The mean score of the BSL-23 in the sample of 241 patients that had a current diagnosis of BPD (BPD_CAL) was 1.87, with a standard deviation (SD) of .8. As illustrated in figure 1, the distribution of the BSL-23 scores was approximately symmetrical around the mean (see histogram), and did not significantly deviate from normality (cf. Q-Q plot and histogram). Accordingly, the skewness and the (excess) kurtosis were close to zero (skew=-0.14, kurt=-0.67), and despite the relatively large sample size, the Anderson-Darling test did not indicate a significant deviation from normality (A-Sq = 0.75, p = 0.05).
Based on the distribution of the BSL-23 mean scores in the calibration sample (BPD_CAL), the dimensional scores were divided into 6 grades of severity: (i) none or low: BSL-23 mean scores from 0 (i.e., the theoretical minimum) to less than .28 (i.e., 2 SDs below the mean); (ii) mild: scores between .28 and 1.07 (i.e., 1 SD below the mean); (iii) moderate: scores between 1.07 and 1.87 (i.e., the mean); (iv) high: scores from 1.87 to 2.67 (i.e., 1 SD above the mean); (v) very high: scores from 2.67 to 3.47 (i.e., 2 SDs above the mean); (vi) extremely high: from 3.47 to 4 (i.e., the theoretical maximum).
Correlations of the BPD severity grades with theoretically convergent measurements mostly conformed to theory. In the combined sample, including all of n=1.090 participants from the four groups (BPD_CAL, BPD_VAL, HC, and CC), the correlation between the BPD severity grades and the original BSL-23 scores was rho=0.98 (p<0.001). In the combined subsamples of n=558 participants from the two groups with a diagnosis of BPD (BPD_CAL and BPD_VAL), this correlation was still as large as rho=0.97 (p<0.001). The correlation between the BPD severity grades and the number of BPD-symptoms (as assessed by the IPDE) was available in 908 participants from the BPD_CAL, BPD_VAL, and HC samples. In these 908 samples this correlation was large (rho=0.80, p<0.001); however, in the subsample that included all BPD patients (BPD_CAL and BPD_VAL, n=558), the correlation was only small to medium (rho=0.24, p<0.001). The correlation between BPD severity grades and GAF showed a similar pattern. While this correlation was large (rho=0.80, p<0.001) in the combined sample of BPD and HC participants (BPD_CAL and HC, n=588), the correlation was only small to medium (rho=0.27, p<0.001) in the BPD patients (BPD_CAL, n=235). Conversely, the correlation between BPD severity grades and the global severity index (GSI) was large in the sample that included all available BPD patients (rho=0.77, p<0.001). Formal testing in participants with ratings in all of the four measurements (BSL-23, IPDE, GSI, and GAF, n=233) revealed that the absolute value of the correlation of the BSL-23 severity grades and the GSI was significantly larger than the correlation between the BSL-23 severity grades and the number of diagnostic criteria (rho=0.77 vs -0.26, z=8.11, p<0.001), and also larger than the absolute value of the correlation between the BSL-23 severity grades and the GAF (rho=0.77 vs. -0.27, z=8.02, p<0.001).
