What type of psychological treatments would be most appropriate for treating personality disorders?

Cochrane Database Syst Rev. 2012 Aug; 2012(8): CD005652.

Abstract

Background

Psychotherapy is regarded as the first‐line treatment for people with borderline personality disorder. In recent years, several disorder‐specific interventions have been developed. This is an update of a review published in the Cochrane Database of Systematic Reviews (CDSR) in 2006 and it was superseded by a new review (with the same title) in the CDSR in 2020.

Objectives

To assess the effects of psychological interventions for borderline personality disorder (BPD).

Search methods

We searched the following databases: CENTRAL 2010(3), MEDLINE (1950 to October 2010), EMBASE (1980 to 2010, week 39), ASSIA (1987 to November 2010), BIOSIS (1985 to October 2010), CINAHL (1982 to October 2010), Dissertation Abstracts International (31 January 2011), National Criminal Justice Reference Service Abstracts (15 October 2010), PsycINFO (1872 to October Week 1 2010), Science Citation Index (1970 to 10 October 2010), Social Science Citation Index (1970 to 10 October 2010), Sociological Abstracts (1963 to October 2010), ZETOC (15 October 2010) and the metaRegister of Controlled Trials (15 October 2010). In addition, we searched Dissertation Abstracts International in January 2011 and ICTRP in August 2011.

Selection criteria

Randomised studies with samples of patients with BPD comparing a specific psychotherapeutic intervention against a control intervention without any specific mode of action or against a comparative specific psychotherapeutic intervention. Outcomes included overall BPD severity, BPD symptoms (DSM‐IV criteria), psychopathology associated with but not specific to BPD, attrition and adverse effects.

Data collection and analysis

Two review authors independently selected studies, assessed the risk of bias in the studies and extracted data.

Main results

Twenty‐eight studies involving a total of 1804 participants with BPD were included. Interventions were classified as comprehensive psychotherapies if they included individual psychotherapy as a substantial part of the treatment programme, or as non‐comprehensive if they did not.

Among comprehensive psychotherapies, dialectical behaviour therapy (DBT), mentalisation‐based treatment in a partial hospitalisation setting (MBT‐PH), outpatient MBT (MBT‐out), transference‐focused therapy (TFP), cognitive behavioural therapy (CBT), dynamic deconstructive psychotherapy (DDP), interpersonal psychotherapy (IPT) and interpersonal therapy for BPD (IPT‐BPD) were tested against a control condition. Direct comparisons of comprehensive psychotherapies included DBT versus client‐centered therapy (CCT); schema‐focused therapy (SFT) versus TFP; SFT versus SFT plus telephone availability of therapist in case of crisis (SFT+TA); cognitive therapy (CT) versus CCT, and CT versus IPT.

Non‐comprehensive psychotherapeutic interventions comprised DBT‐group skills training only (DBT‐ST), emotion regulation group therapy (ERG), schema‐focused group therapy (SFT‐G), systems training for emotional predictability and problem solving for borderline personality disorder (STEPPS), STEPPS plus individual therapy (STEPPS+IT), manual‐assisted cognitive treatment (MACT) and psychoeducation (PE). The only direct comparison of an non‐comprehensive psychotherapeutic intervention against another was MACT versus MACT plus therapeutic assessment (MACT+). Inpatient treatment was examined in one study where DBT for PTSD (DBT‐PTSD) was compared with a waiting list control. No trials were identified for cognitive analytical therapy (CAT).

Data were sparse for individual interventions, and allowed for meta‐analytic pooling only for DBT compared with treatment as usual (TAU) for four outcomes. There were moderate to large statistically significant effects indicating a beneficial effect of DBT over TAU for anger (n = 46, two RCTs; standardised mean difference (SMD) ‐0.83, 95% confidence interval (CI) ‐1.43 to ‐0.22; I2 = 0%), parasuicidality (n = 110, three RCTs; SMD ‐0.54, 95% CI ‐0.92 to ‐0.16; I2 = 0%) and mental health (n = 74, two RCTs; SMD 0.65, 95% CI 0.07 to 1.24 I2 = 30%). There was no indication of statistical superiority of DBT over TAU in terms of keeping participants in treatment (n = 252, five RCTs; risk ratio 1.25, 95% CI 0.54 to 2.92).

All remaining findings were based on single study estimates of effect. Statistically significant between‐group differences for comparisons of psychotherapies against controls were observed for BPD core pathology and associated psychopathology for the following interventions: DBT, DBT‐PTSD, MBT‐PH, MBT‐out, TFP and IPT‐BPD. IPT was only indicated as being effective in the treatment of associated depression. No statistically significant effects were found for CBT and DDP interventions on either outcome, with the effect sizes moderate for DDP and small for CBT. For comparisons between different comprehensive psychotherapies, statistically significant superiority was demonstrated for DBT over CCT (core and associated pathology) and SFT over TFP (BPD severity and treatment retention). There were also encouraging results for each of the non‐comprehensive psychotherapeutic interventions investigated in terms of both core and associated pathology.

No data were available for adverse effects of any psychotherapy.

Authors' conclusions

There are indications of beneficial effects for both comprehensive psychotherapies as well as non‐comprehensive psychotherapeutic interventions for BPD core pathology and associated general psychopathology. DBT has been studied most intensely, followed by MBT, TFP, SFT and STEPPS. However, none of the treatments has a very robust evidence base, and there are some concerns regarding the quality of individual studies. Overall, the findings support a substantial role for psychotherapy in the treatment of people with BPD but clearly indicate a need for replicatory studies.

Plain language summary

Psychological therapies for borderline personality disorder

People with borderline personality disorder often have difficulties controlling their emotions and impulses, and find it hard to keep relationships. They can experience feelings of emptiness, suffer quick changes in mood and they may harm themselves. Problems coping with abandonment and a rapidly changing view of other people can form part of their difficulties. All of these things make it hard for them to engage with any treatment they may be offered. Those who are able to engage often find it hard to stick with the treatment and leave before the end. Certain types of psychological treatment ('talking therapies') have been developed in recent years to help people with this disorder. This review summarises what is currently known about the effects of these treatments. It updates a review published in the Cochrane Database of Systematic Reviews (CDSR) in 2006 and it was superseded by a new review (with the same title) in the CDSR in 2020.

We found 28 studies that had involved a total of 1804 people with borderline personality disorder. These studies examined various psychological treatments. Some of these are called 'comprehensive' treatments because the person talks one‐to‐one with a professional for at least part of the time. Other treatments are called 'non‐comprehensive' because they do not involve this one‐to‐one work.

A number of studies have been carried out on one particular type of comprehensive treatment, called Dialectical Behaviour Therapy. For this treatment, there were sufficient studies for us to pool the results and draw conclusions. The results indicate Dialectical Behaviour Therapy is helpful for people with borderline personality disorder. Effects included a decrease in inappropriate anger, a reduction in self‐harm and an improvement in general functioning.

There were generally too few studies to allow firm conclusions to be drawn about the value of all the other kinds of psychotherapeutic interventions evaluated. However, single studies show encouraging findings for each treatment that was investigated, both 'comprehensive' and 'non‐comprehensive' types. More research is needed.

Summary of findings

Background

Description of the condition

According to current diagnostic criteria, borderline personality disorder (BPD) is characterised by a pervasive pattern of instability in affect regulation, impulse control, interpersonal relationships and self‐image. Clinical hallmarks include emotional dysregulation, impulsive aggression, repeated self‐injury and chronic suicidal tendencies (Lieb 2004). Whereas some authors have suggested that BPD is a variant of affective disorders (Akiskal 2004), others claim only partially overlapping aetiologies (Paris 2007). Despite its controversial nature, borderline personality disorder is the focus of great interest. Its importance stems from the huge suffering of the persons concerned, the functional impairment (Skodol 2002) caused and from the significant impact it has on mental health services (Zanarini 2004Zanarini 2004a).

The Diagnostic and Statistical Manual of Mental Disorders (DSM‐IV‐TR) (APA 2000a) definition of BPD comprises nine criteria that cover the above features; for a definite diagnosis five of these must be met, and a probable diagnosis requires four. The competing International Classification of Diseases in its 10th edition (ICD‐10) refers to the condition 'Emotionally Unstable Personality Disorder' (F60.3), of which there is an impulsive type (F60.30) and a borderline type (F60.31) (WHO 2003). The latter essentially overlaps with the DSM‐IV definition of BPD. A significant problem with this type of definition is that it is possible for two people to satisfy the criteria and yet have very different personalities. This heterogeneity is a major problem in assessing the impact of an intervention. Additional to the specific BPD criteria, DSM‐IV‐TR and ICD‐10 provide general diagnostic criteria for any personality disorder that must also be met.

The prevalence of BPD is estimated at about 1.5% in the general population (Torgersen 2005), but higher (up to 20%) among psychiatric inpatients and prison populations, and it is predominantly diagnosed in women (75%; APA 2000APA 2000a). There are particular problems with its diagnosis in adolescents and young adults where existential dilemmas may mistakenly be classified as BPD (DSM‐IV). BPD commonly co‐occurs with mood disorders, substance misuse, eating disorders and post‐traumatic stress disorder (PTSD), and is also associated with other personality disorders (McGlashan 2000). Recent findings also suggest a close association between BPD and adult attention‐deficit/hyperactivity disorder (ADHD). Prevalence estimates of comorbid adult ADHD in people with BPD range from 16% (Philipsen 2008) up to 38% (Ferrer 2010), and genetic analyses underline a genetic correlation of traits of both disorders (Distel 2011). Suicidal behaviour is reported to occur in up to 84% of patients with BPD (Soloff 2002), with comorbid mood disorders or substance use being the most relevant risk factors for completion (Black 2004).

Although the short‐ to medium‐term outcome of BPD is poor, similar to that of schizophrenia, there is some evidence that the long‐term follow‐up course is more favourable with remission rates of about 88% within 10 years (Zanarini 2007). However, remission here only means that diagnostic criteria are not fulfilled and not the absence of any symptoms. Indeed, whereas acute symptoms, such as self‐mutilation, help‐seeking suicide threats or attempts and impulsivity decrease with time in most cases, affective symptoms reflecting areas of chronic dysphoria, such as chronic feelings of emptiness, intense anger or profound abandonment, largely remain (Zanarini 2007). The majority of people with BPD thus continue to experience significant levels of symptoms. Risk factors for a poorer long‐term outcome are a comorbid substance use disorder; PTSD; an anxious cluster disorder (Zanarini 2005, Zanarini 2007); family history of psychiatric disorder (especially mood disorder and substance use disorder), and demographic factors, such as older age, longer treatment history, pathological childhood experiences, temperament problems and poor adult psychosocial functioning (Zanarini 2007). It is estimated that about 60% to 70% of patients with BPD make suicide attempts, although the rate of completed suicides is far less. Zanarini and colleagues found suicide rates of 4% during a 10‐year follow‐up (Zanarini 2007).

Description of the intervention

Psychotherapeutic interventions for BPD encompass a broad range of treatments. As for any other mental disorder, established psychotherapies from the major psychotherapeutic schools are used, such as psychodynamic psychotherapy, cognitive behaviour therapy (CBT) or client‐centered therapy. In addition, several specific psychotherapeutic approaches have been developed in the last decades to meet the challenges of BPD treatment. These disorder‐specific approaches are based on principles of the established psychotherapeutic schools, but they are usually precisely structured and manualised. Strategies are provided for addressing interpersonal difficulties, which are a core problem for people with BPD and lead to difficulties in forming a therapeutic alliance. Most BPD‐specific psychotherapies include treatment contracts, actively take measures to minimise premature non‐completion of treatment and provide a crisis intervention protocol (De Groot 2008).

Among the psychological interventions used with people with BPD, the most commonly used are transference‐focused therapy (TFP), dialectical behaviour therapy (DBT), mentalisation‐based treatment (MBT), schema‐focused therapy (SFT) and the systems training for emotional predictability and problem solving (STEPPS). Most of these treatments are designed as outpatient psychotherapies of six to 12 months duration with one or two weekly individual sessions. Some also include additional group therapy sessions.

In this review, we categorise therapies involving substantial one‐to‐one work as 'comprehensive' and those without this element as 'non‐comprehensive'.

How the intervention might work

According to the treatment guidelines of the American Psychiatric Association (APA), "clinical experience suggests that most patients with borderline personality disorder will need some form of extended psychotherapy in order to resolve interpersonal problems and attain and maintain lasting improvements in their personality and overall functioning" (APA 2001, p.18), whereas drug treatment is accredited an adjunctive role.

Psychotherapeutic treatments are based on their specific assumptions about the aetiology and maintenance of the disorder. Broadly speaking, psychoanalytic therapies aim to help their patients understand and reflect on their inner mental processes and make links between their past and current difficulties. Treatments based on CBT place emphasis on self‐directed learning processes; patients are encouraged to identify their core beliefs, evaluate and modify their behaviour accordingly and gain new experiences.

As mentioned previously, numerous disorder‐specific approaches have been derived from the major psychotherapeutic schools. Some of the most prominent are dialectical behavioural therapy, mentalisation‐based therapy, schema‐focused therapy and transference‐focused psychotherapy.

Dialectical behavioural therapy (DBT) (Linehan 1993) is a complex psychological intervention that was developed using some of the principles of CBT in combination with mindfulness‐based and systemic strategies. It aims to change behaviour and the ability to contain difficult feelings by focusing on improving skills, stress tolerance, emotion regulation, interpersonal behaviour and mindfulness.

Mentalisation‐based therapy (MBT) (Bateman 2004) is a complex psychoanalytically‐based psychological intervention that aims to increase the reflective or mentalising capacity of the individual, helping the person to understand and recognise the feelings they evoke in others and the feelings they experience themselves.

Schema‐focused therapy (SFT) (Young 2003) draws from both behavioural and psychoanalytic theories and helps people with BPD to identify their self‐defeating core themes arising from unmet emotional needs in childhood and presenting as maladaptive coping styles in adulthood. The goal of SFT is to aid patients in getting their needs met.

Transference‐focused psychotherapy (TFP) (Clarkin 1999) strives to achieve integrated representations of self and others, modification of primitive defence operations and resolution of identity diffusion by analysis of the transference within the therapeutic relationship. Primitive object relations (for example, split, polarised) may be transformed to advanced ones (for example, differentiated, integrated).

In summary, psychotherapeutic approaches claim slightly different mechanisms of action, according to their underlying specific aetiology models. A common element is that they aim to ameliorate BPD pathology by use of verbal communication. Psychoanalytically‐based therapies usually emphasise and use the therapeutic relationship as a model for other relationships, whereas cognitive behaviour therapy‐based therapies primarily aim at acquiring new learning experiences and general self‐management skills.

Why it is important to do this review

In addition to the suffering experienced by people with BPD and their relatives, considerable direct costs arise from the significant demands they make on health professionals. In medical settings, people with BPD often present after having self‐harmed or in suicidal crisis and are treated in emergency settings. In many cases, repeated psychiatric hospitalisations occur. Additionally, according to a US study more than 80% of patients with BPD are in individual psychotherapy for at least half of a six‐year period (Zanarini 2004), though treatment settings and provisions for patients with BPD vary across different countries. However, it remains unclear which treatments are helpful. This review aims to provide a systematic summary of the evidence from randomised controlled trials to help people with BPD and their health care workers make informed decisions about their treatment.

This is an update of a Cochrane review previously published in 2006 (Binks 2006). At that time, the authors concluded that "some of the problems frequently encountered by people with borderline personality disorder may be amenable to talking/behavioural treatments but all therapies remain experimental and the studies are too few and small to inspire full confidence in their results." A number of new studies on this topic has become available in the meantime, so that an update of the evidence seems timely.

Objectives

To evaluate the effects of psychological interventions for people with borderline personality disorder (BPD).

Methods

Criteria for considering studies for this review

Types of studies

Randomised studies involving parallel‐arm comparisons examining psychotherapeutic interventions for people with BPD were included. Data from randomised cross‐over studies up to the point of first cross‐over were eligible for inclusion (first phase only). We excluded data from subsequent phases of cross‐over trials because of the characteristically unstable course of BPD. In studies where participants served as their own controls (within‐subject comparisons), we used first period data only (Elbourne 2002). For further details, see Unit of analysis issues.

Types of participants

Adults (aged 18 years or over) with a diagnosis of BPD according to DSM criteria (see table below), however diagnosed. Since the introduction of the diagnosis in 1980, the operational criteria have only changed marginally. Studies in which at least 70% participants had a formal diagnosis of BPD were included.

Types of interventions

Experimental interventions

Experimental interventions comprised any well‐defined, theory‐driven psychotherapeutic treatment. According to the index of Medical Subject Headings (MeSH), the thesaurus of the US National Library of Medicine's controlled vocabulary, psychotherapy was defined as "treatment of mental illness or emotional disturbances primarily by verbal or nonverbal communication" (NLM 2009). We considered all types of psychotherapy, regardless of theoretical orientation or treatment setting, including, for example, psychodynamic therapy, CBT, systemic therapy or eclectic therapies designed for BPD treatment. We also included any kind of treatment setting, that is, inpatient, outpatient or partially hospitalised. Trials on relaxation techniques and patient education programs were also eligible.

After all relevant trials had been identified and the variety of types of interventions became clear, we decided to arrange the results according to classes of interventions, and defined these classes as follows:

a) comprehensive psychotherapies: includes individual psychotherapy as substantial part of the intervention; additional group therapy may or may not be included; duration of at least three months;

b) non‐comprehensive psychotherapeutic interventions: does not include individual psychotherapy as a substantial part of the intervention.

If the same interventions were tested in different settings (that is, outpatient, inpatient, partially hospitalised), we did not pool them together but treated them as different kinds of interventions.

Comparator interventions

Eligible comparator interventions were grouped as follows:

• unspecific control interventions; includes clinical management (CM), standard care (SC), treatment as usual (TAU) or waiting list (WL) (these are conditions without any specific mode of action);

• comparative specific psychotherapeutic interventions; well‐defined and theory‐driven (head‐to‐head comparison).

Concomitant drug treatment was allowed if applied to both treatment conditions.

Types of outcome measures

Outcomes were either self‐rated or interviewer‐assessed. Only adequately validated measures were included. Studies were only included if they provided data that could be used for effect size calculation for at least one of the primary or secondary outcomes defined below.

If a study provided more than one measure for the same construct (for example, depression), the measure most often used in the whole pool of included studies was used for effect size calculation in order to minimise heterogeneity of outcomes in form and content. If a study reported the data of two assessment instruments that were equally frequently used, two review authors (JS, BV) discussed the issue and chose the one that they judged was most adequate for assessment of patients with BPD. Self‐rated measures were preferred. We combined self‐ and observer‐rated measures in the same analysis if no self‐reported measure was available and it seemed appropriate to do so in terms of content validity. The possibility of heterogeneity was examined by considering I2 values and a visual inspection of forest plots (see Assessment of heterogeneity).

Primary outcomes

The following outcomes were defined as primary outcomes.

• Overall BPD severity

• Severity of single BPD criteria according to DSM, subsumed into the following symptom clusters:

  • affective dysregulative cluster symptoms

    • anger

    • affective instability

    • chronic feelings of emptiness

  • impulsive cluster symptoms

    • impulsivity

    • suicidality: severity of intent of killing oneself. Includes continuous measures of intent severity or numbers of suicidal acts and dichotomous measures of proportions of participants with suicidal acts.

    • parasuicidality: tendency or severity of intent of doing self‐inflicted harm. Includes continuous measures of severity and dichotomous measures of proportion of participants with parasuicidal episodes.

  • interpersonal cluster symptoms

    • interpersonal problems general

    • avoidance of abandonment

  • cognitive cluster symptoms

    • identity disturbance

    • dissociation/stress‐related paranoid ideation

'Summary of findings' tables are provided for the primary outcomes.

Secondary outcomes

• Depression

• Anxiety

• General psychopathology: composite measures of current general psychpathology)

• Mental health status/functioning: measures of general psychosocial functioning on a hypothetical continuum of mental health to mental illness or full function to disability.

• Leaving the study early

• Adverse effects

Search methods for identification of studies

Electronic searches

We searched the following databases.

• Cochrane Central Register of Controlled Trials (CENTRAL), 2010, Issue 3, part of The Cochrane Library

• MEDLINE, 1950 to current, searched 15 October 2010

• EMBASE, 1980 to 2010, Week 39, searched 15 October 2010

• ASSIA, 1987 to current, searched 17 November 2010

• BIOSIS, 1985 to current, searched 12 October 2010

• CINAHL, 1982 to current, searched 8 October 2010

• Dissertation Abstracts International, searched 31 January 2011

• ICTRP, searched August 2011

• metaRegister of Controlled Trials, searched 15 October 2010

• National Criminal Justice Reference Service Abstracts, searched 15 October 2010

• PsycINFO, 1872 to October Week 1 2010

• Science Citation Index, 1970 to 10 October 2010

• Social Science Citation Index, 1970 to 10 October 2010

• Sociological Abstracts, 1963 to current, searched 11 October 2010

• Zetoc (conference search), searched 15 October 2010

We included search terms for all types of personality disorder in the search strategy as this is one of a series of personality disorder reviews. Search terms and syntax were modified as necessary for each database (Appendix 1). There were no restrictions on language, date or document format.

Searching other resources

Relevant journals such as the Journal of Personality Disorders, the American Journal of Psychiatry, Archives of General Psychiatry, the British Journal of Psychiatry and the Journal of Clinical Psychiatry were surveyed on a regular basis. Additionally, we contacted the lead authors of published RCTs via e‐mail and kept track of any developments presented at relevant international meetings including the conferences of the International Society for the Study of Personality Disorders (ISSPD; ISSPD 2007; ISSPD 2009; ISSPD 2011) and the 1st International Congress of the European Society for the Study of Personality Disorders (ESSPD 2010). We also searched trial registers in order to identify any ongoing research (see Appendix 2). Cross‐references from relevant literature were also traced.

