Exploring the Relations Between Interpersonal Risk and Adolescent Suicidality During Treatment

Abstract

Objective:

Despite considerable evidence that supports perceived burdensomeness (PB) and thwarted belongingness (TB) as risk factors for suicidal ideation (SI), far less is known about the direction of effects between these constructs in treatments for suicidal adolescents. The current study examined bidirectional relations between PB, TB, and adolescents’ suicidal ideation (SI) during a 16-week randomized clinical trial.

Method:

129 depressed and suicidal adolescents completed PB, TB, and SI measures at three time points: baseline (T1), mid-treatment (T2), and treatment completion (T3). Random-intercept cross-lagged panel models (RI-CLPM) examined within-subject direction of effects between interpersonal variables (PB & TB) and suicidal ideation (SI) in the first and second halves of treatment.

Results:

Within-subjects, autoregressive paths indicated significant carryover in PB and SI. In the first half of treatment, a significant cross-lagged path indicated that T1 PB predicted change in T2 SI, and in the last half of treatment change in T2 SI predicted change in T3 PB. There were no significant auto-regressive or cross-lagged effects for TB.

Conclusions:

In the first half of treatment, baseline PB predicted fewer reductions in SI suggesting that PB initially moderated adolescents’ response to treatment. However, in the last half of treatment, initial reductions in SI predicted subsequent reductions in PB suggesting that adolescents’ initial response to treatment decreased their perceptions of burdening others. The clinical and treatment implications of these bidirectional findings are discussed.

Suicide is the second leading cause of death in those between the ages of 10 and 24, and despite expansions in research and treatment development, the total suicide rate in the United States has increased 31% since 2001 (CDC, 2017). The interpersonal-psychological theory of suicide (IPTS; Joiner, 2005; Van Orden et al., 2010) is a predominant theory for conceptualizing risk for adolescent suicidality (Stewart et al., 2017). The IPTS posits that suicidality is precipitated by two interpersonal beliefs: thwarted belongingness (TB), or the belief that one is alone or isolated from social circles, and perceived burdensomeness (PB), or the belief that one’s life is a burden on loved ones and/or society (i.e., “they would be better off without me”). The IPTS has been employed in over 100 samples with a recent meta-analysis indicating moderate effect sizes between interpersonal beliefs and suicidal symptoms, with larger effects for PB than for TB (Chu et al., 2017). However, while PB and TB beliefs have been tested as proximal predictors of suicide ideation (SI), few studies have examined bidirectional relations between IPTS variables and SI during treatments for suicidal adolescents. More generally, studies of adolescents may need to reconsider IPTS variables within the context of parent-adolescent relationships (Nock, et al., 2008) and in light of the possible strain that adolescents’ suicidality generates for their family caregivers. From this perspective, not only may IPTS variables increase risk for SI but adolescents’ SI may increase their perceptions of burdensomeness and thwarted belongingness. The current study tests this possibility by examining IPTS variables and SI in an RCT for suicidal adolescents.

Recent studies of adult community and in-patient samples have used intensive longitudinal designs to assess bidirectional relations between IPTS variables and SI. Kleiman and colleagues (2017) assessed these variables 4 times per day for 28 days. Only PB predicted subsequent SI, but this effect did not remain significant after controlling for prior SI. In a study of inpatient suicidal adults that used three daily samples, Kyron, Hooke, and Page (2018) reported bidirectional effects showing that IPTS variables predicted subsequent change in SI and that SI also predicted subsequent changes in PB and TB. Rogers and Joiner (2019) provided further evidence for bidirectional effects in a community sample of high-risk adults. Using repeated online measurements (6 daily time points sampled at 3-day intervals), they reported bidirectional effects between PB and SI but not between TB and SI. Given that the IPTS posits unidirectional predictions of PB and TB influencing subsequent suicidal ideation and behaviors, this emerging evidence for the bidirectional effects of SI on PB warrants further investigation.

