Mother–child interactions in adolescents with borderline personality disorder traits and the impact of early life maltreatment

Although remission rates for BPD are promising with an average of 60% in the course of 5–15 years [2], individuals with BPD often continue to suffer from severe interpersonal impairments [42, 58, 83, 90]. The cumulation of stressful interpersonal life events can further aggravate poor psychosocial functioning in adult BPD patients [68]. Also, acute BPD symptoms like self-harm behavior, intense anger or depressive symptoms often occur in interpersonal contexts [44, 66, 69, 71] and the perceived quality of interpersonal relationships influences current BPD symptomatology and vice versa [44]. Interpersonal experiences and processes therefore seem to influence present BPD symptoms but also the long-term psychosocial development of BPD individuals. As early relationship experiences form our later expectation of and behavior in social interactions [1, 37, 40], a closer look at caregiver-child relationships and their dyadic interaction is warranted.

Many adolescent BPD patients still live at home and experience maladaptive relationships with their caregivers [46]. This becomes especially relevant when considering the role of transgenerational transmission of mental disorder and trauma in the development of the disorder (see [48]. Several therapeutic approaches for adolescent BPD (e.g. [73]: dialectic-behavioral therapy, DBT, [76]: mentalization-based treatment, MBT, [35]: adolescent identity training, AIT) have already addressed this fact by including caregivers in the treatment of the disorder. However, although first evidence from these treatments suggests an enhancement of psychosocial functioning for adolescent BPD patients (e.g. [79], they still seem to profit less from these therapies than adult BPD patients [86]. More research is needed in order to understand familial interactional patterns, detect risk and protective factors and to identify potential windows for interventions.

According to Linehan’s Biosocial Theory (1993; see also [23], BPD develops as a result of child vulnerability (e.g. impulsivity, emotional sensitivity) interacting with an invalidating social environment. Important social risk factors include parental psychopathology, poor quality of the parent–child relationship, dysfunctional parenting practices and early life maltreatment [8, 15, 84].

Past research has mainly focused on the influence of parental caregiving, i.e. parent-driven effects, on the development of BPD pathology. In community samples, maladaptive parenting (e.g. chaotic parenting, physical maltreatment) was identified as predictive of adolescent BPD symptoms [5, 8]; validating parenting (emotional support, involvement), on the other hand, could have a protective effect on developing BPD symptoms [39]. Also in clinical samples, adolescents and young adults with BPD report that their parents displayed several problematic parenting practices (e.g. emotional withdrawal, parental inconsistencies, invalidation of thoughts and feelings; [7, 10, 14, 46, 63, 80, 89, 91]. Studies investigating child-driven effects have focused on temperament-related features: a review of Boucher et al. [14] summarized that parents of children with BPD often describe their child as “unusually sensitive” or with a “difficult temperament” early on.

The studies reviewed above relied on the retrospective reports of BPD patients and/or their parents by applying self-report questionnaires and interviews. Self-reported experiences, however, may be influenced by recall bias and at least in parts by the BPD symptomatology itself [18, 30]. The observation and professional coding of behavior during caregiver-child interactions offers a chance to address this problem. Regarding parent-driven effects, longitudinal studies using high-risk community samples showed that maternal withdrawal and hostility displayed during parent-toddler interactions predicted BPD symptoms in early adulthood [18, 61]. Maternal insensitivity during mother–child interactions at infancy, preschool and adolescence was associated with adolescent BPD traits [17]. On the child’s side, disorganized-controlling behavior at age 8 was predictive of early adult BPD symptoms [61].