Data collection and analysis

Selection of studies

Database and clinical trial register searches yielded 38,701 records, which were imported into ProCite reference management software. After electronic and manual deduplication, there were 28,535 records. We used the global search in ProCite to identify obviously irrelevant records (for example, about "borderline hypertension" or "borderline fractures"). The remaining 2458 records were divided up amongst various review authors (NH, MF, NS, MP) to assess study eligibility using titles and abstracts. All were double‐checked by a second review author (JD). After exclusion of 1955 records, the remaining 503 references were made available in full text and assessed by two review authors independently (JD, JS). At this stage, multiple reports of the same study were linked together. If the two review authors disagreed, a third person (BV) adjudicated upon inclusion or exclusion. At the end of this process, 91 records for 28 RCTs were included in quantitative synthesis (see Figure 1)

Data extraction and management

Data were independently extracted by two review authors (JS, BV) using standardised data extraction forms and double‐entry into the Review Manager software. Any discrepancies were resolved through discussion or, if required, by an adjudicator (KL). We contacted study authors where publications reported incomplete data or where relevant subsample data were lacking.

Assessment of risk of bias in included studies

Risk of bias was assessed using The Cochrane Collaboration's tool for assessing risk of bias (Higgins 2011). Assessments included ratings of the likelihood for selection bias (random sequence generation, concealment of allocation), detection bias (blinding of outcome assessors), reporting bias (selective reporting), performance bias (treatment adherence), bias due to allegiance effects and attention bias.

Selection bias and reporting bias were assessed using the criteria for judging respective risks of bias as delineated in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

Though the importance of blinding and the possibility of bias due to lack of blinding are beyond question, it is still unclear how this issue should best be dealt with in research practice (Boutron 2008). We did not judge the likelihood of detection bias due to inadequate blinding of patients and personnel, since in psychotherapy outcome research it is quite impossible to blind therapists and patients, and both also need to be informed about theoretical frameworks of the therapy provided to gain an in‐depth understanding of the conditions and postulated mechanisms of change. However, if interviewer‐rated measures were used, we assessed the likelihood of detection bias due to inadequate blinding of outcome assessors.

In addition, we assessed the likelihood of performance bias due to inadequate treatment adherence. The risk of bias was judged low if any means had been undertaken to assure adequate treatment adherence, for example, by regular supervision or use of adherence ratings of videotaped or audiotaped therapy sessions.

Psychotherapy trials are especially prone to bias due to "allegiance bias". That means, that "despite care in design, the therapeutic allegiance of the experimenter might in some way influence the results" (Luborsky 1975, p. 1003). However, the existence of the allegiance bias hypothesis, that advantageous findings result from the investigators' allegiances rather than from their inherent superiority, has yet to be proved (Leykin 2009).

Last, the likelihood of bias due to different amounts of attention given to the treatment groups was rated (attention bias). Findings of beneficial effects by one treatment may primarily result from simply being paid attention to or being provided with any kind of intervention rather than from a specific mechanism of action. If there was a substantial difference of attention, this was regarded as possibly introducing bias, irrespective of other treatment options the participants may have used from other providers.

The 'Risk of bias' tool was applied by two review authors independently (JS, BV), and discrepancies were discussed in order to arrive at a consensus. A third person (KL) could have been called upon, but that was not necessary. All 'Risk of bias' domains mentioned above were included in a graph (see Figure 2) and summary (Figure 3). Trials were included irrespective of risk of bias, but possible impacts on effect estimates are discussed (see Quality of the evidence).

'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study

Measures of treatment effect

Continuous data

Standardised mean differences (SMDs) were calculated on the basis of post‐treatment group results. If the direction of a scale was opposite to most of the other scales, the corresponding mean values were multiplied by ‐1. Following (Cohen 1988), SMD scores around 0.20 were regarded small, scores of 0.50 as moderate and scores of 0.80 or more as large.

In single studies, for example Giesen‐Bloo 2006, effect sizes could not be calculated for some outcomes since data were not reported in a format usable for SMD calculation.

Dichotomous data

Regarding dichotomous outcomes, the risk ratio (RR) was computed on an intention‐to‐treat (ITT) basis. We made the conservative assumption that all participants who were lost to post‐treatment assessment had an unfavourable outcome, for example, they had left because the treatment had not been acceptable for them. All calculations were done using the latest release of the Review Manager software (RevMan 2011).

Unit of analysis issues

Repeated observations

Study estimates were calculated on basis of post‐treatment group results. Interim observations were not used.

Cross‐over trials

Elbourne 2002

Though cross‐over studies were eligible, no such study was available for inclusion. See Appendix 3 for information about future updates of this review.

Cluster‐randomised trials

We intended to follow the guidance on statistical methods for cluster‐randomised trials described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). We would have sought direct estimates of the effect (for example, an odds ratio with its confidence interval) from an analysis that properly accounted for the cluster design. Alternatively, we would have extracted or calculated effect estimates and their standard errors as for a parallel group trial and adjusted the standard errors to account for the clustering (Donner 1980). This would have required information on an intra‐class correlation coefficient (ICC), which describes the relative variability in outcome within and between clusters (Donner 1980). We would have extracted this information from the articles if available; otherwise, we would have contacted the authors or used external estimates obtained from similar studies. We would have searched for closest matching scenarios (with regard to both outcome measures and types of clusters) from existing databases of ICCs (Ukoumunne 1999) and if we had been unable to identify any, we would have performed sensitivity analyses using a high ICC of 0.1, a moderate ICC of 0.01 and a small ICC of 0.001. We recognise that these values are relatively arbitrary but would have preferred to use them to adjust the effect estimates and their standard errors due to the implausibility that the ICC is actually zero. Subsequently, we would have combined the estimates and their corrected standard errors from the cluster‐randomised trials with those from parallel designs using the generic inverse variance method in Review Manager 5 (RevMan 2011).

None of the included studies applied a cluster‐randomisation procedure.

Multi‐arm trials

We would have included all eligible outcome measures for all trial arms in this review. If there had been more than two arms of a trial meeting the inclusion criteria, each referring to a different treatment, we would have included them as different comparisons (Higgins 2011a). If two or more arms referred to the same type of treatment, we would have combined these groups to create a single pair‐wise comparison (Higgins 2011b).

There were no multi‐arm trials.

Dealing with missing data

In cases of incomplete reporting of outcomes stated as having been assessed, we contacted the study authors.

Effect sizes were preferably calculated on the basis of ITT data. If only available case analysis (ACA) data were reported, we calculated effect sizes on this basis. In case of dichotomous data that were not presented on the basis of ITT data, the number of participants lost in each group were added to the participants with unfavourable results, acting on the assumption that most patients with BPD do not get lost at random.

For continuous outcomes we used per protocol analysis as available from the reports (that is, results are based on the number of patients at follow‐up). If data were not reported in an immediately usable way but required processing before being analysed, a statistician (GR) was consulted. This was the case for the study of Blum 2008 where standard deviations of mean values had to be calculated from standard errors.

Assessment of heterogeneity

We assessed studies for clinical homogeneity with respect to type of therapy, therapy setting and control group. For any studies judged as clinically homogeneous and adequate for pooling, statistical heterogeneity was investigated by both visual inspection of the graphs and the I2 statistics (Higgins 2003). An I2 score of > 75% was regarded as representing considerable heterogeneity (Higgins 2011c).

We intended to carry out meta‐analyses even if there was substantial concern about heterogeneity, but to interpret the results with caution, discuss possible reasons and investigate them by conducting subgroup analyses.

Assessment of reporting biases

We planned to use funnel plots for comparisons with sufficient primary studies. No single comparison included sufficient effect estimates to allow for drawing of a conclusive funnel plot.

However, we drew a funnel plot for the one outcome (parasuicidality) for which data were available from 18 out of the 22 included controlled comparisons. Some studies reported continuous and some dichotomous measures, but the majority used continuous measures; we therefore decided to re‐express dichotomous outcomes as SMDs using the approach of Chinn 2000.

Data synthesis

If several effect estimates were available, these were pooled and their 95% confidence interval (CI) calculated. The random‐effects model was used, as some degree of clinical heterogeneity was present in most cases.

Separate comparisons were performed by type of intervention, that is, all approaches were analysed separately (for example, DBT versus control, MBT versus control), as were comparisons of the same type of psychotherapy to different kinds of controls (for example, DBT versus treatment as usual; DBT versus general management). In addition, separate comparisons were done if the same psychotherapy was delivered in different treatment settings (that is, MBT‐partially hospitalised setting versus control, MBT‐outpatient versus control).

Subgroup analysis and investigation of heterogeneity

Pre‐planned subgroup analyses according to participant characteristics (sex, presence of distinct psychiatric comorbidities such as depression, addictive behaviour or post‐traumatic stress disorder, having undergone psychotherapy previously) or treatment characteristics (duration less or more than one year) could not be performed due to lack of studies.

There were only single studies available per comparison except for DBT versus TAU (Analysis 1.2; Analysis 1.8; Analysis 1.17; Analysis 1.18). The DBT versus TAU studies studies did not vary clinically with regard to predefined subgroup criteria, so we did not perform subgroup analyses.

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 2: Inappropriate anger

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 8: Parasuicidality

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 17: Mental health status/functioning

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 18: Leaving the study early

Statistical heterogeneity in terms of an I2score exceeding 75% was only found for one analysis (Analysis 1.18). We discussed reasons (Effects of interventions, 1.1.9) and undertook a post‐hoc sensitivity analysis (Analysis 1.19).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 19: Leaving the study early: sensitivity analysis (non‐rural areas only)

Sensitivity analysis

Sensitivity analyses for the primary outcomes were planned to be performed as follows.

• Trials requiring participants to have a certain psychiatric comorbidity in addition to BPD were to be excluded.

• Only ITT‐data based outcomes were to be included.

Given the small numbers of effect estimates per comparison and outcome, we did not conduct sensitivity analyses, as this would only have led to omitting results. Instead, we strived to make trial characteristics as well as all potential shortcomings of methodological quality explicit (compare Characteristics of included studies tables and Risk of bias in included studies section of the Description of studies section) and to critically discuss all findings.

Results

Description of studies

Results of the search

All electronic databases and search periods are listed in the Methods section (see Electronic searches) and in Appendix 1 and Appendix 2. There were no language, date or document format restrictions. This review is part of a series of reviews on interventions for personality disorders and so a very comprehensive search strategy was used covering all psychotherapeutic and/or pharmacological treatment of any personality disorder.

The searches generated 38,701 records altogether, of which 10,166 were identified as duplicates. After screening of titles and abstracts, 503 citations merited closer inspection. Assessment of full texts lead to the exclusion of 362 records: 164 were not trials; 143 did not have psychotherapy as the intervention; 30 were not RCTs, and 25 were excluded on the grounds of sample characteristics. This left 141 records. 32 of those records related to 28 different ongoing studies (see Ongoing studies). Eighteen records related to eight different RCTs that either had not assessed any of the pre‐defined outcomes of interest of this review or did not provide usable data were excluded. This left 91 eligible papers altogether (see Included studies), that is, primary or secondary publications of primary studies, conference proceedings and trial register entries, covering 28 different RCTs that were included in quantitative syntheses (see Figure 1).

Included studies

For essential characteristics of the 28 included studies, please see Characteristics of included studies. The studies were published between 1991 and 2010: three during the 1990s, another three between 2000 and 2005 and a much larger number (22) between 2006 and 2010.

Setting

All but two trials were conducted in an outpatient setting. In contrast, participants of the Bateman 1999 study were partially hospitalised if allocated to the experimental group, but remained outpatients if allocated to the control group. The study of Steil 2010 tested DBT‐PTSD, an adaption of DBT for the treatment of patients with BPD with comorbid posttraumatic stress disorder. It includes several stages from a diagnostic outpatient phase to an inpatient stay and an additional outpatient booster session after the end of inpatient treatment. However, the main interventions were conducted in an inpatient setting.

Participants

Exclusion criteria

People with evidence of mental impairment, organic brain disorder, insufficient commend of the language spoken, severe disabling organic conditions, dementia or neurologic diseases, place of residence too far from the study centre or being in coercive treatment were not eligible for most studies. Comorbid personality disorders were no reason for exclusion from most trials, but Koons 2001; Giesen‐Bloo 2006; Cottraux 2009; Carter 2010; Doering 2010; and Nadort 2009 excluded those with antisocial features or full antisocial personality disorder and Bellino 2010 did not include people with any comorbid personality disorder. Of the axis‐I disorders, schizophrenia, schizoaffective and other psychotic disorders were reasons for exclusion from all studies except Bos 2010. Bipolar disorder, mostly not further specified, was also a very common reason for exclusion, except for the studies of Blum 2008; Gregory 2008; Cottraux 2009; Farrell 2009; McMain 2009; Bos 2010; Morey 2010. Gratz 2006; Zanarini 2008; Bateman 2009; McMain 2009 specified that only people with bipolar I disorder were excluded. Other common reasons for exclusion were substance‐related disorders. People with current substance abuse were not eligible for the studies of Bateman 1999; Bellino 2006; Bellino 2007; Bellino 2010; Morey 2010. Dependence was a reason for exclusion from the trials of Bateman 2009 (opiates); Linehan 1991; Linehan 1994; Koons 2001; Davidson 2006; Gratz 2006; Weinberg 2006; Zanarini 2008; Cottraux 2009; McMain 2009; Soler 2009; Doering 2010 (any substance). People with acute substance dependence were excluded from the studies of Giesen‐Bloo 2006 and Nadort 2009 if clinical detoxification was required but could enter the trial afterwards. In contrast, all participants of Gregory 2008 had a current diagnosis of alcohol abuse or dependence. Both Giesen‐Bloo 2006 and Nadort 2009 excluded possible participants with a dissociative identity disorder and those with attention‐deficit/hyperactivity disorder. Soler 2009 was the only study explicitly excluding people with a current major depressive episode. Severity‐related reasons for exclusion were rare. For the Soler 2009 study, people with a CGI‐S score of ≧ 4, that is, moderately ill or worse, were not eligible. Acute danger to self and/or others as reason for ineligibility was explicitly specified only by Gratz 2006; Weinberg 2006; Bos 2010.

Severity of illness/level of functioning

There was no standard way of reporting or describing the overall severity of illness or the level of functioning of study participants.

The Global Assessment of Functioning Scale (GAF; APA 1987) was used by five trials. On a scale from one (persistent danger of severely hurting self or others ) to 100 (superior functioning), participants of Bateman 2009 and Farrell 2009 scored between 41 and 50 and thus had "serious symptoms (for example, suicidal ideation, severe obsessional rituals, frequent shoplifting) OR any serious impairment in social, occupational, or school functioning (for example, no friends, unable to keep a job)". Participants of Zanarini 2008; McMain 2009; Doering 2010 scored between 51 and 60 and therefore had "moderate symptoms (for example, flat affect and circumstantial speech, occasional panic attacks) OR moderate difficulty in social, occupational, or school functioning (for example, few friends, conflicts with peers or co‐workers)".

Another trial used the Global Assessment Scale (GAS; Endicott 1976), the scale that the GAF originally was derived from. It also uses a 1 to 100 continuum, with 1 indicating the lowest level of functioning in a hypothetical continuum of mental illness to health. Participants of Linehan 1994 scored between 31 and 40 on average at baseline, which indicates "major impairment in several areas, such as work, family relations, judgment, thinking or mood OR some impairment in reality testing or communication OR single serious suicide attempt".

Another widely used measure was the Clinical Global Impression Scale (CGI, Guy 1976). Scores on its Severity of Illness Subscale (CGI‐S) range from 1 = "not ill at all" to 7 = "among the most extremely ill". In primary studies included here, the average CGI‐S baseline scores described samples as "mildly to moderately ill" (Bellino 2007), "moderately to markedly ill" (Bellino 2006) or "markedly to severely ill" (Blum 2008; Cottraux 2009; Bellino 2010).

An adaptation of CGI for BPD (Clinical Global Impression‐Scale for Borderline Personality Disorder Patients; CGI‐BPD; Pérez 2007) was used in one trial (Soler 2009), indicating a "moderately to markedly ill" sample of patients on average.

Another indicator of severity of illness was the reported mean number of fulfilled BPD criteria (five out of nine are required for a diagnosis). These were 6.8 for the samples of Koons 2001 and Nadort 2009, 6.9 for Giesen‐Bloo 2006, 7.4 for Van den Bosch 2005 and 7.5 for Gratz 2006.

Davidson 2006 used the SFQ (Social Functioning Questionnaire; Tyrer 2005). The score range is 0 to 24, and a score of 10 or more indicates poor social functioning. The mean SFQ score of the sample was 14.6, comparable to psychiatric emergencies.

The remaining studies reported no standardised measures of severity.

Interventions

The duration of trial interventions ranged from 1.5 to 36 months. The mean duration was 10.19 months (SD = 7.02). Seven trials with interventions of predominantly psychoeducational character had a duration of less than six months (Gratz 2006; Weinberg 2006; Blum 2008; Zanarini 2008; Soler 2009; Bos 2010; Morey 2010). In addition, the inpatient treatment in the study of Steil 2010 was also of less than six months duration, that is, three months. The studies of Koons 2001; Bellino 2006; Bellino 2007; Carter 2010 lasted six months, those of Farrell 2009 and Bellino 2010 eight months. The major part of 10 studies (Linehan 1991; Linehan 1994; Turner 2000; Van den Bosch 2005; Davidson 2006; Linehan 2006; Gregory 2008; Cottraux 2009; McMain 2009; Doering 2010) were of one year's duration. The three studies of Bateman 1999; Bateman 2009; Nadort 2009 lasted 18 months, and the trial of Giesen‐Bloo 2006 was the longest with a duration of 36 months.

In the following, the interventions tested in the included RCTs will be described briefly. However, it is not possible to provide a comprehensive account of each psychological intervention. Only very short descriptions will be given and relevant references indicated.

Comprehensive psychotherapies

As depicted above, comprehensive psychotherapies were defined as including individual psychotherapy as substantial part of the intervention. Duration must have been three months or more. Group therapy may or may not be delivered.

Dialectical behaviour therapy (DBT) and modified DBT‐related treatments (eight studies: Carter 2010; Koons 2001; Linehan 1991; Linehan 1994; Linehan 2006; McMain 2009; Turner 2000; van den Bosch 2005)

Dialectical behavioural therapy (DBT; Linehan 1993a; Linehan 1993b) is a multi‐module psychological intervention that was developed using some of the principles of CBT in combination with mindfulness‐based techniques. It aims to change behaviour by improving skills and the ability to contain difficult feelings, but also applies supportive elements, for example, in the principles of validation of emotions and acceptance. Problem‐oriented behaviour "skills" are targeted at enhanced stress tolerance, interpersonal situations, emotion regulation and mindfulness. DBT was originally designed as an outpatient treatment. It includes weekly individual sessions along with weekly "skills group training", and telephone access for patients to individual therapists in times of crisis. Therapists meet in teams for regular supervision and exchange. Usually DBT is delivered for a period of 12 months.

Dialectical behaviour therapy (DBT) according to Linehan (Linehan 1993a; Linehan 1993b) was used in eight studies (Linehan 1991; Linehan 1994; Turner 2000; Koons 2001; Van den Bosch 2005; Linehan 2006; McMain 2009; Carter 2010). A summary of major modifications and/or adaptations is given in the table below.

DBT for Borderline Personality Disorder with Severe Posttraumatic Stress Disorder after Childhood Sexual Abuse (DBT‐PTSD; one study: Steil 2010)

DBT‐PTSD aims at meeting the special demands of people with both BPD and chronic post‐traumatic stress disorder (PTSD) after sexual abuse in childhood (Steil 2010a). It draws from and employs principles and modules of standard DBT but enhances them by PTSD specific cognitive restructuring and exposure techniques. DBT‐PTSD was delivered in an inpatient setting of three months duration in the study of Steil 2010.

Mentalisation‐based treatment (MBT; two studies: Bateman 1999; Bateman 2009)

Mentalisation‐based therapy (MBT; Bateman 2004; Bateman 2006) is a complex psychoanalytically and attachment theory‐based treatment. It aims to increase the reflective or mentalising capacity of the participant, helping them to understand and recognise the feelings they evoke in others and the feelings they experience themselves as a result of others. Thus, "mentalising" means the implicit and explicit interpretation of the actions of oneself and others as meaningful in view of intentional mental states such as desires, feelings, and needs.

MBT was used in two trials, one in a partial hospitalised setting (MBT‐PH; Bateman 1999) and the other outpatient (MBT‐OUT; Bateman 2009).

Cognitive‐behavioural approaches (CBT; three studies: Bellino 2007; Cottraux 2009; Davidson 2006)

Cognitive‐behavioural therapy (CBT) as a generic term refers to treatments that combine cognitive and behavioural techniques. Today, neither of them is practiced without the other, so the combined term has prevailed. Originially, cognitive techniques were compiled by AT Beck who developed his approach initially for the treatment of depression (Beck 1979). Current CBT is problem‐focused, with therapeutical processes strongly oriented towards pre‐defined and both patient‐ and therapist‐agreed‐upon treatment targets. Overall, the major goal is to help patients to develop adaptive (instead of maladaptive and dysfunctional) beliefs about self and others by self‐directed learning processes. Participants are encouraged to identify their core beliefs, evaluate and modify their behaviour accordingly and gain new experiences (Beck 1990; Beck 1995). Cognitive‐behavioural approaches (CBT) were used in three RCTs.

In the trial of Bellino 2007, the participants of which all had an additional major depressive episode, CBT was delivered according to the basic works of Beck (Beck 1990; Beck 1995). Cottraux 2009 delivered CBT based on further developments of A T Beck's working group for the treatment of people with personality disorders (esp.: Layden 1993; Young 1994). Davidson 2006 used a CBT approach based on AT Beck work that she has specially formulated for the practical treatment of those with Cluster B personality disorders (Davidson 2000).

