Anger-Related Affect and Suicidal Thoughts and Behaviors: A Systematic Review and Meta-Analysis

Abstract

Objective:

Growing evidence indicates anger-related affect (i.e., anger, hostility, and irritability) is a transdiagnostic risk factor for suicide. The goal of this meta-analysis was to systematically review the literature and calculate the effect size of these relationships.

Method:

We examined the relationships between suicidal thoughts and behaviors and anger, irritability, and hostility with data from 107 published studies (total N = 141,434).

Results:

Our analyses uncovered a moderate-sized average correlation (r = .31) between anger-related affect and suicidal thoughts and behaviors. There were no significant differences between the effect size estimates of the three separate anger constructs. Studies assessing suicide attempts were associated with significantly lower effect sizes than studies assessing death by suicide or suicidal ideation. Studies that assessed suicidal thoughts and behaviors using a clinical interview were associated with lower effect size estimates than those assessing via self-report or chart review. Additionally, studies that assessed current suicidal thoughts and behaviors were associated with higher effect size estimates versus those that assessed lifetime occurrence. Studies that assessed suicide in an outpatient setting were associated with lower effect sizes than those in community, inpatient, or forensic settings. There were no other significant moderators (i.e., sociodemographic or study design variables) of the relationship between overall anger and suicide.

Conclusions:

Findings highlight the importance of anger as a potential risk factor for suicide. Future research is needed to examine these relationships longitudinally and to look at different facets of anger (e.g. experience vs. expression).

There is accumulating evidence suggesting that anger may be a transdiagnostic risk factor for suicide. Many psychiatric disorders associated with increased suicide risk involve dysregulated anger [e.g., borderline personality disorder, posttraumatic stress disorder (PTSD), intermittent explosive disorder; Nepon et al., 2010; Nock et al., 2010; Nock et al., 2014]. This association between anger and suicide has been observed across populations and study designs. For example, among psychiatric inpatients, anger during hospitalization was found to prospectively predict a non-fatal suicide attempt (SA) in the year following discharge, above and beyond other well-established risk factors such as psychiatric symptoms and recent SA (Sadeh & McNiel, 2013). Retrospective studies of individuals who died by suicide have also found evidence of anger difficulties in their medical records (Dobscha et al., 2014; Morgan & Priest, 1984). Cross-sectional studies have found that anger is associated with both suicidal ideation (SI; Hawkins & Cougle, 2013; Hawkins et al., 2014; Horesh et al., 1997; Jang et al., 2014; Kachadourian et al., 2018; Kotler et al., 1993; Wilks et al., 2019) and SA history (Hawkins & Cougle, 2013; Hawkins et al., 2014). Anger has also been implicated as a risk factor for suicide in longitudinal studies. For example, in a large, nationally representative sample, problematic anger was associated with SI and SA three years later, even when controlling for common risk factors such as history of suicidal thoughts and behaviors, psychiatric comorbidities, and demographics (Dillon et al., 2020).

Constructs Related to Anger (Hostility and Irritability)

There has also been research examining the relationships between suicide and constructs closely related to anger, such as hostility and irritability. These are considered distinct concepts, though there is considerable overlap among them. Anger is an emotional state composed of physiological, cognitive, and affective elements that often emerges in response to a perceived threat or thwarted goal (Harmon-Jones, 2008). Hostility is a general predisposition or tendency to mistrust and dislike others (Norlander & Eckhardt, 2005). Hostility is attitudinal, and hostile individuals demonstrate a proneness to cynicism and propensity to interpret others’ behaviors as hurtful, which often results in a combination of anger and disgust (Orth & Wieland, 2006; Ramirez & Andreu, 2006). Irritability is defined as a sensory sensitivity that predisposes individuals to respond to sensory stimuli (e.g., noise, heat) with anger or aggressive behavior (Grondal et al., 2023). In general, anger is an emotion, whereas hostility and irritability are predispositions. Individuals with high levels of irritability or hostility have an increased tendency to experience anger across situations. Anger and hostility can also be precursors to aggressive behavior (Wilkowksi et al., 2010), and there are measures of aggression that conceptualize anger and hostility as emotional and cognitive aspects of aggression, respectively (Buss & Perry, 1992; Buss & Warren, 2000; Morey, 1991). Importantly, anger and hostility do not always lead to aggression and aggression can occur in the absence of anger and hostility; thus, these are related, yet separate constructs.

