Emotional processing, attachment and suicidal ideation in attachment-based family therapy

Executive Summary Statement and Final Scientific Report for BSF Grant 2011279 - Emotional processing, attachment and suicidal ideation in attachment-based family therapy Principal Investigators: Prof. Gary M. Diamond, Ben-Gurion University, Psychology, Beer-Sheva, Israel; Prof. Diamond Guy S., Drexel University, Psychiatry, Philadelphia, PA, USA; and Prof. Kobak R. Rogers, University of Delaware, Department of Psychology, Newark, DE, USA.

The purpose of this study was to examine the role of emotional processing, and physiological arousal during episodes of emotional processing, in transforming attachment schema and decreasing depressive symptoms and suicidal ideation among a sample of 66 suicidal adolescents receiving 16 weeks of attachment-based family therapy. We hypothesized that: 1) Amount (in seconds) of productive emotional processing during the first two reattachment sessions would be associated with improvements in adolescents' attachment schema from pre- to post-treatment; 2) amount of productive emotional processing during the first two reattachment sessions would be associated with trajectories of depressive symptoms and suicidal ideation over the course of treatment; and 3) the positive association between the amount of productive emotional processing and changes in attachment schema, depressive symptoms and suicidal ideation, would be moderated by levels of physiological arousal during such episodes of emotional processing, measured using objective voice parameters.

Method Sample Participants were recruited and treated as part of the parent study housed in the Center for Family Intervention Science at Drexel University. In the context of this NIMH funded randomized clinical trial, the U.S. investigators recruited and collected data from 129 adolescents, ages 12-18 years old (M = 14.87, SD = 1.68), who were randomized to either 16 weeks of attachment-based family therapy (ABFT) (n=66) or family-enhanced non-directive supportive therapy (FE-NST) (n=63). Participants were recruited from hospital primary care centers, emergency departments, outpatient facilities, inpatient units, schools, churches and the general community. The protocol was approved by the Children’s Hospital of Philadelphia and Drexel University. The study was monitored quarterly by a data safety and monitoring board. All participants provided written informed assent and/or consent. Eligibility for the clinical trial included severe suicidal ideation (SIQ-JR>31), and moderate to severe depression (BDI-II>20), measured at two consecutive pre-treatment assessment time points. At least one primary caregiver was required to participate in treatments and assessment. Exclusion criteria included: a) evidence of imminent risk of harm to self or others than could not be safely treatment on an outpatient basis, b) evidence of psychotic features, c) evidence of severe cognitive impairment based on educational records, parent report and/or clinical impression, d) non-English speaking participating parent. Additionally, participants who began psychiatric medication within three weeks of the initial pre-treatment assessment were ineligible for participation.

Fifty-five percent of the sample identified as African American, 31.9% White, 5.2% American Indian or Alaskan Native, 1.7% Asian, 86% Native Hawaiian or Pacific Islander and 11.2% other. The majority of the sample was non-Hispanic/Latino (74.5%). A large majority of the sample was female (81.9%). Only 41.2% met criteria for major depressive disorder, 3.9% met criteria for dysthymia, and 36.3% met criteria for an anxiety disorder on the Diagnostic Interview Schedule for Children (DISC-IV). Forty-two percent of participants had made a suicide attempt in their lifetime and 57.5% reported engaging in non-suicidal self-injury as indicated by the Columbia-Suicide Severity Rating Scale (C-SSRS). For the purposes of this study, we only examined the data from the 66 adolescents participating in the ABFT group. Outcome (BDI-II, SIQ-JR) and attachment (ECR) data was gathered at weeks 0 (baseline assessment), 4, 8, 12 and 16 (post-treatment assessment) by independent assessors naïve to treatment condition.

Procedures and measures After the completion of treatment, the first two Task 4 sessions from each ABFT was selected and coded for productive emotional processing using the Productivity Scale (Auszra, Greenberg & Herrmann, 2010). Only 47 out of the 66 cases could be coded for emotional processing during Task 4. Of the 19 cases that could not be included: seven never actually began therapy, though they had been randomized; two only came to three sessions, and thus never reached Task 4; and another nine cases came to four or more sessions, but the therapy never progressed to the stage of Task 4. Finally, one case did not have any video or voice recordings due to technical reasons, so that their Task 4 sessions could not be analyzed.