Interpersonal Psychotherapy (IPT) and related treatments (IPT‐BPD; three studies: Bellino 2006; Bellino 2007; Bellino 2010)

Interpersonal Psychotherapy (IPT) is a brief and highly structured manual‐based psychotherapy that addresses interpersonal issues. It was originally developed for the treatment of depression, to the exclusion of all other foci of clinical attention (Klerman 1984). It acts on the assumption that interpersonal factors play an important role in the development and maintenance of psychological problems, and emphasises interpersonal processes rather than intrapsychic ones. Participants are encouraged to acquire interpersonal skills and thus adapt their behaviour to current roles and situations.

IPT was used in the studies of Bellino 2006 and Bellino 2007. In a more recent trial (Bellino 2010), a BPD‐specific adaptation of IPT, called IPT‐BPD, was used (Markowitz 2005; Markowitz 2006).

Transference‐Focused Psychotherapy (TFP; two studies: Doering 2010; Giesen‐Bloo 2006)

Another popular psychodynamically‐based approach is the one of transference‐focused psychotherapy (TFP; Clarkin 1999). By analysis of the transference within the therapeutic relationship, integrated representations of self and others, modification of primitive defensive operations and resolution of identity diffusion are to be achieved. Primitive object relations (for example, split, polarised into 'good' and 'bad') are to be transformed to advanced ones (for example, differentiated, integrated).

TFP was used in two trials. Doering 2010 compared TFP with a control condition, and it served as comparison treatment for SFT in the study of Giesen‐Bloo 2006.

Schema‐focused therapy (SFT; two studies: Giesen‐Bloo 2006; Nadort 2009)

Schema‐focused therapy (SFT; Young 2003) draws from both behavioural as well as psychoanalytic theories and helps patients to identify their self‐defeating core themes that evolved from unmet emotional needs in childhood and implicate maladaptive coping styles in adulthood. The goal of SFT is to aid patients to get their needs met.

SFT was tested in two trials, those of Giesen‐Bloo 2006 and Nadort 2009. The first one strongly stuck to the original SFT treatment regimen. Nadort 2009 investigated the effects of standard SFT in a regular mental health care setting and SFT plus additional crisis telephone support by individual therapists outside the office hours (SFT+).

Client‐centered therapy (CCT) and related treatments (two studies: Cottraux 2009; Turner 2000)

Client‐centered (or patient‐centered) therapy (CCT; Rogers 1951) is a non‐directive approach that aims at encouraging participants to find their own solutions to their problems. This is achieved by creating a safe, non‐judgmental therapy setting in which therapists meet their clients with un‐conditional respect, demonstrating warmth, empathy, and genuineness while "active listening". Participants are encouraged to explore their own experiences in‐depth and express their feelings, so they can decide for themselves in what ways they need to change.

CCT based on C Rogers' work was one of the conditions of the RCT of Cottraux 2009. Turner 2000 used R Carkhuff's approach, a further development of Rogers' CCT (Carkhuff 1969; Carkhuff 1976), emphasising therapeutic core conditions as crucial for the client's growth and development.

Dynamic deconstructive psychotherapy (DDP; one study: Gregory 2008)

Dynamic deconstructive psychotherapy (DDP) has been developed to treat those people with BPD having co‐occurring disorders, such as substance use disorders or additional personality disorders (Gregory 2008a; Gregory 2010). BPD is regarded a disorder of aberrant processing of emotional experiences. Therefore, DDP aims to enhance neurocognitive self‐capacities by elaborating affect‐laden interpersonal experiences and integrate attributions by narrative construction through association techniques. Thus, a differentiated view of self versus others is to be developed, supported by novel experiences within the patient‐therapist relationship.

DDP was used in a RCT conducted by its developer (Gregory 2008).

Non‐comprehensive psychotherapeutic interventions

As depicted above, non‐comprehensive psychotherapies were defined as not including individual psychotherapy as substantial part of the intervention. They mainly focus on psychoeducation as delivered in a group‐therapy settings, are of limited duration (that is, beyond six months) and focus on impartation of knowledge rather than individual therapy. These interventions may or may not be administered as an adjunctive module to ongoing individual treatments, but individual psychotherapy is not part of these interventions themselves.

DBT‐skills training only (DBT‐ST; one study: Soler 2009)

One trial investigated the effects of DBT‐skills training only (Soler 2009): three months of standard‐DBT based skills training sessions were delivered to people who did not receive any of the remaining usual DBT components.

Emotion regulation group training (ERG; one study: Gratz 2006)

ERG is an eclectic treatment approach that draws from Acceptance and Commitment Therapy (ACT; Hayes 1999), DBT and includes aspects of emotion‐focused psychotherapy (Greenberg 2002) as well as behavioural therapy. ERG was delivered for a duration of 3.5 months in the trial of Gratz 2006.

Schema‐focused therapy‐group (SFT‐G; one study: Farrell 2009)

Farrell 2009 used a shortened group‐only format of SFT (SFT‐G), based on schema change work according to J. Young (Young 1994; Young 2003) in combination with BPD psychoeducation, emotional awareness training and distress management training (Farrell 1994). This intervention was delivered for eight months.

Manual‐assisted cognitive treatment (MACT; two studies: Morey 2010; Weinberg 2006)

Weinberg 2006 and Morey 2010 tested MACT, a bibliotherapeutic approach for acutely self‐harming patients. The intervention comprises six individual sessions that are each structured around a chapter of a self‐help book (Schmidt 2004). Weinberg 2006 compared MACT with a control group, whereas Morey 2010 tested the effects of both standard MACT as well as MACT enhanced by a therapeutic assessment (MACT+TA). The therapeutic assessment comprised an individualised collaborative assessment including development of questions the client would like to be answered by the test results and development of individualised treatment goals.

Systems training for emotional predictability and problem solving for borderline personality disorder (STEPPS; two studies: Blum 2008; Bos 2010)

Two studies (Blum 2008; Bos 2010) applied the "Systems training for emotional predictability and problem solving for borderline personality disorder" (STEPPS) training. STEPPS is a 20‐week seminar‐like group treatment program. It combines cognitive‐behavioural elements and skills training with a systematic approach by involving participants' relatives and other treatment providers. Blum 2008 employed the generic STEPPS training as an add‐on to any ongoing individual treatments, whereas Bos 2010 combined STEPPS with a complementary individual therapy (STEPPS‐IT) aiming to help consolidate the skills that had been acquired during STEPPS group into the individual lives of patients. However, the authors explicitly refer to the individual sessions as "limited individual [...] developed as an adjunct to STEPPS to help consolidate the newly acquired skills and to stimulate their use" (Bos 2010, p. 300). Thus, the STEPPS group remains the core element, and we decided to classify STEPPS+IT as non‐comprehensive psychotherapeutic intervention.

Psychoeducation only (PE; one study: Zanarini 2008)

Zanarini 2008 used a very generic form of psychoeducation. People newly diagnosed with BPD participated in a single, half‐day workshop. Topics of the workshop curriculum were phenomenology, aetiology, treatment and course of BPD.

Comparisons

Comparisons included both active as well as classic control conditions.

Controls

• Treatment as usual (TAU): TAU was the most commonly used control condition. TAU means that participants are allowed to use any kind of treatment they would or would not have used in case they had not been involved in the actual trial. However, TAU conditions differ slightly, especially in the extent to which the use of any alternative treatment was obligatory or not, the extent to which alternative treatments reflect the supposed usual treatment and the extent to which alternative treatments were homogenous among all participants assigned to that condition.

  • obligatory/all control participants in some kind of treatment:

    • Bateman 1999: homogeneous; standard treatment in the general psychiatric services, excluding formal psychotherapy

    • Bateman 2009: homogeneous; structured clinical management according to best generic practice for BPD offered by non‐specialist practitioners within UK psychiatric services

    • Bos 2010: homogeneous; standard treatment for BPD as offered at the participating sites

    • Davidson 2006: homogeneous; standard treatment a patient would have received if the trial had not been in place as offered at the study sites

    • Farrell 2009: homogeneous; participants had to be in any individual psychotherapy to be eligible for study participation, and they had to continue this treatment for the duration of the study

    • Gratz 2006: homogeneous; participants had to be in any individual therapy and continue during the study

    • Koons 2001: homogeneous; participants received 60 minutes of weekly individual therapy with a clinician of the participating clinics of different orientations (none DBT) and were offered attendance of several supportive and psychoeducational group

    • Van den Bosch 2005: homogeneous; clinical management from the original referral source, that is, addiction treatment centres or psychiatric services; generally no more than two sessions per month

    • Weinberg 2006: heterogeneous; all participants took part in additional, ongoing treatments

  • optional/treatment usage:

    • Blum 2008: heterogeneous; participants "encouraged to continue their usual care" of any kind

    • Carter 2010: heterogeneous; treatment as usual plus waiting list (six months)

    • Gregory 2008: heterogeneous; participants (all with either alcohol abuse or dependence) were referred to an alcohol rehabilitation centre and given names of psychiatric clinics and therapists in the community but were also allowed to keep their current psychotherapist, if any

    • Linehan 1991: heterogeneous; participants were given alternative therapy referrals

    • Linehan 1994: heterogeneous; participants received alternative therapy referrals and were allowed to participate in any type of treatment available in the community

    • Steil 2010: heterogeneous; treatment as usual plus waiting list (six months)

• Community treatment by experts (CTBE): all participants allocated to the control condition received treatment by experienced psychotherapists in the community. Thus, treatment was obligatory, and it was standardised in that all psychotherapists were known experts through long‐term experience in that field.

  • Doering 2010: treatment by community psychotherapists who were known as experienced and particularly interested in people with BPD

  • Linehan 2006: community treatment by experts who had been nominated by community mental health leaders; the treatment provided was uncontrolled by the study but therapists' characteristics (sex, experience, mean number of clients etc.) were balanced among both groups

• Non‐specific comparison programmes: treatments mainly designed to control for attention biases or the like by providing an alternative programme. Specific techniques or putative efficacious interventions are explicitly avoided. Those treatments are subsumed in an extra category since they will usually not be found in usual health care settings.

  • Soler 2009: "Standard group therapy" (SGT); "oriented to provide a relational experience allowing people with BPD to share their characteristic difficulties"; therapists present to conduct group interaction

• Clinical management (CM):

  • Bellino 2006: appointments with a psychiatrist, fluoxetine medication (both groups received fluoxetine)

  • Bellino 2010: appointments with the local Service for Personality Disorders, fluoxetine medication (both groups received fluoxetine)

• Waiting list without treatment (WL):

  • Zanarini 2008: waiting list, weekly screening appointments but no psychiatric treatment for the duration of the waiting period (12 weeks)

Active comparators

The following studies compared two active treatments (description of interventions see above).

• Bellino 2007: cognitive therapy versus interpersonal therapy

• Cottraux 2009: cognitive therapy versus client‐centered therapy

• Giesen‐Bloo 2006: schema‐focused therapy versus transference‐focused therapy

• McMain 2009: dialectical behaviour therapy versus APA guidelines‐based treatment algorithm (general psychiatric management derived from the American Psychiatric Association (APA) guideline recommendations; combination of psychodynamically informed therapy and symptom‐targeted medication management)

• Morey 2010: manual assisted cognitive therapy versus manual‐assisted cognitive therapy plus therapeutic assessment

• Nadort 2009: schema‐focused therapy versus schema‐focused therapy plus therapist telephone crisis support outside office hours

• Turner 2000: dialectical behaviour therapy versus client‐centered therapy

Outcomes

In the case of availability of several measures relating to the same outcome construct (for example, data from several questionnaires for the assessment of depression), the one most often used in the whole pool of included studies was used for effect size calculation, in order to minimise heterogeneity of outcomes in form and content. If a study reported data of two assessment instruments that were equally frequently used, two review authors (JS, BV) discussed the issue and chose the one which was in its content most appropriate for assessing BPD‐relevant pathology. Self‐rated measures were preferred. If there was more than one post‐treatment assessment available, we used the one which was assessed closest to the end of psychotherapeutic treatment. For the studies of Steil 2010 and Zanarini 2008, several assessments were considered for use as post‐treatment data: Steil 2010 reported data that were assessed immediately at end of the three months inpatient treatment and, additionally, data that were assessed after a subsequent booster session six weeks after dismissal. As the booster session was a pre‐defined part of treatment, we used the post‐booster session data. Zanarini 2008 conducted a singular psychoeducation workshop and assessed data on a weekly basis up to week 12. We decided to use the data from week 12 to ensure observation periods throughout the whole pool of studies were as homogeneous as possible.

All available outcomes that were included in this review are listed below as well as specific measures and the respective studies that used them.

Primary outcomes  

1. BPD severity

   2. Borderline personality disorder checklist‐40 (BDP‐40) ‐ total score: Bos 2010

   4. Borderline symptom list (BSL): Steil 2010

   6. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ global: Soler 2009

   8. Personality assessment inventory ‐ borderline features scale total (PAI‐BOR‐total): Morey 2010

   9. Zanarini rating scale for borderline personality disorder (ZAN‐BPD) ‐ total score: McMain 2009

2. Affective dysregulative cluster symptoms

   1. anger

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ anger: Bellino 2010

      2. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ anger: Soler 2009

      4. Spielberger anger expression scale (STAXI) ‐ trait anger: Linehan 1994

      5. Target behaviour rating (TBR) ‐ anger: Turner 2000

   2. affective instability

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ affective instability: Bellino 2010

      2. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ affective instability: Soler 2009

      3. Diagnostic Interview for BPD‐Revised (DIB‐R) ‐ affect subscale: Farrell 2009

      4. Difficulties in emotion regulation scale (DERS) ‐ emotion dysregulation: Gratz 2006

      5. Personality assessment inventory ‐ affective instability (PAI‐BOR‐A): Morey 2010

      6. Zanarini rating scale for borderline personality disorder (ZAN‐BPD) ‐ affective instability: Blum 2008

   3. chronic feelings of emptiness

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ emptiness: Bellino 2010

      2. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ emptiness: Soler 2009

3. Impulsive cluster symptoms

   1. impulsivity

      1. Barrett impulsiveness scale (BIS): Blum 2008

      3. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ impulsivity: Soler 2009

      4. Diagnostic Interview for BPD‐Revised (DIB‐R) ‐ impulsive subscale: Farrell 2009

      5. Difficulties in emotion regulation scale (DERS) ‐ impulse dyscontrol: Gratz 2006

      6. Eysenck impulsivity venturesomeness empathy questionnaire (IVE) ‐ impulsivity: Cottraux 2009

      7. Number of participants scoring above BPDSI‐IV‐impulsivity cut‐off score: Bos 2010

      8. Target behaviour rating (TBR) ‐ impulsiveness: Turner 2000

      9. Zanarini rating scale for borderline personality disorder (ZAN‐BPD) ‐ impulsivity: Zanarini 2008

   2. suicidality

      3. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ suicidality: Soler 2009

      4. Number of participants with suicide attempt during previous six‐month period: Bateman 1999; Bateman 2009

      6. Personality assessment inventory ‐ suicidal ideation (PAI‐SI): Morey 2010

• 1. parasuicidality

     1. Borderline personality disorder severity index (BPDSI‐IV) ‐ parasuicidal behaviour score: Bellino 2010

     2. Deliberate self‐harm inventory (DSHI) ‐ frequency score: Gratz 2006

     3. LPC‐self‐mutilative acts during previous three‐month period: Van den Bosch 2005

     4. Mean number of suicidal and self‐injurious episodes: McMain 2009

     5. Mean number of self‐harming acts during previous three‐month period: Koons 2001

     6. Number of patients scoring above BPDSI‐IV‐parasuicide cut‐off: Bos 2010

     7. Number of patients with parasuicide during previous three‐month period: Gregory 2008

     10. Personality assessment inventory ‐ borderline features scale self‐harm (PAI‐BOR‐S): Morey 2010

     11. Parasuicide history interview (PHI) ‐ deliberate self‐harm frequency: Weinberg 2006

     12. Self‐harming behaviours checklist (SHBCL): Cottraux 2009

     13. Target behaviour rating (TBR) ‐ frequency of parasuicide: Turner 2000

1. Interpersonal cluster symptoms

   1. interpersonal problems general

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ interpersonal relationships: Bellino 2010

      2. Clinical global impression scale for borderline personality disorder patients (CGI‐BPD) ‐ unstable relations: Soler 2009

      3. Diagnostic interview for BPD‐revised (DIB‐R) ‐ interpersonal subscale: Farrell 2009

      5. Inventory of interpersonal problems (IIP‐C): McMain 2009

      6. Inventory of interpersonal problems‐short form (IIP‐SC): Davidson 2006

      7. Personality assessment inventory ‐ borderline features scale negative relationships (PAI‐BOR‐N): Morey 2010

      8. World Health organization quality of life assessment‐Bref (WHOQOL‐Bref) ‐ social relationships score: Bos 2010; Carter 2010

      9. Zanarini rating scale for borderline personality disorder (ZAN‐BPD) ‐ disturbed relationships score: Blum 2008; Zanarini 2008

   2. avoidance of abandonment

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ abandonment: Bellino 2010

2. Cognitive cluster symptoms

   1. identity disturbance

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ identity disturbance: Bellino 2010

      2. Personality assessment inventory ‐ borderline features scale ‐ identity disturbance: Morey 2010

   2. dissociation/stress‐related paranoid ideation

      1. Borderline personality disorder severity index (BPDSI‐IV) ‐ paranoid ideation: Bellino 2010

      3. Diagnostic interview for BPD‐revised (DIB‐R) ‐ cognitive subscale: Farrell 2009

      5. Zanarini rating scale for borderline personality disorder ‐ cognitive subscale: Blum 2008

Secondary outcomes

1. Depression

   3. Depression anxiety stress scales (DASS) ‐ depression: Gratz 2006

2. Anxiety

   2. Depression anxiety stress scales (DASS) ‐ anxiety: Gratz 2006

   4. Spielberger state‐trait anxiety inventory (STAI) ‐ state: Steil 2010

3. General psychopathology

4. Mental health status/functioning

   1. Brief disability questionnaire (BDQ) ‐ days out of role: Carter 2010

   3. Clinical global impressions scale (CGI) ‐ improvement, patient‐rated (CGI‐I‐SR): Soler 2009

   6. Social functioning questionnaire (SFQ): Davidson 2006

5. Leaving the study early

   1. Number of participants lost after randomisation for any reason; available or calculable for all primary studies

6. Adverse effects: no data available from any primary study

Excluded studies

The main primary studies that readers might be expected to be included but in fact were not are listed in the Characteristics of excluded studies table with individual reasons for exclusion. Please note that only one reason may be listed, though actually several inclusion criteria may not have been met.

Risk of bias in included studies

Figure 2 and Figure 3 show the review authors' judgements about each risk of bias item for each single study and across all studies.

Allocation

Blinding

It is almost impossible to keep psychotherapy trials participants blind to treatment allocation. Thus, the risk of bias due to non‐blindness of participants is present in psychotherapy trials throughout.

One trial report stated clearly that there was no blinding. Since only self‐rated outcomes were used, we judged there was no risk of bias due to lack of blinding (Gratz 2006). The majority of trials report that outcome assessors were intended to be kept blind to treatment allocation (Linehan 1991; Linehan 1994; Turner 2000; Koons 2001; Van den Bosch 2005; Bellino 2006; Davidson 2006; Giesen‐Bloo 2006; Gratz 2006; Linehan 2006; Weinberg 2006; Bellino 2007; Blum 2008; Gregory 2008; Bateman 2009; Cottraux 2009; McMain 2009; Soler 2009; Bellino 2010; Carter 2010; Doering 2010; Morey 2010; Steil 2010). Some of them ( Van den Bosch 2005; Giesen‐Bloo 2006; Gratz 2006; Blum 2008; Gregory 2008; Cottraux 2009; McMain 2009; Soler 2009; Carter 2010) also discussed the issue of partial non‐maintenance of blindness throughout the whole course of the study, since patients may have unintendedly indicated which group they were in, although they had been told not to do so. These studies were, nevertheless, rated as having a 'low' risk of bias due to partial break of blindness, since we felt it was not appropriate to downgrade these studies that discuss issues that may also have concerned those studies that did not discuss them. Nadort 2009 clearly reported that outcome assessors could not be kept blind to treatment conditions. However, since most outcomes were self‐assessed by the trial participants, the risk of bias was judged 'unclear'.

Two trials did not mention blinding of outcome assessors (Bateman 1999; Zanarini 2008), and the risk of bias was judged 'unclear'. One trial (Farrell 2009) providing mainly interviewer‐assessed outcomes reported that the interviewers were not blind to treatment allocation, and the risk of bias was judged likely.

Other potential sources of bias

Attention bias

Only 10 studies (Koons 2001; Giesen‐Bloo 2006; Linehan 2006; Bellino 2007; Bateman 2009; Cottraux 2009; Nadort 2009; Soler 2009; Doering 2010; Morey 2010) were rated as providing similar amounts of attention as obligatory components of the study protocol to all trial groups.The participants allocated to the control group of Gregory 2008 did not get an obligatory control treatment, but were referred to alternative treatments in the community and had not less but markedly more professional contact hours during most of the study period. The risk of attention bias was therefore also judged low for this trial. All remaining trials provided more attention (that is, in terms of frequency of appointments, involvement in additional group treatments etc.) to one group, usually the experimental group (EG).