Further, there are several distinctions that have been made within the anger literature regarding different facets of anger (e.g., state anger, trait anger, anger expression; Spielberger, 1999). State anger is defined as the extent to which an individual is currently experiencing anger. Trait anger is the individual predisposition to experience elevations in state anger. Anger expression refers to how anger tends to be expressed when it is experienced. Anger-in is the tendency to hold in or suppress anger, whereas anger-out is the tendency to express anger verbally or physically. Anger-control is the tendency to control or manage angry feelings rather than expressing them. Broadly, these constructs can be divided into two categories: anger experience and anger expression. Anger experience includes state and trait measures of anger and refers to the extent to which an individual experiences the emotion of anger. Anger expression includes anger-in, anger-out, and anger-control, and refers to the way that an individual expresses their anger when it is felt. This latter construct is likely influenced by many other variables (e.g., impulsivity, substance use). Thus, the current review will focus on the associations between suicide and anger experience.

Hostility and Suicide

Across diverse samples, hostility has been associated with SI (Guidotti et al., 2024; Kachadourian et al., 2018; Martin et al., 2021; Palmu & Partonen, 2023) and SA history (Christodoulou et al., 2017; Mann et al., 1999; Michaelis et al., 2004; Oquendo et al., 2004). Notably, however, there have been exceptions. For example, among treatment-seeking veterans with PTSD, there was no association between hostility and SI when accounting for severity of PTSD and depressive symptoms (Wells et al., 2021), indicating that this relationship may be less straightforward and may differ among different populations.

Irritability and Suicide

Studies on the relationships between irritability and SI or SA have been similarly inconsistent. In one study of patients with anxiety and mood disorders, irritability was associated with SI and SA (Baldessarini et al., 2016). Other studies conducted in community samples with near-even gender distributions have also found irritability to be associated with SI (Cooper et al., 2015; Mitsui et al., 2017). However, in a study examining sex differences in a community sample of young adults, irritability was associated with SI in male, but not female participants (Skala et al., 2012). On the other hand, another study found female but not male individuals attempting suicide to have higher levels of irritability, though it is worth noting that just 19.6% of the sample was male (Ardani et al., 2017). A two-year longitudinal study of outpatients with bipolar I or schizoaffective disorder also found that irritability is associated with increased SI over time (Berk et al., 2017). The inconsistencies of these findings suggest that the association between irritability and suicide may vary based on the sample being examined (e.g., community vs. clinical), demographics (e.g., sex/gender), and suicide outcome examined (e.g., SI vs. SA). This highlights the importance of conducting moderator analyses to better understand the conditions under which anger-related affect is linked to suicidal outcomes.

Theoretical Basis for the Link between Anger and Suicide

The link between anger and suicide is consistent with two prominent conceptual models of suicide: fluid vulnerability theory (FVT; Rudd, 2006) and the interpersonal-psychological theory of suicide (IPTS; Joiner, 2005). FVT posits that risk for suicide fluctuates as a function of both chronic and acute risk factors. According to FVT, there are individual differences in both the propensity to be triggered to acute periods of SI and the capacity to recover from urges to die by suicide. As a result, anger-prone individuals may experience frequent situational triggers. Subsequent increased anger may in turn cause them to push away social support, increase isolation, and lead to heightened negative thoughts and feelings, putting them at further risk for SI. Additionally, it may be more difficult for anger-prone individuals to recover from these episodes (Tafrate et al., 2002; Wilkowski & Robinson, 2010).

FVT is not incompatible with the interpersonal-psychological theory of suicide (IPTS; Joiner, 2005; Wolfe-Clark, 2017), and researchers have suggested integration of the two approaches (Wolfe-Clark, 2017). IPTS posits that perceived burdensomeness (feeling one is a burden on others) and thwarted belongingness (feeling a lack of connection with others) put individuals at the greatest risk for SI. Individuals with heightened anger may experience frequent interpersonal conflict and relationship difficulties, putting them at increased risk of feelings of thwarted belongingness and perceived burdensomeness, and therefore increased risk of SI. There is evidence of an indirect relationship between anger and SI via both of these factors (Hawkins et al., 2014; Rogers et al., 2017). When ideation transitions to suicidal behavior, Joiner’s IPTS indicates an acquired capability for suicide that is garnered over the course of repeated exposure to painful stimuli. Hawkins et al. (2014) have demonstrated such a link between anger and the acquired capability for suicide through the mediation of repeated exposure to painful stimuli.

