اعتقادات در اختلال شخصیتی: یک آزمون با پرسشنامه اعتقاد اختلال شخصیت

Abstract The hypothesis that each personality disorder (PD) is characterized by a specific set of beliefs was tested in a sample of 643 subjects, including non-patient controls, axis-I and axis-II patients, diagnosed with SCID-I and -II interviews. Beliefs of six PDs (avoidant, dependent, obsessive–compulsive, paranoid, histrionic, borderline) were assessed with the Personality Disorder Belief Questionnaire (PDBQ). Factor analyses supported the existence of six hypothesized sets of beliefs. Structural equation modeling (SEM) supported the hypothesis that each PD is characterized by a specific set of beliefs. Path coefficients were however in the medium range, suggesting that PDs are not solely determined by beliefs. Nevertheless, empirically derived cutoff scores of the six belief subscales were reasonably successful in classifying subjects, percentages ranging form 51% to 83%. It appeared that there was a monotonical increase in scores on each belief subscale from non-patient controls, to patients without any PD, to patients with PDs (other than the pertinent PD), to patients with the pertinent PD. This suggests that PD-related beliefs are at least partly associated with (personality) psychopathology in general. Another explanation is that many patients’ position on the underlying dimensions is not high enough to lead to a DSM PD diagnosis, but high enough to lead to an elevated belief score.

1. Introduction In cognitive views of personality disorders, an important role is given to the patient’s beliefs or schema’s that are assumed to underlie the patient’s dysfunctional behavior and emotions (Beck, Freeman et al., 1990 and Beck, Butler, Brown, Dahlsgaard, Newman and Beck, 2001). These views hypothesize that each personality disorder (PD) is characterized by a specific set of beliefs. Various suggestions have been made about the beliefs that are central in PDs. Young (1990) has suggested that 18 themes like self-sacrifice and entitlement are fundamental dimensions in personality pathology (see Schmidt, Joiner, Young, & Telch, 1995 for a psychometric evaluation of Young’s Schema Questionnaire). So far, it is unclear how these themes exactly relate to the PDs as defined by the DSM. Beck et al. (1990) offer an extensive list of beliefs for most DSM-III-R PDs. Arntz, Dietzel, & Dreessen (1999) have suggested that specific beliefs characterize borderline PD (BPD). Based on Beck et al. (1990), Beck and Beck (1991) have developed the Personality Belief Questionnaire (PBQ). The PBQ has been investigated in several studies. Trull, Goodwin, Schopp, Hillenbrand, and Schuster (1993) tested the PBQ in a sample of 188 students, and found that reliability of subscales was good (Cronbach alpha’s 0.77–0.93). The authors concluded that intercorrelations between subscales were too high (up to 0.65, median 0.40), and correlations with other PD measures too low (median correlation with corresponding PDQ-R scale 0.37). Factor analyses did not reproduce the hypothesized subscales. This may have been caused by the non-clinical sample, leading to a limited number of factors (e.g., ‘psychopathology’) dominating the variance. Fydrich, Schmitz, Hennch, and Bodem (1996) investigated a German version in 282 psychiatric patients, and found good evidence for the subscales in terms of internal consistency (Cronbach alpha 0.78–91). But, no data on factorial validity were reported. The PBQ subscales correlated poorly (0.09, antisocial) to good (0.57, dependent) with corresponding SCID-II trait scores (median correlation 0.32), but tests of specificity were not reported. A study in a clinical sample by Beck et al. (2001) also found promising results. This study largely confirmed that each of five PDs (avoidant, dependent, obsessive–compulsive, narcissistic and paranoid) was specifically characterized by the corresponding beliefs as assessed with the PBQ. But, this study did not investigate the factor analytic structure of the PBQ, thus, in a sense, omitting a first step in investigating a self-report instrument. Independently from Beck and Beck’s PBQ, Dreessen and Arntz (1995) developed the Personality Disorder Belief Questionnaire (PDBQ). For each PD, 20 beliefs were formulated, partly on the basis of the Appendix in Beck et al. (1990; with permission), excluding items describing symptoms, impulses, emotions, and behaviors, and partly on the basis of hypotheses of the constructors. All beliefs hypothesized to be specific to BPD were constructed by the authors, since the Appendix of Beck et al.’s (1990) book does not give a list of BPD beliefs (see Arntz, 1994 and Arntz, Dietzel and Dreessen, 1999 for hypothesized themes of BPD beliefs). Arntz et al. (1999) investigated some of the properties of a short PDBQ version (assessing avoidant, dependent, obsessive–compulsive, paranoid, histrionic, and