Effects of interventions

See: Table 1; Table 2; Table 3; Table 4; Table 5; Table 6; Table 7; Table 8; Table 9; Table 10; Table 11; Table 12; Table 13; Table 14; Table 15; Table 16; Table 17; Table 18

Summary of findings 1

Dialectical Behaviour Therapy (DBT) versus treatment as usual (TAU) for people with borderline personality disorder

Summary of findings 2

Dialectical Behaviour Therapy (DBT) versus general management (GM) according to APA guidelines for people with borderline personality disorder

Summary of findings 3

Dialectical Behaviour Therapy (DBT) versus community treatment by experts (CTBE) for people with borderline personality disorder

Summary of findings 4

Dialectical Behaviour Therapy for BPD with post‐traumatic stress disorder (DBT‐PTSD) versus waiting list (WL) for people with borderline personality disorder

Summary of findings 5

Mentalisation‐Based Treatment‐partial hospitalisation (MBT‐PH) versus treatment as usual (TAU) for people with borderline personality disorder

Summary of findings 6

Mentalisation‐Based Treatment‐outpatient (MBT‐out) versus treatment as usual (TAU) for people with borderline personality disorder

Summary of findings 7

Transference‐Focused Psychotherapy (TFP) versus community treatment by experts (CTBE) for people with borderline personality disorder

Summary of findings 8

Cognitive‐Behavioural Therapy (CBT) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 9

Dynamic‐Deconstructive Psychotherapy (DDP) versus treatment as ususal (TAU) for people with borderline personality disorder

Summary of findings 10

Interpersonal Psychotherapy (IPT) versus clinical management (CM) for people with borderline personality disorder

Summary of findings 11

Interpersonal Psychotherapy adapted for BPD (IPT‐BPD) versus clinical management (CM) for people with borderline personality disorder

Summary of findings 12

Dialectical Behaviour Therapy‐skills training only (DBT‐ST)versus standard group (SG) for borderline personality disorder

Summary of findings 13

Emotion regulation group training (ERG) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 14

Schema‐Focused Therapy‐Group (SFT‐G) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 15

Systems training for emotional predictability and problem solving for borderline personality disorder (STEPPS) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 16

Systems training for emotional predictability and problem solving for borderline personality disorder + individual therapy (STEPPS+IT) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 17

Manual‐assisted cognitive treatment (MACT) versus treatment as usual (TAU) for borderline personality disorder

Summary of findings 18

Psychoeducation (PE) versus waiting list (WL) for borderline personality disorder

No adverse effects data were available for any included study.

1 Comprehensive psychotherapies versus control

1.1 Dialectical behaviour therapy (DBT) versus treatment as usual (TAU)

Five studies were included in this comparison: Carter 2010 (female outpatients; six months treatment; N = 73), Koons 2001 (female outpatients; six months treatment; N = 28); Linehan 1991 (female outpatients; 12 months treatment; N = 61); Linehan 1994 (female outpatients; 12 months treatment; N = 26); Van den Bosch 2005 (female outpatients; high proportion of substance abusers, that is, 53% of those who actually started treatment; 12 months treatment; N = 64).

1.1.1 BPD total severity

Koons 2001 provided data on BPD total severity indicating no significant difference but a tendency in favour of DBT treatment (standardised mean difference (SMD) ‐0.29; N = 20, one RCT, 95% confidence interval (CI) ‐1.17 to 0.59; Analysis 1.1).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 1: BPD total severity

1.1.2 Affective dysregulative cluster symptoms

Anger symptoms were reported in two studies (Linehan 1994; Koons 2001). The pooled effect estimates indicate a large, significant effect of DBT in terms of anger reduction (SMD ‐0.83; N = 46, two RCTs, 95% CI ‐1.43 to ‐0.22; Analysis 1.2). No specific data were available from relevant studies regarding affective instability or chronic feelings of emptiness.

1.1.3 Impulsive cluster symptoms

BPD‐specific impulsivity in general was reported upon by Van den Bosch 2005. The data indicated a small, non‐significant difference (SMD ‐0.17; N = 48, one RCT, 95% CI‐0.74 to 0.39). One study (Koons 2001) indicated a very large significant effect in favour of DBT concerning the reduction of suicidality (SMD ‐1.26; N = 20, one RCT, 95% CI ‐2.24 to ‐0.29; Analysis 1.6). Three studies reported on parasuicidality in terms of the mean number of parasuicidal acts (Linehan 1991; Koons 2001; Van den Bosch 2005). Their pooled effect estimates indicated a moderate significant effect of parasuicidality reduction by DBT (SMD ‐0.54; N = 110; three RCTs, 95% CI ‐0.92 to ‐0.16; Analysis 1.8). Carter 2010, however, found no significant difference in the proportion of participants with self‐harm between both groups (risk ratio (RR) 1.11; N = 51, one RCT; 95% CI 0.78 to 1.57; Analysis 1.9).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 6: Suicidality

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 9: Parasuicidality

1.1.4 Interpersonal cluster symptoms

Carter 2010 report data indicating no statistically significant difference between treatment and control conditions regarding interpersonal problems, resulting in a negligible effect size (SMD ‐0.04, N = 48, one RCT, 95% CI ‐0.54 to 0.61; Analysis 1.10). No data were available from any relevant studies on avoidance of abandonment.

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 10: Interpersonal problems

1.1.5 Cognitive cluster symptoms

Koons 2001 report data indicating no statistically significant difference but favouring DBT with regard to dissociative symptoms, resulting in a large effect size (SMD ‐0.90; N = 20, one RCT, 95% CI ‐1.83 to 0.03; Analysis 1.13). No data are available for any other cognitive cluster symptoms.

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 13: Dissociation/psychoticism

1.1.6 Depression

Koons 2001 report data indicating a large, statistically significant difference between treatment and control conditions in depression (SMD ‐1.12; N = 20, one RCT, 95% CI ‐2.08 to ‐0.16; Analysis 1.14).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 14: Depression

1.1.7 Anxiety

Koons 2001 report data indicating a large, statistically significant difference in favour of DBT (SMD ‐1.22; N = 20, one RCT, 95% CI ‐2.20 to ‐0.25; Analysis 1.15).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 15: Anxiety

1.1.8 Mental health status/functioning

Carter 2010 and Linehan 1994 report on the level of functioning. The pooled effect estimates yield a moderate, significant effect in favour of DBT (SMD 0.65; N = 74, two RCTs, 95% CI 0.07 to 1.24; Analysis 1.17).

1.1.9 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 1.25; N = 252; five RCTs; 95% CI 0.54 to 2.92; Analysis 1.18). However, the corresponding I2 score of 77% suggests substantial heterogeneity. A possible reason for this may be a higher likelihood of drop‐out in rural areas. A post‐hoc sensitivity analysis restricting the results to non‐rural areas, that is, leaving out the results of Carter 2010, resulted in a more homogenous effect (RR 0.80; N = 179; four RCTs; 95% CI 0.47 to 1.36; I2 = 35%; Analysis 1.19) supporting this hypothesis.

No data were available for the outcome of general psychopathology. See Table 19 for an overview of primary effect outcome effect estimates.

1

Spreadsheet of primary outcome effect estimates: Dialectical behaviour therapy (DBT) versus control groups (comprehensive therapies only)

1.2 Dialectical behaviour therapy (DBT) versus general management (GM) according to APA guidelines

One study was included in this comparison: McMain 2009 (outpatients, 92% female; 12 months treatment; N = 180).

1.2.1 BPD total severity

McMain 2009 reported data on this outcome indicating a marginal, non‐statistically significant difference (SMD ‐0.04; N = 180, 95% CI ‐0.33 to 0.25; Analysis 1.1).

1.2.2 Affective dysregulative cluster symptoms

Data on anger symptoms indicated a very small, statistically non‐significant difference between the two treatments (SMD ‐0.03; N = 180, 95% CI ‐0.32 to 0.26; Analysis 1.2).

1.2.3 Impulsive cluster symptoms

The findings indicate no significant difference in terms of the mean number of suicidal and/or self‐injurious episodes, but favour DBT (SMD ‐0.23; N = 180, 95% CI ‐0.52 to 0.06; Analysis 1.8) with a small difference. No data were available for effects on impulsivity in general.

1.2.4 Interpersonal cluster symptoms

McMain 2009 report data indicating a marginal, statistically non‐significant difference between treatment and control conditions in this regard (SMD ‐0.03, N = 180, 95% CI‐0.32 to 0.26; Analysis 1.10).

1.2.5 Depression

McMain 2009 report data indicating a small, statistically, non‐significant difference between both groups (SMD ‐0.17; N = 180, 95% CI ‐0.46 to 0.12; Analysis 1.14).

1.2.6 General psychopathology

McMain 2009 report data indicating a marginal, non‐significant difference between both groups (SMD ‐0.01; N = 180, 95% CI ‐0.30 to 0.28; Analysis 1.16).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 16: General psychopathology

1.2.7 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 1.03; N = 180; 95% CI 0.71 to 1.49; Analysis 1.18).

No data were available for the following outcomes: cognitive cluster‐related symptoms, anxiety and mental health status/functioning. See Table 19 for an overview of primary effect outcome effect estimates.

1.3 Dialectical behaviour therapy (DBT) versus community treatment by experts (CTBE)

One study was included in this comparison: Linehan 2006 (female outpatients; 12 months treatment; N = 101).

1.3.1 Impulsive cluster symptoms

Linehan 2006 report data indicating a small, statistically non‐significant difference between treatment and control conditions in terms of suicidality (SMD ‐0.12; N = 89, 95% CI ‐0.54 to 0.30; Analysis 1.6).

1.3.2 Depression

Linehan 2006 report data indicating a small to moderate, statistically non‐significant difference between groups but favouring DBT (SMD ‐0.39; N = 89, 95% CI ‐0.81 to 0.04; Analysis 1.14).

1.3.4 Leaving the study early

Reported data indicate a statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised, favouring DBT (RR 0.43; N = 101; 95% CI 0.28 to 0.67; Analysis 1.18).

No data were available for the following outcomes: BPD total severity; affective dysregulative, interpersonal and cognitive cluster symptoms; anxiety, general psychopathology, mental health status/functioning. See Table 19 for an overview of primary effect outcome effect estimates.

1.4 DBT adapted for BPD + post‐traumatic stress disease (DBT‐PTSD) versus waiting list (WL)

One trial was included in this comparison: Steil 2010 (inpatients; 100% females with comorbid PTSD from childhood sexual abuse; three months inpatient treatment plus additional booster session 6 months after discharge; N = 32).

1.4.1 BPD total severity

Steil 2010 report data indicating a moderate to large, statistically significant difference between experimental and control conditions in terms of overall BPD severity after treatment,favouring DBT‐PTSD (SMD ‐0.74; N = 31; 95% CI ‐1.47 to ‐0.01; Analysis 1.1).

1.4.2 Cognitive cluster symptoms

Reported data indicate a small to moderate, statistically non‐significant difference between experimental and control conditions in terms of dissociation (SMD ‐0.34; N = 30; 95% CI ‐1.06 to 0.38; Analysis 1.13).

1.4.3 Depression

Steil 2010 provided data indicating a large, statistically significant difference between both conditions, with less pathology for the DBT‐PTSD‐treated group (SMD ‐1.06; N = 30; 95% CI ‐1.84 to ‐0.29; Analysis 1.14).

1.4.4 Anxiety

Data also indicate a large, statistically significant difference between experimental and control conditions for the outcome of anxiety, favouring the experimental intervention (SMD ‐0.96; N = 30; 95% CI ‐1.72 to ‐0.20; Analysis 1.15).

1.4.5 General psychopathology

No statistically significant difference is indicated by the reported data for the outcome of general psychopathology (SMD ‐0.70; N = 30; 95% CI ‐1.45 to 0.04; Analysis 1.16). However, the data indicate a favourable moderate to large effect.

1.4.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised exclusive of those that were excluded due to meeting exclusion criteria, which became clear after randomisation but before start of treatment (RR 1.94; N = 32; 95% CI 0.44 to 8.57; Analysis 1.18).

No data were available for the following outcomes: affective‐dysregulative, impulsive and interpersonal cluster symptoms; mental health status/functioning. See Table 19 for an overview of primary effect outcome effect estimates.

1.5 Mentalisation based treatment ‐ partial hospitalisation (MBT‐PH) versus treatment as usual (TAU)

One study was included in this comparison: Bateman 1999 (partially hospitalised versus outpatient participants, 58% females; 18 months treatment; N = 44).

1.5.1 Impulsive cluster symptoms

No data available on general impulsivity. Bateman 1999 report data indicating large, significant effects on both less suicidality (RR 0.08; N = 38, 95% CI 0.01 to 0.58; Analysis 1.7) and parasuicidality (RR 0.44; N = 38, 95% CI 0.24 to 0.81; Analysis 1.9) favouring the experimental group.

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 7: Suicidality

1.5.2 Interpersonal cluster symptoms

Bateman 1999 report data indicating a very large, significant effect of less interpersonal pathology in partially‐hospitalised MBT patients (SMD ‐2.22; N = 38, 95% CI ‐3.04 to ‐1.39; Analysis 1.10). No further data on avoidance of abandonment available.

1.5.3 Depression

Bateman 1999 report data indicating a very large, statistically significant difference in favour of MBT‐PH (SMD ‐1.98; N = 38, 95% CI ‐2.78 to ‐1.19; Analysis 1.14).

1.5.4 Anxiety

Bateman 1999 report data indicating a moderate, statistically non‐significant difference in terms of anxious pathology (SMD ‐0.49; N = 38, 95% CI ‐1.14 to 0.16; Analysis 1.15).

1.5.5 General psychopathology

Bateman 1999 report data indicating a small to moderate, statistically non‐significant difference (SMD ‐0.39; N = 38, 95% CI ‐1.03 to 0.26; Analysis 1.16).

1.5.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 1.00; N = 44; 95% CI 0.23 to 4.42; Analysis 1.18).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative and cognitive cluster symptoms; mental health status/functioning. SeeTable 20 for an overview of primary effect outcome effect estimates.

2

Spreadsheet of primary outcome effect estimates: Mentalisation‐based treatment (MBT) versus control groups

1.6 Mentalisation based treatment ‐ outpatient (MBT‐out) versus treatment as usual (TAU)

One study was included in this comparison: Bateman 2009 (outpatients, 80% females; 18 months treatment; N = 134).

1.6.1 Impulsive cluster symptoms

No data available on general impulsivity. Bateman 2009 report data indicating large, significant effects of smaller proportions of participants with both suicidality (RR 0.11; N = 134, 95% CI 0.03 to 0.46; Analysis 1.7) and parasuicidality (RR 0.56; N = 134, 95% CI 0.34 to 0.92; Analysis 1.9) in the experimental group.

1.6.2 Interpersonal cluster symptoms

Bateman 2009 report data indicating a large, statistically significant difference in favour or MBT‐outpatient treatment in terms of interpersonal pathology (SMD ‐0.95; N = 134, 95% CI ‐1.30 to ‐0.59; Analysis 1.10). No further data on avoidance of abandonment available.

1.6.3 Depression

Bateman 2009 report data indicating a moderate, statistically significant difference in favour of MBT‐out (SMD ‐0.45; N = 134, 95% CI ‐0.79 to ‐0.10; Analysis 1.14).

1.6.4 General psychopathology

Bateman 2009 report data indicating a moderate to large statistically significant difference between the two treatment conditions in favour of MBT‐outpatient treatment (SMD ‐0.67, N = 134, 95% CI ‐1.02 to ‐0.33; Analysis 1.16).

1.6.5 Mental health status/functioning

Bateman 2009 report data indicating a statistically significant difference between the two treatment conditions in favour of MBT‐outpatient treatment (SMD 0.55, N = 134, 95% CI 0.20 to 0.89; Analysis 1.17).

1.6.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 1.05; N = 134; 95% CI 0.59 to 1.87; Analysis 1.18).

No data were available for the following outcomes: BPD severity; affective‐dysregulative and cognitive cluster symptoms; anxiety. SeeTable 20 for an overview of primary effect outcome effect estimates.

1.7 Transference‐focused therapy (TFP) versus community treatment by experts (CTBE)

One study was included in this comparison: Doering 2010 (female outpatients; 12 months treatment; N = 104).

1.7.1 BPD total severity

Data reported by Doering 2010 indicate a moderate, statistically significant effect in favour of TFP concerning overall BPD severity reduction (SMD ‐0.55; N = 104, 95% CI ‐0.95 to ‐0.16; Analysis 1.1).

1.7.2 Impulsive cluster symptoms

Doering 2010 report data indicating no statistically significant difference in the proportion of participants with any suicidal act (RR 0.64; N = 104, 95% CI 0.27 to 1.51; Analysis 1.7) or any self‐harming behaviour (RR 1.09; N = 104, 95% CI 0.84 to 1.40; Analysis 1.9). No data available on impulsivity in general.

1.7.3 Depression

Doering 2010 report data indicating a very small, statistically non‐significant difference between the groups in terms of depression (SMD 0.12; N = 104, 95% CI ‐0.26 to 0.51; Analysis 1.14).

1.7.4 Anxiety

Doering 2010 report data indicating a marginal, statistically non‐significant difference between the groups regarding anxiety (SMD 0.04; N = 104, 95% CI ‐0.35 to 0.42; Analysis 1.15).

1.7.5 General psychopathology

Doering 2010 report data indicating a marginal, statistically non‐significant difference between the groups in terms of general psychopathology(SMD 0.08; N = 104, 95% CI ‐0.31 to 0.46; Analysis 1.16).

1.7.6 Mental health status/functioning

Doering 2010 report data indicating a small, statistically non‐significant difference between the groups in terms of overall functioning (SMD 0.34; N = 104, 95% CI ‐0.05 to 0.73; Analysis 1.17).

1.7.7 Leaving the study early

Reported data indicate a statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised, favouring TFP (RR 0.57; N = 104; 95% CI 0.39 to 0.85; Analysis 1.18).

No data were available for the following outcomes: Affective‐dysregulative, interpersonal and cognitive cluster symptoms. See Table 21 for an overview of primary effect outcome effect estimates.

3

Spreadsheet of primary outcome effect estimates: Transference‐focused psychotherapy (TFP) versus control groups

1.8 Cognitive‐behavioural therapy (CBT) versus TAU

One study was included in this comparison: Davidson 2006 (outpatients, 84% females; 12 months treatment; N = 106).

1.8.1 Impulsive cluster symptoms

Reported data indicate a smaller proportion of participants with any suicidal act in the CBT group (RR 0.78; N = 101; 95% CI 0.47 to 1.27; Analysis 1.7, but a higher proportion of participants with any self‐mutilation (RR 1.17; N = 99; 95% CI 0.86 to 1.60; Analysis 1.9). Neither of these effects was statistically significant.

1.8.2 Interpersonal cluster symptoms

Data indicate a small, but statistically non‐significant difference in terms of interpersonal pathology between the two conditions (SMD 0.23; N = 99; 95% CI ‐0.16 to 0.63; Analysis 1.10).

1.8.3 Depression

We found only a very small, statistically non‐significant difference between both groups in terms of depression (SMD ‐0.11; N = 99; 95% CI ‐0.50 to 0.29; Analysis 1.14).

1.8.4 Anxiety

There was no indication of a statistically significant difference between both conditions in terms of anxious pathology, the effect size was only marginal in size (SMD ‐0.03; N = 99; 95% CI ‐0.42 to 0.37; Analysis 1.15).

1.8.5 General psychopathology

General psychopathological burden was not found to differ statistically significant between both groups, and was also marginal in size (SMD ‐0.03; N = 99, 95% CI ‐0.43 to 0.36; Analysis 1.16).

1.8.6 Mental health status/functioning

Data indicated no statistically significant difference between both groups in terms of the level of functioning at the end of treatment (SMD 0.00; N = 99, 95% CI ‐0.39 to 0.39; Analysis 1.17). The numerical effect size indicated no difference between the two treatment groups.

1.8.7 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.48; N = 106; 95% CI 0.09 to 2.52; Analysis 1.18).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative and cognitive cluster symptoms. See Table 22 for an overview of primary effect outcome effect estimates.

4

Spreadsheet of primary outcome effect estimates: Cognitive behavioural therapy (CBT) versus control groups

1.9 Deconstructive dynamic psychotherapy (DDP) versus treatment as usual (TAU)

One study was included in this comparison: Gregory 2008 (outpatients, 80% females, all with active alcohol abuse or dependence; 12 months treatment; N = 30)

1.9.1 BPD total severity

Reported data indicate a moderate, statistically non‐significant difference between both conditions (SMD ‐0.44; N = 30; 95% CI ‐1.16 to 0.29; Analysis 1.1).

1.9.2 Affective dysregulative cluster symptoms

No data available.

1.9.3 Impulsive cluster symptoms

Gregory 2008 report data that indicate no statistically significant difference between both conditions in terms of proportion of participants with self‐mutilating behaviour (RR 0.89; N = 30; 95% CI 0.47 to 1.67, Analysis 1.9).

1.9.4 Cognitive cluster symptoms

As for dissociation, reported data indicate no statistically significant difference between the two conditions (SMD 0.25; N = 30; 95% CI ‐0.47 to 0.97; Analysis 1.13).

1.9.5 Depression

Reported data indicate a moderate but statistically non‐significant difference between the conditions in terms of depression (SMD ‐0.52; N = 30; 95% CI ‐1.24 to 0.21; Analysis 1.14).

1.9.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.83; N = 30; 95% CI 0.32 to 2.15; Analysis 1.18).

No data were available for the following outcomes: interpersonal cluster symptoms; anxiety, general psychopathology, mental health status/functioning. See Table 23 for an overview of primary effect outcome effect estimates.

5

Spreadsheet of primary outcome effect estimates: Dynamic Deconstructive psychotherapy (DDP) versus control groups

1.10 Interpersonal psychotherapy (IPT) + fluoxetine versus clinical management (CM) + fluoxetine

One trial was included in this comparison: Bellino 2006 (outpatients, all with current major depressive episode, 60% females; six months treatment; additional antidepressive medication was given (fluoxetine 20‐40mg/d) to participants of both conditions; N = 39).

