A Prospective Examination of the Interpersonal-Psychological Theory of Suicidal Behavior in Adolescents

ABSTRACT

Objective

Given the large and complex array of suicide risk factors, theoretical frameworks are critical to furthering our understanding of risk. This study prospectively examined several key constructs of the interpersonal-psychological theory of suicidal behavior (IPTS) in a large, geographically diverse sample of U.S. adolescents.

Method

Conducted in collaboration with the Pediatric Emergency Care Applied Research Network, adolescents, ages 12 to 17, were recruited from emergency departments. Baseline and 6-month follow-up samples were comprised of 6,448 (59% female sex) and 2,009 (64% female sex) adolescents, with self-identified race/ethnicity as follows (baseline/follow-up): White (52%/54%), Black (22%/23%), Multiracial (6%/6%), American Indian (3%/3%), other/unknown race (15%/14%), and Latinx (25%/23%). Youth and parents completed adolescent suicide risk surveys at baseline and 6-month follow-up (retention, 69%). Latent class analysis was used to identify classes of painful and provocative events (PPE), considered a precursor to acquired capability.

Results

In keeping with IPTS tenets, thwarted belongingness (TB), perceived burdensomeness (PB), and the interaction between TB and PB were each significant predictors of suicidal ideation at baseline and follow-up. However, only PB and PPE were significant predictors of cross-sectional suicide attempts and only TB and PPE were significant predictors of prospective suicide attempts in models that adjusted for baseline suicidal ideation. The three-way interaction among PB, TB and PPE was nonsignificant.

Conclusions

Results from this large-scale prospective study suggest the importance of TB, PB, and PPE to our understanding of suicidal thoughts and suicide attempts among adolescents, pointing to promising prevention and intervention targets.

Suicide is the second leading cause of death among adolescents in the U.S (Centers for Disease Control and Prevention, 2020), and nearly 9% of U.S. high school students report having attempted suicide within the past 12 months (Ivey-Stephenson et al., 2020). As such, adolescent suicide risk is an urgent public health concern. It is imperative that we improve our understanding of adolescent suicide risk and of how we identify and intervene with youth who are at risk. While decades of research have made valuable contributions to our knowledge of suicide risk and protective factors (see reviews; Bridge et al., 2006; Cha et al., 2018; Spirito & Esposito-Smythers, 2006), only a small number of studies have examined the multi-variable prediction of suicidal behavior (e.g., C. A. King et al., 2019), and single risk factors are known to be poor predictors of such behavior (Franklin et al., 2017; J. Ribeiro et al., 2016). Given the complexity and heterogeneity of suicide risk factors, further advances in knowledge to guide our understanding of suicide risk may benefit from a grounding in rigorously validated theoretical frameworks of suicidal behavior.

Interpersonal-Psychological Theory of Suicide (IPTS)

The interpersonal-psychological theory of suicidal behavior (IPTS; Joiner, 2005; Van Orden et al., 2010) has made important theoretical contributions to our understanding of the complex phenomena of suicidal ideation and behavior (suicide and suicide attempts). IPTS proposes that suicidal desire (i.e., suicidal ideation) is produced by the simultaneous presence of thwarted belongingness (TB) and perceived burdensomeness (PB). The theory further proposes that more active suicidal ideation develops when individuals experience hopelessness regarding the possibility of improvement in their experienced TB and PB. Suicidal behavior, in turn, is hypothesized to occur in the presence of both suicidal desire and acquired capability to enact lethal self-injury. TB involves a perception that one is alienated from others while PB is the perception that one is a burden on others. Acquired capability (AC), proposed to account for the transition from suicidal desire to action, is conceptualized as increased physical pain tolerance and fearlessness about death, acquired through exposure to painful and provocative events or experiences (PPE). IPTS proposes that risk for suicidal behavior occurs when an individual experiences TB and PB and has a history of PPE (e.g., physical abuse, nonsuicidal self-injury, previous suicidal behavior) that has resulted in AC (Van Orden et al., 2010). This hypothesized proximal pathway to suicidal action points to potentially valuable prevention, assessment, and intervention targets that may be particularly useful in streamlining efforts to prevent youth suicide. However, more evidence regarding the predictive validity of IPTS in identifying suicide risk among adolescents is needed.

Previous Studies of IPTS and Adolescent Suicide Risk

In the more than 15 years since its original publication (Joiner, 2005), IPTS has received considerable attention and been the focus of research involving college, military, community, and clinical populations (Chu et al., 2017; Ma et al., 2016). A growing number of studies also have focused on the application of IPTS to adolescents. Stewart and colleagues’ meta-analysis of 15 adolescent-focused studies (Stewart et al., 2017) identified indicators of acquired capability (AC) as an important predictor of suicide attempts, yet also reported mixed findings regarding the roles of perceived burdensomeness (PB) and thwarted belongingness (TB). They recommended further research to assess the IPTS with diverse adolescents using developmentally appropriate scales that include non-suicidal self-injury (NSSI). In a large prospective study examining IPTS constructs in relation to suicidal ideation among high school students, Barzilay et al. (2019) obtained only limited support for the theory. They found that parental low belongingness, but not peer belongingness or perceived burdensomeness, predicted greater suicidal ideation across a 12-month period. Similarly, Hains et al. (2019) study of adolescents receiving treatment for suicide risk yielded mixed findings, with reductions in PB, but not TB, associated with reductions in suicide risk. In a community-based study of Australian adolescents, the interaction between PB and TB was found to be significantly associated with the severity of suicidal ideation, but neither PB nor TB were significantly associated with suicide attempt (Calear et al., 2021). Similarly, two studies of psychiatrically hospitalized adolescents found no prospective association between the three-way interaction of IPTS constructs and suicide attempts (Czyz et al., 2015; C. D. King et al., 2019). These previous studies suggest that TB and PB may each be important to our understanding of suicidal ideation; and TB, PB, and AC may each be important to our understanding of suicide attempts; however, the combinations purported by the theory may not be required for an adolescent to experience suicidal desire or to engage in suicidal behavior.

