Non‐Randomized Clinical Trial of a Short‐Term Therapy Group for Adults Diagnosed With Borderline Personality Disorder: An Intermediate Step in Care

ABSTRACT

Objective

The aim of this study was to evaluate the effectiveness of Road Maps, a short‐term therapy group program (i.e., 12 weeks) as part of an intermediate step in care for adults with a diagnosis of borderline personality disorder (BPD).

Method

A pragmatic non‐randomized clinical trial study design was utilized to examine the effectiveness of Road Maps in a publicly funded community mental health care setting. First, we examined whether participation in the short‐term group intervention (n = 80) resulted in significantly greater reductions in psychopathology and improved psychosocial functioning compared to waiting for the intervention (n = 41). The waitlist condition was naturalistic (> 28 days) in that there was no specific allocation to conditions and all people on the waitlist had the opportunity to participate in the next available group. Second, we examined whether therapeutic gains were maintained at 6‐month follow‐up across the entire sample (N = 121). Finally, we analysed the proportion of respondents who demonstrated clinically significant change.

Results

Relative to people on the waitlist for the intervention, those who participated in the group demonstrated a significantly greater reduction in the primary outcomes of borderline symptom severity and personality dysfunction (e.g., both self and interpersonal functioning) and secondary outcomes (e.g., emotion regulation, reflective functioning, and psychosocial functioning). Therapeutic gains were maintained at 6‐month follow‐up.

Conclusion

Short‐term therapy groups such as Road Maps may be a useful intermediate step in a broader model of stepped care aimed at increasing treatment accessibility in resource‐constrained environments.

Trial Registration

Australian New Zealand Clinical Trials Registry ACTRN12622000849796/retrospectively registered 16th June 2022.

Borderline personality disorder (BPD) is a serious mental health condition characterized by core difficulties in affect, identity, and relationships (American Psychiatric Association ²⁰¹³). Associated with psychosocial functioning difficulties (Álvarez‐Tomás et al. ²⁰¹⁹), non‐suicidal self‐injury (Zanarini et al. 2008) and suicide (Broadbear et al. 2020), BPD can have a devastating impact on individuals and their families. Meta‐analytic studies demonstrate the effectiveness of psychological therapies in the treatment of BPD (Cristea et al. 2017; Stoffers‐Winterling et al. ²⁰²²). However, evidence‐based therapies are often intensive (involving weekly group and individual therapy), and the recommended length of treatment can range between 12 and 24 months.

Access to evidence‐based treatment has been highlighted as a significant problem by service users and the scientific community (Iliakis et al. 2019; Lawn and McMahon ²⁰¹⁵). The broad implementation of specialized evidence‐based therapies for BPD is limited by treatment duration, cost, and access to intensive training for therapists (Iliakis et al. 2019). This can result in delayed treatment and people, allocated to lengthy waitlists, left untreated, increasing the likelihood of adverse events. In recognition that not all people with a diagnosis of personality disorder require long‐term care (Laporte et al. 2018) and evidence that people can benefit from easily accessible psychoeducational programs (Gunderson et al. 2020; Zanarini and Frankenburg ²⁰⁰⁸), recommendations have shifted towards stepped care models in which a variety of treatment options of varying intensities are available and patients are matched to treatment based on their presenting needs (Choi‐Kain et al. ²⁰¹⁶; Grenyer ²⁰¹⁴).

Historically, there has been a reluctance to offer short‐term therapy to people with a diagnosis of BPD, due to suggestions this may exacerbate difficulties with mistrust in relationships and abandonment fears (National Institute for Health and Care Excellence ²⁰⁰⁹). Recently, Spong et al. (2021) conducted a meta‐analysis of short treatments for BPD. They found that across 27 studies, the average length of the interventions was 14 weeks (range: 2–24 weeks), with an intensity of 27 h (range 7.5–60 h). The review indicated that group‐only interventions may be effective in reducing self‐harm, and that short interventions may be a useful adjunct to ongoing support. Additionally, it was suggested that generalist support may be as effective as specialist interventions in reducing borderline symptoms and improving social functioning. There were limitations to the findings as discrepancies in the measurement of variables restricted the pooling of data for some outcomes. Additionally, there weren't sufficient data to estimate if outcomes were maintained at follow‐up. However, this study challenged recommendations to avoid short‐term psychological interventions and highlighted the need for further service development and research in this field.

