Participants and procedure

DSM-5 Section II BPD criteria

The DSM-5 Section II criteria for BPD were systematically assessed using the Structured Clinical Interview for DSM–IV Axis II Personality Disorders [SCID-II] [23]. BPD features were measured dimensionally by adding the number of fulfilled criteria for each category (criterion count). All SCID-II interviews were performed and recorded independently of the computerized administration and scoring of the self-reported childhood traumas and pathological personality traits. Thirteen randomly selected SCID-II interviews were inter-rated by a blinded psychologist, and optimal interrater reliability was identified on the basis of criterion-counts, with a two-way random intra-class correlation coefficient for single rater of .977 (confidence interval = .927-.993; p > .001). See Additional file 1: Table S1 and Additional file 2: Table S2 for base rates and distribution of BPD diagnostic criteria.

DSM-5 Section III BPD traits

The DSM–5 Section III traits for BPD were assessed using the Personality Inventory for DSM-5 [PID-5] [24, 25]. The PID-5 is a 220-item self-report inventory measuring the Criterion B (i.e., 25 trait facets and five higher-order trait domains) of the alternative DSM–5 model for PDs. Patients were required to rate each PID-5 item on a 4-point likert-scale from 0 (Very False or Often False) to 3 (Very True or Often True). We used the Danish version of the PID-5, which has demonstrated acceptable psychometric properties [26, 27] and continuity with categorical PDs [28], including specific ability to differentiate BPD from other PDs [29, 30]. In the current study we exclusively employed 9 PID-5 trait facets that were designated to describe features of BPD. First, we included the 7 trait facets that are specified for BPD in the official DSM-5 Section III trait-to-disorder cross-walk (i.e., Emotional Lability, Anxiousness, Separation Insecurity, Depressivity, Impulsivity, Risk Taking, and Hostility [31]). Additionally, we included Suspiciousness and Cognitive & Perceptual Dysregulation reflecting the ninth Section II diagnostic criterion of BPD comprising stress-induced paranoid ideations and/or dissociative experiences, which has been supported by previous findings [29, 30, 32–34]¹. In the present study, alpha coefficients for the nine facets ranged from .78 (Emotional Lability) to .92 (Depressivity) with a mean alpha of .87.

Level of suicidal risk

Suicidal risk was assessed using the “Suicidality” module of Mini International Neuropsychiatric Interview 6.0 (MINI; Sheehan et al. [35]). In this structured interview module, the patients are probed for various aspects of suicidality and suicide risk factors within the last month (e.g., suicidal thoughts or plans, deliberate self-harm with or without expectation to die, and history of suicidal attempts). Subsequently, the obtained information are added up and scored as “Low”, “Moderate”, or “High” suicide risk. The suicidality module of the MINI diagnostic interview has demonstrated sound psychometric and predictive features in relation to suicidality and risk assessment [35, 36].

Childhood trauma

Childhood traumas were measured with the Childhood Trauma Questionnaire (CTQ; Bernstein & Fink, [37]). The CTQ is a 25-item self-report inventory (excluding 3 validity items) containing retrospective questions concerning childhood abuse and neglect. On a 5-point Likert-type scale (“Never true” to “Very often true”) the respondent indicates how well each item reflects what they have experienced during their childhood. In the current study we computed a total score by averaging all item responses. A number of international studies suggests that the reliability and validity of the CTQ is excellent [38]. In the present study, the total CTQ scale score had an alpha coefficient of .93, and a mean corrected item-total coefficient of .57 with all coefficients above .40 (except “item 9” [.37] describing severe physical violence).

Bivariate associations

Mediational analysis

We employed mediation analysis to explore the mediating effects of DSM-5 Section II BPD criterion-count and DSM-5 Section III BPD traits, separately, in the relationship between childhood trauma and level of suicidal risk. In mediation analysis different effects are considered: The total effect of an independent variable (IV) on a dependent variable (DV) is composed of the direct effect of the IV on the DV and the indirect effect through a proposed mediator variable. First, we examined the direct and indirect effect of childhood trauma on level of suicidal risk through DSM-5 Section II BPD criterion count. Next, we examined the direct and indirect effect of childhood trauma on level of suicidal risk through the nine designated DSM-5 Section III BPD traits, simultaneously, while controlling for the potential mediating effects of the other traits. Based on recommendations by Hayes [39] and Mackinnon et al. [40], a bootstrapping sampling procedure was applied for assessing effects. Importantly, bootstrapping is a nonparametric approach that can provide more accurate inferences when the data are not well behaved, as in the current study, in terms of distributional assumptions (e.g., non-normality) and when the sample size is small [41]. The reported coefficients and effects were considered significant if zero was not included in the 95% bias-corrected confidence interval. All effects were adjusted for the influence of gender, age, and educational level (see Table 1). Multicollinearity of independent variables was unlikely to be an issue (all VIF < 2.5; tolerances > .40). In the current study we used PROCESS version 1.12.1 for SPSS provided by Hayes [39].

