Abstract
Comorbidity of bulimia nervosa (BN) and borderline personality disorder (BPD) is common, with estimates that one-quarter of women with BN also have BPD [1]. A clinical bias that the presence of comorbid BPD in women with BN is associated with poorer eating disorder treatment outcome is partially supported by outcome studies [2–5]. Other studies, however, report fewer adverse impacts of BPD on the severity of eating disorder symptoms or treatment outcome than previously believed [6], [7]. Furthermore, several long-term follow-up studies of ≥2 years show that the presence of a PD at pretreatment assessment is not necessarily associated with a worse outcome of the eating disorder [8], [9] A naturalistic prospective study found that the presence or severity of BPD, avoidant PD and obsessive–compulsive PD did not significantly affect the natural course of BN [10].
In addition to outcome differences reported in some studies, clinical symptoms have been found to differ between women with BN with and without PDs. Those with a comorbid PD have been characterized as having poorer general psychological functioning (depression, anxiety, self-mutilation, substance abuse), interpersonal problems, and poorer social functioning [11]. Despite this, many studies show no differences in eating disorder symptoms such as binge eating, or attitudes such as drive for thinness and body dissatisfaction [12], [13].
Recent research on personality in eating disorders has examined personality traits and its impact on the aetiology, symptomatic expression, and maintenance of the disorder [14]. One of the measures used to assess this is Cloninger's Temperament and Character Inventory (TCI) [15]. Temperament is thought to reflect emotional responses that are partially genetic and stable throughout life [15]. The four temperament dimensions in the TCI are novelty seeking (NS), the propensity to explore or initiate; harm avoidance (HA), the propensity to worry and withdraw; reward dependence (RD), the propensity for social attachment/dependence; and persistence (P), the propensity to persevere in spite of unrewarding behaviour. Character refers to an individual's self-concept, goals and values that develop through experience [15]. The TCI character dimensions include self-directedness (SD), identifying oneself as an autonomous individual; cooperativeness (C), how one sees oneself in relationship to others; and self-transcendence (ST), how one sees oneself in relationship to all things (the universe). Women with BN are characterized by high NS, high HA and low SD [16–18]. Cloninger et al. link these personality features to pessimistic thoughts, rumination, immaturity, and an inability to set and achieve goals [19]. There is considerable overlap in TCI profiles between BN and BPD. Individuals with BPD also exhibit low SD, high NS, high ST and high HA [20]. The trait of HA appears to be in contrast to the deliberate self-harm that is known to occur in a high percentage of patients with BPD [21]. Korner et al. address this contradiction by suggesting that self-harming behaviours in these patients may be as an alternative to experiencing greater internal emotional harm [22]. In addition, BPD patients have low C scores, often making them socially intolerant and unhelpful [23]. Past research has shown SD and HA to be positively affected in women with ED (by treatment) over time [24].
The aims of the present study were to (i) examine the impact of BPD on BN outcome 1 year and 3 years after cognitive therapy for BN; (ii) compare pretreatment clinical characteristics of women with BN with either BPD, other PD or no PD; and (iii) examine how TCI scales differ at pretreatment and over time in women with BN with either BPD, other PD or no PD.
Method
Overview
Women with BN were recruited for a randomized clinical trial with long-term follow up. The trial evaluated the additive efficacy of exposure-based versus non-exposure-based behavioural treatments to a core of cognitive behaviour therapy.
All participants received eight sessions of cognitive therapy before being randomized to a further eight sessions of one of three forms of behavioural therapy: (i) exposure to pre-binge cues with binging being prevented (B-ERP); (ii) exposure to pre-purge cues with purging being prevented (P-ERP); or (iii) relaxation training (RELAX). Details of the study design and outcome, and 3 year follow-up data have been presented elsewhere [25], [26].
Participants
Participants were 134 women, aged 17–45 years, with a current DSM-III-R diagnosis of BN. Of the 135 participants entering the study, one was excluded from the analyses because axis II data were missing. Exclusion criteria were current anorexia nervosa, current obesity (body mass index > 30), current severe major depression, psychoactive substance use disorder, bipolar I disorder, schizophrenia, current severe medical illness or severe medical complications of BN, current use of psychoactive medications and unwillingness to undergo a supervised drug wash-out period.
Procedure
This study received ethics approval from the Southern Regional Health Authority (Canterbury) and the University of Canterbury Ethics Committee. Participants provided written informed consent.