Many of the 3‐month interventions reported by Spong et al. (2021) were derived from specialist interventions which have significant training requirements (Morton et al. 2012; Soler et al. ²⁰⁰⁹) and pose a potential barrier for broader implementation (Iliakis et al. 2019). It has been argued that less resource‐intensive treatments could be developed through integrating common factors from effective BPD therapies (Bateman et al. 2015; Beatson and Rao ²⁰¹⁴; Choi‐Kain et al. ²⁰¹⁶; Weinberg et al. ²⁰¹¹). Additionally, generalist approaches delivered in individual‐therapy formats have been described and recommended (Kramer et al. 2014; Laporte et al. ²⁰¹⁸). Driven by a local need to increase access to treatment we drew on common factor principles to inform the development of a 12‐week therapy group intervention for people with a diagnosis of BPD.

The intervention, called Road Maps, was developed through the iterative integration of participant, peer educator, and facilitator feedback. Road Maps is a generalist intervention that could be reasonably delivered by a qualified mental health professional. It was designed around principles of functional analysis to support people with BPD to increase their understanding of their experiences, patterns, and symptoms and to develop skills to tolerate distress and foster agency.

1. Objectives

The intervention was developed and trialled in a statewide mental health service which was established in 2019 to increase access to treatment for people with a diagnosis of BPD. A pragmatic approach to evaluation was undertaken to balance principles of scientific rigor with the complexities of service delivery in real‐world clinical settings. First, we wanted to understand if the short‐term intervention was superior to “waiting” for an intervention (i.e., a minimum of 28 days) in acknowledgment that offering the short‐term intervention aimed to increase treatment access. It is common within tertiary mental health services for people to be referred to therapy, but experience wait times before the next intake (i.e., a closed group with intake at the beginning of each term); also patients may postpone their start date due to life circumstances. To approximate these conditions a naturalistic waitlist condition was derived, where people who had waited more than 28 days between initial assessment and intervention were asked to recomplete the assessment battery before intervention commencement. This enabled us to test whether participation in the short‐term group had a greater impact on psychological and psychosocial functioning relative to the “waiting” condition in which we did not anticipate a change in symptoms.

Second, we aimed to examine whether the gains from the short‐term intervention were maintained at 6‐month follow‐up. Due to the clinical setting, it was not ethical to delay intervention for those in the waitlist condition to ensure data timepoints could be directly compared to the intervention group. They were offered the next available group. Therefore, a separate analysis was conducted to examine whether outcomes were maintained, including all participants whilst accounting for the condition (waitlist vs intervention) statistically. Finally, we will explore the clinical significance of change on the primary outcome of borderline symptoms severity, as well as the probability of specific behavior change. More specifically we hypothesized:

• (1)Participants in the intervention condition will report significantly greater reduction in the primary outcomes of borderline symptom severity, self‐destructive/impulsive behaviour, self‐ and interpersonal dysfunction, compared to participants waiting for treatment.
• (2)As above, we also anticipate a similar pattern of results relating to improvement in secondary outcomes (i.e., emotional regulation, reflective functioning, psychosocial functioning).
• (3)When examining change in primary and secondary outcomes across the entire sample, we hypothesize that posttreatment therapeutic gains will be maintained at 6‐month follow‐up.

2. Methods

2.1. Participants

The study was approved by the Southern Adelaide Clinical Human Research Ethics Committee (reference number HREC/20/SAC/92). The trial was retrospectively registered on the Australian New Zealand Clinical Trials Registry: ACTRN12622000849796. We utilized the TREND checklist for non‐randomized clinical trials to guide the reporting of this project (Des Jarlais et al. 2004).

The group was developed and delivered by clinicians at Borderline Personality Disorder Collaborative (BPD Co), a state‐wide mental health service in South Australia. Participants were recruited to the evaluation of the group between June 2020 and June 2023. The main source of referral was self (83%), followed by other mental health services/clinicians (14%), or in a few instances a family member (3%). Inclusion criteria were defined as a person aged over 18 years of age, a resident of South Australia and met diagnostic criteria for BPD. Exclusion criteria included a high level of risk requiring urgent assessment and follow‐up, acute psychosis, active substance misuse that would impact participation, antisocial behaviour, and/or impaired cognitive capacity likely to impact ability to understand materials (due to developmental disability or medical condition).

People referred to the group attended an initial assessment with a senior mental health clinician. Before assessment participants were provided with a questionnaire that included an explanation of the research study and participant information sheet. Consent to research was optional and non‐consent did not impact participants' access to the group program. Participants completed a battery of self‐report assessments including a 10‐item screening instrument for BPD symptoms (MSI‐BPD; Zanarini et al. ²⁰⁰³), with a cut‐off of 7 or more items recommended for a possible diagnosis of BPD among adults. The clinician reviewed responses to the screener in a semi‐structured interview and assessed the level of risk, diagnostic criteria, and capacity to effectively participate within a group therapy context. Following this interview, the assessment, clinicians' recommendations, suitability, and eligibility for group therapy were considered and endorsed via a multi‐disciplinary clinical review attended by the team Consultant Psychiatrist.