Studies of adolescent samples have yielded few conclusive findings about the relative degree of risk that PB and TB confer on suicidality (Stewart et al., 2017). For instance, although TB and PB mediated the relationship between depression and SI in an undergraduate sample (Kleiman, Liu, & Riskind, 2014), a clinical sample of suicidal adolescents found that only PB and not TB was associated with concurrent suicidal ideation (Miller, Esposito-Smythers, & Leichtweis, 2016). In a treatment study using a pre-post design, King and colleagues (2018) reported that while neither pre-treatment TB nor PB predicted post-treatment or follow-up suicide risk, an interaction between post-treatment reductions in PB and TB was associated with reductions in subsequent suicide risk. Another treatment study indicated that reductions in adolescents’ and young adults’ PB but not TB mediated the effects of depression and hopelessness on post-treatment suicide risk (Hains, Janackovski, Deane, & Rankin, 2018). Taken together, these treatment studies support Chu and colleagues’ (2017) summary of the literature suggesting that PB confers relatively greater risk for suicidality than TB. However, none of these adolescent studies addressed the possibility of bidirectional relations between SI and either PB or TB.

Intensive longitudinal designs with adult community and inpatient samples also indicate that PB confers greater risk than TB. Kleiman and colleagues (2017) assessed these variables 4 times per day for 28 days. Only PB predicted subsequent SI, but this effect did not remain significant after controlling for prior SI. In a study of inpatient suicidal adults that used three daily samples, Kyron, Hooke, and Page (2018) reported bidirectional effects indicating bidirectional effects showing that IPTS variables predicted subsequent change in SI and that SI also predicted subsequent changes in PB and TB. Rogers and Joiner (2019) provided further evidence for bidirectional effects in a community sample of high-risk adults. Using repeated online measurements (6 daily time points sampled at 3-day intervals), they reported bidirectional effects between PB and SI but not between TB and SI. Given that the IPTS posits unidirectional predictions of PB and TB influencing subsequent suicidal ideation and behaviors, this emerging evidence for the bidirectional effects of SI on PB warrants further investigation.

In a recent treatment study of women with eating disorders, Bodell and colleagues (2020) parsed within-subject or state-like fluctuations in IPTS and SI variables from more trait-like or between-subject differences in these measures. Time-lagged analyses indicated that within-subjects’ fluctuations of PB, predicted SI one week later and SI predicted the subsequent week’s PB (Bodell, Smith, & Witte, 2020). Similar effects were not evident for TB. Increasing evidence for a direction of effect in which SI influences subsequent change in PB has often been attributed to intrapersonal factors that covary with SI such as stigma, self-isolation, and perceived worthlessness that bias adults toward perceiving themselves as a burden to others. Less often considered is the possibility that suicidal adolescents perceive their suicidality as creating interpersonal strain for their loved ones consistent with of the high levels of strain reported by their caregivers (Barksdale et al., 2009). Adolescents’ implicit or explicit awareness of their caregivers’ strain may in turn, contribute to their more general perceptions of burdening others.

Surprisingly, treatment studies have not considered how adolescents’ high levels of SI may increase their perceptions of burdening others. An initial step for exploring this possibility would be to examine whether early reductions in adolescents’ SI in treatment predicts subsequent reductions in PB later in treatment. Treatment studies of suicidal adolescents provide a unique opportunity to employ cross-lagged panel designs to further explore bidirectional relations between SI and IPTS variables in different phases of treatment. However, these cross-lagged panel designs require separating trait or between-subject difference in these variables from state or within-subject fluctuations in IPTS and SI variables in order to examine bidirectional effects over the course treatment.

The Current Study

The current study uses a three-panel cross-lag panel design (T1 baseline, T2 mid-treatment, T3 end of treatment) using secondary analyses of a sixteen-week randomized control trial (RCT) for depressed and suicidal adolescents (Diamond et. al, 2019). In the first panel, we tested bidirectional effects between baseline variables and mid-treatment changes in those variables. In the second panel, we tested whether changes in IPTS and SI variables at mid-treatment predicted subsequent reductions in those variables by the end of treatment. A random-intercept cross-lagged panel model (RI-CLPM) distinguished between-person (trait-like) variance in the three repeated measures from the within-person (state-like) components. Isolating state-like fluctuations from the more stable between person components allowed us to examine the cross-lagged effects and reciprocal relationships between the IPTS and SI variables. In accord with IPTS theory, we anticipated that pre-treatment IPTS risk would moderate initial treatment response and predict less reductions in mid-treatment SI. In the second half of treatment, we anticipated that mid-treatment reductions in IPTS and SI variables would predict cross-lagged reductions in post-treatment IPTS and SI variables. We also examined treatment condition as a potential moderator of autoregressive and cross-lagged paths in the RI-CLPM model.