Baseline assessment
Participants were assessed for axis I and II disorders using the Structured Clinical Interview for DSM-III-R Patient version (SCID-I) and Personality Disorders version (SCID-II) [27]. These assessments were done by non-treating psychiatrists and psychologists. The treating therapist was blind to PD diagnosis. There was some overlap in raters across the time points. All clinicians had training and experience administering the clinician interview. Eating disorder symptoms such as binging and purging frequency were assessed using the Comprehensive Bulimia Severity Index (CBSI). The CBSI is a clinician-rated instrument designed to measure the frequency and intensity of bulimic symptoms (including food restriction and body dissatisfaction) and general functioning (mood, anxiety, substance use, and social and occupational functioning) [28]. Questions reflected concepts from the Eating Disorders Examination [29]. The clinician completed the 17-item Hamilton Depression Rating Scale (HDRS) [30] and the Global Assessment of Functioning scale (GAF) [31]. The GAF is a measure of general psychosocial functioning over the past week. Participants completed self-report questionnaires including the Eating Disorder Inventory (EDI) [32] and the TCI [15]. The TCI version 8 is a 238-item scale used to assess the personality dimensions of temperament and character [15].
Follow-up assessment
Participants were reassessed at 1 and 3 years follow up. Post-treatment assessment consisted of re-evaluating diagnosis of eating disorders, CBSI, HDRS, GAF, EDI 64 (short version), and TCI.
Statistical analysis
SPSS version 12 (SPSS, Chicago, IL, USA) was used to analyse data. Participants were divided into three groups: (i) BPD, (ii) other PD, and (iii) no PD. The division into these three groups follows previous studies that indicate differences in severity of psychopathology across these groups [4], [23], [33], [34]. χ2 tests were conducted on dichotomous variables. Analysis of variance (ANOVA) among the groups was used to compare normally distributed continuous variables. t-Tests were used to compare different time points for the groups. A repeated measures ANOVA was used to assess differences among the three groups over four time points. To reduce the risk of Type I error we have opted for a more stringent statistical significance level of p < 0.01.
Results
Pretreatment differences between BPD, other PD and no PD
The mean age of the sample was 26.1 years (SD = 6.1), and 91% were New Zealand Caucasian. Sixty-two per cent had never been married. Diagnosis of an axis II disorder was present in 56% of the sample. The most common PDs were borderline (28%), avoidant (28%) and paranoid (24%). Of the 75 individuals with a PD diagnosis, 38 met criteria for BPD and 37 were diagnosed with having other PDs. BPD often co-occurred with other PDs, as seen in Table 1. In the cases where BPD existed with another PD, these individuals were categorized in the BPD group.
Subject details for bulimic patients vs type of PD
AN, anorexia nervosa; BPD, borderline personality disorder; GAF, Global Assessment of Functioning; HRDS, Hamilton Depression Rating Scale; PD, personality disorder.; †Kruskal–Wallis tests; ‡ANOVA; §χ2 test. Results for Lifetime axis II comorbidity are shown without statistical significance because the presence of a personality disorder determined the division of the group.
There was no difference among the groups in age of onset of BN, duration, or history of anorexia nervosa. Table 1 shows that those with BPD had marginally poorer general functioning and more severe clinician rated depressive symptomatology (HDRS), and greater axis I comorbidity at presentation than those with no PD. Participants with BPD had significantly lower GAF scores, and more frequent bipolar II disorder, simple phobia, and substance abuse/dependence diagnoses than those with other PD and no PD. No significant differences emerged in the prevalence of depressive disorder, social phobia, obsessive compulsive disorder, panic disorder and agoraphobia among BPD, other PDs or no PDs.
Eating attitudes and behaviours and psychological functioning at pretreatment assessment are presented in Table 2,Table 3. On the EDI both BPD and other PD groups had significantly higher levels of disturbance than the no PD group on EDI drive for thinness and body dissatisfaction subscales. Eating disorder symptoms such as binge eating and purging did not differ significantly among the three groups at presentation.
Outcome measures of bulimic patients vs type of PD
BPD, borderline personality disorder; GAF, Global Assessment of Functioning; HRDS, Hamilton Depression Rating Scale; PD, personality disorder, n = the numbers used for the repeated measures ANOVA that were available at all time points.
EDI scores for bulimic patients vs type of PD
BPD, borderline personality disorder; EDI, Eating Disorder Inventory; PD, personality disorder, n = the numbers used for the repeated measures ANOVA that were available at all time points.
