Effectiveness of Predominantly Group Schema Therapy and Combined Individual and Group Schema Therapy for Borderline Personality Disorder

This randomized clinical trial examines whether schema therapy is more effective than optimal treatment as usual for borderline personality disorder and whether it is more effectively delivered in a predominantly group or combined individual and group format.


M
eta-analytic reviews have found specialized therapies for borderline personality disorder (BPD) to be associated with greater reductions in overall BPD severity than treatment as usual (TAU). 1,2One specialized therapy is schema therapy (ST).Early trials of ST demonstrated effectiveness when delivered in an individual format. 3,4A large effect was found for group ST delivered as an adjunct to TAU. 5 However, this group ST was delivered by its developers, and therefore, the generalizability of this finding to other therapists and other settings was unclear.Also unknown was the extent to which group ST would be effective as a stand-alone treatment and the relative merits of combining individual and group ST.
An international workgroup, in considering these issues, speculated that the group-only format might encourage patients to bring all issues to the group, thus optimizing group therapeutic processes.Conversely, an argument was made that combining individual with group ST would better meet patients' needs for attention and attachment and better facilitate the processing of severe childhood adverse experiences, a hallmark of ST. 6,7 The workgroup decided to test these competing positions through an international randomized clinical trial (RCT) with 3 arms: predominantly group ST (PGST), combined individual and group ST (IGST), and TAU. 8 We decided to use optimal TAU as a comparator-that is, the optimal treatment that was available at the treatment site according to the usual local practice (excluding ST).
This RCT examined the effectiveness of ST in 2 formats compared with optimal TAU and also compared the effectiveness of the 2 ST formats.We investigated effectiveness for the primary outcome measure (BPD severity).Secondary outcomes included treatment retention, specific BPD and general psychiatric symptoms, quality of life, schemas and schema modes, and psychosocial functioning.We report the results of the planned analyses 8 : first, both group ST formats were combined and compared with TAU, and then the 3 arms were compared with each other.The first hypothesis was that the 2 ST formats combined would be superior to TAU in reducing BPD severity.The second hypothesis was that the 3 treatment types would differ in effectiveness.Moreover, we explored differences in treatment retention and other secondary outcomes.

Study Design
This international multicenter RCT had 3 arms: optimal TAU, PGST, and IGST.Participants were randomized 1:1 in cohorts of 16 to 18 participants per site either to PGST or TAU or to IGST or TAU (NTR2392) (trial protocol in Supplement 1).The order of these cohort types was randomly assigned across sites.There were 15 sites: 2 in Australia, 3 in Germany, 1 in Greece, 7 in the Netherlands, and 2 in the UK (details are provided in eAppendix 1 in Supplement 2).Because some sites could not finance participation or recruit a second cohort, other sites compensated by including extra cohorts, deviating from the plan 8 (eAppendix 2 in Supplement 2).The number of cohorts per site varied from 1 to 5. Thirty cohorts were recruited.The mean rank order of cohort type (PGST vs IGST) was equal (eAppendix 1 in Supplement 2).The planned minimum number of participants was 448 from 28 cohorts based on a power analysis (power 90%; medium effect size, α = 0.05) and taking attrition into account. 8Ethical approval was obtained at each participating site.Participants provided written informed consent.This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline.