2.2. Intervention

The short‐term therapy group, Road Maps, was offered weekly over 12 weeks, with 2.5‐h sessions (including a 15‐min break) equating to approximately 30 h intervention. Road Maps content was developed in alignment with the common factors of evidence‐based therapies for treatment of BPD (Bateman et al. 2015; Weinberg et al. ²⁰¹¹). With a clearly defined structure, therapeutic framework, and active facilitators who promoted participants' agency, the group aimed to address symptom severity, self‐destructive behaviours, emotion dysregulation, personality functioning, and self‐reflective capacity by incorporating exploratory and change orientated interventions. Initial sessions provided psychoeducation on the biopsychosocial model of BPD, neurobiology, and physiological states and strategies to manage distress. Participants were then introduced to functional and solution analysis to target behavioral change. The remaining sessions were focused on the steps of functional analysis and teaching strategies to manage each step (e.g., vulnerabilities, crisis behaviours, emotions, thoughts, problem‐ and emotion‐focused coping).

Another key feature of the intervention included a carer information session which aimed to provide support people with psychoeducation about BPD and the intervention, as well as linkage to carer resources. The intervention utilized multimedia resources (e.g., PowerPoint presentations to deliver content, videos to aid psychoeducation, and audio to support experiential exercises) and a fictional character to demonstrate teaching points across sessions. Additionally, the road map metaphor was central to providing a consistent narrative to increase understanding and connection to fundamental concepts and adaptability to a wider audience. Additional details about the feasibility and safety of Road Maps, as well as a table outlining session by session content, have been published elsewhere (Bartsch, Cooke‐O'Connor, et al. ²⁰²⁴).

Road Maps content delivery was manualized, and handouts and worksheets were provided to participants. In total eight experienced mental health clinicians from different disciplinary backgrounds (i.e., 2 x social workers, 1 x nurse, 2 x occupational therapists, and 3 x psychologists) delivered the intervention across the study period. They had pre‐existing training in various therapeutic approaches such as dialectical behaviour therapy (DBT), schema‐focussed therapy (SFT) and mentalization‐based therapy (MBT). As a generalist intervention, specific specialist clinician training was not a requirement for facilitation. Weekly clinician debriefs and peer supervision occurred to support adherence to the Road Maps manual and reflect on group dynamics. Groups were closed, facilitated by two clinicians and size ranged between 7 and 13 participants with some online delivery during the Covid pandemic as required.

There were no fees charged to access the group, nor were any financial incentives offered for participation in the research. The group was offered within the context of a broader public‐health model of stepped care for people with BPD situated between a brief crisis intervention (Bartsch, McLeod Everitt, et al. ²⁰²⁴) and specialized therapies for BPD in their local community mental health teams or private health settings as available (e.g., DBT, SFT, MBT). Road Maps was provided in addition to treatment as usual for all participants (i.e., there were no restrictions on what other care people may have been receiving in the community nor did we control for the effects of any medication participants may have been taking during their participation in the Road Maps group).

2.3. Outcomes

Outcomes were tracked utilizing a self‐report battery of assessments. Participants who consented to research and commenced group were contacted by a team researcher via phone 6 months after the intervention and invited to participate in a scripted follow‐up phone interview. To reduce social desirability bias, the researchers who conducted the interviews were not involved in the group intervention delivery. The 6‐month follow‐up interview repeated the assessment battery, and participants were asked additional questions about their service use in the preceding 6 months. Reliability estimates for all measures in the current study are listed in Table 1.

Table 1.

Demographic and clinical characteristics of participants at baseline among the sample split by condition.

	Total n = 121		Intervention n = 80		Waitlist n = 41
	n M	% SD	n M	% SD	n M	% SD	p	αa
Gender
Woman	107	(88)	71	(89)	36	(88)	—
Man	11	(9)	7	(9)	4	(10)	—
TGNC	3	(3)	2	(3)	1	(2)	—
Age	32.60	13.26	32.77	12.39	32.27	14.97	0.42
ATSI	6	(5)	4	(5)	2	(5)	—
MSI‐BPD	8.92	1.35	8.90	1.43	8.95	1.18	0.80	0.55, 0.60, 0.45
Primary outcomes
BSL‐23	2.50	0.94	2.55	0.91	2.38	0.99	0.40	0.96, 0.96, 0.97
BSL‐supp	0.70	0.60	0.71	0.64	0.67	0.50	0.84	0.75, 0.78, 0.68
LPFS ‐ total	36.88	6.80	37.29	6.64	36.05	7.12	0.40	0.85, 0.85, 0.85
Self	19.94	3.46	20.11	3.39	19.59	3.62	0.40	0.79, 0.78, 0.79
Interpersonal	16.94	4.23	17.18	4.20	16.46	4.32	0.46	0.79, 0.80, 0.76
Secondary outcomes
DERS	65.01	13.79	65.60	14.02	63.83	13.40	0.41	0.91, 0.92, 0.90
RFQc	0.36	(0.49)	0.28	(0.39)	0.53	(0.59)	0.04	0.67, 0.62, 0.69
RFQu	1.71	(0.87)	1.80	(0.83)	1.54	(0.92)	0.19	0.81, 0.80, 0.81
WSAS	29.13	(7.14)	30.05	(6.58)	27.30	(7.90)	0.05	0.69, 0.64, 0.74

