Skip to main content


Swipe om te navigeren naar een ander artikel

18-04-2017 | Empirical Research | Uitgave 4/2018

Journal of Youth and Adolescence 4/2018

Epidemiology of Suicide Attempts among Youth Transitioning to Adulthood

Journal of Youth and Adolescence > Uitgave 4/2018
Martie P. Thompson, Kevin Swartout


Both fatal and nonfatal suicide attempts among youth are significant public health concerns. In 2014, the most recent year for which there are official mortality data, suicide was the tenth most common cause of death, resulting in 42,773 lives lost. Although people of all ages are at risk for suicide, it is more common among young people. Suicide was the second leading cause of death for 12–25 year olds and the third leading cause of death for 26–34 year olds (Centers for Disease Control and Prevention [CDC] 2017). A nonfatal suicide attempt is the strongest predictor of a suicide completion (Bostwick et al. 2016). Research shows that attempters are approximately 40 times more likely to commit suicide than those who have never attempted suicide (Harris and Barraclough 1997). Nonfatal suicide attempts are 25–60 times more prevalent than fatal ones. Approximately 91,209 12–17 year olds and 103,524 18–25 year olds were treated in emergency rooms following suicide attempts in 2014 (CDC 2017). In 2013, 8% of high school students reported that they had attempted suicide in the past year (Kann et al. 2014). The morbidity and mortality associated with suicidal behavior among young people, as well as its impact on family members and friends, makes it imperative that effective interventions be developed and targeted to appropriate high-risk groups. Increased understanding of risk factors for suicide attempts can ultimately help reduce the public health burden of death by suicide.

Emerging Adulthood and a Developmental Perspective

Investigations into the epidemiology of suicidal behavior among adolescents require “developmentally dynamic models that characterize emotional and behavioral maladjustment in terms of both severity and change over time” (Kerr et al. 2013, p. 51). The transition from adolescence to young adulthood represents a key developmental time period for youth, and has been referred to as “emerging adulthood” (Arnett 2000). This developmental period between the approximate ages of 18 and 25 represents a time of identity exploration when young people experience frequent changes in life goals, romantic attachments, work aspirations, and worldviews. Emerging adulthood is a time where life circumstances and choices can set youth on certain trajectories, either positive or negative. An epidemiological study of suicidal behavior during emerging adulthood should take into account that known risk factors for suicide attempts may themselves not be static, and as they change during this developmental time period, so too might suicide risk.
Longitudinal data are needed to investigate how suicide risk changes as youth transition from adolescence into young adulthood. Some studies have examined the longitudinal trajectory of suicide risk during emerging adulthood, but these studies typically have assessed change for the sample as a whole and have not investigated differential change among subgroups (Marshal et al. 2013; Needham 2012). An exception to this was a study of 552 adolescents whose suicide ideation was assessed at ages 14, 15, and 17. Results revealed three unique trajectories of ideation; approximately 74% reported no ideation at any time point, 11% showed a decreasing likelihood of ideation, and 15% showed an increasing likelihood of ideation (Rueter et al. 2008). A study by Thompson and colleagues (Thompson et al. 2009) using three assessments spanning 7 years found that youth could be classified into three latent classes representing degree of suicide risk. Youth in the low-risk group in late adolescence remained at low risk 7 years later. Although some youth who were classified as high-risk transitioned to a lower risk group 7 years later, a significant proportion remained at high risk. Also, although there are several studies on prospective predictors of suicidal behavior in community-based samples of adolescents (Kerr et al. 2013; Miranda et al. 2014; Nkansah-Amakra 2013), these studies have not assessed if and how suicide likelihood changes as risk factors change. A handful of studies have examined how changes in suicide risk factors are associated with changes in suicidal behavior over time. Studies that have taken this type of developmental approach have found support for the importance of looking at not only the presence or severity of a particular risk factor, but also its level of change (Kerr et al. 2013).