Method

Participants

Participants were 129 adolescents enrolled in an RCT for the treatment of suicidal and depressed adolescents (NCT01537419: Attachment-Based Family Therapy for Suicidal Adolescents; Diamond et al., 2019). The RCT was designed to test the superiority of Attachment-Based Family Therapy (ABFT) to Family-Enhanced Non-Directive Supportive Therapy (FE-NST). ABFT links adolescents’ suicidality to ruptures in the parent-teen relationship and seeks to repair these ruptures through improved parent-teen communication (Diamond, Russon, & Levy, 2016; Kobak, Zajac, Herres, & Ewing, 2015). FE-NST uses supportive client centered individual sessions to explore adolescents’ distressing thoughts and feelings (Brent & Kolko, 1991). This Non-Directive Supportive Therapy (NST) was enhanced with a joint parent-adolescent session as well as parent psychoeducation that focused on safety-planning and managing the adolescents’ suicidal symptoms. Although adolescents in both treatment conditions showed marked reductions in SI, there were no significant differences between the two treatment conditions (Diamond et. al, 2019).

The average age of participants was 14.87 (SD = 1.68) and most (81.9%) were female. The sample was racially diverse with almost half (49.7%) of adolescents identifying as black, 28.7% as white, and 21.6% as American Indian, Pacific Islander, biracial, or “other.” Adolescents were referred from schools, hospitals, and outpatient clinics in a large mid-Atlantic city. To be included in the study, adolescents endorsed severe suicidal ideation (≥ 31 on the SIQ-JR) and clinically significant depression (> 20 on the BDI-II) and had the participation of at least one primary caregiver. Adolescents were excluded if they did not speak English, had initiated psychotropic medication within 3 weeks of study enrollment, or required a higher level of care (e.g., imminent risk of harm, severe cognitive impairment). More details on the sample, including a CONSORT table, is presented elsewhere (Diamond et al., 2019).

Measures

Interpersonal Needs Questionnaire (INQ; Van Orden, Cukrowicz, Witte, & Joiner, 2012).

The INQ is a self-report measure assessing the two major constructs of the IPTS: Thwarted Belongingness (TB) and Perceived Burdensomeness (PB). The measure’s 15 items are scored on a 7-point Likert scale from 1 (“Not at all true for me”) to 7 (“Very true for me”). Nine of the items measure TB (e.g., “other people care about me” and “I feel like I belong,” reverse coded), which demonstrated acceptable internal consistency (α = .76). Six items measure PB (e.g., “the people in my life would be happier without me” and “I think I am a burden on society”). Adolescents completed the INQ at sessions 0, 8, and 16 reporting on how they “have felt recently.” In the current sample, these PB items demonstrated good internal consistency (α = .92).

This 2-factor model (Van Orden et al., 2012) has been widely utilized in adult samples and to a more limited extent in adolescent samples (Baams, Grossman, & Russell, 2015; Buitron et al., 2016; Grossman, Park, & Russell, 2016; Podlogar, Žiberna, Poštuvan, & Kerr, 2017). A previous study with the current sample suggested that both a 2-factor and a 3-factor model could be derived from the Van Ordern’s items with acceptable fit (Hunt et al., 2019). However, this third factor, Perceived Isolation, has only 2 items that are typically included in TB factor (Hunt et al., 2019) and removing these two items reduced the overall internal consistency of the TB scale. Although promising, further tests of the 3-factor model will require developing additional items for the third factor and testing the revised scale in adolescent samples. Therefore, for the current study, we used the 2-factor model for the sake of parsimony and for purposes of comparison with the prior literature.

Suicidal Ideation Questionnaire-Junior (SIQ-JR-Monthly; Reynolds & Mazza, 1999).

The SIQ-JR-Monthly is a self-report measure assessing frequency of thoughts about suicide in the past month. Total scores range from 0 to 90 with scores equal to or above 31 indicating severe suicidal ideation. Examples of the 15 items include “I wished I were dead” and “I thought about killing myself,” rated from 0 (“I never had this thought”) to 6 (“Almost every day”). The scale demonstrated good internal consistency (Cronbach’s α = .86) in this sample of suicidal adolescents.

Beck Depression Inventory (BDI-II; Beck, Steer, & Brown, 1996).

