Introduction

Borderline personality disorder (BPD) is a serious psychiatric condition associated with instability in affect and identity, significantly impaired interpersonal relationships, and self-injurious behaviors [1–3]. Though initially considered a disorder limited to adults, recent longitudinal and genetic findings establish BPD in adolescents as a valid diagnostic entity, warranting empirical attention [4–6]. Much like the adult counterpart, BPD in adolescence is a severe disorder associated with functional impairment [7], low life satisfaction [8], and a disproportionately large percentage of admissions to inpatient psychiatric facilities [9].

There has been a recent effort to study borderline personality traits (rather than diagnostic status) among community and at-risk adolescents [10–12]. Indeed, a debate about dimensional versus categorical representations of personality pathology is ongoing, representing an issue of particular relevance given the recent release of the newest Diagnostic and Statistical Manual in 2013 [2]. A categorical approach assumes that BPD is a discrete syndrome, clearly delineated from other psychiatric disorders as well as normal personality functioning [1, 13, 14]. A dimensional, trait-based perspective suggests instead that BPD traits may be distributed continuously within the normal population and may denote psychopathology only at the severe end [15–17]. Thus, a dimensional perspective allows for the examination of borderline personality traits and their relation to other constructs across the full latent trait of borderline pathology and across varying levels of severity.

Several influential theories explain the developmental trajectory of BPD as a complex transaction between environmental and biological risk factors [17, 18]. For example, Linehan’s biosocial theory of BPD emphasizes the centrality of emotional vulnerability in the development of the disorder. She posits that children with a genetically based, emotionally sensitive, and reactive temperament are at an elevated risk of developing BPD when reared in an invalidating (e.g., neglectful, abusive, and/or dismissive) early environment. These environments fail to impart adequate emotion management skills, so the individual often resorts to short-term avoidance strategies (e.g., self-harm, impulsive behavior) when experiencing unpleasant internal states. This process of avoidance of internal thinking and feeling states is also identified as a risk process for the development of BPD in Fonagy’s [19, 20] mentalization-based account of BPD. In this account, an insecure attachment relationship between primary caregiver and child results in avoidance of internal thinking and feeling states in the child because acknowledging such internal states would further threaten an already vulnerable relationship with the caregiver. In turn, the child is at increased risk of developing BPD.

A construct that accurately captures the tendency to avoid internal thinking and feeling states (highlighted in both of the aforementioned developmental models) is experiential avoidance (EA)—a construct that was developed within the context of acceptance and commitment therapy (ACT; [21]). EA refers to an “unwillingness to remain in contact with uncomfortable private events (e.g., thoughts, emotions, sensations, memories, urges)” that often manifests in behaviors that serve to avoid unpleasant experiences [p. 1154, 21]. Typical EA behaviors include thought suppression, denial, self-distraction, substance abuse, and self-injury [22]. While these behaviors alleviate distress in the short term, avoidance of unpleasant thoughts and sensations actually increases the likelihood of experiencing them again in the future, elevating physiological arousal and distress [23]. This sets into motion a vicious cycle of using more avoidance-based strategies, thereby thwarting healthy and effective problem solving. Repeated long-term engagement in EA is posited to have damaging effects such as high rates of psychopathology, high negative affect and low positive affect, high emotional reactivity, as well as a number of problem behaviors [e.g., substance abuse, self-injury, and eating disorders; 22, 23].

As suggested by Fonagy’s mentalization-based account of BPD [19, 20] and Linehan’s biosocial theory of BPD [18], research has already identified a reliable link between EA and BPD features [22]. For example, Iverson et al. [24] found that EA uniquely predicted BPD features when controlling for depressive symptoms and emotion dysregulation, signifying that EA may uniquely underlie BPD symptom severity. Similarly, Gratz et al. [25] found that, compared with individuals with no personality disorder, BPD outpatients were less willing to allow the experience of emotional distress in the pursuit of goals. Additionally, Bijttebier and Vertommen [26] reported that individuals with BPD engaged in more avoidant coping strategies compared to controls. Taken together, this evidence suggests that EA is an important correlate of BPD.