Risk Factors

Researchers have identified several significant risk factors for suicide, and most of these are promulgated in different theories of why people engage in suicidal behavior (see Barzilay and Apter 2014 for a review). These theories identify risk factors reflecting psychological, behavioral, cognitive, interpersonal, and personal history variables. Because our study focused on how time-varying risk factors corresponded with suicide attempt trajectories, we included only risk factors that were assessed using similar measurements over time. These included psychological, behavioral, and personal history variables. One key psychological risk factor for suicide attempts is depressive symptoms (Brent et al. 1999; Goldsmith et al. 2002). One study examined contemporaneous associations between suicide attempts and various psychiatric disorders in a sample of psychiatrically hospitalized youth. Findings based on repeated assessments of youth as they transitioned from adolescence to young adulthood revealed that major depressive disorder was associated with a five-fold increase in contemporaneous risk for suicide attempt, even after controlling for other comorbid disorders (Goldston et al. 2009). In another study with a community sample of adolescents, suicide attempt history assessed through age 25 was significantly associated with higher levels of depressive symptoms in fifth grade and significantly associated with relative increases in externalizing behaviors and depressive symptoms from grades 5 to 10 (Kerr et al. 2013).
Three important behavioral risk factors for suicide attempts are delinquency (Thompson et al. 2007), impulsivity (Kasen et al. 2011; Simon et al. 2001), and alcohol problems (Norstrom and Rossow 2016). In a study using longitudinal data from a community sample of 770 youth, those who had attempted suicide reported impulsivity levels that were six times higher than their counterparts who had not attempted suicide. Further, although both attempters and non-attempters showed expected decreases in impulsivity as they aged out of adolescence, only non-attempters showed significant annual declines that plateaued with age (Kasen et al. 2011). A psychological autopsy study based on 645 people aged 11–87 years who died by suicide revealed that higher levels of impulsivity were associated with death by suicide at a younger age, suggesting that impulsivity may be a particularly relevant risk factor among adolescents and young adults (McGirr et al. 2008).
In terms of the link between delinquency and suicidal behavior, a prospective study using nationally-representative data revealed that males who died by suicide were more likely to have engaged in prior delinquent behavior compared to their living male counterparts (Feigelman et al. 2016). The role of delinquency as a risk factor for suicidal behavior among adolescents was also found using nationally representative data from the Youth Risk Behavior Surveillance System (Thompson et al. 2006). Delinquent youth were approximately 5 times more likely to have seriously considered suicide, 5 times more likely to have made a suicide plan, 10 times more likely to have attempted suicide, and 15 times as likely to have required medical treatment after attempting suicide compared to their nondelinquent counterparts.
Alcohol use has been found to be a significant risk factor for suicide ideation (Lamis et al. 2016), attempts (Windle 2004), and completions (Kolves et al. 2006). In a systematic review of research, there was consistent support for both individual- and population-level associations between alcohol use and suicidal behavior. Increased risk of suicide was found among alcohol abusers and increased rates of alcohol use in a population were associated with increased rates of suicide (Norstrom and Rossow 2016). Another study found that in a sample of people who had completed suicide, 50% had had alcohol dependence (Kolves et al. 2006).
Two important personal history factors for suicide attempts are having a family member or friend attempt or complete suicide and experiencing partner violence. Researchers have found that having a friend who committed suicide increased the risk of suicide attempts (Bearman and Moody 1994). In another study investigating the short- and long-term consequences of loss of a friend to suicide, Feigelman and Gorman (2008) found that a friend’s suicide was associated with increased risk for both suicidal ideation and attempts during the first year after loss. However, the loss of a friend to suicide did not affect the likelihood of ideation or attempts 6 years later. Researchers also have found that a family history of suicidality is an important predictor of youth suicide attempts. In a prospective study of children whose parents were being treated for mood disorders, youth who had a parent attempt suicide were approximately five times more likely to make a suicide attempt compared to their counterparts, even after accounting for their own past and current mood disorders and suicide attempt history (Brent et al. 2015). In a case-control study comparing individuals who had committed suicide to population-based controls, a family history of completed suicide increased the risk for suicidality, even after controlling for psychiatric variables (Qin et al. 2002).
In terms of risk for suicidality conferred by partner violence, a longitudinal study assessing the association between physical dating violence and suicidal ideation among adolescents found that youth who had experienced physical dating victimization were twice as likely to report suicidal ideation compared to their peers who had not experienced dating violence (Nahapetyan et al. 2014). In a national sample of adolescents, youth who experienced dating violence were almost three times more likely than their counterparts to have attempted suicide (Swahn et al. 2008). A link between partner violence and suicide attempts also has been found in adult samples, as reflected in a review that found that 36 of the 37 papers investigating the link between partner abuse victimization and suicidality found significant associations (McLaughlin et al. 2012).