The extent to which IPTS is valid and useful to our understanding of adolescent suicide risk is unclear due to the research designs and measurement strategies of previous research. First, the majority of this research focused on individual IPTS constructs, i.e. main effects, with considerably fewer studies testing the theory-specific combinations of constructs (Ma et al., 2016). More specifically, few investigations have examined the effect of the full IPTS model (i.e., the simultaneous presence of TB, PB, and PPE or AC) on suicidal behavior – a key outcome the theory is concerned with predicting. Among the handful of studies investigating the interaction between all three components with adults, nearly all examined retrospectively assessed association with previous suicide attempts (Anestis et al., 2011; Joiner, 2005; Monteith et al., 2013) or with a composite variable that included previous attempts together with other related outcomes (e.g. suicidal ideation, likelihood of future attempts; Bryan et al., 2010). An exception is a recent study of adult inpatients, which indicated that the combination of IPTS constructs was not associated with suicide attempts 12 months after discharge (Forkmann et al., 2020). To our knowledge, only two adolescent studies (Czyz et al., 2015; C. D. King et al., 2019), described above, examined indicators of all three IPTS components in a single model. The limited number of prospective IPTS studies has precluded a more stringent test of the theory as a predictive framework.

The gap in our understanding of IPTS in relation to adolescent suicide risk is also due to the limited measurement of PPE and AC, key IPTS constructs, in previous studies. In the Czyz et al. (2015) study, PPE, referred to as AC in the study, was measured solely by the presence or absence of previous multiple suicide attempts, which had previously been established as a risk factor for suicidal behavior (Goldston et al., 1998; Gould et al., 2003). Similarly, in the C. D. King et al. (2019) study, AC was defined by lifetime NSSI, one type of PPE. This narrow measurement strategy did not capture the broader range of painful or fear-provoking experiences that the IPTS conceptualizes as being important (Stewart et al., 2017).

Purpose of the Present Study

In this study, we sought to address these gaps by examining cross-sectional and longitudinal associations between thwarted belongingness (TB), perceived burdensomeness (PB), painful and provocative events (PPE, a key theorized precursor to acquired capability), and suicide attempts in a large sample of adolescents recruited from 14 emergency departments (EDs) in the United States. Although TB and PB have been found to vary over time (Zullo et al., 2021) and associations between these constructs and suicide-related outcomes may be particularly strong when measured close in time, there may also be a more chronic or stable component to these interpersonal perceptions. We used latent class profile analysis to develop a measure of PPE based on multiple indices, with the goal of identifying profiles of PPE among adolescents that may be characterized by different types or combinations of PPE and/or by different severity levels. Consistent with IPTS, we hypothesized that the simultaneous presence of TB, PB, and PPE would contribute to the baseline cross-sectional and the 6-month prospective prediction of suicide attempts. We also hypothesized that the combination of TB and PB would be a stronger predictor of suicidal ideation at these time points than either single construct.

Method

Sample

Participants were recruited from the Emergency Department Screen for Teens at Risk for Suicide (ED-STARS) Study One sample (C. A. King et al., 2019). Adolescents, ages 12 to 17, with a parent/legal guardian present, were recruited from 13 emergency departments within the Pediatric Emergency Care Applied Research Network (PECARN) and one Indian Health Service hospital. Exclusion criteria were: 1) ward of the State, 2) prior study enrollment, 3) non-English speaking, 4) medically unstable, and 5) severe cognitive impairment. Among the 10,664 eligible patients approached for participation, 6,641 were enrolled (62.3%) and 6,448 adolescents completed at least 80% of the baseline survey (60.5%). This baseline sample was 41% male (n = 2,640), with a mean age of 15.0 (SD = 1.64), and a racial composition as follows: American Indian or Alaska Native (3%), Asian, Native Hawaiian, Pacific Islander (2%), Black or African American (22%), White (52%), Multiracial (6%), and Unknown (15%). Twenty-five percent of youth self-identified as Hispanic or Latino. Levels of education for mothers and fathers, respectively, were: high school graduate or less (32%; 41%), some college/technical training (27%; 20%), college graduate/professional training (39%; 29%), and don’t know/not applicable (3%; 9%). Approximately two-fifths of the sample (43%) reported currently receiving public assistance.

The follow-up sample was comprised of 2,009 youth who completed at least 80% of the baseline survey (n = 6,448 of 6,641, 97.1%), were randomized to follow-up (n = 2,897 of 6,448, 44.9%) and for whom 6-month follow-up data are available (n = 2,009 of 2,897, 69.3%). The algorithm for randomization, which incorporated enrichment for suicide risk, was described previously (C. A. King et al., 2019). Based on adolescents’ baseline survey responses, we randomized to follow-up a higher percentage of “high risk” adolescents (83.6%; defined as one of: suicidal ideation with intent/plan, history of suicide attempt, NSSI 5 or more times in past year, homicidal ideation with intent/plan), than “moderate risk” (68.0%) and “low risk” (16.3%) adolescents. Written youth assent and parental/guardian consent were obtained from all participants. Youth and a parent/guardian completed a baseline self-report survey assessing the youth’s risk for suicide in the ED. Youth and/or a parent/guardian also completed a telephone interview at 6 months. Youth were remunerated with a $15 gift card. Study procedures were approved by the Institutional Review Board at all sites (IRB approval # at lead site: HUM00097200).

Measures

Demographics

Youth and parent/guardian demographic information was collected in the baseline self-report survey.

Suicidal Ideation, Lifetime Suicide Attempts

The Columbia-Suicide Severity Rating Scale (C-SSRS) was used to assess severity of suicidal ideation and suicide attempts (Posner et al., 2011) at baseline and follow-up. In addition, the four-item Ask Suicide Questionnaire (ASQ) was administered at baseline to assess suicidal ideation and lifetime history of suicide attempts (Horowitz et al., 2012).

