On the Genetic and Environmental Relationship Between Suicide Attempt and Death by Suicide

Abstract

Objective:

To examine the extent to which the genetic and environmental etiology of suicide attempt and suicide death is shared versus unique.

Methods:

The authors used Swedish national registry data for a large cohort of twins, full siblings, and half siblings (N=1,314,990) born 1960-1990 and followed through 2015. They conducted twin-family modeling of suicide attempt and suicide death to estimate heritability for each outcome along with genetic and environmental correlations between them. They further assessed the relationship between suicide attempt by young people versus adults.

Results:

In bivariate models, suicide attempt and death were moderately heritable among both women (A_attempt=0.52 [95% confidence intervals 0.44, 0.56]; A_death=0.45 [0.39, 0.59]) and men (A_attempt=0.41 [0.38, 0.49]; A_death=0.44 [0.43, 0.44]). The outcomes were substantially, but incompletely, genetically correlated (rA=0.67 [0.55, 0.67] for women and rA=0.74 [0.63, 0.87] for men). Environmental correlations were weaker (rE=0.36 [0.29, 0.45] for women and rE=0.21 [0.19, 0.27] for men). Heritability of suicide attempt was stronger among people aged 10-24 (A=0.55-0.62) than among adults aged 25 and older (A=0.36-0.38), and the genetic correlation between attempt during youth versus adulthood was stronger for women (rA=0.79 [0.72, 0.79]) than for men (rA=0.39 [0.26, 0.47]).

Conclusions:

The genetic and environmental etiologies of suicide attempt and death are partially overlapping, exhibit modest sex differences, and shift across the life course. These differences must be considered when developing prevention efforts and risk prediction algorithms. Where feasible, suicide attempt and death should be considered separately rather than collapsed, including in the context of gene identification efforts.

Introduction

Suicidal behavior, which includes non-fatal suicide attempts and death by suicide, is a significant public health concern. Suicide accounts for approximately 800,000 deaths per year worldwide (1), and non-fatal attempts are estimated to be up to 30× more common than suicide deaths (2). A prior attempt is one of the most prominent predictors of future death by suicide (3, 4) and 4-7% of individuals with a history of self-harm die by suicide in the ensuing 5-9 years (5, 6). While suicide death is more common among men, women attempt more often (7). Furthermore, while attempts and suicide death share many predictors (8), there is some evidence of differential associations between risk factors with attempt versus death (9). For example, suicide attempts are more common among women and youth, while deaths are more common among men and adults (10). In addition, social isolation and anxiety disorders are associated with suicide attempts more strongly than with death (10). Clarification of common versus distinct contributions to etiology may improve the ability to assess risk and thereby inform prevention and intervention efforts.

Prior studies have revealed familial clustering and a modest to moderate heritable component to suicidal behavior (11–16), including recent efforts to identify specific genetic variants associated with risk (17–19). While many such studies limited their analyses to one outcome (e.g., suicide death), others have collapsed manifestations of suicidality (e.g., ideation, attempt, death) into one variable using inconsistent approaches (20–23). Results from such studies may be driven by only one contributing measure (e.g., ideation), potentially obscuring important differences in the genetic underpinnings of different suicidal outcomes. Twin/family studies with a single outcome of interest have reported heritability estimates of 0.17-0.55 for suicide attempt (24–26) and 0.43 for suicide death (27), though as noted previously (28) limited sample sizes contribute to widely varying heritability rates. Some analyses have been further complicated by sample selection, for example where participants are ascertained on the basis of a prior personal/family history of suicidal behavior (21, 29) or specific psychiatric disorder (17, 23, 30), precluding generalization to the overall population. Notably, Erlangsen et al. (31) found that adjusting for psychiatric comorbidity reduced the SNP-based heritability (h²_SNP) of suicide attempt from 0.05 to 0.02. Based on findings from a recent study of suicide attempt within psychiatric disorders, h²_SNP also varies as a function of the disorder in question, e.g., from h²_SNP=0.03 within major depressive disorder to h²_SNP=0.10 within schizophrenia (17). Recent support for classifying suicidal behavior as a distinct mental disorder (32) suggests that viewing suicidal behavior primarily through the lens of psychiatric illness has important limitations.

