Association of suicidal behavior with exposure to suicide and suicide attempt: A systematic review and multilevel meta-analysis

Background Exposure to suicidal behavior may be associated with increased risk of suicide, suicide attempt, and suicidal ideation and is a significant public health problem. However, evidence to date has not reliably distinguished between exposure to suicide versus suicide attempt, nor whether the risk differs across suicide-related outcomes, which have markedly different public health implications. Our aim therefore was to quantitatively assess the independent risk associated with exposure to suicide and suicide attempt on suicide, suicide attempt, and suicidal ideation outcomes and to identify moderators of this risk using multilevel meta-analysis. Methods and findings We systematically searched MEDLINE, Embase, PsycINFO, CINAHL, ASSIA, Sociological Abstracts, IBSS, and Social Services Abstracts from inception to 19 November 2019. Eligible studies included comparative data on prior exposure to suicide, suicide attempt, or suicidal behavior (composite measure—suicide or suicide attempt) and the outcomes of suicide, suicide attempt, and suicidal ideation in relatives, friends, and acquaintances. Dichotomous events or odds ratios (ORs) of suicide, suicide attempt, and suicidal ideation were analyzed using multilevel meta-analyses to accommodate the non-independence of effect sizes. We assessed study quality using the National Heart, Lung, and Blood Institute quality assessment tool for observational studies. Thirty-four independent studies that presented 71 effect sizes (exposure to suicide: k = 42, from 22 independent studies; exposure to suicide attempt: k = 19, from 13 independent studies; exposure to suicidal behavior (composite): k = 10, from 5 independent studies) encompassing 13,923,029 individuals were eligible. Exposure to suicide was associated with increased odds of suicide (11 studies, N = 13,464,582; OR = 3.23, 95% CI = 2.32 to 4.51, P < 0.001) and suicide attempt (10 studies, N = 121,836; OR = 2.91, 95% CI = 2.01 to 4.23, P < 0.001). However, no evidence of an association was observed for suicidal ideation outcomes (2 studies, N = 43,354; OR = 1.85, 95% CI = 0.97 to 3.51, P = 0.06). Exposure to suicide attempt was associated with increased odds of suicide attempt (10 studies, N = 341,793; OR = 3.53, 95% CI = 2.63 to 4.73, P < 0.001), but not suicide death (3 studies, N = 723; OR = 1.64, 95% CI = 0.90 to 2.98, P = 0.11). By contrast, exposure to suicidal behavior (composite) was associated with increased odds of suicide (4 studies, N = 1,479; OR = 3.83, 95% CI = 2.38 to 6.17, P < 0.001) but not suicide attempt (1 study, N = 666; OR = 1.10, 95% CI = 0.69 to 1.76, P = 0.90), a finding that was inconsistent with the separate analyses of exposure to suicide and suicide attempt. Key limitations of this study include fair study quality and the possibility of unmeasured confounders influencing the findings. The review has been prospectively registered with PROSPERO (CRD42018104629). Conclusions The findings of this systematic review and meta-analysis indicate that prior exposure to suicide and prior exposure to suicide attempt in the general population are associated with increased odds of subsequent suicidal behavior, but these exposures do not incur uniform risk across the full range of suicide-related outcomes. Therefore, future studies should refrain from combining these exposures into single composite measures of exposure to suicidal behavior. Finally, future studies should consider designing interventions that target suicide-related outcomes in those exposed to suicide and that include efforts to mitigate the adverse effects of exposure to suicide attempt on subsequent suicide attempt outcomes.


Conclusions
The findings of this systematic review and meta-analysis indicate that prior exposure to suicide and prior exposure to suicide attempt in the general population are associated with increased odds of subsequent suicidal behavior, but these exposures do not incur uniform risk across the full range of suicide-related outcomes. Therefore, future studies should refrain from combining these exposures into single composite measures of exposure to suicidal behavior. Finally, future studies should consider designing interventions that target suicide-related outcomes in those exposed to suicide and that include efforts to mitigate the adverse effects of exposure to suicide attempt on subsequent suicide attempt outcomes.