Outcome at follow up for the three groups
Of the 134 women, follow-up data were available for 101 women (75%) at 1 year follow up and 112 (84%) at 3 year follow up. There was no differential dropout rate across the groups at 1 year (BPD = 11, other PD = 8, no PD = 14, χ2=1.1, p = 0.57) or 3 year follow up (BPD = 5, other PD = 8, no PD = 9, χ2=1.1, p = 0.58). Ninety-two participants were available for all three time points. All groups had improvement in eating disorder symptoms at 1 year follow up. Women with BN and BPD did not differ significantly from the other PD and no PD groups in eating disorder symptoms and attitudes at 1 year and 3 year follow up.
At 1 year follow up (Table 3) none of the differences among the groups seen at pretreatment assessment on the EDI drive for thinness and body dissatisfaction subscales were found. There were no differences among the three groups on eating disorder symptoms such as binge eating and purging at 1 year follow up. Overall no significant differences were found for the presence of any eating disorder diagnosis at 1 year follow up among the three groups (Table 4). At 1 year follow up, however, there were several specific differences found in the BPD group. They were more likely to still have a BN diagnosis but had lower prevalence of eating disorder not otherwise specified than the other two groups. Anorexia nervosa at 1 year follow up was highest in the other PD group.
Eating-related diagnoses of bulimic patients vs type of PD
AN, anorexia nervosa; BN, bulimia nervosa; BPD, borderline personality disorder; ED, eating disorder; EDNOS, eating disorder not otherwise specified; PD, personality disorder.
General and psychiatric functioning as measured on the GAF and HDRS (Table 2) showed improvements for all three groups at 1 year follow up. No significant differences among the groups were found at 1 year follow up.
Results at 3 year follow up were similar to those at 1 year follow up. At 3 year follow up eating disorder symptoms were improved in all three groups and no differences were found among the three groups on EDI drive for thinness, bulimia and body dissatisfaction subscales, eating disorders symptoms (Figure 1), or eating disorder diagnoses. General psychiatric functioning did not differ among the three groups.
Changes in (a) binge frequency and (b) purging frequency from pretreatment to 3 year follow up. (––) Borderline personality disorder; (– – –) other personality disorder (PD); (• • •) no PD.
Changes in the TCI over 3 years
Pretreatment TCI scores for participants with BPD (Table 5) showed lower C than those with other PDs or no PDs. The BPD and other PD groups had significantly higher HA and lower SD than the no PD group. NS, RD, and P scores did not differ among the groups.
Personality characteristics of bulimic patients vs type of PD
BPD, borderline personality disorder; PD, personality disorder; TCI, Temperament and Character Inventory; a = participant numbers at pretreatment for all variables were: BPD (n = 38), other PD (n = 37), No PD (n = 57); b = participant numbers at 1 year follow up for all variables were: BPD (n = 24), Other PD (n = 24), No PD (n = 44); C = participant numbers at 3 year follow up for all variables were BPD (n = 25), Other PD (n = 24), No PD (n = 46).
At 1 year follow up only the C subscale scores had changed from pretreatment assessment, with the BPD group no longer significantly different from the other PD and no PD groups. The BPD and other PD groups remained higher in HA and lower in SD than the no PD group at 1 year follow up.
At 3 year follow up the other PD group had higher TCI HA. At 3 years there were no differences between the BPD and no PD groups on any TCI subscales.
Changes in the TCI were examined within groups over 3 years. A one-way repeated measures ANOVA compared HA and SD scores at 1 year and 3 year follow ups. There was a significant effect for HA in the BPD (Wilks’ λ = 0.34, F(2,14) = 13.88, p<.001, multivariate partial η2=0.67) and no PD groups (Wilks’ λ = 0.67, F(2,34) = 8.5, p<.001, multivariate partial η2=0.33). SD also showed significant within-group effects in the no PD group across 3 years (Wilks’ λ = 0.51, F(2,34) = 16.36, p<.001, multivariate partial η2=0.49). Despite an increase of one standard deviation in SD, the BPD group had a smaller effect size than the no PD group (Wilks’ λ = 0.59, F(2,14) = 4.8, p<.03, multivariate partial η2=0.41). The other PD group had no significant within-group changes in HA or SD across 3 years.
Discussion
This follow-up study examined the specific impact of BPD in a sample of women with BN and outcome at 1 and 3 year follow ups for BPD, other PD and no PD groups. Overall, despite having a marginally poorer clinical presentation at pretreatment assessment, women with BN and comorbid BPD did not have a worse eating disorder or general functioning outcome at 3 years after treatment than those with other or no PDs.