Efforts to directly compare the etiology of suicide attempt and death are difficult where longitudinal data on both outcomes are not available for a large, representative sample, e.g., when samples are selected as described above. Furthermore, the relatively low prevalence of attempts and, in particular, suicide deaths has impeded efforts to obtain reliable heritability estimates for these outcomes due to inadequate statistical power. In the near term, these constraints mean that molecular genetic studies are unlikely to be able to elucidate shared versus outcome-specific aspects of etiology. Accordingly, it is necessary to employ other methodologic approaches to this important area of research.

Here, we leverage national Swedish registry data, including the recent addition of primary care data, to evaluate the genetic relationship between suicide attempts and death by suicide. While prior reports have employed substantial samples sizes to study familial aggregation of suicidal behavior (11–13), the current study is, to our knowledge, the largest to date that applies formal twin/family modeling, particularly with a goal of estimating genetic and environmental correlations between suicide attempt and death. We previously demonstrated that these outcomes should be treated independently: A liability threshold model, in which attempt and death lie on the same continuum of risk and differ only as a function of severity, does not fit the data well (33).

We expand upon those efforts here, by conducting an extended twin-family model of liability to suicide attempt and death. Our use of nationwide registry data, with a high degree of completeness due to universal healthcare, which yields a representative and statistically powerful sample. Given previous evidence of sex differences in heritability for psychiatric and substance use outcomes (34–36), we also benefit from this large sample by modeling sources of variance separately by sex. The longitudinal nature of the data further enables us to test whether genetic and environmental influences on suicide attempt are stable or dynamic from youth (here, age 10-24) into adulthood (age 25 and older), which may be relevant given shifts in the heritability of both internalizing and externalizing symptoms across the life course (37).

Materials and Methods

Sample

We collected information on individuals from Swedish population-based mortality, inpatient, and outpatient registers (38) with national coverage linking each person’s unique personal identification number which, to preserve confidentiality, Statistics Sweden replaced with a serial number. We secured ethical approval from the Regional Ethical Review Board of Lund University (No. 2008/409, 2012/795, and 2016/679). We double entered all same sex twin pairs with known zygosity and birth years between 1960 and 1990 from the Swedish Twin Registry, and all Swedish-born same sex full- and half-sibling pairs born between 1960 and 1990, and within 5 years of each other, from the Swedish Multi-Generation Register. This cohort was selected to optimize data coverage across time and sample size. An individual could be included several times if he/she had several siblings or different types of siblings; note that this could result in narrowed confidence intervals, but did not substantively impact variance component estimates. Zygosity was assigned using standard self-report items, which, when validated against DNA markers, were 95% to 99% accurate.

Using the Swedish national census and population registers, we assessed cohabitation status for full- and half-sibling pairs as the proportion of possible years they lived in the same household until the oldest turned 16. Among monozygotic (MZ) and dizygotic (DZ) twins and full siblings, we only included pairs reared together for ≥80% of their possible years. Half siblings residing together for ≥80% of the possible years were classified as having been reared together. Those residing together for ≤20% of the possible years were classified as having been reared apart. See below for implications within the twin/family model.

Phenotypes

The following ICD codes were used to identify suicide attempts and deaths of both known and undetermined intent: ICD-8 codes E950-E959, E980-987, ICD-9 codes E950-E959, E980-987, ICD-10 codes X60-X84, Y10-Y34. Details are available through the World Health Organization (39). Distinctions between suicide death and non-fatal attempts, where necessary, were determined using the Cause of Death Register. Prior studies have provided support for the inclusion of events of undetermined intent (40, 41), with one reporting that excluding these cases would result in approximately 25% of true suicide attempts being misclassified (41). Suicide attempt and suicide death were treated as binary, and the registration for either outcome (i.e., the date that the ICD code was entered into the medical registry) could occur at any time during the follow-up period, which began when individuals turned 10 and ended in December 2015. Note that an individual could be registered for both suicide attempt and suicide death.

Statistical Analysis

We used classical bivariate twin/sibling modeling, which assumes three sources of liability to non-fatal suicide attempts and death by suicide: additive genetic (A), shared environment (C), and unique environment (E). The model assumes that MZ twins share 100% of their genes; DZ twins and full siblings share, on average, 50% of their genes; while half siblings share, on average, 25% of their genes. The model also assumes that the shared environment, which reflects family and community experiences, is equal between MZ twins, DZ twins, and full siblings, while for half siblings C equaled 1 for pairs reared together and 0 for pairs reared apart. Finally, the unique environment reflects experiences not shared by twins/siblings, random developmental effects, and random measurement error.