Author summary
Why was this study done?
• Exposure to suicidal behavior in others has been linked to increased risk of suicidal behavior, but it is not known whether the association differs between types of exposure (suicide versus suicide attempt) or different outcome measures of suicidal thoughts and behaviors.
• Distinguishing the relationships of different exposure types with outcomes is important for the development of targeted interventions and public health approaches to suicide prevention.
What did the researchers do and find?
• We conducted a systematic review and meta-analysis of 34 studies that investigated the independent associations between exposure to different types of suicidal behavior and subsequent suicide, suicide attempt, and suicidal ideation outcomes.

Introduction
Suicide attempt and suicide are leading causes of global morbidity and mortality. Approximately 800,000 people die by suicide annually [1], of which about one-third are under the age of 30 [2]. The prevalence of suicide attempt is significantly greater than that of suicide death and is associated with heightened risk of later death by suicide [3,4] as well as psychosocial adversities that persist later in life [5]. For every suicide death, it is estimated that approximately 135 people are affected [6]. Over the course of a lifetime, the proportion of people exposed to the suicide of a relative, friend, or acquaintance is approximately 21% [7]. Exposure to suicide has been linked to increased risk of physical disease and adverse mental health including depression, posttraumatic stress disorder, and complicated grief [8,9]. The deleterious effects associated with exposure to suicide may also render some people, particularly adolescents and young adults, at increased risk of suicide and suicide attempt [10]. Combined, the large number of people exposed to suicide and the potential increased risk of suicide-related outcomes (suicide, suicide attempt, and suicidal ideation) in others mean that exposure to suicide is a significant public health concern [1]. This is reflected in several national suicide prevention strategies that recommend postvention interventions for those bereaved by suicide [11], as well as several international frameworks for the prevention of suicide-related contagion, and the management of suicide and self-harm clusters [12][13][14][15]. These public health strategies have largely focused on exposure to suicide, despite a growing body of evidence that suggests that exposure to suicide attempt, the behavior most proximal to suicide, may also be associated with increased risk of suicide-related outcomes [16][17][18][19].
Distinguishing between the potential independent effects of exposure to suicide and suicide attempt is important since measures of morbidity and mortality have markedly different public health implications. Yet evidence regarding the independent effects of exposure to suicide and suicide attempt on subsequent suicide-related outcomes is unclear. A systematic review and meta-analyses by Geulayov and colleagues [20] showed that exposure to suicide and exposure to suicide attempt of a parent were associated with increased risk of suicide and suicide attempt in offspring. However, the authors pooled mean effect sizes across subgroups within studies and did not take into account the dependencies between effect sizes, an approach that may distort the results of the meta-analyses [21]. Another systematic review by Crepeau-Hobson and Leech [19] reported that both exposure to suicide and exposure to suicide attempt were associated with subsequent suicide-related behavior among friends or acquaintances. But the authors did not adequately control for studies that reported lifetime prevalence, leaving the causal direction between exposure to suicide attempt and subsequent suicide-related outcomes unclear.
Lack of guiding evidence has impeded translation of the evidence into practice. For example, it is not currently clear which populations may be at risk, nor whether the risk differs across outcomes involving suicide, suicide attempt, and suicidal ideation. Sveen and Walby [22] found inconclusive evidence supporting a relationship between exposure to suicide and increased risk of suicide-related behavior in others. However, the authors combined studies reporting exposure in relatives and friends or acquaintances, which may incur different suicide risk. More recently, systematic reviews that investigated exposure to suicide in friends and acquaintances have reported a positive association between exposure to suicide and subsequent suicide-related outcomes [19,23]. Yet, as noted previously, the causal direction between exposure and outcome measures were confounded by the inclusion of studies that reported lifetime prevalence of exposure and outcome measures. Lastly, some studies included outcome measures that combined suicidal ideation with suicide attempt [24,25] or combined exposure to suicide and exposure to suicide attempt as a composite measure of exposure to suicidal behavior [26][27][28]. Composite measures of exposure to suicidal behavior prevent us from identifying whether the observed effect is influenced by a true association or the result of a cumulative effect.
Consequently, the effects of prior exposure to suicide and suicide attempt on suicide-related outcomes have not been reliably quantified, and the factors that moderate this risk are not currently known. We therefore aimed to conduct a systematic review and multilevel meta-analysis investigating the independent association between prior exposure to suicide, suicide attempt, and suicidal behavior (composite measure-suicide or suicide attempt) and subsequent suicide, suicide attempt, and suicidal ideation in relatives, friends, and acquaintances. In doing so, we aimed to quantify the association between exposure to suicide and suicide attempt and the full range of suicide-related outcomes, and to identify whether factors such as relationship to the person who engaged in the initial suicidal act, age of the study population, and study design characteristics moderate this risk. By using multilevel meta-analyses, we were able to account for dependencies among multiple effect sizes taken from the same cohort within a study, an extremely common and challenging aspect of conducting meta-analyses of epidemiological studies [29].