Note: Continuous variables were compared using Mann Whitney U test. Missing data on scales MSI‐BPD (n = 1), LPFS (n = 2), WSAS (n = 1), RFQ (n = 30; data missing on RFQ subscale as it was introduced later).

Abbreviations: ATSI, Aboriginal and/or Torres Strait Islander; BSL‐23, Borderline Symptom List – 23; BSL‐supp, Borderline Symptom List – Supplement; DERS, Difficulties in Emotion Regulation Scale; LPFS, Levels of personality functioning; MSI‐BPD, McLean's Screening Instrument – Borderline Personality Disorder; RFQc, Reflective Functioning Questionnaire – Certainty; RFQu, Reflective Functioning Questionnaire ‐ Uncertainty; TGNC, Transgender and gender nonconforming; WSAS, Work and Social Adjustment Scale.

^a First figure is Cronbach alpha for total sample, second figure is Cronbach alpha for the Intervention and Waitlist condition in consecutive order.

3. Primary Outcomes

3.1. Borderline Symptom List (BSL‐23)

The BSL‐23 (Bohus et al. 2009) is a self‐report measure of borderline symptom severity and includes items such as “My mood rapidly cycled in terms of anxiety, anger and depression.” Participants rated 23 items utilizing a Likert scale ranging from 0 (not at all) through to 4 (very strong) and a total mean score was calculated. This measure has adequate reliability and validity and is sensitive to the effects of therapy (Bohus et al. 2009).

3.2. Borderline Symptom List – Supplement (BSL‐Supp)

The BSL‐supp was also included in the current study to measure the frequency and type of self‐destructive and/or impulsive behavior in the previous week. The supplement includes 11 items which describe behaviors such as “I hurt myself by cutting, burning, strangling, head‐banging etc.” and “I took drugs.” Items were measured on a 4 ‐point Likert scale ranging between 0 (not at all) and 4 (daily or more often). These items were reviewed by clinicians in practice to assess risk and track outcomes.

Despite the clinical utility of this measure there are few studies available to provide scoring guidelines and psychometric information. We followed the example of another local study which utilized the total mean score as a continuous measure of self‐destructive/impulsive behavior frequency (pre intervention α = 0.77 and posttreatment α = 0.84; Heerebrand et al. ²⁰²¹). However, we also acknowledge that conceptually, this scoring does not account for potential complexity that arises from an individual endorsing different types of behaviors at different frequencies (e.g., one person could be engaging in multiple self‐destructive behaviors once a week, compared to someone engaging in one form daily). For this reason, we also undertook an additional analysis to examine change across individual items.

3.3. Levels of Personality Functioning Scale – Brief Version 2.0 (LPFS‐BF 2.0)

The LPFS‐BF 2.0 (Weekers et al. 2019) was included as a measure of general personality functioning and included subscales relating to self and interpersonal functioning. It aligns with the alternative model of personality disorders (AMPD) which considers personality dysfunction as a dimensional construct (American Psychiatric Association ²⁰¹³). This measure includes a total of 12 items which are rated on a 4‐point Likert scale ranging from 1 (very untrue of me) to 4 (very true of me). The total score ranges between 12 and 48 (Subscale scores range between 6 and 24) with higher scores representing greater dysfunction. The scale has demonstrated adequate internal consistency, construct validity and sensitivity to change (Weekers et al. 2019).

4. Secondary Outcomes

4.1. Difficulty With Emotion Regulation Scale ‐ 18 (DERS‐18)

The DERS‐18 (Victor and Klonsky 2016) was included to assess participants' capacity to regulate emotions given emotion dysregulation has been proposed as a core feature of BPD (Linehan 1993). Items were rated on a 5‐point scale ranging from 1 (almost never) to 5 (almost always). Three items (1, 4, 6) were reverse scored. Higher total scores reflected greater difficulty regulating emotions. The scale has high reliability and adequate validity (Victor and Klonsky 2016).