BPD beliefs) in a small sample of borderline and cluster-C PD patients, and non-patient controls. It was demonstrated that the six subscales of the PDBQ were highly reliable (internal consistency) and stable (despite a mood induction), and that the BPD subscale was a very good predictor of BPD as assessed with the SCID-II. The borderline PDBQ subscale also mediated the relationship between childhood trauma reports and SCID-II BPD diagnosis. But, a factor analytic investigation was not done (the sample was too small) and the specificity of the other PDBQ subscales was not tested. The aim of the present study was to test the hypothesis that six PDs are characterized by specific beliefs, using the PDBQ to assess beliefs1. By means of factor analytic techniques, the hypothesis was tested that factors could be derived from the PDBQ that correspond to the hypothesized subscales. Next, we attempted to improve psychometric properties by item selection. Internal consistencies of the new subscales were estimated. Then, the hypothesis was tested that each subscale was specifically related to (i.e., characteristic of) the corresponding PD, as assessed with the SCID-II. Next, it was investigated whether only the pertinent PD shows elevated scores on the belief subscale related to the PD, or that there is a monotonically increasing response in the following groups: (1) non-patient controls, (2) patients without PDs, (3) patients with a PD but not with the pertinent PD, and (4) patients with the pertinent PD. The reason for the last possibility is that patients without PD may score higher than non-patient controls, because dysfunctional personality characteristics, including dysfunctional beliefs, are higher in subjects with than without psychopathology. Similarly, patients meeting any PD may score higher on non-pertinent PDBQ scales than patients without a PD, because they have more general dysfunctional personality characteristics than the latter group. Lastly, cutoff scores were derived for each belief scale and the percentages of correctly classified subjects were calculated.

. Results 3.1. Factor analysis and item selection An initial principal component analysis on the 120 PDBQ items yielded 19 factors with eigenvalues larger than 1. There were two breaches in the scree plot, one between the first and the second component, the second in the area of the 5th to the 7th component. Eigenvalues were for component one 39.5, for component five 3.4, for component six 2.2, for component seven 1.9, and for components eight and nine 1.6. Since we were interested in specific sets of beliefs, rather than in a global belief factor related to psychopathology in general, the one factor solution (32.9% variance) was disregarded, and the 5 (47.8% variance), 6 (49.6%) and 7 (51.2%) factor solutions were further investigated. Extractions followed by direct oblimin rotation yielded solutions roughly corresponding to the hypothesized PDBQ dimensions, with the six factor solution the most promising in terms of interpretability in light of the hypothesized structure. After item selection using the criteria mentioned above, a solution clearly corresponding with the hypothesized structure was obtained on the 71 selected items. Ten items loaded on the avoidant beliefs factor (11.8% explained variance), 13 on the dependent (13.2%), 11 on the obsessive–compulsive (7.3%), 20 on the paranoid (17.5%), 11 on the histrionic (6.3%), and 6 on the borderline factor (11.9%). In total, 56.1% of the variance was explained, whereas the sum of the explained variances of the six factors was 68.0%, implying that the overlap of the factors was 68.0%−56.1%=11.9% of the total variance. The factors did not correlate highly with each other, as is evident from the raw and the disattenuated correlations between the factors (Table 1). Taken together, the results of the factor analysis supported the hypothesized structure of the PDBQ. The selection of specific PDBQ borderline items appeared the most problematic: only 6 of the original 20 remained. Table 1. Correlations between factors derived from the final factor analysis on 71 PDBQ items (above the diagonal). Internal consistencies (Cronbach alpha) are depicted in italics on the diagonal. Disattenuated correlations below the diagonal PDBQ factor AV DEP OC PAR HIS BOR AV 0.93 0.37 0.25 0.40 0.03 0.40 DEP 0.40 0.93 0.30 0.35 0.23 0.30 OC 0.28 0.34 0.85 0.23 0.18 0.21 PAR 0.42 0.37 0.25 0.96 0.21 0.48 HIS 0.03 0.26 0.21 0.24 0.83 0.16 BOR 0.44 0.33 0.24 0.53 0.19 0.87 AV, avoidant; DEP, dependent; OC, obsessive–compulsive; PAR, paranoid; HIS, histrionic; BOR, borderline. Table options 3.2. Internal consistency Internal consistencies as estimated by Cronbach alpha’s coefficients of the six scales constructed on the basis the final factor analysis (after item selection) are given in Table 1 (diagonal). They are all in the good–excellent range (0.83–0.96). 