The Current Study

Despite mounting evidence of an association between anger (and the similar constructs of hostility and irritability) and SI and SA, a comprehensive and robust statistical estimate of the combined magnitude of this relationship has yet to be calculated. The purpose of the current meta-analysis was to calculate the effect size of the relationship between anger, hostility, and irritability with suicide variables (i.e., SI and SA). Based on variability in prior findings, we also sought to determine whether there were any demographic (e.g., military status, gender) or study design (e.g., timing or format of assessment) moderators of this association.

Method

The meta-analysis was pre-registered through Prospero on September 5, 2022 [CRD42022270153] and reported in accordance with PRISMA guidelines (Moher et al., 2009). See Supplementary Table 1 for concordance with PRISMA checklist for meta-analyses.

Search Strategy

A medical librarian with expertise in systematic searching composed a search utilizing a mix of keywords and subject headings to represent the concepts of suicide, anger, and adults. The databases MEDLINE via PubMed, Embase via Elsevier, Scopus via Elsevier, PTSD Pubs via ProQuest, and APA PsycINFO via EBSCO were searched from inception to September 30, 2021, with an update on November 16, 2023. All search results were compiled in EndNote and imported into Covidence for deduplication and screening. All search strategies are available in the supplementary materials.

This search resulted in the identification of 8,873 studies, from which 4,535 duplicates were removed (4,338 unique studies; see Figure 1 for PRISMA flow chart). Screening procedures were conducted using the web-based platform Covidence systematic review software. Titles and abstracts for these 4,338 studies were screened for relevance twice independently by seven of the co-authors (MG, KHD, TAP, STL, DM, PA, SS); conflicts were resolved by the lead or senior authors (KHD or TFH). Title and abstract screening resulted in the exclusion of 3,097 irrelevant studies. For the remaining 1,241 studies, full texts were retrieved, and screening of full text articles was conducted to make inclusion decisions. All full text articles were screened for relevance twice independently by seven of the co-authors (KHD, STL, MG, TAP, SS, TFH, DM). Conflicts were resolved by an individual who had not conducted the full text review in question (either KHD, TFH, or MG). Full text review resulted in the exclusion of 1,134 studies. A total of 107 studies (108 unique study samples) were included.

Figure 1.

Consort diagram.

Eligibility Criteria

The following criteria were used to determine inclusion at each stage:

Abstract Stage

For inclusion in this phase, articles were required to: 1) be written in English; 2) be published in peer-reviewed scientific journals; 3) provide quantitative results; 4) include participants (or a subset of participants separately reported) aged 18 years or older; 5) assess suicidal ideation, suicidal behavior/attempt, suicide risk, or self-directed violence; and 6) assess anger, hostility, irritability, or general negative affect.

Articles were excluded from this phase if they violated a condition for inclusion, or: 1) represented case study, case series, or review articles; or 2) if the research clearly exclusively targeted behavioral aggression (i.e., aggressive actions). If it was unclear how anger or aggression were operationalized within a specific study, it was included at this stage. If no abstract was provided, articles were included for review in the subsequent phase.

Full Text Stage

For inclusion in this phase, articles were required to: 1) be written in English with full text available; 2) be peer-reviewed; 3) present empirical/quantitative results; 4) include participants (or a subgroup reported separately) aged 18 years or older; 5) assess the construct of anger, hostility, or irritability; 6) assess the construct of suicidal ideation, behavior, or risk; and 7) report a bivariate measure of association between a variable of anger, hostility, or irritability and at least one suicide variable (e.g., correlation, t-test, odds ratio).

Articles were excluded from this phase if they violated a condition for inclusion or: 1) were duplicated articles (including multiple articles that published on overlapping samples); 2) were non-empirical (e.g., review paper, book chapter, qualitative analysis); 3) were not peer-reviewed (e.g., opinion papers, dissertations, white papers, etc.); 4) assessed behavioral aggression rather than an anger-related emotion; 5) assessed nonsuicidal self-injury rather than suicidality; or 6) did not report a measure of association between suicide variables and at least one of the anger-related variables. In the case of multiple articles that were published on overlapping samples, effect sizes were retained for the largest sample size.

Anger Domains.

