EXAMINING THE CONSTRUCT VALIDITY OF BORDERLINE PERSONALITY TRAITS USING FAMILIAL AGGREGATION AND OTHER EXTERNAL VALIDATORS

Abstract

Numerous studies have questioned the reliability and validity of borderline personality disorder’s (BPD) categorical conceptualization. DSM-5 Section III’s alternative trait-based model of BPD may better capture borderline pathology, but aspects of its validity should be further established. Thus, the authors examined whether a latent BPD factor derived from Section III traits exhibits (1) familial aggregation among siblings and (2) association with constructs related to borderline pathology. The authors also tested whether gender moderated associations. A total of 498 community-recruited adults completed the Personality Inventory for DSM-5, a behavioral assessment of risk-taking, and reported their history of childhood maltreatment, substance use, nonsuicidal self-injury, and suicidal ideation. Familial aggregation was assessed among 232 sibling pairs. Siblings’ BPD scores were significantly correlated and most external validators were significantly associated with BPD, with the exception of risk-taking. Results did not vary by gender. Findings further support the construct validity of Section III’s BPD trait profile.

Borderline personality disorder (BPD) is a debilitating psychiatric illness characterized by persistent instability of self-image, interpersonal relationships, and affect (American Psychiatric Association [APA], 2013). BPD is common across clinical settings, with prevalence rates up to 22% in inpatient psychiatric settings (Ellison et al., 2018); is associated with heavy health care utilization (Bender et al., 2001); and leads to functional impairment that is on par with primary affective and certain psychotic disorders (Temes & Zanarini, 2018). BPD is also associated with high rates of suicidality (Lieb et al., 2004), and its burden is exacerbated by suboptimal treatment strategies (Levy et al., 2018; Lieb et al., 2010).

Poor treatment outcomes may be, in part, due to the conceptualization of BPD as a categorical construct. Indeed, the validity of the categorical model of BPD has been questioned extensively (Hudziak et al., 1996; Paris, 2007). First, taxometric studies have shown that borderline pathology is best captured by a dimensional model (Haslam et al., 2012; Trull et al., 1990). Second, longitudinal studies have shown that BPD symptomatology fluctuates substantially over time (Temes & Zanarini, 2018; Videler et al., 2019), which conflicts with the definition of personality disorders as stable and enduring (APA, 2013). Third, BPD nearly always co-occurs with other psychiatric conditions, calling into question whether BPD is a discrete entity that is separate from other clinical phenotypes (Shah & Zanarini, 2018; Zanarini et al., 1998; Zanarini et al., 2004). Together, these findings cast doubt on the traditional categorical model of BPD in favor of models that incorporate traits that can better account for symptom heterogeneity and co-occurrence (Conway et al., 2018).

One prominent model of BPD that incorporates dimensional traits is included in the Section III Alternative DSM-5 Model of Personality Disorders (APA, 2013). The Section III model of BPD, which is a hybrid categorical-dimensional model, characterizes BPD as personality-related impairment in self and interpersonal functioning (Criterion A) with extreme deviations in personality traits that are normally distributed in the general population (Criterion B). The specific aberrant traits included in Criterion B are Emotional Lability, Separation Insecurity, Anxiousness, and Depressivity (within the domain of negative affectivity); Impulsivity and Risk-Taking (within the domain of disinhibition); and Hostility (within the domain of antagonism). These traits can be assessed with the Personality Inventory for DSM-5 (PID-5; Krueger et al., 2012) and are informed by prior work on the relation between the Five-Factor Model (FFM) of personality and categorical personality disorders (Samuel & Widiger, 2008).

The psychometric properties of BPD models that incorporate dimensional traits have been examined in several studies, and results indicate that these models, including the Section III model of BPD, exhibit good reliability and validity. For example, Sellbom et al. (2014) reported Cronbach’s alphas for PID-5 borderline traits ranging from 0.85 (Risk-Taking and Separation Insecurity) to 0.93 (Depressivity), and Morey (2019) found strong interrater agreement for Section III-defined BPD using clinical vignettes (intraclass correlation coefficient [ICC] = 0.33).

Several studies have examined the construct validity of the Section III model of BPD by assessing relationships between the BPD traits and external correlates known to be associated with the construct of interest (Anderson & Sellbom, 2015; Anderson et al., 2016; Munro & Sellbom, 2020; Sajjadi et al., 2022). For example, Anderson et al. (2016) examined associations between the Section III BPD traits and an array of expected external correlates (e.g., self-harm, maltreatment, aggressive behaviors, contact with care providers), identifying relationships with most correlates across both an undergraduate and psychiatric inpatient sample. Similarly, Anderson and Sellbom (2015) found that among undergraduate students, external correlates of BPD (e.g., substance use, dissociative experiences, sexual activity, trauma history) were related to Section III BPD trait scores (i.e., an average of the seven facet scores). Critically, research has found that the Section III model of BPD also predicts external correlates of BPD above and beyond existing categorical models (Anderson et al., 2016; Boland et al., 2018). For example, Anderson et al. (2016) found that the Section III model of BPD predicted more incremental variance in many external correlates of BPD than the Section II categorical diagnosis of BPD, including substance abuse, self-doubt, low positive emotions, and dysfunctional negative emotions.

