In face of the clinically challenging heterogeneous presentation of BPD and the enduring scientific debate about whether the heterogeneity can be best explained by categorical or dimensional differences between individuals, the current study applied LCA (investigating qualitatively distinct subtypes), FA (investigating dimensional differences) and FMM (allowing a latent structure to have both categorical and dimensional aspects) to the DSM-IV BPD criteria in a sample of adolescent outpatients with risk-taking and/or self-harming behavior. The main result that emerged from the study was that help-seeking adolescents with BPD features are best represented as two qualitatively distinct subgroups, with sex significantly explaining group membership, and both a latent factor and age explaining heterogeneity within groups. As the latent factor in our best fitting model explained within-class variability only, the two identified groups cannot be compared with regard to mean differences in the factor. As implied by the class-varying item thresholds, the two groups were based on the responses to the BPD criteria rather than the factor mean and variance (25). There was a majority group with relatively high probabilities for all BPD criteria (“the borderline group”), and a minority group with relatively low probabilities for all BPD criteria, except from impulsivity and anger (“the impulsive group”). The class-varying covariance matrix allowed for different levels of heterogeneity within each class, resulting in the borderline group having a greater range of symptoms compared with the impulsive group. Considering sex and age as covariates significantly improved the model fit, indicating that these variables should be taken into account when explaining heterogeneity of BPD among adolescents. Being female was associated with a greater likelihood of belonging to the borderline group, being male with a greater likelihood of belonging to the impulsive group. Within each group, older adolescents were more likely to meet a BPD criterion than younger adolescents, which is in line with the epidemiological finding that BPD first emerges during adolescence and peaks during early adulthood (2). The two identified groups demonstrated meaningful differences in predisposing factors and clinical variables, supporting their validity.
From a developmental perspective (54), it could be argued that the borderline group included individuals who had experienced emotional abuse / neglect or sexual abuse early in life and then developed a personality characterized by high negative emotionality, stress sensitivity, and social inhibition (39, 55), which in turn made them more susceptible to severe psychopathology, functional impairments, and life dissatisfaction. In contrast, the impulsive group may have consisted of people characterized by an attitude of lack of regard for others (39, 55), which in turn predisposed them to dissocial behavior and substance-related and behavioral addictions, resulting in a phenotype resembling ASPD in adulthood. The developmental pathway appeared to be crucially influenced by sex, with females rather belonging to the borderline group and males to the impulsive group. The question arises whether the impulsive group actually represents a “true” BPD group or rather a precursor or early manifestation of ASPD in adulthood. Evidence indicates that BPD and ASPD share common biological vulnerabilities (e.g., trait impulsivity derived from dopaminergic and serotonergic dysfunctions) and environmental risk factors (e.g., disrupted attachment, abuse and neglect), with sex moderating the phenotypical expression of biology x environment interactions to produce BPD overproportionately in females and ASPD overproportionately in males (56). For instance, some high risk genes may confer differential vulnerability to internalizing behaviors among girls versus externalizing behavior among boys. Additionally, deviant peer group affiliations may emerge during adolescence, leading girls to become exposed to self-injurious behaviors of peers and boys to delinquent behaviors (56). Future examination of the stability of the two identified adolescent groups over time is needed.
Our findings are most consistent with the LCA results reported by Fossati et al. (20) and Thatcher et al. (22). Both reported an impulsive class that endorsed symptoms of impulsivity and anger only, along with two (20) or three (22) BPD classes differing in severity. Comparably to our findings, the impulsive group in Thatcher et al. (22)’s study included an overproportionally large number of males and was characterized by high rates of CD, while the severe BPD group was distinguished by high rates of depression. Our findings stand in contrast to previous studies suggesting that the heterogeneous clinical presentation of BPD can be best understood in terms of individual differences on a single underlying trait (“BPD-ness”) or subgroups that lie on a continuum of BPD severity (18, 19, 21, 27, 28).