Three independent raters were trained to apply the Productivity Scale in order to identify episodes of productive emotional processing within the selected sessions. Raters were undergraduate psychology students, naïve to the purpose of the study. The training consisted of six meetings of two hours each. During the first three meetings, raters were taught about the theory behind emotional processing. During the final three meetings, they were taught to use the Productivity Scale and watched exemplars of productive emotional processing taken from sessions not included in the study. Also, between meetings, the two coders separately and independently coded practice sessions and received personal feedback from the lead trainer. Only after they had coded five practice sessions not included in the study sample, and evidenced sufficient reliability (kappa > 0.60), a kappa considered substantial (Landis & Koch, 1977) and good (Altman, 1991), did they begin coding the selected sessions. Two of the three coders rated each session, with raters being assigned sessions to code on the basis of random, rotating pairs.

Of the 47 cases submitted for observational ratings of adolescents’ productive emotional processing, 13 had only one Task 4 session. Out of the 47 sessions coded in total, 17 were found to have no episodes of productive emotional processing. From the remaining 30 cases found to have had a least one Task 4 session with at least some productive emotional processing, all such segments of productive emotional processing were submitted to voice analyses in order to examine the average rate of speech (syllables per second), fundamental voice frequency range and voice amplitude range during the moments of emotional processing in each session. The average value for each voice parameter was calculated by subtracting the average score of all matched baseline episodes drawn from the same session from the average score of all of the productive emotional processing episodes in that same session. However, because it is impossible to analyze male and female voices together, the three cases with male adolescents were removed and only the 27 cases with female adolescents were ultimately analyzed. Among these 27 cases, three were eliminated because the quality of the voice recording was insufficient, and an additional three were eliminated because there were too few syllables to analyze reliably, leaving a total sample of 21 cases for voice analyses.

Results Reliability estimates for emotional processing ratings The inter-rater reliability estimate for coding productive emotional processing was calculated using Cohen’s kappa statistic. Cohen’s kappa is the most commonly used statistic for estimating the reliability of nominal data and represents an improvement upon percentage of absolute agreement, in that it accounts for chance agreement (McHugh, 2012). Our results produced a kappa of .64 (95% CI, 0.57-0.77, p

Differences between emotional processing and baseline episodes in terms of voice parameters To test whether the voice indices (SPS, F0 range, and amplitude range) actually reflected physiological arousal, we examined whether scores were higher during the emotional processing episodes than during the baseline episodes from the same sessions. To do so, we ran three 2-level multilevel models (episode nested within patient), in which the outcome was one of the three voice indices and the predictor was a dichotomous variable indicating whether this voice index was sampled from a baseline or an emotional processing episode. As expected, clients’ F0 range was higher during the emotional processing episodes than during the baseline episodes (est.=34.79, SE=7.35, p

Changes in depressive symptoms, suicidality and attachment schema over time, and the moderating effects of amount of emotional processing and fundamental frequency range during episodes of emotional processing In order to examine the trajectories of depressive symptoms, suicidal symptoms and attachment over the course of therapy, and whether observer rated emotional processing or physiological arousal during moments of emotional processing moderated such trajectories, we conducted multi-level regression models (MLM), with level 1 as the session level and level 2 as the client level. We used such models to account for the non-independence of our data, since sessions were hierarchically nested within clients (Krull & MacKinnon, 2001, Laurenceau & Bolger, 2012).

Prior research has shown that a log-linear transformation of session numbers can parsimoniously approximate the average pattern of change over the course of treatment (e.g., Gibbons et al., 1993; Stulz et al., 2013). Thus, at level 1, we modeled patients’ outcome (i.e., BDI, SIQ, or attachment anxiety and avoidance for each parent separately) as a function of the log-transformed session number. At level 2, the main effect of the moderator (amount of emotional processing or fundamental frequency range during episodes of emotional processing), as well as the cross-level interaction effect between the moderator and session number, were introduced. In all estimated models, the intercept and the session effect were considered to be random. In addition, to account for autocorrelation between adjacent observations, a first-order autoregressive structure was imposed on the covariance matrix for the within-person residuals.

The generic model’s equations were: Level 1: Outcomesp = β0p + β1p*Log10(Sessionsp) + esp Level 2: β0p = γ00 + γ01*Moderatorp + u0p β1p = γ10 + γ11*Moderatorp + u1p where the outcome at session s of patient p was modeled at level 1 as a function of the patient’s intercept (β0p), this patient’s rate of change during the treatment (β1p), and a level-1 residual term (esp ). At level 2, the patient’s intercept was modeled as a function of the sample’s intercept (i.e., fixed effect; γ00), the main effect of the moderator (γ01), and a level-2 residual term quantifying patient-level deviation from these effects (u0p); finally, the patient’s session slope was modeled as a function of the sample’s slope (γ10), the moderator effect on the patient’s slope (γ11; i.e., cross-level interaction effect), and a level-2 residual term quantifying patient-level deviation from these effects (u1p).