Surprisingly few studies have observed (and reported) parent and child behavior at the same time in the context of current BPD symptomatology. It has also been suggested to observe behavior not only on an individual but also on a dyadic level, as mother and child are most likely influencing each other during interactions [92]. In at-risk community samples with adolescents, dyadic negative escalation and disoriented/role-confused behavior during a conflict discussion between mothers and their adolescent children were associated with more BPD traits [60, 92]. Positive dyadic behavior, however, was related to decreases in adolescent girls’ BPD severity scores over time [92]. Dixon-Gordon et al. [28] identified adolescent negative affect during interaction as a possible risk factor for BPD traits but only when mothers showed low support/validation and high problem solving during a conflict discussion task. In a clinical sample comparing young adults with BPD with healthy controls, BPD patients and their mothers displayed more disorganized behavior during a conflict discussion task [54]. Whilst overall collaboration as a marker for dyadic behavior during conflict discussions did not predict adolescent BPD traits in a high-risk community sample [60], it contributed to a more secure attachment profile in the clinical sample of young adults [54]. Fleck et al. [34] observed interactions in two community samples. At age 9, they found less maternal structuring and more child withdrawal during a conflict discussion task to be associated with BPD traits, but no relations during a fun day planning task. At age 14, less maternal sensitivity and structuring, more maternal intrusiveness and less child engagement during the conflict discussion task and more child withdrawal during the fun day planning was related to BPD traits. Associations between dyadic behavior and BPD traits in adolescents were significant during the conflict discussion task, but not during the fun day planning.

In summary, parental and child behavior in caregiver-child transactions seem to be altered and more conflict-driven when child or adolescent BPD traits are present. This seems to be especially relevant in stressful contexts (e.g. [34], when problems with emotion regulation would become noticeable. There is also first evidence that positive parental and dyadic behavior might mitigate the development of the disorder [39, 92]. Although many studies have identified specific maladaptive parenting practices that seem to foster the development of BPD, parental behavior was rarely investigated during actual mother–child interactions. Studies that have observed interactional behavior mainly focused on either parent- or child-driven effects in community samples.

To the best of our knowledge, only four studies have studied the influence of current BPD traits on parent, child and dyadic behavior during the same interaction [34, 54, 60, 92]; the only study including a case–control design focused on young adults rather than adolescents [54]. Only one study compared observed mother–child interactions in a positive versus a stressful context [34], although context seems to have an impact on relations between mother–child interactions and child behavioral problems [27].

The present study addresses these limitations by comparing a clinical adolescent sample with BPD symptoms and their mothers with healthy control dyads, observing maternal, adolescent and dyadic behavior during a positive interaction and a stress paradigm.

An overwhelming body of literature has identified ELM as a contributing factor to the development of BPD traits in children, adolescents and adults [16, 45, 70, 94]. Lyons-Ruth et al. [61] suggested that the assessment of parent–child interactions should include ELM as a possible influential factor. Previous research has focused mainly on maternal ELM and its influence on maternal behavior in mother–child interactions: mothers who have experienced ELM are more likely to show maladaptive parenting, including parental behavior that was previously associated with the development of BPD, such as psychological control, maternal hostility or harsh punishment [77, 88].

Although some of the above-described studies have reported child or adolescent ELM and have shown associations with the development of BPD traits (e.g. [18], the influence of child ELM on behavior in mother–child interactions was rarely considered. Maternal withdrawal during a mother-toddler interaction, child disorganized-controlling behavior at age 8 and less collaborative, more mutual punitive and disoriented/out-of-context behavior during discussions between young adults and their parents seem to be influenced by the severity of the trauma the child/adolescent has experienced in the past [60, 61]. Further research is needed to disentangle the effects of maternal/adolescent ELM and BPD pathology on caregiver-child interactions and how this may facilitate the development of BPD.

With the present study, we aimed at expanding prior research on the observation of caregiver-child transactions by using a case–control group design. Specifically, we wanted to know how a clinical group of adolescents with BPD traits (BPD-G) differs from a healthy control group (HC-G) in maternal, adolescent and dyadic behavior during mother-adolescent interactions in a positive versus a stress context. Additionally, we wanted to investigate the role of maternal and adolescent ELM and how it might independently or additionally to adolescent BPD symptomatology contribute to behavior in this context.