1.10.1 Depression

Reported data indicate a large, statistically significant difference between both conditions favouring combined treatment (that is, IPT plus medication; SMD ‐0.90; N = 32; 95% CI ‐1.63 to ‐0.16; Analysis 1.14).

1.10.2 Anxiety

There is no indication of a statistically significant difference between both groups (SMD 0.20; N = 32; 95% CI ‐0.49 to 0.90; Analysis 1.15). Data indicated a small effect.

1.10.3 Mental health status/functioning

Again, there is no indication of a statistically significant difference, with a very small effect size (SMD 0.12; N = 32; 95% CI ‐0.57 to 0.81; Analysis 1.17).

1.10.4 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.79; N = 39; 95% CI 0.20 to 3.07; Analysis 1.18).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative, impulsive, interpersonal and cognitive cluster symptoms; general psychopathology. See Table 24 for an overview of primary effect outcome effect estimates.

6

Spreadsheet of primary outcome effect estimates: interpersonal therapy (IPT) versus. control groups

1.11 IPT adapted to BPD (IPT‐BPD) + fluoxetine versus clinical management + fluoxetine

One trial was included in this comparison: Bellino 2010 (outpatients, 67% females; eight months treatment, additional antidepressive medication was given (fluoxetine 20‐40 mg/d) to participants of both conditions; N = 55).

1.11.1 BPD total severity

Bellino 2010 report data that indicate no clinical or statistically significant difference between both conditions (SMD ‐0.03; N = 44; 95% CI ‐0.62 to 0.56; Analysis 1.1).

1.11.2 Affective dysregulative cluster symptoms

Reported data indicate no clinical or statistically significant differences between both conditions in terms of inappropriate anger (SMD 0.01; N = 44; 95% CI ‐0.58 to 0.60; Analysis 1.2) and chronic feelings of emptiness (SMD 0.09; N = 44; 95% CI ‐0.50 to 0.68; Analysis 1.4). A large, statistically significant difference is indicated for affective instability, favouring IPT‐BPD (SMD ‐0.92; N = 44; 95% CI ‐1.54 to ‐0.30; Analysis 1.3).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 3: Affective instability

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 4: chronic feelings of emptiness

1.11.3 Impulsive cluster symptoms

Bellino 2010 report data indicating only a marginal, statistically non‐significant difference between both groups in terms of parasuicidality (SMD 0.02; N = 44; 95% CI ‐0.58 to 0.61; Analysis 1.8) but in terms of general impulsivity, favouring IPT‐BPD with a large effect (SMD ‐0.91; N = 44; 95% CI‐1.53 to ‐0.28; Analysis 1.5).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 5: Impulsivity

1.11.4 Interpersonal cluster symptoms

There is indication for a statistically significant difference between both groups in terms of interpersonal problems, favouring IPT‐BPD with a large effect (SMD ‐0.82; N = 44; 95% CI ‐1.44 to ‐0.20; Analysis 1.10). However, both groups did not different in terms of avoidance of abandonment (SMD 0.01; N = 44; 95% CI ‐0.58 to 0.60; Analysis 1.11).

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 11: Avoidance of abandonment

1.11.5 Cognitive cluster symptoms

Reported data indicate neither a statistically significant difference between both groups for the outcome of identity disturbance (SMD ‐0.03; N = 44; 95% CI ‐0.62 to 0.56; Analysis 1.12) nor for the outcome of paranoid ideation (SMD 0.10; N = 44; 95% CI ‐0.49 to 0.70; Analysis 1.13), with very small effect sizes for both outcomes.

Analysis

Comparison 1: Comprehensive psychotherapies: active vs. control conditions, Outcome 12: Identity disturbance

1.11.6 Depression

No indication is given for a statistically significant difference between both conditions in terms of depression (SMD ‐0.05; N = 44; 95% CI ‐0.64 to 0.55; Analysis 1.14).

1.11.7 Anxiety

There is no indication for a statistically significant difference between both groups (SMD ‐0.52; N = 44, 95% CI ‐1.12 to 0.08; Analysis 1.15). However, data indicate a moderate favourable effect of IPT‐BPD.

1.11.8 Mental health status/functioning

Reported data indicate no statistically significant difference between both groups with regard to this outcome (SMD ‐0.04; N = 44; 95% CI ‐0.63 to 0.55; Analysis 1.17).

1.11.9 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.35; N = 55; 95% CI 0.08 to 1.57; Analysis 1.18).

No data were available for the outcome of general psychopathologiy burden. See Table 24 for an overview of primary effect outcome effect estimates.

2 Non‐comprehensive psychotherapeutic interventions versus control

2.1 Dialectical behaviour therapy‐skills training only (DBT‐ST) versus standard group (SG)

One study was included in this comparison: Soler 2009 (outpatients, 83% females; three months treatment; N = 60).

2.1.1 BPD total severity

Reported data indicate a very large, statistically significant difference at end of treatment between both conditions, favouring DBT‐ST (SMD ‐1.01; N = 59; 95% CI‐1.55 to ‐0.47; Analysis 2.1).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 1: BPD total severity

2.1.2 Affective dysregulative cluster symptoms

Data indicate large to very large statistically significant differences in favour of DBT‐ST concerning inappropriate anger (SMD‐0.84; N = 59; 95% CI ‐1.37 to ‐0.30; Analysis 2.2) and affective instability (SMD ‐1.07; N = 59; 95% CI ‐1.61 to ‐0.52; Analysis 2.3). No statistically significant difference was found for the outcome of chronic feelings of emptiness, with a moderate effect size (SMD ‐0.43; N = 59; 95% CI ‐0.95 to 0.09; Analysis 2.4).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 2: Inappropriate anger

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 3: Affective instability

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 4: Chronic feelings of emptiness

2.1.3 Impulsive cluster symptoms

Available data indicate a moderate, statistically significant difference favouring DBT‐ST in terms of impulsivity (SMD ‐0.61; N = 59; 95% CI ‐1.14 to ‐0.09; Analysis 2.5) but not suicidality (SMD ‐0.10; N = 59; 95% CI ‐0.61 to 0.41; Analysis 2.7). For suicidality, the effect size indicates only a very small to marginal difference.

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 5: Impulsivity

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 7: Suicidality

2.1.4 Interpersonal cluster symptoms

There was no indication of a statistically significant difference concerning interpersonal pathology, with a small effect size (SMD ‐0.29; N = 59; 95% CI ‐0.80 to 0.23; Analysis 2.10).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 10: Interpersonal problems

2.1.5 Cognitive cluster symptoms

Data indicate a moderate to large statistically significant difference favouring DBT‐ST in terms of psychotic symptoms (SMD ‐0.66; N = 59; 95% CI ‐1.18 to ‐0.13; Analysis 2.11).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 11: Dissociation/psychoticism

2.1.6 Depression

A large, statistically significant difference is indicated by the reported data favouring DBT‐ST with regard to depressive pathology (SMD ‐0.97; N = 59; 95% CI ‐1.51 to ‐0.43; Analysis 2.12).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 12: Depression

2.1.7 Anxiety

There is also an indication of statistically superiority with regard to anxious pathology with a moderate effect size (SMD ‐0.67; N = 59; 95% CI ‐1.20 to ‐0.15; Analysis 2.13).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 13: Anxiety

2.1.8 General psychopathology

A small to moderate, but statistically non‐significant difference was found for general psychopathology (SMD ‐0.42; N = 59; 95% CI ‐0.93 to 0.10; Analysis 2.14).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 14: General psychopathology

2.1.9 Mental health status/functioning

Data indicated a small but statistically non‐significant difference regarding the overall level of functioning (SMD ‐0.29; N = 59; 95% CI ‐0.80 to 0.22; Analysis 2.15).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 15: Mental health status/functioning

2.1.10 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.58; N = 60; 95% CI 0.34 to 1.00; Analysis 2.16).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 16: Leaving the study early

See Table 25 for an overview of primary effect outcome effect estimates.

7

Spreadsheet of primary outcome effect estimates: Dialectical behaviour therapy skills training(DBT‐ST) versus control

2.2 Emotion regulation group (ERG) versus treatment as usual (TAU)

One trial was included in this comparison: Gratz 2006 (female outpatients; 3.5 months treatment; N = 24)

2.2.1 BPD total severity

Reported data indicate a very large, statistically significant difference favouring ERG treatment (SMD ‐1.02; N = 22; 95% CI ‐1.92 to ‐0.11; Analysis 2.1).

2.2.2 Affective dysregulative cluster symptoms

Available data indicate a very large, statistically significant difference favouring ERG over TAU alone regarding affective instability (SMD ‐1.65; N = 22; 95% CI ‐2.65 to ‐0.55; Analysis 2.3).

2.2.3 Impulsive cluster symptoms

Data indicate very large, statistically significant differences between both conditions favouring ERG regarding the outcomes of impulsivity (SMD ‐1.30; N = 22; 95% CI ‐2.24 to ‐0.36; Analysis 2.5) and parasuicidality (SMD ‐0.98; N = 22; 95% CI ‐1.88 to ‐0.09; Analysis 2.8).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 8: Parasuicidality

2.2.4 Depression

Reported data indicate very large, statistically significant difference (SMD ‐1.20; N = 22; 95% CI ‐2.13 to ‐0.28; Analysis 2.12).

2.2.5 Anxiety

There was also indication of a large, statistically significant difference regarding anxious pathology (SMD ‐0.89; N = 22; 95% CI ‐1.78 to ‐0.01; Analysis 2.13).

2.2.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.58; N = 60; 95% CI 0.34 to 1.00; Analysis 2.16).

No data were available for the following outcomes: interpersonal and cognitive cluster symptoms; general psychopathology, mental health status/functioning. See Table 26 for an overview of primary effect outcome effect estimates.

8

Spreadsheet of primary outcome effect estimates: Emotion regulation group therapy (ERG) versus control condition

2.3 Schema‐focused therapy ‐ group intervention (SFT‐G) versus treatment as usual (TAU)

One study was included in this comparison: Farrell 2009 (female outpatients; 8 months treatment; N = 32).

2.3.1 BPD total severity

Farrell 2009 report data indicating a very large, statistically significant effect in favour of SFT‐G (SMD ‐1.66; N = 28, 95% CI ‐2.54 to ‐0.78; Analysis 2.1).

2.3.2 Affective dysregulative cluster symptoms

Reported data indicate a very large, statistically significant difference with regard to general affective instability (SMD ‐1.41; N = 28, 95% CI ‐2.26 to ‐0.57; Analysis 2.3).

2.3.3 Impulsive cluster symptoms

Farrell 2009 found a very large, statistically significant difference favouring SFT‐G with regard to BPD‐specific impulsivity (SMD ‐1.92; N = 28, 95% CI ‐2.85 to ‐1.00; Analysis 2.5).

2.3.4 Interpersonal cluster symptoms

There was a very large, statistically significant difference between both groups with regard to interpersonal pathology in favour of SFT‐G (SMD ‐1.94, N = 28, 95% CI ‐2.87 to‐1.02; Analysis 2.10).

2.3.5 Cognitive cluster symptoms

Reported data indicate a very large, statistically significant difference in favour of SFT‐G with regard to cognitive‐cluster symptoms (SMD ‐1.37, N = 28, 95% CI ‐2.21 to ‐0.53; Analysis 2.11).

2.3.6 General psychopathology

Farrell 2009 report data indicating a very large, statistically significant difference between the two conditions with better outcomes for the SFT‐G group (SMD ‐1.06, N = 28, 95% CI ‐1.87 to ‐0.25; Analysis 2.14).

2.3.7 Mental health status/functioning

Again, Farrell 2009 report data indicating a very large, statistically significant difference in favour of SFT‐G (SMD 1.20; N = 28, 95% CI 0.38 to 2.03; Analysis 2.15).

2.3.8 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.11; N = 32; 95% CI 0.01 to 1.91; Analysis 2.16).

No data were available for the following outcomes: depression, anxiety. See Table 27 for an overview of primary effect outcome effect estimates.

9

Spreadsheet of primary outcome effect estimates: Schema‐focused group therapy (SFT‐G) versus control condition

2.4 Manual‐assisted cognitive treatment (MACT) versus treatment as usual (TAU)

One trial was included in this comparison: Weinberg 2006 (female outpatients; 1.5 months treatment; N = 30)

2.4.1 Impulsive cluster symptoms

Reported data indicate statistically significant differences favouring MACT for the outcomes of suicidality (SMD ‐0.86; N = 28; 95% CI ‐1.64 to ‐0.07; Analysis 2.7) and parasuicidality (SMD ‐0.88; N = 28; 95% CI ‐1.67 to ‐0.10; Analysis 2.8) with large effects.

2.4.2 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.20; N = 30; 95% CI 0.01 to 3.85; Analysis 2.16).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative, interpersonal and cognitive cluster symptoms; depression, anxiety, general psychopathology, mental health status/functioning. See Table 28 for an overview of primary effect outcome effect estimates.

10

Spreadsheet of primary outcome effect estimates: manual‐assisted cognitive treatment (MACT) versus controls

2.5 Systems training for emotional predictability and problem solving for BPD (STEPPS) versus treatment as usual (TAU)

One trial was included in this comparison: Blum 2008 (outpatients, 83% females; five months treatment; N = 124).

2.5.1 BPD total severity

Reported data indicate a small, statistically non‐significant difference between the treatment groups (SMD ‐0.17; N = 124; 95% CI ‐0.52 to 0.19; Analysis 2.1).

2.5.2 Affective dysregulative cluster symptoms

There was a small, statistically non‐significant difference between conditions in terms of affective instability (SMD ‐0.32; N = 124; 95% CI ‐0.67 to 0.04; Analysis 2.3).

2.5.3 Impulsive cluster symptoms

Available data indicate a small, statistically non‐significant difference between both groups for BPD‐speciifc impulsivity (SMD ‐0.29; N = 124; 95% CI ‐0.64 to 0.07; Analysis 2.5).

2.5.4 Interpersonal cluster symptoms

We found a small to moderate statistically significant difference favouring STEPPS treatment for the outcome of interpersonal pathology (SMD ‐0.42; N = 124; 95% CI ‐0.78 to ‐0.06; Analysis 2.10).

2.5.5 Cognitive cluster symptoms

There was also a small to moderate statistically significant difference with better outcomes of cognitive cluster symptoms in the STEPPS group (SMD ‐0.42; N = 124; 95% CI ‐0.78 to ‐0.06; Analysis 2.11).

2.5.6 Depression

Data indicate a small, but statistically non‐significant difference between conditions for the outcome of depression (SMD ‐0.24; N = 124; 95% CI ‐0.59 to 0.11; Analysis 2.12).

2.5.7 General psychopathology

There was a small, statistically non‐significant difference for the outcome of general psychopathology (SMD ‐0.29; N = 124; 95% CI ‐0.64 to 0.07; Analysis 2.14).

2.5.8 Mental health status/functioning

The treatment groups differed statistically significantly at the end of treatment with a small to moderate effect indicating better results for STEPPS, that is, better mental health status (SMD 0.38; N = 124; 95% CI 0.02 to 0.73; Analysis 2.15).

2.5.9 Leaving the study early

Reported data indicate a statistically significant difference between both condition in terms of different proportion of those leaving the study early (RR 2.27; N = 124; 95% CI 1.08 to 4.76; Analysis 2.15), with a higher drop‐out rate for STEPPS.

No data were available for the outcome of anxiety. See Table 29 for an overview of primary effect outcome effect estimates.

11

Spreadsheet of primary outcome effect estimates: systems training for emotional predictability and problem solving for borderline personality disorder (STEPPS) versus controls

2.6 Systems training for emotional predictability and problem solving for BPD + individual therapy (STEPPS) versus treatment as usual (TAU)

One study was included in this comparison: Bos 2010 (outpatients, 86% females; 6 months treatment; N = 79)

2.6.1 BPD total severity

Bos 2010 report data indicating a moderate, statistically non‐significant difference between both conditions, but the boundaries of the 95% CI are very close to significance, favouring STEPPS + IT (SMD ‐0.55; N = 52; 95% CI ‐1.11 to 0.00; Analysis 2.1).

2.6.2 Impulsive cluster symptoms

There are no statistically significant differences between both conditions in terms of proportion of those participants beyond the cut‐off scores of clinically relevant impulsivity (RR 0.93; N = 58; 95% CI 0.66 to 1.29; Analysis 2.6) and clinically relevant parasuicidality (RR 1.32; N = 58; 95% CI 0.78 to 2.22; Analysis 2.9).

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 6: Impulsivity

Analysis

Comparison 2: Non‐comprehensive psychotherapeutic interventions: active vs. control conditions, Outcome 9: Parasuicidality

2.6.3 Interpersonal cluster symptoms

Data indicate a small, statistically non‐significant difference between both treatment groups in terms of interpersonal pathology (SMD ‐0.27; N = 53; 95% CI ‐0.81 to 0.27; Analysis 2.10).

2.6.4 General psychopathology

There was a moderate, statistically significant difference between both conditions in favour of STEPPS + IT (SMD ‐0.60; N = 51; 95% CI ‐1.16 to ‐0.04; Analysis 2.14).

2.6.5 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 1.47; N = 79; 95% CI 0.59 to 3.65; Analysis 2.16).

No data were available for the following outcomes: affective‐dysregulative and cognitive cluster symptoms; depression, anxiety, mental health status/functioning. See Table 29 for an overview of primary effect outcome effect estimates.

2.7 Psychoeducation (PE) versus waiting list (WL)

The only study included in this comparison was Zanarini 2008 (newly‐diagnosed female outpatients; three months treatment; N = 50)

2.7.1 Impulsive cluster symptoms

Data indicate a small to moderate but statistically non‐significant difference between both groups in terms of impulsivity baseline to endpoint change scores (SMD ‐0.47; N = 50; 95% CI ‐1.04 to 0.10; Analysis 2.5).

2.7.2 Interpersonal cluster symptoms

We found a moderate to large, statistically significant difference between both groups in terms of interpersonal pathology baseline to endpoint change scores (SMD ‐0.75; N = 50; 95% CI ‐1.33 to ‐0.16; Analysis 2.10).

2.7.3 Leaving the study early

All participants attended every visit, there were no non‐attendances in either group.

No data were available for the following outcomes: BPD total severity; affective‐dysregulative and cognitive cluster symptoms; depression, anxiety, general psychopathology, mental health status/functioning. See Table 30 for an overview of primary effect outcome effect estimates.

12

Spreadsheet of primary outcome effect estimates: Psychoeducation (PE) versus control

3 Comprehensive psychotherapies: active versus active conditions

3.1 Dialectical behaviour therapy (DBT) versus Client‐centered therapy (CCT)

This comparison included one trial: Turner 2000 (outpatients, 79% females; 12 months therapy; N = 24)

3.1.1 Affective dysregulative cluster symptoms

Data indicate a tendency towards better results of DBT with a large effect for the outcome of anger, but no statistically significant difference (SMD ‐0.79; N = 24; 95% CI ‐1.62 to 0.05; Analysis 3.2).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 2: Inappropriate anger

3.1.2 Impulsive cluster symptoms

Reported data indicate large to very large statistically significant differences between both groups for impulsive symptom cluster outcomes, all favouring DBT over CCT (impulsivity: SMD ‐1.05, N = 24, 95% CI ‐1.92 to ‐0.19, Analysis 3.3; suicidality: SMD ‐0.87, N = 24, 95% CI ‐1.71 to ‐0.02, Analysis 3.4; parasuicidality: SMD ‐1.28, N = 24, 95% CI ‐2.17 to ‐0.38, Analysis 3.5).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 3: Impulsivity

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 4: Suicidality

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 5: Parasuicidality

3.1.3 Cognitive cluster symptoms

There was also a very large, statistically significant difference for the outcome of psychotic symptoms, favouring DBT over CCT treatment (SMD ‐1.11; N = 24; 95% CI ‐1.98 to ‐0.24; Analysis 3.6).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 6: Psychoticism

3.1.4 Depression

Available data also indicate a very large, significant difference for the outcome of depression, with lower burden in DBT‐treated participants (SMD ‐1.26; N = 24; 95% CI ‐2.15 to ‐0.37; Analysis 3.7).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 7: Depression

3.1.5 Anxiety

Data indicate no significant difference between both conditions for the outcome of anxiety, but a moderate to large difference (SMD ‐0.70; N = 24; 95% CI ‐1.53 to 0.13; Analysis 3.8).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 8: Anxiety

3.1.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.50; N = 24; 95% CI 0.16 to 1.55; Analysis 3.11).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 11: Leaving the study early

No data were available for the following outcomes: BPD total severity; interpersonal cluster symptoms; general psychopathology, mental health status/functioning.

3.2 Schema‐focused therapy (SFT) versus Transference‐focused therapy (TFP)

One trial was included in this comparison: Giesen‐Bloo 2006 (outpatients, 93% females; 36 months treatment; N = 88).

3.2.1 BPD total severity

Giesen‐Bloo 2006 report data indicating a moderate, statistically significant difference between both treatment groups. SFT‐treated participants showed significantly lower levels of BPD total severity (SMD ‐0.45; N = 86; 95% CI ‐0.88 to ‐0.02; Analysis 3.1)

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 1: BPD total severity

3.2.2 General psychopathology

Data indicate no statistically significant difference between both conditions for the outcome of general psychopathology (SMD ‐0.09; N = 86; 95% CI ‐0.52 to 0.33; Analysis 3.9), with a marginal effect size.

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 9: General psychopathology

3.2.3 Leaving the study early

Reported data indicate a statistically significant difference between both condition in terms of a different proportion of those leaving the study early (RR 0.52; N = 88; 95% CI 0.30 to 0.92; Analysis 3.11), with a smaller drop‐out rate for SFT.

No data were available for the following outcomes: affective‐dysregulative, impulsive, interpersonal and cognitive cluster symptoms; depression, anxiety, mental health status/functioning.

