Introduction
Suicide is a major and growing public health concern in the U.S. Per capita, age-adjusted suicide rates increased by 35% in the 20 years preceding the COVID-19 pandemic (CDC, 2021a), dropped slightly during the pandemic (CDC, 2021a), then trended upward as the pandemic subsided (CDC, 2021b). Younger U.S. cohorts are particularly vulnerable. They are the only demographic for whom observed suicide rates were greater than expected during the pandemic (Bridge et al., Reference Bridge, Ruch, Sheftall, Hahm, O’Keefe, Fontanella, Brock, Campo and Horowitz2023), and they showed the steepest growth in years before the pandemic, including a 300% increase for 10–14-year-old girls and a 66% increase for 10–17-year-old children and adolescents of color (CDC, 2021b; Hedegaard et al.Reference Hedegaard, Curtin and Warner2020). Increasing rates of suicide ideation, suicide attempts, and suicide among Black youth are of utmost concern given the much lower risk historically than among White youth (Bridge et al., Reference Bridge, Horowitz, Fontanella, Sheftall, Greenhouse, Kelleher and Campo2018; Sheftall et al., Reference Sheftall, Vakil, Ruch, Boyd, Lindsey and Bridge2022).
For girls, rates of nonsuicidal self-injury (NSSI), defined by acts of intentional harm to one’s own body without intent to die, are also increasing, while the modal age of onset is declining (Ammerman et al., Reference Ammerman, Jacobucci, Kleiman, Uyeji and McCloskey2018; Zetterqvist et al., Reference Zetterqvist, Jonsson, Landberg and Sved2021). These trends are alarming as well given the high risk for later suicide associated with NSSI, particularly for early starters (Ammerman et al., Reference Ammerman, Jacobucci, Kleiman, Uyeji and McCloskey2018; Ribeiro et al., Reference Ribeiro, Franklin, Fox, Bentley, Kleiman, Chang and Nock2016). Age-period-cohort analyses attribute self-injury and suicide trends among U.S. youth to cohort effects (Joe et al., Reference Joe, Banks and Belue2016; Martínez-Alés et al., Reference Martínez-Alés, Pamplin, Rutherford, Gimbrone, Kandula, Olfson, Gould, Shaman and Keyes2021; Phillips, Reference Phillips2014). In contrast to age and period effects, which diminish with maturation and time, cohort effects are enduring group-level attributes (e.g., health behaviors, social mores), worldviews (e.g., beliefs about fairness and justice), and patterns of cognition (e.g., attributions about others’ behavior) that derive from cultural experiences (e.g., social movements) and major events (e.g., pandemics, catastrophes) that affect most members of a generation (Twenge et al., Reference Twenge, Cooper, Joiner, Duffy and Binau2019). Recent cohorts of young women, including U.S. high school seniors, report higher levels of anxiety and depression, greater external and lower internal locus of control, and lower levels of agency than members of any cohort since the mid-1970s, when surveys began (Haidt, Reference Haidt2023; Twenge, Reference Twenge2020). Rising cohort-level suicides and worsening mental health will almost certainly exert enduring upward pressure on U.S. suicide rates in years ahead (Joe et al., Reference Joe, Banks and Belue2016), contrary to the widespread narrative of expected declines (e.g., CDC, 2021b). Reversing U.S. suicide trends is more likely to be achieved through scientific advances that specify heterogeneous pathways of vulnerability and risk, and by translation of such advances into targeted prevention programs. These programs must be scalable, more effective, and easier to implement than current approaches, which collectively exert small effects (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Wolff et al., Reference Wolff, Thompson, Thomas, Nesi, Bettis, Ransford, Scopelliti, Frazier and Liu2019).
Primary versus secondary prevention
Reducing U.S. suicide rates and forestalling future attempts among the youngest U. S. cohorts will require a sharpened focus on primary prevention–disrupting developing disease states by identifying and altering precursors to those disease states – in this case, by identifying premorbid markers of first attempts (AbdulRaheem, Reference AbdulRaheem2023). At present, both U.S. systems of care and federal funding agencies invest disproportionate resources on secondary prevention – halting episodic recurrence and/or progression of acute disease states – in this case, suicidal behaviors (SBs) – among those who have already attempted (Ganz et al., Reference Ganz, Braquehais and Sher2010). In contrast, primary prevention assesses more distal vulnerability and risk to determine who is at sufficiently elevated odds of developing future suicidal behaviors. These individuals can then be targeted for enrollment in prevention programs, months to years before a likely event. Predictive time windows may be a few months to many years, which constrains prediction because of the low base rate of suicide (Baldessarini et al., Reference Baldessarini, Finkleste and Arana1983). Suicidal behaviors include past histories of attempts and preparatory actions to attempt (Hamza et al., Reference Hamza, Stewart and Willoughby2012; David Klonsky et al., Reference David Klonsky, May and Glenn2013). Person-level traits (e.g., impulsivity; see Crowell et al., Reference Crowell, Beauchaine and Linehan2009) and environmental events (e.g., physical and sexual abuse; see Hinshaw et al., Reference Hinshaw, Owens, Zalecki, Huggins, Montenegro-Nevado, Schrodek and Swanson2012) that predictably emerge/occur before first attempts are potential targets for primary prevention, even if they continue after attempts and are therefore legitimate targets for secondary prevention as well. This approach parallels common practices in medicine for preventing physical diseases. Obesity and blood sugar, for example, are targets of both primary and secondary prevention of diabetes, as are cholesterol and blood pressure for heart disease.
Following this rationale, we distinguish between markers of vulnerability and risk that emerge before SBs versus after SBs, even when these markers overlap and are useful for both primary and secondary prevention. We acknowledge the age-old problem of using single vulnerabilities and risk factors for predicting future suicide given low base rates (see Franklin et al., Reference Franklin, Ribeiro, Fox, Bentley, Kleiman, Huang, Musacchio, Jaroszewski, Chang and Nock2017), resulting in large numbers of false positives. We therefore focus our attention on multiple, intersecting vulnerabilities and risk factors, which, when taken together, can improve prediction by wide margins (see Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019). Alone, even NSSI, the most potent prospective predictor of first attempts besides SBs themselves (Hamza et al., Reference Hamza, Stewart and Willoughby2012; David Klonsky et al., Reference David Klonsky, May and Glenn2013), identifies many youth who later desist with no attempts.
Nonsuicidal self-injury and prevention of suicidal behaviors and suicide
Nonsuicidal self-injury (NSSI), defined by acts of intentional harm to one’s own body without intent to die, is the strongest univariate predictor of later suicide other than suicidal behavior (SB) (Hamza et al., Reference Hamza, Stewart and Willoughby2012; David Klonsky et al., Reference David Klonsky, May and Glenn2013). NSSI is therefore of central interest to both primary and secondary prevention researchers, whose opinions often differ on whether or not NSSI should be included on a spectrum of suicidal behaviors (Griep & MacKinnon, Reference Griep and MacKinnon2022; Hamza & Willoughby, Reference Hamza and Willoughby2014; Nock et al., Reference Nock, Borges, Bromet, Cha, Kessler and Lee2008; Victor & Klonsky, Reference Victor and Klonsky2014).
