Suicidal behaviour and memory: A systematic review and meta-analysis.

Objectives. Suicidal behaviour results from a complex interplay between stressful events and vulnerability factors, including cognitive deficits. It is not yet clear if memory impairment is part of this specific vulnerability. Therefore, the objective of this study was to examine the association between memory deficits and vulnerability to suicidal acts. Methods. A literature review was performed using Medline, Embase, and PsycInfo databases. Twenty-four studies (including 2,595 participants) met the selection criteria. Four different types of memory (i.e., working memory, short- and long-term memory, and autobiographical memory) were assessed in at least three different studies. Results. Autobiographical memory was significantly less specific and more general in patients with a history of suicide attempt relative to those without such a history (Hedges’ g = 0.8 and 0.9, respectively). Long-term memory and working memory were both more impaired in suicide attempters than in patient and healthy controls. Only short-term memory did not differentiate suicide attempters from patient controls. Conclusions. Memory may play a significant role in the risk of suicidal acts, perhaps by preventing these individuals from using past experiences to solve current problems and to envision the future, and by altering inhibitory processes. More studies are necessary to better clarify these relationships.


Introduction
What makes a person consider and eventually commit a suicidal act while another person in the same context of stressful life events or major depression, for instance, will never do so? it is now largely thought that the first person may present with vulnerability factors which increase the risk in stressful conditions of experiencing mental pain, hopelessness and suicidal ideas and, in some cases, of acting out (Mann 2003). neuropsychology may shed light on some aspects of these vulnerability factors.
a growing literature suggests that suicide attempters, in comparison to patient and healthy controls, show several cognitive deficits that may play a significant role in the vulnerability to suicidal acts (Jollant et al. 2011). We recently conducted a metaanalysis of seven different neuropsychological tests (Richard-devantoy et al. 2013) and suggested that impairment in value-based decision-making and cognitive inhibition, two deficits that show little overlap in suicide attempters (Richard-devantoy et al. 2013), may be part of this specific vulnerability.
We did not assess memory in this previous study. yet, memory is a major brain function playing a significant role in what we are and do. consequently, investigating memory deficits in relation to the vulnerability to suicidal behaviour is particularly relevant. Several studies have been published reporting memory deficits in suicide attempters (see below). however, both positive and negative results have been published, one major limitation of several studies being a small sample size. also, it is not clear if these deficits may be associated with suicidal behaviour or may be more closely related to co-morbid psychiatric disorders such as major depression. to this aim, we have systematically reviewed the literature on

Study selection
abstract selection was based on the Strengthening the Reporting of oBservational studies in epidemiology (StRoBe) checklist (von elm et al. 2008) which describes items that should be included in reports of cohort studies. abstracts identified through the literature search were independently evaluated by two reviewers (SRd and FJ) and selected by a consensus from all authors.
Studies that met the following inclusion criteria were included in this systematic review: (1) published in an english or French language peerreviewed journal; (2) including at least one memory task; and (3) including at least one group of patients with a history of suicide attempt. a suicidal attempt was defined as any act carried out with a certain intent to die and different from nonsuicidal self-injury (Mann 2003). Full articles were then obtained for final review. the study selection process is shown on a chart flow diagram in Figure 1.
of the 334 originally identified abstracts, 31 studies met the inclusion criteria for this systematic review (see table i). the quality of each study was assessed independently by two reviewers (SRd and FJ) using the crombie criteria adapted by petticrew et al. (2006). then, studies that met the additional inclusion criteria of comparing at least two groups of which one comprised patients with a history of suicide attempts were included in this meta-analysis. of the 31 studies of the systematic review, 24 studies met this criterion (see table ii). although eligible, two studies were not included, because precise means and standard deviations were not available in the papers and could not be obtained directly from the authors (Williams et al. 1986;taylor et al. 2010).