In summary, the extremely large positive correlations of the BPD severity grades and the original BSL-23 suggest that the loss of information related to the grouping is minor. The pattern of correlations between the BPD severity grades and theoretically convergent measures also suggests that that BSL-23 severity grades assess psychopathology that is not included when using a very narrow assessment of BPD-symptoms, such as the number of BPD-symptoms (defined in the DSM-5), or when using a very broad and unspecific measure such as the GAF. The pattern of correlations further supports the convergent validity of the BSL-23 severity grades. Validity of the BPD severity grades is also supported by the cross-tabulation of the 6 classes established in the calibration sample (BPD_CAL) and the means ± standard deviations of the above mentioned external measures (i.e., the number of BPD-symptoms, the GSI, and the GAF) in the combined sample of all BPD and HC participants (BPD_CAL, BPD_VAL, and HC, n=914). As presented in table 1, the mean values of these external measures suggest a rather functional level in individuals from the two lowest BSL-23 classes (none or low, and mild). Individuals from the class labelled none or low are virtually free from BPD-symptoms as defined in the DSM-5, and were typically found to have a global severity index (GSI) of 0.11, which is below the mean of 0.31 in the mixed adult norm sample described by Franke et al. [41]. These individuals also presented a GAF of 88.35, which (by definition) corresponds to a high level of global functioning and few or no symptoms [32]. According to the variables used for validating the BSL-23 grades, individuals with a severity classification of BPD-symptoms labelled mild show some symptoms. The mean number of 2.88 ± 2.76 BPD-symptoms indicate that the majority of participants do not meet the diagnostic criteria for BPD. However, a minority of these individuals do reach the threshold for 5 diagnostic criteria, which qualify these individuals for a diagnosis of BPD. Similarly, a mean of 0.62 in the GSI corresponds to a value of 0.62, which indicates the threshold for caseness according to Franke [41] and the mean of 67.79 in the GAF falls into the category indicating “mild symptoms in one area”. As further shown in table 1, individuals from the BSL-23 classes moderate, high, very high, and extremely high all have mean number of 5 or more diagnostic criteria for BPD and mean GSI-scores clearly indicating a need for psychiatric treatment and moderate to serious symptoms according to the mean GAF-values. The BSL-23 classes ranging from moderate to extremely high showed clear differentiation with respect to both the number of BPD-criteria (for details see table 1) and with respect to GSI-scores. The observed mean of 1.17, with respect to the GSI-score in the class labelled moderate corresponds to the mean GSI-scores of 1.19-1.32 observed in mixed samples of adults receiving psychiatric treatment (Franke [41]), while the mean scores of 1.62, 1.99, and 2.76 observed in the BSL-23 classes labelled high, very high, and extremely high, respectively, also correspond to high, very high, and extremely high mean values in the GSI. In contrast, the four highest BSL-23 classes showed little differentiation with respect to the GAF (for details see table 1).
Table 1: Means and SD of GAF and GSI in every category of severity for BPD_CAL and HC
BSL-23 in the BPD calibration sample (BPD_CAL)
|
Values of external measures (BPD and HC samples) across the BSL-23 classes of severity
|
Severity classification (BSL-23)
|
Range of BSL-23 mean scores
|
Number of BPD-symptoms (IPDE)
|
Global Severity Index (GSI, SCL-90-R)
|
Global Assessment of Functioning (GAF)
|
None or low
|
0-.28
rounded: [0 , .3)
|
.19 ± .95
|
.11 ± .0.11
|
88.35 ± 10.5
|
Mild
|
.28-1.07
rounded: [.3 , 1.1)
|
2.88 ± 2.76
|
.62 ± .3
|
67.79 ± 15.57
|
Moderate
|
1.07-1.87
rounded: [1.1 , 1.7)
|
5.22 ± 2.04
|
1.17 ± .37
|
53.69 ± 9.55
|
High
|
1.87-2.67
rounded: [1.7 , 2.7)
|
5.90 ± 1.75
|
1.62 ± .46
|
50.51 ± 9.61
|
Very high
|
2.67-3.47
rounded: [2.7 , 3.5)
|
6.5 ± 1.73
|
1.99 ± .44
|
47.49 ± 8.5
|
Extremely high
|
3.47-4
rounded: [3.5 , 4]
|
7.2 ± 1.23
|
2.76 ± 0.39
|
49.8 ± 10.71
|
In order to further characterize the BPD severity levels, two additional analyses implying the 23 items of the BSL-23 have been carried out. First, to provide a more nuanced differentiation within the lower end of severity levels (i.e. those ranging from severity level 1 (none or low) to severity level 3 (moderate)), the level (range: 1 to 3) was correlated with each of the BSL-23 items. Second, to study symptomatic differentiation within the higher end of severity levels (i.e. those ranging from 4 (high) to 6 (extremely high)) these levels (range: 4 to 6) were again correlated to all of the items from the BSL-23. These post-hoc analyses were applied to the combined sample of all BPD patients (BPD_CAL and BPD_VAL) and carried out using Pearson correlations (rho) which were classified as “large effect sizes” if rho was equal or larger than 0.5 (Cohen, 1988). Within the lower range of severity classes effect-sizes were large for all items but item 15 (i.e. “I suffered from voices and noises from inside and/or outside my head”) (rho=0.27 for item 15 and 0.51-0.74 for the other 22 items). This means that (with the exception of item 15) the items of the BSL-23 clearly differentiated patients across the lower levels of the severity spectrum. On the higher end, a somewhat different pattern emerged. While all BSL-23 items were again positively related with the level of severity (now ranging from 4 (high) to 6 (extremely high)), correlations were large for only three of the items (“I didn’t believe in my right to live”: rho=0.52, “I hated myself”: rho=0.54, and “I wanted to punish myself”: rho=0.56). When also considering the three items for which the effect-size was approaching a large effect (i.e. between 0.45 and 0.49) (“I thought of hurting myself”, “I felt disgusted by myself”, and “everything seemed senseless to me”), a consistent interpretation is getting obvious: A clinically important feature differentiating those BPD patients with a high over very high to extremely high symptomatic level relates to potentially dangerous self-damaging action tendencies that are grounded in self-contempt and self-hatred.