While foundational evidence has accrued in support of the Section III model of BPD, it is important that the validity of this model be further established. Notably, none of the previous studies mentioned above assessed familiality or used laboratory-based behavioral measures as external correlates of the model. In addition, only a minority of the studies have drawn on community samples (e.g., Munro & Sellbom, 2020). Although the use of clinical samples is essential for determining the validity and utility of the Section III model of BPD, it is also important to identify the degree to which the Section III model of BPD relates to correlates of BPD in the broader population, given that the traits are purported to be distributed along a continuum in all individuals. Thus, drawing on a large community sample oversampled for psychopathology, the present study sought to build on and extend previous findings by testing (1) familiality (i.e., whether the model “runs in families”) and associations with well-established correlates of BPD, including (2) maltreatment in childhood, (3) clinical phenotypes (i.e., suicidal behaviors and alcohol and substance use disorders), and (4) a laboratory probe of risk-taking.

Given the strength with which family history of a disorder predicts the presence of disorders among probands (Klein et al., 2003; Nierenberg et al., 2007; Shankman & Klein, 2003), familiality (i.e., familial aggregation) has long been viewed as an important validator of psychiatric disorders and constructs (Robins & Guze, 1970). It has already been established that personality disorders (broadly) and categorically defined BPD more specifically tend to aggregate within families (Distel et al., 2008; Gunderson et al., 2011; Gunderson, Herpertz, et al., 2018; Kendler et al., 2011; Sanchez-Roige et al., 2018; Vukasović & Bratko, 2015); however, this has yet to be determined for the Section III model of BPD.

Regarding important external correlates of BPD, numerous studies have shown that a significant portion of individuals with BPD report having experienced maltreatment as children (Battle et al., 2004; Fossati et al., 1999; Ibrahim et al., 2018). For example, Zanarini et al. (1997) found that an overwhelming majority of BPD patients reported a history of either abuse (91%) or neglect (92%) prior to age 18. BPD relates to several forms of maltreatment (Battle et al., 2004), and there may be a dose–response relationship between borderline symptomatology and maltreatment severity (Ibrahim et al., 2018). In line with this, a slightly revised version of the Section III model of BPD was previously found to relate to cumulative childhood maltreatment (Bach & Fjeldsted, 2017), although relationships with specific subtypes of maltreatment (e.g., sexual, physical, emotional) were not assessed.

Suicidality, nonsuicidal self-injury, and alcohol and substance use disorders (AUD, SUD) are each robustly associated with borderline psychopathology. Suicidal behavior is a nonrequired criterion in the Section II categorical diagnosis of BPD (APA, 2013), and individuals with BPD are more likely to engage in nonsuicidal self-injury (Levine et al., 2020; Paris, 2019) and attempt and complete suicide (Soloff et al., 1994), with a standardized mortality rate 45 times greater than the general population (Chesney et al., 2014). Aouidad et al. (2020) recently showed that adolescents with BPD and multiple lifetime suicide attempts were more likely to self-mutilate, have attempted suicide at an earlier age, and made attempts that were more severe. Regarding alcohol and substance use, one review concluded that 57.4% of patients with BPD had a co-occurring SUD (Trull et al., 2000). Notably, many of these clinical phenotypes have previously been found to correlate with the Section III model of BPD in undergraduate and clinical samples (Anderson & Sellbom, 2015; Anderson et al., 2016; Bach & Fjeldsted, 2017).

Finally, increased risk-taking is thought to be a core component of BPD. Engagement in risky behaviors, such as reckless driving, promiscuous sexual activity, and excessive spending, appear among the Section II and Section III models of BPD (APA, 2013; Lieb et al., 2004; Paris, 2007). However, data supporting the relationship between BPD and risk-taking are mixed. Whereas women with BPD have been found to make riskier decisions than healthy controls on the Game of Dice Task (Svaldi et al., 2012) and a two-choice gambling task (Endrass et al., 2016), other studies have found that the performance of those with BPD and controls does not differ on the Balloon Analog Risk Task (BART; Coffey et al., 2011; Hüpen et al., 2020), which is a widely used and well-validated laboratory-based measure of risk-taking. Furthermore, PID-5 trait Risk-Taking has been found to only weakly correlate with Section II BPD (Anderson et al., 2014; Hopwood et al., 2012; Sellbom et al., 2014; Watters et al., 2019; Yam & Simms, 2014).

In sum, the primary goal of the present study is to evaluate the construct validity (and support the “nomological network”; Cronbach & Meehl, 1955) of the Section III model of BPD by assessing the familiality of the model and comparing it to external correlates related to borderline personality pathology. More specifically, the study focused on the Criterion B traits, which were measured with the PID-5. Using a community sample of adult biological siblings who were oversampled for psychopathology, we derived BPD latent factor scores from the PID-5 traits.