In the full text stage, articles were included if they measured anger experience, hostility, or irritability. Each study measure was examined to ensure that it assessed for one of these emotional or attitudinal constructs, rather than a behavioral measure of aggression (e.g., number of aggressive acts). There were several studies that included an aggression measure (e.g., Buss-Perry Aggression Questionnaire) but were included because they reported results for subscales that were assessing anger or hostility. Measures of behavior or anger expression were not included. See Supplementary Table 3 for list of anger measures identified.

Suicidal Thoughts and Behavior Domains.

In the full text stage, articles were included if they measured suicidal thoughts or behaviors. They were broadly divided into three groups: SI, SA, and death by suicide. Based on our review, articles that assessed “suicide risk” were primarily focused on thoughts/ideation and were, thus, included as SI. See Supplementary Table 3 for list of measures of SI, SA, or death by suicide.

Effect Size Relationships

Overlapping samples were identified by: 1) cross-referencing all authors and research groups, and 2) cross-referencing grant numbers. Samples with potential overlap underwent a second round of full-text inspection and consensus. Samples with a probable degree of overlap were dealt with in the following ways: estimates from the largest sample were retained if overlapping samples presented identical suicide and anger relationships (12 studies were excluded for presenting identical suicide and anger relationships from a smaller sample); overlapping effects were included if they presented unique suicide-anger relationships but these effects were considered as part of the same sample so dependency among effects could be modeled.

One hundred and ten studies met inclusion criteria for the present review (see Supplementary Materials for a full reference list of included studies and Supplementary Table 2 for study characteristics). One study included two independent samples (Jha et al., 2020) for a total of 108 study samples. If studies reported multiple suicide-anger outcomes within the same domains (e.g., multiple effect sizes for SA and irritability), correlations were pooled. Studies were allowed to contribute multiple suicide-anger outcome effects across unique domains (e.g., SA and irritability; SA and hostility) resulting in a total of 140 effect sizes.

Coding Procedure

Article coding categories were defined prior to article review and were based on categories recommended by PRISMA guidelines (Moher et al., 2009). Articles were coded for: year of publication, country of publication, type of study design (between group comparisons or within participants), type of effect size reported (e.g., t-test, Pearson’s r), sample size, type of sample (e.g., civilian, Veterans), percent male, percent White, mean age, study setting (e.g., outpatient, inpatient), study inclusion criteria, sample comorbidities, research focus of the study (i.e., anger, suicide, or both/equal), type of anger measure and timeframe assessed (i.e., current or general), and type of suicide measure and timeframe assessed (i.e., current or lifetime).

For ease of interpretation, only the baseline effect size was utilized for studies with a longitudinal design, so all effect sizes are cross-sectional. “Within-participant effect sizes” refer to correlation coefficients. “Between-subject effect sizes” refer to odds ratios comparing groups (e.g., individuals with SI vs. individuals without SI). The research focus of the study was defined as a rating of whether the study focused primarily on anger, on suicide, or both constructs equally. This information was determined by whether either, both, or neither construct was mentioned in the title and/or abstract. Timeframes for anger measures were divided into two categories (current or general), based on whether the instructions of the measure identified a timeframe to consider when filling out the items (current, e.g., “in the past week”) or asked participants to consider how they felt more generally, without a timeframe provided (general). Timeframes for the suicide measures were also divided into two categories (current or lifetime) based on whether suicide measures covered the participants entire lifetime (lifetime) or a more recent timeframe (current). Supplementary Table 2 provides coding categories and corresponding study values.

All articles were double coded by two independent coders. Discrepancies in objective categories (e.g., study sample size) were reconciled by consulting original articles. Discrepancies in more subjective categories (e.g., research focus of the study) were reconciled through a consensus meeting.

Statistical Methods

Bivariate relationships between suicide and anger constructs formed the indices of effect size. Most studies presented effect sizes as correlations (54.00%, n = 77). Subsequently, correlation coefficients (r) were chosen as the indices of effect size with follow-up sensitivity analyses performed to determine if type of effect size reported or type of study design significantly predicted effect size estimates. All effect sizes were converted using Fisher’s r-to-z transformation to stabilize variance and estimate confidence intervals (Fisher, 1922). Effect sizes were then backtransformed to correlation coefficients (r) for ease of interpreting the relationship between suicide and anger prior to all analyses (Viechtbauer, 2010).