3.3 Schema‐focused therapy (SFT) versus Schema‐focused therapy + Therapist Telephone Availability (SFT+TTA)

One trial was included in this comparison: Nadort 2009 (outpatients, 9´7% females; 18 months treatment; N = 62).

3.3.1 BPD total severity

Data indicate no clinically or statistically significant difference between both conditions for the outcome of BPD total severity (SMD ‐0.03; N = 61; 95% CI ‐0.53 to 0.47; Analysis 3.1).

3.3.2 General psychopathologiy

Data indicate a very small, statistically non‐significant difference between both conditions for the outcome of general psychopathology (SMD 0.14; N = 61; 95% CI ‐0.37 to 0.64; Analysis 3.9).

3.3.3 Leaving the study early

Reported data indicate no statistically significant difference between both conditions in terms of a different proportion of those leaving the study early (RR 0.91; N = 62; 95% CI 0.35 to 2.41; Analysis 3.11).

No data were available for the following outcomes: affective‐dysregulative, impulsive, interpersonal and cognitive cluster symptoms; depression, anxiety, mental health status/functioning.

3.4 Cognitive therapy (CT) versus Client‐centered therapy (CCT)

One trial was included in this comparison: Cottraux 2009 (outpatients, 77% females; 12 months treatment; N = 65)

3.4.1 Impulsive cluster symptoms

Data were available for the outcomes of impulsivity, suicidality and parasuicidality. Neither indicated a statistically significant difference between the experimental conditions (impulsivity: SMD ‐0.22, N = 38, 95% CI ‐0.86 to 0.41, Analysis 3.3; suicidality: SMD 0.13, N = 38, 95% CI ‐0.51 to 0.77, Analysis 3.4; parasuicidality: SMD 0.59, N = 38, 95% CI ‐0.06 to 1.24, Analysis 3.5). The data favoured CT for impulsivity with a small effect, and CCT for the outcomes of suicidality (very small effect) and parasuicidality (moderate effect).

3.4.2 Depression

Data indicate a very small, statistically non‐significant difference between the two conditions (SMD 0.10; N = 38; 95% CI ‐0.54 to 0.73; Analysis 3.7).

3.4.3 Anxiety

There was also no indication of a statistically significant difference between the two conditions in terms of anxious pathology, with a moderate effect favouring CT (SMD ‐0.51; N = 38; 95% CI ‐1.16 to 0.13; Analysis 3.8).

3.4.4 Mental health status/functioning

The two treatments did also not differ in a statistically significant way regarding the outcome of general psychopathological burden, with a small effect favouring CT (SMD 0.30; N = 38; 95% CI ‐0.34 to 0.94; Analysis 3.10).

Analysis

Comparison 3: Comprehensive psychotherapies: active vs. active conditions, Outcome 10: Mental health status/functioning

3.4.5 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 0.90; N = 65; 95% CI 0.51 to 1.60; Analysis 3.11).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative, interpersonal and cognitive cluster symptoms; general psychopathology.

3.5 Cognitive therapy (CT) versus Interpersonal psychotherapy (IPT)

One study was included in this comparison: Bellino 2007 (outpatients with concurrent major depressive disorder, 79% females; six months treatment; N = 32).

3.5.1 Depression

Data indicated no statistically significant difference between both conditions with regard to the outcome of depression, with a marginal effect size (SMD ‐0.07; N = 26; 95% CI ‐0.84 to 0.70; Analysis 3.7).

3.5.2 Anxiety

There was also no indication of a statistically significant difference between the two conditions in terms of anxious pathology (SMD ‐0.53; N = 26; 95% CI ‐1.32 to 0.26; Analysis 3.8). However, the effect was moderate in size, favouring CT.

3.5.3 Mental health status/functioning

Data indicated no statistically significant difference between both conditions with regard to the outcome of the level of functioning, favouring CT with a small effect (SMD 0.19; N = 26; 95% CI ‐0.58 to 0.97; Analysis 3.10).

3.5.4 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for leaving the study early if considering all participants randomised (RR 2.00; N = 32; 95% CI 0.42 to 9.42; Analysis 3.11).

No data were available for the following outcomes: BPD total severity; affective‐dysregulative, impulsive, interpersonal and cognitive cluster symptoms; general psychopathology.

4. Non‐comprehensive psychotherapeutic interventions: active versus active conditions

4.1 Manual‐assisted cognitive therapy (MACT) versus manual‐assisted cognitive therapy plus therapeutic assessment (MACT+TA)

One study was included in this comparison: Morey 2010 (outpatients, 81% females; six weekly sessions; N = 16)

4.1.1 BPD total severity

Data indicated a small, statistically non‐significant difference between both conditions with regard to the outcome of BPD total severity (SMD ‐0.33; N = 16; 95% CI ‐1.32 to 0.66; Analysis 4.1).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 1: BPD total severity

4.1.2 Affective dysregulative cluster symptoms

Available data indicated a moderate to large, statistically non‐significant difference in terms of affective instability scores after treatment, favouring MACT (SMD ‐0.71; N = 16; 95% CI ‐1.73 to 0.31; Analysis 4.2).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 2: Affective instability

4.1.3 Impulsive cluster symptoms

Morey 2010 reported both on suicidality as well as parasuicidality. Both rendered marginal, statistically not different between‐group effects (suicidality: SMD ‐0.03; N = 16; 95% CI ‐1.01 to 0.95; Analysis 4.3; parasuicidality: SMD 0.08; N = 16; 95% CI ‐0.90 to 1.06; Analysis 4.4).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 3: Suicidality

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 4: Parasuicidality

4.1.4 Interpersonal cluster symptoms

Data indicated a marginal, statistically non‐significant difference between both conditions with regard to the outcome of interpersonal problems (SMD ‐0.04; N = 16; 95% CI ‐1.02 to 0.94; Analysis 4.5).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 5: Interpersonal problems

4.1.5 Cognitive cluster symptoms

There was also no indication of a statistically significant difference between the two conditions in terms of identity disturbance, with a small to moderate effect favouring MACT (SMD ‐0.45; N = 16; 95% CI ‐1.44 to 0.55; Analysis 4.6).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 6: Identity disturbance

4.1.6 Leaving the study early

Reported data indicate no statistically significant difference between treatment and control conditions for the risk of non‐completing all six therapy sessions (RR 0.80; N = 16; 95% CI 0.33 to 1.92; Analysis 4.7).

Analysis

Comparison 4: Non‐comprehensive psychotherapeutic interventions: active vs. active conditions, Outcome 7: Leaving the study early

No data were available for the following outcomes: depression, anxiety, general psychopathology, mental health status/functioning.

Discussion

Summary of main results

The following summary is ordered according to the overall number of trials available for a certain treatment or adaptations of the "original" treatment. For a summary of primary outcome effect estimates of controlled comparisons, see Table 19 to Table 30.

Dialectical Behaviour Therapy (DBT) and DBT‐derived treatments have been studied most intensively among the included trials. Findings from five studies comparing DBT to treatment as usual (TAU) (Koons 2001; Linehan 1991; Linehan 1994; Van den Bosch 2005; Carter 2010) indicate statistically significantly beneficial effects for the broadest range of outcomes relative to remaining treatments. The comparison of DBT to TAU was the only comparison that allowed for meta‐analytic pooling of effect estimates from several studies, but only for the outcomes of anger, parasuicidality and mental health status. Standard DBT was found to have beneficial effects in terms of anger (large effect), suicidal behaviour (very large effect) and parasuicidality (moderate effect), associated psychopathology such as depression (very large effect) and anxiety (very large effect) and overall mental health status (moderate to large effect). The remaining statistically non‐significant findings were based on single study results each, yielding a large effect for the outcome of dissociation and small effects for the outcomes of overall BPD severity, impulsivity and interpersonal problems.

However, if compared to more rigorous control conditions, that is, general management according to the APA guidelines (McMain 2009) or community treatment by experts (CTBE) (Linehan 2006), there were no statistically significant group differences for pathology‐related outcomes, but DBT was statistically superior to CTBE in terms of treatment retention. DBT was only marginally superior to APA guidelines general management (BPD severity, anger, interpersonal problems, general psychopathology), and small differences were found for the outcomes of parasuicidality and depression, favouring DBT. Compared with CTBE, there was a very small effect favouring DBT in terms of suicidality, and a small to moderate effect in terms of less depression in DBT‐treated patients.

DBT‐PTSD, a modified DBT approach developed to meet the specific needs of those with comorbid PTSD, was also found to be effective in the reduction of overall BPD severity, depression and anxiety (Steil 2010), with moderate (BPD severity) and very large (depression, anxiety), statistically significant effects. Statistically non‐significant effects were found for the outcomes of general psychopathology (moderate to large effect) and dissociation (small effect).

As compared to Client‐Centered Therapy CCT (the effects of which have not been tested against a control condition in BPD treatment up to now, and the efficacy of which thus is unclear), DBT showed better results in the reduction of impulsivity, suicidality, parasuicidality, dissociative pathology and depression, with large to very large, statistically significant effects throughout (Turner 2000). DBT had also favourable results for the outcomes of anger, depression and anxiety, with moderate to very large, but statistically non‐significant effects.

These findings are consolidated by trials of short‐term interventions derived from or including elements of DBT. A trial of DBT skills training only (DBT‐ST) as compared to a non‐specific standard group (Soler 2009) resulted in statistically significant results favouring DBT‐ST for BPD severity, anger, affective instability, impulsivity, dissociation, depression and anxiety with moderate to very large effects. Small to moderate, statistically non significant effects were found for the outcomes of chronic feelings of emptiness, suicidality, interpersonal problems, general psychopathology and overall mental health status, each favouring DBT‐ST.

Emotion Regulation Group Training (ERG), which was developed as a short‐time, group‐based approach and combines elements of DBT and Acceptance and Commitment Therapy (ACT), also showed encouraging, statistically significant results in terms of amelioration of BPD severity, affective instability, impulsivity, parasuicidality, depression and anxiety, with large to very large effects (Gratz 2006).

In sum, DBT and related treatments provide the most solid (but not sufficiently robust) evidence of efficacy relative to all treatments that have been investigated in RCTs so far.

Mentalisation‐Based treatment (MBT) is, second to DBT, the one treatment that provides most robust evidence of its efficacy. It has been compared with control groups in two trials, one testing MBT in a partial hospitalisation (MBT‐PH; Bateman 1999) and one testing MBT in an outpatient setting (MBT‐out; Bateman 2009). Both show consistently beneficial effects concerning the reduction of suicidality, parasuicidality, interpersonal problems and depression, with very large, statistically significant effects. In addition, data indicate beneficial effects also in terms of amelioration of general psychopathology and overall functioning for the outpatient setting. Small to moderate, statistically non‐significant effects were found for the reduction of anxiety and general psychopathology by MBT‐PH.

Transference‐Focused Psychotherapy (TFP) was tested in an outpatient study against an unspecific control therapy (CTBE; treatment period 12 months; Doering 2010) and directly against SFT (treatment period three years; Giesen‐Bloo 2006). Though tested against a rigorous comparison treatment (CTBE), TFP was found to have encouraging effects for the reduction of BPD severity (statistically significant, moderate effect) and treatment retention. There were small, unfavourable but statistically non‐significant effects for the outcomes of parasuicidality, depression, anxiety and general psychopathology, and a small, statistically non‐significant effect in favour of TFP for the outcome of overall functioning. As compared to Schema‐Focused Therapy (SFT), results indicated statistical superiority of SFT over TFP in the reduction of overall BPD severity (moderate effect) and treatment retention. As concerns general psychopathology, both treatments differed only marginally, resulting in a non‐significant finding.

SFT and derived treatments were subject to three trials: First, SFT was compared with TFP (Giesen‐Bloo 2006), with the above reported results of SFT being more effective in reducing overall BPD severity and keeping patients in treatment, but not in terms of reducing general psychopathology. Second, modified SFT in a group only format (SFT‐G) was subject to another trial (Farrell 2009). The authors report findings resulting in very large, statistically significant effects for all reported pathology‐related outcomes, that is, BPD severity, affective instability, impulsivity, interpersonal problems, dissociative/psychotic symptoms, general psychopathology, and overall mental health status. Third, SFT was compared with a modified form of SFT with additional therapist telephone availability in times of crisis (SFT+TTA; Nadort 2009). Both treatments did not differ in a statistically significant way, with marginal effects in terms of BPD severity and general psychopathology.

Interpersonal Psychotherapy (IPT) (Bellino 2006) and IPT adapted for BPD (IPT‐BPD; Bellino 2010) were supported by statistically significant findings of single trials each. IPT had statistically significant beneficial effects in terms of an amelioration of depression, with a large effect, though effects on BPD core pathology were not assessed and remain unclear. IPT‐BPD, however, showed large, statistically significant effects in the reduction of affective instability impulsivity and interpersonal problems. The remaining statistically non‐significant effects were marginal in size (anger, chronic feelings of emptiness, parasuicidality, avoidance of abandonment, identity disturbance, dissociative/psychotic symptoms, depression and mental health status). However, there was a moderate, statistically non significant effect favouring IPT‐BPD for the outcome of anxiety.

Cognitive‐Behaviour Therapy (CBT) was compared with TAU in a single trial (Davidson 2006). There were no statistically significant between‐group differences at post‐treatment for any out outcome. There was a trend of better results for CBT for the outcomes of suicidality, depression, anxiety, general psychopathology and mental health status. For two outcomes, the effects were opposite, favouring the control group with very small (parasuicidality) and small (interpersonal problems), statistically non‐significant effects.

Cognitive Therapy (CT) was compared with two alternate active treatments, that is, CCT (Cottraux 2009) and IPT (Bellino 2007). Both comparisons yielded no statistically significant differences for any outcome. As compared to CCT, results indicated a superiority of CT over CCT for the outcomes of impulsivity, suicidality, anxiety and mental health status with very small to moderate effects, and superiority of CCT for parasuicidality (moderate effect) and depression (very small effect). As compared to IPT, better results were found for CT for the outcomes anxiety (moderate effect) and mental health status (small effect). For the outcome of depression, the difference was marginal in size.

Dynamic Deconstructive Psychotherapy (DDP) was also investigated in a single trial only (Gregory 2008). There were no statistically significant results, but DDP was indicated to be superior to the control group in terms of BPD severity and parasuicidality and depression with small to moderate effects.

Systems training for emotional predictability and problem solving for borderline personality disorder (STEPPS) was subject to two trials. First, STEPPS was compared with TAU (Blum 2008). All results favoured STEPPS with statistically significant, small to moderate effects for interpersonal problems, dissociation and mental health status, and small, statistically non‐significant effects for the outcomes of overall BPD severity, affective instability, impulsivity, depression and general psychopathology. Notably, there was a statistically significant higher drop‐out rate in the STEPPS group. Another trial compared STEPPS plus limited individual therapy (STEPPS + IT; Bos 2010) to TAU. There was a statistically significant, moderate effect indicating beneficial effects of STEPPS in terms of general psychopathology. Statistically non‐significant, favourable effects were found for the reduction of BPD severity (moderate in size), impulsivity (small effect), and interpersonal problems (small effect). For the outcome of parasuicidality, there was statistically non‐significant effect indicating more such pathology in the STEPPS + IT group.

Manual‐Assisted Cognitive Treatment (MACT) was also tested in two trials: First, it was compared with TAU (Weinberg 2006). There were two large, significant effects favouring MACT over TAU in terms of suicidality and parasuicidality. Second, MACT was compared with MACT enhanced by pre‐treatment individual therapeutic assessment (Morey 2010). There were no statistically significant differences, with a moderate to large effect for affective instability, a moderate for identity disturbance and a small one for BPD severity, all favouring MACT. Differences in terms of suicidality and parasuicidality were marginal.

At last, a very generic psychoeducative intervention (PE) was compared with a waiting list (WL) control group (Zanarini 2008). There was a nearly large, statistically significant effect for the reduction of interpersonal problems, and an almost moderate but non‐significant effect for the reduction of impulsivity by this intervention.

Overall completeness and applicability of evidence

Participants

Most trial participants were women. Overall, 89% of included participants were female, and there was not a single study with more men than women, or even balanced proportions. This pattern may reflect reality in clinical settings, where about 75% of all BPD diagnoses are given to women (APA 2000). However, some doubt that the 'real' prevalence of BPD is actually higher in women than men, and there are contradictory findings of balanced proportions or even higher prevalence rates in men (Torgersen 2009). Men with BPD may exhibit another clinical picture than women, and especially antisocial features may be more prevalent in men than women with BPD (Skodol 2009). From a clinical point of view, this may yield different treatment approaches or special requirements that may not be reflected in the actual findings from the RCT evidence. In contrast, antisocial features or full antisocial personality disorder were reasons for exclusion in several trials (see Types of participants). Thus, the applicability of findings of this review, and most BPD research findings, to men may be confined.

Objective measures of levels of functioning at baseline were not available from all studies, and if so, different measures were used, rendering comparability of single study samples difficult. Overall, drawing from those studies that reported objective measures such as GAF, GAS, CGI or SFQ, the severity of illness ranged between major impairment, comparable to psychiatric emergencies, and mild illness, with the majority of samples exhibiting serious to moderate levels of illness. Regarding comorbidity, those with comorbid "severe mental disorders" such as psychotic and/or bipolar disorders were not eligible for study participation in most trials, as were those with mental impairments, organic brain disorder and severe organic conditions. Substance‐related disorders were also common reasons for exclusion (17 studies; some differentiated between substance abuse and addiction or specific substances, some did not (see. Types of participants). Some studies, however, concentrated on participants with distinct comorbid conditions, such as alcohol abuse or dependence (Gregory 2008), post‐traumatic stress disorder (Steil 2010) or major depressive disorder episodes (Bellino 2006; Bellino 2007).

Notably, acute endangerment of self and/or others was explicitly specified a reason for exclusion only in three trials. However, it seems not improbable that others may also have excluded patients in acute crisis. However, anxiety‐related disorders or eating disorders were not explicit reasons of exclusion in any study. There was only one trial that explicitly did not include patients with concurrent major depressive episode (Soler 2009).

Most studies were conducted either in Western Europe (14 trials: Bateman 1999; Van den Bosch 2005; Bellino 2006; Davidson 2006; Giesen‐Bloo 2006; Bellino 2007; Bateman 2009; Cottraux 2009; Nadort 2009; Soler 2009; Bellino 2010; Bos 2010; Doering 2010; Steil 2010) or North America (13 trials: Linehan 1991; Linehan 1994; Turner 2000; Koons 2001; Gratz 2006; Linehan 2006; Weinberg 2006; Blum 2008; Gregory 2008; Zanarini 2008; Farrell 2009; McMain 2009; Morey 2010). One study was conducted in New Zealand (Carter 2010). Thus, the applicability of findings to other locations and ethnicities, that is, as found in South America, Asia, Africa and also other parts of Europe and Oceania remains unclear.

In summary, the findings may be mostly applicable to a female, moderately to severely ill BPD patient without any comorbid severe mental condition such as psychotic disorder or substance‐related disorder, mental retardation or severe organic condition. However, we tried to exactly specify and describe all studies with regard to their crucial characteristics (see Description of studies, Characteristics of included studies) in order to let the reader decide about applicability of relevant study characteristics to his or her decisive situation.

Interventions    

All major psychotherapeutic treatments for BPD (DBT, MBT, SFT, TFP) have been tested in RCTs so far, though the number of RCTs varies with most trials investigating DBT. Treatment periods range broadly. Even if looking at long‐term treatments only (defined here as covering a period of more than six months), there is a broad range from six up to 36 months. It may surely depend of the specific national mental health care setting which treatment periods fit into the context of current practice, that is, which "amounts" of psychotherapeutic treatment a patient may be enabled to use.

A new trend of group‐based short‐time interventions can be observed in recent trials, with all referring studies dating from 2006 or later. Those short‐time interventions draw from already‐established interventions, especially DBT, and combine them with new elements and/or modify them according to group settings. Thus, a certain eclecticism can be observed, and in some cases it remains unclear which sources have been used. Short‐term interventions show encouraging results, however, the long‐term stability of effects has to be established. From a clinical point of view, it should also be made clear which "experience of treatment" a patient must already have to profit from those interventions, or if they should only be recommended to those patients who can use them as "booster treatment" on basis of their already‐gained treatment experience or as add‐on to concurrent individual psychotherapies only.

Most trials allowed concurrent psychotropic treatments. With the exception of the trials of Bellino 2006, Bellino 2007 and Bellino 2010, the participants of which were all given the SSRI antidepressant fluoxetine, most trialists did not prevent participants from taking concomitant medication. Some cited the APA guidelines as a rough orientation for medication regimens (Giesen‐Bloo 2006; Gregory 2008; Bateman 2009, McMain 2009 for control group patients). Most DBT trials explicitly encouraged tapering‐off of medications as an explicit treatment goal (Linehan 1991; Linehan 1994; Linehan 2006; McMain 2009). Thus, the actual use of medications during the study period differed between DBT and control groups as a consequence of the psychotherapy treatment goal of tapering‐off medications ("rely on skills over pills"), but the study groups did not differ at pretreatment. Overall, the wide‐spread use of pharmacotherapy in patients with BPD (Ansell 2007; Hörz 2010) is adequately reflected by the included primary studies.

During the time‐span that the included trials cover (that is, publication years range from 1991 to 2011), the understanding of the relative efficacy of medications in BPD has changed. For example, SSRIs that were once regarded the first‐line treatment for BPD (compare APA guidelines: APA 2001; Rush 2005) for the treatment of various symptom clusters lack corresponding evidence, whereas mood stabilisers and second‐generation antipsychotics are supported by some evidence (Stoffers 2010). It is therefore possible that  the efficacy of concurrent medications may have changed throughout  this long time‐span and may have confounded the corresponding findings. Within this review, this could possibly concern the comparison of DBT to TAU as this was the only comparison for which several study findings were pooled, covering a time‐span of publication years 1991 to 2010. However, in examining more closely the actual concomitant drug treatment in these studies, it becomes clear that antidepressants were most frequently used in all studies, rendering them studies comparable and limiting the risk of any confounding medication effects.

