Change in Emotional Arousal Variance in Clients with Borderline Personality Disorder during a Brief Psychiatric Treatment Open Access

Although somewhat dated, Kazdin’s [1] observation that we still lack an evidence-based explanation for how or why psychotherapy produces change remains valid. As a result, we are left with treatments that are effective, yet largely unspecific, and few strategies to optimize them. To address this gap, research must identify and study the core active ingredients of psychological interventions. One promising candidate is emotional processes [2, 3] – particularly emotional arousal (EA), which has been linked to both the experience of psychological distress and its resolution in therapy [4].

EA can be defined as a combination of expressive behaviours (e.g., postures, gestures, facial and vocal expressions) and physiological responses (involving the somatic, autonomic, endocrine, and immune systems) triggered by emotionally arousing stimuli [5]. While the importance of EA in therapy is often discussed, findings on its exact role remain mixed. Some studies using validated observer-rated tools to assess patients’ in-session EA suggest that both too little and too much arousal may hinder the effective processing of distressing emotions [6‒8]. These findings support the idea of an inverted-U relationship between emotional intensity and therapeutic outcomes, rather than a simple linear one [9].

Clinically, this concept aligns with Siegel’s “window of tolerance” [10]. When individuals operate within this window, they are better able to manage their emotions. Outside of that, they may experience hyperarousal (e.g., anxiety, anger, overwhelm) or hypoarousal (e.g., numbness, detachment). In borderline personality disorder (BPD), where dysregulation of EA is considered a core feature [11], helping patients stay within this window has been linked to symptom reduction [12]. Yet, many authors argue that it is not only the intensity of EA, but also how it is processed once activated, that facilitates change [13].

According to the interpersonal sensitivity hypothesis of BPD [14], individuals with the diagnosis experience heightened, prolonged, and easily triggered emotions due to intense needs for closeness combined with fears of rejection [15‒17]. Improving emotion regulation – and thereby reducing excessive EA – has been shown to alleviate symptoms [18]. For instance, Goodman et al. [19] found that emotion regulation skills improved through Dialectical Behaviour Therapy (DBT), with corresponding reductions in symptoms. Similarly, McMain et al. [20] highlighted the importance of emotion regulation mechanisms in the successful treatment of BPD. Kramer et al. [21] also found that BPD patients experienced a moderate decrease in subjective arousal during brief treatment, which was associated with symptom improvement.

Recent reviews emphasize that psychological treatments should be the first-line approach for BPD [22]. Among these, brief interventions have shown promise in quickly alleviating acute symptoms such as non-suicidal self-injury (NSSI) and suicidality [23‒25], and some studies report they are not inferior to longer-term treatments [26, 27]. One such intervention is Good Psychiatric Management (GPM) [28], which focuses on patients’ interpersonal hypersensitivity [29]. Kramer and colleagues [30] found GPM to be more effective than treatment as usual in improving relationship problems, impulsivity, and treatment retention.

In their study, Kramer et al. [30] assessed emotional processing using the experiential Two-Chair Task (TCT), which focused on self-criticism. Emotional responses were coded using the Classification of Affective Meaning States (CAMS) [31], which distinguishes nine emotion types and organizes them by degree of transformation using the Degree of Transformation Scale [32]. They found that early changes in emotional processing – assessed outside of sessions – predicted improvements in relationship functioning, highlighting the importance of EA as a mechanism of change in BPD treatment.

However, studying EA poses several methodological challenges. It can be measured through various means (physiological, self-report, observer-rated) and conceptualized in different ways (e.g., peak, mean, variance). EA is also influenced by context, stimuli, and the measurement method itself. Pascual-Leone et al. [33] argue for individualized research designs, emphasizing the importance of the emotional response’s validity over the stimulus itself. They advocate using idiosyncratic stimuli – rather than standardized sets like validated images – to better capture relevant emotional reactions [34].