A 3-level random-effects meta-analysis using restricted maximum likelihood (REML) was conducted to estimate a sample-weighted average effect size using the metafor package (Viechtbauer, 2010) in R version 4.4.0. A 3-level meta-analysis was chosen because 23.15% (n = 25) of the included study samples reported more than one effect size (i.e., multiple types of suicide or anger outcomes). Random effects meta-analyses account for heterogeneity introduced by different methods or samples and allow for multiple effect estimates from a single sample (Hasselblad & Hedges, 1995; Hedges & Vevea, 1998; Viechtbauer, 2010). Dependent effect sizes (i.e., multiple outcome measures reported from the same sample) are nested within samples (level 2) and then pooled across all studies (level 3). This 3-level model, hereafter referred to as “multilevel”, allows for estimates of the heterogeneity quantified as the within (τ²₍₂₎) and between (τ²₍₃₎) variance in study samples. The I² statistic is also estimated and represents the proportion of variance in true effects out of total variance for each level of the model. The I² statistic is estimated for both the within study samples (I²₍₂₎) and between studies (I²₍₃₎).

An initial multilevel meta-analysis was conducted on all available effect sizes to estimate the overall relationship between all suicide and anger outcomes. Follow-up mixed effects analyses were conducted to examine moderators of effect size heterogeneity. Forest plots were generated to display summary effect sizes for ease of interpretation. Funnel plots and regression tests for funnel plot asymmetry were generated to examine publication bias (Egger et al., 1997).

Results

Sample Characteristics

The meta-analysis included 107 studies with 108 independent samples contributing a total of 140 effect sizes (k = 140) from 141,434 participants. Participants had a mean age of 52.74 years (SD = 14.53); 51.76% of the sample was male, and 65.41% was White. A notable minority of participants (15.75%) identified as military service members or veterans, and 1.49% were justice-involved (e.g., participated in a forensic setting). Most participants were recruited from community settings (76.93%) followed by outpatient or specialty clinic settings (13.99%), inpatient settings (3.69%), or forensic settings (3.68%).

Study Characteristics

Effects sizes were based on both within-subjects (54.00%, k = 77) and between-subjects (45.00%, k =63) designs. SI was the most common suicidal construct (63.57% of effect sizes, k = 89), followed by SA (32.86%, k = 46), and death by suicide (3.57%, k =5). Anger was the most common construct measured (45.71%, k = 64), followed by hostility (31.43%, k = 44), and irritability (22.86%, k = 32).

Moderator data were available for most studies, including year of publication (100%), country of publication (99.07%), sample size (100%), type of sample (100%), percent male (94.39%), mean age of sample (76.64%), study setting (100%), study inclusion criteria (92.52%), and research focus of the study (100%). See Supplementary Table 1 for moderator variables presented by study. Few studies reported the race and/or ethnicity of participants (31.78%) and/or mental health comorbidities (28.97%). Moderator analyses were conducted for overall anger and suicide construct relationships to ensure adequate power. To maximize observations in each moderator analysis, moderators were applied to overall models individually with Bonferroni correction for multiple comparisons.

There were 52 different measures of anger constructs and 47 different measures of suicide constructs (see Supplementary Table 3) precluding moderator analyses based on measurement instrument or approach. Instead, moderator analyses included the timeframe (i.e.., current, general, lifetime history) and source (e.g., self-report, clinical interview) of anger and suicide constructs. In terms of anger, most effect sizes were based on participants’ “typical” level of anger (71.43%, k = 100) followed by current levels of anger (28.57%, k = 40). Anger was most frequently assessed by self-report (91.43%, k = 128) and less frequently assessed by a clinical interview (7.14%, k = 10) or informant-report (1.42%, k = 2). In terms of suicide constructs, effect sizes captured current (53.57%, k = 75) or a lifetime history (46.43%, k = 65) of suicide constructs (i.e., SI, SA, or death by suicide). Suicide constructs were most frequently assessed by self-report (63.57%, k = 89) followed by clinical interviews (27.14%, k = 38), chart reviews (7.86%, k = 11), and informant-report (1.42%, k = 2).

Publication Bias

Publication bias was examined through funnel plot inspection (see Supplementary Figure 1) and a formal assessment of funnel plot asymmetry using a multilevel analogue of Egger’s test, which evaluates the relationship between the effect sizes and the inverse of standard errors. Inspection of the funnel plot and the multilevel analogue of Egger’s test did not reveal evidence of a small study effect or publication bias (b = −0.01, SE = 0.01, p = .91).