The model is based on the idea of an unobserved distribution of liability underlying each of the two binary outcomes, non-fatal suicide attempt and death by suicide. The correlation within each twin/sibling pair corresponds to the proportion of variance explained by the genes (A) and environment (C) they share. The bivariate model was built using the Cholesky decomposition where the first factor loads on both suicide attempts and death by suicide while the second loads only on the latter. Based on results from the univariate ACE models of suicide attempts and death by suicide, we restricted the bivariate model by setting the two C-paths to death by suicide to 0. All analyses were separated by gender. The OpenMx software (42, 43) was used to fit the models.

We conducted a sensitivity analysis in which events of undetermined intent were excluded to account for the possibility that these impacted estimates in the primary analysis. ICD-8 and ICD-9 codes corresponding to events of undetermined intent are E980-E987; for ICD-10, the codes are Y10-Y34. A total of N=18,436 suicide attempts and N=1063 suicide deaths were excluded in these analyses, corresponding to 42.4% and 25.6% of all attempts and deaths, respectively.

In a secondary analysis, we estimated the genetic and environmental correlations between suicide attempt during youth and adulthood, using the World Health Organization’s age range for “young people” – ages 10-24 (44). We selected this age range to capture neurodevelopment of the prefrontal cortex, which persists into the early 20’s (45, 46) and may be particularly relevant to impulsive behavior (47, 48), including suicidality (49). We imposed a restriction on the data to avoid misclassifying persistent suicidal behavior toward the end of the earlier developmental period as “adult”: Those with a suicide attempt at age 24 were censored for one year, such that additional attempts occurring during that year (into age 25) were not counted. Beyond that year-long censoring period, subsequent attempt registrations were classified as adult attempts. We identified N=916 individuals with an attempt at age 24 and within the subsequent year; of these, N=443 had an additional attempt at a later time and were coded as attempting as a young person and an adult, while the remaining N=473 were classified as attempting only as a young person. We lacked sufficient power to conduct a corresponding analysis of suicide death.

Results

Descriptive Statistics

Table 1 provides details on the number of pairs for each relative group (monozygotic or dizygotic twins, full siblings, and half siblings) that contributed to the analyses and corresponding descriptive statistics. The total sample size was N=1,314,990. There were N=21,664 women and N=21,854 men who attempted suicide; N=1048 women and N=3109 men died by suicide. Suicide attempts were more common among women (p<0.0001), while death by suicide was more common among men (p<0.0001). Individuals with half-siblings exhibited higher rates of both suicide attempts and deaths compared to twins or full-siblings.

Table 1.

Descriptive statistics for sibling pairs born in Sweden 1960-1990. Individuals may belong to more than one sibling pair.

Women	Monozygotic twins			Dizygotic twins			Full siblings			Half siblings reared together			Half siblings reared apart
N Pairs	3,186			2,570			275,668			9,828			19,166
% Suicide attempt	3.2%			3.5%			3.7%			6.5%			7.1%
% Suicide death	0.2%			0.2%			0.2%			0.4%			0.4%
Median (Q1, Q3) age at suicidal event	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3
Suicide attempt	24	19	32	26	20	35	23	18	31	22	17	30	22	17	32
Suicide death	39	35	42	26.5	20	33.5	30	23	39	27	23	34.5	28	23	38
Tetrachoric correlations (95% CI)
SA between siblings	0.57	0.49; 0.65		0.25	0.13; 0.27		0.28	0.27; 0.29		0.17	0.12; 0.21		0.12	0.09; 0.15
SD between siblings	n/a	n/a		n/a	n/a		0.25	0.18; 0.32		n/a	n/a		0.17	−0.03; 0.37
Within sibling SA and SD	0.50	0.28; 0.73		0.47	0.21; 0.73		0.52	0.50; 0.54		0.45	0.36; 0.54		0.43	0.37; 0.50
Between sibling SA and SD	0.33	0.04; 0.61		0.21	−0.15; 0.57		0.16	0.12; 0.19		0.15	0.02; 0.27		0.01	−0.09; 0.11
Men	Monozygotic twins			Dizygotic twins			Full siblings			Half siblings reared together			Half siblings reared apart
N Pairs	2,646			2,294			310,569			10,138			21,430
% Suicide attempt	3.2%			3.0%			3.4%			6.5%			6.3%
% Suicide death	0.3%			0.4%			0.5%			0.9%			0.9%
Median (Q1, Q3) age at suicidal event	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3	Median	Q1	Q3
Suicide attempt	28	22	35	28.5	23	35	26	20	34	26	20	33	26	20	35
Suicide death	31	26	36	29.5	26	39	29	23	37	29	24	38	29	23	37
Tetrachoric correlations (95% CI)
SA between siblings	0.41	0.30; 0.52		0.29	0.16; 0.43		0.30	0.29; 0.31		0.19	0.15; 0.24		0.12	0.09; 0.16
SD between siblings	0.83	0.69; 0.96		n/a	n/a		0.20	0.17; 0.24		0.14	−0.02; 0.30		0.12	0.01; 0.23
Within sibling SA and SD	0.45	0.25; 0.66		0.60	0.43; 0.76		0.44	0.42; 0.45		0.36	0.29; 0.43		0.38	0.33; 0.43
Between sibling SA and SD	0.23	−0.04; 0.50		0.23	−0.04; 0.51		0.15	0.13; 0.18		0.13	0.04; 0.22		0.09	0.02; 0.15