Depressive symptoms were assessed using the BDI-II, a 21-item self-report measure. Items are rated on a 4-point scale (0: absence of symptom, 3: most severe symptom) and inquire about a range of depressive symptoms, including sadness, irritability, loss of interest, and worthlessness in the past month. This measure has been validated for use with adolescents (Osman, Kopper, Barrios, Gutierrez, & Bagge, 2004). In the current sample, the BDI-II demonstrated good internal consistency (Cronbach’s α = 0.85).

Treatment Condition.

Adolescents were randomly assigned to receive either Attachment-Based Family Therapy (ABFT; Diamond, Diamond, & Levy, 2014) or Family-Enhanced Non-Directive Supportive Therapy (FE-NST; adapted from Brent & Kolko, 1991). Both treatments were 16-weeks long. ABFT is rooted in attachment theory and involves individual sessions with the adolescent, individual sessions with the parent, and joint sessions to address an attachment rupture in the relationship. The primary goal is to increase the adolescent’s use of the parent as a resource to help manage depressive and suicidal thoughts. FE-NST is a modified version of supportive therapy in which the therapist validates, empathizes, and attends to the adolescent’s concerns in individual sessions. To balance ABFT, there were also individual sessions with the parent that involved psychoeducation and joint parent-adolescent sessions focused on safety planning.

Procedure

Following a baseline assessment, participants were randomized to 16 weeks of ABFT or FE-NST. Adolescents’ self-reports of SI, PB, and TB were collected at baseline (week 0), half-way through treatment (week 8), and upon completion of treatment (week 16) by an outcomes study team blind to treatment condition. This study was approved by the Institutional Review Boards of the participating sites.

Data Analytic Plan

We conducted preliminary analyses to examine if between subject differences in IPTS variables moderated adolescents’ reductions in SI over the course of treatment. These initial analyses included adolescents’ depressive symptoms as a between-subject control variable. Hierarchical Linear Modeling was used to account for the retention of participants despite missing data and estimate robust standard errors that better account for non-normality and outliers in the data, providing increased accuracy in significance testing (Raudenbush & Bryk, 2002). The multilevel modeling approach allows for the specification of within-subject change in SI and between-subjects moderators of treatment. The rate of change in SIQ demonstrated a non-linear trend, therefore a log-10 transformation of the time variable was used to account for faster rates of change early in treatment (Young et al., 2016).

To test our primary hypotheses about direction of effects between IPTS variables and SI in the first and second halves of treatment, we tested a random-intercept cross-lagged path model (RI-CLPM; Hamaker, Kuiper, & Grasman, 2015) according to procedures outlined by Ponzini and colleagues (2019). Three time points were specified: T1 (pre-treatment), T2 (8 weeks into treatment), and T3 (completion of treatment at 16 weeks). Analyses were run using Mplus Version 7.0 (Muthén & Muthén, 1998-2010). Model fit was assessed using the Comparative Fit Index (CFI), the Root Mean Square Error of Approximation (RMSEA), the Standardized Root Mean Square Residual (SRMR), and chi-square statistics. According to Kaplan (2000), acceptable values for CFI are above 0.9 (above 0.95 is considered good fit) and acceptable values for RMSEA and SRMR are below 0.05. Because we used maximum likelihood estimation (MLR) to obtain robust estimates with missing data, model comparisons were tested using the Satorra–Bentler Scaled Chi-square difference test.

The RI-CLPM addresses a major limitation of traditional cross-lagged path models (CLPM), which do not distinguish between within-person and between-person variability (Hamaker, Kuiper, & Grasman, 2015). CLPM does not estimate between-person differences, allowing individuals to vary across the group means. However, this can result in biased and inaccurate estimates of regression coefficients, as it is rarely the case that there is no between-person variation in traits (Hamaker, Kuiper, & Grasman, 2015). The RI-CLPM addresses this limitation through distinguishing the between-person (i.e., trait-like) and within-person (i.e., state-like) variance by creating one latent variable for each construct across the measurement time points, with factor loadings constrained to 1. These new random-intercept latent variables are added to the traditional CLPM model to account for the between subject trait-like variability in each of the repeated measures. The correlations between these latent factors therefore represent the relationship between these trait-like portions of the variance. After accounting for the trait-like component, interpretations of regression coefficients are then relative to each individual’s own mean or state variations in the constructs. Auto-regressive paths in RI-CLPM models reflect within-person carryover from one time period to the next, or the extent to which one’s later scores are predicted by earlier states. The cross-lagged paths indicate the degree to which earlier states predict future states in another construct. In other words, they reflect the degree to which adolescents’ states on one variable can be predicted by prior states on another variable, controlling for other variables in the model. Within-person correlations indicate cross-sectional correlations between adolescents’ states on one variable and their states in another variable at the same time point. In Figure 1, we illustrate a full RI-CLPM using two of our three variables for the sake of visual simplicity. Finally, we tested for an effect of treatment condition by assessing whether model fit was equivalent across ABFT and FE-NST.