Results of the Multilevel Meta-Analysis

A random-effects meta-analysis yielded a moderate-sized average correlation between overall anger (i.e., anger, hostility, and irritability) and suicide constructs (i.e., SI, SA, or death by suicide; r = .31, 95% CI [.29, .34], p <.001) based on 140 effect sizes (see Figure 2). Heterogeneity within studies was 20.93% (τ²₍₂₎ = .004, I²₍₂₎ = 20.93), while heterogeneity between studies was 73.59% and substantially larger (τ²₍₃₎ = .013, I²₍₃₎ = 73.59, Q(139) = 3182.08, p <.001). Sensitivity analyses revealed that the type of study (i.e., between or within design) did not significantly influence effect size estimates (b = .01, SE = .03, Z = 0.35, p = .72). Similarly, the type of effect size estimate (e.g., t-test, r) did not significantly influence effect size estimates (b = .01, SE = .03, Z = .36, p = .72). See Figure 2 for a forest plot summarizing effect sizes by suicide and anger outcomes.

Figure 2.

Forest plots displaying correlation coefficients by suicide and anger constructs.

Note: Diamond polygon represents the summary correlation coefficient estimates among all anger and suicide constructs; Number of effect sizes (k) included in estimates; relationships with k ≤ 3 not presented; 95% CI = 95% confidence interval.

Results of moderator analyses are presented in Table 1. Of note, the anger construct assessed (i.e., anger, hostility, and irritability) did not moderate the effect size. Only four variables significantly moderated the relationship between anger and suicide outcomes. The type of suicide outcome assessed was a significant moderator of effect size; assessing SA (versus death by suicide or SI) was associated with a decrease in effect size (b = −.068 [95% CI −.124, −.013], Z = 2.42, p = .016). Assessing suicidal thoughts and behaviors with a clinical interview (versus self-report or chart review) was associated with a decrease in effect size estimate (b = −.094 [95% CI −.152, −.036], Z = 3.17, p = .002). The timeframe for assessing suicide constructs was also a significant moderator of effect size; assessing current suicide constructs (versus lifetime occurrence) was associated with an increase in effect size (b = .073 [95% CI .023, .125], Z = 2.82, p = .005). Finally, measuring the relationship between suicide constructs and anger in an outpatient setting (versus community, inpatient, or forensic) was associated with a decrease in effect size (b = −.079 [95% CI −.136, −.023], Z = 2.74, p = .006).

Table 1.

Results of moderator analyses of relationships between anger and suicide outcomes.

	Effect Size (k)	b [95% CI]	Z	p-value
Overall Effect Size	140	.31 [.29, .34]	-	<.001
Moderators (Reference Group)
Anger Outcome
Construct Assessed (Anger)	64
Irritability	32	−.009 [−.072, .052]	0.31	.759
Hostility	44	.022 [−.025, .069]	0.92	.358
Source of information (Self-Report)a	128
Clinical Interview	10	−.037 [−.139, .065]	0.71	.480
Timeframe Assessed (General)	100
Current	40	.039 [−.018, .096]	1.33	.183
Suicide Outcome
Construct Assessed (Ideation)	89
Attempt	46	−.068 [−.124, −.013]	2.42	.016
Death by Suicide	5	−.071 [−.222, .080]	0.92	.358
Source of Information (Self-Report)a	89
Clinical Interview	38	−.094 [−.152, −.036]	3.17	.002
Chart Review	11	−.026 [−.129, .076]	0.50	.618
Timeframe Assessed (Lifetime)	65
Current	75	.074 [.023, .125]	2.82	.005
Study Characteristics
Mean Age	115	.001 [−.001, .00]	0.80	.424
Percent of Male Participants	134	−.001 [−.001, .001]	1.04	.298
Country Published (USA, n = 70)	139	.021 [−.035, .078]	0.73	.465
Year Published	140	.001 [−.001, .003]	0.591	.554
Study Design (Within, n = 76)	140	−.009 [−.063, .044]	0.36	.720
Study Sample (Civilian)	96
Service Members/Veterans	33	−.030 [−.089, .029]	1.01	.313
Forensic	11	−.085 [−.205, .035]	1.39	.163
Setting (Community)	51
Outpatient	55	−.079 [−.136, −.023]	2.74	.006
Inpatient	23	−.047 [−.124, .023]	1.21	.228
Forensic	11	−.117 [−.238, .005]	1.88	.059
Research Focus (Anger & Suicide)	29
Anger	12	−.020 [−.126, .086]	0.37	.711
Suicide	99	.001 [−.058, .077]	0.27	.790

Note:

^aInformant-report used for 2 effect sizes, since k ≤ 3 not included in moderator analysis.