The results of the MLM models are presented in the table below, with the top section of the table reflecting models in which amount of emotional processing served as the moderator, and the bottom section of the table reflecting models in which fundamental frequency range during episodes of emotional processing served as the moderator. As can be seen in this table, clients’ BDI and SIQ scores showed a decline over the course of treatment. Clients’ attachment scores, in contrast, showed no changes over the course of therapy. Neither amount of emotional processing, nor fundamental frequency range during episodes of emotional processing, moderated the trajectories of clients’ BDI, SIQ, or attachment scores.

Outcome: BDI SIQ M_Avd M_Anx F_Avd F_Anx Est.(SE) p Est.(SE) p Est.(SE) p Est.(SE) p Est.(SE) p Est.(SE) p Intercept 1.796 (0.057)

Discussion This study examined the role of productive emotional processing, and physiological arousal during episodes of productive emotional processing, in attachment-based family therapy for depressed and suicidal adolescents. Results from the randomized clinical trial, from which the participants in this study were drawn, indicated that adolescents receiving ABFT, on average, evidenced significant decreases in depressive symptoms and suicidal ideation over the course of 16 weeks of therapy (Diamond et al., under review). This study was designed to examine whether the amount of productive emotional processing occurring during Task 4 of therapy, the reattachment task, and the level of physiological arousal during such productive emotional processing, moderated these effects.

Our findings suggest that neither amount of productive emotional processing during Task 4, nor the level of arousal during such moments of processing, moderated trajectories of changes in depressive symptoms, suicidality or attachment anxiety or avoidance over the course of the treatment. These findings are contrary to our hypotheses. They are also contrary to findings from previous research we conducted on ABFT for young adults with unresolved anger. In that previous study, we found that greater amounts of young adults’ productive emotional processing during attachment/identity episodes predicted greater decreases in psychological symptoms (Diamond et al., 2016), and that young adults themselves attributed positive treatment outcomes to being able to productively process their emotions, and to being vulnerable in the presence of their parents (Steinmann et al., 2017).

Our findings are also inconsistent with reports from prior studies demonstrating the positive role of emotional processing across a wide range of individual psychotherapeutic approaches, including cognitive-behavior therapy (Foa et al., 2006), psychodynamic therapy (Diener et al., 2009) and emotion-focused therapy (Greenberg, 2010). For example, in one study of emotion-focused therapy and client-centered therapy for depressed clients, increases in expressed emotional arousal from early to mid-therapy, and levels of mid-therapy emotional arousal, were found to be correlated with treatment outcome, particularly when such emotional processing occurred in conjunction with reflective processing (Missirlian, Toukmanian, Warwar & Greenberg, 2005). Along the same lines, Greenberg, Auszra, and Herrmann (2007) found that the arousal of primary adaptive emotions discriminated between better and poorer outcome cases, while simple emotional arousal per se did not. In yet another study, Auszra, Greenberg and Herrmann (2013) found that productive emotional processing during the working stage of therapy predicted symptom reduction, above and beyond both high emotional expression and the quality of the therapeutic relationship. In a study comparing cognitive-behavioral and process-experiential therapy for depression, Watson and Bedard (2006) found that higher levels of emotional processing were associated with better outcome across treatment groups and that, on average, clients receiving process-experiential therapy evidenced higher levels of emotional processing than those receiving cognitive-behavioral therapy. More recently, Kramer, Pascual-Leon, Despland & de Roten (2014) examined the link between emotional processing and outcome in short-term dynamic therapy for individuals presenting with adjustment disorder with depressive mood. They found that the amount of adaptive grief/hurt expressed by clients distinguished between good and poor outcome cases.

There is no ready explanation for our non-findings. One possibility is that, unlike individual therapy, emotional processing in and of itself is not a primary change mechanism in ABFT. Instead, in ABFT, corrective attachment is the result of both emotional disclosure and subsequent benevolent, validating, supportive parental responses. It could be that, since in this study of family therapy emotional processing may have been followed by either positive or negative parental responses, the effects of such processing are context dependent. With that said, we did find in one prior study that productive emotional processing was predictive of outcome in ABFT for young adults, above and beyond quality of parental response. Another possibility is that for this specific client population – severely depressed adolescents - productive processing was less important than the effects of being attended to by therapist, parent and the research team. The inner workings of the therapy may have also been overshadowed by actual events in these kid’s lives, including positive changes in school setting, family setting, etc. that occurred as a result of the therapy. More research examining the role of emotional processing with this population are required.