Recently, an interest in extending research on EA to youth samples has emerged, likely because EA is implicated in developmental models of BPD like Fonagy’s [19] and Linehan’s [18]. The only existing measure of EA in youth is the Avoidance and Fusion Questionnaire for Youth [AFQ-Y; 27], a self-report measure that assesses EA in children and adolescents. The AFQ-Y was adapted from the Acceptance and Action Questionnaire [28], an extensively researched and widely accepted adult EA measure. Greco et al. [27] developed and validated the AFQ-Y through a comprehensive, multiphasic, statistic-, and theory-based approach in a large community sample. The results of this evaluative process yielded support for the construct and convergent validity of the AFQ-Y in youth as well as adequate internal consistency (α = 0.90).

To our knowledge, only one empirical study has investigated EA in relation to borderline pathology in youth. In adolescent inpatients with BPD, Schramm et al. [29] found that EA made a significant and independent contribution to the variance in borderline features, while partially mediating the relation between difficulties in emotion regulation and borderline features. This study is important in providing evidence that EA may be an important feature in adolescent BPD—as has been documented in the aforementioned adult studies. However, given that this study utilized an inpatient sample, the relation between EA and BPD across the full latent trait of borderline pathology was not examined, suggesting the need for studying these relations in a community sample. Additionally, the sample was not demographically diverse, and thus, additional research is needed. Furthermore, potential confounding variables were not considered. For example, depression, anxiety, and sex are associated with BPD and EA [23, 30], but these constructs have not been investigated simultaneously in the same sample, precluding a thorough understanding of the relations between BPD and EA. Most importantly, no longitudinal work has ever been conducted to examine the predictive role of EA for borderline features over time.

Against this background, the current study had two aims. First, we sought to examine the concurrent relations between EA and borderline features in adolescents in a large and diverse community sample. When examining this relation, we also aimed to control for known correlates of EA (depression, anxiety, and sex) that may confound the relation between EA and borderline features. Second, we examined the prospective relation between EA and borderline features over a 1-year follow-up period, controlling for the baseline level of borderline features. Positive findings would support the notion that the presence of EA in adolescence may predict increases in borderline features over time, thus identifying EA as an important factor for targeted preventative intervention efforts.

Methods

Participants

Adolescents were recruited from seven schools in a large metropolitan school district, resulting in a sample of 881 adolescents who completed the AFQ-Y as well as measures of borderline features, depression, and anxiety at baseline; 55.9 % of participants were female and 44.1 % were male. The age breakdown for the sample was as follows: 0.4 % were 14 years, 23 % were 15 years, 50 % were 16 years, 23 % 17 years olds, and 3.5 % were 18 years and older. Racial identification was as follows: 31.7 % Hispanic, 30.3 % White, 26.6 % African American, 3.4 % Asian or Pacific Islander, and 8 % Other. Out of the 881 who completed baseline measures, 730 (82.8 %) adolescents completed a follow-up assessment of borderline features exactly 1 year later.

Measures

Avoidance and Fusion Questionnaire for Youth (AFQ-Y)

The AFQ-Y [27] is a 17-item self-report questionnaire that measures EA in youth. It was modeled after the Acceptance and Action Questionnaire, a measure that assesses the same construct in adults [28]. Sample items include “If I feel sad or afraid, then something must be wrong with me,” and “I am afraid of my feelings.” Item responses are rated on a 5-point Likert Scale ranging from 0 (not at all true) to 4 (very true). A total score is obtained by summing item responses, with higher scores indicating a higher degree of EA. Previous studies provide evidence that the AFQ-Y has adequate internal consistency and construct, convergent, and concurrent validity [31]. In the present sample, this measure was administered only at baseline and internal consistency of this measure, as measured by Cronbach’s alpha, was 0.82.