Methods
This work adheres to PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [30] and MOOSE (Meta-analysis of Observational Studies in Epidemiology) [31] guidelines (S1 Text) and was prospectively registered with PROSPERO (CRD42018104629). Deviations from the protocol include the use of exposure to suicidal behavior (composite) and statistical analyses using multilevel meta-analyses. The association between exposure to suicide and suicide attempt and grief and mental health outcomes will be reported in a separate systematic review and meta-analysis.

Study selection and eligibility criteria
Eligible studies reported dichotomous events (both the exposure and outcome were reported as having occurred or not occurred, yielding a 2 × 2 matrix) or odds ratios (ORs) for exposure to suicide, suicide attempt, or suicidal behavior and subsequent suicide, suicide attempt, or suicidal ideation. Exposure to suicide, suicide attempt, or suicidal behavior was determined from self-reported measures, informant interviews, official records (such as hospital admission records), or data linkage to death certificates. Outcomes involving suicide, suicide attempt, or suicidal ideation were determined from self-reported measures, informant interviews, or official records, such as death certificates, coroner reports, or hospital admission records. Cohort, case-control, and cross-sectional study designs were eligible if the study was reported in a peer-reviewed journal and the temporal sequence between the exposure and outcome was specified. For cross-sectional studies, the temporal sequence between exposure and outcome was established if the outcome measurement occurred after the exposure (e.g., the study asked participants if they had made a suicide attempt after exposure to the suicide of another). Participants of any age who were exposed to prior suicide or suicide attempt were eligible if the sample was mainly, or solely, drawn from the general population, as opposed to a clinical or other high-risk population (e.g., inpatients or prison detainees). Eligible control groups included individuals who did not report prior exposure to suicide, suicide attempt, or suicidal behavior in others.
Studies were excluded if findings from a non-exposed (control) group were not reported, or the control group was composed of participants exposed to other modes of death (e.g., accident or natural causes). Studies that reported estimates of lifetime prevalence as well as studies that did not establish the temporal sequence between exposure and suicide-related outcomes (e.g., the study reported 12-month prevalence of the outcome, but prior exposure to suicide was not indicated) were excluded. Finally, studies that reported outcomes following exposure to media reports of suicide (including fictional and non-fictional portrayals) or non-suicidal self-injury were excluded.