The mediational model for DSM-5 Section II BPD criterion-count is displayed in Fig. 1a. Both the a-path and the b-path were significant, while the c-path (total effect) did not remain significant when the effect of BPD was taken into account (c'-path; direct effect). As presented in Additional file 3: Table S3, the BPD criterion-count score explained 67% of the total effect of childhood trauma on suicide risk. We further tested the specificity of BPD as mediator by redoing the analyses using each of the remaining 9 PDs as mediating variables, individually, which only yielded significant but inferior effects for Dependent PD (.002), Paranoid PD (.002), and Schizotypal PD (.003). Subsequently, when performing a mediation analysis with Borderline, Dependent, Paranoid, and Schizotypal PDs in one simultaneous model, only the effects of Borderline (.010) and Dependent (.002) remained significant. This suggests that BPD is the core mediator of the relationship in question.

The mediational model for DSM-5 Section III BPD traits is displayed in Fig. 1b. The trait facets of Depressivity and Suspiciousness showed significant a-paths and b-paths, and the c-path (direct effect) did not remain significant when controlling for the effect of BPD trait facets. As presented in Additional file 3: Table S3, the total DSM-5 Section III BPD composite trait score accounted for 82% of the total effect of childhood trauma on suicidal risk. Specifically, the trait facet of Depressivity explained 52% whereas Perceptual Dysregulation explained 37% of this effect. As the PID-5 facets of Suspiciousness and Perceptual Dysregulation are not included in the official definition of the DSM-5 Section III BPD type, we tentatively retested the model without these two facets. Consequently, the total indirect effect remained significant and relatively stable (decreased from .012 to .010), and the direct effect remained non-significant. Apart from the facet of Depressivity no other facets emerged as significant mediators in this model.

As both DSM-5 Section II (BPD Criterion 5; "recurrent suicidal behavior, gestures, or threats, or self-mutilating behavior") and Section III (PID-5 Depressivity facet; items 81, 119, and 178) explicitly refer to potential suicidality, it may not seem surprising that these features are associated with suicidal risk. Therefore, to test for possible circularity we also performed the analyses without this suicidality-related content. First, we computed a BPD criterion count without criterion number 5, and included this modified score in the mediation model. Notably, the mediating effect of BPD remained significant and was only reduced from .009 to .007. Second, when excluding the PID-5 Depressivity items that explicitly refer to suicidality (items 81, 119, and 178), the mediating effect of Depressivity remained significant but decreased from .007 to .005, whereas the mediating effect of perceptual dysregulation increased slightly from .0047 to .0053; no other facets emerged as significant mediators in this model. In this modified model, the total indirect effect of all 9 PID-5 facets decreased slightly from .0118 to. 0113 and remained significant, and the direct effect remained non-significant.

As an additional post hoc analysis, we also examined the mediational effect of DSM-5 Section II BPD criterion count versus DSM-5 Section III BPD composite score in one joint parallel mediation model (see Fig. 2) while controlling for the effect of one another. In this model, only DSM-5 Section III traits showed significant a-paths and b-paths, whereas the DSM-5 Section II traits only showed significant a-paths (association between childhood trauma and BPD). The contrast between the two mediational paths (.011) was statistically significant suggesting that the explanatory power of DSM-5 Section III traits is superior to DSM-5 Section II criterion-counts in terms of mediational effect.

Strengths, limitations, and future directions

The main strength of the present study is the appropriate inclusion of psychiatric patients with BPD features, the utilization of DSM-5 Section II and III, and the employment of both self-report and interview-rated measures. This is both a significant and novel addition to understanding the potential role of BPD in relation to suicidal risk. However, certain limitations and recommendations for future research should be emphasized.

First, the use of mediation analysis on cross-sectional data has been questioned as one can find evidence for indirect effects in cross-sectional data, even when the true indirect effect in longitudinal data is zero, and vice versa [51]. However, in the present study mediation analysis was employed in accordance with recommendations by Hayes [39] and MacKinnon [40] as an initial attempt to test a proposed model. Moreover, our objective was also to compare the mediational effect exerted by DSM-5 Section II and II BPD features, respectively. Thus, the findings provided preliminary support for the model, but longitudinal research is necessary for a conclusive test of mediation. A more definitive test of our mediational model would require that childhood traumas be recorded during childhood and early adolescence, that BPD pathology be assessed during mid-adolescence, and that level of suicidal risk be assessed during adulthood. However, taken together with the many studies indicating that childhood traumas are associated with BPD, and that BPD predicts elevated suicidal risk, our findings suggest that future research may be likely to confirm that BPD, in particular features of Depressivity (i.e., shame, guilt, pessimism, and self-loathing) and Perceptual Dysregulation (i.e., dissociation) mediate a substantial part of the effect of childhood traumas on suicidal risk in adulthood.

Second, because childhood traumas were measured with a questionnaire, we were unable to differentiate between the effects of actual and perceived childhood events. It is possible that the results would have been different if reports from multiple informants (e.g., parents, siblings) had been available. Moreover, the retrospectively recollected reporting may have been influenced by state-dependent memory and recall bias reflecting personality and current mental condition. However, in a comparative study Scott et al. [52] found no significant difference between prospective records and retrospective self-reports of childhood maltreatment. Additionally, in the current study we did not assess inter-rater reliability of the MINI Suicidal Risk measure.