Following this line of work [33, 35], we adopted a standardized procedure based on the TCT to elicit and assess emotional responses to the idiosyncratic stimulus of self-criticism [36]. This approach builds on methods described by Kramer et al. [30, 37]. In line with current recommendations for measuring emotion [38], we used a multimethod approach, combining self-reported and observer-rated measures. Given that mean EA alone does not capture the complexity of emotional responses, we focused on variance as a key indicator. This measure reflects the range of fluctuation from the mean and serves as a useful proxy for emotional dysregulation – capturing both extremes of the emotional rollercoaster.

The group of participants diagnosed with BPD will show more (a) self-reported and (b) observer-rated EA variance during the individualized emotion-focused task at the beginning of the assessment compared to the control group.

Participants diagnosed with BPD will see a significant decrease in (a) self-reported and (b) observer-rated EA variance over time compared to those in the control group.

Participants diagnosed with BPD in the GPM subgroup will see a significant decrease in (a) self-reported and (b) observer-rated EA variance over time compared to those in the TAU subgroup.

This decrease in the GPM subgroup on (a) self-reported and (b) observer-rated EA variance predicts a reduction in BPD and other psychopathological symptoms between the beginning and the end of the brief treatment.

The present article is framed within a greater RCT (study register number NCT0317818) that investigated the effectiveness of a brief psychiatric treatment (GPM) for the treatment of BPD compared with an equally brief nonspecific psychiatric treatment (that is not focused on BPD), as well as the changes in neurofunctional activation in networks associated with emotion and sociocognitive processing (for details and the complete methodology see [37]).

The total sample (N = 105) consisted of healthy controls (HC; n = 29) and clients with a diagnosis of BPD (N = 76, randomized into GPM or TAU). Twenty-one participants were excluded because of missing data resulting from a final sample of N = 84 (HC; n = 29, GPM group: n = 28; TAU group: n = 27). Participants in the BPD group (n = 55) were recruited from clients seeking treatment at a university outpatient clinic. Inclusion criteria were as follows: (1) must be aged between 18 and 65 years old; (2) must have a diagnosis of BPD according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5; [39]); and (3) must have no additional diagnoses of neurocognitive disorders, psychosis and bipolar disorder 1. The diagnosis process was carried out by psychiatrists or psychologists using the Structured Clinical Interview for DSM-5 Personality Disorders (SCID-5-PD). The clients’ ages ranged between 19 and 54 years old with a mean of 29.33 (SD = 8.33; see Table 1).

The participants from the control group were recruited to assess baseline differences and did not receive therapy. Recruitment was conducted through advertisement flyers distributed across two universities as well as through convenience sampling. Inclusion criteria were as follows: (1) aged between 18 and 65 years old; and (2) no diagnosis of a mental disorder. Participants reported no current DSM diagnosis, which was further confirmed through a clinical interview conducted by a psychologist. While recruitment was open to all eligible individuals, only participants who self-identified as female chose to enrol in the study, resulting in a final sample of 29 participants (100% female). Their ages ranged between 20 and 31 years old, with a mean of 22.62 (SD = 2.60).

A total of N = 28 therapists participated in the trial. They were psychiatrists (n = 23) and psychologists (n = 5) with a mean of 3 years of clinical experience (range was between 1 and 20 years of experience). All the therapists received at least 1 day of intensive workshop in GPM, as well as online material offered to learn GPM principles. They also all received weekly supervision by certified GPM trainers, which guarantees the quality control of GPM. Therapists all intervened in both treatment conditions (GPM and TAU) and received explicit instruction for both.

Brief psychiatric treatment followed the principles of GPM [28, 40] and encompassed communication about the BPD diagnosis and co-morbidities, the elaboration of links between problem areas, interpersonal hypersensitivity, the work on treatment focus and the treatment of treatment-interfering problems. The first session involved specifically exploration, clarification of problem areas and background information. In keeping with the GPM protocol, psychotropic medication was allowed under certain circumstances, according to national guidelines [41]. The last session (after 4 months) involved the synthesis of the main achievements and a decision on next steps in the treatment program. Patients who required more treatment were referred to (a) additional GPM, or (b) evidence-based psychotherapy programs.