The present study also had several secondary goals. First, we examined whether gender moderated any of the effects, given that the prevalence of BPD is typically higher in women (APA, 2013; Torgersen et al., 2001; although see Holthausen & Habel, 2018, for a review of inconsistent gender differences across samples). Not only do prevalence rates commonly differ by gender, but categorically diagnosed BPD has been found to differently predict maladaptive behaviors—such as substance use and disordered eating (Johnson et al., 2003)—and expected correlates—such as aspects of social functioning (Silberschmidt et al., 2015)—as a function of gender. Thus, it is plausible that the extent to which the Section III BPD model relates to external correlates may vary by gender.

In addition, although the primary analyses focused on the Criterion B traits that are included in the current Section III model of BPD, a growing body of evidence indicates that additional trait facets may augment the BPD trait profile (Bach & Sellbom, 2016; Bach et al., 2016; Watters et al., 2019). Thus, in follow-up analyses, we tested a version of the Section III model of BPD that also included Suspiciousness and Perceptual Dysregulation. Finally, while the primary focus of the study was on BPD traits (Criterion B) rather that Criterion A (impairment in the domains of self- and interpersonal functioning), we also examined the degree to which the Section III model of BPD scaled with broad functional impairment as indexed by the World Health Organization Disability Assessment Schedule (WHODAS 2.0).

METHODS

PARTICIPANTS

The sample included 498 participants from a large family study of transdiagnostic mechanisms of psychopathology. Participants were recruited from the community and mental health clinics in Chicago, Illinois. Individuals qualified if they were between 18 and 30 years old and had a biological sibling also between 18 and 30 years old who was interested in participating in the study. Exclusion criteria were left-handedness, inability to read or write in English, a personal history of head trauma with loss of consciousness, and a personal history or first-degree family member with a history of manic, hypomanic, or psychotic symptoms. Participants were not recruited based on particular DSM-defined disorders, but were oversampled for severe internalizing psychopathology using the Depression Anxiety Stress Scales (DASS; Lovibond & Lovibond, 1995) during an initial phone screen. Efforts were also made to enroll a demographically and clinically diverse sample, including individuals with depression, anxiety, and substance use disorders (for full method details, see Correa et al., 2019; Weinberg & Shankman, 2017). Participant demographic and clinical characteristics are presented in Table 1.

TABLE 1.

Demographic and Clinical Characteristics

Variable	Mean (SD)
Age	22.4 (3.2)
Gender	63.9% female
Race/Ethnicity	White: 42.4%
	Black/African American: 15.7%
	Hispanic: 20.9%
	Asian: 11.8%
	Middle Eastern: 3.0%
	Mixed: 5.8%
	Other: 0.4%
CTQ
Total	35.96 (11.27)
Sexual Abuse	5.84 (2.97)
Physical Abuse	6.67 (2.45)
Emotional Abuse	8.34 (3.86)
Physical Neglect	6.59 (2.48)
Emotional Neglect	8.53 (3.96)
PID-5
Anxiousness	1.09 (0.74)
Emotional Lability	0.78 (0.72)
Hostility	0.70 (0.56)
Risk-Taking	1.31 (0.49)
Impulsivity	0.69 (0.63)
Separation Anxiety	0.59 (0.60)
Depressivity	0.37 (0.52)
SCID-5
Lifetime Suicidal Ideation	15.3% (n = 76)
Lifetime Nonsuicidal Self-injury	6.2% (n = 31)
Lifetime Alcohol Use Disorder	29.5% (n = 147)
Lifetime Substance Use Disorder	21.9% (n = 109)
BART
Total Pumps	21.00 (14.56)
Total Cash	72.70 (21.69)

Note. CTQ = Childhood Trauma Questionnaire; PID-5 = Personality Inventory for DSM-5; SCID-5 = Structured Clinical Interview for DSM-5; BART = Balloon Analogue Risk Task.

MEASURES

Personality Inventory for the DSM-5 (PID-5 Adults).

The PID-5 (Krueger et al., 2012) is an empirically derived 220 item self-report measure of pathological personality traits across five domains and 25 facets. It has demonstrated strong psychometric properties and high convergent validity with other personality measures (Thomas et al., 2013; A. G. C. Wright & Simms, 2015). Each item is rated on a 4-point Likert scale from 0 (very false or often false) to 3 (very true or often true). Personality facet subscales that are included in Criterion B of the Section III model of BPD were used to calculate BPD factor scores (APA, 2013; Anderson et al., 2016; Sellbom et al., 2014). These facets include Emotional lability, Anxiousness, Separation Insecurity, Depressivity, Impulsivity, and Hostility. PID-5 Risk-Taking has also been proposed to be a component of the Section III model of BPD, but as reported below was excluded in our calculations because it did not load onto a latent factor model of BPD. Cronbach’s alphas for the six facets ranged from 0.84 (Separation Insecurity) to 0.93 (Depressivity), all within the commonly accepted range (Nunnally, 1978), and was 0.96 for the BPD sum score.

Childhood Trauma Questionnaire (CTQ).