Overview and Hypotheses

Although a substantial empirical literature base has identified important risk factors of suicidal behavior, it is not well understood if changes in risk factors correspond with changes in suicide risk. To address this knowledge gap, we addressed the following research questions using a nationally representative sample: (a) What are the different trajectories of suicidal behavior across an approximately 13-year period as youth transition into young adulthood; and (b) do variations on established risk factors over this 13 year period predict different suicide behavior trajectories? We hypothesized that there would be four classes of suicidal behaviors—persistently low, persistently high, decreasing, and increasing. We also hypothesized that levels on certain risk factors would correspond with suicide trajectories (i.e., low levels on risk factors in both adolescence and young adulthood would predict being in a persistently low suicide risk group; high levels on risk factors in both adolescence and young adulthood would predict being in a persistently high suicide risk group; high levels on risk factors only in young adulthood would predict being in an increasing trajectory group; and high levels on risk factors only during adolescence would predict being in a decreasing trajectory group). Because we were interested in assessing how changes on risk factors corresponded with changes in suicide risk, we only included risk factor variables that were assessed at early (Wave 1 or Wave 2) and later (Wave 4) waves of data collection. These variables included depression, impulsivity, delinquency, alcohol problems, family and friend suicide history, and partner abuse.


Sampling Procedures and Sample

This study used restricted data from the National Longitudinal Study on Adolescent Health (Add Health). Add Health utilized a multistage stratified cluster design to sample public and private high schools in the United States (Harris et al. 2009). Of the 26,666 eligible high schools, a stratified random sample of 80 high schools was selected. Schools were stratified by region, urbanicity, school type, and percentage white. For each high school selected, the largest feeder school was also recruited to participate. Seventy-nine percent of the recruited schools agreed to participate, resulting in a sample size of 132 schools. All students listed on the school rosters were eligible to be selected for the in-home survey. Informed consent was obtained from all individual participants included in the study. Wave 1 in home surveys were completed in 1995 with 20,745 7th–12th graders. Of respondents eligible for follow-up, 14,738 completed a Wave 2 in-home survey 1 year later (89%), 15,197 completed a Wave 3 in-home survey approximately 7 years later (77%), and 15,701 completed a Wave 4 in-home survey approximately 13 years later (80%). Respondents who were 12–18 years old at Wave 1, participated in all four waves of data collection, and had sample weights were used in our analyses (n = 9027). Local IRB approval was obtained for the analysis of these restricted data.
At Wave 1, participants’ mean age was 15.26 years (SD = 1.76), 50% were male, and 17% were Hispanic. Race categories were: 58% White, 21% African American, 7% Asian, 1% American Indian/Alaskan Native, 5% more than one race, and 8% unknown.


Suicidal behavior

Suicide attempts were assessed by asking respondents “During the past 12 months, how many times did you actually attempt suicide?” We computed dichotomous variables that measured if a respondent had attempted suicide at Wave 1, Wave 2, Wave 3, and Wave 4. Measures across all four waves were used to identify suicide risk trajectories.

Depressive symptoms

Seven items derived from the Center for Epidemiologic Studies on Depression scale (Radloff 1977) were used to assess depressive symptoms. These seven items were available at both Waves 1 and 4, and showed good internal consistency at both Waves (α = .81 at Wave 1 and α = .84 at Wave 4). A sample item was “How often was the following true during the past week? You felt depressed?” Items were answered on a 4-point scale (0–3), with higher scores indicative of higher levels of depressive symptoms.


At Waves 2 and 4, respondents were asked if when making decisions, they usually went with their “gut feeling” without thinking too much about the consequences of each alternative. Responses were answered on a 5-point scale (1–5), with higher scores indicative of higher levels of impulsivity.