For the TAU, nonspecific crisis management as usual, safety management, and patient contact, in accordance with the minimal ethical standards and following the clinic-internal guidelines of good medical practice, were proposed. Supervision was offered and supervisors received detailed instructions about what areas to focus that are consistent with these internal guidelines of good medical practice. Psychotropic medication was allowed. Discussion about the BPD diagnosis was prohibited. Discussion about the links between problem areas and the patient’s interpersonal hypersensitivity, and the discussion of any interpersonal factors within BPD, were prohibited.

Therapist adherence to GPM principles was assessed using the self-assessment by therapist using Gunderson’s (2016; personal communication) questionnaire for adherence to GPM. This questionnaire was given to therapists in both conditions, in the end of each treatment. The therapists self-reported in the GPM condition an average score to of adherence to GPM principles of 73.21 points (SD = 12.79). The therapists self-reported for the TAU condition an average adherence to GPM principles of 53.77 (SD = 11.51). This difference is statistically significant (t = 4.15, p = 0.001, d = 1.60; [42]).

The TCT (also known as two-chair dialogue) is an individualized therapeutic intervention from emotion-focused therapy designed to increase EA and emotional processing and resolve self-criticism [35, 43‒45]. In this study, the TCT was used as an assessment tool during videotaped out-of-therapy assessment sessions, that were then coded with the Client Expressed Emotional Arousal Scale (see below).

During this task, the participant is first invited to imagine a personal situation of failure in their life as vividly as possible (without reporting verbally). Then, the researcher asks them to change chairs. On this new “self-critical” chair, participants should adopt the stance of the inner self-critical voice and express self-criticism to the self, as imagined on the initial chair. Finally, the participant (who is now back again in the initial chair) describes their current emotional reaction to the self-criticism (for a complete description of the two-chair dialogue used in research, see [44]).

The relationship between the imagination task and the TCT is sequential and complementary:

• 1.

  The imagination task establishes the emotional context by activating self-critical schemas and arousal through recall of a past failure.

• 2.

  The TCT builds upon the emotional activation elicited during the imagination task, facilitating the expression, exploration, and potential transformation of self-critical emotional states.

Thus, the imagination task serves a crucial role in preparing participants for the two-chair dialogue by providing an individually relevant and meaningful stimulus. This approach is grounded in prior research [35], which emphasizes the importance of eliciting personally significant emotional material to ensure ecological validity and enhance the relevance of the TCT.

The ZAN-BPD [46] is a clinician-administered measure used to assess BPD severity. Each item reflects one of the nine DSM-5 BPD criteria and is rated from “no symptoms” = 0 to “severe symptoms” = 4. Total scores range from 0 to 36 with higher scores indicating greater BPD severity. The ZAN-BPD is a widely used measure and has demonstrated its validity and reliability [46]. Training in the use of this scale for this study was provided by its developer.

The Outcome Questionnaire 45 (OQ-45) [47] is a self-report questionnaire comprising 45 items assessing symptom distress (or subjective discomfort; intrapsychic functioning with an emphasis on depression and anxiety), interpersonal relationships (loneliness, conflict with others and marriage and family difficulties) and social role (difficulties in the workplace, school or home duties). Total scores range from 0 to 180 with a clinical cut-off at 64 and above. The OQ-45 has been translated and validated in French [48]. The reliability of the OQ-45 was assessed using Cronbach’s alpha, which yielded a value of 0.83, indicating good internal consistency.

The Self-Assessment Manikin (SAM) [49] is a self-assessed questionnaire using a single item to measure the momentary level of arousal using a 9-point Likert scale, ranging from “not excited at all” (1) to “very excited” (9). The scale is illustrated as a series of human-shaped figures displaying varied levels of activation. It is widely used in emotion research and has proven its validity and reliability [50].

