Psychometric properties of the French borderline symptom list, short form (BSL-23)

Background The short form of the Borderline Symptom List (BSL-23) is a self-rating instrument used to assess specific symptoms of borderline personality disorder (BPD). The original German version has shown good psychometric proprieties. The BSL-23 can also be used to measure the effects of therapy on patients with BPD. The aim of this study was to assess the psychometric properties of the French version of the BSL-23. Methods The French version of the BSL-23 was given to 265 subjects with BPD. Factor structure, reliability, test-retest stability, convergent validity, divergent validity, and sensitivity to change were analysed. Forty-five subjects suffering from attention-deficit hyperactivity disorder (ADHD) were used as controls to evaluate the specificity of BSL-23. Results A one-factor structure was obtained in the French version of the BSL-23, showing high internal consistency (Cronbach’s alpha = .94) and test-retest reliability (r = .841). The French version of the BSL-23 was highly correlated with depression severity, hopelessness, anger, motor impulsiveness, and BPD diagnosis. It was an efficient tool to discriminate between BPD patients and ADHD patients, and showed good sensitivity to change in a group of BPD patients who took part in a one-month DBT intervention. Conclusions The French version of the BSL-23 shows similar psychometric properties as the original German version. This study therefore provides clinicians and researchers with a French instrument to measure BPD symptomatology.


Impulsivity
Background Borderline personality disorder (BPD) is characterized by emotional dysregulation, impulsivity, self-damaging and suicidal behaviours, interpersonal difficulties, and identity disturbance. The lifetime prevalence of BPD among the general population varies according to surveys, diagnostic instruments and rules, but it is estimated to be in between 0.7 % and 2.7 % in recent studies [1][2][3]. BPD is a severe condition that causes major impairments in a variety of contexts and is associated with poor socio-economic and familial outcomes [4]. BPD patients are frequent users of mental health services and their mortality rate by suicide reaches 10 % [5]. The relevance of early diagnosis has been demonstrated by Kaess et al. [6], who showed that BPD can be identified during adolescence.
In order to establish a diagnosis for BPD based on the DSM-5 criteria [7], clinicians and researchers commonly use structured interviews, such as the Structured Clinical Interview for DSM-5 Axis II Personality Disorders (SCID-II) [8]. In addition to the interviews, clinician-rated scales, such as the Zanarini Rating Scale for Borderline Personality Disorder [9] and the Borderline Personality Disorder Severity Index [10], are used as screening tools for BPD symptomatology or as instruments assessing changes in the severity of the disorder. Finally, self-report scales have been developed to take into account the subjective view patients have of their disorder. These scales include, but are not limited to, the Borderline Evaluation of Severity over Time [11] and the Borderline Symptom List (BSL) [12,13], both quantifying borderline-specific symptomatology.
The initial BSL [12,13] included 95 items based on the criteria of the DSM-4, the Diagnostic Interview for Borderlines-Revised (DIB-R) [14], and the opinions of clinical experts and BPD patients. Each item describes a complaint frequently made by BPD patients, such as "I was lonely" or "I experienced stressful inner tension". The patients are asked to evaluate the intensity of each complaint over the previous week on a 5-point Likert scale, ranging from 0 (none) to 4 (very strong). The BSL-95 showed good psychometric properties [13]. No particular effect of gender, age or level of education was found. Based on the 95-item scale in German, a shorter, 23item version of the BSL was developed [15,16]. The BSL-23 was validated on different samples representing a total of 659 BPD patients. The psychometric properties of the BSL-23 were similar to those of the BSL-95 and the correlation between the two versions of the scale was high (range: 0.958-0.963 in five different samples). The factor analysis showed a one-factor structure, and the internal consistency of the BSL-23, as well as its test-retest reliability, were more than satisfactory (Cronbach's alpha: 0.935-0.969 and r = 0.82; p < 0.0001). Furthermore, the BSL-23 discriminated BPD patients from healthy subjects and from patients suffering from other psychiatric disorders. It showed a positive correlation with measures of psychopathology, depression and anxiety, and a negative correlation with a measure of global well-being. Finally, the BSL-23 was sensitive to change [15,16] after three months of dialectical behavior therapy (DBT) [17]. The BSL-23 was also translated and validated in Spanish. The Spanish BSL-23 [18] replicated the one-factor structure of the original version and was found to be a reliable and valid instrument for assessing BPD severity and sensitivity to change. Moreover, it also correlated with depressive symptomatology, state and trait anxiety, hostility and impulsivity scores.
Only a few validated instruments are available in French to specifically assess the severity of BPD symptomatology, and to our knowledge there are no validated self-report scales currently available. The BSL-23 is a brief, sensitive, easy to use and specific instrument, which can be repeated to assess changes in the severity of the disorder over time. Our aim was to examine the structure and psychometric properties of a French version of the BSL-23 on a sample of BPD patients. Factor structure, internal consistency, and test-retest reliability were assessed. Correlations between the French BSL-23 and other psychiatric symptoms were explored. Furthermore, the instrument's relevance in discriminating BPD patients from a sample of patients suffering from attention deficit and hyperactive disorder (ADHD) was also examined. Finally, sensitivity to change after a four-week Intensive DBT (I-DBT) [19,20] was tested on a sample of 92 BPD patients.