The secondary prevention literature is organized primarily around ideation-to-action models, such as the Three-Step Theory (Klonsky et al., Reference Klonsky, Pachkowski, Shahnaz and May2021), Interpersonal Theory (Van Orden et al., Reference Van Orden, Witte, Cukrowicz, Braithwaite, Selby and Joiner2010), and Fluid Vulnerability Theory (Klonsky et al., Reference Klonsky, Saffer and Bryan2018). These models focus on transitions to higher levels of risk, under the assumption that understanding transitions will eventually lead to prediction and control. Important questions are therefore, “What explains initiation of NSSI, and, what explains transitions from SI to SAs” (e.g., Park & Ammerman, Reference Park and Ammerman2023; O’Loughlin et al., Reference O’Loughlin, Burke and Ammerman2021)? Within-person prediction of such transitions has obvious implications for halting SB progression. Given its association with later SBs, NSSI is of central interest to prevention researchers, but opinions sometimes differ on whether or not NSSI itself should be placed on a spectrum of SBs (Griep & MacKinnon, Reference Griep and MacKinnon2022; Hamza & Willoughby, Reference Hamza and Willoughby2014; Nock et al., Reference Nock, Borges, Bromet, Cha, Kessler and Lee2008; Victor & Klonsky, Reference Victor and Klonsky2014). In the secondary prevention literature, ideation-to-action models predominate, including Three-Step Theory (Klonsky et al., Reference Klonsky, Pachkowski, Shahnaz and May2021), Interpersonal Theory (Van Orden et al., Reference Van Orden, Witte, Cukrowicz, Braithwaite, Selby and Joiner2010), and Fluid Vulnerability Theory (Klonsky et al., Reference Klonsky, Saffer and Bryan2018). These models focus on transitions to higher levels of risk (e.g., from NSSI to SAs), under the assumption that understanding such transitions will eventually lead to prediction and control. Important questions are therefore, “What explains initiation of NSSI, and what explains transitions from SBs, including ideation, to SAs” (e.g., Park & Ammerman, Reference Park and Ammerman2023; O’Loughlin et al., Reference O’Loughlin, Burke and Ammerman2021)? Within-person prediction of such transitions has obvious implications for halting SB progression.
Much of the secondary prevention literature distinguishes sharply between NSSI and SBs, placing the former outside the SB continuum (Muehlenkamp, Reference Muehlenkamp2006). By definition, NSSI lacks suicidal intent and portends about 10% lifetime suicide risk (Berman et al., Reference Berman, Jobes and Silverman2006). Although orders of magnitude above population-level risk, 10% is still insufficient for prospective prediction (Franklin et al., Reference Franklin, Ribeiro, Fox, Bentley, Kleiman, Huang, Musacchio, Jaroszewski, Chang and Nock2017). Moreover, although up to 30% of girls report engaging in NSSI at least once (Sornberger et al., Reference Sornberger, Heath, Toste and McLouth2012; Swannell et al., Reference Swannell, Martin, Page, Hasking and St John2014), most desist (see Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019). Finally, many who die by suicide have no history of self-injury (Bresin & Schoenleber, Reference Bresin and Schoenleber2015; Victor et al., Reference Victor, Muehlenkamp, Hayes, Lengel, Styer and Washburn2018). For these and other reasons, NSSI is usually not included among SBs.
Nonsuicidal self-injury and suicide risk in primary prevention
Yet certain developmental pathways to SAs almost invariably include NSSI that predates SBs (see Crowell & Kaufman, Reference Crowell and Kaufman2016; Linehan, Reference Linehan1993). Among girls who engage in NSSI at least once (Swannell et al., Reference Swannell, Martin, Page, Hasking and St John2014), a highly vulnerable subset progresses predictably from NSSI to SBs. Much like illicit progression vs. desistence of substance experimentation (NIAAA, 2023) and binge-purge behavior (Solmi et al., Reference Solmi, Sonneville, Easter, Horton, Crosby, Treasure, Rodriguez, Jarvelin, Field and Micali2015), age-of-onset, frequency, and severity of NSSI differentiate those whose self-injury persists and transitions to SBs from those whose self-injury subsides. Girls who initiate self-injury before age 10 show an especially pernicious course, including greater frequency of NSSI, use of more diverse and dangerous methods, higher hospitalization rates, greater general impairment, and wider-ranging psychiatric morbidities (see Ammerman et al., Reference Ammerman, Jacobucci, Kleiman, Uyeji and McCloskey2018; Klonsky, Reference Klonsky2011). In turn, self-injury severity, including lethality and frequency – especially of cutting – is a strong predictor of later ideation and attempts (O’Loughlin et al., Reference O’Loughlin, Burke and Ammerman2021; Stewart et al., Reference Stewart, Eaddy, Horton, Hughes and Kennard2017b). Among adolescent psychiatric inpatients who self-injure, 70% also report ideation (Nock & Kessler, Reference Nock and Kessler2006). This number rises to over 85% for adolescent girls who are recruited based solely on histories of at least moderate lethality NSSI (3 or more episodes in 6 months or 5 or more lifetime; see Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Crowell et al., Reference Crowell, Beauchaine, McCauley, Smith, Stevens and Sylvers2005). Coding NSSI dichotomously, as is common, obscures these very strong frequency and severity effects.
Given this discussion, our view is that both positions regarding NSSI are correct. Although NSSI affects up to 30% of adolescents, especially girls (Sornberger et al., Reference Sornberger, Heath, Toste and McLouth2012), most of the burden of self-harm is driven by a much smaller number of early-starting, highly symptomatic girls who experience persistent and escalating self-harm that almost always transitions to SBs. For these girls, NSSI belongs on the SB continuum. Others engage in NSSI but mature out. For them, NSSI does not belong on the SB continuum.
More generally, preoccupation with pathognomonic signs, or symptoms and behaviors that always mark single disease states and predict their course, belies the overwhelming multifactorial complexity of self-injury, SBs, and almost all forms of psychopathology. Sustained searches for large, independent effects that predict all cases of a condition or disorder have hampered and will continue to hamper our understanding of etiology, and our effectiveness in developing and implementing precision care (Beauchaine & Constantino, Reference Beauchaine and Constantino2017; Beauchaine & Hinshaw, Reference Beauchaine and Hinshaw2020).