4.2. Reflective Functioning Questionnaire – 8 (RFQ‐8)

The RFQ‐8 (Fonagy et al. 2016) was included as a brief self‐report assessment of reflective functioning (i.e., the capacity to interpret both one's own and others' internal mental states). A failure to establish robust mentalizing capacity has been proposed as one of the mechanisms underlying the development of BPD (Fonagy and Bateman 2008). The RFQ‐8 is comprised of two subscales; the RFQu which assesses uncertainty about mental states (hypomentalizing) and the RFQc which measures certainty about mental states (hypermentalizing). Each subscale includes 6‐items, 4 that overlap and 2 unique to the scale. Respondents rate the measure on a 7‐point Likert scale ranging from strongly agree (1) to strongly disagree (7). A nonlinear recoding of scaled scores was used to measure the strength of response at either end of the scale as recommended by Cucchi et al. (2018). The final scores on each subscale were averaged and ranged between 0 and 3. Scores closer to 0 represent more genuine mentalizing (i.e., the ability to acknowledge the opaqueness of mental states) whereas higher scores represent greater difficulties mentalizing.

4.3. Work and Social Adjustment Scale (WSAS)

The WSAS (Mundt et al. 2002) was included as a brief measure of psychosocial functioning. It included five items scored on a 9‐point scale ranging from 0 (not at all impaired) to 8 (severe impairment). A total score is also reported and ranges between 0 and 40. The scale has acceptable internal consistency and is sensitive to clinical change (Mundt et al. 2002).

5. Sample Size

Outcome data (i.e., BSL‐23) from a local evaluation (Heerebrand et al. 2021) of a DBT skills training group was utilized to calculate the sample size needed to power an analysis at 80% with an alpha of 0.05. Forty‐one participants were needed per condition taking into consideration the potential for 50% non‐completion.

6. Assignment Method

Given the group was delivered within a tertiary mental health service, randomization to a waitlist condition was not appropriate (Des Jarlais et al. 2004), thus a naturalistic waitlist condition was adopted. Initial assessments were offered in a rolling manner and participants who consented to research and were assessed 28 days or more before the next available group comprised the “waitlist” condition. Participants in this condition were then asked to complete a second assessment at the end of the waiting period (e.g., before the commencement of their first group session). Participants assessed less than 28 weekdays before the start of the next group were assigned directly into the intervention condition and completed the battery of measures before their first group session and after the final group session. Given the pragmatic nature of our quasi‐experimental study design, in which participants were either waiting or receiving the intervention in a real‐world mental health setting, it was not feasible to mask conditions for clinicians who were coordinating the group referrals, undertaking the initial assessments, and delivering the intervention.

7. Statistical Methods

The data were cleaned and reviewed for missing items, outliers, and normality. Baseline comparisons between the waitlist and treatment group were analyzed using the Mann Whitney U test as most scales were not normally distributed. Linear mixed‐effects regression modelling (LMM) was conducted using the MIXED procedure in SPSS v 29. Mixed‐effects models have several advantages over traditional ANOVAs including (a) the ability to include all available data; (b) the approach is robust to uneven sample sizes; (c) it accounts for repeated measurements within individuals; and (d) handles missing data using maximum likelihood estimation (Field 2018). The majority of patient‐reported outcomes were measured continuously, and separate models were evaluated for each dependent variable. Statistical analyses included all available data with an alpha level p < 0.05.

To test the first hypothesis, we treated time as a continuous variable in which the number of weeks between the first assessment (Baseline/T0 for both groups) and the second assessment (waitlist = start of group; posttreatment = completion of group) accounted for variation in time between assessments given the naturalistic conditions. Fixed effects were specified for time in weeks, condition (waitlist/intervention), and for their interaction (time x condition). Random intercepts for subjects were specified to account for repeated measures. The time x condition interaction effect was the primary variable of interest in each analysis as it demonstrated a differential rate of change in outcome per week in the intervention group relative to the waitlist condition. The model was estimated using Maximum Likelihood (ML) and temporal correlations were modelled using the AR1 covariance structure. Post‐hoc pairwise comparisons of condition at each level of time (Weeks = 0, 4, 8, 12) were performed using estimated marginal means (EMMs) with Least Significant Difference (LSD) adjustment. We estimated effect sizes using the mean differences of the groups while taking into consideration unequal sample sizes within the pre‐post‐control design, also known as d_ppc2 (Morris 2008). We utilized Cohen's d to guide the interpretation of the effect sizes as follows: small = 0.2, medium = 0.5 and large = 0.8 (Cohen 1992).

To assess whether predicted therapy gains were maintained 6‐months posttreatment, we utilized LMM to compare each of the dependent variables at 6‐month follow up to their posttreatment and baseline assessments. We included time in the model as a fixed factor (coded as discrete timepoints baseline, posttreatment, and 6‐month follow up). Condition (waitlist vs intervention) was also included as a fixed factor to account for potential variation that might occur from being in the waitlist condition before attending the group. Random intercepts for participants accounted for repeated measures. The covariance structure for repeated measures was modelled using autoregressive (AR1). We estimated post‐hoc comparisons pairwise comparisons using EMMs comparing time and condition, adjusting for multiple comparisons. All participants who had at least one timepoint available were included in the analysis. Effect sizes were estimated utilizing Cohens d _av as recommended by Cumming (2012).