Delinquent behavior

Seven items from both Waves 1 and 4 assessed for risky/delinquent behavior. These items asked respondents if they had done the following in the past 12 months: damaged property, hurt someone, stole something worth more than $50, gone into a house or building to steal, threatened to use a weapon, sold drugs, and stole something worth less than $50. Items were coded as no or yes and then summed. Due to positive skewed scores, we then categorized the summed scores as follows: 0 = 0, 1 = 1, > 1 = 2.

Alcohol problems

Five items were used to assess for alcohol problems. Questions used a reporting time frame of the past 12 months and included: (1) How often drinking interfered with their responsibilities at work or school; (2) how often they had problems with family or friends because of their drinking; (3) how many times they had problems with someone they were dating because of drinking; (4) how many times they were hung over; and (5) how many days they had gotten drunk. Responses were coded such that a 0 = never, 1 = once, and 2 = two or more times. All of these questions were asked at Wave 1. Questions assessing problems with partner and being hungover were assessed at Wave 3 but not at Wave 4. Thus, the scale score for Wave 4 was based on responses from Waves 3 and 4. Responses were z-scored prior to averaging. The scales showed adequate internal consistency at both Waves (α = .75 at Wave 1 and α = .69 at Wave 4).

Family and friend history of suicidal behavior

Respondents were asked at Wave 1 if a friend had attempted suicide, if a friend had completed suicide, if a family member had attempted suicide, and if a family member had completed suicide in the past 12 months. At Wave 4, respondents were asked if a family member or friend had attempted suicide and if a family member or friend had completed suicide. In order to make the measures consistent across Waves 1 and 4, variables were constructed that reflected if a respondent had a family member or friend who had attempted or completed suicide in the last year (0 = no, 1 = yes).

Partner abuse

Respondents were asked at Waves 2 and 4 if a partner had threatened them with violence, pushed or shoved them, or thrown something at them that could hurt them. If a respondent experienced any of these forms of violence, they were considered as having experienced partner abuse (0 = no, 1 = yes).


Sex and race were included as covariates. Sex was coded as male = 0, female = 1. Race was coded as white = 0 and non-white = 1.

Data Analytic Strategy

We used the recommended longitudinal weighting variable to ensure that results were nationally representative with unbiased estimates (Chen and Chantala 2014). As a result, our analyses assumed that missing data due to attrition were missing at random. This sampling weight was only available for 9421 respondents. Further, we limited the sample to those respondents who were aged 12–18 at Wave 1, which resulted in a final analysis sample of 9027. Variance estimates were adjusted to account for the complex sample design.
We used MPlus v.7.4 to conduct the analyses (Muthén and Muthén 1998–2017). We used latent class growth analysis (LCGA), a type of a growth mixture model where within-class variances are fixed at zero (Muthén 2004; Muthén and Shedden 1999), to determine if there were qualitatively different subgroups that displayed varying trajectories of suicide attempt likelihood across time. This analytic approach can handle highly skewed and categorical data (Feldman et al. 2009), which are often characteristics of suicide data. In our study, each indicator represented the presence or absence of a suicide attempt during that timeframe and was treated as categorical. We estimated continuous latent intercepts as well as linear and quadratic slopes for each class.
LCGA analyses were conducted using maximum likelihood estimation and robust standard errors to account for missing observations (i.e., full information maximum likelihood). In our first step, we estimated one-, two-, three-, and four-class models and compared the models using Bayesian information criterion estimates (BIC), which is an accurate method of determining latent class model fit (Nagin 1999, 2005; Nylund et al. 2007). We also considered entropy estimates to compare how cleanly each model classified cases into discrete suicidal behavior trajectories (Muthén 2004). Likelihood ratio-based tests are often used to assess growth mixture model fit, but they are not appropriate for use with complex survey data (Muthén 2016). Initial model tests and BIC comparisons indicated that LCGA models—without intercept and slope variance estimates—fit the data better than competing growth mixture models that allowed within-class variation.
Our second step was to determine if time-varying risk factors corresponded with time-varying suicide attempt likelihood by evaluating how risk factors assessed in adolescence (i.e., at Wave 1 or 2) and in young adulthood (i.e., at Wave 4) were associated with suicide attempt trajectory membership. We used the three-step multinomial logistic regression-based approach to predict class membership (Asparouhov and Muthén 2014; Vermunt 2003). This approach allowed us to predict individuals’ most likely trajectory classification while accounting for classification uncertainty based on posterior probabilities. The “adolescent limited” trajectory group served as the reference group when comparing it to the consistently “low” group and the “persistently high” group, and the “persistently high” group served as the reference group when comparing it to the “low” group. The models included each predictor assessed at Wave 1 (or Wave 2) and Wave 4 and controlled for effects of race and sex.