Data collection and coding
Two independent reviewers (NTMH and KK) extracted data using a standardized data collection form. A description of the a priori moderators of risk included in the study are presented in S1 Table. Dichotomous data were favored over ORs. When dichotomous events were not available, unadjusted ORs were recorded. For studies with multiple follow-up time points, only data from the longest time point were extracted [32]. Studies that included participants from the same population during overlapping time periods (e.g., nationwide data registry studies that reported suicide deaths from overlapping time periods) were included only if the studies reported different relationships (e.g., relatives and friends and acquaintances) or different suicide-related outcomes. When studies combined measures of exposure to the suicide of a relative or friend, we contacted primary authors for disaggregated data. If these data were not available, the relationship between the exposed individual and the individual(s) who engaged in suicidal behavior was determined by a majority rule (the relationship that occurred most frequently as indicated in >50% of the total sample). Similarly, if the age of participants included a combination of youths and adults, the age of the population was categorized in favor of the age group that exceeded 50% of the overall population. Study-level data are provided as S1 Data.

Multilevel meta-analysis rationale and data analysis
Since 16/34 (47%) studies reported multiple exposure and/or outcome measures in the same sample of participants, the assumption of independent estimates for a traditional meta-analysis was not met. We therefore used a 3-level meta-analysis, which parallels traditional random effects meta-analyses. The main difference is that dependent effect sizes (due to multiple subgroups or outcome measures within studies) are nested within studies (level 2) before these are pooled across studies (level 3). Thus, t 2 ð2Þ is the variance within studies while t 2 ð3Þ is the variance between studies. This approach allows for the investigation of heterogeneity not only between but also within studies [33]. For clarity, we use the general term "multilevel" throughout to describe our analyses.
We conducted a multilevel meta-analysis with the maximum likelihood estimation method using the metaSEM package [34] for R version 3.6.0. For the main analysis, we used dichotomous event data to calculate the pooled OR with the accompanying 95% confidence interval (CI) for risk of suicide, suicide attempt, and suicidal ideation within exposed and non-exposed individuals. When event data were not available, we used unadjusted ORs. Meta-analyses were conducted separately for exposure to suicide, suicide attempt, and suicidal behavior. Heterogeneity was quantified as variance in true effects within (t 2 ð2Þ ) and between (t 2 ð3Þ ) studies. We also report the I 2 statistic, which represents the proportion of variance in true effects out of total variance for each level (i.e., I 2 ð2Þ and I 2 ð3Þ ), along with its 95% confidence interval. Maximum likelihood mixed-effects analyses were used to examine effect moderators via subgroup analysis and to explain heterogeneity (quantified as R 2 ) for each level. Since the multilevel model does not provide study-level effect estimates, forest plots present the mean OR of each study but report the pooled 3-level estimate. Small study effect ("publication bias") was assessed by visually inspecting funnel plots of mean log ORs against standard error for asymmetry [35]. When at least 10 studies were available for analysis, we formally assessed funnel plot asymmetry using a multilevel analogue of Egger's test of the intercepts [36].

Risk of bias and quality assessment
Study quality was assessed using the National Heart, Lung, and Blood Institute quality assessment tool for observational studies [37]. The original tool contains 14 criteria that determine potential sources of bias in the study population and selection of participants, outcome and exposure measurement, blinding, confounding, and attrition. An overall rating of "good," "fair," or "poor" was provided for each independent study. Three independent reviewers conducted assessments (NTMH, AP, and AC), and any discrepancies were settled through discussion and finalized by the primary author (NTMH).

Study selection
The initial search identified 21,868 records, of which 8,320 were duplicates. A total of 13,548 records were screened based on title and abstract (Fig 1). The full-text versions of 760 records were assessed, 10 of which were obtained from searching the reference lists of existing reviews. The authors of 6 studies were contacted [38][39][40][41][42][43], and information or additional data provided for 2 studies [39,41]. A total of 167 records reported outcomes relating to suicide, suicide attempt, or suicidal ideation. Of these, 73 articles reported lifetime prevalence estimates, 35 studies involved overlapping populations or superseded time points, and 2 studies did not report ORs or accompanying effect sizes: These articles were therefore excluded from the metaanalysis. One study [44] was excluded because it reported an OR of 36.4, and 1 study [45] was excluded because it reported an OR of 18; both studies were prone to artifacts introduced by quasi-separation (S2 Table; S3 Text). The final dataset included 34 independent studies, which comprised 71 effect sizes (exposure to suicide: k = 42 across n = 22 studies; exposure to suicide attempt: k = 19 across n = 13 studies; exposure to suicidal behavior: k = 10 across n = 5 studies).