Methods
Participants BPD patients were recruited in two specialized outpatient units (Geneva and Fribourg) treating patients suffering from BPD and/or ADHD and relying on DBT as a first-line treatment. Patients were interviewed by general practitioners or psychiatrists to assess emotion dysregulation, impulsive behaviours, self-damaging behaviours, and/or suspicion of BPD. Each patient was interviewed first by a trained psychiatrist or psychologist, and then assessed for psychiatric disorders with the Diagnostic Interview for Genetic Studies (DIGS) [21], as part of a broader study investigating genetic and epigenetic correlates of BPD [22]. BPD was assessed with the BPD part of the SCID-II [8]. Only subjects filling the criteria for BPD (5 or more DSM-5 criteria) were included in the study. Psychotic disorder, bipolar affective disorder type 1, and pervasive developmental disorder were used as exclusion criteria. In order to test sensitivity to change of the BSL-23, 92 BPD patients were reassessed after a four-week I-DBT program [19,20]. I-DBT is an original adaptation of DBT skills training which combines, in a short and intensive format, individual sessions with the primary therapist and skills-training groups based on the traditional DBT modules: mindfulness, interpersonal effectiveness, emotion regulation, and distress tolerance. Patients are also offered telephone assistance with therapists between 9 am and 6 pm. All therapists attend weekly meetings with the consultation team.
Forty-five patients suffering from ADHD were also recruited in order to test the discriminant validity of BSL-23. The ADHD diagnosis was based on a clinical evaluation by a trained psychiatrist and on the Diagnostic Interview for ADHD in adults (DIVA 2.0) [23]. In addition, patients completed the following questionnaires: the Adult ADHD Self-Report Scale (ASRS v1.1) [24], which assesses severity of adult ADHD, and the Wender Utah Rating Scale (WURS) [25,26], featuring a subset of 25 questions on a five point Likert-scale. Following Fossati et al. [27], we used a very stringent cut-off score of 46 to indicate the existence of ADHD in childhood. BPD was clinically excluded by the same clinicians.

Assessment
At admission, each patient completed the following self-report scales:

The borderline symptom list (BSL-23)
Each subject completed the French version of the BSL-23 to assess BPD symptomatology. BSL-23 [15,16] was translated from English to French by NP and PP and an independent English-speaking translator back-translated the French version into English. In its original form, the BSL-23 is a 23-item self-rated scale presenting a onefactor structure and high internal consistency (Cronbach's alpha = .935). The original BSL-23 also boasts good reliability for BPD diagnosis and discriminates BPD patients from other psychiatric patients (mean effect size =1.13). It has also shown a sensitivity to change through therapy.

Other self-report measurements
The Beck Depression Inventory II (BDI-II) [28] assesses the current severity of depression symptoms. It includes 21 items that are rated on a four-point scale (0 to 3), with scores ranging from 0 to 63. High scores indicate greater severity.
The Beck Hopelessness Scale (BHS) [29] was used to estimate the degree of pessimism and negativity about the future. Featuring 20 true-false statements, the scores of the scale range from 0 to 20. High scores indicate a greater sense of hopelessness.
The Barrat Impulsivity Scale (BIS-10) [30] is a measure of impulsiveness that includes 34 items rated on a four-point scale (rarely/never, occasionally, often, almost always/ always. The scoring of the items reveal three factors: motor impulsivity, cognitive impulsivity, and nonplanning impulsivity. High scores indicate a greater level of impulsiveness. The State-Trait Anger Expression Inventory (STAXI) [31] is a 44-item self-report measure of the experience and expression of anger. Items are rated on a four-point frequency scale and scores range from 0 to 132. Five subscales are calculated: state anger, trait anger, anger-in, anger-out, and anger control, which assesses the intensity of the angry feelings or the frequency at which anger is experienced, expressed, or controlled.
The demographic data (Table 1) were obtained from a standard questionnaire given to all participants. The study was approved by the ethics committees of Fribourg and of the Geneva University Hospital. Patients signed an informed written consent form.