At baseline, suicidal ideation was indicated by past month C-SSRS Suicidal Ideation (SI) Severity Scale scores. For baseline cross-sectional analyses, suicidal ideation was indicated by a score ≥2 (scored on a 5-point severity scale) or endorsement of the ASQ item “In the past week, have you been having thoughts about killing yourself?” At follow-up, suicidal ideation was indicated by a SI Severity Scale score ≥2 for the past 3 months (or since last assessment).

At baseline, two items from an adapted C-SSRS Suicide Behavior scale were used to assess lifetime history of suicide attempts, which was coded as Yes or No: Have you ever in your life made a suicide attempt? Have you ever in your life tried to harm yourself because you were at least partly trying to end your life? At follow-up, whether the adolescents had engaged in one or more suicide attempt since baseline was coded as Yes or No. This was assessed in multiple ways, including adolescent self-report: “Since the last assessment, have you made a suicide attempt?” or “Since the last assessment, have you tried to harm yourself because you were at least partly trying to end your life?” These data were available at the 6-month follow-up for 2,009 participants (69.3%). The presence of a suicide attempt outcome was also assessed by adolescent or parent report that the adolescent visited an ED or was hospitalized because of a suicide attempt, or by a record of suicide death collected through follow-up or medical chart review.

Depression

Depressive symptoms were assessed with the 9-item Patient Health Questionnaire (PHQ-9) (Kroenke et al., 2001). A sample item includes: “Over the last 2 weeks, how often have you been bothered by … feeling down, depressed, or hopeless?” Scores on the PHQ-9 range from 0 to 27, with higher scores reflecting a greater number and/or severity of depressive symptoms. The PHQ-9 has demonstrated good psychometric properties (Richardson et al., 2010).

IPTS Constructs

Each of the items and brief scales used to assess IPTS constructs has been validated prospectively (C. A. King et al., 2019).

Perceived burdensomeness (PB)

One item from the Ask Suicide-Screening Questions (ASQ) 4-item instrument (Horowitz et al., 2012) was used as a proxy for PB: “In the past few weeks, have you felt that you or your family would be better off if you were dead?” Responses options were “Yes,” “No,” and “No response.”

Thwarted belongingness (TB)

Six self-report items assessing family, school, and social connectedness were used as create a Connectedness Scale to assess TB at baseline. Family connectedness was assessed with two items adapted from the Parent-Family Connectedness Scale (Resnick, 1997): “How much do people in your family understand you?” “How much does your family pay attention to you?” Two items, adapted from the School Connectedness Scale, were used to measure youth sense of belonging in school and closeness to other classmates (Resnick, 1997): “You feel close to people at your school,” and “You feel like you are part of your school.” Similarly, two items assessing social connectedness were adapted from Hemingway’s Adolescent Connectedness Scale (Karcher & Sass, 2010): ‘I have friends I’m really close to and trust completely.” and “Spending time with my friends is a big part of my life.’ Responses to each item were on a 5-point Likert scale and coded as integer values 1 through 5. For youth with at least 5 of the 6 items completed, the mean of the completed items was used as a proxy for TB, with higher values representing more belongingness. Each of the adapted two-item scales has shown evidence of predictive validity for suicide attempt outcomes (C. A. King et al., 2019). The internal consistency of the 6-item Connectedness Scale in the present study sample was 0.79.

Painful and provocative events (PPE)

A categorical variable was developed to assess the painful and provocative events (PPEs) that are considered to be precursors to AC (see latent class analysis [LCA] details below). Lifetime suicide attempts were assessed with the C-SSRS Screen item “How many times in your life” following affirmative responses to: “Have you ever in your life made a suicide attempt?” or “Have you ever in your life tried to harm yourself because you were at least partly trying to end your life?” We categorized lifetime suicide attempts as zero, one, or multiple attempts. Non-suicidal self-injury (NSSI) frequency was assessed with an item that queried the number of incidents within the past 12 months (Centers for Disease Control and Prevention, 2016): possible responses were “0 times,” “1–2 times,” “3–4 times,” and “5 or more times.” One item from the Functional Assessment for Self-Mutilation Scale (FASM; Lloyd-Richardson et al., 2007) was used to assess the number of NSSI methods: “Over the last 12 months, which method(s) have you used to hurt yourself (select all that apply)?” Sample response items include “cutting or carving on skin,” “picking at a wound,” and “hitting self.” The number of NSSI methods was categorized as zero, one to four, or more than four methods. Physical aggression was measured with a dichotomous item adapted from the Impulsive/Premeditated Aggression Scale (Stanford et al., 1995): “During these times, have you … physically attacked someone?” Youth also responded to an item about past-year fighting frequency from the Youth Risk Behavior Scale (YRBS) (Centers for Disease Control and Prevention, 2016). Responses were categorized as 0, 1, 2, 3, or ≥4. Finally, youth completed the brief Two-Item Screener to assess childhood abuse (Thombs et al., 2007). Dichotomous items included “People in my family have hit me so hard that it left me with bruises or marks” and “Someone has tried to touch me in a sexual way or tried to make me touch them.” The past-year fighting and physical abuse items were not included in the final LCA model due to considerations described in the LCA details below.

Design

This study’s prospective design included 6-month follow-up assessments conducted by staff blind to baseline evaluations. Interviews were obtained from adolescent and parent (n = 1,666, 82.9%), adolescent only (n = 257, 12.8%), or parent only (n = 86, 4.3%). Table 1 presents demographic data for adolescent samples at baseline and follow-up. Non-white adolescents, Latinx adolescents, and those with lower parent education were less likely to be retained. None of these variables were a predictor of suicide attempt.

Table 1. Demographics.