Descriptive Data

Almost one in ten respondents (n = 764; 8.5%) reported having attempted suicide at least once across the four waves (3.9% at Wave 1, 3.4% at Wave 2, 1.6% at Wave 3, and 1.4% at Wave 4). Among the 764 in the sample who attempted suicide, 81.2% (n = 620) attempted at only one wave, 15.8% (n = 121) attempted at two waves, 2.9% (n = 22) attempted at three waves, and only 0.1% (n = 1) attempted at all four waves. Table 1 shows descriptive statistics for study variables.
Table 1
Weighted sample descriptives
Wave 1/2
Wave 3/4
Alcohol problems, mean (SD)
−.03 (.67)
.01 (.68)
Depressive symptoms, mean (SD)
6.65 (4.69)
6.05 (4.68)
Family/friend suicide history, %
Partner abuse, %
Delinquency, mean (SD)
.80 (1.25)
.19 (.60)
Impulsivity, mean (SD)
2.96 (1.14)
2.56 (1.04)
SD standard deviation

Model Estimation and Selection

We estimated models with and without quadratic effects to determine if change was nonlinear; models without quadratic effects had smaller BIC estimates, suggesting the quadratic terms did not significantly improve model fit. Intercepts and linear slopes were therefore fit in the final series of suicidal behavior trajectory models. Each of the one- through four-class models was successfully estimated without errors associated with model misspecification. We selected the three-class model as the final model based on its lowest BIC estimate and strong entropy (see Table 2).
Table 2
Fit statistics for each class-structure estimated
BIC Bayesian Information Criterion

Latent Trajectories of Suicidal Behavior

The majority of the analysis sample (93.6%) was classified into Trajectory 1. The intercept was fixed at zero for reference, and the linear slope did not statistically differ from zero. Figure 1 illustrates that this trajectory remained consistently low across time; therefore, youth in this group had a low likelihood of suicide risk from adolescence to emerging adulthood (Table 3). Trajectory 2 accounted for approximately 5.1% of the sample. The intercept of Trajectory 2 was significantly higher than that of the consistently low trajectory, but the linear slope was significantly less than zero (Table 3). Figure 1 illustrates that youth in this trajectory generally had an elevated likelihood of attempting suicide early in adolescence, but this likelihood decreased as youth transitioned into emerging adulthood, which prompted the “adolescent limited” label. Finally, Trajectory 3 accounted for 1.3% of the sample. The intercept of Trajectory 3 was significantly greater than Trajectory 1, but did not significantly differ from Trajectory 2; the linear slope of Trajectory 3 did not significantly differ from zero. Figure 1 indicates that youth in this group typically had a persistent, elevated likelihood of attempting suicide across adolescence and emerging adulthood, thereby prompting the “persistently high” label. Based on participants’ most likely trajectory classification, no one who attempted suicide at more than one wave was classified in the “low” trajectory, and all participants who attempted suicide at three or more waves were classified in the “persistently high” trajectory. In addition, all participants classified in the “persistently high” trajectory attempted suicide at two or more waves. Table 3 contains frequencies of suicide attempts for each trajectory at each wave, which indicates the frequency and incidence trends are strongly related.
Table 3
Characteristics of the three-class model of suicide risk
Latent trajectories
% of samplea
Intercept (SE)
Linear slope (SE)
Wave 1b
Wave 2
Wave 3
Wave 4
1. Low
−.05 (.04)
2. Adolescent limited
4.18 (.27)***
−.61 (.18)***
3. Persistent
4.2 (.49)***
−.05 (.05)
a Based on estimated posterior probabilities
b Average frequency of suicide attempts per assessment
c Fixed as reference
*p < .05; **p < .01; ***p < .001