Discussion

Given the growing evidence indicating that anger-related affect (i.e., anger, hostility, and irritability) is a transdiagnostic risk factor for suicide, the goal of this meta-analysis was to examine the literature and calculate the effect size of these relationships. We included data from 108 published studies (total N = 141,434 participants) to examine relationships between anger, irritability, and hostility and suicidal thoughts and behaviors (i.e., SI, SA, and death by suicide). Our analyses uncovered a moderate-sized average correlation (r = .31) between anger-related affect (i.e., anger, hostility, and irritability) and suicidal thoughts and behaviors. There were no significant differences between the effect size estimates when looking at the three anger constructs separately, suggesting that the three constructs are similarly related to suicide.

Studies assessing SA were associated with significantly lower effect sizes than studies assessing death by suicide or SI. These findings indicate that anger is more strongly associated with increased ideation than attempts. Within the ideation-to-action framework (Klonsky & May, 2014), anger may be a stronger predictor of ideation, whereas other factors contribute more strongly to suicidal behavior. This finding is consistent with prior research that has found that experiencing emotion dysregulation is associated with thoughts of suicide, whereas difficulties managing and expressing those emotions are more strongly associated with suicidal behavior (see Law et al., 2015). It is possible that if we had only selected studies that examined anger expression rather than experience, we would have found a stronger relationship with SA. Further research is needed to examine this possibility. Regarding our finding that death by suicide had a higher association with anger, there were not enough studies that examined suicide death (i.e., only five) to be able to draw any firm conclusions about the strength of the effect size for these studies, so this result should be interpreted with caution.

We examined several potential moderators of the association between overall anger and suicidal thoughts and behaviors. There were moderation effects for source of suicide assessment, method and timeframe of suicide assessment, and setting of the study. Specifically, studies that assessed suicide constructs using a clinical interview were associated with lower effect size estimates than those assessing via self-report or chart review. This may be because individuals are more forthcoming about SI and SA on self-report questionnaires. Additionally, this may be due to the shared measure variance, considering that the majority of the anger-related measures were also self-report. Interestingly, studies that used chart review to assess suicide constructs primarily included samples of participants who were hospitalized or presenting to the emergency room for an SA. The fact that chart review-based studies were generally observed to demonstrate stronger associations between anger and suicidal thoughts and behaviors compared to studies of SA suggests that this relationship may be particularly strong in the acute aftermath of a SA. It is also possible that in these instances, anger was experienced as a result of the SA. Additionally, studies that assessed current suicidal thoughts and behaviors were associated with higher effect size estimates versus those that assessed lifetime occurrence of suicidal thoughts and behaviors, suggesting that increasing the time frame of the suicide assessment may dilute the association with anger.

Other potential moderators of effect size, such as source of anger assessment, anger timeframe assessed, age, gender, country of study, year of publication, study design (within or between), study sample (civilian vs. military vs. forensic), or research focus of the publication (anger vs. suicide vs. both) were not significant moderators of the relationship between overall anger and suicide. There were too many different measures of anger and suicidal thoughts and behaviors used to examine whether measurement instrument moderated the effect size estimate; however, it is possible this may have influenced effect sizes.