	Baseline^a (N = 6448)	Follow-up^a (N = 2009)
Age at enrollment (years): Mean (SD)	15.0 (1.64)	15.1 (1.60)
Gender (male)	2640 (41%)	731 (36%)
Race
American Indian or Alaska Native	194 (3%)	66 (3%)
Asian, Native Hawaiian, Pacific Islander	125 (2%)	32 (2%)
Black or African American	1444 (22%)	454 (23%)
White	3335 (52%)	1082 (54%)
Multi-racial	355 (6%)	116 (6%)
Unknown or unavailable	995 (15%)	259 (13%)
Ethnicity
Hispanic or Latino	1506 (25%)	439 (23%)
Not Hispanic or Latino	4084 (68%)	1341 (71%)
Unknown	406 (7%)	115 (6%)
Education – mother/stepmother
High school graduate or less	1964 (32%)	540 (28%)
Some college/technical training	1635 (27%)	575 (30%)
College graduate/professional training	2381 (39%)	770 (40%)
Don’t know/Not applicable	156 (3%)	51 (3%)
Education – father/stepfather
High school graduate or less	2507 (41%)	751 (39%)
Some college/technical training	1230 (20%)	404 (21%)
College graduate/professional training	1802 (29%)	593 (31%)
Don’t know/Not applicable	571 (9%)	180 (9%)
Family receives public assistance	2615 (43%)	858 (45%)

^aCategories are not mutually exclusive.

Latent class analysis (LCA) was used to identify classes of adolescents who varied in painful and provocative events (PPE). We chose variables for the LCA a priori, based on IPTS theory and the types of PPEs reported by adolescents in the parent study. Previous attempt and NSSI variables were first included in the LCA model because they are PPEs involving self-harm. Next, we added other candidate variables to the model and assessed for fit and contribution, aiming to include at least one physical fighting and/or aggression variable and one physical and/or sexual abuse variable. LCA were conducted using Mplus version 7.4. Full information maximum likelihood was used to handle missing data. We considered solutions with three or four classes. We considered solutions with more than four classes to be unnecessarily complicated for meaningful translation to clinical practice. We evaluated the Bayesian Information Criterion (BIC) and entropy and set a minimum class size threshold of 5% of the sample. For analyses involving PPE, adolescents were placed in their most likely latent class membership based on posterior probabilities. In a secondary analysis, we conducted the LCA using only the three self-harm variables: suicide attempt, number of NSSI incidents, and number of NSSI methods.

Logistic regressions, controlling for baseline depression, were used to examine perceived burdensomeness (PB) and thwarted belongingness (TB) as cross-sectional predictors of suicidal ideation at baseline and longitudinal predictors of suicidal ideation at 6-month follow-up. For suicide attempt outcomes, we used logistic regression models and controlled for baseline suicidal ideation to test whether all three constructs together predict suicide attempt better than any single construct or any combination of two of the constructs. We used the Akaike information criterion (AIC) to compare models. Additionally, we used separate models to investigate interaction effects between PB and TB when predicting suicidal ideation as well as two- and three-way interactions between PB, TB, and painful and provocative events (PPE) when predicting suicide attempts. These interactions were investigated at baseline and follow-up. For all regression models, we calculated adjusted odds ratios. Adjusted odds ratios were calculated with and without additionally adjusting for depression at baseline in models predicting suicidal ideation and with and without additionally adjusting for suicidal ideation at baseline in models predicting suicidal attempts.

For analyses of associations with suicide attempts at baseline, we excluded the “outcome” suicide attempt in the derivation of PPE classes. If adolescents reported a suicide attempt in the past month at baseline, one suicide attempt was subtracted from the adolescent’s total number of past attempts and then the adolescent was assigned to the class matching the pattern of PPE variables. In this way, each adolescent’s PPE was estimated without consideration of the attempt during the past month. For the prospective analyses displayed in Table 4, only baseline SA information was used in the derivation of PPE classes. Summaries and analyses were performed using SAS version 9.4.

Results

Descriptive Data – Primary Study Variables

Suicidal Ideation, Suicide Attempt History, Non-Suicidal Self-Injury

Adolescents’ mean lifetime C-SSRS Suicidal Ideation Severity Score at baseline was 1.1 (SD = 1.77). Twenty-nine percent of adolescents (n = 1,882) had a lifetime C-SSRS Suicidal Ideation Severity Score ≥ 2, which indicates thoughts of killing oneself; and 22% of adolescents (n = 1,411) had a lifetime C-SSRS Suicidal Ideation Severity Score ≥3, which indicates suicidal ideation with intent, thoughts of a method, and/or thoughts of a plan. 15.7% of adolescents had a history of at least one suicide attempt (n = 1,010), and 11.5% of adolescents had a history of multiple suicide attempts (n = 738). As reported by adolescents and/or parents, 146 adolescents (7.3%) made at least one suicide attempt between baseline and 6-month follow-up. Adolescents’ mean C-SSRS Suicidal Ideation Severity Score at follow-up was 0.6 (SD = 1.26); 16% of adolescents (n = 305) had a C-SSRS Suicidal Ideation Severity Score ≥2 and 10% of adolescents (n = 195) had a C-SSRS Suicidal Ideation Severity Score ≥3.

Perceived Burdensomeness (PB), Thwarted Belongingness (TB), Depression

As measured by adolescent responses at baseline to the burdensomeness question (ASQ question #2), 10.5% (n = 677) of adolescents reported perceived burdensomeness (PB), 82.8% (n = 5342) did not report PB, 6.3% (n = 407) selected the “no response” option and 0.3% (n = 22) did not respond to the question. Adolescents’ mean scores on the Connectedness Scale and the PHQ-9 at baseline were 3.9 (SD = 0.8) and 6.2 (subthreshold level, SD = 6.6), respectively.

Development of Painful and Provocative Events (PPE) Index

The LCA included five baseline indicators of PPE (lifetime suicide attempts, NSSI frequency, NSSI methods, sexual abuse, recent history of physically attacking someone) and yielded three PPE classes: low, 81% (n = 5222); moderate, 12.1% (n = 779); and high, 6.9%, (n = 447). As indicated in Table 2, this three-class model had a low BIC while still maintaining high entropy when compared to models with more or fewer classes. Regarding the exclusion of the physical abuse item, when we compared models that used either the physical abuse item or the sexual abuse item or both, the model which used the sexual abuse item had the best Bayesian Information Criterion and entropy. The distribution of the five variables across the three classes of PPE is provided in Supplement Table S1. The likelihood of each type of PPE (e.g., multiple suicide attempts, NSSI incidents, sexual abuse) increases across low, moderate, and high levels.