Firstly, and consistent with prior literature, we expected maternal, adolescent and dyadic behavior to be of less quality in the BPD-G than in the HC-G. During stress, we assumed this group difference to be larger than during the fun day planning, i.e. the BPD-G was expected to show more dysfunctional behavior during stress -when emotion regulation difficulties might come into play- than during fun day whilst we did not expect the HC-G to significantly change behavior between contexts. We were also interested in exploratory analyses on subscale level to identify specific behavior (e.g. maternal intrusiveness vs. maternal sensitivity), that might explain differences between groups and/or contexts.

Secondly, we assumed that BPD-G mothers and adolescents experienced more ELM than subjects in the HC-G. As a large body of research suggests (see [88], we expected maternal ELM to influence maternal and dyadic behavior in both interactions; no specific assumptions were made for adolescent behavior. Regarding the influence of adolescent ELM on behavior we expected all behavioral outcomes to be affected, following indications of prior research [60, 61]. Lastly, we explored whether adolescent ELM has an effect on interactional behavior above and beyond the effect of group membership.

A-priori power analyses for group comparisons utilizing an anticipated effect size of d = 0.8, a desired statistical power level of 80% and a probability level of α = 0.05 revealed a minimum sample size of n = 26 per group. We therefore aimed at recruiting a total of 30 clinical dyads (BPD-G) and 30 healthy control dyads (HC-G). Recruitment took place from 06/2018 to 01/2021. BPD-G adolescents were recruited in our outpatient clinic for risk-taking and self-harm behavior (AtR!Sk; [50, 51]. Patients had to meet ≥ 3 criteria of BPD, which was assessed by trained clinical psychologists using the Structured Clinical Interview for DSM-5-Personality Disorders (SCID-5-PD; [33]. BPD-adolescents were also screened for other psychiatric disorders using the Mini International Neuropsychiatric Interview for Children and Adolescents (MINI-KID; [82]. HC-G were recruited via advertising and the local residents’ registration office and matched to BPD-G according to adolescent sex and age, and adolescent and maternal education. HC-G dyads were excluded if adolescents fulfilled criteria for any current or lifetime disorder (assessed with the MINI-KID or if mothers had received any psychotherapeutic/psychiatric treatment in the 2 years prior to the study. HC-G adolescents were also screened for BPD traits using the SCID-5-PD [33]. Further exclusion criteria for all mothers and adolescents were a diagnosis of schizophrenia and/or autism. As the study included biological measures (not reported in this manuscript), exclusion criteria for both groups were serious somatic illness, neurological disorder or cardiac/hypothalamus–pituitary–adrenal system dysfunction. Also, mothers had to be the primary caregiver.

From a pool of 353 possible participants (BPD-G = 161, HC-G = 192), 294 (83.3%) were contacted and screened for exclusion criteria (reasons for not being contacted: no phone number/email available, BPD-G: still waiting for or in the process of the clinical assessment of AtR!Sk, HC-G: no match for BPD-G). 205 of the contacted dyads [BPD-G: 102(63.4%); HC-G: 103(53.6%)] could not be included due to lack of interest (BPD-G: 51, HC-G: 46), somatic illness (BPD-G: 12; HC-G: 17), lack of time (BPD-G: 11; HC-G: 10), being too young or too old (BPD-G: 11; HC-G: 5), mother not being the primary caregiver or not being available (BPD-G: 9), no match for BPD-G (HC-G: 5), insufficient language skills (BPD-G: 4; HC-G: 2), psychiatric illness according to exclusion criteria (BPD-G: 2; HC-G: 12) or giving wrong contact information (BPD-G: 2; HC-G: 6). From 89 (25.2%) included dyads, 16 (18%) became dropouts during the course of the study: One BPD-G turned 18, one BPD-G mother reported a somatic illness, 9 of HC-G adolescents reported psychopathology of any kind, and some dyads lost interest in the study in both groups (BPD-G: 3; HC-G: 2).