Patients
Participants were recruited between June 29, 2010, and May 18, 2016.Figure 1 shows the CONSORT flow diagram; 495 patients were randomly assigned, 1 withdrew consent, and 494 were analyzed.Inclusion criteria were a primary diagnosis of BPD, age of 18 to 65 years, a Borderline Personality Disorder Severity Index IV (BPDSI-IV) score greater than 20 (range, 0-90; a score >20 denotes clear BPD; 15 is the cutoff for BPD, and <15 is used as recovery criterion), and the willingness and ability to participate in 2 years of treatment.Exclusion criteria were an inability to speak, read, and/or understand the study site's language, an IQ less than 80, a psychotic disorder (except reactive episodes; BPD criterion 9 per the DSM-5), bipolar disorder 1, dissociative identity disorder, untreated attentiondeficit/hyperactivity disorder, addiction needing clinical detoxification (inclusion after detoxification was allowed), full or subthreshold narcissistic or antisocial PD, a serious and/or unstable medical illness, and having received ST for more than 3 months during the previous 3 years.Ethnicity was selfreported.Diagnoses were made with the Structured Clinical Interview for DSM-IV Axis I and II Disorders.Sites were eligible if they offered treatment for BPD, had therapists willing to be trained in group ST, agreed to adhere to the study protocol, and could organize and finance the study.

Randomization and Masking
When a cohort of 16 to 18 participants was accrued, participants of that cohort were randomly assigned by a central independent research assistant using computerized randomization in blocks of 2 based on the order of completion of baseline assessments.The type of ST (PGST or IGST) per cohort was randomized across sites. 8Assessors were blinded to the condition.The study blind was removed after the last data became

Key Points
Question Is group schema therapy for borderline personality disorder (BPD) more effective than optimal treatment as usual, and is predominantly group schema therapy or combined individual and group schema therapy more effective?Findings In this randomized clinical trial, which included 495 adult participants with BPD in 5 countries, combined individual and group schema therapy was significantly more effective than optimal treatment as usual and predominantly group schema therapy in reducing BPD severity.

Meaning
The findings add to the evidence for the effectiveness of schema therapy for BPD and indicate that the combination of individual and group schema therapy is the more effective schema therapy format.
available.Given the unequal sample sizes in the 3 arms, blinding during analysis was not feasible.

Procedures
Treatments and Therapists Group ST followed a protocol. 9,10In PGST, 2 group sessions per week were provided in year 1, with a maximum of 12 individual sessions available on patient request.In year 2, group ST was provided once per week in months 13 to 18, once per 2 weeks in months 19 to 21, and once monthly in months 22 to 24, with a maximum of 5 individual ST sessions available on request.
For IGST, 2 sessions (1 individual 7,10 and 1 group) per week were delivered in year 1.In the first 6 months of year 2 (months 13 to 18), frequencies of both individual and group ST were biweekly.In months 19 to 21, group ST was provided biweekly and individual ST was provided monthly.In months 22 to 24, both were offered monthly (details are shown in eTable 1 in Supplement 2).
Group ST was closed and provided by 2 therapists.For IGST, most patients had an individual therapist other than their group therapists.All ST therapists were licensed mental health professionals trained in individual ST for BPD.Group ST therapists were additionally trained by I.S. and J.F. 9 Initially, therapists in both ST formats received online supervision by I.S. weekly; after 6 months, supervision was reduced to biweekly or monthly depending on the therapists' experience.
Treatment as usual was the optimal psychological treatment available at the site.The most frequently offered was dialectical behavior therapy, with an intensity matched to that of ST.An overview of the types of TAU offered is shown in eTable 2 in Supplement 2.

Treatment Integrity and Assessments
A scale was designed to assess whether ST discriminated from TAU (eAppendix 3 in Supplement 2).Independent evaluators

Outcomes
The primary outcome was the slope over time of BPD severity during the past 3 months as assessed at baseline, 6 months, 12 months, 18 months, 24 months, and 36 months with the BPDSI-IV total score, a reliable and valid index of BPD severity. 12econdary outcomes included 9 BPDSI subscales (representing the BPD criteria); suicidality and the number of suicide attempts during the past 3 months, assessed with the BPDSI-IV; the BPD Checklist