We conducted additional analyses to further examine the clinical significance of the primary outcomes. Clinically significant change (CSC) refers to the degree to which individuals achieve normal functioning after treatment (Jacobson and Truax 1991) and enables categorization of outcomes as improved, remained unchanged, or deteriorated. Calculation considers both the reliability (RCI cut‐off 1.96) and clinical significance of change, which in the current study was the midpoint between the clinical and general population means (Method C). We were only able to calculate CSC could only be calculated for the BSL‐23 due to a lack of published Australian clinical and general population norms for the BSL‐supp and LFPS‐BF. To inform the analysis of the BSL‐23 we drew from two Australian studies to gain estimates for the non‐help‐seeking (Meaney et al. 2016) and help‐seeking populations (Heerebrand et al. 2021).

While Australian clinical and general population norms are not available for the mean BSL‐supp score, we were able to examine the probability that participants ceased engaging in specific self‐destructive/impulsive behaviors over time (baseline, posttreatment and 6‐month follow‐up). Each behavioral item was recoded into a binary variable “not at all = 0” and “once or more in the previous week = 1” to examine the likelihood of people engaging in the specified behavior over time. Generalized estimating equation (GEE) models with binomial distribution and logit link function were used to account for repeated measures over time. The reference category was set to 1 (once or more in the previous week) allowing interpretation of the model in terms of the predicted probability that people “did not engage in the behavior” in the previous week (not at all = 0). Again, condition (waitlist vs intervention) was included as a fixed factor to account for potential variation between conditions and autoregressive (AR1) was defined for the covariance structure. Model fit was evaluated using Wald chi‐square t‐tests and odds ratios (Exp [B]) described the relative likelihood of “not” engaging in the behaviour over time. EMMs were calculated for each timepoint and pairwise comparisons were undertaken with LSD adjustment for multiple comparisons.

8. Results

8.1. Participant Flow

One‐hundred and forty participants were assessed for eligibility to participate in the group between July 2020 and August 2022. A total of 121 participants consented to participate in the research evaluation. Eighty participants were assigned to the intervention condition. Forty‐one participants were assessed 28 days or more before the intervention started and were assigned to the uncontrolled waitlist. All participants who participated in the waitlist (n = 41) were offered the intervention after their waiting period and 83% commenced group. Treatment completion was defined as attending 60% or more of the group sessions.

The participant flow chart (see Figure 1) highlights the number of people in the waitlist and intervention conditions that completed the group, 6‐month follow up interviews and the number of completed assessments included in the analysis. In total, n = 99 participants commenced the treatment, and 62% met the criteria for completion. Between February 2021 and May 2023, 6‐month follow‐up interviews were offered to people who commenced the group and 37% completed this assessment. The majority (87%) of those who completed the 6‐month follow‐up interview had completed the treatment (intervention n = 21, waitlist n = 11). All participants enrolled in the study were included in the statistical analyses.

Figure 1.

Participant flow chart.

8.2. Baseline Comparisons

Participant demographic and baseline clinical characteristics are displayed in Table 1. The total sample consisted largely of women (89%), followed by men (9%), and transgender or gender non‐conforming people (3%; TGNC). A small proportion of the sample were Aboriginal or Torres Strait Islander (5%). There were no statistically significant differences in age between the waitlist (Md = 26.00, n = 41) and intervention (Md = 29.50, n = 80) conditions U = 1492, z = −0.811, p = 0.42, d = 0.15.

Comparison between baseline scores across primary and secondary outcomes did not reveal statistically significant differences between the waitlist and intervention conditions in most cases. We identified a statistically significant difference in hypermentalizing at baseline, with those in the waitlist condition (Md = 0.42, n = 32) rating significantly higher compared to the intervention condition (Md = 0.17, n = 59), U = 1180, z = 2.05, p = 0.04, d = 0.42. Further, baseline psychosocial dysfunction was lower in the waitlist condition (Md = 28.00, n = 40) compared to the intervention condition (Md = 30.50, n = 80) and this difference approached significance U = 1244, z = −1.98, p = 0.05, d = 0.37.