Predicting Trajectory Membership

Table 4 presents results of the multinomial logistic regression models. These models tested if the hypothesized time-varying risk factors predicted trajectory membership. As hypothesized, the “adolescent limited” trajectory group and the “low” trajectory group significantly differed on risk factors assessed in adolescence (Wave 1 or 2). Specifically, during adolescence, those in the “adolescent limited” group had significantly higher levels of alcohol problems, depressive symptoms, delinquency, and impulsivity, and a greater likelihood of experiencing partner abuse and having a history of family or friend suicidal behavior compared to their lower risk counterparts. By Wave 4 when the sample was young adults, the “adolescent limited” group still had significantly more depressive symptoms and a higher likelihood of partner abuse, but had fewer alcohol problems, compared with the “low” group.
Table 4
Multinomial logistic regressions predicting latent class trajectory group
Low vs. adolescent limiteda
Persistently high vs. adolescent limiteda
Low vs. persistently highb
OR 95% CI
OR 95% CI
OR 95% CI
Wave 1 or 2
 Alcohol problems
.27, .44
.34, .75
.49, .94
 Depressive symptoms
.73, .82
.79, .97
.81, .97
 Family/friend history
.06, .22
.18, 1.76
.09, .45
 Partner abuse
.15, .72
.45, 4.73
.08, .61
.38, .54
.57, 1.06
.45, .77
.57, .93
.49, 1.14
.70, 1.35
Wave 4
 Alcohol problems
1.11, 3.19
1.45, 5.30
.46, .97
 Depressive symptoms
.89, 1.00
1.03, 1.20
.79, .90
 Family/friend history
.40, 7.32
.62, 27.97
.84, 6.94
 Partner abuse
.29, .94
.12, 1.88
.37, 3.29
.62, 1.24
1.29, 3.45
.28, .61
.65, 1.12
.99, 2.86
.34, .76
Control variables
.69, 1.91
.24, 2.57
.56, 3.78
.06, .55
.17, 4.34
.08, .57
SE Standard Error, OR Odds Ratio
a Reference group is “adolescent limited”
b Reference group is “persistently high”
c Coded as white = 0, nonwhite = 1
d Coded as male = 0, female = 1
+p = .05; *p < .05; **p < .01; ***p < .001
In terms of comparisons between the “persistently high” and the “adolescent limited” trajectory groups, there were few significant differences on risk factor levels in adolescence, with the exception that those in the “persistently high” group had significantly lower levels of alcohol problems and depressive symptoms. By young adulthood however, the “persistently high” group had significantly higher levels of alcohol problems, depressive symptoms, delinquency, and impulsivity than those in the “adolescent limited” group.
When compared with those in the “low” trajectory group, those with a “persistently high” risk of suicide attempts were likely to have significantly higher levels of alcohol problems, depressive symptoms and delinquency, and to have a family member or friend with a history of suicidal behavior, as well as to have experienced partner abuse during adolescence. At Wave 4, those in the “persistently high” group still had significantly higher levels of alcohol problems, depressive symptoms, and delinquency, and had higher levels of impulsivity than those in the “low” trajectory group.
Gender significantly differentiated trajectories, such that males were more likely than females to be in the consistently “low” group than either the “adolescent limited” or “persistently high” groups. Race was not a significant predictor of trajectory membership.

Sensitivity Analysis/Alternate Model Analysis

We estimated several alternatively specified models to assure our final model best represented the data. These included a 3-class model with time-varying observations (i.e., by age), which did not converge, due ostensibly to lack of fit. We also estimated models with non-linear change across time (i.e., quadratic effects); a 3-class model with an additional quadratic effect did not fit the data as well (BIC: 8672.22) as the model without the quadratic effect (BIC: 8652.08). Additionally, we estimated a model indicated by the raw frequencies of suicide attempts at each assessment, rather than dichotomous suicide attempts; this frequency-based model (BIC: 10813.04) also did not fit the data as well as the dichotomous model.