Table 2. Painful and Provocative Events (PPE): fit statistics and classification coefficients for Latent Class Analysis Models.

Number Classes	LL	AIC	BIC	Entropy	Smallest Class Size
2	−10684.028	21406.055	21534.714	0.996	18.848%
3	−10552.902	21163.805	21360.179	0.942	6.878%
4	−10465.509	21009.108	21273.108	0.792	6.925%

As a secondary analysis, we conducted the LCA with only three types of PPE: history of suicide attempt, NSSI frequency, and NSSI method, which have been widely studied and are established risk factors for suicide attempts (C. A. King et al., 2019), the results were similar. Considering only these self-harm experiences, the model was stronger when both NSSI frequency and method were included (entropy = 94.2%) versus when only one of these was included (NSSI frequency only: entropy = 77.8%; NSSI methods only: entropy = 71.5%).

Examination of IPTS Theory of Suicide

Prospective Prediction of Suicide Attempt

Table 3 indicates that higher levels of PPE were associated with a greater likelihood of suicide attempt at baseline and at 6-months. Table 4 presents the prediction models and their odds ratios (with confidence intervals) for the prediction of suicide attempts. Each IPTS construct was significant and important to the prediction of suicide attempts even after adjusting for baseline suicidal ideation. When also adjusting for the other IPTS constructs, only TB and PPE remained significant. At follow-up, the model with all three constructs (adjusting for baseline suicidal ideation) had the lowest Akaike information criterion (AIC).

Table 3. Suicide attempt predicted by Painful and Provocative Events (PPE).

Outcomes, Predictors	Suicide Attempt	Adjusted Odds Ratio (95% CI)^a	P-value
Previous suicide attempt during past month (at baseline)
PPE			<.001
Low	100/5204 (2%)	Reference
Moderate	121/796 (15%)	2.03 (1.48, 2.77)
High	204/420 (49%)	5.18 (3.76, 7.12)
Suicide attempt at 6-months
PPE			<.001
Low	37/1328 (3%)	Reference
Moderate	43/414 (10%)	2.07 (1.27, 3.36)
High	66/264 (25%)	3.50 (2.13, 5.73)

^aAdjusted for suicidal ideation at baseline (not shown); In models that did not adjust for suicidal ideation at baseline, the pattern of findings was similar, with significant differences in likelihood of suicide attempt (baseline and 6-months) across PPE levels.

Table 4. Suicide attempt predicted by Perceived Burdensomeness (PB), Thwarted Belongingness (TB), and Painful and Provocative Events (PPE).

	Suicide Attempt		Adjusted Model 1^a		Adjusted Model 2^b
Outcomes, Predictors	Yes	No	Odds ratio (95% CI)	P-value	Odds ratio (95% CI)	P-value
Previous suicide attempt during past month (at baseline)
PB^c				<.001		<.001
No	88/5307 (2%)	5219/5307 (98%)	Reference		Reference
Yes	278/669 (42%)	391/669 (58%)	7.56 (5.65, 10.12)		6.01 (4.41, 8.19)
No response	55/403 (14%)	348/403 (86%)	3.03 (2.05, 4.46)		2.50 (1.67, 3.74)
TB^d Median (Q1, Q3)	3.2 (2.7, 3.7)	4.2 (3.5, 4.5)	0.66 (0.57, 0.77)	<.001	0.93 (0.78, 1.11)	0.415
PPE				<.001		<.001
Low	99/5173 (2%)	5074/5173 (98%)	Reference		Reference
Moderate	120/791 (15%)	671/791 (85%)	2.04 (1.49, 2.79)		1.69 (1.22, 2.34)
High	202/415 (49%)	213/415 (51%)	5.26 (3.81, 7.24)		3.53 (2.52, 4.95)
Suicide attempt at 6-months
PB				<.001		0.127
No	52/1448 (4%)	1396/1448 (96%)	Reference		Reference
Yes	74/362 (20%)	288/362 (80%)	2.39 (1.58, 3.63)		1.58 (1.01, 2.49)
No response	18/184 (10%)	166/184 (90%)	1.63 (0.91, 2.92)		1.18 (0.64, 2.16)
TB Median (Q1, Q3)	3.2 (2.5, 3.7)	3.8 (3.2, 4.3)	0.62 (0.49, 0.78)	<.001	0.70 (0.55, 0.89)	0.003
PPE				<.001		<.001
Low	36/1322 (3%)	1286/1322 (97%)	Reference		Reference
Moderate	43/412 (10%)	369/412 (90%)	2.11 (1.29, 3.43)		1.89 (1.15, 3.10)
High	65/260 (25%)	195/260 (75%)	3.50 (2.12, 5.75)		2.80 (1.67, 4.71)

^aAdjusted for suicidal ideation at baseline (not shown), but unadjusted for other predictors; In a model that did not adjust for suicidal ideation at baseline, a similar pattern of findings emerged with stronger relationships (larger odds ratios).

^bAdjusted for suicidal ideation at baseline (not shown) and for other predictors. In a model that did not adjust for suicidal ideation at baseline, a similar pattern of findings emerged with stronger relationships (larger odds ratios, all p-values < .01).

^cIn the past few weeks, have you felt that you or your family would be better off if you were dead?” (ASQ Item #1).

^dHigher scores on TB represent higher connectedness and less thwarted belongingness.

The 3-way interaction between the three IPTS constructs (perceived burdensomeness [PB], thwarted belongingness [TB], painful and provocative events [PPE]) and suicide attempt outcomes at 6-month follow-up was nonsignificant (p = .403, odds ratios ranging from 0.31 to 1.32). Similarly, the bivariate interactions between PPE and, separately, PB and TB in relation to suicide attempt outcomes were nonsignificant (PB*PPE, p = .705, odds ratios ranging from 0.45 to 0.61; TB*PPE, p = .953, odds ratios 0.91 and 0.92).

