Study design

The study was approved by the ethics commission of the Medical University Innsbruck, Austria, on 24 March 2004 (ID: UN1950) and was registered at Clinicaltrials.gov (NCT00714311). Participants were recruited at the out-patient units of the Departments of Psychiatry and Psychotherapy, Technical University of Munich, Germany, and the Psychoanalysis and Psychotherapy Department, Medical University Vienna, Austria. People who fulfilled the inclusion criteria were given a complete description of the study. Those who gave written informed consent were assessed by trained local research assistants. The results of the first assessments were sent to a researcher outside the two study centres who performed the randomisation. Participants were randomly assigned to either transference-focused psychotherapy or experienced community psychotherapists. After randomisation participants were referred to a therapist; medication was registered continuously by the local administrators. Medication treatment was not standardised, its type and amount were decided on an individual basis by the individuals' psychiatrists in the community in both groups. One year after treatment started, outcome was assessed by the local research assistants, who were masked to the therapy delivered. The community therapists were interviewed by the local administrator to record number of sessions and frequency of supervision that took place as well as the drop-out rate and participants' reasons for terminating. In the transference-focused psychotherapy group these data were recorded by the supervisors.

Participants

Participants were screened by the local administrators at the outpatient units. Inclusion criteria were age 18–45 years, female gender, diagnosis of borderline personality disorder (DSM–IV) ¹¹ and sufficient knowledge of the German language. Exclusion criteria were diagnosis of antisocial personality disorder, schizophrenia, bipolar I and II disorder with a major depressive, manic or hypomanic episode during the previous 6 months, substance dependency (including alcohol) during the previous 6 months, organic pathology or mental retardation. After written informed consent was obtained from participants, inclusion and exclusion criteria were examined again within the baseline assessment. Antisocial personality disorder was an exclusion criterion if three or more DSM–IV criteria were met. Comorbidity with all other personality disorders and with Axis I disorders except those mentioned above were allowed as well as medication.

Assessments

Four research assistants, who were masked for the therapy delivered, conducted assessments before randomisation and after 1 year of treatment. Two performed pre- and post-tests, two pre- or post-tests only. Prior to the study they received comprehensive interview training and demonstrated satisfactory reliability for the Structured Clinical Interview for DSM (SCID–I and –II) ^{Reference Wittchen, Zaudig and Fydrich12,Reference Fydrich, Renneberg, Schmitz and Wittchen13} and Structured Interview for Personality Organization (STIPO) ^{Reference Clarkin, Caligor, Stern and Kernberg14} (intraclass correlation coefficients 0.72 to 0.78).

Two primary outcome measures – drop-out and suicide attempts – were chosen because they cover fundamental issues in the treatment of people with borderline personality disorder: individuals' adherence to the treatment while refraining from taking their own life. Suicidality was not chosen as the only primary outcome measure since data on it might not be available for those who dropped out.

Treatment and therapists

Psychotherapy and medication were reimbursed by the German and Austrian health insurance companies at no expense to the participants. Psychotherapy was delivered in the private practices of licensed psychotherapists or at the psychotherapy out-patient units of university hospitals. The study period was 1 year; participants and therapists were instructed to maintain the treatment for at least this duration. Psychotherapies were to be continued if deemed necessary by the therapist and the participant and if paid by the insurance company. Additional non-study psychotherapy was not permitted in the transference-focused psychotherapy group.

Statistical analyses

For sample size calculation (continuity corrected) a 1-year prevalence of 30% suicide attempters among individuals with borderline personality disorder was assumed. ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen and van Asselt6,Reference Gunderson and Ridolfi26} To detect a difference between a reduction of 10% in the transference-focused psychotherapy group and of 20% in the experienced community psychotherapists group with a power of 80% (P<0.05), 41 participants per group were needed.

Intention-to-treat group comparisons were calculated for the outcome analyses. Chi-squared statistics were employed for dichotomous variables, the McNemar test for within-group changes of dichotomous variables, Mann–Whitney U-tests for ordinal data, and t-tests for continuous data. Cohen's d was computed to measure effect sizes for continuous data, and for dichotomous measures effect size (d*) was calculated according to Hasselblad & Hedges. ^{Reference Hasselblad and Hedges27} To compare the time effects of the two interventions, analyses of covariance (ANCOVA) were performed. For the group comparison of change over time of presence or absence of suicide attempts and self-harming acts during the treatment year (dichotomouos variables) additional variables were created: 1, positive change; 0, no change; –1, negative change. Group differences on these variables were tested by means of the Mann–Whitney U-test. For the primary outcome variables (drop out, suicide attempt during the treatment year) Bonferroni adjustment was performed, which set the level of significance to P<0.025. P for secondary outcome measures are explorative. The SPSS 15.0 software package for Windows was used for statistical analyses.