This study aimed to extend the research base on the epidemiology of suicidal behavior as youth transition into young adulthood. This research is important because suicide is the second leading cause of death among adolescents and young adults (CDC 2017) and understanding what risk factors predict suicide attempts can ultimately help reduce the public health burden of death by suicide. As noted at the beginning of the article, several risk factors for suicide attempts have been consistently identified in the literature. Few of these studies however have investigated if and how suicide likelihood changes as risk factors change. Studies that have taken a developmental approach have found support for the importance of looking at not only the presence or severity of a particular risk factor, but also its level of change (Kerr et al. 2013). Our study extends the knowledge base by using data from a nationally representative sample of non-clinical youth to investigate how suicide risk changes over the course of 13 years as youth transition from adolescence into young adulthood. Further, by taking a developmental approach, we were able to examine if changes in risk factors accounted for changes in suicidality.
Our investigation yielded at least two important findings. First, our results indicated that there were qualitatively different subgroups that displayed varying trajectories of suicide attempt likelihood across time. We had hypothesized four trajectory classes, a consistently low risk group, a persistently high risk group, an increasing risk group, and a decreasing risk group. However, a three-class model fit the data best. Interpreting the classes revealed that the majority of the sample (93.6%) had a low likelihood of suicide attempts across all four waves of data. One in twenty (5.1%) of the youth had an elevated likelihood of attempting suicide early in adolescence, but this likelihood decreased as they aged. A small percentage (1.3%) of the sample had a persistent, elevated likelihood of attempting suicide across adolescence and emerging adulthood. Counter to hypothesis, we did not find support for an increasing suicide risk group. This could be because our trajectory analyses only extended to when the sample was approximately 25 to 31 years old. It is possible that if one were able to assess trajectories over a lifespan, an increasing trajectory would have emerged. Fortunately, a fifth wave of data collection with the Add Health sample is underway. This will enable us to test empirically this supposition, at least in terms of assessing trajectories through middle adulthood.
A second important finding to emerge from our study is that changes in suicide risk over time corresponded with changes on certain risk factors. In general, youth with a relatively high risk of suicide in adolescence that decreased as they got older had higher levels of depressive symptoms, delinquency, impulsivity, and alcohol problems, and a higher likelihood of experiencing partner abuse and having a family member or friend attempt or complete suicide in adolescence compared to youth in the consistently low suicide risk group. The adolescent limited group only differed on depressive symptoms and alcohol problems in adolescence compared with the persistently high group. Those with a relatively high risk of suicide in both adolescence and young adulthood had higher levels on five of the six risk factors in adolescence compared with their low suicide risk counterparts, and in young adulthood, they had higher levels on four of the six risk factors. By young adulthood, youth whose suicide risk had decreased (i.e., adolescent limited group) had risk factor levels equivalent to their consistently low risk counterparts with the exception of depressive symptoms and partner abuse, and had lower levels of alcohol problems. By adulthood, the adolescent-limited group had lower levels of depressive symptoms, delinquency, and alcohol problems than their counterparts in the persistently high risk group. These data reflect that adolescents who had a relatively high likelihood of suicidal behavior in adolescence but decreased in their risk as they transitioned into young adulthood showed concurrent decreases in time-varying risk factors; yet adolescents who had a relatively high risk for suicidal behavior in adolescence and young adulthood showed high levels on several risk factors across time.
There were some limitations that should be noted. First, we focused on suicide attempts rather than completions. However, research indicates that a nonfatal suicide attempt is the strongest predictor of a fatal attempt (Bostwick et al. 2016), and risk factors for attempts and completions are similar (Gould and Kramer 2001). There are also potential limitations posed by how Add Health measured suicide attempts. This question was only asked of those who acknowledged engaging in suicidal ideation. Thus, a respondent who attempted suicide without first considering suicide would be misclassified as a false negative. Third, data were based on self-report only. Fourth, we only included risk factor variables that were assessed similarly at earlier and later waves in Add Health, given we were interested in how changes in risk factors corresponded with changes in suicide risk. As a result, many risk factor variables that the literature indicates increase likelihood of suicide attempts, such as lack of belongingness and perceived burdensomeness (Joiner 2005), were not assessed. Also, due to the limited assessment of protective factors and because the few that were assessed were not measured at earlier and later waves, we were not able to determine what factors protected respondents from being in a high risk trajectory. Finally, the data were collected between 1995 and 2008, and were thus somewhat dated. However, while prevalence of the variables under study might change over time (e.g., degree of depression in youth), there is little reason to expect that the associations of the variables under study (e.g., depression’s association with suicide attempts) would change. Thus, despite the aging data of Add Health, the findings are still useful.
Despite these limitations, this work extends the literature base in several ways. Taking a person-centered, developmental, longitudinal perspective, we were able to determine how suicide risk changed as youth matured into young adulthood. Youth who are at risk in adolescence did not necessarily remain at high risk. Most reduced their suicidal behavior likelihood as their risk factors correspondingly declined. However, some youth at high suicide attempt risk in adolescence maintained their high risk status through emerging adulthood, and this corresponded with the continuation of higher levels on certain risk factors, most notably in this study, depression, delinquency, impulsivity, and alcohol problems. The identification of time-varying risk factors whose levels correspond with suicide attempt risk provides insights for possible targets for interventions. Interventions designed to decrease depression, delinquency, impulsivity, and alcohol problems among youth at high risk for suicidal behavior hold promise, as our data indicate that reductions in these risk factors over time corresponded with desistance or categorization in the adolescent limited group. Further, prevention efforts designed to minimize risk factors that differentiated the low risk group from the other trajectories can offset the initiation of suicidality. In sum, study results suggest that suicidal behavior can be mitigated; not all youth who show high risk in adolescence will continue to be at high risk in young adulthood. Intervention programs need to focus on risk factors that not only are amenable to change but also show evidence of altering suicide attempt likelihood.