Study quality
Studies were most commonly rated fair (13/34) and good (13/34), followed by poor (8/34; S3  Table). The 13 good-quality studies tended to comprise cohort or case-control study designs and had clearly defined and valid exposure and outcome measures that were verified using official hospital or death records. The 8 studies that were rated poor tended to combine  Official records: Cause of death register. Exposure(s) determined by ICD codes for suicide and self-inflicted injury (ICD-8/9: E950-E959), intentional self-harm (ICD-10: X60-X84), and sequelae of intentional self-harm (ICD-10: Y870).
All participants were hospitalized for deliberate self-harm at the time that prior exposure was measured. exposure to suicide and suicide attempt into a composite measure of exposure to suicidal behavior, did not provide adequate definitions of exposure to suicide or suicide attempt, and did not provide information on case ascertainment for suicide-related outcomes.

Discussion
Based on findings from 34 studies of mostly good and fair quality, encompassing 13,923,029 participants and 71 effect sizes, we found that prior exposure to suicide was associated with significantly greater odds of suicidal behavior (suicide or suicide attempt; OR = 2.94). Results of the moderator analysis revealed that prior exposure to suicide was associated with 3.23-fold increased odds of suicide and 2.91-fold increased odds of suicide attempt, while there was no evidence of an association between exposure to suicide and subsequent suicidal ideation. These findings remained robust across cohort, case-control, and cross-sectional studies, as well as exposure and outcome measurements encompassing informant interview, self-report, and official records (e.g., coroner reports, hospital admission records, or data linkage with birth and death registries).
Exposure to suicide attempt was associated with increased odds of suicidal behavior (OR = 2.99). However, moderator analyses revealed that the association of exposure to suicide attempt with suicide-related outcomes was significant only for suicide attempt (OR = 3.53), not for suicide death (OR = 1.64). These findings were demonstrated across 19 effect sizes from 13 studies of mostly fair quality, and corroborated by 3 large population-based studies using data linkage or hospital admission records for suicide attempt [48,59,62]. Exposure to suicidal behavior (suicide or suicide attempt) was associated with a 2.58-fold increased odds of suicidal behavior, but moderator analysis revealed that this was significant only for outcomes relating to suicide death (OR = 3.83), not suicide attempt (OR = 1.10). These findings were demonstrated across 10 effect sizes from 5 studies, the majority of which involved psychological autopsy methodologies.
Our analyses update and further specify the findings from previous systematic reviews, which included estimates from studies reporting lifetime prevalence or did not differentiate between the independent effects associated with exposure to suicide and exposure to suicide attempt [19,22,23]. The finding that exposure to suicide was associated with an increased odds of suicide and suicide attempt-in contrast to exposure to suicide attempt, which was associated with an increased odds of suicide attempt only-indicates that exposure to suicide and suicide attempt do not incur uniform risk across the range of suicide-related outcomes. This was corroborated by our analysis of exposure to suicidal behavior, which found that this composite measure was associated with increased odds of suicide but not suicide attempt, a finding that was inconsistent with our separate analyses of exposure to suicide and exposure to suicide attempt. Taken together, the present findings raise questions about the conceptual value of combining suicide and suicide attempt as a composite measure of suicidal behavior, and suggest that future research and public health policies should refrain from combining these exposures and outcomes into 1 composite measure of suicidal behavior.