The Client Expressed Emotional Arousal Scale-III (CEAS-III) [6] is a standardized, 7-point observer-rated assessment of the intensity of observable, expressed emotional intensity, including levels of affect and emotional restriction displayed verbally and nonverbally. The higher levels of the scale indicate higher EA intensities whereas the lower ones suggest emotional restriction. Based on these criteria, each defined amount of time of the selected recording (in our study, these were intervals of 2-min) is assigned one of seven ordinal ratings of expressed arousal, ranging from no EA (1) to extreme EA (7). Six raters assessed both the modal (most frequent) and peak (highest) levels of intensity of the client’s expressed EA. The raters did consider participants’ baseline EA by observing them during the Relationship Anecdotes Paradigm (RAP) interviews that were also part of the psychological assessment. The raters in our study were two Ph.D. students in clinical psychology, one master-level psychologist and three master-level psychology students, all extensively trained in the use of the scale. In order to ensure sufficient inter-rater agreement, prior to coding the study’s material, the raters practiced on other videos. The two-way mixed effects, single rater, absolute agreement intra-class coefficient correlation between four raters on 59 cases was 0.820 with a 95% confidence interval from 0.714 to 0.893 (F(43,129) = 5.841, p < 0.001) which can be interpreted as good [51].

The assessment was recorded and lasted approximately 30 min (shown in Fig. 1). Two PhD students in clinical psychology (one being the first author of this article) and two master’s level psychologists were responsible for the data collection. They were all trained in the use of the TCT by an emotion-focused therapy licensed PhD level clinical psychologist. Prior to participating, each participant provided informed consent.

Before coming to the session, participants were asked to fill an online or paper and pencil OQ-45. (1) At the beginning of the session, participants filled out a SAM. (2) Then, they were asked to think about a personal meaningful situation of failure. (3) Once the imagination task was completed, participants were asked to fill out the SAM again. Then, participants were invited to sit in a new chair in front of them and imagine themselves still sitting in the first one. We asked them to voice their own inner critic (4) before asking them to fill out a SAM (5) for the last time.

After the TCT, the research conducted a ZAN-BDP with the participant. The recorded session was used by the raters to code the observed peaked EA during the experiential task by 2-min chunks with the CEAS-III. Each participant’s session was coded by at least two raters whose scores were then combined. In the end, each participant’s session had 2-min segments of combined scores ranging from 1 to 7.

The design of the study includes three sessions that follow the outlined protocol. The first session (T1) served to establish a clinical baseline at the start treatment. The second session (T2) took place at mid treatment (after 2 months) and did not include the control group. The final third session (T3) took place at the end of the brief treatment for participants in the BPD group and after 2 months for the HC and captures treatment outcomes.

Excluding the non-existent T2 for controls and 7 missing observations from sessions with participants diagnosed with BPD who did not complete all measurements, the final dataset consists of 216 individual study out-of-therapy sessions (described in Fig. 1) from 84 participants. After some descriptive and preliminary analyses, we first predict EA variance on the self-reported (SAM) and observer-rated (CEAS-III) measures (see H1, H2, H3).

This approach was intended to capture within-subject variability in EA across multiple time points. We chose to focus on variance as an index of within-subject variability, which reflects fluctuations in EA but does not account for temporal dependency or directionality. Our primary focus was on comparing EA variance between groups, with variance calculated as described above and aggregated for analysis. The inclusion of within-subject variability was intended to provide insight into individual emotional dynamics within sessions, while the between-group comparisons allowed us to examine broader patterns.

For H1, we fit separate multilevel OLS regression models with the study group (0 = control; 1 = BPD) and session (0 = T1, 1 = T2, or = T3) as dummy predictors along with their interaction terms. We cluster standard errors at the participant level for all analyses to correct for potential within-subject correlations in responses. To adjust for differences in the extent of participants’ EA and psychological distress before each session, we include participants’ initial score of the OQ45 and their baseline measures as statistical controls. We test H2 and H3 by repeating the same model while replacing the study group dummy with two separate dummy variables for the treatment arms (0 = control, 1 = GPM or TAU). For H4, we then fit the same models predicting participants’ borderline and other psychopathology symptoms (ZAN-BPD, OQ45), adding the EA as separate measures. All analyses were conducted with the statistical software R (version 4.2.1).

The control group consisted entirely of female participants (100%), whereas the treatment groups (TAU and GPM) had 67% and 75% female participants, respectively. Statistical testing revealed no significant differences in gender distribution between the TAU and GPM groups (χ² = 0.008, p = 0.93).