This article used four waves of data from the Add Health study to investigate how suicide attempt likelihood changed as a sample of 9027 youth aged 12–18 matured into young adulthood. This study elucidates the importance of taking a developmental perspective in assessing the epidemiology of suicidal behavior. Using longitudinal data, we were able to uncover qualitatively different subgroups with varying trajectories of suicide attempt likelihood across time. Fortunately, most youth are at low risk for suicide attempts during adolescence and remain at low risk through young adulthood. It is promising to note that for most youth who were at high risk for suicide attempts during adolescence, their suicide risk declined. This study supported the utility of a developmental perspective by elucidating how changes on established risk factors for suicidality during the emerging adulthood period had contemporaneous associations with suicide risk. Psychological, behavioral, and personal history variables are not necessarily static, and as youth transition into young adulthood, these variables can change in direction and magnitude. It is promising to note that for youth whose suicide risk decreased over time, this decrease corresponded with declines in impulsivity and delinquency, and a lower likelihood of past year family/friend suicide history to equivalent levels as youth at low suicide risk, and corresponded with reductions in depressive symptoms, impulsivity, delinquency, and alcohol problems relative to youth who remained at high suicide risk. This suggests that suicide risk is not static and interventions designed to influence these time-varying risk factors can help prevent suicide morbidity and mortality.


This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations.


This research was supported by a Distinguished Investigator Grant (0-138-13) awarded by the American Foundation for Suicide Prevention to the first author.

Authors’ Contributions

M.T. conceived of the study, acquired the data, participated in the study plan, reviewed statistical analyses, and drafted the manuscript; K.S. participated in the design of the study plan, performed the statistical analysis, and helped to draft the manuscript. All authors read and approved the final manuscript.

Compliance with Ethical Standards

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Conflicts of Interest

The authors declare that they have no competing interests.

Ethical Approval

The analysis of the restricted data presented in the study detailed in this manuscript was approved by Clemson University’s IRB.

Informed Consent

Add Health participants provided written informed consent for participation in all aspects of Add Health in accordance with the University of North Carolina School of Public Health Institutional Review Board guidelines that are based on the Code of Federal Regulations on the Protection of Human Subjects 45CFR46.

Onze productaanbevelingen

BSL Psychologie Totaal

Met BSL Psychologie Totaal blijft u als professional steeds op de hoogte van de nieuwste ontwikkelingen binnen uw vak. Met het online abonnement heeft u toegang tot een groot aantal boeken, protocollen, vaktijdschriften en e-learnings op het gebied van psychologie en psychiatrie. Zo kunt u op uw gemak en wanneer het u het beste uitkomt verdiepen in uw vakgebied.

Over dit artikel

Andere artikelen Uitgave 4/2018

Journal of Youth and Adolescence 4/2018 Naar de uitgave