Specifying multiple pathways to suicide: a necessary step for effective prevention
Effective prevention is only possible when heterogeneous pathways to suicide are well described, providing mechanistic targets. Suicide is an equifinal (phenocopied) outcome that cannot be accounted for by any single etiology (Beauchaine et al., Reference Beauchaine, Klein, Crowell, Derbidge and Gatzke-Kopp2009). This is a problem for ideation-to-action models, which are designed to be universal – to account for psychological states faced by all or nearly all who attempt suicide (Klonsky et al., Reference Klonsky, Pachkowski, Shahnaz and May2021). In the paragraphs to follow, we present one common pathway that extends Linehan’s (Reference Linehan1993) theoretical framework and was derived and validated from work with samples of primarily urban-dwelling White adolescent girls (Crowell et al., Reference Crowell, Beauchaine, McCauley, Smith, Stevens and Sylvers2005, Reference Crowell, Beauchaine, McCauley, Smith, Vasilev and Stevens2008). This was the first youth population to show rapidly accelerating NSSI converting to SBs in the early 2000s (Klonsky, Reference Klonsky2011). One objective of this article is to test the model described herein in a more diverse sample. We note, however, that such validation does not obviate the need for models specific to vulnerable subpopulations, as called for in the NIMH Strategic Framework (NIMH, 2023). Contexts, motivations, and even neuro-biological correlates of NSSI and SBs differ by race and sex in ways that cannot be understood fully without sex- and culture-specific research (Sheftall & Miller, Reference Sheftall and Miller2021).
A developmental transactional model
In developmental psychopathology, contemporary etiological models are invariably transactional, assuming bidirectional influences among person-level vulnerabilities (e.g., genes, neural functions, temperament) and environmental risk factors (e.g., invalidation, abuse, deviant peer affiliations) over time. Such models accommodate etiological complexity and have been applied fruitfully to internalizing disorders (Hankin et al., Reference Hankin, Snyder, Gulley, Schweizer, Bijttebier, Nelis, Toh and Vasey2016), externalizing disorders (Beauchaine et al., Reference Beauchaine and Constantino2017), developmental disabilities (Dawson, Reference Dawson2008), and more. Suicide, however, is a specific behavior observed across disorders, emerging within heterogeneous developmental contexts of those disorders. It is therefore unsurprising that the first transactional model of NSSI and suicide risk was disorder-specific (it has since been extended).
Linehan (Reference Linehan1993) was the first to propose a transactional model of NSSI and SBs in her highly influential model of borderline personality development. Based on extensive clinical work with hundreds of patients, Linehan observed that most suicidal adult women engaged in NSSI before engaging in SBs (Brent, Reference Brent2011; Groschwitz et al., Reference Groschwitz, Plener, Kaess, Schumacher, Stoehr and Boege2015), a pattern subsequently observed among adolescents and other samples of adults (Andover et al., Reference Andover, Morris, Wren and Bruzzese2012; Bryan et al., Reference Bryan, Bryan, May and Klonsky2015; Kiekens et al., Reference Kiekens, Hasking, Boyes, Claes, Mortier, Auerbach, Cuijpers, Demyttenaere, Green, Kessler, Myin-Germeys, Nock and Bruffaerts2018; Taliaferro et al., Reference Taliaferro, Almeida, Aguinaldo and McManama O’Brien2019).Footnote 1 Linehan proposed that in most cases, NSSI emerges from emotion dysregulation (ED), the inability to modulate negative affect, especially anger, fear, and sadness. She proposed that ED develops through thousands of transactions between a then-unknown predisposing biological vulnerability and coercive, invalidating, and sometimes abusive family environments that shape and maintain physiological arousal and emotional lability (Linehan, Reference Linehan1993). Within troubled families, ED is often effective in terminating aversive interactions, providing relief from associated psychological distress through negative reinforcement/escape conditioning (see Crowell et al., Reference Crowell, Baucom, McCauley, Potapova, Fitelson, Barth, Smith and Beauchaine2013; Liu, Reference Liu2017).
Over time, ED generalizes to social relationships and other interpersonal contexts outside the family (Beauchaine & Zalewski, Reference Beauchaine, Zalewski, Dishion and Snyder2016), and is further shaped in deviant peer groups (Crowell et al., Reference Crowell, Baucom, McCauley, Potapova, Fitelson, Barth, Smith and Beauchaine2013). For example, many adolescents who engage in NSSI face difficulties establishing positive peer relationships, and in the relationships that they do establish, they show extreme sensitivity to peer rejection and engage in frequent conflict (Adrian et al., Reference Adrian, Zeman, Erdley, Lisa and Sim2011; de Luca et al., Reference de Luca, Giletta, Nocentini and Menesini2022). This conflict is often followed by increased urges and episodes of NSSI, which become a go-to coping mechanism by providing relief from the emotional fallout of social relationship difficulties and other sources of negative affect (see Crowell et al., Reference Crowell, Baucom, McCauley, Potapova, Fitelson, Barth, Smith and Beauchaine2013; McKenzie & Gross, Reference Joiner, Van Orden, Witte, Selby, Ribeiro, Lewis and Rudd2014; Nixon et al., Reference Nixon, Cloutier and Aggarwal2002).
Emotion dysregulation in social relationships and NSSI as escape from negative emotion have different effects on physiological function. ED reflects sensitization to emotional pain (Linehan, Reference Linehan1993). Self-injuring adolescent girls show strong neural and psychophysiological responses during social conflict and peer rejection, indexed by fMRI, respiratory sinus arrhythmia reactivity, and skin conductance reactivity (e.g., Crowell et al., Reference Crowell, Butner, Wiltshire, Munion, Yaptangco and Beauchaine2017; Haines et al., Reference Haines, Bell, Crowell, Hahn, Kamara, McDonough-Caplan, Shader and Beauchaine2019). These girls are similarly reactive to strong emotion inductions when social stimuli are used (e.g., Crowell et al., Reference Crowell, Beauchaine, McCauley, Smith, Stevens and Sylvers2005), and when excluded or rejected by age-matched confederates (Groschwitz et al., Reference Groschwitz, Plener, Groen, Bonenberger and Abler2016).
In contrast, despite some mixed findings, NSSI often produces desensitization to physical pain and associated stress responding, as reflected in lower β-endorphin levels (van der Venne et al., Reference van der Venne, Balint, Drews, Parzer, Resch, Koenig and Kaess2021), blunted HPA axis responding (e.g., Beauchaine et al., Reference Beauchaine, Crowell and Hsiao2015), and both increased pain thresholds and lower self-reported perceptions of pain during lab-based exposures (e.g., Franklin et al., Reference Franklin, Ribeiro, Fox, Bentley, Kleiman, Huang, Musacchio, Jaroszewski, Chang and Nock2017; van der Veene et al.). The combination of sensitization to emotional pain, use of NSSI to escape such pain, and desensitization to physical pain can be lethal given more severe NSSI and stronger NSSI-SA associations among those who experience lower subjective pain (Ammerman et al., Reference Ammerman, Kleiman, Uyeji, Knorr and McCloskey2015). This is consistent with acquired capacity models (Joiner et al., Reference Joiner, Van Orden, Witte, Selby, Ribeiro, Lewis and Rudd2009).