Our meta-analysis used a rigorous methodology. The literature search was conducted using five major databases (Embase, Medline, Scopus, PsycINFO and PTSD Pubs) and a thorough article selection and extraction process. Additionally, our entire process adhered to PRISMA guidelines. Nevertheless, there are some limitations. First, as stated above, there were too many anger assessment measures used to be able to examine whether the anger measurement instrument used impacted the observed relationship between anger and suicide constructs. Although we compared three anger-related constructs (anger, irritability, and hostility), additional facets of anger exist, for which there was insufficient data to include in present analyses. For example, there were not enough studies of anger expression (i.e., anger-in, anger-out, anger-control) to examine these constructs separately. These constructs could be differentially associated with suicide risk. Further research is needed to examine whether these additional facets of anger or specific anger measures are more strongly associated with suicide. Second, the majority of studies included used a cross-sectional design, which limited our ability to examine the directionality of the relationship between anger and suicide. Additional longitudinal research is needed to examine these relationships over time and explore the temporal dynamics that might influence these relationships. Another limitation is that we were unable to examine whether specific psychiatric disorders moderated the association between anger and suicidal ideation or attempts. Unfortunately, there was a significant amount of heterogeneity across studies in how psychiatric comorbidities were assessed and reported—if they were assessed and reported at all—which made it impossible to determine which participants met criteria for psychiatric diagnoses and which did not. Given this heterogeneity and the possibility that moderator analyses for specific psychiatric disorders would likely not reflect true clinical presentations, we were not able to complete this analysis. Therefore, we are unable to determine to what extent the association between anger and suicide is accounted for by comorbid depression and/or anxiety, for example. Rather than complete moderator analyses based on specific diagnoses, the severity of clinical presentation is better captured with the available data by examining the type of sample setting with inpatient and forensic settings likely reflective of more severe or complex psychopathology compared to community or outpatient settings. Future research should continue to examine the role of comorbid psychopathology in the association between anger and suicide.

The mean age of 52.74 is somewhat older than expected in studies of suicidal behaviors, which may be due to the use of weighting based on sample size. Some of the largest studies included in this meta-analysis were epidemiological, meaning the study samples were recruited to be representative of a population rather than a specific clinical concern, and accordingly also had higher mean ages (e.g., Komulainen et al., 2020, Palmu et al., 2023, Stefanovics et al., 2023).

Our findings have implications for understanding the association between anger and suicide. They confirm that there is a moderate relationship between these constructs, which may be stronger for ideation. This is consistent with prominent models of suicide (Joiner, 2005; Rudd, 2006) and indicates that, particularly if directionality of anger on SI can be confirmed, treatments for anger may be indicated as a means for reducing SI. In particular, treatments targeted at reducing the tendency to experience anger, hostility, and irritability may be helpful to reduce SI, rather than simply teaching patients how to better manage anger-related emotions that arise. Meta-analytic reviews of cognitive-behavioral interventions for anger have found moderate to large effect sizes for their effects on anger (Del Vecchio & O’Leary, 2004; DiGiuseppe & Tafrate, 2003). Modification of cognitive or attributional biases that contribute to anger may be a helpful target for reducing anger experience. Treatments that target hostile interpretation bias, paranoid ideation, or other negative attribution biases may help to reduce SI. Targeted interventions such as the Mobile Anger Reduction Intervention (MARI; Dillon et al., 2023) have shown initial promise for reducing hostile interpretation bias. Future research should examine whether treatments such as these are also associated with reductions in SI. Findings also highlight the need to broaden risk assessment for suicidal thoughts and behaviors to include anger. Increasingly, motivation for suicidal thoughts and behaviors is being assessed (e.g., Moselli et al., 2021) and future research should examine the potential role of anger as a motivation for these behaviors.

Conclusion

Our systematic review and meta-analysis revealed a moderate-sized average correlation between overall anger (i.e., anger, hostility, and irritability) and suicidal thoughts and behaviors (i.e., SI, SA, or death by suicide), with no differences between these three anger-related constructs. We found that studies assessing non-fatal SA were associated with significantly lower effect sizes than studies assessing death by suicide or SI. Studies that assessed suicidal thoughts and behaviors using a clinical interview were associated with lower effect size estimates than those that assessed suicidal thoughts and behaviors via self-report or chart review. Additionally, studies that assessed current suicidal thoughts or behaviors were associated with higher effect size estimates versus those that assessed lifetime occurrence of suicidal thoughts or behaviors. Studies that assessed suicidal thoughts and behaviors in an outpatient setting were associated with lower effect sizes than those in a community, inpatient, or forensic setting. There were no other significant moderators of the relation between overall anger and suicide. These findings highlight the importance of anger as a potential risk factor for suicide. Future research focused on the relationship between anger-related coping or expression and suicidal thoughts and behavior is still needed, as is research aimed at examining these relationships longitudinally.

Public Health Significance.

This study found a moderate-sized relationship between anger-related emotions and suicidal thoughts and behaviors. The relationship between anger and suicidal ideation was stronger than the relationship between anger and non-fatal suicide attempts. These findings suggest that anger may be an important risk factor for suicidal ideation.