Finally, 38 adolescent patients between 12;0 and 17;0 years (mean = 15.6, sd = 1.13) and their mothers formed the BPD-G; 35 healthy dyads formed the HC-G (adolescents aged between 14;0 and 17;0, mean = 15.5, sd = 1.25). Adolescents were mostly female (BPD-G: 84.2%, HC-G: 80%) and on track for higher education. Mothers were well educated and the majority part-time or full-time employed. All participants were of European ancestry. For a detailed sample description see Table 1.

Dyads were invited to our laboratory in Heidelberg for two appointments (t1 and t2) over a 3 week period. At t1, clinical assessment (interviews and questionnaires) and a computer task were performed. At t2, two 10-min-long standardized mother-adolescent interactions (a positive interaction that was, after a resting period, followed by a stress task) were videotaped. During the positive interaction, dyads were asked to plan positive activities both individuals would benefit from and enjoy. The stress task was loosely based on the Parent–Child-Challenging Task (PCCT) by Lunkenheimer et al. [59] and has, to the best of our knowledge, not been used before. During the stress task, the adolescent was presented with a tangram. The dyad was told that other adolescents were able to solve the tangram without any issues, when in fact it was too difficult to work out the tangram in the allotted timeframe. Additionally, the examiner would carefully observe their approach and make notes about their performance. Mothers were instructed to support their child but not to solve the puzzle for them. After 5 min, it was stated that the child was unsuccessful in completing the task and therefore an easier tangram would be presented (which was actually even more difficult to solve than the first one).

Preliminary analyses confirmed that mothers and adolescents reported significantly more negative affect [mothers: t(72) = -5.731, p < 0.001; adolescents: t(72) = -6.706, p < 0.001] and less positive affect (mothers: t(72) = 2.66, p < 0.01; adolescents: t(72) = 3.15, p < 0.01) in the stress task in comparison to the positive interaction.

Before, during and/or after interactions, physiological data (functional near-infrared spectroscopy, electrocardiography, saliva sampling) was retrieved. The physiological data as well as the computer task were not analyzed in the present study and will therefore not be described further.

All analyses were carried out with R (v1.4.1717; [72]. Per group, one stress task was missing due to malfunction of the video camera [n = 2 (2.74%)]. Welch’s t-tests were applied to calculate group differences for continuous variables [24], chi-square-tests for categorical variables.

For the analyses of the CIB data (BPD-G vs. HC-G and positive vs. stress context), robust two-way mixed Analyses of Variance (ANOVA) using 20% trimmed means were calculated via the WRS2 package [64], as variance homogeneity could not be met. Although normal distribution could be assumed (sample sizes > 30), it was visually verified and tested with the Shapiro–Wilk-Test. Greenhouse–Geisser corrections were made when the assumption of sphericity was violated. In our main analyses, Bonferroni-Holm correction for multiple testing was applied across effects that were applying to the same hypothesis (i.e. for group differences, context differences and group x context interactions we controlled for three comparisons per hypothesis).

To ensure that the group effect on behavior is not an effect of general psychopathology but specific to BPD traits, we transformed all values into z-values and then calculated Pearson correlation coefficients to determine the association between SDQ and the CIB scales. When significant correlations were found, we calculated regression analyses with group as a predictor and the respective CIB scale as an outcome variable, controlling for SDQ.

To determine how much additional variance in behavior could be explained by maternal and/or adolescent ELM, again all values were z-transformed and Pearson correlation coefficients were calculated. Significant correlations were further investigated in hierarchical linear regression models (step1: group as a single predictor, step2: group and CTQ as predictors).

When homoscedasticity could not be met in the above-described regression models, HC4-method for robust standard errors was applied.