The short form version of the CTQ assesses childhood and adolescent (i.e., before the age of 18) experiences of maltreatment across five domains: sexual abuse, physical abuse, emotional abuse, physical neglect, and emotional neglect (reverse scored) and has demonstrated excellent psychometric properties in nonclinical and clinical samples (Bernstein & Fink, 1998; Bernstein et al., 2003). Each domain contains five items, scored on 5-point Likert scales from 1 (never true) to 5 (very often true). CTQ subscales can be summed into a total score or considered individually as dimensional and/or categorical measures of childhood maltreatment (e.g., “mild-to-moderate emotional neglect”). CTQ subscale Cronbach’s alphas ranged from .60 to .93, and for the full scale was .89 (see Supplemental Table S1). Cronbach’s alphas for the physical neglect and physical abuse subscales (.60 and .69, respectively) did not reach the commonly accepted cutoff of .70 (Nunnally, 1978), although this is consistent with previous findings (Scher et al., 2001).

Structured Clinical Interview for DSM-5 (SCID-5).

The Structured Clinical Interview for DSM-5 (SCID; First et al., 2015) was used to assess lifetime occurrence of suicidal ideation and nonsuicidal self-injury, as well as lifetime diagnoses of AUD and other SUD. SCID raters were required to (a) view a training video (First & Gibbon, 1996), (b) observe two to three interviews performed by an experienced rater, and (c) perform three interviews under the supervision of a licensed clinical psychologist or other advanced rater, wherein the trainee and supervisor reached 100% diagnostic agreement. As shown elsewhere (Shankman et al., 2018), diagnoses demonstrated high test–retest reliability.

Balloon Analog Risk Task (BART).

Participants completed the Balloon Analogue Risk Task (BART)–Auto Pump, a modified version of the original BART (Lejuez et al., 2002). In this task, 30 computerized balloons were presented one at a time, and participants typed in the number of “pumps” that they wanted to administer between 1 and 128. Participants were told they would receive two cents for each pump. Each balloon exploded beyond a certain number of pumps, but participants were not informed what the explosion point was. If the participant typed in a number that exceeded the explosion point, the balloon popped and the participant received no money. By contrast, if the entered number was below the explosion point, the balloon inflated and money was deposited into the participant’s “bank account.” Participants were told the amount of prize money they would receive at the end of the session depended on the amount of money accumulated during the task. Pumps total (i.e., the sum of pumps administered across all trials) and cash total (i.e., the sum of money won across all trials) were used as behavioral indicators of risk-taking propensity.

Functional Impairment.

Functional impairment was captured using items from the WHODAS 2.0 (Üstün et al., 2010), which has excellent psychometric properties (Üstün et al., 2010). Specifically, we measured impairment in work/school functioning (four items: e.g., “Because of your health condition, in the past 30 days, how much difficulty did you have in your day-to-day work/school?”) and interpersonal functioning (four items: e.g., “In the past 30 days, how much difficulty did you have maintaining a friendship?”).

DATA ANALYSIS

Confirmatory Factor Analysis (CFA) of BPD-Related Traits.

To identify whether a single-factor model provided a good fit for the BPD traits included in the Section III model, CFA was implemented using the lavaan R package (Rosseel, 2012). The BPD-related PID-5 scores (Anxiousness, Emotional Lability, Hostility, Risk-Taking, Impulsivity, Separation Insecurity, and Depressivity) were included as dependent variables in a single-factor model. Follow-up analyses also included Suspiciousness and Perceptual Dysregulation, given evidence that additional trait facets may augment the BPD trait profile (Watters et al., 2019). To account for the familial (nonindependent) relationship between siblings in the model, siblings were randomly assigned as either Sibling 1 or Sibling 2 and a latent BPD factor model was fit for each sibling group (BPD 1 and BPD 2) with manifest means, residual manifest variances, and cross-sibling manifest covariances constrained to be equivalent across sibling pairs (Funkhouser et al., 2021). Participants with missing sibling data were included by randomly assigning them as either Sibling 1 or Sibling 2, and the missing sibling data were accounted for by using full-information maximum likelihood (FIML) estimation. PID-5 data were available for 498 participants.

To assess model fit, the comparative fit index (CFI), root mean square error of approximation (RMSEA) index, and standardized root mean square residual (SRMR) index were used. CFI values closer to 1 indicate better fit, with values ≥ .95 considered to reflect a “good fit” (Hu & Bentler, 1999; West et al., 2012). RMSEA values ≤ .06 are considered acceptable (Hu & Bentler, 1999), whereas values ≥ .10 reflect a poor fit (Browne & Cudeck, 1992). SRMR values < .10 are indicative of acceptable fit, and values < .05 are indicative of good model fit (Hu & Bentler, 1999; Iacobucci, 2010). Given that each participant was randomly assigned to sibling group 1 or 2 to estimate the latent BPD factor, which can suppress model fit estimates, model fit values were adjusted following Olsen and Kenny (2006). Factors weights from the best-fitting model were carried forward to the ICC and regression analyses.

Intraclass Correlation Coefficients (ICCs) to Test Familial Associations.

ICCs were computed for the BPD-related traits and BPD factor score among siblings (232 sibling pairs). Agreement between the sibling pairs (i.e., siblings rated themselves, not each other) was assessed with a one-way random effects model (ICCs [1,1]). We also estimated narrow-sense heritability (i.e., the portion of phenotypic variance accounted for by additive genetic effects), which can be estimated using correlations among sibling pairs (Visscher et al., 2008) using the following formula: h² = 2*r_xy, where r_xy is the ICC between siblings.

Regression Analyses.