Data extraction and analyses
a standardized form was used to extract data, which included authors, date of publication, study design, settings, study population, memory tests used, definition of suicidal behaviour, and memory scores (mean and Sd). Five variables encompassing four memory domains that have been explored in at least three different studies, were analysed: (1) Working memory, (2) Short-term memory, (3) long-term memory, (4) General autobiographical memory, and (5) Specific autobiographical memory.
analyses were performed using comprehensive Meta-analyses version 2.0 (Biostat, englewood, nJ, USa), and iBM SpSS version 20 (iBM corporation, chicago, il, USa). memory in suicidal behaviour, and then conducted a meta-analysis.
as no particular memory test has been used in at least three different studies, we had to use a cognitive domain approach (Keefe 1995). therefore, different tests had to be analysed together in the meta-analysis. Moreover, defining categories of memory is not a straightforward task. Based on previous publications in the field of suicide, we chose to consider the four following types: (1) working memory, (2) short-term memory, (3) long-term memory, and (4) autobiographical memory. Working memory is the function that actively holds multiple pieces of transitory information in mind, where they can be manipulated (conway et al. 1996;Goldman-Rakic 1996), whereas short-term memory is defined as the capacity for holding a small amount of information in mind in an active, readily available state for a short period of time (engle et al. 1999; tetzlaff et al. 2012). in contrast, long-term memory can hold a large amount of information for a long period of time. Finally, autobiographical memory is the system involved in collecting, storing and retrieving episodes from an individual's life, and comprising a combination of episodic (e.g., personal experiences and specific objects, people and events experienced at particular time and place) and semantic (general knowledge and facts about the world) memory (Williams et al. 1986).
this meta-analysis therefore aimed at clarifying the relationship between each of these four domains and the vulnerability to suicidal behaviour, and disentangling potential deficits associated with suicidal behaviour from those related to comorbid disorders.

Methods and material
Data sources an english and French systematic literature search of Medline, embase, and psycinfo databases was performed for human studies published until 31 June 2013. the Medical Subject heading (MeSh) term "suicide" was combined with the MeSh terms "neuropsychology", "neuropsychological tests", "executive Function", "Memory", "Memory, long-term", "Memory, Short-term", "Working Memory", and "autobiographical Memory". an iterative process was used to ensure that all relevant articles were obtained. a further hand search of the bibliographical references of the selected papers and existing reviews was conducted to identify additional potential studies. References were also selected from our research group's online database (www.bdsuicide.disten.com).
We used a random-effects model as we assumed that the true effect sizes had likely varied between the included studies (Riley et al. 2011). pooled hedges' g effect sizes for the subjects' neuropsychological three groups were compared: suicide attempters (patients with a history of suicide attempt), patient controls (i.e., patients with no personal history of suicidal act but with a history of mental disorders) and healthy controls. When two groups of suicide attempters were reported in one study (e.g. low vs. high lethality acts (McGirr et al. 2011;Keilp et al. 2013)), the combined means and standard deviations were calculated to obtain a global group, using the following formula:   (cooper et al. 2009). values of P  0.10 for the former and  35% for the latter were deemed as indicative of study heterogeneity. Finally, we used Funnel plots, Rosenthal's Fail-Safe n (Rosenthal 1979) and egger's Regression intercept (egger et al. 1997) to test for the presence of publication bias (cooper et al. 2009).
autobiographical memory was only tested with the autobiographical Memory test (aMt; Williams et al. 1986). in the aMt, five positive words (happy, safe, interested, successful, and surprised) and five negative words (sorry, angry, clumsy, hurt (emotional), and lonely) are successively presented to the participant. Subjects are given 1 min to produce a specific personal memory in response to each word. Responses are usually audio-recorded. the specificity of the memories is then determined on the basis of details given to describe the event (Williams et al. 1986).
Working memory was assessed with six different tests (i.e., n-Back task, dual-task of Baddeley, can-taB Spatial Working Memory, consonant trigram test, WMS-iii letter-number Sequencing task; Wechsler Memory Scale 3rd edition), short-term memory with 11 different tests (Buscke selective ( * )contrast became significant after excluding studies responsible for heterogeneity.
[ * ]contrast became non-significant after excluding studies responsible for heterogeneity.
performances in suicide attempters compared to patient controls (Kim et al. 2003;nangle et al. 2006) and three studies in mood disorders found worse performance in suicide attempters, especially in high-lethality attempters (Keilp et al. 2001(Keilp et al. , 2013. Similarly, three studies found no between 4.
group differences in long-term memory, but two studies in schizophrenia found better performances in suicide attempters compared to patient controls (Kim et al. 2003;potkin et al. 2003), and two studies in mood disorders found worse performance in suicide attempters, especially in high-lethality attempters (Keilp et al. 2001(Keilp et al. , 2013.