In sum, evidence that ED is socialized within families and confers vulnerability to NSSI is now overwhelming (e.g., Beauchaine & Zalewski, Reference Beauchaine, Zalewski, Dishion and Snyder2016; Xu et al., Reference Xu, Spinrad, Cookston and Matsumoto2020), and supported by work across neural, physiological, behavioral, and observational levels of analysis (e.g., Crowell et al., Reference Crowell, Butner, Wiltshire, Munion, Yaptangco and Beauchaine2017; Wolff et al., Reference Wolff, Thompson, Thomas, Nesi, Bettis, Ransford, Scopelliti, Frazier and Liu2019). Yet as Linehan anticipated in proposing an unspecified biological vulnerability, ED alone cannot explain NSSI or SBs. Most forms of psychopathology are characterized by ED (Beauchaine & Cicchetti, Reference Beauchaine and Cicchetti2019; Beauchaine & Thayer, Reference Beauchaine and Thayer2015) but most are not characterized by NSSI or SBs. Alone, ED is therefore of very limited value in identifying prospective risk for suicide, or for targeted recruitment into primary prevention programs.
An elaborated transactional model
In collaboration with Linehan, we refined her theory, describing trait impulsivity as the specific predisposing biological vulnerability. Based on the current literature, we proposed that ED confers vulnerability to, but is insufficient for the development of NSSI and SBs (Crowell et al., Reference Crowell, Beauchaine and Linehan2009). We further proposed that NSSI is most likely to develop when highly heritable trait impulsivity (h 2 ` ∼ .8) – indexed by hyperactive-impulsive symptoms of ADHD – is coupled with socialized deficiencies in emotion regulation (Beauchaine et al., Reference Beauchaine, Klein, Crowell, Derbidge and Gatzke-Kopp2009; Crowell et al.; for updates see Beauchaine et al., Reference Beauchaine, Hinshaw and Bridge2019; Beauchaine Reference Beauchaine, Lejuez and Gratz2020). According to this perspective, preexisting trait impulsivity is magnified by ED (or mollified by emotion regulation) to increase the risk for NSSI considerably. Like ED, however, impulsivity alone is insufficient vulnerability for the development of NSSI and SBs. It is important to emphasize that impulsivity and self-regulation are independent in the population (Friedman et al., Reference Friedman, Hatoum, Gustavson, Corley, Hewitt and Young2020).
There is growing support for the Impulsivity × ED interaction hypothesis (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019). A major source of evidence comes from studies of girls who are diagnosed in childhood with ADHD and later experience maltreatment versus those diagnosed with ADHD but do not experience later maltreatment. Physical abuse, sexual abuse, and neglect are potent sources of ED (Dvir et al., Reference Dvir, Ford, Hill and Frazier2014), which often mediates associations between maltreatment and psychopathology (Jennissen et al., Reference Jennissen, Holl, Mai, Wolff and Barnow2016). As shown in Figure 1, rates of NSSI and suicide attempts are elevated somewhat among adolescent girls with ADHD only (impulsivity), and among girls who have been maltreated only (ED). The risk associated with the combination of ADHD and maltreatment, however, is alarming. Girls who are diagnosed with ADHD at ages 6-12 and later incur maltreatment—an environmental risk factor with potent effects on ED—are at 50% risk for NSSI at ages 17–23, and over 33% risk for suicide attempts (Guendelman et al., Reference Guendelman, Owens, Galan, Gard and Hinshaw2016; Hinshaw et al., Reference Hinshaw, Owens, Zalecki, Huggins, Montenegro-Nevado, Schrodek and Swanson2012).
Self-injury and Suicide Attempt Rates by Risk Status.
Stress
Broadly speaking, stressful events are experiences that prompt aversive emotional reactions (WHO, n.d.). Ideally, these reactions motivate adaptive engagement to either eliminate the stressor or strategically cope. In contexts of extreme, protracted, or life-threatening stress, humans often behave at the behest of strong emotions, often unable to consider alternative responses. This is ED defined, but it is appropriately situational—it is a state, not a trait. As described initially by Linehan (Reference Linehan1993), many who self-injure, whether or not they have attempted suicide—are highly sensitive to stress and respond with ED asituationally. ED transitions into their primary learned coping strategy—eventually becoming automated with the stability of a psychological trait (e.g., Daros et al., Reference Daros, Daniel, Boukhechba, Chow, Barnes and Teachman2020).
A sizable literature on NSSI, including studies of samples in which most adolescents engage in SBs, demonstrates stronger than normal behavioral, physiological, and neural responses to stressors. Excessive physiological reactivity can constrain attentional resources needed for situationally adaptive behavior, as described above (Zimmer-Gembeck & Skinner, Reference Zimmer-Gembeck, Skinner and Cicchetti2016). Findings are particularly consistent for interpersonal stressors such as parent-child relationship difficulties, peer conflict, and romantic relationship problems (Baetens et al., Reference Baetens, Claes, Onghena, Grietens, van Leeuwen, Pieters, Wiersema and Griffith2015; Bendezú et al., Reference Bendezú, Calhoun, Patterson, Findley, Rudolph, Hastings, Nock and Prinstein2022; Crowell et al., Reference Crowell, Beauchaine, McCauley, Smith, Stevens and Sylvers2005; Reference Crowell, Butner, Wiltshire, Munion, Yaptangco and Beauchaine2017; de Luca et al., Reference de Luca, Giletta, Nocentini and Menesini2022; Kaess et al., Reference Kaess, Hooley, Klimes-Dougan, Koenig, Plener, Reichl, Robinson, Schmahl, Sicorello, Westlund Schreiner and Cullen2021; Koenig et al., Reference Koenig, Rinnewitz, Warth, Hillecke, Brunner, Resch and Kaess2017; Sauder et al., Reference Sauder, Derbidge and Beauchaine2015).
Although longitudinal studies are lacking, stressful life events may precipitate transitions from NSSI to SAs (e.g., Vergara et al., Reference Vergara, Jobes and Brausch2023). Qualitative research suggests increased risk for SBs when NSSI no longer modulates stress-related negative affect (Taliaferro et al., Reference Taliaferro, Almeida, Aguinaldo and McManama O’Brien2019). This is unsurprising given similar effects of stress on the progression of most psychiatric disorders (see e.g., Cicchetti & Rogosch, Reference Cicchetti and Rogosch2002).
Associations between stress and SAs in adolescence have been described for decades (e.g., Wilson et al., Reference Wilson, Hicks, Foster, McGue and Iacono2015). Among vulnerable adolescents, stronger psychological and physiological stress responses to emotionally evocative life events may potentiate suicidal tendencies (Miller & Prinstein, Reference Miller and Prinstein2019). Certain stressors, particularly social stressors, tend to precede SAs in adolescence. These include difficulties with romantic relationships (Sheftall et al., Reference Sheftall, Asti, Horowitz, Felts, Fontanella, Campo and Bridge2016), bullying (Koyanagi et al., Reference Koyanagi, Oh, Carvalho, Smith, Haro, Vancampfort, Stubbs and DeVylder2019), family conflict (Oppenheimer et al., Reference Oppenheimer, Glenn and Miller2022), and both discrimination and minority stress (AACAP, 2022).
The current study
Here we extend research on the Impulsivity × ED pathway to a diverse prospective cohort of 344 adolescents who were treated for major depressive disorder (MDD). Using longitudinal data collected at four time points (baseline, 6 months, 1 year, 2 years), we construct theoretically informed multilevel models (MLMs) to evaluate both distal (before study entry) and proximal (during the two-year observation) predictors and correlates of NSSI, SI, and SAs among vulnerable adolescents, as described in detail below.