Cross-Sectional Association of Thwarted Belongingness (TB), Perceived Burdensomeness (PB), and Painful and Provocative Events (PPE) with Baseline Suicide Attempt History

As presented in Table 4, PB and PPE were both significantly related to baseline suicide attempt history in models that adjusted for suicidal ideation only and in the model that adjusted for suicidal ideation and other predictors. TB was significantly related to baseline suicide attempt history in the model that adjusted for suicidal ideation only but not in the model that adjusted for suicidal ideation and other predictors. At baseline, the model with all three constructs (adjusting for baseline suicidal ideation) had the lowest AIC.

Prospective Prediction of Suicidal Ideation and Cross-Sectional Association of perceived burdensomeness (PB) and thwarted belongingness (TB) with Suicidal Ideation

PB and TB were each significantly related to suicidal ideation at baseline and at 6-month follow-up in unadjusted models, in models that adjusted for baseline depression, and in models that adjusted for baseline depression, PB, and TB. Table 5 presents the prediction models and their odds ratios (with confidence intervals) for the prediction of suicidal ideation.

Table 5. Suicidal ideation predicted by Perceived Burdensomeness (PB), and Thwarted Belongingness (TB).

	Suicidal Ideation		Unadjusted Models		Adjusted Model 1 ^a		Adjusted Model 2^b
Outcomes, Predictors	Yes	No	Odds ratio (95% CI)	P-value	Odds ratio (95% CI)	P-value	Odds ratio (95% CI)	P-value
Suicidal ideation at baseline
PB^c				<.001		<.001		<.001
No	253/5313 (5%)	5060/5313 (95%)	Reference		Reference		Reference
Yes	532/674 (79%)	142/674 (21%)	74.93 (59.89, 93.75)		17.64 (13.66, 22.77)		15.95 (12.31, 20.67)
No response	136/398 (34%)	262/398 (66%)	10.38 (8.15, 13.23)		3.94 (2.97, 5.21)		3.56 (2.67, 4.73)
TB^d Median (Q1, Q3)	3.2 (2.7, 3.8)	4.2 (3.7, 4.5)	0.25 (0.23, 0.28)	<.001	0.58 (0.51, 0.66)	<.001	0.76 (0.65, 0.87)	<.001
Suicidal ideation at 6-months
PB^c				<.001		<.001		<.001
No	115/1418 (8%)	1303/1418 (92%)	Reference		Reference		Reference
Yes	144/352 (41%)	208/352 (59%)	7.84 (5.90, 10.44)		3.17 (2.24, 4.48)		2.85 (2.01, 4.05)
No response	44/177 (25%)	133/177 (75%)	3.75 (2.54, 5.54)		2.20 (1.45, 3.35)		2.02 (1.32, 3.08)
TB^d Median (Q1, Q3)	3.2 (2.7, 3.7)	3.8 (3.3, 4.3)	0.42 (0.36, 0.49)	<.001	0.66 (0.55, 0.79)	<.001	0.73 (0.60, 0.88)	<.001

Denominators and percentages are based on row totals.

^aAdjusted for depression at baseline (not shown).

^bAdjusted for depression at baseline (not shown) and other predictor.

^c“In the past few weeks, have you felt that you or your family would be better off if you were dead?” (ASQ Item #1).

^dConnectedness Scale; higher scores represent more connectedness and less thwarted belongingness.

The interaction between Thwarted Belongingness and Perceived Burdensomeness significantly predicted suicidal ideation at follow-up when adjusting for baseline depression (p = .005, odds ratios 1.15 and 1.88) and without adjusting for baseline depression (p < .001, odds ratios 1.21 and 2.04). This interaction was also significant in predicting baseline suicidal ideation both with (p < .001, odds ratios 1.77 and 1.98) and without (p < .001, odds ratios 1.77 and 2.13) adjusting for baseline depression).

Discussion

This study examined three constructs of the interpersonal-psychological theory of suicidal behavior (IPTS) – thwarted belongingness (TB), perceived burdensomeness (PB) and painful and provocative events (PPE) – in a large and geographically diverse sample of U.S. adolescents. The study addressed important gaps in our knowledge by exploring these constructs prospectively and simultaneously, and by assessing PPE broadly. In keeping with study hypotheses and IPTS tenets, baseline PB, TB, and the interaction between PB and TB were significant predictors of suicidal ideation at baseline and at 6-month follow-up. Although the three-way interaction between TB, PB, and PPE was not associated with suicide attempts at 6-month follow-up, TB and PPE were each significant prospective predictors of suicide attempts in multivariable models that controlled for baseline suicidal ideation, and all three constructs were significant predictors in models that did not control for baseline suicidal ideation, which is perhaps not surprising given the close association between TB, PB and suicidal ideation proposed by IPTS. Findings suggest that all three IPTS constructs need not be present for an adolescent to engage in suicidal behavior, although they may have an additive impact on the likelihood of suicide attempt.

IPTS Constructs and Our Understanding of Adolescent Suicide Attempts

Study findings largely converge with and extend those of previously published studies of the IPTS (as reviewed by Chu et al., 2017; Stewart et al., 2017), which has been generally supported in studies that have sampled participants of differing ages from a variety of settings (e.g., military, school/community, psychiatric hospital) and used different instruments to assess key constructs. In the Chu et al. (2017) meta-analysis of 122 published and unpublished studies examining IPTS constructs and suicide-related outcomes, 48.3% of the studies were of college students, who may not be representative of many populations of interest, and only 6.9% of the studies were prospective. In Stewart et al.’s systematic review of the smaller number of studies (n = 17) that examined the IPTS with adolescents (Stewart et al., 2017), five studies were longitudinal, three studies examined interpersonal (thwarted belongingness, perceived burdensomeness) and painful and provocative events components of the theory, and only the Czyz et al. study (Czyz et al., 2015) examined indicators of all three of these components. The present prospective study sampled a larger and more diverse sample of adolescents and used a broader assessment of PPE. As such, study findings regarding the importance of these constructs to our understanding of adolescent suicidal ideation and suicide attempts are particularly compelling.