Meta-analysis
a total of 24 studies were included (table ii), comprising 2,595 participants, of whom 984 were suicide attempters (mean age  40.8  10.0 years; 43.5 % males), 942 were patient controls (40.1  9.4 years; 46.9 % males), and 669 were healthy controls (41.8  8.6 years; 44.1 % males). table iii presents the results of the contrasts between the three groups for the five variables, and table iv provides a summary of the main findings. detailed information on heterogeneity and publication bias can be found in the table in the Supplemental data available online at http://informahealthcare.com/ doi/abs/10.3109/15622975.2014.925584.

Suicide attempters vs. patient controls
Suicide attempters had significantly less specific and more general autobiographical memories than patient controls (Figure 2, a & B), all representing high effect sizes. the Fail-Safe n, i.e., the number of unpublished or missing null-findings that would be needed to render the results non-significant, was 18 for the general autobiographical memory and 26 for the specific autobiographical memory. depression level did not differ between the two groups, thus ruling out depression intensity as a confounding factor.
Besides, suicide attempters had significantly worse performances than patient controls in working memory but not short-and long-term memory (Figure 2, c, d, and e). heterogeneity exceeded that expected by chance at P  0.05 for working memory and short-and long-term memory implying that the variance among the effect sizes was greater than expected by sampling error (table in Supplemental data available online at http://informahealthcare.com/doi/abs/ 10.3109/15622975.2014.925584). three studies reminding test, dementia-Rating Scale, Wechsler Memory Scale 3rd edition immediate memory index, dual task of Baddeley, california verbal learning task-ii, memory battery of Signoret, Rey auditory verbal learning test immediate recall, paired associate learned, verbal list learning immediate recall, verbaler lern-and Merkfähigkeitstest trial 1-5, n-Back test 0 back), and longterm memory with five different tests (memory battery of Signoret, verbaler lern-and Merkfähigkeitstest trial 7, Wechsler Memory Scale 3rd edition, Rey auditory verbal learning test delayed recall, and verbal list learning delay recall).
cific and more general autobiographical memories in suicide attempters compared to patient (Williams et al. 1986(Williams et al. , 1996pollock et al. 2004;Kaviani et al. 2005;arie et al. 2008;pettersen et al. 2010) and healthy controls (Williams et al. 1996;pollock et al. 2004;Kaviani et al. 2005, arie et al. 2008, Maurex et al. 2010, across different age and diagnosis groups. one study in psychosis reported the opposite results (taylor et al. 2010), and one study reported no betweengroup difference (Williams et al. 2005). Finally, four studies reported a significant association between poor performance in autobiographical memory and a higher number of suicide attempts (Startup et al. 2001;Sinclair et al. 2007;Rasmussen et al. 2008;crane et al. 2009), especially in those with an early age of onset of history of childhood abuse (crane et al. 2009). among 10 studies on working memory, no 2.
difference between suicide attempters and patient controls were found in five studies. Four studies reported worse performances in suicide attempters with mood disorder (Raust et al. 2007;Richard-devantoy et al. 2012;Jollant et al. 2013;Keilp et al. 2013). one study (Keilp et al. 2013) found highlethality attempters to outperform lowlethality attempters although it is not clear if this result is a true positive result, if this is related to the high frequency of violent attempters in the low-lethality group, or to a sampling bias. one study reported better performance in suicide attempters with schizophrenia (Kim et al. 2003) relative to patient controls. no between group differences in short-term 3. memory were found in 10 studies, but two studies in schizophrenia found better general autobiographical memory due to the small number of studies available. it was 8, 77, and 29, respectively, for the other memory domains. heterogeneity exceeded that expected by chance for four domains. two studies (Williams et al. 1996;Maurex et al. 2010) were likely responsible for the heterogeneity related to the specific autobiographical memories performance. after excluding these studies, results became significant. two studies (Malloydiniz et al. 2009;Richard-devantoy et al. 2012) were likely responsible for the heterogeneity related to short-term memory performance. after excluding these studies, results remained significant. two studies (Martino et al. 2010;Richard-devantoy et al. 2012) were likely responsible for heterogeneity related to the long-term memory and the working memory, respectively. after excluding these studies, results remained significant.