Statistical Analysis
Data were analyzed from June 4, 2019, to December 29, 2021.Given skewed distributions, the BPDSI total and subscale scores were analyzed with generalized linear mixed model (GLMM) gamma regression with a log link, with random effects of treatment for site and of treatment and time for the cohort nested under the site if estimation allowed (eAppendix 5 in Supplement 2).For the repeated part, an autoregressive moving average (1,1) model was used.The fixed part contained time, treatment, and their interaction, with the time × treatment interaction (the difference in slopes) testing the differential effectiveness hypothesis.Because gamma regression cannot be conducted with 0, a small value (0.01) was added to scores.Sensitivity analyses were performed to test the robustness of this choice.Effect sizes were expressed as r = ͙(t 2 /[t 2 + df]) for fixed effects, and Cohen d was the conventional effect size, with the numerator based on estimated means from GLMM gamma regression and the denominator as baseline SD in transformed scale (based on GLMM gamma regression with a log link, without random parts, and with an intercept for the fixed part) (eAppendix 6 in Supplement 2).Treatment retention was analyzed using GLMM survival analysis of treatment dropout per quarter (every 3 months) during 2 years with the site as a random intercept.The number of suicide attempts during the previous 3 months was analyzed using generalized estimating equations with a Tweedie distribution (because of many 0 counts) with a log link and firstorder autoregression for the repeated part.Depending on the distribution, secondary outcomes not based on the BPDSI were analyzed with appropriate LMM or GLMM.The time model was either linear or piecewise (if an inflection point was observed in all at 1 year).The choice of a piecewise model was supported by superior fit, tested with χ 2 tests.For all piecewise models, the time × treatment interaction was not significant and did not lead to improved fit; thus, it was deleted, implicating that treatment effects were assessed by the difference in slopes in years 2 and 3.In accordance with the preset plan, 8 first, both ST formats were jointly compared with TAU; next, the 3 arms were compared.Binomial tests (with a probability of 0.05) were used to assess the likelihood of the number of significant tests of secondary outcomes per comparison. 21nalyses were conducted in SPSS, version 25 (IBM), as intention-to-treat after the last data became available.Significance was set at 2-tailed P = .05.Different sample sizes made blind analysis impossible.

Results
A

Assessment of Treatment Integrity
The Treatment Integrity Scale discriminated ST from TAU (Table 1).The quality of individual and group ST was rated between good and very good by independent raters 11 (eAppendix 3 in Supplement 2).

Primary Outcomes
Forty-two recordings of BPDSI interviews were randomly selected; these were rated by an independent assessor, and the BPDSI total scores were then compared with those reported by the original interviewer.Interrater agreement was excellent (intraclass correlation coefficient, 0.99).
Results of a loc-f sensitivity analysis are given in eTable 3 in Supplement 2. Results of sensitivity analyses for different offset values are shown in eTable 4 in Supplement 2. Descriptive statistics of raw BPDSI scores and results in a transformed scale are shown in eTable 5 in Supplement 2.

Severity per BPD Trait
Predominantly group ST and IGST combined were superior to TAU on 6 of the 9 BPDSI subscales (binomial test, P < .001)(eTable 6 in Supplement 2); IGST was superior to TAU on 7 subscales (binomial test, P < .001)and to PGST on 2 subscales (binomial test, P = .07).PGST was not significantly different from TAU on any of the subscales (binomial test, P > .99).The effects of PGST were between those of TAU and IGST for all subscales apart from emptiness, for which the effect of PGST was smaller than that of TAU.

Suicidality and Suicide Attempts
For scores on suicidality items of the BPDSI (para)suicide scale, PGST and IGST combined were superior to TAU (eTable 6 in Supplement 2).Only IGST was superior to TAU; IGST and PGST did not differ significantly in reducing suicidality scores.
Overall, the number of suicide attempts during the previous 3 months reduced with time, with a nonsignificant difference between ST and TAU but with IGST being significantly superior to TAU in reducing suicide attempts (eTable 6 in Supplement 2).