In this study, perceived burdensomeness (PB) and thwarted belongingness (TB) had a much stronger role in relation to suicidal ideation than suicide attempts among adolescents, which is in keeping with the IPTS tenet that PB and TB are critical precursors of suicidal ideation or suicidal desire, whereas painful and provocative events and, more broadly, acquired capability, are critical to whether or not an individual with suicidal desire engages in suicidal behavior. Contrary to the tenets of IPTS, however; findings suggest that PB and TB need not both be present for an adolescent to engage in suicidal behavior. As one possible example of this, an adolescent with a history of PPE (or, more broadly AC) who is drinking alcohol (depressant) and experiencing strong negative emotions, perhaps related to a personal failure (e.g., important exam, sports competition, job interview) may experience perceived burdensomeness due to failures, yet have some positive level of connectedness with others.

Our results also support Franklin et al.’s proposal that time varying risk factors have stronger effects in the near term than across longer time periods (Franklin et al., 2017). In fully adjusted models, we found a stronger relationship between PB and suicidal ideation at baseline (OR ~ 16) than between PB at baseline and suicidal ideation at 6 months (OR ~ 3). As noted by Horton et al. (2016), who found relatively strong cross-sectional associations between PB, TB, and suicidal ideation in a sample of youth shortly after hospitalization for acute suicide risk, these closer in time measurements may be capturing a cognitive state more associated with acute suicide risk. Our study findings suggest there also may be a stable component to PB, which is consistent with Bryan et al. (2010) who proposed that “PB may be indicative of a chronic suicide risk dimension … ” By examining both cross-sectional (baseline) and prospective relations in our study, we were able to document empirically the varying strength of these relationships.

Some previous research suggests that perceived burdensomeness (PB) may be more important than thwarted belongingness (TB) to our understanding of suicide risk (e.g., Chu et al., 2017; Ma et al., 2016), although findings have been mixed (e.g., Czyz et al., 2015) and few prospective studies have examined these constructs with adolescents. For example, Glenn and colleagues examined proximal risk in adolescents who had received acute psychiatric care for suicide risk and found that TB both predicted greater suicidal thoughts later in the day and mediated the association between interpersonal negative life events and suicidal thoughts the next day (Glenn et al., 2021), suggesting its importance to understanding suicidal thoughts. It may also be important to consider the measurement of TB, which is akin to connectedness, in multiple domains that include family, peers, and schools (e.g., Arango et al., 2021; C. A. King et al., 2019). In this study, we developed a composite measure with items tapping each of these forms of connectedness and found that both PB and TB were important to understanding suicidal ideation. In multivariable models that adjusted for baseline suicidal ideation; however, only PB and PPE were important to the cross-sectional prediction of suicide attempts and only TB and PPE were important to the prospective prediction of suicide attempts.

Perhaps this study’s most striking finding related to prospectively predicting suicide attempts is the importance of painful and provocative events (PPE), particularly history of suicide attempts, number of NSSI incidents, and number of NSSI methods, as these were key to the LCA classification. At baseline, 49% of adolescents in the high PPE group, 15% in the moderate PPE group, and 2% in the low PPE group had made a suicide attempt during the past month. Similarly, even with the study’s relatively brief 6-month follow-up period, 25% of adolescents in the high PPE group made a suicide attempt during follow-up versus only 3% of adolescents in the low PPE group. These findings build on those that emerged in Stewart and colleagues’ meta-analysis (Stewart et al., 2017) and suggest the potential clinical importance of considering PPE in suicide risk assessments.

In addition to the potential importance of IPTS constructs to clinical risk assessment and prediction (e.g., Joiner et al., 2021), the IPTS has an important explanatory role. As one example, a recent study of the multi-variable prediction of adolescent suicide attempts reported four variables that were most important to prediction: past week suicidal thoughts, lifetime severity of suicidal ideation, lifetime history of suicidal behavior, and school connectedness (C. A. King et al., 2019). Within the framework of IPTS, low levels of school connectedness can be conceptualized as a developmentally attuned aspect of thwarted belongingness (TB) for adolescents. This theory enables us to consider conceptual links among risk factors, moving beyond a “strong arm” probabilistic or prediction model.

Theory-Related Measurement Issues

The IPTS theory proposes that more active suicidal ideation – movement on the trajectory toward suicidal behavior – occurs when individuals experience hopelessness regarding the possibility of reducing their perceived thwarted belongingness (TB) and perceived burdensomeness (PB) (Van Orden et al., 2010). Although this construct was not examined in this study, nor, to our knowledge, in previous research with adolescents, the extant research with adults has yielded only weak support for the construct. Hopelessness about TB and PB have been identified as distinct constructs that capture experiences apart from TB, PB, and overall hopelessness (Mitchell et al., 2023). Nevertheless, in a cross-sectional study of undergraduate students, Tucker et al. (2018) reported that the three-way interaction of TB by PB by interpersonal hopelessness predicted only 1.7% of suicide risk above and beyond the single variables and two-way interactions, which accounted for 32.4% of suicide risk in their model. This suggests a relatively small predictive impact, and two more recent studies in adults did not find support for the construct (Forkmann et al., 2021; Gerner et al., 2023). Thus, findings are mixed regarding the importance of interpersonal hopelessness to active suicidal ideation in adults, and further study is recommended in adolescents.

Acquired Capability (AC) has been conceptualized as being comprised of two components, lower fear of death and greater physical pain tolerance, which develop from exposure to painful and provocative events (PPE) such as previous suicide attempts, NSSI, and physical pain and trauma (Joiner, 2005). In this study, we assessed PPE, a theorized precursor to AC, rather than positing questions pertaining to fearlessness of death or self-harming behaviors (e.g., J. D. Ribeiro et al., 2014; Van Orden et al., 2008). Similarly, we did not examine the three distinct contributors to capability for suicide proposed by the 3-Step Theory: dispositional (e.g., temperament, genetically driven factors), practical (e.g., access to lethal means) and AC (Klonsky & May, 2015). We note that rather than studying “capability” for suicide in the broad sense, we examined how PPEs combine with or contribute to thwarted belongingness (TB) and perceived burdensomeness (PB) in the prediction of suicide attempts cross-sectionally and prospectively.