Comparisons

There was a large heterogeneity of control group treatments. Even "treatment as usual", a very common control condition, varied between studies. Some TAU participants were completely free to use or not to use any kind of care they would have used if not included into the study. Some TAU participants, however, received some minimum standard of care, for example, referrals to other providers (that participants could follow up or not) or a rough, guideline‐oriented but non‐specific treatment regimen. What is more, TAU will certainly vary internationally, the "usual" standards of mental health care are known to be different depending on national health care systems.

There is a clear trend towards new kinds of control conditions such as CTBE (Linehan 2006; Doering 2010) or guideline‐oriented treatments (Bateman 2009; McMain 2009). Reasons may be ethical considerations taking into account that severely ill patient should not be left untreated if established treatments such as DBT are available. In addition, it is interesting for the assessment of a certain therapy if this therapy not only works compared with TAU but also if compared with experienced therapists or therapists using guidelines. However, such comparisons will yield smaller between‐group effects if using more rigorous comparison conditions, and this should be taken into account if comparing data from such studies with those studies that use less rigorous control groups. From a reviewer's point of view, it is sometimes difficult to classify these newer comparison treatments. One may argue to group them as controls or as active comparisons, as well. Here, we decided to classify them as controls, since they do not use specific treatment manuals but only rough guidelines, so that there could have been variation in the type of treatment received by participants within the same control group. .

Another problem is the head‐to‐head testing of treatments before investigating the mere effects of the two treatments against control treatments (for example, CCT, SFT). Thus, it remains unclear what a between‐group difference really means. What is more, not only controlled studies but also pilot studies are mostly missing. This is in some ways astonishing taking into account the burden conducting a psychotherapy RCT to both investigators and participants.

Outcomes

Starting from the phenomenological diversity of BPD, there is also a high variety of possible outcomes, and by now there is only a small consensus about really important outcomes. Identifying theory‐inherent outcomes related to putative mechanisms of change is a major problem especially in psychotherapy research. The real core symptoms that people with BPD and caregivers may be interested in when looking for a helpful treatment option are sometimes neglected. Trialists should keep in mind that anyone concerned will certainly not be interested in how a change will putatively be achieved but if a change will be achieved at all, to what extent, and in terms of which pathology. The development of new outcome scales such as BPDSI‐IV, CGI‐BPD or ZAN‐BPD allows for distinct assessment of BPD symptoms. These scales, encouragingly, have potential to measure outcomes such as change in chronic feelings of emptiness or in avoidance of abandonment, which are significant traits in many people with BPD.

Attrition data must be interpreted with caution. There is a high risk of bias especially in small sample studies with, for example, also small numbers of therapists. What is more, there is a substantial likelihood of attrition rates depending on the kind of control groups used. For example, waiting list participants seem to be more prone to leaving the study early than TAU participants. In addition, geographical conditions and accessibility of study centres can play an important role (for example, if it is a rural region with high costs and burden for participants to go study assessments, higher drop‐out rates can be expected than in urban settings, cf. Blum 2008; Carter 2010).

None of the studies considered adverse or undesired effects. Some argue that psychotherapy, constituing a potent intervention, may both have the potential to cure and harm. A broad range of possible adverse outcomes has been discussed so far, including, e.g., the lack of significant improvement, acceleration of ongoing deterioration or increase of substance use if a certain intervention inducing high emotional arousal (Berk 2009).

Quality of the evidence

Overall, a total of 28 studies involving 1804 participants have been included in this review. Study sample sizes ranged between 16 and 180. However, except for DBT, the review findings are based on single study effects only. There were either single trials available for a certain intervention (for example, CBT, DDP, ERG, MACT), or the intervention had actually been tested in several trials, but in modified forms (for example, MBT and MBT‐PH; SFT and SFT‐G, STEPPS and STEPPS + IT), and/or it had been compared with different conditions (for example, CT, IPT), so the study effects could not be pooled either. 'Summary of findings' tables (see Table 1 to Table 18) are provided for all active to control group comparisons, focusing on primary outcomes. The comparison of DBT with TAU was the only comparison that allowed for pooling of several effect estimates. The overall quality of evidence ranges between moderate and low, depending on overall sample sizes. According to the GRADE system, a rating of 'high quality' requires as a rule of thumb large sample sizes of at least 400 due to statistical precision issues. To date, there are no such sample sizes available for any psychotherapy.

Key methodological limitations were, as usual in psychotherapy research, the lack of caregiver and patient blinding. Another limitation was potential bias due to allegiance effects, which were present in the majority of trials. However, it seems inevitable that psychotherapists who must undergo a thorough, often time‐ and cost‐intensive training to be able to deliver a certain psychotherapy properly, are allied to a certain psychotherapy they have "invested" in. On the other hand, a certain allegiance may also motivate them to deliver the treatment as properly as possible. However, treatment developers who also act as main investigators of their treatments should be prevented from suspicions that positive results stem from biased study designs or bias study conducting, so there is an urgent need for independent research endeavours to undermine available findings.

Another major limitation of most studies was attention bias. In the majority of trials, control group participants did not receive comparable amounts of professional attention as obligatory elements of their treatment regimen. Findings of beneficial effects by one treatment may then primarily result from simply being paid attention to or being provided with some kind of intervention rather than from a specific mechanism of action.

Most studies had comparatively small samples (12 out of 28 trials included 39 participants or less). As a consequence, the experimental groups may seem to be imbalanced at baseline, which was, for example, the case in the RCT of Gregory 2008. Dr Gregory, the main investigator of this trial, has raised concerns about the appropriateness of using endpoint data alone for effect size calculation, and referred to "substantially greater baseline psychopathology" in the active group, which may have led to an underestimation of positive treatment effects and a Type II error (Gregory 2017). However, group allocation was randomised (as in any here‐included study), so any imbalances at baseline are regarded to be due to chance alone, and the the statistical methods used in this review allow for such chance differences.

Nevertheless, we are aware that the power of studies including smaller samples (such as Gregory 2008) may be too small to detect a real effect as statistically significant if it exists. Therefore, we have not only reported statistically significance throughout the text but also referred to their magnitude ("small, moderate, large").

A funnel plot was drawn for all controlled comparisons and the outcome of parasuicidality (cf. Figure 4). The funnel plot is rather symmetric in shape and indicates no selection bias (for example, publication, delayed‐publication or location bias) or spurious inflation of effects in smaller studies due to poor methodological quality.

Funnel plot of comparison of all controlled comparisons for the outcome of parasuicidality

Potential biases in the review process

With regard to our inclusion criteria, we tried to retain a homogeneous pool of primary studies. However, there were some inconsistencies between studies, particularly pertaining to psychiatric comorbidity of study participants. For example, presence of a substance‐related disorder was a common exclusion criterion (see Types of participants), whereas one study (Gregory 2008) required participants to have such a disorder, and another study included a mixed sample of participants with and without substance abuse problems (Van den Bosch 2005). In addition, the severity of illness varied between studies, covering a range of severe to mild.

Many studies provide outcome measures that appear to reflect theory‐inherent, putative mechanisms of change than consumer‐relevant data. We tried to deal with this by first defining all patient‐relevant outcome variables that are directly (primary outcomes) or indirectly (secondary outcomes) associated with BPD, that is, all outcome variables that a consumer and his or her therapist are likely to be interested in, and took BPD pathology as defined in the DSM‐IV criteria (APA 2000) as a guideline for primary outcomes. We feel that these are most likely to be shared by the majority of patients with BPD, and did not regard cost‐related issues.

We decided to focus on RCTs only as they provide the only way to prevent systematic differences and confounders. One major concern against uncontrolled trials of BPD treatment outcome pertains to the characteristically unstable course of BPD, findings of amelioration over time and high affective responsiveness, which render simple pre‐post comparisons (within‐subject) difficult and prone to bias. Another practical consideration was the availability of reliable methods for identifying studies of another type. The identification of 28 relevant RCTs seems to vindicate this strategy.

A potential bias in the review process may have resulted from the decision not to include RCTs that do not provide any outcome of interest. Thus, the trials of Linehan 1999 and Linehan 2002 that both concentrate on outcome measures related to substance‐related disorders were excluded. In addition, the lack of usable data led to exclusion of Clarkin 2007 (see Characteristics of excluded studies).

We strived to identify all relevant published and unpublished RCT evidence (see Search methods for identification of studies). The search was not restricted to any language. In spite of great efforts to minimise publication bias, we were able to include only one unpublished study, the publication of which is impending (Steil 2010).

Agreements and disagreements with other studies or reviews

This review differs from its preceding version (Binks 2006) insofar as more RCTs are available for inclusion. The previous version included seven studies compared to the 28 in the current version. However, the statement that "the studies are too few and small to inspire full confidence in their results" (Binks 2006) remains relevant. The variety of available treatments has increased, whereas robustness has not, with the exception of DBT and MBT. On the other hand, the earlier conclusion that "some of the problems frequently encountered by people with borderline personality disorder may be amenable to talking/behavioural treatments" (Binks 2006) has been strengthened.

Though this Cochrane Collaboration review is not a guideline, its findings are likely to be checked against guideline recommendations and the following may be relevant.

The current APA guidelines (APA 2001; Oldham 2005) are substantially outdated, with the most recent psychotherapy RCT evidence included dating from 2003. In their 2005 update, the guideline authors conclude: "All in all, the database is growing, and further evidence is accumulating that BPD is a condition that can be effectively treated by a combination of psychotherapy and symptom‐targeted pharmacotherapy." (Oldham 2005, p. 4). We can not draw any distinct conclusions about the combination of psychotherapy and pharmacotherapy from this review. However, there were two RCTs in which all participants were given fluoxetine, and a psychotherapeutic approach (Bellino 2006: IPT; Bellino 2010: IPT‐BPD) was compared with CM in each case. Both trials consistently indicate that the group receiving combined treatment of psychotherapy and pharmacotherapy had superior results as compared to medication plus CM only. This finding advocates in particular for the conclusion of the APA guidelines that "...psychotherapy represents the primary, or core, treatment for this disorder and that adjunctive, symptom‐targeted pharmacotherapy can be helpful." (Oldham 2005, p. 3).

The comparison of pharmacotherapeutic and psychotherapeutic interventions was neither the scope of this review nor can it be answered from the here‐included evidence. The effects of pharmacotherapy of BPD is the subject of two other Cochrane Collaboration reviews (Lieb 2010; Stoffers 2010). One may argue that in the light of comparable effect sizes and a by and large comparable robustness of underlying evidence, it is not clear if either drug treatment or psychotherapy should be regarded the first‐line treatment and which one the adjunctive one. Again, this question can neither be answered from this review nor can the results of the two reviews be directly compared with each other. However, some observations from the two reviews: the pharmacotherapy review (Stoffers 2010) included three trials testing psychotherapy plus drug against psychotherapy plus placebo (Simpson 2004; Soler 2005; Linehan 2008). The findings were not conclusive in terms of a clear superiority of combined treatments. In contrast, there were indeed favourable results for psychotherapy plus placebo over psychotherapy plus drug, especially with regard to self‐harming and dissociative behaviour. As discussed previously, the two trials included here testing drug treatment plus psychotherapy against drug treatment plus clinical management only (Bellino 2006; Bellino 2010), indicated superior results for those participants who received additional psychotherapy. As a consequence, drug treatment plus psychotherapy seems not clearly superior to psychotherapy alone, whereas psychotherapy plus drug treatment had favourable results as compared to drug treatment alone. These findings support the role of psychotherapy as the core treatment, as also suggested by the APA guidelines (APA 2001; Oldham 2005).

However, the effectiveness of therapies in combination still remains unclear on basis of this review and the available RCT evidence.

The UK National Institute for Health and Clinical Excellence (NICE) published their guidelines in 2009 (NICE 2009), covering relevant evidence available up to April 2008. This review now includes 21 more studies. However, the conclusions that "There are few studies; low numbers of patients and therefore low power; multiple outcomes with few in common between studies; and a heterogeneous diagnostic system which makes it hard to target specific treatment on patients with specific sets of symptoms." (NICE 2009, p. 204) remain relevant. This review is also in line with NICE concerning the valuation that especially DBT and MBT are supported by the current evidence. However, TFP and SFT must be added to the class of treatments that showed beneficial effects in at least one methodologically sound, medium‐size trial. NICE concluded that "very brief interventions (less than 3 months) do not appear to be effective in the treatment of borderline personality disorder" (NICE 2009, p. 204). From our findings, based also on more recent RCTs, short‐time interventions of up to six months duration show at least encouraging results in small studies, though the necessary contexts that patients can profit from these interventions remains unclear as previously discussed (see Overall completeness and applicability of evidence).

Authors' conclusions

Implications for practice

In sum, the up‐to‐date available randomised controlled trial (RCT) evidence for psychotherapeutic treatments is scarce, and replicative studies would be most desirable for each type of treatment. Therefore, conclusions have to be drawn carefully.

Most "robust" conclusions can be drawn from evidence of dialectical behaviour therapy (DBT). It is the only psychotherapy for which, in comparison with treatment as usual (TAU), data could be accumulated from several trials. For all remaining comparisons and psychotherapies, single study effects are available only. Thus, the evidence is not as robust as would be desirable, but the findings indicate the usefulness of both comprehensive psychotherapies and non‐comprehensive psychotherapeutic interventions in the treatment of both core borderline personality disorder (BPD) pathology and its associated pathology.

Comorbid conditions have been recognised by RCT research, but the evidence base for BPD treatment in the presence of defined comorbidities (for example, substance‐related disorders, post‐traumatic stress disorder (PTSD), depressive episodes) is still small and only single studies are available. Studies of DBT‐PTSD for patients with BPD with comorbid PTSD and dynamic deconstructive psychotherapy (DDP) for patients with concurrent addictive disorders found encouraging effects on core BPD pathology as well as associated pathology, indicating that psychotherapy may have beneficial effects in severely ill groups of patients. Interpersonal psychotherapy (IPT) which was developed for people with depressive disorders was successful in reducing depressive pathology, but the effects on BPD core pathology remain unclear as no such data were assessed.

Trials of non‐comprehensive psychotherapeutic interventions suggest large effects for several approaches and indicate that these rather short‐term interventions may (at least as add‐ons to long‐term treatments) be helpful; further research is required. To date, it is rather unclear if these interventions are equally effective in patients with BPD who have already experienced several individual therapies as in patients who are "psychotherapy naive".

In sum, it can be concluded that disorder‐specific treatments should be used. Although nonspecific treatments were scarcely investigated, those that were (cognitive behavioural therapy (CBT), client‐centered therapy (CCT), IPT) showed no encouraging effects for the treatment of BPD core pathology. Beneficial findings from group‐only interventions are mostly based on studies of participants with another ongoing individual treatment, and it is therefore unclear which treatment exerts the major effect. The optimal length of treatment is unclear from the up‐to‐date RCT evidence. On the basis of the available findings, a treatment of 12 to 18 months seems to be appropriate.

Implications for research

First, replication studies would be most desirable, especially from independent researchers not involved in treatment development and/or delivery. Future studies should focus on male patients with BPD who have been neglected in BPD treatment studies so far, and treatment efficacy in patients with BPD with defined comorbid conditions should be investigated in more detail. In addition, it remains unclear how psychotherapy and pharmacotherapy interact. There is some evidence that psychotherapy may enhance pharmacotherapy effects (Lieb 2010; Stoffers 2010). There is a need for agreement on a minimum core battery of BPD outcomes. There is currently a huge heterogeneity of outcome variables and assessment instruments. A consensus on a minimum set of therapy outcome variables that are most likely to be of interest for any patient with BPD would be desirable. Outcome assessment should be more specific and sensitive to BPD pathology. Assessment instruments have been developed lately to reflect BPD core pathology as described precisely by the DSM‐IV criteria (for example, the BPDSI scale by Arntz 2003, the CGI‐BPD scale by Perez 2007, or the ZAN‐BPD scale by Zanarini 2003a). Further investigation of the impact of different treatment settings, that is, inpatient, outpatient, and day hospital, during acute and non‐acute stages of the course of illness would also be helpful. Some researchers advocate for the development of an integrated psychotherapy on the basis of effective treatments that combines methods which work from all therapies (for example, Livesley 2007; Livesley 2012). However, as discussed previously, there is still an urgent need for additional evidence to answer the question of which treatments are really effective, and which work for whom.

Feedback

Comments on the Gregory et al 2008 study, February 2017

Summary

On 14 February 2017, Dr Robert Gregory submitted the following comments, having identified himself as an author of an included study within the present review.

I have a number of concerns regarding the 2012 review of “Psychological Therapies for People with Borderline Personality Disorder”, specifically in regards to your review of Dynamic Deconstructive Psychotherapy (DDP). I realize that several years have passed since the Cochrane review. However, since the time of its publication, both clinical and academic colleagues have repeatedly brought the review to my attention, citing it as evidence that DDP is ineffective. So I felt I needed to respond.

I have a great respect for Cochrane reviews and the time and effort needed by the Cochrane authors to make a fair and accurate assessment. And in such a large review (over 200 pages), it is possible for oversights to occur. Nevertheless, since busy clinicians are making decisions that affect vulnerable patients based on the review, it’s important to get it right. I have listed my concerns below:

1. My largest concern is how the data from my 2008 study were analyzed and reported. The 2008 study by Gregory and colleagues was a small randomized controlled trial (n = 30) comparing DDP to TAU for participants suffering from co‐occurring borderline personality disorder (BPD) and alcohol dependence. In their Discussion section, Summary of Main Results, the authors conclude that “There were no statistically significant results, but DDP was indicated to be superior to the control group in terms of BPD severity and parasuicidality, with small to moderate effects” (p 74). The authors draw the same conclusion on pages 2 and 31.

This would be a reasonable conclusion based on the Cochrane authors’ method of analysis, which calculated standardized mean differences “on the basis of post‐treatment results” (p 15). The difficulty of that method of analysis for the 2008 study was that there were substantial differences in baseline scores between the two groups, which were not accounted for in the analysis. Participants assigned to DDP through randomization had substantially greater baseline psychopathology than control participants (seeGregory 2008, Table 4, p 37). In fact, on the continuous outcome variables, participants receiving DDP had baseline scores averaging more than half a standard deviation higher than the control group! By not taking into account baseline differences in either the analysis, summary of results, or discussion, the Cochrane authors markedly understate positive treatment effects and leave clinicians with the wrong impression that DDP was shown to be an ineffective treatment in this RCT, a potential Type II error.

2. There was also a simple error in the same section of the report summarizing results in the wrong direction. The Cochrane authors state: “For depression, a non‐statistically significant effect favouring the control group was found” (p 74). This conclusion contradicts the authors’ own analysis of treatment effects for depression as outlined on p 164 and p 184 of the report, and is clearly a misstatement and an honest mistake in a very lengthy report. Nevertheless, it’s important to get it corrected.

3. The Cochrane authors mention concern regarding attentional bias in the 2008 RCT (see p 14, p 26, and p 120). A strength of the RCT (Gregory 2008) is that the control participants were assigned to the best alternative treatment available in the community. In regards to total treatment contact hours, there were no statistically significant differences between the groups; three of the time points indicated contact hours favoring the control group (3, 6, and 12 months) and one time point (9 months) favored the DDP group (see Table 2, p 33).

Admittedly, I am focusing on concerns and am omitting all the aspects that the authors did extremely well in this complex and large review. For that I am very appreciative. Nevertheless, Cochrane has an enormous responsibility given its well‐deserved reputation among clinicians for accurate and unbiased reporting of treatment reviews. The patient population with BPD is high risk, with substantial morbidity and mortality. Clinicians need to get the most accurate information possible and I hope you will consider amending your report to take these concerns into account.

Sincerely,

Robert J. Gregory, MD

Do you have any affiliation with or involvement in any organisation with a financial interest in the subject matter of your comment?:I do not have any affiliation with or involvement in any organisation with a financial interest in the subject matter of my comment.​
_____________________________________________________________________________________________________________________

Reply

On 20 February the Feedback Editor of the Cochrane Developmental, Psychosocial and Learning Problems Group (CDPLPG) sent the above comments to Professor Julian Higgins, Statistical Editor of the CDPLPG, for advice / comment, which was duly supplied on 27 February. This was then circulated to other members of CDPLPG's editorial team, approved and sent on to the author team.

On 27 March 2017, Dr Jutta Stoffers‐Winterling submitted the following comments on behalf of the author team as a whole:

First, we would like to thank Dr Gregory for his most respectful and appreciative feedback. We understand his concerns, and take them seriously.

Though an update of this review is currently being finalised by our working group, we would like to address Dr Gregory’s comments and concerns. In the following, we will try to reply to each of the points raised by Dr Gregory.

Ad 1:We agree that obviously greater pathology scores were observed in the experimental group (DDP) at baseline for most measures (cf.Gregory 2008, Table 4, p 37). We have sought statistical advice and were reassured that this is a situation that often occurs in samples of smaller size. However, as this is a randomised trial (using a minimization procedure), the trialists have gone to some lengths to try and ensure balance. That means, if baseline imbalance occurs, it is due to chance. The statistical methods used in our review to compare the groups allow for such chance differences, and we were advised not to change them (Higgins 2017 [pers comm]).

We understand your concerns regarding the possibility of Type II errors, and have, therefore, inserted the following amendment (see 'Discussion > Quality of the evidence'):

"Most studies had comparatively small samples (12 out of 28 trials included 39 participants or less). As a consequence, the experimental groups may seem to be imbalanced at baseline, which was, for example, the case in the RCT of Gregory 2008. Dr Gregory, the main investigator of this trial, has raised concerns about the appropriateness of using endpoint data alone for effect size calculation, and referred to "substantially greater baseline psychopathology" in the active group, which may have led to an underestimation of positive treatment effects and a Type II error (Gregory 2017). However, group allocation was randomised (as in any here‐included study), so any imbalances at baseline are regarded to be due to chance alone, and the the statistical methods used in this review allow for such chance differences." See Quality of the evidence.