SA=suicide attempt; SD=suicide death; CI=confidence intervals; n/a=cannot be estimated due to sample size).

Among individuals with a prior suicide attempt, 2.1% (N=457) of women and 3.9% (N=846) of men died by suicide during the observation period. Among those who died by suicide, 56.0% (N=581) of women and 72.6% (N=2,250) of men had no known prior attempt. The median (Q1, Q3) time interval between first attempt and suicide death was 4 years (1, 10) for women and 3 (1, 8) for men.

Tetrachoric correlations (used for binary data such as diagnostic status) within and across outcomes, both within individuals and across members of the sibling pairs, are reported in Table 1. In general, correlations were higher for full siblings (including monozygotic and dizygotic twin pairs) than for half siblings, although estimates were imprecise for some correlations. The low prevalence of suicide death prevented us from estimating sibling correlations for some groups. The correlation patterns suggest that genetic factors contribute to resemblance across siblings.

Univariate Twin and Family Models

We first fit standard ACE models for suicide attempt and suicide death separately, stratified by sex. Variance component estimates are presented in Table 2. Heritability estimates (A) were moderate for both outcomes. We observed low but significant contributions from shared environmental factors for suicide attempt but not death, though the latter could be a false negative finding in light of a relatively limited number of suicide deaths. We therefore elected to specify an ACE structure for attempt and an AE structure for suicide death in subsequent models.

Table 2.

Variance components (95% confidence intervals) from univariate and bivariate twin-family modeling results.

		Univariate Analyses				Bivariate Analyses
Suicide Attempt	Women Estimate	95% CI	Men Estimate	95% CI	Women Estimate	95% CI	Men Estimate	95% CI
A	0.50	0.42; 0.54	0.41	0.32; 0.49	0.52	0.44; 0.56	0.41	0.38; 0.49
C	0.03	0.01; 0.03	0.09	0.06; 0.12	0.02	0.00; 0.06	0.09	0.09; 0.12
E	0.47	0.43; 0.51	0.50	0.45; 0.55	0.46	0.44; 0.46	0.50	0.50; 0.50
Suicide Death
A	0.47	0.36; 0.61	0.45	0.32; 0.53	0.45	0.39; 0.59	0.44	0.43; 0.44
C	0.01	0.00; 0.27	0.00	0.00; 0.06	-	-	-	-
E	0.52	0.39; 0.80	0.55	0.47; 0.63	0.55	0.42; 0.67	0.56	0.50; 0.57

A=additive genetic variance component (heritability); C=common/shared environment variance component; E=unshared environment variance component; HR=hazard ratio; CI=confidence intervals

Bivariate Twin and Family Models

Final parameter estimates from the bivariate model of suicide attempt and death by suicide are presented in the Figure, and variance components are presented in Table 2. We tested whether thresholds for both outcomes for each sibling group, plus genetic and shared environment path estimates could be equated across men and women, and they could not (ΔX²=4973.8, Δdf=15, p<0.0001). The genetic correlation between suicide attempt and death was rA=0.67 (0.55; 0.67) for women and rA=0.74 (0.63; 0.87) for men. The environmental correlation, attributable to environmental experiences not shared by siblings, was rE=0.36 (0.29; 0.45) for women and rE=0.21 (0.19; 0.27) for men.

Figure.