In the BPD-G (range = 3–8 BPD criteria), 19 (50%) adolescents fulfilled the BPD diagnosis. 9 (23.7%) BPD-G adolescents were diagnosed with F10-F19 diagnoses, 29 (76.3%) with F30-F39 diagnoses, 16 (42.10%) with F40-F49 diagnoses and 8 (21.05%) with F90-F99 diagnoses. 30 (78.95%) patients fulfilled at least two or more diagnoses. Adolescents of the BPD-G reported significantly more emotional and behavioral problems in the SDQ than the adolescents of the HC-G (p < 0.001); BPD-G mothers reported significantly more psychopathology in the SCL-90-R (p = 0.001), significantly more interpersonal sensitivity in the respective SCL-90-R subscale (p = 0.033) and significantly more attachment insecurity in the VASQ (p = 0.028) than HC-G mothers (for details see Table 2). However, on additional exploratory analyses we did not find any significant correlations between maternal interpersonal sensitivity and attachment insecurity and maternal, adolescent or dyadic behavior (r = 0.00 to r = 0.14).

BPD-G adolescents reported a significantly higher CTQ total score in comparison to the HC-G (p < 0.001). Mothers’ CTQ, however, did not significantly differ between groups (p = 0.365). Details can be obtained from Table 2.

Means and standard deviations of CIB scores from both interactions can be obtained from Table 3. For our main analyses, robust two-way mixed-model ANOVAs revealed a significant group effect for maternal (p_adj = 0.003), adolescent (p_adj = 0.003) and dyadic behavior (p_adj = 0.003): the BPD-G showed a significantly lower quality of maternal, adolescent and dyadic behavior in both interactions than the HC-G. For context, no significant main effects were found. Group x context interactions were also not significant (details can be obtained from Table 4).

In exploratory analyses on subscale level, significant main effects for group were found for maternal sensitivity (p = 0.003) and maternal structuring (p = 0.004) but not for maternal intrusiveness (p = 0.317), all child scales [engagement: p = 0.002; compliance: p = 0.002; withdrawal: p = 0.006] and both dyadic scales [reciprocity: p < 0.001; negativity: p = 0.019]: in the BPD-G, mothers were less sensitive and structuring; adolescents behaved less engaged and compliant and more withdrawn. BPD-G dyads showed less reciprocity and more negativity in comparison to the HC-G. Context effects were significant for maternal sensitivity (p < 0.001), child engagement (p < 0.001), child withdrawal (p = 0.025) and dyadic negativity (p = 0.049): in both groups, mothers behaved less sensitive in the stress task compared to the positive interaction whilst adolescents displayed less engagement and less withdrawal. Dyadic negativity scores were higher in the stress task than in the positive interaction task. Context effects for maternal structuring and intrusiveness, child compliance and dyadic reciprocity were not significant. For dyadic reciprocity a significant group x context interaction (p = 0.029) was found: whilst HC-G dyads increased reciprocity under stress, BPD-G dyads decreased their reciprocal behavior. Details of all calculated ANOVAs are described in Table 4.

We next investigated, if SDQ total values were associated with CIB behavior. Exploratory analyses showed that SDQ was significantly negatively correlated with child and dyadic behavior during both interactions (r = − 0.24 to r = − 0.36) but not with maternal behavior. We then ran different regression analyses with group as a predictor and adolescent and dyadic behavior as outcome variables, controlling for SDQ total values. Whilst group still significantly predicted adolescent behavior during both tasks (positive interaction: p = 0.024, stress task: p = 0.044), the significant effect of group to dyadic behavior disappeared (positive interaction: p = 0.108, stress task: p = 0.141).

Maternal CTQ total scores did not significantly correlate with any of the CIB scales and were therefore not considered in further analyses. Adolescent CTQ total scores were significantly negatively correlated with all CIB total scales (r = − 0.24 to r = − 0.40). Hierarchical linear regressions revealed that adolescent CTQ did not significantly explain any additional variance in CIB scores above and beyond group as a predictor (changes in R²: p = 0.078 to p = 0.425; see Table 5).