Secondary Outcomes Not Based on BPDSI
For all 11 secondary outcomes, PGST and IGST combined were superior to TAU (binomial test, P < .001);IGST was superior    to TAU for 9 of the 11 outcomes (binomial test, P < .001)(Table 2 and Table 3).PGST was superior to TAU for only 2 outcomes (the Work and Social Adjustment Scale and happiness; binomial test, P = .10).IGST was superior to PGST for 1 of 11 secondary outcomes (binomial test, P = .43),the World Health Organization Quality of Life Assessment.The overall ST effect was largely attributable to IGST, with PGST outcomes between those of TAU and IGST.
During the study period, 3 patients died of suicide (1 from each treatment arm).The patient from the PGST arm applied successfully to a euthanasia clinic during the 3 months (quarter 1) of treatment.The patient from the IGST arm died by suicide approximately 8 months after discontinuing treatment and study during quarter 2 (months 4 to 6).A TAU recipient died by suicide during quarter 3 (months 7 to 9).In the TAU arm, an additional patient died in quarter 2 of an unknown cause; suicide was suspected but not confirmed.

Discussion
This international RCT compared 2 ST formats (PGST and IGST) with TAU and with each other in the treatment of BPD across 15 sites in 5 countries.Although the PGST and IGST combined appeared to be superior to TAU in reducing BPD severity (the primary outcome) and for most secondary outcomes, most effects were attributable to IGST; there were also fewer suicide attempts and treatment dropouts in this arm.IGST was superior to TAU in reducing BPD severity and for 7 BPDSI subscales.Moreover, IGST was superior to PGST in reducing BPD severity and for 2 BPDSI subscales.In summary, the results support IGST as an effective, acceptable (as indexed by treatment retention), and safe treatment for BPD, whereas the results were more equivocal for PGST.
The 3-year effects were generally large in all arms.Treatment as usual was effective given the within-condition effect sizes.Compared with the pooled pretreatment-posttreat-Table 3. Comparison of Time Effects for First and Secondary Outcomes Between the ST Groups Combined vs the TAU Group, the PGST Group vs the TAU Group, the IGST Group vs the TAU Group, and the IGST Group vs the PGST Group a (continued)  d Analyzed with GLMM gamma regression with a log link.Estimated means are −46.99 points lower than on the original scale.Scores were transformed by subtracting 46.99 to bring the minimum to just greater than 0 to enable gamma regression.e Piecewise gamma regression with a general slope for time and an additional slope for the second and third year was used.Deleting the general time × treatment interaction from the model increased model fit; thus, the final model was time (general), time (years 2 and 3), treatment, and treatment × time (years 2 and 3).Interactions for the second time slope are shown (second and third year).To enable GLMM gamma regression with a log link, 0.1 was added to raw scores.f Piecewise regression with a general slope for time and an additional slope for the second and third year was used.Deleting the nonsignificant general time × treatment interaction from the model did not reduce model fit; thus, the final model was treatment, time (general), time (years 2 and 3), and time (years 2 and 3) × treatment.Interactions for the second time slope are shown (second and third year).g Analyzed with GLMM logistic regression.Estimated means are the proportion of working to studying.ment effect size of 0.5 for dialectical behavior therapy, 22 usually found at 1 year, the TAU effect sizes at 1 year in our study were similar (0.80 for the primary outcome and a mean effect size of 0.56 for secondary non-BPDSI outcomes).This indicates that optimal TAU was offered.
With regard to the reason that IGST was superior to PGST, IGST had relatively large effects on schemas, schema modes, and affective instability, variables that are assumed to underlie change processes. 23,24Core emotional needs such as safe attachment and positive attention are often not adequately met during childhood in patients with BPD in both individual and group relationships.The combined ST format to meet needs in both contexts, whereas PGST provided less individual attention.Moreover, addressing severe problems and childhood trauma might be easier for therapists in individual treatment than in group treatment.This was reported by participating therapists in a qualitative ancillary study. 25Consistent with these hypotheses, patients in another ancillary qualitative study 26 stated that individual ST was important to discuss sensitive topics on a deeper, more personal level.Most patients (75%) emphasized the importance of individual ST in conjunction with group ST, and 70% of the participants in receiving PGST expressed the need for more individual sessions.Also indicative of the importance of individual ST, PGST participants in the current study generally used the maximum number of individual sessions.How well individual ST compares with IGST and which format is optimal for whom are topics for future studies.
When comparing, to our knowledge, the first RCT on individual ST for BPD 3 with the present study, there appeared to be equivalent effect sizes.There were more sessions in the former study (approximately double), but the dropout rate was lower.Group treatment might be associated with increased dropout for a variety of reasons, including distrust, frightening group processes, or lack of an individualized agenda.Some patients in the ancillary qualitative study 26 described hostility from group members as particularly difficult.Individual sessions might mitigate the problem of aversive group dynamics.
The present study has some implications about the sufficient dosage of ST sessions.A finding of note was the continu-ation of symptom improvement during years 2 and 3.In the ST arms, session frequency was reduced to only once a month using a tapered schedule, and in year 3, no further treatment was offered.Although patients often resist this tapering, feeling not ready and expressing fear of relapse, 26 in our study, they tended to do well.At the 3-year follow-up, some participants expressed how helpful it was to discover that they were able to apply what they learned after therapy was discontinued.They commented that this helped to increase their self-confidence and changed their self-view as a patient with chronic mental illness who is dependent on long-term mental health care.
Despite the generally positive results, there were 3 suicides and 1 death of unknown cause during the study, with 1 suicide in each treatment arm.One of the deaths was related to a preset plan to apply for euthanasia that the person did not share with the treating team.Another participant died by suicide after dropping out of treatment.Nevertheless, in general, suicidality was reduced during treatment, as were suicide attempts, with IGST being superior to TAU in reducing suicidality and suicide attempts.