Parameter and correlation estimates (95% confidence intervals) from the bivariate twin-family model of non-fatal suicide attempt (SA) and suicide death (SD), for women (top panel) and men (bottom panel). The sources of variance are: additive genetic factors (A), common environmental factors (C; attempt only), and unique environmental factors (E).

Sensitivity Analysis

To determine whether parameter estimates were heavily influenced by attempts and deaths of undetermined intent, we conducted a sensitivity analysis excluding those registrations (N=18,436 suicide attempts and N=1063 suicide deaths). Among women, 15.0% of attempts and 12.5% of deaths were excluded; for men, 30.8% of attempts and 12.6% of deaths were excluded. Variance component estimates changed little for women and confidence intervals overlapped those in the primary analysis (Table 3). Similarly, rA and rE were comparable (rA=0.71 [0.71; 0.87]; rE=0.36 [0.34; 0.46]). For men, exclusion of these events resulted in a lower heritability estimate for suicide attempt, and confidence intervals did not overlap those from the primary analysis. Estimates for suicide death were less affected by the exclusion of these events. The genetic and environmental correlations across attempts and suicide death were modestly increased but with overlapping confidence intervals (rA=0.86 [0.86; 1.00]; rE=0.25 [0.2; −0.26]). Parameter estimates could not be equated across sex (ΔX²=9667.8, Δdf=15, p<0.0001).

Table 3.

Variance components (95% confidence intervals) from bivariate twin-family modeling results where events of undetermined intent were excluded from the analysis.

Suicide Attempt	Women Estimate	95% CI	Men Estimate	95% CI
A	0.55	0.55; 0.56	0.35	0.30; 0.35
C	0.00	0.00; 0.00	0.06	0.06; 0.06
E	0.44	0.42; 0.47	0.59	0.59; 0.62
Suicide Death
A	0.46	0.32; 0.61	0.44	0.44; 0.52
C	-	-	-	-
E	0.54	0.40; 0.68	0.56	0.48; 0.61

A=additive genetic variance component (heritability); C=common/shared environment variance component; E=unshared environment variance component; HR=hazard ratio; CI=confidence intervals

Secondary Analysis of Suicide Attempt as Young Person and/or Adult

We identified N=20,893 individuals with an attempt between ages 10-24, and N=22,625 individuals with an attempt after age 24. There were N=2752 individuals with attempts during both periods, of which 58% were women and 42% were men. Results of bivariate twin-family models are presented in Table 4. Heritability estimates were higher for attempt as a young person than as an adult for both sexes. The genetic correlation (rA) between non-fatal attempt as a young person and as an adult was higher among women than among men, and the Olkin-Pratt test for homogeneity of rA across sexes was significant (Q=28.2, df=1, p<0.0001). The shared environmental correlation (rC) between young and adult attempt was in opposite directions across sexes; however, C accounted for very little of the total variance, making substantive interpretation of these disparate estimates difficult.

Table 4.

Variance components (95% confidence intervals) from bivariate twin-family modeling results of suicide attempt among young people aged 10-24 and adults (age 25+).

Young People	Women Estimate	95% CI	Men Estimate	95% CI
A	0.62	0.59; 0.65	0.55	0.44; 0.65
C	0.01	0.00; 0.02	0.07	0.05; 0.07
E	0.37	0.35; 0.41	0.38	0.37; 0.42
Adults
A	0.36	0.30; 0.42	0.38	0.37; 0.44
C	0.09	0.09; 0.09	0.10	0.10; 0.10
E	0.55	0.52; 0.55	0.52	0.49; 0.55
Correlations between Young People and Adults	Correlation	95% CI	Correlation	95% CI
rA	0.79	0.72; 0.79	0.39	0.26; 0.47
rC	−0.88	−1.00; 0.00	0.99	0.95; 1.00
rE	0.17	0.17; 0.17	0.22	0.16; 0.26

A=additive genetic variance component (heritability); C=common/shared environment variance component; E=unshared environment variance component; rA=genetic correlation; rC=common/shared environmental correlation; E=unshared environmental correlation; CI=confidence intervals

Discussion

Our primary research aim was to quantify the genetic relationship between suicide attempts and death by suicide using a representative national Swedish cohort. We found that attempts and suicide death were moderately heritable, with estimates from our twin-family approach (A=0.41-0.52) considerably higher than those from molecular genetic studies (h²_SNP=0.03 for suicide attempt (17); h²_SNP=0.16 for suicide death (19)), as is common across these different methodological approaches due in part to limited sample sizes and the inclusion of only common variants in the latter (50, 51). These outcomes are substantially genetically correlated (rA=0.67 for women and rA=0.74 for men), while the unique environmental correlation was more modest (rE=0.36 for women and rE=0.21 for men). Our findings support prior evidence that the etiologies of suicide attempt and suicide death are incompletely overlapping and may thus present distinct opportunities for prevention. In a secondary analysis, we found that the genetic factors contributing to risk of suicide attempt are potentially dynamic across the life course, underscoring the complex roles of biology and development in suicidality.