Analyses were conducted in R using the lme4 package (Bates et al., 2015). The lmer function was used for the linear mixed effects regression models, and the glmer function was used for the logistic mixed effects regression models. For all regression analyses, gender was included as a covariate and family was included as a random effects factor to account for familial relation among sibling pairs.

Six separate linear mixed effects regression models were implemented for the childhood maltreatment total score and the five CTQ subtypes. In the first model, dimensional CTQ scores were summed together across all maltreatment domains (i.e., CTQ total). Total (dimensional) scores were also used for each domain except sexual abuse, which was dichotomized (coded as no sexual abuse history: CTQ sexual abuse score < 6 = −1, and sexual abuse history: CTQ sexual abuse score ≥ 6 = 1) due to low dimensional score variance. CTQ data were available for 490 participants. Four separate logistic mixed effects regression models were implemented for suicidality, nonsuicidal self-injury, AUD, and SUD. From the SCID-5, suicidality, AUD, and SUD data were available for 498 participants, and nonsuicidal self-injury data were available for 497 participants. Two separate linear mixed effects regression models were implemented for (1) total pumps and (2) cash total. BART behavioral data were available for 441 participants.

Finally, three separate linear mixed effects regression models predicting WHODAS overall functioning were implemented specifically assessing: (1) work/school functioning, (2) interpersonal functioning, and (3) a model with both functioning domains in the same model to examine specificity to separate domains of functioning.

Gender Moderation.

All regression analyses were rerun with gender as a moderator to identify whether relationships between BPD and the dependent variables differed for females and males.

Operationalizing BPD as Sum of Traits.

ICC and regression analyses were repeated using a z-scored composite measure of the BPD-related traits to identify whether results differed based on the approach used to compute BPD factor scores.¹

RESULTS

SAMPLE DEMOGRAPHICS

The average age of participants was 22.4 years old (SD = 3.2), 64% self-identified as female, and the sample was racially/ethnically heterogeneous (42.4% self-identified as White; see Table 1). Means and standard deviations for the personality measure, childhood maltreatment, and BART performance are reported in Table 1. Frequencies for the presence/absence of lifetime suicidal ideation, nonsuicidal self-injury, and for a DSM-5 diagnosis of AUD and SUD are also reported in Table 1. Childhood maltreatment scores were slightly higher than those of previous studies that assessed maltreatment within community samples (e.g., CTQ Total scores for males and females were in the 75th percentile range of community norms; Scher et al., 2001). Mean scores for all BPD traits on the PID-5 were within one standard deviation of published community norms (Krueger et al., 2012).

LATENT FACTOR MODEL: CONFIRMATORY FACTOR ANALYSIS

None of the model fit indices for the single-factor model with seven BPD indicators (i.e., including risk-taking) were in the good range, and only one value (SRMR) fell within the acceptable range (0.81, 0.11, and 0.09 for CFI, RMSEA, and SRMR, respectively). The individual factor loadings indicated that poor fit was primarily driven by one facet, Risk-Taking, which had a standardized factor loading of 0.07 (see Figure S1). By contrast, all other items had loadings greater than 0.40 (range: 0.43–0.85). After removing Risk-Taking, all model fit indices were in the good-to-acceptable range (0.95, 0.06, and 0.05 for CFI, RMSEA, and SRMR, correction applied). Factor weights from the single-factor model with six indicators were carried forward to compute ICCs and to the regression analyses (see Figure 1).

FIGURE 1.

Latent factor model of borderline personality disorder (BPD) with six indicators (Emotional Lability, Anxiousness, Separation Insecurity, Depressivity, Impulsivity, and Hostility–Risk-Taking) removed.

FAMILIAL CONCORDANCE OF BPD

There was significant sibling agreement for all individual BPD-related traits except Emotional Lability, with significant ICCs ranging between 0.12 and 0.19 (with narrow-sense heritability estimates ranging from 0.23 to 0.38; see Table 2). There was also significant sibling agreement for the BPD factor score, with an ICC of 0.16 (and a narrow-sense heritability estimate of 0.32; see Table 2).

TABLE 2.

Within-Family Associations for Borderline Personality Disorder (BPD) Latent Factor With Six Indicators and Constituent Traits

Statistic	Anxiousness	Emotional Lability	Hostility	Impulsivity	Separation Insecurity	Depressivity	BPD Latent Factor
ICC(1)	0.175	0.080	0.116	0.123	0.185	0.190	0.158
F test [222,222]	F = 1.43, p = .004	F = 1.17, p = .112	F = 1.26, p = .038	F = 1.28, p = .030	F = 1.45, p = .002	F = 1.47, p = .002	F = 1.37, p = .008
95% CI	0.048 < ICC < 0.297	−0.049 < ICC < 0.206	0.013 < ICC < 0.241	0.005 < ICC < 0.248	0.058 < ICC < 0.306	0.063 < ICC < 0.311	0.030 < ICC < 0.280
h2	0.35	0.16	0.23	0.25	0.37	0.38	0.32

CHILDHOOD MALTREATMENT

BPD latent factor scores were significantly related to cumulative childhood maltreatment and to all childhood maltreatment subtypes, except for sexual abuse (see Table 3). All results were in the same direction, with higher levels of BPD related to more childhood maltreatment. Results did not differ by gender (see Table 4).