Limitations
This study had several limitations.First, the worldwide financial crisis of 2007-2008, followed by an economic crisis (in most countries), interfered with study implementation, leading some sites that had planned to participate to withdraw owing to cutbacks.This had to be compensated by recruiting additional sites and by sites including additional cohorts.Although the mean rank order of the ST formats was equal, the distribution deviated from the original plan.Second, as with all RCTs, participants dropped out of the study, which can affect the validity of the results.Third, rating of TAU recordings to assess differentiation between ST and TAU was impossible.However, patients' self-reports supported that ST was different from TAU.Moreover, assessments of adherence to ST by independent raters supported that ST was generally delivered at a good level of quality.Fourth, group ST was generally delivered by novice group ST therapists who were specifically trained for the study without having had the opportunity to learn the intervention before the study commenced.Some aspects of the group in-

Conclusions
In this RCT, IGST was more effective in reducing the severity of BPD compared with TAU and PGST.However, PGST was not more effective than TAU.The international multicenter design, the size of the study population, and execution of the study in regular mental health services that were treating BPD support the generalizability of the results.However, this study did not compare ST with another specialized psychotherapy, and no conclusions in this respect can be made.The cost-effectiveness of ST is a subject for further study.Future research should focus on the direct comparison of individual and combined ST formats; the testing of less extreme combinations of group and individual ST; direct comparisons of ST with other evidence-based treatments, including the study of factors associated with better treatment allocation; and the testing of ST in countries in other regions of the world.