Figure 1.

Illustration of a RI-CLPM for Thwarted Belongingness (TB)/Perceived Burdensomeness (PB) and Suicidal Ideation (SI) across three measurements: baseline (T1), mid-treatment (T2), and treatment completion (T3). Only two of the three variables are included for illustrative purposes. Squares represent observed variables and circles represent latent variables. Straight arrows represent direct effects and curved lines represent correlations between variables.

Results

Means and standard deviations of SI, PB, and TB at T1 (baseline), T2 (mid-treatment) and T3 (end of treatment) are presented in Table 1.

Table 1.

Descriptive statistics between study variables

	M	SD	Range	N
SI T1	49.89	15.16	27-87	129
SI T2	28.52	20.62	0-90	104
SI T3	20.89	16.80	0-72	108
PB T1	3.95	1.57	1.0-7.0	121
PB T2	3.42	1.91	1.0-7.0	97
PB T3	2.39	1.57	1.0-7.0	105
TB T1	3.99	1.04	1.0-6.6	121
TB T2	3.91	1.23	1.4-6.3	97
TB T3	3.28	1.30	1.0-6.3	105

SI = Suicidal Ideation; PB = Perceived Burdensomeness; TB = Thwarted Belongingness; T1 = baseline, week 0; T2 = mid-treatment, week 8; T3 = treatment end, week 16

Missing Data

Most adolescents (80.6%) completed all 3 measurements of suicidal ideation. The number of completed measures was not significantly correlated with baseline measurements of suicidal ideation (r = 0.22, p = .82). Eight adolescents missed INQ data at baseline due to the measure being added to study protocol after the initiation of the RCT. Adolescents with missing INQ data did not differ from adolescents with complete data on any of the study variables. Further, Little’s test confirmed the data were MCAR (χ²(37) = 40.56, p = .32). Therefore, restricted maximum likelihood estimation was used.

Preliminary Between-Subject Analysis

Unconditional models of depression and suicide ideation were examined in order to determine the within-persons and between-persons variability of the weekly measures. The Intraclass Correlation Coefficient (ICC) for SI was .356, indicating that around 36% of the variability in suicide ideation is accounted for by between-person stability, and 64% was due to within-subject fluctuation. The within-subjects model estimated a fixed intercept of 48.66 (t(119) = 34.17, p < 0.001) as the average starting level of suicidal ideation symptoms. A fixed slope of −41.80 (t(119) = −14.95, p < 0.001) indicated that, on average, adolescents reported a significant decrease in suicidal ideation from the beginning to end of treatment. This rate of non-linear change corresponded to a total decline of 29 points on the SIQ-JR between baseline and post-treatment. Next, level 2 models were specified to examine IPTS variables on SI controlling for BDI. PB and BDI were significantly positively associated with SI at baseline (β_PB = 2.97, t(116) = 3.05, p = 0.003; β_BDI = 0.53, t(116) = 3.30, p = 0.001). TB was not significantly related to the intercept for SI (β_TB = −1.00, t(116) = −0.78, p = 0.437). Neither between subject TB, PB, nor BDI significantly moderated the rate of change in SI across treatment (all p’s > 0.10). The addition of the interaction between PB & TB also yielded non-significant results for the intercept (β_PBxTB = 0.31, t(115) = 0.46, p = 0.645) and slope (β_PBxTB = −1.04, t(115) = −0.72, p = 0.475).