TABLE 3.

Linear Regression Results for the BPD Latent Factor With Six Indicators (Emotional Lability, Anxiousness, Separation Insecurity, Depressivity, Impulsivity, With Hostility–Risk-Taking Removed)

Childhood Maltreatment (CTQ)
Variable	BPD (latent trait)
Total	B = .25, p < .001
Sexual Abuse	B = .02, p = .595
Physical Abuse	B = .16, p < .001
Emotional Abuse	B = .34, p < .001
Physical Neglect	B = .13, p = .004
Emotional Neglect	B = .23, p < .001
SCID-5 Symptom and Diagnòstic Status (Lifetime)
Variable	BPD
Suicidal Ideation	B = .28, p < .001
Nonsuicidal Self-Injury	B = .15, p = .001
Alcohol Use Disorder	B = .20, p < .001
Substance Use Disorder	B = .20, p < .001
Risk-Taking (BART)
Variable	BPD
Total Pumps	B = −.06, p = .163
Total Cash	B = −.02, p = .730

Note. CTQ = Childhood Trauma Questionnaire; SCID-5 = Structured Clinical Interview for DSM-5; BART = Balloon Analogue Risk Task.

TABLE 4.

Linear Regression Results for Gender Moderation (Latent Factor With Six Indicators: EmotionalLability, Anxiousness, Separation Insecurity, Depressivity, Impulsivity, With Hostility–Risk-Taking Removed)

Childhood Maltreatment
Variable	BPD (latent trait)	Gender	BPD × Gender
CTQ Total	B = .25, p < .001	B = .01, p = .762	B = −.01, p = .837
CTQ SA	B = .03, p = .454	B = .10, p = .027	B = −.02, p = .532
CTQ PA	B = .16, p < .001	B = .01, p = .894	B = −.04, p = .326
CTQ EA	B = .34, p < .001	B = .03, p = .487	B = .03, p = .493
CTQ PN	B = .13, p = .004	B = −.05, p = .251	B = .01, p = .890
CTQ EN	B = .23, p < .001	B = −.04, p = .367	B = −.02, p = .726
SCID-5 Symptoms and Diagnostic Status (Lifetime)
Variable	BPD	Gender	BPD × Gender
Suicidal Ideation	B = .28, p < .001	B = .04, p = .332	B = .04, p = .346
Nonsuicidal Self-injury	B = .15, p < .001	B = .15, p < .001	B = .08, p = .069
Alcohol Use Disorder	B = .19, p < .001	B = −.18, p < .001	B = .04, p = .382
Substance Use Disorder	B = .20, p < .001	B = −.10, p = .003	B = −.01, p = .809
Risk-Taking (BART)
Variable	BPD	Gender	BPD × Gender
Total Pumps	B = −.06, p = .169	B = −.25, p < .001	B = −.03, p = .517
Cash Total	B = −.02, p = .721	B = −.07, p = .136	B = .02, p = .662

Note. CTQ = Childhood Trauma Questionnaire; SCID-5 = Structured Clinical Interview for DSM-5; BART = Balloon Analogue Risk Task. SA = Sexual Abuse; PA = Physical Abuse; EA = Emotional Abuse; PN = Physical Neglect; EN = Emotional Neglect.

SUICIDALITY, NONSUICIDAL SELF-INJURY, ALCOHOL USE, AND SUBSTANCE USE

Significant relationships emerged between BPD factor scores and lifetime endorsement of suicidal ideation and nonsuicidal self-injury, and a lifetime diagnosis of AUD and SUD (see Table 3). Results were all in the same direction, with higher levels of BPD related to higher lifetime incidence of suicidal ideation, nonsuicidal self-injury, AUD, and SUD. Results did not vary as a function of gender (see Table 4).

RISK-TAKING

Borderline PD did not predict either pumps total or cash total on the BART (see Table 3). Results did not vary by gender (see Table 4).

OPERATIONALIZING BPD AS SUM OF TRAITS

The above results using a latent factor score for BPD did not differ when using a z score composite of the BPD traits (i.e., all significant/nonsignificant results remained; for full results, see Supplemental Tables S2–S4).

FOLLOW-UP ANALYSES

Section III Traits Plus Suspiciousness and Perceptual Dysregulation.

The revised model did not outperform the six-factor model. However, model fit indices were in the good-to-acceptable range (0.93, 0.07, and 0.05 for CFI, RMSEA, and SRMR; see Supplemental Table S5 and Figure S2). Familiality was slightly higher for the six-trait latent factor model than the revised model (ICCs: 0.158 vs. 0.128; see Table S6), while the external correlate results were generally comparable (see Table S7).

Functional Impairment.

In separate models, the six-trait latent factor strongly predicted greater impairment in (1) work/school functioning, t(476) = 9.96, B = .41, p < .001, and (2) interpersonal functioning, t(476) = 13.47, B = .52, p < .001. When both domains of functioning were included, the association with the six-trait latent BPD factor remained for work/school impairment, t(475) = 3.31, B = .14, p = .001, and interpersonal impairment, t(475) = 8.82, B = .32, p < .001.