Discussion
to our knowledge, this is the first meta-analysis on mnesic functions associated with the vulnerability to suicidal behaviour. overall, three domains (i.e., autobiographical memory reported as being over-general and less specific; long-term memory and working memory) were found to be altered in suicide attempters vs. patient controls, and in patient controls vs. healthy controls suggesting greater alterations in those with a vulnerability to suicidal behaviour than in those with co-morbid disorders but no history of suicidality. one domain (short-term memory) was only altered in the comparison between patient control groups and healthy controls suggesting that impairment in this domain is mostly related to co-morbid disorders (in this case, mainly major depression). Finally, no memory alteration was found to be specifically associated with the vulnerability to suicidal behaviour (i.e., in the contrast between suicide attempters and both patient and healthy control groups, but not between patient and healthy controls).
Some limitations of this meta-analysis need to be highlighted. First, studies included in this review examined various populations in terms of sociodemographic and/or clinical variables. Most studies examined middle-aged samples, although some focused on elderly participants. Some studies were only conducted in males while most included both genders. diagnoses varied from bipolar disorder, major depressive disorder (or a combination of both subtypes of mood disorder), to schizophrenia and borderline personality disorder. also, some studies were conducted in patients who were acutely depressed, while others focused on those in remission. Furthermore, participants in some studies were on medica- (Richard-devantoy et al. 2012;Jollant et al. 2013;Keilp et al. 2013) were likely responsible for the heterogeneity related to working memory. after excluding these studies, the heterogeneity disappeared, and results remained significant. no specific study explained the heterogeneity related to the short-term memory analyses. Between-group difference in long-term memory became significant after excluding the two studies responsible for the heterogeneity (Keilp 2001(Keilp , 2013. the Funnel plots were reasonably symmetrical for all types of memory, suggesting a low risk of publication bias. Moreover, the more conservative egger's regression intercept suggested no publication bias.

Suicide attempters vs. healthy controls
Suicide attempters had significantly more general and less specific autobiographical memories, and worse working memory and short-term memory performance than healthy controls. the Fail-Safe n was, respectively, 63, 171, 142, and 40. heterogeneity exceeded that expected by chance for all domains. one study (Maurex et al. 2010) was likely responsible for the heterogeneity related to the general autobiographical memory performance. after excluding this study, results became not significant, and the associated Funnel plots were reasonably symmetrical. egger's regression intercept test suggested no publication bias. two studies (Williams et al. 1996;Maurex et al. 2010) were likely responsible for the heterogeneity related to the specific autobiographical memory. after excluding these studies, results remained significant. three studies (Richard-devantoy et al. 2012;Jollant et al. 2013;Keilp et al. 2013) were likely responsible for the heterogeneity related to the short-term memory performances, and one study (Kim et al. 2003) for the heterogeneity related to the long-term memory performances. after excluding these studies, results remained significant, funnel plots were reasonably symmetrical. egger's regression intercept test suggested no publication bias. Finally, two studies (Richard-devantoy et al. 2012;Jollant et al. 2013) were likely responsible for the heterogeneity in working memory. after excluding these studies, results remained significant, funnel plots were reasonably symmetrical, and egger's regression intercept test suggested no publication bias.