Statistical analysis
Data analysis was carried out using SPSS version 22 and STATA release 13. Descriptive statistics were used to describe the demographic and clinical characteristics of the sample. To test internal consistency, a global Cronbach's alpha was estimated and the split-half method was applied. In addition, Cronbach's alphas were estimated, with each of the 23 items removed one-by-one from the scale. Test-retest reliability was evaluated by paired-sample correlations.
To measure the appropriateness of the factor analysis, the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and the Bartlett's test of sphericity were used. An exploratory factorial analysis (EFA) of principal components with a Promax rotation was performed to examine the factorial structure of the scale. A confirmatory factor analysis (CFA) was then performed to test the adequacy of the one-factor model proposed by Bohus et al. [16]. The accuracy of the fit with the original version of BSL-23 was tested with chi-squares; as chi-squares are dependent on sample size, other indexes recommended by Hu and Bentler [32] were also used: the standardized root mean square residual (SRMR) and the root mean square error of approximation (RMSEA). Schermelleh-Engel et al. [33] consider that an RMSEA between 0 and .05 indicates a good fit, whereas an RMSEA between .05 and .08 is considered an acceptable fit, and values between .08 and .10 are considered as a mediocre fit. Values > .10 are not acceptable. The SRMR should be less than .05 for a good fit [32], whereas values smaller than .10 are still deemed acceptable. The comparative fit index (CFI) and the goodness of fit index (GFI) were also used to test how well the model fits the data. A CFI or GFI value over .90 generally indicates a reasonable fit between the model and the data but Hu and Bentler [32] recommend the use of a more severe criterion (≥.95) to describe a good fit. Correlations between BSL-23 and other psychological scales (BDI-II, BHS, BIS-10 and STAXI) were analysed, using the Bonferroni correction for multiple correlations. Correlation for ordinal data (Spearman's rho) was performed to assess the association between BSL-23 and number of BPD criteria in the SCID-II [8]. Discriminant validity with a group of patients suffering from ADHD and no co-occurring BPD was also tested. Finally, in order to assess BSL-23 sensitivity to change through therapy, scores before and after a four-week I-DBT [19,20] were compared. The change of the BSL-23 scores before and after I-DBT was evaluated by paired-sample t-tests and Cohen's d effect size.

Demographic data
The clinical characteristics of the sample of 265 BPD patients (239 women, 90.8 % and 26 men, 9.8 %) are shown in Table 1 were neither studying nor working at the time of the study. Since no gender differences were found for all measures, the analyses were computed for the entire sample (men and women).
To study the test-retest reliability of the French BSL-23, a sub-sample of 61 BPD patients were asked to complete the instrument again after one week. This revealed a high correlation (r = .841; p < .001) between the first (m = 1.89; SD = 1.00) and second time (m = 1.73; SD = .91) the scale was completed, which suggests high test-retest reliability. In the original study, Bohus et al. [16] already found high test-retest reliability (r = .82; p < .001) in a sample of 35 subjects.

Factor structure
The KMO measure of our data's sampling adequacy was very high (.936) and Bartlett's test of sphericity (3263.2) was highly significant (p < .001). Both measures indicated that the factor analysis is appropriate for our data. The factor analysis of the original BSL-23 [15,16] suggested a one-factor structure, and both the principal component analysis and the scree plot of eigenvalues supported the dominance of a single factor, accounting for 40.6 % of the total variance. In our data, the single factor explained 44.7 % of variance. Although the EFA showed four factors with eigenvalues greater than 1.0 (10.285, 1.445, 1.201 and 1.131), cumulatively accounting for 61.1 % of the variance, the scree plot (Fig. 1) indicated a one-factor solution. Retaining all factors with eigenvalues greater than 1.0 is often an overestimation of the number of factors to be retained and Floyd and Widaman [34] suggest that the scree plot is a more useful guide. Another factor analysis was therefore conducted, specifying that a single factor should be identified. When one factor was fixed, all items showed factorial loadings equal or superior to .40, which is an acceptable level for a central factor ( Table 2).
The goodness of fit test was good (chi square = 765.25, df = 230, p < .001), but results of the CFA with the onefactor and recommended fit indexes were less than satisfactory. The values of RMSEA (.114) and RMSR (.116) were above .10, which is usually considered to be unacceptable. The CFI (.82) and the GFI (.78) were inferior to .90 indicating a poor fit. Models with 2, 3 or 4 factors were examined but they didn't provide for a better fit. The inadequacy of fit of our basic CFA model could be explained by the fact that several items, namely items 5, 7, 11, 12, 21, and 23, were highly intercorrelated (≥.70) ( Table 3).

Convergent validity
The following scales were used to analyse the convergent validity of the French BSL-23: BDI-II, BHS, BIS-10 and STAXI. All correlations are reported in Table 4. Because of the large number of correlations calculated from this sample, we applied the Bonferroni correction and focused on correlations that were significant at p ≤ .005. As found in previous reports [15,16], the French BSL-23 score was highly correlated with depression severity, as measured by the BDI-II (r = .550), and with hopelessness, as measured by the BHS (r = .350). In addition, high scores on the BSL-23 were associated with stateanger (r = .482), trait-anger (r = .285), anger-in (r = .284) and anger-out (r = .194) subscales of the STAXI. Correlations between BSL-23 and motor impulsivity (sub-score of the BSI; r = .281) were also found. A positive correlation (r s = .200, n = 265, p = .001) was found between number of positive criteria at the SCID-II and severity of symptoms in BSL-23.