Evidence from 2 studies [46,57] suggests that exposure to suicide may be associated with increased risk of suicidal ideation, especially in older adults [46]. Conversely, results from a single cohort study in youths [57] indicate higher risk for suicide attempt than for suicidal ideation, pointing once more to lack of uniformity across populations and outcomes. Moreover, theoretical and empirical accounts suggest that while exposure to suicide may contribute to subsequent suicidal ideation to some extent, its effect on people with a history of suicidal ideation may be more pronounced [71], as this experience might reduce cognitive and practical barriers to acting on one's suicidal thoughts [46,72,73]. A more comprehensive look at this interaction may have important practical implications for developing specific interventions for this high-risk population, in particular interventions guided by the "ideation-to-action framework" [71] that aim to reduce acquired capability for suicidal behavior among individuals exposed to suicide.
The increased risks associated with exposure to suicide for outcomes relating to suicide and suicide attempt in the current meta-analyses suggest that further consideration should be given towards developing interventions that target suicide-related outcomes in those bereaved by suicide. To date, interventions targeting those exposed to suicide have largely focused on bereavement-related factors such as grief, reduced social support, and stigma [74,75]. Although previous studies have shown that these factors are elevated among those bereaved by suicide as opposed to other modes of death, there remains a dearth of studies that investigate the effectiveness of interventions on suicide and suicide attempt behavior. A recent review by Andriessen and colleagues [74], for example, found 3 controlled studies [76][77][78] that investigated the effectiveness of an intervention on suicidal ideation and found no studies that included outcomes related to suicide or suicide attempt.
Although we did not observe a significant association between exposure to suicide attempt and subsequent suicide, the specific relationship between exposure to suicide attempt and subsequent suicide attempt is noteworthy, since suicide attempt is associated with significant disruptions to an individual's milieu, and has been linked to adverse psychosocial and mental health stressors that persist later in life [5]. The findings from our analysis of exposure to suicide attempt also provide some insight into the mechanisms underlying the observed association between exposure to suicide and exposure to suicide attempt and the suicide-related outcomes. Arguably, the absence of bereavement-related factors and the specific association between exposure to suicide attempt and subsequent suicide attempt support the hypothesis that suicidal individuals may model, or imitate, suicide-related behavior that they see in others [10]. An imitation model is consistent with previous studies that have shown that increased risk of suicide-related behavior following exposure to both suicide and suicide attempt is not significantly moderated by preexisting risk factors such as depression, anxiety, and hospital admission for mental health [79,80]. The finding that exposure to suicide is associated with significant increased odds of suicide attempt is important since public health approaches for the prevention of behavioral contagion of both suicide and suicide attempt, such as frameworks for the prevention of suicide and self-harm clusters [12][13][14][15], have focused largely on mitigation efforts following exposure to suicide and therefore may benefit from the inclusion of exposure to suicide attempt in future mitigation efforts.