We purposefully select MLM over other longitudinal methods, including growth mixture modeling (GMM) and cross-lag panel modeling (CLPM). GMM, which divides samples into latent subgroups based on differences in growth trajectories, almost always yields multiple group solutions even with dimensional data and often identifies more groups than present when the group structure is known (Shader & Beauchaine, Reference Shader and Beauchaine2022). Moreover, we have no strong group hypotheses with this high-risk sample.
CLPM is often used to infer directions of effect and reciprocal effects in multiwave data, but the method does not account for time-invariant (trait-like) effects (Hamaker, et al., 2015), in part because trait variance is partialled out of Time 1 to Time 2 paths, making it unavailable for testing effects in later paths (see Beauchaine & Slep, Reference Beauchaine and Slep2018). Finally, neither approach is well-suited for binary, zero-inflated data, such as SAs. MLM can readily handle such data by specifying Bernoulli distributions.
Few studies distinguish between proximal and distal vulnerabilities and risk factors for NSSI and SBs, and none that we know of do so with more than two timepoints. In addition, most related studies include girls who are highly impulsive (ADHD) or who self-injure at baseline – not both – and most studies lack diversity. We recruit a more diverse sample of adolescents with histories of depression. This provides an opportunity to replicate and extend the Impulsivity × ED hypothesis, putting it to a broader test. We note that trait impulsivity and impulsive decision-making are not uncommon among depressed adolescents (e.g., Amlung et al., Reference Amlung, Marsden, Holshausen, Morris, Patel, Vedelago, Naish, Reed and McCabe2019; McDonough-Caplan et al., Reference McDonough-Caplan, Klein and Beauchaine2018; Zisner & Beauchaine, Reference Zisner and Beauchaine2016), so restricted range is of limited concern. Our primary hypotheses are (1) impulsivity combined with ED (the Impulsivity × ED interaction) will portend more persistent NSSI, SI, and SAs, over-and-above main effects of impulsivity and ED alone, and over-and-above other predictors; (2) NSSI will mark both distal and proximal risk for SI and SAs; and (3) stress will covary with longitudinal trajectories of NSSI, SI, and SAs, Further details about analyses appear below.
Method
Sample
Participants were recruited from a major metropolitan children’s hospital inpatient and outpatient facilities and affiliated community behavioral health clinics. Potentially eligible youth (N = 724) were those who screened positive for elevated depressive symptoms using an adapted version of the Patient Health Questionnaire for Adolescents (PHQ-A; Johnson et al., Reference Johnson, Harris, Spitzer and Williams2002) and whose parents or legal guardians agreed to be contacted about research opportunities. Patients indicating at least moderate levels of depression on the PHQ-A were contacted by study staff with information. Further eligibility for this study included a diagnosis of MDD and at least one parent or legal guardian willing to participate. Exclusion criteria included a history of brain injury, a diagnosis of schizophrenia, schizoaffective, and/or bipolar disorder, a significant neurological condition such as multiple sclerosis or epilepsy, an IQ <70, and inability of youth or parent(s) to speak and read English fluently.
There were 344 adolescents, ages 12–15 years (31% male, 37% non-White) at baseline. As in most multiwave studies, some participants were lost to attrition at each assessment. Attrited and retained participants did not differ by age, sex, or race. Low income, however, was associated with attrition at both the one-yr, χ2(1) = 14.54, p = .02, and 2-yr follow-ups, χ2(6) = 14.95, p = .02, but not at six-months follow-up, χ2(6) = 4.82, p = .56. Data imputation procedures appear below. The study was approved by the Institutional Review Board at Nationwide Children’s Hospital.
Assent and informed consent were obtained from all participants and a parent/guardian. Participants were informed in the consent and assent processes that acceptance or refusal to participate would not influence their receipt of healthcare at recruitment sites. Interviews were approximately 1.5 – 2 h long for both youth and parents/guardians, and the behavioral tasks took 30 – 60 min. Both adolescents and caregivers were compensated $30 – $40, depending on the number of tasks completed at each assessment. Participant caregivers and adolescents could earn up to $150 and $190, respectively, over the course of the two-year study.
Measures
Demographics
Baseline demographic data were collected using the General Information Sheet (Brent et al., Reference Brent, Perper, Moritz, Allman, Friend, Roth, Schweers, Balach and Baugher1993). This includes the number of people living in the youth’s home, annual household income, and the participating caregiver’s education, relationship status, race, ethnicity, sex, gender, and parental role (e.g., mother, father, other biological caregiver, etc.). Adolescent race (1 = White, 0 = not White), sex (1 = female, 0 = male), and family income before taxes (0 = less than 5,000, 1, 2, 3 = $15,000 to $30,000, 4, 5, 6 = more than 90,000) were also collected. We recognize that dichotomizing race is not ideal and that more diverse representation is needed to build culturally relevant models (e.g., NIMH, 2023; Sheftall & Miller, Reference Sheftall and Miller2021). For this study, however, breaking race down further would result in cell sizes as small as n = 2, violating assumptions of inferential statistics, and creating circumstances in which findings could be driven by single “outliers”. Such findings hold the potential to reify and generalize spurious outcomes, with potential for harm. We therefore see our study as a first step toward diversifying a research area that to date has focused largely on White girls.
Nonsuicidal self-injury, suicide ideation, and suicide attempts
At baseline, lifetime history for NSSI, SI, and SAs were quantified using the Columbia Suicide Severity Rating Scale (C-SSRS; Posner et al., Reference Posner, Brent, Lucas, Gould, Stanley, Brown, Fisher, Zelazny, Burke, Oquendo and Mann2008), a semi-structured interview designed for this purpose. Ideation and attempts were determined from adolescent- or parent-reports on the CSSRS: In instances where adolescents did not want to or chose not to answer questions about their histories of NSSI or SAs, caregiver-reports were used. Lifetime histories of NSSI and SAs were quantified as 1 = yes or 0 = no. At follow-ups, NSSI was defined at each timepoint as self-injury without suicidal intent at baseline (Time 1) or self-injury since the last contact with the research team (Times 2 – 4). SAs were determined by a team of trained researchers based on complete interview data. Thus, all outcomes were binary, as dictated by available data.
Emotion lability/dysregulation
Adolescent ED was indexed at baseline, One-year follow-up, and Two-year follow-up using parent-report means from the angry/depressed subscale of the Affective Lability Scale for Children, ages 6 – 17 (Gerson et al., Reference Gerson, Gerring, Freund, Joshi, Capozzoli, Brady and Denckla1996). This 13-item subscale (α = 0.91) measures anger and sadness dysregulation such as “Suddenly loses temper,” and “Suddenly starts to cry”, respectively, on 5-point Likert scales (0 = never/rarely, 1 = l – 3 times/month, 2 = occurs l – 3 times/week, 3 = occurs 4 – 6 times/week, and 4 = occurs 1 or more times/day). The mean ED was higher at baseline (M = 1.33, SD = 0.83) compared to One-year (M = 0.99, SD = 0.70) and Two-year follow-up (M = 0.94, SD = 0.67).