Treatment of missing data

As presented in Fig. 1, a number of participants were lost to the follow-up evaluation. Since this study contains two assessments only, we decided to address the problem of missing values by combining two strategies of analysis: observed cases and last observation carried forward (LOCF). The rationale for this combination is the fact that the latter tends to underestimate within-group changes and the former to overestimate these changes. ^{Reference Prakash, Risser and Mallinckrodt28} To prevent unrealistically high effect sizes because of missing values on the one hand, and to rule out a bias due to different drop-out rates in the two groups on the other hand, the complete intention-to-treat analysis was performed twice. The first analysis contained only participants who attended the 1-year follow-up, and in the case of the second analysis, missing data were replaced by the baseline assessment.

Fig. 1

Participant flow in the randomised controlled trial.

Participant characteristics at baseline

Participants were recruited between October 2004 and August 2006. The participant flow is presented in Fig. 1. In total, 104 women with borderline personality disorder gave written informed consent and were included in the study. Their characteristics are shown in online Table DS1. There were no significant differences between the groups with regard to sociodemographic and clinical variables at baseline.

Treatment received

The drop-out rate was significantly higher (Bonferroni adjusted) in the experienced community psychotherapists group (67.3% v. 38.5%, χ² = 8.683, d.f. = 1, P = 0.003). The point of time when individuals dropped out of therapy can be seen from the Kaplan–Meier survival curve in Fig. 2 (log-rank statistic 8.453, d.f. = 1, P = 0.004). In total, 72 participants were included into the observed cases analyses (43 in the transference-focused psychotherapy group and 29 in the experienced community psychotherapist group); the LOCF analyses contained the whole sample.

Fig. 2

Proportion of participants remaining in transference-focused psychotherapy (TFP) and treatment by experienced community psychotherapists (ECP).

Participants in the transference-focused psychotherapy group received 48.5 (s.d. = 34.2) sessions and those in the experienced community psychotherapists group 18.6 (s.d. = 24.0) sessions of individual psychotherapy within the 1-year study period. The difference is significant (t = 5.163, d.f. = 101, P<0.001). Among the completers, transference-focused psychotherapy participants (n = 32) received 72.9 (s.d. = 14.5) sessions, and the experienced community psychotherapists group (n = 17) 39.4 (s.d. = 26.3) sessions, which also differs significantly (t = 4.984, d.f. = 48, P<0.001).

Medication

There were no significant differences between the groups with regard to medication at baseline and during the 1-year treatment period (Fig. 3). The only participant who received amphetamines was in the transference-focused psychotherapy group. There was no significant influence of psychotropic medication on the outcome variables with the exception of a worse BSI global severity index in medicated participants (F = 43.927, d.f. = 1,101, P = 0.04).

Fig. 3

Proportion of participants taking psychotropic medication. TFP, transference-focused psychotherapy; ECP, experienced community psychotherapists.

Treatment outcomes

Selection of participants

This study involved people with less severe borderline personality disorder, with higher GAF scores, fewer comorbid Axis I and II disorders and fewer self-harming acts than other treatment studies on this patient group. This can be attributed to the healthcare situation in Germany and Austria, where more severely ill people with borderline personality disorder receive in-patient treatment paid for by healthcare insurance companies. Moreover, people with comorbid antisocial personality disorder and substance dependency were excluded. This patient selection limits the generalisability of the results.

Drop-out rates

The drop-out rate of participants in the transference-focused psychotherapy group (38.5%) was somewhat higher than that reported for out-patient dialectic behavioural therapy (25%) ^{Reference Linehan, Comtois, Murray, Brown, Gallop and Heard29} and schema-focused therapy (26.7%). ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen and van Asselt6} This can be attributed to the healthcare situation in Germany and Austria, where insurance companies cover psychotherapy for everyone in need. As a consequence, the study participants received psychotherapy at no cost to them and were free to choose another therapist if they were not satisfied with the one they were assigned to. Individuals who do not receive manualised borderline-specific psychotherapy drop out of therapy much more often. Linehan et al ^{Reference Linehan, Comtois, Murray, Brown, Gallop and Heard29} reported a 59.2% drop-out rate in their community treatment by experts group; which is close to the 67.3% in our study.