The validity of the BSL-23 classes was further supported by a cross-tabulation of the severity classes against the samples investigated in this study. In the group of healthy controls (HC), 89% showed none or low symptoms, while 11% showed mild symptoms. In contrast, the majority (70%) of treatment seeking patients with a diagnosis of BPD (BPD_VAL) presented with a high, very high or extremely high severity of BPD-symptoms (for details see table 2). In the sample of clinical controls, the mode (44%) was in the category of mild symptoms followed by the adjacent categories (i.e., moderate: 24% and none or low: 15%). A complete list of percentiles of BSL-23 mean scores in the four subsamples investigated in this study is provided in supplementary table 1.
Table 2: Means, standard deviations, and percentiles of BSL-23 mean scores
Sample
|
N
|
BSL-23 Mean ± SD
|
Severity classification (BSL-23)
|
|
|
|
none or low
|
mild
|
moderate
|
high
|
very high
|
extremely high
|
BPD_CAL
|
241
|
1.87
± 0.8
|
2.90%
|
15.77%
|
31.95%
|
30.71%
|
17.43%
|
1.24%
|
BPD_VAL
|
317
|
2.34
± 0.86
|
0.63%
|
9.15%
|
20.19%
|
31.55%
|
29.65%
|
8.83%
|
CC
|
176
|
1.08
± 0.79
|
14.77%
|
43.75%
|
23.86%
|
13.07%
|
4.55%
|
0.00%
|
HC
|
356
|
0.12
± 0.17
|
88.76%
|
10.96%
|
0.28%
|
0.00%
|
0.00%
|
0.00%
|
The accuracy of the BSL-23 in correctly identifying the diagnostic status (BPD yes vs. no as diagnosed by clinical psychologists using the IPDE) is displayed in figure 2. As illustrated in the receiver operator curve (ROC) to the left, the BSL-23 was very accurate in discerning BPD patients from healthy individuals. Accordingly, the area under the curve (ROC AUC) for the BSL-23 mean scores and for the BSL-23 severity classes were as high as 0.997 and 0.991, respectively. The Youden index reached 0.95 and reached the highest values in the BSL-23 severity class indicating mild symptoms (at a BSL-23 mean score of 0.64) where the simultaneous pair of sensitivity and specificity was 0.97 and 0.98, respectively. For a complete list of sensitivities and specificities at all BSL-23 mean scores, see supplementary table 2. As illustrated in the ROC curve to the right of figure 2, the BSL-23 also separated BPD patients from individuals with a diagnosis of a mental disorder other than BPD. The area under the curve in this sample was 0.85 for the BSL-23 mean scores and 0.84 for the BSL-23 severity categories, indicating a medium to high level of accuracy in both cases. The Youden index reached 0.55, with the highest values in the BSL-23 severity class indicating moderate symptoms (at a BSL-23 mean score of 1.5) where the simultaneous pair of sensitivity and specificity was 0.83 and 0.72, respectively. The complete table of sensitivities and specificities in BPD vs. clinical controls for all values of the BSL-23 is displayed in the supplementary table 3.