in suicidal depressed elders. Further, impaired decision-making may contribute to inability to adhere to treatment, manage one’s finances, or remain in independent housing – these problems are associated with prefrontal dysfunction (Fan, Royall, Chiodo, Polk, & Mouton, 2003
; L. S. Mann et al., 1999
; Royall, Chiodo, & Polk, 2005
) and often encountered in depressed elders, and may also be amenable to targeted interventions.
Several limitations of the current study should be noted. We used a cross-sectional design, and it is possible that the psychological profile of suicide completers may differ from the suicide attempters that we have characterized here. Sample sizes were modest, particularly in the ideator group, and the lack of significant direct comparisons against the ideators may be due to a lack of statistical power. It is notable that a significant difference between attempters and ideators was revealed on Quality of Decision-Making after excluding three attempters with possible or likely neurological damage as a consequence of the attempt. As a further caveat, we cannot comment on the specificity of these effects to the decision-making domain. Quality of Decision-Making was positively associated with years of education, and whilst the four groups did not differ on general cognitive status assessed with the MMSE, it is possible that this scale lacked sensitivity in these populations. Further research is required to assess the specificity of decision-making variables against other cognitive constructs implicated in suicidality, such as executive function (Dombrovski, Butters, et al., 2008
; Dombrovski, et al., 2010
; Keilp, et al., 2008
). Nevertheless, the present data add to a growing number of studies that highlight risk-sensitive decision-making as an important cognitive feature of suicidal behavior in depression. A disposition to make decisions that ignore outcome probabilities may interact with psychosocial stressors such as physical illness (Waern et al., 2002
), financial hardship (Duberstein, Conwell, Conner, Eberly, & Caine, 2004
) and interpersonal loss (Szanto, Prigerson, Houck, Ehrenpreis, & Reynolds, 1997
), to precipitate and perpetuate suicidal crisis in depressed elders.