Borderline Personality Disorder Features Scale for Children (BPFS-C)

The BPFS-C [32] was developed from the BPD scale (BOR) of the Personality Assessment Inventory [PAI; 3] and consists of 24-items measuring borderline personality features in childhood (for ages 9 and older). These items assess how participants feel about themselves and other people and are rated on a 5-point Likert Scale ranging from 1 (not at all true) to 5 (always true). Items reflect four domains of core borderline personality disorder features: affective instability, identity problems, negative relationships, and self-harm. Sample items include “I want to let some people know how much they’ve hurt me,” and “When I’m mad, I can’t control what I do.” Construct validity has been established [32, 33], and borderline personality features detected by the BPFS-C have been shown to be moderately stable across time. Specifically, BPD features were found to be moderately stable over the course of 1 year across three time points: fall of year 1, spring of year 1, and fall of year 2. In the present sample, this measure was administered at baseline and 1-year follow-up and internal consistency, as measured by Cronbach’s alpha, was 0.88 at both time points.

The Center for Epidemiologic Studies Depression Scale (CES-D Scale)

The CES-D scale [34] was developed to study depressive symptomology in the general population and is routinely used in adolescent samples from the community [35]. This self-report measure includes 20-items, selected from previously validated depression scales, and targets the major components of depression (i.e., depressed mood, loss of appetite, and feelings of hopelessness). Participants report frequency of depressive symptoms as they occurred during the past week and rate them on a 4-point Likert scale ranging from 0 (rarely or none of the time) to 3 (most or all of the time). Sample items include “I thought my life had been a failure,” “I felt sad,” and “I felt hopeful about the future.” Reliability, validity, and factor structure have been found to be similar across a range of demographic characteristics within the general population. This measure was administered at baseline only, and internal consistency, as measured by Cronbach’s alpha, was 0.75 in the present sample.

Screen for Child Anxiety Related Emotional Disorders (SCARED)

The SCARED [36] is a 66-item self-report questionnaire that measures symptoms of DSM-IV anxiety disorders in children using five subscales. For the current study, only the generalized anxiety disorder subscale, consisting of 9-items, was used to measure anxiety symptoms. Items such as “I worry about things working out for me” and “People tell me I worry too much” were rated by participants on a 3-point Likert scale indicating symptom frequency (i.e., almost never, sometimes, or often). Discriminant validity [37] and concurrent validity [36, 38] have been established in prior studies. Previous research supports the five-factor structure of the SCARED and demonstrates good internal consistency of the GAD subscale [α = 0.86; 39]. In the present study, this measure was administered at baseline only and internal consistency, as measured by Cronbach’s alpha, was 0.85.

Procedures

This study was approved by the appropriate institutional review board, and recruitment was conducted within seven schools, representing five school districts, in a diverse major metropolitan area. On two separate occasions, research staff presented the study to potential student participants and answered any questions that arose. A take-home packet was then sent home with general information about the study and parental consents for signed approval. Students whose parents completed the forms were assented and completed the measures in a private classroom during school hours.

Results

Attrition analyses

Attrition analyses compared adolescents for whom baseline data were collected but did not participate at 1-year follow-up to adolescents who participated at both time points. There were no significant differences between adolescents who did and did not complete the follow-up assessments with regard to sex (χ 2 = 0.07, p = 0.80) or racial identification (χ 2 = 5.13, p = 0.27). There were no significant differences with regard to EA (AFQ-Y total score, t = 1.51, p = 0.13), depression symptoms (CES-D total score, t = −1.16, p = 0.25), or anxiety symptoms (SCARED total score, t = −1.34, p = 0.18) at baseline. There was a significant difference between adolescents who did and did not complete the follow-up assessments with regard to borderline features (BPFS-C total score, t = 1.96, p = 0.05). Adolescents who did not complete assessments at follow-up had significantly higher mean borderline features at baseline (M = 59.94, SD = 14.97) than those who did complete assessments at follow-up (M = 57.15, SD = 14.29).