Primary RI-CLPM Analysis

The Random Intercept-Cross Lag Panel Model demonstrated good fit (χ²(3) = 2.88, p = .41, RMSEA = 0.00, CFI = 1.00, and SRMR = 0.011), however, the random intercepts for PB and SI both had negative variances. By setting the PB and SI variances to zero, the model retained good fit (χ²(5) = 2.67 (p = .75), RMSEA = 0.00, CFI = 1.00, and SRMR = 0.012; Ponzini et al., 2019). Removing the PB and SI random intercepts from the model did not reduce model fit (Sattora-Bentler Scaled Chi-Square Difference: Δχ²(5, 8) = 64.29, p = .23, cd = .88) indicating that these variables were due to state-like and not between-subject effects. The final adjusted model provided good fit indicators (χ²(8) = 6.73, p = .57; RMSEA = 0.00, CFI = 1.00, and SRMR = 0.01) and is displayed in Figure 2, although only significant paths are presented for the sake of visual parsimony.

Figure 2.

RI-CLPM for Thwarted Belongingness (TB), Perceived Burdensomeness (PB), and Suicidal Ideation (SI). Only statistically significant paths are shown for simplicity.

Standardized estimates of the autoregressive and RI-CLPM model are displayed in Table 2. Autoregressive paths for PB were significant, indicating that within-person state fluctuations in PB were positively predicted by PB at earlier timepoints (i.e., when an adolescent had higher state PB at T1, he/she also has higher state PB at T2 and T3). This was also true for SI. The autoregressive paths for TB were not statistically significant, providing little evidence that fluctuations in state TB at earlier time points predicted fluctuations at subsequent timepoints when TB was estimated as a between subject random intercept. In other words, state-like effects of PB and SI did endure over time, while TB is likely explained by a more stable trait-like component. Two significant cross-lagged paths emerged between PB and SI. T1 PB predicted T2 change in SI (β = 0.28, p = .01), indicating that within-person states of PB at T1 predicted change in SI in the first half of treatment. In other words, when adolescents had higher PB states at the beginning of treatment, they tended to report less reduction in suicidal ideation at T2. However, in the second half of treatment, this cross-lagged path was not significant. The opposite cross-lagged effect was evident with change in state levels of SI between T1 and T2 significantly predicting changes in state levels of PB between T2 and T3 (ß = 0.27, p = .019). There were no significant cross-lagged paths involving TB.

Table 2.

Standardized autoregressive and cross-lagged paths

	T1 → T2			T2 → T3
	ß	SE	p	ß	SE	p
Autoregressive
PB→PB	.39	.14	.005**	.46	.16	.003**
SI→SI	.23	.09	.015*	.77	.07	<.001***
TB→TB	.31	.18	.093	.15	.27	.569
Cross-lagged
PB→SI	.28	.11	.010*	.09	.13	.464
PB→TB	.13	.15	.373	.27	.19	.145
TB→PB	.04	.14	.768	−.06	.16	.710
TB→SI	.08	.09	.383	−.17	.14	.211
SI→PB	.05	.10	.602	.27	.11	.019*
SI→TB	−.05	.11	.655	−.05	.11	.655

^*p <.05

^**p <.01

^***p<.001

SI = Suicidal Ideation; PB = Perceived Burdensomeness; TB = Thwarted Belongingness.

Within-person, or state-like, correlations are presented in Table 3. At baseline (T1), within levels of PB were positively correlated with state levels of TB and SI. By comparison, within levels of SI and TB were not significantly correlated with each other at T1. However, all three state level variables were significantly positively correlated with each other at mid-treatment (T2) and end of treatment (T3) with medium to strong effects. Lastly, there were no differences in the model based on treatment modality (ABFT vs. FE-NST); Δχ²= 7.80, Δdf= 11, p= .731.

Table 3.

Summary of correlations for state variables

	r	p
PB T1 – TB T1	.46	<.001**
PB T1 – SI T1	.36	<.001**
SI T1 – TB T1	.14	.202
PB T2 – TB T2	.66	<.001**
PB T2 – SI T2	.41	<.001**
SI T2 – TB T2	.44	<.001**
PB T3 – TB T3	.56	<.001**
PB T3 – SI T3	.60	<.001**
SI T3 – TB T3	.36	<.001**

Note. All pathways were freely estimated.

^*p <.05

^**p <.01

^***p<.001

SI = Suicidal Ideation; PB = Perceived Burdensomeness; TB = Thwarted Belongingness.