Nevertheless, we are aware that the power of studies that include smaller samples (such as Gregory 2008) may be too small to detect a real effect as statistically significant if it exists. Therefore, we have not only reported statistical significance throughout the text but also referred to their magnitude ("small, moderate, large").

Ad 2:We apologise for this mistake! Actually, this finding was reported correctly elsewhere in the text but erroneously in the 'Discussion > Summary of main results' section. We have corrected this, so that it now reads:

“Dynamic Deconstructive Psychotherapy (DDP) was also investigated in a single trial only (Gregory 2008). There were no statistically significant results, but DDP was indicated to be superior to the control group in terms of BPD severity, parasuicidality and depression with small to moderate effects.” See Summary of main results.

Ad 3:Thank you for bringing this to our attention. We agree that there is actually a low risk of bias due to a higher amount of attention spent to the participants of the treatment of investigation (DDP).

We have now changed the judgement of attention bias from 'high risk' to 'low risk'. The 'Risk of bias' table now reads:

Figures 2 (“Risk of bias graph”) and 3 (“Risk of bias summary) were updated accordingly.

We also changed the text in the 'Results > Risk of Bias in included studies > Other potential sources of bias > Attention bias' section:

"Only 10 studies (Koons 2001; Giesen‐Bloo 2006; Linehan 2006; Bellino 2007; Bateman 2009; Cottraux 2009; Nadort 2009; Soler 2009; Doering 2010; Morey 2010) were rated as providing similar amounts of attention as obligatory components of the study protocol to all trial groups. The participants allocated to the control group of Gregory 2008 did not get an obligatory control treatment, but were referred to alternative treatments in the community and had not less but markedly more professional contact hours during most of the study period. The risk of attention bias was therefore also judged low for this trial. All remaining trials provided more attention (that is, in terms of frequency of appointments, involvement in additional group treatments etc.) to one group, usually the experimental group (EG)." See Other potential sources of bias.

We hope we have addressed your concerns adequately. An update of this review is currently under way, and publication is intended for the end of this year/early 2018. Of course, any changes incorporated now will also be included in the updated version.

We are aware that without the most valuable, time‐consuming and hard work of the primary study authors, no meta‐analysis or systematic review would exist. What is more, the study sample in Dr Gregory’s study included persons with BPD plus substance‐related disorder (alcohol abuse or dependence), who are usually considered as most difficult to treat (or to be kept in treatment at all), and we can understand how much work and effort this study must have meant.

Sincerely,

Jutta Stoffers‐Winterling, University Medical Center, Mainz, Germany

Contributors

Robert J Gregory, MD, SUNY Upstate Medical University, New York, USA
Email address:

Jutta Stoffers‐Winterling, University Medical Center, Mainz, Germany
Email address:

(on behalf of the review author team — Jutta Stoffers‐Winterling, Birgit A Völlm, Gerta Rücker, Antje Timmer, Nick Huband, Klaus Lieb)

From the editorial board of the CDPLPG:

Professor Julian Higgins, University of Bristol, UK (Statistical Editor, CDPLPG)
Email address:

Jane Dennis, PhD, University of Bristol, UK (Feedback Editor, CDPLPG)
Email address:

Professor Geraldine Macdonald, Coordinating Editor, University of Bristol, UK (Coordinating Editor, CDPLPG)
Email address:

Joanne Wilson, PhD, Managing Editor, Queens University Belfast, UK (Managing Editor, CDPLPG)
Email address:

What's new

History

Review first published: Issue 1, 2006

Notes

This is an update of a published review (Binks 2006), but it is no longer being updated because it was superseded by a new review in 2020 (Storebø 2020).

Acknowledgements

We are grateful to Jane Dennis, Margaret Anderson, Laura MacDonald, Elaine McKay, Chris Champion and Geraldine MacDonald from the Cochrane Developmental, Psychosocial and Learning Problems Group. In particular, we thank Jane Dennis, Margaret Anderson, Nick Huband, Mike Ferriter, Nadja Smajlaigic and Melanie Powney for their help in identifying relevant studies. We would also like to thank all primary study authors who responded to our requests and provided further information, and Mr H A Glazener who helped translating a Dutch paper.

We would like to thank the previous authors of this review, Claire Binks, Mark Fenton, Lucy McCarthy, Tracy Lee, Clive Adams and Conor Duggan. Additionally, we thank the German Cochrane Centre for supporting this work. We are grateful to Gerd Antes, director of the German Cochrane Centre, who made contact with CDPLPG and helped in gaining grants for financing this work. We are also grateful to Martin Schumacher, director of the Institute of Biostatistics and Medical Informatics at the University Medical Center Freiburg, who gave support in application submission. We are grateful to the German Ministry of Education and Research (BMBF; grant no. 01KG0609), the research committee of the University Medical Center Freiburg and the NHS Cochrane Collaboration Programme Grant Scheme (NIHR), UK for supporting this work. JS would like to thank the Ministry of Science, Research and Arts of the federal state of Baden‐Württemberg for supporting this work by a stipend.

Appendices

Appendix 1. Search strategies

Cochrane Central Register of Controlled Trials (CENTRAL)

#1 MeSH descriptor Personality Disorders explode all trees
#2 (moral near/2 insanity)
#3 (DSM and (axis and II))
#4 (ICD and (F60 or F61 or F62))
#5 ((Odd* or eccentric* or dramatic* or emotional* or anxious* or fearful*) near/5 cluster*)
#6 ("Cluster A" or "Cluster B" or "Cluster C")
#7 ((aggressiv* or anxious* or borderline* or dependent* or emotional* or passiv* or unstable) near/5 personalit*)
#8 anankastic* or asocial* or avoidant*or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or narciss* or
obsessiv* or paranoi* or psychopath* or sadist* or schizoid* or schizotyp* or sociopath*
#9 personalit* near/5 disorder*
#10 character disorder*
#11 anal* next (personalit* or character* or retentiv*)
#12 #1 or #2 or #3 or #4 or #5 or #6 or #7 or #8 or #9 or #10 or #11

MEDLINE

1 randomized controlled trial.pt.
2 controlled clinical trial.pt.
3 randomi#ed.ab.
4 placebo.ab.
5 drug therapy.fs.
6 randomly.ab.
7 trial.ab.
8 groups.ab.
9 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8
10 exp animals/ not humans.sh.
11 exp Personality Disorders/
12 (moral adj2 insanity).tw.
13 (DSM and (axis and II)).tw.
14 (ICD and (F60 or F61 or F62)).tw.
15 ((odd$ or eccentric$ or dramatic$ or emotional$ or anxious$ or fearful$) adj5 cluster$).tw.
16 ("Cluster A" or "Cluster B" or "Cluster C").tw.
17 ((aggressiv$ or anxious$ or borderline$ or dependent$ or emotional$ or passiv$ or unstable) adj5 personalit$).tw.
18 (personalit$ adj5 disorder$).tw.
19 character disorder$.tw.
20 (anal$ adj (personalit$ or character$ or retentiv$)).tw.
21 9 not 10
22 (anankastic$ or asocial$ or avoidant$ or antisocial$ or anti‐social$ or compulsiv$ or dissocial$ or histrionic$ or
narciss$ or obsessiv$ or paranoi$ or psychopath or psychopaths or psychopathic or sadist$ or schizoid$ or schizotyp$ or
sociopath$).tw.
23 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 22
24 21 and 23

EMBASE

1 (moral adj2 insanity).tw.
2 (DSM and (axis and II)).tw.
3 (ICD and (F60 or F61 or F62)).tw.
4 ((odd$ or eccentric$ or dramatic$ or emotional$ or anxious$ or fearful$) adj5 cluster$).tw.
5 ("Cluster A" or "Cluster B" or "Cluster C").tw.
6 ((aggressiv$ or anxious$ or borderline$ or dependent$ or emotional$ or passiv$ or unstable) adj5 personalit$).tw.
7 (anankastic$ or asocial$ or avoidant$ or antisocial$ or anti‐social$ or compulsiv$ or dissocial$ or histrionic$ or
narciss$ or obsessiv$ or paranoi$ or psychopath$ or sadist$ or schizoid$ or schizotyp$ or sociopath$).tw.
8 (personalit$ adj5 disorder$).tw.
9 character disorder$.tw.
10 (anal$ adj (personalit$ or character$ or retentiv$)).tw.
11 exp personality disorder/
12 or/1‐11
13 random$.tw.
14 factorial$.tw.
15 crossover$.tw.
16 cross over$.tw.
17 cross‐over$.tw.
18 placebo$.tw. (1
19 (doubl$ adj blind$).tw.
20 (singl$ adj blind$).tw.
21 assign$.tw.
22 allocat$.tw.
23 volunteer$.tw.
24 Crossover Procedure/
25 double‐blind procedure.tw.
26 Randomized Controlled Trial/
27 Single Blind Procedure/
28 or/13‐27
29 12 and 28

ASSIA

((DE=("personality disorders" or "antisocial personality disorder" or "avoidant personality disorders" or "borderline
personality disorder" or "dependent personality" or "depressive personality disorders" or "gender identity disorder" or
"histrionic personality disorder" or "identity crisis" or "kleptomania" or "multi impulsive personality disorder" or
"multiple personality disorder" or "narcissistic personality disorder" or "passive aggressive personality disorder" or
"sadistic personality disorder" or "schizotypal personality disorders" or "selfdefeating personality disorder")) or((moral
within 2 insanity) or(DSM and (axis and II)) or(icd and (F60 or F61 or F62)) or((odd* or eccentric* or dramatic* or
emotional* or anxious* or fearful*) within 5 cluster*) or("Cluster A" or "Cluster B" or "Cluster C") or((aggressiv* or
anxious* or borderline* or dependent* or emotional* or passiv* or unstable) within 5 personalit*) or(anankastic* or
asocial* or avoidant* or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or narciss* or obsessiv* or
paranoi* or psychopath* or sadist* or schizoid* or schizotyp* or sociopath*) or(personalit* within 5 disorder*) or(character
disorder*) or(anal* within 1 (personalit* or character* or retentiv*)))) and((DE=("randomized controlled trials" or "clinical
randomized controlled trials" or "cluster randomized controlled trials" or "double blind randomized controlled trials" or
"randomized consent design" or "single blind randomized controlled trials" or "urn randomization" or "random testing"
or "randomization" or "unequal randomization")) or(TI=randomi* or AB=randomi*) or((TI=(single or double) near
blind*)or (AB=(single or double) near blind*)))
BIOSIS

# 15 #14 AND #12
# 14 #13 OR #9 OR #8 OR #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1
# 13 TS=(anankastic* or asocial* or avoidant* or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or narciss* or obsessiv* or paranoi* or psychopath or psychopaths or psychopathic or sadist* or schizoid* or schizotyp* or sociopath*)
# 12 #11 OR #10
# 11 TS=(singl* or doubl* or tripl*or trebl*) OR TS=(mask* or blind*)
# 10 TS=(random* or crossover )
# 9 TS=((anal or anally) SAME (personalit* or character* or retentiv*))
# 8 TS=("Cluster B" or "Cluster C")
# 7 TS=("character disorder*")
# 6 TS=(personalit* SAME disorder*)
# 5 TS=((agressiv* or anxious* or borderline* or dependent* or emotional* or passiv* or unstable) SAME (personalit*))
# 4 TS=((odd* or eccentric* or dramatic* or emotional* or anxious* or fearful*) SAME cluster*)
# 3 TS=(ICD AND (F60 or F61 or F62))
# 2 TS=(DSM SAME (axis SAME II))
# 1 TS=(moral SAME insanity)

CINAHL

S30 S13 and S29
S29 S14 or S15 or S16 or S17 or S18 or S19 or S20 or S21 or S22 or S23 or
S24 or S25 or S26 or S27 or S28
S28 allocat* random*
S27 (MH "Quantitative Studies")
S26 (MH "Placebos")
S25 placebo*
S24 random* allocat*
S23 (MH "Random Assignment")
S22 (Randomi?ed control* trial*)
S21 (singl* mask* )
S20 (doubl* mask* )
S19 (tripl* mask* )
S18 (trebl* mask* )
S17 (trebl* blind* )
S16 (tripl* blind* )
S15 (doubl* blind* )
S14 (singl* blind* )
S13 S1 or S2 or S3 or S4 or S5 or S6 or S7 or S8 or S9 or S10 or S11 or S12
S12 (anal* N1 personalit* or anal* N1 character* or anal N1 retentiv*) and
(S1 or S2 or S3 or S4 or S5 or S6 or S7 or S8 or S9 or S10 or S11)
S11 anal* N1 personalit* or anal* N1 character* or anal N1 retentiv*
S10 character N1 disorder*
S9 personalit* N5 disorder*
S8 anankastic* or asocial* or avoidant* or antisocial* or anti‐social* or
compulsiv* or dissocial* or histrionic* or narciss* or obsessiv* or
paranoi* or psychopath* or sadist* or schizoid* or schizotyp* or
sociopath*
S7 aggressiv* N5 personalit* or anxious* N5 personalit* or borderline* N5
personalit* or dependent* N5 personalit* or emotional* N5 personalit* or
passiv* N5 personalit* or unstable N5 personalit*
S6 ("Cluster A" or "Cluster B" or "Cluster C")
S5 odd* N5 cluster* or eccentric* N5 cluster* or dramatic* N5 cluster* or
emotional* N5 cluster* or anxious* N5 cluster* or fearful* N5 cluster*
S4 (ICD and (F60 or F61 or F62))
S3 (DSM and (axis and II))
S2 moral N2 insanity
S1 (MH "Personality Disorders+")

Dissertation Abstracts

Searched using the Proquest keyword " personality disorders"

ICTRP

condition = borderline personality disorder AND recruitment status = all

metaRegister of Controlled Trials

Searched using the search strings:
antisocial personality
borderline personality
compulsive personality
dependent personality
histrionic personality
hysteria
paranoid personality
passive‐aggressive personality
schizoid personality
schizotypal personality
personality disorder

National Criminal Justice Reference Service Abstracts (NCJRS)

Searched using the search strings:
random* AND compulsive personality
random* AND antisocial personality
random* AND borderline personality
random* AND dependent personality
random* AND histrionic personality
random* AND hysteria
random* AND paranoid personality
random* AND passive aggressive personality
random AND schizoid personality
random* AND schizotypal personality
random* AND personality disorder*

PsycINFO

1 (moral adj2 insanity).tw.
2 (DSM and (axis and II)).tw.
3 (ICD and (F60 or F61 or F62)).tw.
4 ((odd$ or eccentric$ or dramatic$ or emotional$ or anxious$ or fearful$) adj5 cluster$).tw.
5 ("Cluster A" or "Cluster B" or "Cluster C").tw.
6 ((aggressiv$ or anxious$ or borderline$ or dependent$ or emotional$ or passiv$ or unstable) adj5 personalit$).tw.
7 (anankastic$ or asocial$ or avoidant$ or antisocial$ or anti‐social$ or compulsiv$ or dissocial$ or histrionic$ or
narciss$ or obsessiv$ or paranoi$ or psychopath$ or sadist$ or schizoid$ or schizotyp$ or sociopath$).tw.
8 (personalit$ adj5 disorder$).tw.
9 character disorder$.tw.
10 (anal$ adj (personalit$ or character$ or retentiv$)).tw.
11 exp personality disorders/
12 or/1‐11
13 randomi$.tw.
14 singl$.tw.
15 doubl$.tw.
16 trebl$.tw.
17 tripl$.tw.
18 blind$.tw.
19 mask$.tw.
20 (or/14‐17) adj3 (or/18‐19)
21 clin$.tw.
22 trial$.tw.
23 (clin$ adj3 trial$).tw.
24 placebo$.tw.
25 exp PLACEBO/
26 crossover.tw.
27 exp Treatment Effectiveness Evaluation/
28 exp Mental Health Program Evaluation/
29 random$.tw.
30 assign$.tw.
31 allocate$.tw. (
32 (random$ adj3 (assign$ or allocate$)).tw.
33 32 or 28 or 27 or 26 or 25 or 24 or 23 or 20 or 13
34 12 and 33

Science Citation Index

# 15 #14 AND #12
# 14 #13 OR #9 OR #8 OR #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1
# 13 TS=(anankastic* or asocial* or avoidant* or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or
narciss* or obsessiv* or paranoi* or psychopath or psychopaths or psychopathic or sadist* or schizoid* or schizotyp* or
sociopath*)
# 12 #11 OR #10
# 11 TS=(singl* or doubl* or tripl*or trebl*) OR TS=(mask* or blind*)
# 10 TS=(random* or crossover )
# 9 TS=((anal or anally) SAME (personalit* or character* or retentiv*))
# 8 TS=("Cluster B" or "Cluster C")
# 7 TS=("character disorder*")
# 6 TS=(personalit* SAME disorder*)
# 5 TS=((agressiv* or anxious* or borderline* or dependent* or emotional* or passiv* or unstable) SAME (personalit*))
# 4 TS=((odd* or eccentric* or dramatic* or emotional* or anxious* or fearful*) SAME cluster*)
# 3 TS=(ICD AND (F60 or F61 or F62))
# 2 TS=(DSM SAME (axis SAME II))
# 1 TS=(moral SAME insanity)

Social Science Citation Index

# 15 #14 AND #12
# 14 #13 OR #9 OR #8 OR #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1
# 13 TS=(anankastic* or asocial* or avoidant* or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or
narciss* or obsessiv* or paranoi* or psychopath or psychopaths or psychopathic or sadist* or schizoid* or schizotyp* or
sociopath*)
# 12 #11 OR #10
# 11 TS=(singl* or doubl* or tripl*or trebl*) OR TS=(mask* or blind*)
# 10 TS=(random* or crossover )
# 9 TS=((anal or anally) SAME (personalit* or character* or retentiv*))
# 8 TS=("Cluster B" or "Cluster C")
# 7 TS=("character disorder*")
# 6 TS=(personalit* SAME disorder*)
# 5 TS=((agressiv* or anxious* or borderline* or dependent* or emotional* or passiv* or unstable) SAME (personalit*))
# 4 TS=((odd* or eccentric* or dramatic* or emotional* or anxious* or fearful*) SAME cluster*)
# 3 TS=(ICD AND (F60 or F61 or F62))
# 2 TS=(DSM SAME (axis SAME II))
# 1 TS=(moral SAME insanity)

Sociological Abstracts (CSA)
((DE="personality disorders") or(moral within 2 insanity) or(DSM and (axis and II)) or(icd and (F60 or F61 or F62)) or((odd*
or eccentric* or dramatic* or emotional* or anxious* or fearful*) within 5 cluster*) or("Cluster A" or "Cluster B" or "Cluster
C") or((aggressiv* or anxious* or borderline* or dependent* or emotional* or passiv* or unstable) within 5 personalit*)
or(anankastic* or asocial* or avoidant* or antisocial* or anti‐social* or compulsiv* or dissocial* or histrionic* or narciss*
or obsessiv* or paranoi* or psychopath* or sadist* or schizoid* or schizotyp* or sociopath*) or(personalit* within 5
disorder*) or(character disorder*) or(anal* within 1 (personalit* or character* or retentiv*))) and((TI=randomi* or
DE=(randomi?ed controlled trial*) or AB=randomi*) or(TI=(double* blind*) or AB=(double* blind*) or DE=(double blind
studies)) or(single near blind*))
ZETOC (Conference search)
23 conference: hysteria trial*
22 conference: histrionic personality trial*
21 conference: dependent personality trial*
20 conference: borderline personality trial*
19 conference: antisocial personality trial*
18 conference: compulsive personality trial*
17 conference: compulsive personality random*
16 conference: schizotypal trial*
15 conference: schizotypal random*
14 conference: schizoid trial*
13 conference: schizoid random*
12 conference: passive aggressive trial*
11 conference: passive aggressive random*
10 conference: personality disorder* trial*
9 conference: paranoid personality trial*
8 conference: paranoid personality random*
7 conference: paranoid personality
6 conference: hysteria random*
5 conference: histrionic personality random*
4 conference: dependent personality random*
3 conference: borderline personality random*
2 conference: antisocial personality random*
1 conference: personality disorder* random*

Appendix 2. Trial register searches

We searched the WHO International Clinical Trails Registry Platform on 9 August 2011 (http://apps.who.int/trialsearch/), a meta‐register that includes the following trial registration databases:

• Australian New Zealand Clinical Trials Registry, last data file imported on 9 August 2011

• ClinicalTrials.gov, last data file imported on 9 August 2011

• ISRCTN, last data file imported on 9 August 2011

• Brazilian Clinical Trials Registry (ReBec), last data file imported on 26 July 2011

• Chinese Clinical Trial Registry, last data file imported on 26 July 2011

• Clinical Trials Registry ‐ India, last data file imported on 26 July 2011

• Clinical Research Information Service ‐ Republic of Korea, last data file imported on 26 July 2011

• Cuban Public Registry of Clinical Trials, last data file imported on 26 July 2011

• German Clinical Trials Register, last data file imported on 26 July 2011

• Iranian Registry of Clinical Trials, last data file imported on 2 August 2011

• Japan Primary Registries Network, last data file imported on 26 July 2011

• Pan African Clinical Trial Registry, last data file imported on 26 July 2011

• Sri Lanka Clinical Trials Registry, last data file imported on 26 July 2011

• The Netherlands National Trial Register, last data file imported on 26 July 2011

The following phrase was searched:

condition = borderline personality disorder AND recruitment status = all

Appendix 3. Additional methods for future updates

Notes

Shared last authorship

Notes

Stable (no update expected for reasons given in 'What's new')

Data and analyses

Comparison 1

Comprehensive psychotherapies: active vs. control conditions