Limitations
The current systematic review and meta-analysis is the first to our knowledge to quantify the association between exposure to suicide and suicide attempt and the full spectrum of suicide- related outcomes and has many strengths, including the use of multilevel meta-analysis, the large sample size, and the exclusion of estimates of lifetime prevalence that do not take into account the temporal sequence between exposure and suicide-related outcomes. Despite this, several limitations exist. Whilst we conducted an extensive search of 21,868 records, there is the possibility that some relevant studies were not detected. Such studies are likely to create a bias towards the null (i.e., the exposure not having a significant effect). This is a limitation that is common to many systematic reviews and was mitigated to the best of our ability through adherence to a screening protocol developed a priori.

PLOS MEDICINE
Association of suicidal behavior with exposure to suicide and suicide attempt: A multilevel meta-analysis Furthermore, since most studies adjusted for different covariates, we restricted our analysis to unadjusted events and ORs. Whilst this is consistent with previous meta-analyses in the field [81,82], it meant that we could not investigate other risk factors, such as frequency of exposure, duration since exposure, and baseline mental health diagnoses, and how these might moderate the association between exposure to suicide and suicide attempt and suicide-related outcomes. For example, a previous systematic review on pre-and post-loss features of suicide bereavement in young people found evidence of a cumulative effect of exposure to suicide on subsequent suicide risk [83]. In the present meta-analysis, 2 out of 34 studies included in our analyses provided separate estimates for multiple exposures to suicide [62] and suicide attempt [48]. In 1 study [62], exposure to 2 or more suicide deaths affected less than 1% of the population, but was associated with 9.8-fold greater odds of suicide attempt, compared to an OR of 3.8 among those who had been exposed to the suicide of 1 relative. Similarly, those exposed to the suicide attempt of 2 parents were 5.67 times more likely to make a suicide attempt, compared to ORs of 2.89 and 3.89 (for paternal and maternal exposures, respectively) among youths who had been exposed to the suicide attempt of 1 parent [48].
Indeed, in the present multilevel meta-analysis, within-study heterogeneity remained largely unchanged by study-level moderators for both exposure to suicide and exposure to suicide attempt. For example, we did not find evidence of a significant difference in suiciderelated outcomes when the exposure to suicide or suicide attempt occurred in relatives compared to friends and acquaintances. Although previous registry-based studies have shown a 6-fold increase of suicide among biological relatives of adoptees who have died by suicide [84], in the present meta-analysis it was not possible to delineate between relatives who resided in the same household, and therefore shared many of the same environmental risk factors, and relatives who did not [9]. Understanding these factors is important for identifying specifically who within in the general population is most at risk. However, the pooling of observational studies meant that analyses of these factors were outside the scope of the present study. An important next step forward would therefore be examinations of exposure to suicide and suicide attempt while taking these risk factors into account using individual participant data meta-analyses.
In the present multilevel meta-analysis, between-study heterogeneity remained moderate (I 2 ð3Þ = 52.2%) across studies measuring exposure to suicide, which was not sufficiently explained by any of the included study design moderators. By contrast, study design characteristics accounted for 72.8% of between-study heterogeneity (I 2 ð3Þ = 87.8%) across studies measuring exposure to suicide attempt. In this instance, cross-sectional studies reported significantly larger ORs (OR = 8.23) compared to case-control (OR = 2.74) and cohort (OR = 2.69) studies. In general, cross-sectional studies are prone to an inherently greater number of biases, compared to case-control and cohort studies. This may be particularly pronounced in studies that measure suicide attempt because recall of suicide attempt may be less salient than recall of suicide death, and is prone to multiple interpretations and definitions [85].
It is noteworthy that we did not find evidence to support the role of age as a risk moderator, as suggested in previous reviews [9,10]. Yet these results should be interpreted with caution, as the dichotomization of study populations into the categories of youths and adults was based on a majority rule in 13 out of 34 studies [26][27][28]39,58,60,61,[63][64][65][68][69][70]. The finding that age was not a risk moderator may therefore be an artifact introduced by the imprecise age classification of the included population in individual studies. Furthermore, whilst similar patterns were observed across studies examining exposure to suicide attempt in youths versus adults, only 1 out of 13 studies [70] reported outcomes among adults, which may have impacted our ability to detect a statistically significant difference.
Finally, the results of the present study do not allow causality to be inferred, and although we show evidence of a temporal association between prior exposure to suicide and suicide attempt and subsequent suicide-related outcomes, cross-sectional studies, by virtue of study design, do not provide incidence estimates. To account for this limitation, we only included cross-sectional studies where participants were explicitly asked about suicidal acts that occurred after exposure to suicide or suicide attempt. But this approach does not mitigate errors in recall and other biases that are inherently more common in cross-sectional studies.

Conclusions
Our findings suggest that prior exposure to suicide is associated with increased risk of suicide and suicide attempt. By contrast, exposure to suicide attempt is associated with increased risk of suicide attempt, but not suicide death. Future studies should refrain from combining suicidal behaviors into composite measures of suicide exposures and outcomes as the relationships between exposure to suicide and suicide attempt and suicide-related outcomes are markedly different. Lastly, future studies should consider interventions that target suiciderelated outcomes in those exposed to suicide and include efforts to mitigate the adverse effects associated with exposure to suicide attempt.

PLOS MEDICINE
Association of suicidal behavior with exposure to suicide and suicide attempt: A multilevel meta-analysis