Discriminant validity
To determine whether the French BSL-23 discriminates BPD patients from other patient groups, the questionnaire was given to 45 patients (18 women, 40 % and 27 men, 60 %) with ADHD and no comorbid BPD. The demographic characteristics of the two groups showed no statistically significant differences. Independent sample tests (t = 8.084, p < .001) showed that BPD patients had higher BSL-23 scores compared with ADHD patients (m = .78; SD = .49). This result support the fact that the items of the BSL-23 were selected because of their ability to discriminate between BPD patients and patients with different axis I diagnoses [15,16].

Sensitivity to change
We examined changes in the French BSL-23 scores in a sample of 92 BPD patients (87 women and 5 men) who participated in a four-week I-DBT. Patients completed the BSL-23 before and after the four-week program. BSL-23 scores decreased significantly after I-DBT (before: 2.00; SD = .91; after: 1.53; SD = .86) with a medium effect size (d = .53).

Discussion
The aim of this study was to investigate the psychometric properties of a French version of the BSL-23 in a BPD sample. Our results showed that its psychometric properties were similar to those of the original [15,16] and the Spanish [18] versions. The French BSL-23 had a high internal consistency and test-retest reliability and the factor analysis showed one highly dominant factor. The French version of the BSL-23 was correlated with depression (BDI-II), hopelessness (BHS), experience and expression of anger (STAXI), and impulsivity (BIS-10); the strongest correlation was found to be with depression severity (r = .550). Bohus et al. [16] already found moderateto-high correlations between the BSL-23 and depression, as well as general severity of psychopathology and global well-being. This was supported in the study by Soler et al. [18], showing that the Spanish version of the BSL-23 positively correlated with measures of depression and anxiety symptoms, hostility, and impulsivity. The correlation with depression was also among the strongest ones in their study. BPD has a great comorbidity with depression [35] and there are similarities between depressive symptoms assessed by the BDI-II and BSL-23 items assessing suicidal ideations and dysphoria which are diagnosis criteria of BPD. The correlation between BSL-23 and hopelessness (BHS) was also expected as depressed BPD patients tend to exhibit high levels of hopelessness [36]. Associations with impulsivity (BIS-10) and anger (STAXI) measures show that the BSL-23 captures a wide range of BPD symptoms, including emotional and behavioral dysregulations. The positive correlation between severity of symptomatology measured by the BSL-23 and In our study, we also demonstrated the discriminant validity of the French BSL-23 when comparing BPD patients with ADHD patients. Although these two disorders are highly comorbid and share similar characteristics, such as impulsiveness and emotion dysregulation [37], the French BSL-23 was able to discriminate between the two conditions, a fact that demonstrates the specificity of the scale in assessing BPD symptomatology. This is consistent with the findings of Bohus et al. [16], showing the ability of the original version of the BSL-23 to distinguish between patients suffering from various psychiatric disorders, including ADHD, and patients suffering from BPD.
Besides being useful in assessing the disorder's current severity, we found that the French version of the BSL-23 was sensitive to change after a four-week I-DBT intervention. Again, this is consistent with the results of previous studies showing that the original version of the BSL-23, as well as its Spanish version, had a good sensitivity to change either after 12 weeks of specific treatment for BPD or after a three-month DBT therapy, respectively [16,18]. Further studies could investigate the relevance of this instrument to measure changes in borderline symptomatology after a longer psychotherapeutic intervention targeting BPD patients specifically, such as standard DBT [17] or Mentalization-Based Treatment [38].
Some limitations must be considered. First, the BSL-23 is a self-report measure and is obviously dependent on the introspective ability of the person. Nonetheless, the positive correlation between BSL-23 mean score and number of symptoms assessed by the SCID-II suggests that patients' own evaluation was coherent with the clinician's assessment of BPD. Another limitation should be reported with regard to the ADHD patients included in our study. They were clinically assessed to exclude a BPD diagnosis by expert psychiatrists, but they didn't undergo the BPD interview of the SCID-II to confirm the absence of BPD. Conclusions BPD is the most common personality disorder in clinical settings, but, to our knowledge, no self-report instruments in French was available to assess the severity of the disorder and its sensitivity to change following a therapeutic intervention. Our study showed that the French BSL-23 has good psychometric properties, provides a specific assessment of BPD symptomatology and is sensitive to change. This study provides a tool in French that is both easy and quick to use. It will allow clinicians and researchers to effectively measure borderline symptomatology.