Several methodological reasons may account for these diverging results. First, the majority of studies did either apply LCA or FA on the diagnostic criteria when investigating the latent structure of BPD (17, 18, 21, 22), while we systematically compared LCA, FA, and FMM. Second, only a few studies have systematically explored the effects of covariates, such as sex and age. There have been mixed results, with two studies reporting that females were more likely than males to belong to the class with more BPD criteria (18, 19), and one study reporting no sex difference (27). To the best of our knowledge, the impact of age has only been examined in one study (19) that found that the probability of belonging to the borderline group declined with increasing age until the age of 27, from which the probability increased. Our results confirm that sex might have an important impact on latent class membership, with females having a greater likelihood of belonging to the borderline group than males. We could not replicate a direct effect of age on latent group membership, but found that age explains within-class variability, with the probability of endorsing a BPD criterion being higher with increasing age. Third, the studies included various clinical and community samples, with well-known differences in prevalence rates for females and males. In community samples, the sex ratio is 1:1, while clinical samples usually show three times more females than males with the disorder (1). Forth and probably most importantly, because BPD presents differently across the lifespan (14), the majority of studies have examined adults with mean ages ranging between 20 and 42 years (18–22, 27, 28), while our sample consisted of adolescents with a mean age of 15 years. We are aware of three studies investigating subtypes of BPD in adolescence. Two of them identified two subgroups (based on either the personality pattern scales from the Millon Adolescent Clinical Inventory (57), or the Shedler-Westen Assessment Procedure-200 for Adolescents (58)) that were clearly gendered and differed regarding the internalizing-externalizing dimensions of psychopathology (59, 60), with internalizing psychopathology being more common among females and externalizing psychopathology being more common among males (61, 62). The third study examined females only and identified four groups (based on the Borderline Personality Questionnaire (63)) with an increasing number of BPD symptoms and distinct patterns of comorbidities (17). Our results are consistent with the finding of a more female, internalizing group, and a more male, externalizing group.
Clinically, our findings have several important implications. First, they are in favor of early assessment and treatment of borderline features among adolescents, even if they are below the diagnostic threshold (2, 7, 9), as they are associated with co-occurring psychopathology, functional impairments, and high emotional burden (64). The borderline group based on latent group membership was more inclusive than the DSM-IV, with 42% meeting the diagnostic threshold of five DSM-IV criteria at baseline, and the average number of BPD criteria being nearly four (see Table 4). This may suggest that the DSM-IV threshold (which remained the same in DSM-5) is too restrictive to adequately conceptualize the borderline construct in adolescence (65). Second, the low rate of males in our sample along with the well-known 1:1 sex ratio for BPD in adult community samples (1) implies that many young males with BPD features such as impulsivity and anger may not access mental health services, but turn up on other services’ doorsteps, including police services and courts. An integrated treatment approach that involves collaboration between services is needed to improve treatment access and engagement for this particular group. Third, mixed-effects linear regression analyses did not find a group difference in clinical improvement over time, indicting that both groups benefited from the received treatment that included elements from cognitive behavioral therapy and dialectical behavioral therapy (66, 67). However, due to the short follow-up period and the substantial amount of missing data in the current study, this finding has to be considered as preliminary. Future studies examining between and within group variability in clinical changes of the two identified groups over a longer period of time are required to clarify whether or not group-specific treatment adaptations could be beneficial.
The strengths of the current study include a large representative sample of help-seeking adolescents with BPD features, the structured assessment of BPD pathology by trained psychologists, the systematic comparison of different latent models according to the procedure proposed by Clark et al. (25), the consideration of sex and age as covariates in the latent models, and the validation of the identified latent structure using external variables. The study has several limitations that ought to be considered. First, the sample consisted of outpatients, limiting the generalizability of the results to adolescents in the community. Second, there was a substantial amount of missing values in the variables used for post-hoc comparisons of the latent classes. Reasons for the missing values include the nature of the consecutive sample, the omission of questions, and the introduction of additional measures during the running study. Third, as Latent GOLD® does not provide common fit indices for comparison of FA models (e.g., Comparative Fit Index or Root Mean Square Error of Approximation), our selection of the best fitting CFA was based on the BIC and SABIC only. Last, as BPD criteria wax and wane over time, it has been argued that subtyping individuals with BPD features according to underlying pathological mechanisms may be a more promising approach (28, 62, 68).