DISCUSSION

The results of the present study demonstrate that the personality traits included in the Section III model of BPD run in families (separately and in aggregate) and that the BPD trait profile is associated with several expected correlates of borderline personality pathology, as well as with significant functional impairment in work/school and interpersonal domains. These findings contribute to a growing body of evidence supporting the validity of the trait operationalizations of BPD.

Since the classic work of Robins and Guze (1970), familiality has been considered an important validator of diagnostic constructs, and previous work has established heritability of the categorical BPD diagnosis (e.g., Distel et al., 2008; Gunderson et al., 2011; Skoglund et al., 2021). Importantly, in our study, significant sibling agreement emerged for the latent factor BPD score and for all but one of its traits (Emotional Lability). To our knowledge, this is the first study to examine the familiality of a latent BPD construct derived from the DSM-5 Section III traits. Our heritability estimates (which ranged from 23% to 38%, excluding Emotional Lability) are in line with previous heritability estimates of the PID-5 facets that comprise the latent BPD factor (27% to 45%; Z. E. Wright et al., 2017). The heritability estimate of the latent factor BPD model was 32%, which is lower than some existing estimates for categorical BPD diagnosis (e.g., Distel et al., 2008; Skoglund et al., 2021). However, this may be explained, in part, by the approach we used to estimate heritability. Nontwin sibling correlations allow for the estimation of narrow-sense heritability (i.e., the proportion of phenotypic variance accounted for by additive genetic influences), but do not capture the broad-sense heritability (i.e., the proportion of phenotypic variance accounted for by genetic variance, including dominance and gene interaction effects), which can only be estimated using other approaches (e.g., twin studies) (Visscher, 2004; Visscher et al., 2008; e.g., Skoglund et al., 2021). The finding that the Section III BPD trait profile aggregates in families (here, among siblings) in the broader population is novel (if not unexpected) and is consistent with existing literature.

The present study also identified significant relationships between the Section III BPD trait profile and several clinically relevant correlates known to be associated with borderline personality pathology, including (1) cumulative childhood maltreatment and all maltreatment subtypes, except sexual abuse; (2) lifetime suicidal ideation and nonsuicidal self-injury; and (3) lifetime AUD and SUD. Our results are largely consistent with the available studies that have examined relationships between Section III BPD traits and related variables (Anderson & Sellbom, 2015; Anderson et al., 2016; Boland et al., 2018; Miller et al., 2015). For example, although Bach and Fjeldsted (2017) did not look at specific childhood maltreatment subtypes, they found that BPD traits measured via the PID-5 were significantly related to greater cumulative maltreatment. Results are also consistent with Anderson and colleagues (2016), who found significant correlations between a Section III latent factor and a history of various forms of mistreatment (e.g., physical abuse, emotional neglect, physical abuse). In contrast with previous research (for review, see de Aquino Ferreira et al., 2018), however, we did not find a relationship between the Section III BPD trait profile and childhood sexual abuse. This may be attributable to the fact that previous studies have typically used clinical samples, which generally have higher rates of sexual abuse than nonclinical/community samples (Bernstein et al., 2003; Devi et al., 2019). Notably, although Anderson and colleagues (2016) did indeed find a relationship between the Section III BPD trait profile and sexual abuse, the correlation in their clinical sample was small (r = .21). With respect to suicidality, self-harm, alcohol and substance abuse, Anderson and Sellbom (2015) and Anderson et al. (2016) previously identified significant relationships between Section III BPD traits and these variables in clinical and undergraduate samples. Thus, our findings contribute to a growing body of work supporting the construct validity of the Section III BPD trait profile.

We used factor analysis to derive a single-factor model of BPD, which is an innovative approach to capturing and assessing the Section III BPD model (although not unique to this study; see Anderson et al., 2016, and Sellbom et al., 2014). The model fit with all seven BPD-related traits currently included in Section III was poor. This was primarily driven by the weak loading of Risk-Taking, which is consistent with several previous studies (Anderson et al., 2014; Evans & Simms, 2018; Orbons et al., 2019). After Risk-Taking was removed, the model exhibited good fit. Because Suspiciousness and Perceptual Dysregulation have previously been found to account for incremental variance in external correlates (Bach & Sellbom, 2016; Bach et al., 2016; Sellbom et al., 2014; for review, see Watters et al., 2019), we reran analyses with these traits included. The revised model exhibited slightly poorer model fit and lower familiality than the six-factor model; however, there were generally no substantive changes to the results with this expanded set of traits. Although the six-factor model is more parsimonious, the revised model captures features that are not included in Section II’s BPD diagnostic criteria that have been found to discriminate clinical from nonclinical status (e.g., Symptom 9; Bach et al., 2016). Thus, future studies should continue to investigate the model fit and utility of alternative latent factor models.