Regarding age, the mean age of the GPM group (M = 34.46, SD = 8.60) was slightly higher than that of the TAU group (M = 31.22, SD = 7.57), but this difference was not statistically significant (t = 1.50, p = 0.14). Thus, no significant age differences were observed between the treatment groups, and age does not need to be included as a covariate in subsequent analyses. These results suggest that gender and age were not confounding factors between the treatment groups and, therefore, do not require statistical adjustment in our analyses. Participants in the BPD group were on average older than in the control group, t(54) = 6.38, p < 0.001. As expected, the BPD group (M = 83.8, SD = 29.1) showed significantly higher scores of psychological distress at T1 than participants in the control group (M = 46.2, SD = 20.6) as measured by the OQ-45, t(68) = 26.8, p < 0.001). The BPD group (M = 14.6, SD = 5.8) also showed significantly higher scores of borderline and other psychopathology symptoms at T1 than participants in the control group (M = 1.9, SD = 1.8) as measured by the ZAN-BPD, t(60) = 13.7, p < 0.001 – they specifically score higher on the relationship subscale (M = 3.06, SD = 1.88) than those in the control group (M = 0.21, SD = 0.41), t(53) = 10.1, p < 0.001.

At baseline, the mean ZAN-BPD score for the GPM subgroup (M = 16.04) was higher than that of the TAU subgroup (M = 13.16), as was the mean OQ45 score for the GPM subgroup (M = 89.42) than that of the TAU subgroup (M = 78.14). These differences between groups were significant on the ZAN-BPD (t(45) = 1.726, p = 0.046) but not on the OQ45 (t(40) = 1.271, p = 0.105). To assess the magnitude of the difference between groups on the ZAN-BPD, we calculated the effect size using Hedges’ g to adjust for small sample sizes, resulting in a value of g = 0.495. This indicates a moderate effect between the GPM and TAU subgroups, suggesting that the GPM group started with slightly higher symptom severity on the ZAN-BPD scale, although the effect size is modest. Despite this statistical difference, both the GPM subgroup mean of 16.04 and the TAU subgroup mean of 13.2 points are within the expected range of clinical populations (M = 14.3, SD = 6.8, [46]).

The correlation between EA variance of the observer-rated CEAS-III and the self-reported SAM measures was not significant, r(82) = 0.12, 95% CI [−0.03: 0.26], p = 0.13). A comparison of mean levels of EA – not its variance – in our sample shows little difference over time (see Table 2). Pairwise t tests show no significant difference in mean self-reported EA between T1 and T3 in either the TAU group, t(26) = 0.23, p = 0.81, or GPM group, t(26) = 0.53, p = 0.60.

We first test whether participants in the BPD and control group significantly differ at the beginning of the study (T1). In line with our expectation (H1), participants in the BPD group showed significantly higher variance on both the self-reported (b = 2.20, 95% CI [0.47: 3.92], p = 0.013) and the observer-rated (b = 0.22, 95% CI [0.07: 0.37], p = 0.005) EA measures than the control group (See Table 3). These differences in EA variance are no longer significant at 4 months (T3; shown in Fig. 2).

We next assess how EA variance changes over time, expecting a decrease for participants in the BPD group. On the one hand, we find evidence of such a decrease in EA for the self-reported measure (H2a), indicated by a significant time-by-condition interaction term (b = −1.48, 95% CI [−2.90: −0.07], p = 0.023). Panel A of Figure 2 illustrates that this decrease takes place mostly after T2. On the other hand, this pattern does not replicate for the observer-rated EA measure (H2b).

More specifically, our third hypothesis (H3) posited that this decrease in EA would be larger for participants in the GPM treatment group compared to those in the TAU group. Indeed, Figure 3 suggests that a significant portion of decrease in EA variance for people with diagnosed BDP can be accounted for by the GPM treatment group. At T3, participants enrolled in the GPM treatment – but not those in the TAU group – showed significantly lower self-reported EA variance (H3a; b = −2.34, 95% CI [−4.49: −0.19], p = 0.033; see online suppl. Table A1; for all online suppl. material, see https://doi.org/10.1159/000546284 in the supplementary material).