Descriptive statistics

Table 1 summarizes the descriptive statistics for all main study variables. Isolated items on the BPFS-C and AFQ-Y were missing at random and therefore affected the calculation of total scores, bringing the final sample size for these two measures to n = 881 (BPFS-C) and n = 869 (AFQ-Y).

Table 1 Descriptive statistics for main study variables
Full size table

EA relates to concurrent borderline features

Table 2 summarizes the bivariate correlations between main study variables. As is evident in Table 2, significant correlations were found for EA and the total score on the BPFS-C. As expected, EA also showed significant relations with depression and anxiety. Sex significantly correlated with both EA and borderline features, such that higher levels of both were evident for female adolescents.

Table 2 Bivariate correlations between main study variables
Full size table

To examine the concurrent multivariate relations between EA and borderline features while controlling for known correlates of these constructs, sex, anxiety and depression were entered in a linear regression alongside EA as independent variables. The BPFS-C total score served as the dependent variable. The only variable that retained significance in this regression was EA (β = 0.42; p < 0.001), suggesting that the bivariate relations between depression, anxiety, sex, age, and borderline features are explained by shared variance associated with EA.

EA predicts borderline features at 1-year follow-up

Next, we examined the prospective relation between EA and borderline symptoms 1 year after baseline, controlling for baseline borderline symptoms. To this end, we entered baseline borderline symptoms, depression, anxiety, age, and sex as independent variables in a linear regression. One-year follow-up borderline symptoms served as the outcome variable. All independent variables were entered together in one block using the “enter” method, given that we did not hypothesize incremental variance to be explained by any particular variable. The full model was significant (F = 31.32; p < 0.001) and explained 23 % of the variance in the dependent variable (R 2 = 0.23). AFQ-Y retained significance (β = 0.24; p < 0.001; t = 0.34), alongside BPFS-C baseline scores (β = 0.09; p = 0.01; t = 2.47), depression (β = 0.17; p < 0.001; t = 4.58), and anxiety (β = 0.14; p = 0.001; t = 3.32).

Discussion

This study had two aims. First, we sought to examine the concurrent relations between EA and borderline features in adolescents in a large and diverse community sample of adolescents, controlling for depression and anxiety (due to the known relations among depression, anxiety, borderline features, and EA). Second, we examined the prospective relation between EA and borderline features over a 1-year follow-up period, controlling for baseline level of borderline features and symptoms of depression and anxiety. Two main findings were reported. First, we found that EA associated with all three types of emotional disorders (e.g., depression, anxiety, and borderline features) at the bivariate level, but when all disorders were considered together, the effects of EA washed out the effects of depression and anxiety in the relation with BPD, suggesting that the relations between these disorders may be accounted for by EA as an underlying cross-cutting psychological process. Second, we found that EA predicted levels of borderline symptoms at 1-year follow-up, controlling for baseline levels of borderline symptoms and symptoms of depression and anxiety.

Our first finding may be interpreted in the context of the recent emphasis on underlying, cross-cutting constructs expressed in the NIMH Research Domain Criteria (RDoc) initiative. Indeed, rather than forcing comorbid disorders into discrete categories, this approach calls for examination of underlying mechanisms across disorders. That EA washed out the effects of depression and anxiety in the relation with BPD suggests that the relations between these disorders may be accounted for by EA as a cross-cutting underlying psychological process. This finding emphasizes the value of approaches such as the new NIMH RDoC initiative and speaks to the value of transdiagnostic approaches in the treatment of mood-related disorders. These approaches include ACT [40], dialectical behavior therapy [18], and mentalization-based treatment [41]. While the role of EA has been a central organizing principle of ACT [42], it is also implicated in Linehan’s biosocial theory [18] and experimental investigations have demonstrated effects for DBT on EA in adults diagnosed with BPD. For instance, during a recent randomized controlled trial [43], it was demonstrated that in DBT treatment for BPD, a reduction in EA levels related to a reduction in depression levels. Conversely, failure to reduce EA levels impeded the reduction of depression scores. Avoidance of internal experiences is a core feature of the mentalization-based conceptualization of BPD developed by Fonagy and colleagues [20, 41, 44]. Mentalization, in this context, refers to the capacity to reflect on the mind’s content without restriction or judgment. EA stands in clear contrast to the process of effective mentalization, which implies a compassionate and curious stance toward one’s own mental states.