In conjunction with our prior study (33), which found that the genetic distinction between suicide attempt and death was not merely one of severity of liability, our results have implications for studies aimed at the identification of genetic risk variants for suicidal behavior. Genomewide association studies that collapse attempt and death are unable to distinguish whether implicated variants or genes have outcome-specific effects versus contribute to liability to both. This prevents dissection of differences in the biological etiologies of attempt and death, which may hinder prevention efforts. These complications could be extended to apply to suicidal ideation and non-suicidal self-injury: Prior evidence suggests these outcomes are at least partially etiologically distinct from suicide attempts and/or death (52–56), although a recent study revealed strong genetic and unique environmental correlations (rA=0.94 and rE=0.80) (57) in a sample of British twins. These outcomes have also been conceptualized as lying on a continuum of self-harm (20). Recent analytic advances that decompose genetic variance into that which is common across outcomes versus that which is outcome-specific (58–60) may avoid these shortfalls once genotyped samples of sufficient size are available.

In addition, our findings underscore the relevance of evaluating family history of suicidal behaviors. Stigma about suicidal behavior persists even among family members and medical personnel (61, 62), which may lead individuals to feel uncomfortable disclosing their own or their family members’ history. This is turn may prevent health care providers from fully appreciating patients’ risk, particularly if suicide attempts, which are more readily concealed from family members, are not disclosed. Clinicians should also be aware of both continuity of risk across the life course and the relevance of life stage-specific influences: i.e., the impact of environmental exposures experienced during adolescence may persist and be augmented by exposures experienced during adulthood.

We previously reported, in an extended adoption design in Sweden, evidence for substantial environmental transmission of suicide attempt liability from parents to offspring (33), while the current study found only a small contribution from the shared environment. These seemingly disparate findings are likely due in part to the different questions posed by these studies – parent-offspring “vertical” transmission of risk versus “horizontal” transmission between siblings close in age. Different, but potentially overlapping, aspects of the environment are captured across these methods. For example, the “shared environment” of the current study includes school/neighborhood-level exposures outside the home, while the “environment” of our previous study includes parenting behaviors that may differ across offspring (e.g., discipline, warmth). We have elsewhere shown, using empirical estimates of twin-twin and parent-child correlations for major dimensions of parenting, that the correlation in liability among twins for typical psychiatric phenotypes due to shared parenting would range from 2-4% (63), consistent with the findings of the current study.

The modest environmental correlations between suicide attempt and death, which reflect exposures not shared by family members, indicate that extrinsic factors differentially impact risk for these outcomes. One study found that most environmental factors examined, including low income and exposure to stressful legal, interpersonal, and work-related events, increased risk for both attempts and death (10). However, other factors had outcome-specific effects: A poor parental relationship during childhood was associated with suicide but not with serious attempts, while social isolation was associated with attempt and not death. Another study found that gun availability – i.e., access to a high lethality method – was related to suicide death but not attempt (9); gun ownership may also be a proxy for other risk factors such as adult antisocial behavior and exposure to violent crime (64). The dearth of studies that examine risk factors for both suicide attempt and death within the same sample – thereby enabling direct comparisons – precludes a nuanced interpretation of the modest environmental correlations we observed. This is a critical deficit in our understanding of the etiology of these outcomes, as it impacts risk prediction and prevention. Environmental exposures, in contrast to genetic variants, are potentially modifiable risk factors, but our findings reveal that environmental factors impacting risk for both suicide attempt and death represent only an incomplete portrait of prevention targets.