Finally, we assessed the role of EA variance in predicting borderline and other psychopathology symptoms. We posited that lower EA variance would result in reduced symptoms for participants in the GPM treatment group specifically (H4). Focussing on our composite symptom measures (OQ45, ZAN-BPD), we find no evidence to substantiate this expectation as the results show no significant associations between neither of the EA variance measures and symptoms (see online suppl. Table A2 in the supplementary material). Moreover, repeating the model on the ZAN-BPD relationship subscale did not result in a significant association either (see online suppl. Table A3 in the supplementary material).

We further explored this relationship by fitting the model with fixed overall intercepts and only a three-way interaction term between EA variance and the session as well as dummy coded treatment group variables. We then probed the interaction by calculating partial derivatives for the effect of EA variance between groups and over time. The results are depicted in Figure 4 below. EA variance negatively predicts symptoms (measured by OQ45) for the control group at both time points and for both measures (see online suppl. Table A4 in the supplementary material).

The evidence is less clear for the treatment groups. At T1 we find that EA variance plays the opposite role, with self-reported EA variance positively predicting BPD-specific symptoms (indicated by ZAN-BPD) in both treatment groups (TAU: b = 0.84, SE = 0.30, p = 0.005; GPM: b = SE = 0.21, p = 0.001). There is association between self-reported EA variance and OQ45 in either treatment group. The same pattern of effects emerges for observer-rated EA variance, which positively predicts symptoms in both treatment groups and for both symptom measures, though the effect is not significant in the TAU group for the OQ45 measure.

At T3, we find significant effects self-reported EA variance on symptoms. Like at T1, observer-rated EA variance is positively associated with both symptom measures in the TAU group. However, in the GPM group, the role of EA variance changes at T3: participants with higher observer-rated EA variance had significantly fewer symptoms on the ZAN-BPD (b = −6.10, SE = 2.91, p = 0.038).

To enhance the effectiveness of psychological treatments, it is essential to identify their active ingredients. Among emotional processes, EA has emerged as a particularly promising candidate, especially in the context of BPD. Individuals with BPD are thought to experience interpersonal hypersensitivity, which directly influences their EA [28].

In this study, we aimed to assess EA in an individualized yet controlled way. Participants were asked to reflect on a personal failure and then engage in a TCT to verbalize their inner critic. This design allowed us to observe how EA evolved throughout a brief intervention. Guided by existing literature suggesting that therapeutic change occurs within an optimal “window” of EA, we addressed three key methodological challenges: (1) we used a multimethod assessment combining self-report and observer-rated measures; (2) we ensured emotional responses were personally meaningful by employing an individualized stimulus within a standardized setting; and (3) we focused on variance in EA – rather than its mean – as a marker of emotional dysregulation.

First, individuals with BPD exhibited significantly greater variance in EA at baseline compared to controls (supporting H1a and H1b). This suggests that people with BPD experience more extreme fluctuations in arousal during emotionally evocative tasks, reflecting difficulties in regulating emotional states – a finding consistent with prior research on the central role of EA in BPD.

Second, over the course of the intervention, EA variance in the BPD group decreased, as measured by self-report (supporting H2a), though not by observer ratings (H2b). By the end of treatment, BPD participants’ self-reported EA variance was statistically indistinguishable from that of controls. This normalization was observed only in the GPM group – not in the TAU group – supporting H3a but not H3b. These results suggest that GPM may foster improved emotional regulation in individuals with BPD, as reflected in their subjective experience of EA.

Third, although BPD participants in the GPM group showed significant symptom reduction on the OQ-45 between T1 and T3, this improvement was not predicted by changes in EA variance. Neither self-reported nor observer-rated EA variability emerged as significant predictors of symptom change (no support for H4a or H4b). Thus, our data do not support EA variance as a mechanism of change, at least in the current sample.