Stressful life events
Stress was indexed at baseline, One-year follow-up, and Two-year follow-up by mean self-report scores on the Stressful Life Events Schedule (SLES; Pan et al., Reference Pan, Goldstein, Rooks, Hickey, Fan, Merranko, Monk, Diler, Sakolsky, Hafeman, Iyengar, Goldstein, Kupfer, Axelson, Brent and Birmaher2017; Williamson et al., Reference Williamson, Birmaher, Ryan, Shiffrin, Lusky, Protopapa, Dahl and Brent2003), a semi-structured interview modification of the Life-events Checklist and Difficulties Schedule. The SLES quantifies 80 potential stressors as present versus absent in the past year. To reduce the interview burden, ratings of objective impact and behavioral dependence (whether the stressor is dependent or independent of respondent behavior) were omitted. The SLES demonstrates strong internal consistency for self-reported events (ICC = .93). Example questions include, “I had trouble with grades or school” and “My parents divorced or separated.” Among the 80 items are three female-specific (e.g., “I had an abortion”) and two male-specific events (e.g., “My girlfriend was pregnant”).
Trait impulsivity
Impulsivity was quantified at baseline using hyperactive-impulsive symptoms of ADHD from the Mini International Neuropsychiatric Interview for Children and Adolescents (MINI-KID; Sheehan et al., Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavas, Weiller, Hergueta, Baker and Dunbar1998). Suicide risk and other untoward behavioral outcomes are concentrated specifically among those with hyperactive-impulsive and combined ADHD presentations who also experience maltreatment (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Hinshaw et al., Reference Hinshaw, Owens, Zalecki, Huggins, Montenegro-Nevado, Schrodek and Swanson2012). The presence of hyperactive-impulsive and combined ADHD presentations was determined based on an overall number of symptoms endorsed (≥ 6) during the interview, whether these symptoms and problems emerged in childhood, and clinician judgment.
Family history
Histories of SAs and suicide in family pedigrees – particularly among parents – confer considerable risk to offspring (Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019). Family histories of SAs and suicide were therefore included, as indexed by guardian reports (usually parents) of children’s consanguineous parents and sibling histories (1 = history of SB present, 0 = no). We used a series of questions adapted from the Family History Screen (Weissman et al., Reference Weissman, Wickramaratne, Adams, Wolk, Verdeli and Olfson2000).
Medication
Medication use was evaluated using the Services Assessment for Children and Adolescents (SACA; Stiffman et al., Reference Stiffman, Horwitz, Hoagwood, Compton, Cottler, Bean and Weisz2000). We determined whether prescription medications were taken for behavioral or psychiatric purposes at baseline (1 = yes, 0 = no).
Analyses
Changes in NSSI, SI, and SBs over the two years were assessed by constructing two-level multilevel models (MLMs) in HLM 6.06 (Raudenbush et al., Reference Raudenbush, Bryk, Cheong and Congdon2019). All models were specified using Bernoulli distributions given dichotomous outcomes (Landerman et al, Reference Landerman, Mustillo and Land2011). Repeated observations were nested within participants by age at each assessment at Level 1.
First, we modeled Level 2 effects of sex, family history, medication status, race, and income on Level 1 trajectories in NSSI, SI, and SAs. Missing values were imputed across 10 data sets in SPSS 22 following current standards (see Hayes & Enders, Reference Hayes, Enders, Cooper, Coutanche, McMullen, Panter, Rindskopf and Sher2023). Multiple imputation is more accurate than both listwise deletion and mean substitution of missing data and reduces selection biases in dependent measures (Acock 2005). It is effective with considerably more missing data than reported herein (see Graham, Reference Graham2009).
Next, for each Level 1 outcome (NSSI, SI, and SAs at baseline, 6 months, 1 year, and 2 years), we added measures of ED, SLES, and an abbreviated measure of the CSSRS asking questions only about NSSI, SBs, and updated contact information. For CSSRS variables, Level 1 slopes and intercepts were computed from all three waves of available data instead of 4.
In Bernoulli models, outcomes are expressed as changes in log odds at each timepoint. Restricted maximum likelihood models followed the form of this general example, which includes intercepts and slopes in SAs over two years as Level 1 outcomes, and family history of SAs as a Level 2 predictor of Level 1 slopes on SAs (log odds) over time:
Level 1:
p(SA since last assessment = 1/π) = φ
log[φ/(1–φ)] = η
η = π 0 + π 1(age at assessment)
Level 2:
π 0 = β 00 + r 0
π 1 = β 10 + β 10(family history of SAs) + r 1,
We used a parallel model-building approach for NSSI, SI, and SAs as outcomesFootnote 2 . First, we simultaneously modeled Level 2 fixed effects of biological sex at birth, family history of SAs, medication status, race, income, and history of NSSI (for SI and SAs as outcomes). Collectively, these influences, all of which show consistent associations with NSSI and SBs, exert sustained effects on psychological adjustment. Second, we dropped non-significant fixed effects from above and added Level 1 time-varying effects of (a) environmental stress; (b) NSSI (for SI and SAs only); and (c); impulsivity, emotional lability, and the impulsivity × ED interaction in predicting NSSI, SI, and SAs. Although impulsivity is a trait with reasonable stability, modeling it as a time-varying covariate was based on three considerations: (1) slopes of any measure are considerably more reliable than scores at a single time point and therefore improve measurement precision; (2) raw impulsivity scores normatively fall across this age range, even though individual differences are preserved–ignoring this reduces measurement precision as well; and (3), testing changes in the Impulsivity × ED interaction over time requires computing the interaction at every time point. A constant main effect of impulsivity does not preclude strengthening interaction effects over time, which follow from theory (see above).
Importantly, fixed effects entered at Step 1 occurred or were established before the two-year observation period. They are therefore more distal, with developmental implications weighted more heavily toward primary prevention. Time-varying (slope) effects at Step 2 were measured during the observation period. They are therefore proximal and weighted more heavily toward secondary prevention. Within steps, all effects were entered simultaneously, which tests their contribution to outcomes over and above all other effects in the model. Our interest in testing the Impulsivity × ED interaction effect requires that the main effects of Impulsivity and ED be entered and that the Impulsivity × ED interaction account for variance over and above those main effects. In other words, when testing for hypothesized interactions, main effects are entered as necessary statistical controls and are not interpreted (see Aiken & West, Reference Aiken and West1991; Pedhazur, Reference Pedhazur1997).
For all analyses, we report unit-specific effects, which allow intercepts to vary across individuals. With binary outcomes, unit-specific effects are better suited to individual-level prediction – our primary objective – than to estimating population parameters (Hu et al., Reference Hu, Goldberg, Hedeker, Flay and Pentz1998).