9. Acute Treatment Effects

9.1. Primary Outcomes

There was a statistically significant differential rate of change between conditions in borderline symptom severity and total personality dysfunction (including subscales of self and interpersonal functioning) over time (see Figure 2). A significant fixed effect (condition x time) was evident for borderline symptom severity with an estimated 0.07 reduction per week [95% CI: −0.11, −0.04] in the intervention condition relative to the waitlist, F(1, 95.18) = 16.50, p < 0.001. The effect size was estimated to be large d _ppc2  = 0.95. The fixed effect (condition x time) was not significant for impulsive and self‐destructive behaviour, with an estimated 0.02 point reduction per week [95% CI: −0.04, 0.00] in the intervention condition compared to waitlist, F(1, 101.28) = 2.72, p = 0.10, d _ppc2  = 0.65.

Figure 2.

Differential rate of change in primary outcomes between waitlist and intervention over time (N = 121).

There was a statistically significant fixed effect (time x condition) for overall personality dysfunction with an average reduction of 0.58 per week [95% CI: −0.78, −0.38] in the intervention group relative the waitlist F(1, 81.94) = 33.47, p < 0.001, d _ppc2  = 1.07. Closer inspection of subscale scores on this measure also revealed a significant fixed effect (condition x time) in self‐dysfunction with an estimated mean reduction of 0.33 per week [95% CI: −0.45, −0.21] in the intervention compared to waitlist, F(1, 84.02) = 31.13, p < 0.001, d _ppc2  = 1.25; and a mean reduction of 0.25 per week [95% CI: −0.37, −0.13] estimated for interpersonal dysfunction F(1, 83.13) = 17.60, p < 0.001, d _ppc2  = 0.69 for the intervention relative to waitlist. Estimated marginal means (EMMs) from baseline to 12 weeks for each of the primary outcomes are available in Table S1.

9.2. Secondary Outcomes

Similarly, there were statistically significant fixed effects (time x condition) across all secondary outcomes over time. Difficulties in regulating emotions mean total score in the intervention reduced 1.03 per week [95% CI: −1.52, −0.54] relative to the waitlist condition F(1, 92.86) = 17.59, p < 0.001, d _ppc2  = 0.98. Certainty in understanding mental states increased 0.05 per week [95% CI: 0.2, 0.07] in the intervention condition relative to the waitlist F(1, 97.72) = 14.04, p < 0.001, d _ppc2  = 1.41. Uncertainty of mental states reduced 0.08 per week [95% CI: −0.11, −0.05] in the intervention condition relative to the waitlist, F(1, 82.25) = 23.72, p < 0.001, d _ppc2  = 1.07. Finally, problems in psychosocial functioning reduced 0.56 per week [95% CI: −0.89, −0.24] in the intervention condition relative to the waitlist F(1, 120.32) = 11.74, p < 0.001, d _ppc2  = 0.96. Table 2 reports the EMMs per condition over time across each of the secondary outcomes.

Table 2.

Linear mixed modelling analyzing the differential rate of change per week in the intervention relative to the waitlist across secondary outcomes.

Outcome	Week	Intervention EMM (95% CI)	Waitlist EMM (95% CI)	Mean difference (95% CI)	p
DERS	0	65.56 (62.56, 68.56)	63.60 (59.53, 67.68)	1.95 (−3.11, 7.02)	0.45
	4	62.29 (59.50, 65.09)	64.46 (60.63, 68.28)	−2.16 (−6.90, 2.57)	0.37
	8	59.03 (56.06, 62.00)	65.31 (61.10, 69.52)	−6.28 (−11.44, −1.13)	0.02
	12	55.77 (52.28, 59.25)	66.17 (61.06, 71.27)	−10.40 (−16.58, −4.22)	0.001
RFQc	0	0.28 (0.15, 0.42)	0.48 (0.31, 0.65)	−0.20 (−0.42, −0.01)	0.07
	4	0.38 (0.26, 0.50)	0.40 (0.25, 0.55)	−0.02 (−0.21, 0.17)	0.82
	8	0.48 (0.35, 0.61)	0.32 (0.16, 0.49)	0.16 (−0.05, 0.37)	0.13
	12	0.58 (0.42, 0.73)	0.24 (0.03, 0.45)	0.34 (0.08, 0.60)	0.01
RFQu	0	1.80 (1.59, 2.01)	1.52 (1.25, 1.79)	0.28 (−0.06, 0.62)	0.11
	4	1.56 (1.37, 1.76)	1.60 (1.35, 1.85)	−0.04 (−0.35, 0.28)	0.82
	8	1.33 (1.12, 1.53)	1.68 (1.41, 1.95)	−0.35 (−0.69, −0.02)	0.04
	12	1.09 (0.85, 1.33)	1.76 (1.44, 2.09)	−0.67 (−1.07, −0.27)	0.001
WSAS	0	30.07 (28.41, 31.73)	27.25 (25.05, 29.44)	2.83 (0.07, 5.58)	0.04
	4	28.54 (27.09, 29.99)	27.97 (26.00, 29.94)	0.57 (−1.88, 3.01)	0.65
	8	27.00 (25.42, 28.58)	28.69 (26.41, 31.97)	−1.69 (−4.46, 1.08)	0.23
	12	25.47 (23.48, 27.45)	29.41 (26.44, 32.39)	−3.95 (−7.52, −0.37)	0.03

Note: Includes all patients with outcome data available for at least 1 timepoint (N = 121).