Our latent class analysis with five self-reported indicators (lifetime suicide attempts, NSSI frequency, NSSI methods, history of sexual abuse, and recent physical aggression) yielded three classes or levels of painful and provocative events (PPE), with 12.1% and 6.9% of adolescents, respectively, found to have moderate and high levels of PPE. Our goal was to broaden the concept of PPE beyond history of suicide attempt and NSSI, which reflect only one type of PPE (self-harm) and are well-documented risk factors for suicide attempts (e.g., C. A. King et al., 2019). In secondary analyses, however, we found that additional variables may be unneeded, although, importantly, number of NSSI incidents and NSSI methods each contributed meaningfully to PPE classification. The strong relationship between PPE and suicide attempt outcomes in this study suggests the potential importance of PPE in clinical risk assessment. They also suggest that self-harm experiences may be most critical as opposed to a broader array of PPE experiences.

Study Limitations

It is important to interpret study findings within the context of study limitations. Although study strengths include its prospective design in addition to its large and geographically diverse sample of adolescents, this sample was not nationally representative. Moreover, we had a lower retention of racial and ethnic minority youth in this study. These study characteristics suggest that our findings are not generalizable to the larger U.S. population of adolescents. Other limitations pertain to measurement issues. Our measure of painful and provocative events (PPE) tapped a broader range of PPE than previous adolescent studies; however, the self-harm PPE of suicide attempt and NSSI (numbers of incidents and methods) were sufficient for LCA classification, and we did not assess self-perceptions of fearlessness or the contributors to capability of suicide proposed in the 3-Step Theory (Klonsky & May, 2015). We also used a newly developed single proxy item for PB, although this measure has strong content validity for the construct of perceived burdensomeness and has been validated prospectively (C. A. King et al., 2019). Although its importance is unclear, particularly in adolescents, we also did not measure adolescents’ perceived hopelessness about the potential for change in their PB or TB.

Regarding data analyses, we may have been underpowered to detect significant 3-way interactions in models predicting suicide attempts at 6 months. Because TB is a continuous variable, and PB and PPE are categorical with 3 levels each; logistic regression models needed to include many estimates to evaluate interactions. While we had follow-up data for 2,009 adolescents, only 146 of these adolescents had a suicide attempt event prior to follow-up, making evaluation of complex interaction terms difficult. Similarly, given the large number of variables and interactions considered in analyses, it was beyond this study’s scope to examine the invariance of models across subgroups of adolescents defined by sociodemographic variables.

Conclusion

Adolescent suicide risk is an urgent public health concern, and it is imperative that we improve our understanding of this risk as well as our ability to identify adolescents who may benefit from further evaluation and treatment. Given the complexity and heterogeneity of suicide risk factors, research grounded in rigorously validated theoretical frameworks has the most potential to further our knowledge in this area. In this study, we examined three constructs of the interpersonal-psychological theory of suicidal behavior (IPTS) – thwarted belongingness (TB), perceived burdensomeness (PB) and painful and provocative events (PPE) simultaneously and prospectively in a large and geographically diverse sample of U.S. adolescents.

Study results provide the strongest support for the IPTS predictions regarding TB and PB, which were each significant predictors of suicidal ideation, as was the interaction of TB by PB, even when adjusting for baseline depression. Moreover, TB and PPE were each significant predictors of future suicide attempts when adjusting for baseline suicidal ideation, and TB, PB, and PPE were each significant predictors when this adjustment was not included in the model. To some extent, the constructs had an additive impact on the likelihood of suicide attempt outcomes, and the importance of PPE to the prediction was striking. Taken together, several components of IPTS are useful to our understanding of adolescent suicide attempts, may contribute importantly to clinical risk assessment, and may be promising targets for prevention and intervention strategies.

Acknowledgments

We thank the PECARN Data Coordinating Center staff at the University of Utah for data programming, data management, and project management; we also thank the research teams at participating sites and the youth and families who participated.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Supplementary Data

Supplemental material for this article can be accessed online at https://doi.org/10.1080/15374416.2024.2330068.

Additional information

Funding

Dr. King receives research support from NIMH and AFSP, receives royalties from Guilford Press, and is on the Scientific Advisory Board of the American Foundation for Suicide Prevention. Funding from NIMH supported the development of intellectual property for the Computerized Adaptive Screen for Suicidal Youth (CASSY), with no current financial interest. Dr. Brent receives research support from NIMH, AFSP, the Once Upon a Time Foundation, and The Beckwith Institute, receives royalties from Guilford Press, from the electronic self-rated version of the C-SSRS from eRT, Inc., and from performing duties as an UptoDate Psychiatry Section Editor, receives consulting fees from Healthwise, receives Honoraria from the Klingenstein Third Generation Foundation, and is a scientific board member for AFSP. Intellectual Property, currently with no financial interest: Funding from NIMH supported CASSY adaptive screen, a suicide risk machine learning algorithm, the BRITE safety planning app, and the Screening Wizard screening tool.This study was funded primarily by grant [U01 MH104311] from the National Institute of Mental Health. This study was also supported in part by the Health Resources and Services Administration (HRSA), Maternal and Child Health Bureau (MCHB), Emergency Medical Services for Children (EMSC) Network Development Demonstration Program under cooperative agreements [U03MC00008, U03MC00001, U03MC00003, U03MC00006, U03MC00007, U03MC22684, U03MC28845, H3MC26201 and U03MC22685]. The content and conclusions in this manuscript are those of the authors and should not be construed as the official position or policy of, nor should any endorsements be inferred by NIMH, HRSA, HHS or the U.S. Government.