Results
Sample descriptives are presented in Table 1 and sample sizes and frequencies of NSSI, SI, and SAs at each timepoint are presented in Table 2. The sample was primarily non-Hispanic, White (61%), and female (69%) (N = 344, M = 13.8 years, SD = 1.0). The mean score for SLES across the study period was 0.18 (SD = 0.12) and the mean score for ED across the study period was 1.00 (SD = 0.82). From the original sample, 302 adolescents completed at least one follow-up assessment; 46 of these adolescents (15.2%) had at least one SA at some point during the follow-up period. There were 16 adolescents who attempted suicide at both baseline and 6-month follow-up, 0 who attempted at both Six-month and One-year follow-up, and 1 who attempted at both One- and Two-year follow-ups. There were 73 who reported NSSI at both baseline and Six-month follow-up, 37 who reported NSSI at both Six-month and One-year follow-up, and 35 who reported NSSI at both One- and Two-year follow-ups. As expected, squared correlations (percentage of variance accounted for) between impulsivity and ED were small but significant at Baseline, r 2 = .05, p < .01, One-year follow-up, r 2 = .04, p<01, and Two-year follow-up, r 2 = .02, p < .01. This is consistent with findings specifying impulsivity and ED as independent in the population (see above).
Sample characteristics at baseline
Suicidal thoughts and behaviors
Note. Frequencies refer to those who affirmed attempt, ideation, or self-injury.
Distal predictors – Level 2 Fixed effects
Distal effects of biological sex at birth, family history of SAs, medication status, race, income, and history of self-injury (for SI and SAs only) appear in the top half of Table 3. Significant intercepts indicate non-zero scores for NSSI, SI, and SA at Baseline (i.e., study entry or Time 0). Significant slopes indicate longitudinal changes in log-likelihoods of each outcome over two years. Negative slopes indicate falling probabilities of NSSI, SI, and SAs over the two years (see below).
Multilevel models including fixed effects and time-varying predictors of NSSI, SI, and SAs across two years
Note. Fixed effects entered at Step 1 occurred or were established before the 2-year observation period. They are therefore more distal, with implications weighted toward primary prevention. Time-varying effects at Step 2 were measured during the observation period. They are therefore proximal, with implications weighted toward secondary prevention. Within steps, all effects were entered simultaneously. At Step 2, main effects of impulsivity and emotional lability (grayed text) were entered as statistical controls when testing the Impulsivity × ED interaction effect. To be interpreted, interaction effects must be significant over-and- above main effects, which should not be interpreted (e.g., Pedhazur, Reference Pedhazur1997).
a We did not model intercepts for proximal effects at Step 2, as they were not of central interest. This enabled us to preserve degrees of freedom for analyses that were already complex.
Although we do not list all significant effects in Table 3 here, several are noteworthy. As in the population, girls showed slightly more NSSI at baseline than boys (b = −1.37, p < .10) and more SI at baseline than boys (b = −3.92, p < .001), but steeper declines than boys over time (both bs > 0.19, both ps < .001; see Figure 1). No sex difference in SAs was observed. Medication status was associated with more NSSI at intake and less improvement (both bs > |0.26|, both ps < .001). Although race was unassociated with SI, White youth reported more NSSI and SAs at baseline (both bs>3.00, both ps < .001) and steeper improvement over time (both bs < |0.22|, both ps < .001).
Family history was unrelated to NSSI but was associated with more SI and SAs at baseline (both bs>1.94, both ps < .03), and less improvement over time (both bs>3.00, both ps < .01). Income was associated with more NSSI at intake (b = 0.68, p < .001), but less SI and fewer SAs (both bs < −1.94, both ps < .05). Income predicted more improvement in NSSI and SI (both bs<0.04, both ps < .001), but less improvement in SAs (b = 0.03, p = .03).
Finally, our first primary hypothesis, that previous NSSI would predict both SI and SAs, was confirmed. In both cases, previous NSSI predicted higher baseline scores (both bs > 2.83, both ps < .001) and steeper improvement (both bs > −0.12, both ps < .03). Of note, all reported effects, both above and below, are from simultaneous regression models. Significance values are therefore over and above all other effects.
Proximal associations – Level 1 Time-varying effects
Proximal (time-varying) effects of impulsivity, emotional lability, and stress on NSSI, SI, and SAs are reported in the bottom half of Table 3. Significant overall model intercepts indicate NSSI, SI, and SAs scores at baseline that were greater than 0 (all bs > 2.91, all ps < .001). In all cases, significant within-person age effects indicated odds of NSSI, SI, and SAs declined over two years (all bs < 0.32, all ps < .001; see Figure 2).
Probabilities of self-injury, suicide ideation, and suicide attempts by age (left three panels) and probabilities of self-injury, suicide ideation, and suicide attempts as a function of the Impulsivity x Emotion Dysregulation interaction (right three panels). Positive slopes indicate higher risk (stronger associations) when impulsivity and emotion dysregulation co-occur.
Major hypotheses regarding the proximal effects of stress, NSSI, and the Impulsivity × ED interaction were all confirmed. Stress slopes covaried positively with NSSI, SI, and SA slopes (all bs > 2.91, all ps < .001); NSSI slopes covaried positively with both SI and SA slopes (both bs>0.06, both ps < .001), and the Impulsivity × ED interaction was significant for NSSI, SI, and SAs (all bs > 0.29, all ps < .003; see Figure 2). The latter interaction was significant over and above the main effects of impulsivity and ED, which are therefore not interpreted (Aiken & West, Reference Aiken and West1991; Pedhazur, Reference Pedhazur1997). Once again, all effects are independent and therefore significant over and above all other effects in the models.
Discussion
In the present study, we tested distal and proximal predictors of NSSI, SI, and SAs among a high-risk sample of adolescents with a known history of depression over the course of two years. Our primary aim was to extend the Impulsivity × ED hypothesis, advancing our understanding of adolescents’ risk for SAs and the association between changes in SAs and stressful life events. Our results add additional support for the Impulsivity × ED hypothesis. In the present study, we found that adolescents with high ED and high impulsivity were more prone toward SAs, SI, and NSSI, and had worse trajectories over the two years. As expected, stressful life events were also associated with poorer trajectories for all outcomes, and NSSI was associated with future and concurrent SI and SAs. These findings extend our understanding of one pathway to SAs to a more diverse sample of adolescents, with potential implications for prevention.
In support of our first hypothesis, we found that Impulsivity × ED conferred greater risk for NSSI, SAs, and SI, at the beginning of the study and worse, or riskier, trajectories over the two years. Our results add to the growing concern that adolescents with high ED and high impulsivity are at increased risk for SBs (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Crowell et al., Reference Crowell, Beauchaine and Linehan2009). Prior research supports NSSI often functions to regulate negative emotions in adolescents, but many more adolescents self-injure than attempt suicide and less attention has been given to why some adolescents who self-injure also attempt suicide. It is widely accepted that SBs are not constrained to a singular pathology or trait, and our primary aim was not to identify a singular etiology but rather to further understand one pathway through which adolescents engage in SAs, impulsivity × ED. These results suggest that NSSI perpetuates the risk for SAs. NSSI and SBs likely represent the outward behavioral expression of ED, and trait impulsivity could enhance tendencies toward self-harming behaviors for highly dysregulated adolescents (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Crowell et al., Reference Crowell, Beauchaine and Linehan2009). Arguably, most if not all adolescents in the participating study were highly dysregulated given their history of clinically significant depression, still, only 15% of the sample attempted suicide during the follow-up. Notably, while SI, SA, and NSSI declined over the course of the study for the sample overall, impulsivity × ED explained the increased risk for SI, SA, and NSSI. Thus, our results suggest this is likely a potential pathway for SA in adolescents that warrants further attention.