The present study does not support the previous research that concludes that women with bulimia and comorbid BPD have ‘dismal prospects’ and poorer outcome [2], [4], [35], [36]. On the contrary, the present study contributes to a growing body of literature that shows that the presence of BPD is not associated with poorer outcome in BN at long-term follow up [5], [8], [9]. A notable difference in the present study was the rate of improvement in the BPD group. In contrast to findings from other studies [4], [37], [38], the BPD group in the present study had improvements at the same rate as the other PD and no PD groups by 1 year follow up. Overall, the BPD group had the lowest rate of any eating disorder diagnoses at follow up (35% and 24% at 1 and 3 years, respectively; Table 4). The rate of symptomatic improvement may be explained by the ‘acute’ nature of many borderline symptoms, which rapidly resolve with treatment [39]. The lack of adverse effects of BPD comorbidity has also been seen in other clinical groups. For example, BPD has been associated with a more favourable outcome in major depression than those with other PDs [23]. Despite this, these results should be interpreted with some caution. Although statistically significant, the differences seen in general functioning, depressive symptomatology and eating attitudes across the groups at pretreatment assessment were modest and may have limited clinical significance.
The low SD and high HA personality traits seen at pretreatment assessment in both PD groups are characteristic of personality profiles previously described in BN [16]. In the present study the only trait specifically associated with a BPD diagnosis was lower C. SD, reported to be a predictor of outcome in BN [25], improved the most in the no PD and BPD groups at 3 years. Interestingly, there was a trend that verged on significance for SD to be lower in other PDs (p = 0.01) at 3 year follow up. Few differences in personality profiles at 3 year outcome were found between the BPD and no PD group.
A strength of the study was the independent assessment of PD diagnosis by non-treating clinicians at pretreatment assessment. By keeping the treating clinician blind to any PD diagnosis, potential bias that may have occurred during therapy was minimized. Another strength of the study was that the BPD group had the same amount of therapy as those with other PD or no PD. This is in contrast to much of the literature, which recommends longer, more intensive therapy sessions for participants with BN and comorbid BPD [2], [4]. Despite the relatively short course of therapy, those with BPD had caught up in almost all measures to the no PD group by 1 year follow up.
There were several limitations in the present study. It has been reported that clinical samples are more likely to attract participants with comorbid disorders seeking treatment for one disorder or another, therefore results may not be generalizable to a community sample [40], [41]. Exclusion criteria for the present study were designed to be limited in order to maximize generalizability but, as in all clinical trials, this may exclude patients who are more difficult to treat due to multiple or severe comorbid disorders [42]. The PD assessment was completed at pretreatment assessment during the acute phase of the eating disorder, increasing the risk of over-diagnosis of PD because some state variables may be confused with trait variables [40]. Despite this, the timing of the PD assessment allowed examination of change in the groups over time. Although the total sample was large, splitting the sample into three groups decreased the power to detect significant differences, increasing the likelihood of Type II error.
A final consideration when interpreting these results is the greater extent of axis II comorbidity within the BPD group (mean = 3.3) compared to the other PD group (mean = 0.9). Although it is possible that those with an increased number of PDs would have had poorer eating-related outcomes, this was not seen in the present results. Further research is needed to consider the impact of complex PD and specific combinations of PD diagnoses on eating disorder outcome.
These results contribute to a growing body of literature that shows that BPD diagnosis is only moderately stable, and declines over time [43–45]. Further evidence of this can be seen in improvement in SD, which is consistent with the loss of the BPD diagnosis at follow up [23], [46]. Future research should closely examine the factors that may influence remission of BPD.
Although women with BN and BPD have a worse clinical presentation at pretreatment assessment, they do not have a poorer outcome over 3 years of follow up in comparison with those with other comorbid PDs or without a PD. The BPD group improved to the same level after treatment as the other PD and no PD groups. Few differences in personality profiles between the BPD and no PD group were evident at follow up. The present study lends further support to the growing literature suggesting that comorbid BN and BPD do not predict poor outcome.
Footnotes
Acknowledgements
This study was supported by project and programme grants from the Health Research Council of New Zealand and by the University of Otago, Christchurch. We thank Leslie Livingston, Isobel Stevens, Robyn Abbott, and Andrea Bartram for their assistance in the coordination of this study.