Patient controls vs. healthy controls
patient controls had significantly more general autobiographical memory, and worse working memory, and short-and long-term memory than healthy controls. the Fail-Safe n could not be calculated for memory efficiency, may also prevent these individuals from accessing various stored material acquired with experience. in addition, literature suggests that the neural system implicated in imagining the future overlap with the one involved in episodic memory (addis et al. 2007). Reduced memory for long-term experiences may therefore impact the ability both to solve problems (evans et al. 1992; pollock et al. 2001) and to see beyond a problem to form a longterm perspective. consequently, this may lead to a feeling of entrapment and hopelessness combined with the continuation of the problem. one brain structure associated with both longterm memory deficits and suicide diathesis may be the hippocampus. theory of hippocampal function operates as a single auto-association network to enable rapid associations between any spatial location and a reward, and then to provide for recalling memory from any part (Rolls 2013). Furthermore, future event construction, as discussed above, uniquely engaged the right hippocampus, possibly as a response to the novelty of these events (addis et al. 2007). hippocampus is also a major inhibitor of the hypothlamic-pituitary axis that underlies the glucocorticoïd stress system. Regarding suicidal behaviour, postmortem studies revealed several biochemical alterations in the hippocampus of suicide completers (labonte et al. 2013). notably, childhood adversity was associated with epigenetic alterations affecting the glucocorticoid receptor expression in hippocampal neurons (McGowan et al. 2009) suggesting developmental impairments. dysregulation of the glucocorticoid system has been reported in suicide attempters and predicts future suicide (coryell et al. 2001), a dysfunction that may be related to hippocampal alterations. in turn, deactivation of the hippocampus has been correlated with the release of tion while in other studies they were not. Moreover, it was not possible to separate memory function from low and high lethality suicide attempters, and from violent and non-violent suicide attempters, because too few studies distinguished these groups. of note, we were able to rule out the effects of several variables (including, for example, the intensity of depressive symptoms, age, and gender). Meta-analyses have often been criticized for combining heterogeneous studies, for their potential of publication bias, and for the inclusion of poor-quality trials. in the present study, however, these concerns were addressed by the objective examination of both publication bias and heterogeneity. an additional limitation is the small number of studies in some analyses, notably on autobiographical memory, and this may have caused an artificially high effect size (Button et al. 2013). one final limitation is the combined analysis of multiple tasks with the assumption that they measure the same core component of cognition. this may partly explain the heterogeneity observed in several contrasts.
Keeping these limitations in mind, our findings suggest that a set of memory deficits may be in play in the suicidal crisis with some deficits being more particularly involved in facilitating the suicidal act. this is notably the case of autobiographical and long-term memory, two types of memory associated with past experiences. Reduced specificity of autobiographical memory may serve an adaptive function (Raes et al. 2006), limiting access to negative affects and distress associated with some recollections ). For instance, over-general memory has been associated with a history of trauma and abuse ), a significant risk factor for suicidal acts. however, a higher tendency for more general memory retrieval at the expense of specificity, combined with decreased long-term [] contrast became non-significant after excluding studies responsible for heterogeneity.

Conclusion
our findings, although limited by several shortcomings, suggest that memory deficits may be a key player in the cognitive vulnerability to suicidal behaviour. More studies are necessary to confirm these initial results, to examine how memory acts in combination with other cognitive domains to increase the risk of suicidal act, and to understand their neural and molecular basis. Ultimately, cognitive deficits may be the target of future interventions aiming at reducing suicide risk.
cortisol in response to stress (pruessner et al. 2008). in vivo, recall of suicide action through scripts was associated with increased activity in the hippocampus of suicide attempters (Reisch et al. 2010) while less activation at rest was found in the right parahippocampal gyrus of suicide attempters compared to patient controls (Reisch et al. 2010;Wagner et al. 2012;Fan et al. 2013). decreased volume of the right parahippocampal gyrus was additionally found in high-vs. low-lethality suicide attempters with borderline personality disorder (Soloff et al. 2012). hippocampus may therefore be implicated in several long-term memory deficits and, indirectly, in the high sensitivity to stress in patients at-risk for suicide. We also found that working memory was altered in suicide attempters compared to patient controls, in both verbal and visual-spatial tasks (Keilp et al. 2013). according to Baddeley (1992), working memory involves executive control and visuospatial skills, which manipulates visual images, and phonological loop, enabling to store and rehearse speechbased information. We have recently shown that, in suicide attempters, working memory deficits were negatively correlated with cognitive inhibition (Richarddevantoy et al. 2013). although working memory deficits may also impact decision-making performance (Bechara et al. 1998), we did not confirm this association in suicide attempters (Richard-devantoy et al. 2013). thus, deficits in working memory may particularly impact functions implicated in executive control in suicide attempters.
interestingly, memory in suicide attempters may play an opposite role in major depression and in schizophrenia. in major depression, suicide attempters had greater memory impairment compared to patient controls, whereas the opposite was found in schizophrenia. Similar findings have been reported in other cognitive domains. Suicide attempters with schizophrenia tended to outperform non-attempters in executive functioning, namely attention and verbal fluency (nangle et al. 2006), and cognitive flexibility (Kim et al. 2003;nangle et al. 2006), whereas depressed suicide attempters had worse executive performance compared to patients without such a history (Richard-devantoy et al. 2013). Similarly, low iQ has been associated with suicidal behaviour (Sorberg et al. 2013), but high iQ was associated with suicidal behaviour in psychosis (Batty et al. 2010). hence, contrary to the transnosographic model of suicidal behaviour favoured so far and supported by clinical and biochemical studies (Mann 2003), neuropsychological studies suggest potentially different mechanisms between mood and psychotic disorders. one hypothesis could be that these improved cognitions increase insight, a significant risk factor for suicide in schizophrenia (lopez-Morinigo et al. 2012). More investigation is definitely needed to clarify this issue.