Suicidality and self-injury

Participants in the transference-focused psychotherapy group had improved suicidality but self-injury was not reduced, whereas in the comparison group both were almost unchanged; this calls for further interpretation, since other treatment approaches reduce both suicidality and self-injury. ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen and van Asselt6,Reference Linehan, Comtois, Murray, Brown, Gallop and Heard29,Reference Bateman and Fonagy30} A possible explanation for this difference may be found in the study by Verheul et al ^{Reference Verheul, van den Bosch, Koeter, de Ridder, Stijnen and van den Brink31} who reported an improvement in self-injury after 1 year of dialectic behavioral therapy only in individuals with a high severity of these behaviours, whereas individuals with a low severity remained unchanged. In our sample, the median number of self-injurious acts in the year before treatment was 6.0 (range 0–365); Linehan et al ^{Reference Linehan, Comtois, Murray, Brown, Gallop and Heard29} reported a median of 10 acts during the pre-treatment year and Bateman & Fonagy ^{Reference Bateman and Fonagy30} in their paper on partially hospitalised patients reported 8–9 self-injuries in the 6 months prior to treatment. Thus, our participants revealed a comparably low severity of self-injurious behaviour, which – according to the finding of Verheul et al ^{Reference Verheul, van den Bosch, Koeter, de Ridder, Stijnen and van den Brink31} – might have led to the lack of positive change.

Depression, anxiety and general psychopathology

Transference-focused psychotherapy yielded lower effect sizes in its improvement in self-reported depression, anxiety and general psychopathology, and as a result no significant differences between groups were found. This finding is in line with previous RCTs that also did not find significant group differences with regard to depression and anxiety in individuals treated with behavioural therapy ^{Reference Clarkin, Levy, Lenzenweger and Kernberg3,Reference Linehan, Comtois, Murray, Brown, Gallop and Heard29,Reference Verheul, van den Bosch, Koeter, de Ridder, Stijnen and van den Brink31,Reference Linehan, Armstrong, Suarez, Allmon and Heard32} and transference-focused psychotherapy. ^{Reference Clarkin, Levy, Lenzenweger and Kernberg3} In contrast, Bateman & Fonagy ^{Reference Bateman and Fonagy30} reported highly significant group differences in the improvement of depression and anxiety in participants with borderline personality disorder treated with mentalisation-based therapy, ^{Reference Bateman and Fonagy33} but their participants were very severely ill – they started with a BDI score of 36.0 (compared with 25.8 in our transference-focused psychotherapy group), received intense 18-month partial hospitalisation treatment and were compared with a group of participants who only received two psychiatric consultations per month.

Dose–effect relationship

Since completed therapies with experienced community psychotherapists included significantly fewer sessions than completed transference-focused psychotherapies, completer analyses were conducted, controlling for the effect of number of sessions. The results showed that the effect of transference-focused psychotherapy can not be solely reduced to the dose of treatment. The fact that the group differences with regard to suicide attempts and psychiatric admissions were no longer significant can easily be explained by the high number of people who dropped out after suicide attempts and/or hospitalisation, especially in the experienced community psychotherapists group.

Continuation of therapy after the study period

In our study a number of participants continued psychotherapy after the 1-year study period. This has also been the case in previous influential studies ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen and van Asselt6,Reference Bateman and Fonagy34} and has to be regarded as a potential bias to the results. As a consequence we controlled the outcome analyses for continuation of therapy and found the main effects to be stable. Since in both groups only one of the participants who continued treatment attempted suicide, the group time interaction for suicide attempts during the study period was no longer significant. As mentioned above, the rest of the participants who attempted suicide during the study period (six in the transference-focused psychotherapy group and ten in the experienced community psychotherapists group) dropped out from therapy before the end of the 1-year study period.

Strengths and limitations

The strengths of this study are its large sample size, the multisite setting, the treatment by experienced community psychotherapists in the comparison group and the conservative statistical evaluation including observed cases and LOCF analyses. Due to the earlier mentioned healthcare situation in Germany and Austria, which allows individuals to switch therapies at no cost and to abstain from the study follow-up assessments without losing their therapy, there are two major limitations of the study: the high drop-out rate and the low participation in the follow-up assessment. Only 49 participants (47.1%) in the whole sample completed the 1-year treatment phase, end-point data were available for 71 individuals (68.3%). This amount of missing data reduces the validity of the results to a certain degree. Completer analyses and control of the results for dose–effect relationship assured that despite the missing data a genuine and significant efficacy of transference-focused psychotherapy exists in the domains of borderline symptomatology, psychosocial functioning and personality organisation. Evidence for a superiority in the reduction of suicidality and need for psychiatric in-patient treatment should be regarded as preliminary.

Implications of the study

This study strengthens the empirical validation of transference-focused psychotherapy for the treatment of people with borderline personality disorder. Future research should look at long-term follow-up, since effects of psychotherapy seem to take years to develop and to continue after termination of treatment. ^{Reference Levy35} Moreover, given that we now have dialectic behavioural treatment, transference-focused psychotherapy, mentalisation-based therapy and schema-focused therapy as viable treatments for borderline personality disorder, the specific and potentially different mechanisms of change need to be studied in order to determine which treatment is most effective for which individuals with specific problems.