Our secondary analysis examining suicide attempt across young people and adults addresses research questions that have not, to our knowledge, been tested previously: To what extent do genetic factors differentially contribute to risk of suicide attempt across the life course, and is there incomplete overlap of genetic factors across time? We observed that the heritability of suicide attempt was considerably higher among young people, particularly for women (confidence intervals overlapped for men across time). This raises the possibility that temperament and/or psychopathology – that is, intrinsic factors of considerable heritability – may play a more prominent role among young people than adults. In contrast, the variance in liability accounted for by unique environmental factors during adulthood suggests that adverse life experiences are more potent risk factors for suicide attempt among adults. The shared environmental correlation (rC) estimates differed markedly across sexes. These factors accounted for little of the total variance in risk (C = 0.01-0.1) and the rC estimate was imprecise among women; substantive interpretation of the rC estimates is therefore not feasible.

Another key finding from the secondary analysis is the considerable sex difference in the genetic correlation between attempts among young people versus adults (rA_women=0.79 [0.72, 0.79] and rA_men=0.39 [0.26, 0.47]). This indicates that, among women, continuity of risk across the life course is due largely to genetic factors whose impact persists across time. In contrast, for men, liability toward suicide attempt in adulthood is more strongly influenced by a qualitatively different facet of genetic influences than were relevant earlier in life. Similar to our findings on the genetic correlation between suicide attempt and death, the results of our secondary analysis have implications for gene-identification studies. The higher heritability of attempts among young people may facilitate genome-wide association studies (65) relative to studies among adults; however, restricting a GWAS study to only young people is not an ideal solution given the advantages of maximizing sample size. Instead, age at attempt could be included as a covariate along with an interaction term between age and genetic variant.

Limitations

There are several limitations to these analyses. First, we are reliant on ICD codes for suicide attempts and are therefore unable to detect attempts that did not require medical attention or were not otherwise reported to a health care provider, of which there may be a considerable number (7). Accordingly, our data likely captures more severe suicide attempts, which may be more closely related to suicide deaths than attempts overall. We are also reliant on the quality and consistency of the coding process (66, 67). We were able to identify only one study that obtained self-reports of self-harm in a sample for which institutional records (medical registrations) were also available (68), which found incomplete overlap between sources; however, the small sample precludes generalization to other studies. Prior evidence indicates that individuals hospitalized for self-harm may be less likely to complete a self-report survey (68), and the rate of refusal to participate is higher for non-anonymous surveys (69). These factors would likely lead to underestimation of suicide attempt prevalence as assessed via self-report. Such estimates vary widely, e.g. from 2.7-14% (69–71). The limitations of our use of registry records may therefore be offset by the reliability of these records.

Although these findings are representative of the Swedish population, they may not be generalizable to populations in other countries. Until relatively recently, Sweden’s annual suicide rate was consistently higher than that of the US, and became comparable at approximately 12 per 100,000 in 2007-2009. Since 2010, the US rate has increased while Sweden’s rate has remained stable (72). Our secondary analyses expose potentially important shifts in the etiology of suicide attempt across the life course, including those related to environmental exposures, which may differ markedly across countries and cultures. We also note that our selection of cohort, which maximized available registry data, did not include elderly individuals.

Consistent with other studies (41, 73), we included events of undetermined intent in our primary analyses. Our sensitivity analysis excluding those events resulted in shifts in estimates, particularly for suicide attempt among men. While some prior evidence suggests that events of undetermined intent include suicides that would otherwise be undetected (74, 75), the conservative approach is to regard the variance component and correlation estimates from our primary and sensitivity analyses as the bounds within which the “true” estimates fall.

Our models do not account for psychiatric disorders, which are genetically related to suicidal behavior (17). This limitation is offset by the insight provided by the current approach to the genetic and environmental influences on suicidal behavior in the general population. Indeed, the majority of those with registrations for suicide attempt (70.7%) or death (51.3%) did not have a prior registration for a major psychiatric disorder.

Although we demonstrated shifts in suicide attempt heritability, and a genetic correlation <1, across development, our analyses did not otherwise account for the potential effect of age. Given prior evidence of age, period, and cohort effects on suicidal behavior (76–78), additional research is warranted.

Conclusions

Suicide attempt and death are moderately heritable and are substantially–but not completely – genetically correlated. In conjunction with modest environmental correlations, our findings speak to partially distinct etiologies, raising the likelihood of incompletely overlapping opportunities for prevention and risk prediction. Furthermore, as demonstrated by our analysis of suicide attempt across young people and adults, both genetic and environmental influences on risk are temporally dynamic, particularly among men. Efforts to reduce risk of suicidal behavior must therefore consider sex differences and shifts across the life course, and gene identification efforts would benefit from distinguishing between suicide attempts versus death.