We also found support for our second hypothesis, with NSSI signaling both distal and proximal risk for SAs and SI. This finding suggests that lifetime history as well as persistence of NSSI increasingly confers risk for SA and SI. These findings are expected given NSSI frequency and severity are associated with risk for SI and SAs (O’Loughlin et al., Reference O’Loughlin, Burke and Ammerman2021; Stewart et al., Reference Stewart, Esposito, Glenn, Gilman, Bryan, Gold and Auerbach2017a), and as found in previous research, we found NSSI prevalence and persistence were more prevalent in girls (Beauchaine et al., Reference Beauchaine, Bell, Knapton, McDonough‐Caplan, Shader and Zisner2019; Crowell et al., Reference Crowell, Beauchaine, McCauley, Smith, Stevens and Sylvers2005).
In support of our third hypothesis, we found stressful life events covaried with NSSI, SI, and SA over the course of two years. Our findings extend previous research suggesting that stressful life events are concurrently and prospectively associated with SAs and SI (Howarth, O’Connor, Panagioti, et al., 2020). Presently, research on the effect of stressful life events and the transition from NSSI to SBs is limited (Taliaferro et al., Reference Taliaferro, Almeida, Aguinaldo and McManama O’Brien2019; Vergara et al., Reference Vergara, Jobes and Brausch2023). Although our model cannot tease apart temporal precedence or bidirectional effects, our results suggest that greater stressful life events and persistence of NSSI increase the risk for SI and SAs. Notably, the Impulsivity × ED effect remained significant for SI and SAs while controlling for changes in NSSI history and stressful life events. These results provide further insight into mechanistic pathways leading to SA and SI. This finding is especially critical in advancing our understanding of the multiple etiological pathways to SA, suggesting that adolescents are generally at risk for SAs and SI in times of high stress, but adolescents with high Impulsivity and ED may be at heightened risk following stressful life events.
Strengths and limitations
By including a more diverse sample, we add to our understanding of NSSI in boys and girls and our understanding of SI and SA among racial and ethnic minority samples. However, our sample is small, and the findings should be replicated. White, non-Hispanic adolescents had a higher prevalence of NSSI at intake but were more likely to discontinue NSSI over the follow-up compared to peers from other racial and ethnic groups. NSSI in non-White and more racially diverse populations is drastically understudied, and our findings suggest that while NSSI was less prevalent, adolescents from other racial and ethnic groups did not experience the same improvements as their White peers. Given the recent concern for elevated rates of suicide in children from minoritized groups (Bridge et al., Reference Bridge, Horowitz, Fontanella, Sheftall, Greenhouse, Kelleher and Campo2018; Sheftall et al., Reference Sheftall, Vakil, Ruch, Boyd, Lindsey and Bridge2022), understanding the identification and treatment of NSSI in children and adolescents from different racial and ethnic groups may advance the identification of youth at-risk for suicide. Further, we find that a family history of suicide was unrelated to NSSI while being significantly associated with SA and SI. To our knowledge, little is known about inherited risk for NSSI and suicide within the same sample, thus, this specific finding warrants replication and further investigation.
The severity and lethality of NSSI vary between individuals and vary within individuals over time and may importantly be related to escalating risk for SB (Kaess et al., Reference Kaess, Hooley, Klimes-Dougan, Koenig, Plener, Reichl, Robinson, Schmahl, Sicorello, Westlund Schreiner and Cullen2021). In the present study, we could only use persistence or reoccurring episodes of NSSI over time using dichotomous variables, which limits our ability to measure fluctuations in the frequency and severity of NSSI. We were also constrained by the dichotomized variable of impulsivity measured at baseline using the KSADS, limiting our ability to detect greater variability across the sample. The SLES was also abbreviated to alleviate participant burden, and thus we were unable to include additional ratings of how strongly affected by stressors participants felt or how much control participants felt they had over the stressor in the analyses. We also found large racial differences in the prevalence of NSSI but were limited by sample size and could not explore these further. While our sample included a diverse range of clinical and psychiatric problems, an important next step is to test these hypotheses in a more racially and ethnically diverse sample of participants.
Implications
Findings from the present study add to a growing body of research advocating for intervention and prevention efforts that target mechanisms of risk for NSSI and SAs in adolescents. Specifically, there is sufficient evidence to support proactive intervention efforts to prevent NSSI and SAs by targeting ED in adolescents with impulsive behaviors (Asarnow & Miranda, Reference Asarnow and Miranda2014). Emotion regulation is socialized early in development through parents’ emotion socialization practices, peer interactions, and other modifiable mechanisms. Prior interventions have demonstrated efficacy in promoting more adaptive emotion regulation by targeting these mechanisms and may help mitigate the risk of NSSI and SAs. We argue there are identifiable groups at sufficient risk that would benefit from primary prevention, including maltreated adolescent girls with ADHD, or youth more broadly with internalizing disorders and impulsive tendencies. While emotion regulation is socialized, impulsivity is highly heritable and generally stable. Thus, successful interventions should specifically aim to support adolescents’ emotion regulation while also anticipating and supporting the needs of impulsive adolescents.
Moreover, NSSI and SBs are often preceded by stressful events, these results support the need to better support adolescents’ emotional and behavioral coping and perhaps invest in interventions that are sensitive to the developmentally variable stressful life events common to adolescent populations. One challenge for researchers creating interventions that prevent adolescent self-harm is teaching skills to identify situations when they will be most at risk for self-harm (Hasking, Whitlock, Voon, et al., 2017). However, research has found common stressful events may precipitate self-harm. For example, precipitating stressors of adolescent suicide often include difficulties with a romantic peer (Sheftall et al., Reference Sheftall, Asti, Horowitz, Felts, Fontanella, Campo and Bridge2016). Other common precipitating events of NSSI in adolescence include conflict with peers or parents (Kaess et al., Reference Kaess, Hooley, Klimes-Dougan, Koenig, Plener, Reichl, Robinson, Schmahl, Sicorello, Westlund Schreiner and Cullen2021). Interventions that specifically target populations of at-risk adolescents with interventions that foster adaptive coping skills for use during highly salient stressful events warrant further attention. Our research supports more targeted primary prevention efforts that target malleable mechanisms like ED to lessen the social, economic, and emotional burden of NSSI and SAs in adolescents.
Funding statement
Preparation of this article was supported by Grants MH125905 and MH127476 from the National Institutes of Health. The authors have no conflicts of interest to disclose.
Competing interests
The authors have no conflicts of interest to declare.
Open access