3.3. Criterion validity The hypothesis that the six PDBQ scales are specifically related to their corresponding PD was tested with SEM. Two models were tested, the first with PD diagnoses (dichotomous variables) as criteria, the second with PD trait scores (computed as the number of traits met divided by the total number of traits per PD; so that DSM-III-R and DSM-IV trait scores had the same range (0–1)) as criteria. Since comorbidity between PDs is the rule, covariance between PDs was allowed. Fig. 1 depicts the model and the path coefficients (with diagnoses in normal script, with trait scores in brackets and in italics). The model with diagnoses as criteria achieved an excellent fit, root mean square error of approximation (RMSEA)=0.044, non-normed fit index (NNFI)=0.96, goodness of fit index (GFI)=0.98. The model with trait scores achieved higher path coefficients, but slightly lower fit indices, which were still very good: RMSEA=0.062, NNFI=0.94, GFI=0.97. These results indicate that the data fitted very well with the hypothesized model. Hypothesized SEM model and observed path coefficients of belief scores as ... Fig. 1. Hypothesized SEM model and observed path coefficients of belief scores as assessed with the PDBQ specifically predicting PD diagnoses as assessed with the SCID-II. (Path coefficients relating to PD trait scores in parentheses and in italics.) Figure options Despite the excellent fits, it was further investigated whether cross-paths (a path from PDBQ subscale x to PD y) should be added. Since allowance of all cross-paths results in a saturated model, which has by definition perfect fit, and the hypothesized model was already tested against the saturated model in the tests above, another approach was chosen. From the saturated model, the cross-paths were selected that met a Bonferroni-corrected significance level of 0.05/30 (given that there were 30 cross-paths that were non-hypothesized). These were added to the model, so that fit indices could be assessed and the extended model could be compared to the hypothesized model. For the PD diagnoses as criterion variables, only one extra path appeared, from histrionic beliefs to avoidant PD, with a negative path coefficient (−0.18). Adding this to the hypothesized model resulted in a slightly smaller path coefficient (−0.15), and in fit indices that were slightly higher than those of the hypothesized model (RMSEA=0.034, NNFI=0.98, GFI=0.99). With PD traits as dependent variables, none of the cross-paths met the Bonferroni-corrected level of significance. 3.4. Scores in subgroups Table 2 presents means and standard deviations of four groups: non-patient controls, patients without any PD, patients with one or more PDs, but without the pertinent PD, and patients with the pertinent PD. ANOVA trend analysis confirmed the hypothesis that scores increase monotonically over these four groups: linear trends were all significant (p<0.001), whereas the other trends were NS, with one exception in which the quadratic trend was also significant (HIS beliefs) ( Table 2). In case of the histrionic beliefs, the positive quadratic trend was related to the pattern that patients with histrionic PD scored much higher than patients with other PDs ( Table 2). Table 2. Means, standard deviations, and results of trend analyses of the six PDBQ subscales in four subsamples PDBQ subscale Non-patient controls Patients without PD Patients with any PDa Patients with pertinent PD Linear trend m sd m sd m sd m sd t(534) p AV 10.1 8.3 27.1 19.9 41.2 20.8 55.5 23.5 8.77 <0.001 DEP 15.8 9.3 30.2 20.1 40.5 22.3 48.0 21.4 5.79 <0.001 OC 27.5 14.0 38.2 19.8 50.7 19.7 56.2 20.8 5.80 <0.001 PAR 9.2 7.2 21.9 17.9 34.2 20.1 48.9 23.1 7.50 <0.001 HIS 18.1 10.4 21.3 14.8 24.9 14.2 40.9 18.7 3.62 <0.001∗ BOR 5.2 7.2 13.7 14.3 25.2 21.5 48.0 28.1 8.41 <0.001 AV, avoidant; DEP, dependent; OC, obsessive–compulsive; PAR, paranoid; HIS, histrionic; BOR, borderline. a Except the pertinent PD. ∗ Quadratic trend also significant, t(534)=2.47, p=0.014. Table options 3.5. Percentages correctly classified subjects Using Jacobson and Truax’s (1991)c cutoff point as the point halfway between the pertinent and the non-pertinent PD samples, numbers of successfully classified subjects were calculated (in the whole sample). The PDBQ scales did a reasonable job in correctly classifying the subjects: avoidant beliefs 75% correctly classified (77% of non-avoidant and 69% of avoidant subjects); dependent beliefs 68% (69% of non-dependent and 60% of dependent subjects); obsessive–compulsive beliefs 66% (68% of non-obsessive–compulsive and 51% of obsessive–compulsive subjects); paranoid beliefs 75% (76% of non-paranoid and 57% of paranoid subjects); histrionic beliefs 76% (76% of non-histrionic and 57% of histrionic subjects); borderline beliefs 81% (83% of non-borderline and 64% of borderline subjects).