b
Cohen d values are based on the parameters of the generalized linear mixed model analyses (change over time), with the square root of the baseline variance of a model with no random parts and only a fixed intercept as the denominator (ie, SD baseline in transformed scale).
Abbreviations: GLMM, generalized linear mixed model; IGST, combination of individual and group schema therapy; PGST, predominantly group schema therapy; ST, schema therapy arms combined; TAU, (optimal) treatment as usual.a All analyses are based on GLMM with an autoregressive moving average (1,1) covariance structure for the repeated part and a random effect of time for a cohort within a site.Effect sizes d are based on the parameters of the GLMM analyses (change over time), with the square root of the baseline variance of a model with no random parts and only a fixed intercept as the denominator.Effect sizes r are defined as r = ͙(t 2 /[t 2 + df]).These represent the effect size associated with the effect tests in the fixed part of the GLMM.The t values and Cohen d values are positive when there is a positive effect (ie, a reduction over time, a larger reduction in ST than in TAU).The main time effect is the mean across ST and TAU.b Cohen d values are based on the parameters of the GLMM analyses (change over time), with the square root of the baseline variance of a model with no random parts and only a fixed intercept as the denominator (ie, SD baseline in transformed scale).The 95% CIs of the Cohen d are based on the 95% CIs of the slope differences from the GLMM.c The interaction of treatment by time was significant.

Figure 2 .
Figure 2. Borderline Personality Disorder Severity Index (BPDSI) Scores and Duration of Treatment Retention

Table 1 .
Descriptive Statistics at Baseline and Treatment Integrity Test Results b Based on the International Standard Classification of Education, 2011 version, with a range of 0 to 8. c Two patients had missing responses (1 from the TAU group and 1 from the IGST group).dIncluded BPD.e Full antisocial PD was an exclusion criterion, as was narcissistic PD. f

Table 2 .
Estimated Means, 95% CIs, and Effect Sizes for the Primary and Secondary Outcomes

Table 2 .
Estimated Means, 95% CIs, and Effect Sizes for the Primary and Secondary Outcomes (continued) Abbreviations: BPD, borderline personality disorder; BPDSI, Borderline Personality Disorder Severity Index; BSI, Brief Symptom Inventory; GAF, Global Assessment of Functioning; IGST, combination of individual and group schema therapy; NA, not applicable; PGST, predominantly group schema therapy; SMI, Schema Mode Inventory; SOFAS, Social and Occupational Functioning Assessment Scale; TAU, (optimal) treatment as usual; WSAS, Work and Social Adjustment Scale; WHO QOL, World Health Organization Quality of Life assessment; YSQ, Young Schema Questionnaire-short form.
a Estimated means and 95% CIs are in original scale.

Table 3 .
Comparison of Time Effects for First and Secondary Outcomes Between the ST Groups Combined vs the TAU Group, the PGST Group vs the TAU Group, the IGST Group vs the TAU Group, and the IGST Group vs the PGST Group a

Table 3 .
Comparison of Time Effects for First and Secondary Outcomes Between the ST Groups Combined vs the TAU Group, the PGST Group vs the TAU Group, the IGST Group vs the TAU Group, and the IGST Group vs the PGST Group a (continued) jamapsychiatry.com(Reprinted) JAMA Psychiatry April 2022 Volume 79, Number 4 295 Downloaded From: https://jamanetwork.com/ on 09/29/2023 Abbreviations: GLMM, generalized linear mixed model; IGST, combination of individual and group schema therapy; PGST, predominantly group schema therapy; ST, schema therapy arms combined; TAU, (optimal) treatment as usual.a All analyses are based on GLMM with an autoregressive moving average (1,1) covariance structure for the repeated part and a random effect of time for a cohort within a site.Effect sizes d are based on the parameters of the GLMM analyses (change over time), with the square root of the baseline variance of a model with no random parts and only a fixed intercept as the denominator.Effect sizes r are defined as r = ͙(t 2 /[t 2 + df]).These represent the effect size associated with the effect tests in the fixed part of the GLMM.The t values and Cohen d values are positive when there is a positive effect (ie, a reduction over time, a larger reduction in ST than in TAU).
(eg, early conflict management) required a high skill level.For some of the individual ST therapists, it was the first time they applied ST.Thus, ST effects might increase and the number of dropouts may be reduced with experience, repeated training, and supervision.Fifth, TAU was unstandardized.Future RCTs should compare ST with other specialized psychotherapies, controlling for intensity of treatment.