Discussion

The current findings support bi-directional relations between PB and SI in the context of an RCT for suicidal adolescents. The direction of effects were notably reversed in the first and second half of treatment. In the first half of treatment, adolescents with elevated pre-treatment PB showed less mid-treatment reductions in SI. This finding supports and extends evidence for PB not only as a risk factor for subsequent SI, but also as an initial indicator of poorer treatment response. However, in the second half of treatment, adolescents’ initial reductions in SI predicted subsequent reductions in PB. This direction of effect suggests that adolescents’ suicidality may amplify their perception of burdensomeness, such that their suicidality itself is a burden on those around them. Although we do not have measures of caregivers’ concern or anxiety about their adolescents’ suicidal ideation and behavior, it seems likely that these caregivers would be encouraged when their adolescents showed initial positive treatment benefit, which could reduce adolescents’ perception of burdensomeness. With the exception of a prior adult treatment study of eating disorders that reported a similar effect of weekly SI on the subsequent week’s PB (Bodell, 2020), the current findings are the first to suggest that reductions SI may predict subsequent reductions in PB in the context of a treatment that was specifically designed to reduce SI.

By controlling for between person or trait levels of the ITPS and SI variables, these findings distinguished state fluctuations as indicators of change in the treatment context. The nonsignificant random intercepts for PB and SI indicated substantial state fluctuations or change in these variables in both treatment phases. However, significant within-subject auto-regressive paths for these variables indicated substantial continuity or carryover between assessments. The smaller autoregressive paths for SI in the first compared to the second halves of treatment indicated that within person reductions in SI was more likely to occur in the first half than in the second half of treatment. This is consistent with previous findings indicating relatively more rapid decline out of the severe range in the first half of treatment and slower decline in the second half (Diamond et al., 2019).

The relative stability of TB in this treatment context may indicate that adolescents’ perceptions of loneliness and social isolation persisted through treatment. However, it is possible that the lack of findings with TB in this sample is due to psychometric factors. Hunt and colleagues (2019) found preliminary evidence for a subscale of TB called Perceived Isolation (PI) in this sample. Perhaps PI would demonstrate less stability over time, however, future research is needed to explore and further validate this potential third factor by adding more items and validating in bigger, more diverse samples. Further, while extant literature has shown that TB in parent but not peer relationships is associated with SI (Barzilay, 2019), the INQ does not distinguish family and peer relationships. Therefore, it is possible that adolescents’ self-reports of TB are more closely tied to their views of being isolated from peer relationships, whereas family relationships were more explicitly addressed and targeted by the ABFT or FE-NST interventions. Insofar as adolescents’ perceptions of burden are more likely to be focused on more intimate relationships with caregivers or close friends, the RCT’s focus on family involvement and safety planning would more likely allow for increased state fluctuations in PB. Lastly, another possible explanation is that TB is more distally related to suicidal ideation via other pathways that are less relevant to clinically severe samples like the one in the current study (Rogers & Joiner, 2019). More research is needed to better understand TB as a potential mechanism of suicidality especially in adolescents who face unique interpersonal challenges compared to adults.

Change in PB in the first half of treatment did not predict subsequent reductions in SI in the second half of treatment which may be attributed to several factors. First, the relative stability of SI in the second half of treatment suggests it is a less malleable target for change in these data. Second, neither ABFT nor FE-NST identified adolescents’ PB as an explicit target for change. While ABFT viewed repairing a rupture in the parent-adolescent relationship as the primary treatment target, this approach could increase adolescents’ awareness of the potential strain that their SI posed to their caregivers, maintaining perceptions of burden. Similarly, although FE-NST included psychoeducation for parents and supportive individual work with adolescents, neither intervention sought to reduce adolescents’ perceptions of burden. It is notable that the between-person or “trait-like” component of TB indicated considerable stability and failed to yield within-person state fluctuations across the three assessments.