Risk-taking has been considered to be a core aspect of borderline personality pathology (Lieb et al., 2004). However, this has not been consistently borne out in the data, as reviewed above. Moreover, in contrast to other personality facets included in the Section III model of BPD, studies have found relatively weak correlations between PID-5 Risk-Taking and measures of borderline personality (Anderson et al., 2014; Hopwood et al., 2012; Sellbom et al., 2014; Yam & Simms, 2014). This was further demonstrated in a meta-analysis by Watters et al. (2019), which found that risk-taking was the lone proposed BPD trait that did not meaningfully correlate with Section II symptoms (r = .22). It is plausible that the risk-taking behaviors observed in BPD are not well captured by the operationalization of risk-taking in the PID-5. For example, most of the PID-5 items that probe risk-taking require a self-perception characterized as “risky” (e.g., “I do what I want regardless of how unsafe it might be”). Perhaps individuals with borderline personality tend not to view themselves as risk-takers, or only when they are upset (Beauchaine et al., 2019; Klonsky, 2007). Also, none of the PID-5 Risk-Taking items refer specifically to the sorts of risky behaviors often described in BPD (e.g., nonsuicidal self-injury, suicidal gestures, or promiscuous sex). If these factors are indeed responsible for weak associations between PID-5 Risk-Taking and BPD, future studies could perhaps supplement the PID-5 risk-taking scale with additional items.

Furthermore, it is noteworthy that the behavioral measure of risk-taking employed in this study (i.e., the BART) did not exhibit a significant relationship with the Section III latent BPD factor. Although this is consistent with other studies that did not find a difference in BART performance between individuals with categorically defined BPD and controls (Coffey et al., 2011; Hüpen et al., 2020), to our knowledge this is the first study to assess BART performance in relation to the Section III BPD trait profile. One possible explanation for these negative findings is that emotion regulation difficulties (reviewed by Bortolla et al., 2020) may lead patients with BPD to make risky decisions when their affect is dysregulated (Beauchaine et al., 2019; Klonsky, 2007), but not when their affect is regulated—for instance, during a relatively affectively neutral task (such as the BART). Notably, although Hüpen et al. (2020) did not find that individuals with BPD and healthy controls differed on BART performance, higher physiological activity (skin conductance response [SCR]) among individuals with BPD, but not healthy controls, was related to poorer decision-making. To capture state-mediated difficulties in BPD, it may be necessary to “affectively activate” individuals with borderline personality pathology and then probe risk-taking with the BART or other behavioral paradigms.

There are several limitations of the present study. Although not a limitation per se, because we sought to examine the Section III model of BPD in a community sample, the results might not generalize to more severe populations—including inpatient samples, which often include individuals with BPD. However, dimensional models conceptualize psychopathology across a range of severity, and therefore the findings may indeed apply to more severe cases of borderline personality pathology. The sample did not include older adults, which limits generalizability to borderline pathology at later stages of life (see Videler et al., 2019). Furthermore, the exclusion of those with a personal or close family history of manic or psychotic symptoms may select out some individuals with BPD who experience transient stress-related psychotic symptoms (for review, see Barnow et al., 2010). It is also worth considering whether characteristics of the sample may have contributed to the negative findings with respect to risk-taking. Although recruitment efforts also targeted those with alcohol and substance use history (in addition to internalizing), we did not recruit for externalizing more broadly. This is notable because externalizing has been shown to be most strongly associated with disinhibitory Section III traits such as risk-taking (Anderson & Sellbom, 2015).

In addition, our study did not compare the Section III trait operationalization of BPD to the categorical model or other models that incorporate personality traits. Thus, we cannot draw conclusions regarding the incremental validity of the Section III model of BPD, although other groups have pursued this line of investigation (e.g., Anderson et al., 2016; Boland et al., 2018). It is also worth noting again that we did not find a relationship between the BPD latent factor and sexual abuse. Given that rates of sexual abuse in our sample exhibited low dimensional variance, future studies with samples enriched for sexual abuse may be needed.

Finally, despite overlap between the WHODAS items that we used and Criterion A of the Alternative DSM-5 Model of Personality Disorders for BPD, we are unable to draw specific conclusions about Criterion A. Increasingly, the importance of assessing personality-related impairment in addition to the specific personality traits has been emphasized (Anderson & Sellbom, 2018; Hutsebaut et al., 2016; Widiger et al., 2019), and several measures of personality-related impairment have been developed (Alternative Model of Personality Disorders-BPD-specific impairment: Anderson & Sellbom, 2018; DSM-5 Level of Personality Functioning Scale–Brief Form: Hutsebaut et al., 2016, and Weekers et al., 2019). Future work should incorporate these or other measures to assess the full Section III model of BPD (Criterion A and Criterion B), because both criteria are clinically relevant and required for a diagnosis of BPD.

In conclusion, the present findings lend support to the construct validity of DSM-5 Section III’s operationalization of BPD traits. In a large, transdiagnostic community sample, a latent borderline personality pathology factor derived from the traits included in the Section III model of BPD was shown to aggregate in families and relate to expected external correlates of BPD, including childhood maltreatment, suicidal behavior, nonsuicidal self-injury, and alcohol and substance abuse. Future directions include assessing other validators of BPD and exploring whether traits should be added or removed from the Section III model. Ultimately, clinical decisions involve making a yes/no (categorical) decision (e.g., does one treat or not, treat at dose A or dose B, hospitalize or not), and it will be important to determine whether models of BPD that incorporate traits improve treatment decisions of BPD pathology beyond the traditional approach.