The finding of a prospective relation between EA and borderline features, controlling for baseline level of borderline features, tentatively suggests a causal role for EA. Developmental models certainly conceptualize EA in a causal role, although direct evidence for links with early development is lacking. For instance, in the mentalization-based account of BPD [19, 20], it is thought that the capacity to be open and curious about own mental states without becoming distressed by them or trying to control them (that is, EA) comes about in the context of secure attachment with primary caregivers. In this context, the parent is not overwhelmed with negative or distressing expressions of the infant or child’s emotions and allows the child to express such emotions, while at the same time modulating these emotions using “marked mirroring.” Marked mirroring involves a deliberate engagement of the infant or child using eye contact and a certain amount of exaggeration—referred to as ostensive communication. That is, the parent employs a cheerful, sing-song voice or holds the child particularly gently, acknowledging the distress but also communicating that it is the infant feeling distress and not the parent feeling distress over the child. The iterative occurrence of this process teaches the child that she has a mind and that the mind has different states (e.g., happy, sad, frightened, angry) and that these states are allowed and can be owned. A disruption of this process may instead produce the sense that emotional states cannot or should not be experienced, but rather they should be avoided (as in EA), leading to psychopathology over time. Empirical evidence for these processes has been accumulating [45, 46]. While the findings of this study do not explicitly test this developmental model (given the adolescent age range studied in the current study), our findings speak to the possibility that variability in EA is an independent predictor of borderline features prospectively.

Recent approaches in psychopathology research [47, 48] have emphasized the value of studying psychopathology dimensionally across the full latent trait in an unselected sample, which requires community samples as used in the current study. Therefore, we consider the use of a community sample in the current study a strength. However, prospective relations between EA and borderline traits should also be established in clinical samples to justify an interventional focus on EA. While these links have been established cross-sectionally [29], longitudinal work in a clinical setting would be informative. A second limitation of the current study relates to the fact that all measures were self-report based. It is challenging to measure EA through more objective approaches. Even so, several laboratory-based studies have developed EA paradigms, including psychologically challenging [49] or distressing [50, 51] tasks, pain tolerance tests [52], and thought suppression experiments [53, 54]. These measures may assess EA more objectively and should be included in future studies alongside the AFQ-Y. Third, a significant difference in borderline features between adolescents who did and did not complete follow-up assessments indicates that an important subsection of our community sample was not included in the prospective analyses. Specifically, adolescents who did not complete follow-up assessments reported higher borderline features at baseline, perhaps suggesting that the absence of follow-up data is a function of increased pathology at baseline. The latter speaks to the importance of reducing attrition in longitudinal research [55]. Finally, although research supports the validity of the SCARED subscales [39], we are the first study to use only the generalized anxiety disorder subscale of the SCARED. Other anxiety disorders that may relate to EA and/or borderline features were not considered.

Notwithstanding these limitations, the current study makes an important contribution by providing the first evidence of a link between borderline features and EA in a community sample of adolescents. Additionally, this study provides the first evidence for a prospective relation between EA and borderline features. The suggestion that EA may be a cross-cutting psychological variable underlying several forms of psychopathology has important treatment implications. Our findings emphasize the importance of helping adolescents learn to validate and accept their internal states—not only as an important step in their recovery process (if they are struggling with borderline symptoms), but also as a potential preventative measure for the development or exacerbation of borderline symptoms.