The clinical implications of the effect of initial reductions in SI predicting subsequent reductions in PB can be considered at both the intrapersonal and interpersonal levels of analysis. At the intrapersonal level, initial reductions in SI would likely coincide with less negative affect and corresponding changes in self-perception that include perceptions of burdening others. Adult studies provide some support for this view of change in PB as attributable to intrapersonal covariates of SI such as stigma, self-isolation, and perceived worthlessness. However, an exclusive focus on intrapersonal factors neglects the interpersonal context in which families seek treatment for an adolescent’s suicidality. Parents in these families are likely to experience substantial threat to their caregiving role as a protector and support for a suicidal child. As a result, parents are likely to have a considerable emotional investment in determining if treatment is effective in reducing suicide risk. From this perspective, parents would likely experience relief and a reduction of strain as a result of initial reductions in their adolescents’ SI. To the extent that adolescents are aware of the strain that their suicidality places on parents, changes in parents’ strain could influence the adolescent’s perceptions of burden. Conversely, a lack of initial treatment gain could amplify perceptions of burden (e.g., “This isn’t working and I’m continuing to burden my parents”). For clinicians, this potential interpersonal component of adolescents’ perceptions of burden points to the importance of monitoring this dynamic between adolescents and their caregivers to improve treatment efficacy. Future studies might also develop measures that are explicitly designed to assess interpersonal factors such as caregiver strain that may contribute to and shape adolescents’ perceptions of burden.

Limitations and Future Directions

There are several notable limitations of the study design. First, the sample is limited to clinically depressed and suicidal adolescents who had a primary caregiver willing to participate in treatment. The temporal design of the study relies on measures that were assessed at eight-week intervals. A notable measurement limitation is that adolescents were instructed to report their SI symptoms in the past 4 weeks, and their TB and PB in how they were feeling recently. As a result, the instructions for reporting did not fully match the 8-week interval between assessments. Including weekly or bi-weekly assessments that could be aggregated over the 8-week period would improve measurement of the constructs. Although this design allows useful inferences about how change in the first part of treatment effects change the last part of treatment, weekly or daily measurements would be more appropriate for capturing daily or weekly within person fluctuations in IPTS and symptom variables. This endeavor may be assisted by adding measurements of other key aspects of the IPTS, including acquired capability and hopelessness, and by studying how and if these variables predict the transition from suicidal ideation to suicidal behaviors. Future studies might also seek to distinguish between the role caregivers and peers play in shaping adolescents’ perceptions of PB and TB.

A significant strength of this study is the methodological and statistical analysis of a RI-CLPM design. The three-panel cross-lag design provides a strong inferential basis for examining direction of effects and potential causal relationships between variables in the first and second halves of a treatment study. Further, the random intercept design focused on the within-person components which may be conflated in the tradition CLPM (Hamaker, Kuiper, & Grasman, 2015). Clinically, the findings highlight the importance of considering how adolescents’ suicidal ideation may contribute to their perceptions of burdensomeness. The interpersonal nature of PB in adolescent-parent relationships may also be a worthy target for discussion and intervention early in treatment. Future clinical trials should examine the effects of targeting PB directly from the outset of treatment and monitoring PB as a clinical marker throughout treatment. Ultimately, findings from this study extend our understanding of depressed adolescents at risk for suicide and point to PB as an important and dynamic construct in adolescent suicidality.

Public Health Significance Statement:

Identifying factors that predict response to treatment can help practitioners identify targets for intervening with suicidal and depressed adolescents. This study identifies perceived burdensomeness as a predictor of worse treatment response in suicidal adolescents, while also noting that reductions in suicidal ideation early in treatment predict subsequent reductions in perceived burdensomeness.

Appendix

Appendix. Data Transparency

The current manuscript uses data which were collected as part of a larger randomized controlled trial (RCT). Primary and secondary analyses from this RCT have been published elsewhere. Manuscript 1 (MS1) reports primary outcomes from the RCT, specifically treatment differences in change in depressive and suicidal ideation symptoms. MS2, MS3, and MS4 were cross-sectional studies using baseline variables only. MS2 used coded suicide narratives and adolescent attachment styles to predict baseline suicide ideation intensity. MS3 used analysis of adolescents’ social networks and peer deviance to predict baseline suicide ideation intensity. MS4 tested the predictive validity of the Interpersonal Needs Questionnaire-15 (INQ). MS5 used baseline emotion dysregulation and weekly assessments of negative affect, negative interpersonal events, and session insight to examine spillover effects of negative events on adolescents’ abilities to derive insight from weekly therapy sessions. MS6 identified three latent classes of treatment response which are distinguished by diagnosis of Major Depressive Disorder, baseline variables of non-suicidal self-injury, pessimism, and perceived burdensomeness.