The discrepancy between self-reported (SAM) and observer-rated (CEAS) measures of EA warrants further discussion. The reduction in self-reported EA in the GPM group may reflect enhanced emotional awareness, appraisal, and regulation skills. In contrast, the lack of observable change on the CEAS could reflect the fact that external expressions of arousal do not always align with internal states – particularly in individuals who have developed regulation strategies that suppress or mask their emotional expression. Some participants may have experienced high arousal internally but appeared composed externally, leading to underestimation by raters. Moreover, despite its structured nature, the CEAS remains subject to rater interpretation. Future research could incorporate voice analysis tools [52, 53] to enhance objectivity and better capture the multidimensional nature of EA.

Finally, it is worth noting that prior studies have suggested brief interventions for BPD may be especially effective in younger, more severely impaired individuals who do not primarily struggle with interpersonal difficulties [54]. Due to limited power, we were unable to control for these variables, but it is possible that EA variance predicts symptom change only in this specific subgroup.

This article fulfils two capital requirements for demonstrating mechanism of change [1], namely, the use of the TCT in a quasi-experimental design as well as the timeline by assessing the participants several time (three for those with a BPD and twice for the controls).

Despite the focus on a multimethod measurement of EA that does not solely rely on a self-report scale, the lack of a third measure (e.g., a physiological measure) hinders the assessment validity and conditioned our results. Our approach captures variance and does not measure instability per se. Future studies could address this limitation by incorporating metrics specifically designed to assess instability over time. The small sample size and the gender disparity prevents any generalization. Also, while the use of SAM and CEAS-III can capture within-subject variability of EA in the context of a TCT, change in EA across repeated administrations of the task might not be reliable indicator of change. The different structures of the SAM and CEAS-III (self-reported vs. observer-rated, three time points vs. a variable number of 2-min coded intervals, coding EA outside of the two-chair dialogue vs. coding EA only during the two-chair dialogue) may also account for differences in results. Our modelling approach to focus on comparability between SAM and CEAS-III has limitations, particularly in terms of fully capturing the complexity of observer-rated EA variability. Future research should explore more sophisticated methods for modelling dependency and directionality such as a relative variability index, iSD [55], or mean square successive difference MSSD [56].

However, by investigating a sample with healthy controls, our article allows us for some important insight on the relationship between EA and psychological distress not exclusively in a clinical setting. Furthermore, the very same sample size that hinders generalization works also as a strength. Indeed, significant results on such small numbers hint at the presence of a robust effect at the risk of overlooking potential smaller effects.

Whereas these findings do not support the hypothesis that a decrease in EA variance is directly associated with reduced symptoms, they suggest that EA variance may have some relevance in the context of GPM treatment.

Individuals with BPD experience more EA than controls during an individualized emotion-eliciting task. By the end of the brief intervention, participants in the GPM subgroup reported a symptom reduction and their EA fluctuation during the task had normalized overtime to become indistinguishable from the controls.

Taken together, these results confirm that EA is a key concept in the clinical presentation as well as the treatment of BPD. Further research using the criteria to investigate mechanisms of change is needed to clarify the relationship between EA, symptom reduction and BPD.

With respect to the original submitted project on the SwissEthics platform, the competent Ethics Committee, has approved the study (2017–02167). The study is registered (NCT0317818). In keeping with the established, and approved, Data Management Plan, only anonymous data will be kept in the file. All video raw data, where the patient may be identified will be stored separately from patient identifiers and from the main dataset. Written informed consent was obtained from participants to participate in the study.

Ueli Kramer was guest editor of the Topical Article Collection “A Mechanistic Approach to Psychotherapy” at the time of submission. The authors have no further conflicts of interest to declare.

This trial is supported by the Swiss National Science Foundation (SNSF100014_179457/1, to Pr. Kramer). The study was designed independently from the funding source, and the data collection, analysis and interpretation are not influenced by the funding source, in keeping with the regulations of SNSF.

L.G.: major contribution to design, data acquisition, coding, analysis and interpretation, as well as writing and revising article; T.R.: major contribution to data analysis and presentation; A.G., F.L., L.A., and J.B.M.: major contribution to data acquisition and coding, I.C.: significant contribution to data analysis, coding, and revision; H.B.: major contribution to data acquisition; S.K.: contribution to revising article; and U.K.: major contribution to revising article and significant contribution to interpretation.

Data are readily available upon request to the first author (L.G.).