Abbreviations: CI, Confidence Interval; DERS, Difficulties in Emotion Regulation Scale; EMM, Estimated Marginal Means; RFQc, Reflective Functioning Questionnaire – Certainty; RFQu, Reflective Functioning Questionnaire ‐ Uncertainty; WSAS, Work and Social Adjustment Scale.

9.3. Follow‐Up Treatment Effects

Thirty‐seven participants who completed the follow‐up interview were asked what other services they engaged with to support their mental health in the 6 months after Road Maps. The majority had engaged with a practitioner (e.g., psychologist, psychiatrist, psychotherapist) in the private sector (73%) or a general practitioner (70%). A smaller proportion had engaged with nongovernment/private health network funded mental health services (35%) or a tertiary community adult mental health service (34%). Six participants noted that they had attended an emergency department/general hospital (17%), and three participants (8%) reported a psychiatric inpatient admission in this timeframe.

The outcomes comparing baseline and posttreatment scores compared to 6‐month follow‐up are reported in Table 3. Longitudinal analysis demonstrated that ratings at 6‐month follow up were significantly improved across all measures with large (i.e., borderline symptom severity, personality functioning, self‐functioning, and emotional regulation) to medium effects (i.e., impulsive/self‐destructive behaviour, interpersonal functioning, reflective functioning, and psychosocial functioning). There were no statistically significant differences and minimal to small effect sizes across all measures between posttreatment ratings and 6‐month follow‐up, suggesting that gains from the intervention were maintained.

Table 3.

Change in Estimated Marginal Means (EMMs) at 6‐month follow‐up compared to baseline and posttreatment.

Outcome	Time	EMMs (95% CI)	Difference at 6‐month Fu (95% CI)	p	Effect size Cohen's d av
BSL‐23	Baseline	2.46 (2.28, 2.64)	−0.80 (−1.10, −0.49)	< 0.001	0.84
	Posttreatment	1.64 (1.41, 1.87)	0.03 (−0.28, 0.33)	0.87	0.09
BSL‐supp	Baseline	0.69 (0.60, 0.79)	−0.34 (−0.51, −0.18)	< 0.001	0.61
	Posttreatment	0.41 (0.29, 0.53)	−0.06 (−0.22, 0.10)	0.47	0.17
LPFS total	Baseline	36.73 (35.42, 38.04)	−6.18 (−7.77, −4.58)	< 0.001	0.91
	Posttreatment	30.51 (28.91, 32.12)	0.04 (−1.77, 1.86)	0.96	0.08
Self	Baseline	19.85 (19.15, 20.55)	−2.78 (−3.80, −1.76)	< 0.001	0.86
	Posttreatment	16.43 (15.56, 17.30)	0.64 (−0.41, 1.69)	0.23	0.07
Interpersonal	Baseline	16.87 (16.11, 17.64)	−3.41 (−4.39, −2.44)	< 0.001	0.76
	Posttreatment	13.14 (13.10, 14.98)	−0.60 (−1.62, 0.42)	0.25	0.20
DERS	Baseline	64.57 (61.96, 67.18)	−14.35 (−17.67, −11.03)	< 0.001	1.06
	Posttreatment	52.89 (49.56, 56.22)	−2.67 (−6.68, 1.34)	0.19	0.32
RFQc	Baseline	0.28 (0.11, 0.45)	0.40 (0.15, 0.65)	0.002	0.79
	Posttreatment	0.57 (0.39, 0.74)	0.11 (−0.11, 0.34)	0.32	0.20
RFQu	Baseline	1.77 (1.54, 1.99)	−0.71 (−1.00, −0.41)	< 0.001	0.68
	Posttreatment	0.97 (0.74, 1.19)	0.09 (−0.18, 0.37)	0.50	0.04
WSAS	Baseline	28.63 (27.06, 30.20)	−5.32 (−8.20, −2.44)	< 0.001	0.50
	Posttreatment	23.77 (21.73, 25.81)	−0.46 (−3.25, 2.34)	0.75	0.13

Note: Includes all patients with outcome data available for at least 1 timepoint (N = 121).

Abbreviations: BSL‐23, Borderline Symptom List – 23; BSL‐supp, Borderline Symptom List – Supplement; CI, Confidence Interval; DERS, Difficulties in Emotion Regulation Scale; LPFS, Levels of personality functioning; RFQc, Reflective Functioning Questionnaire – Certainty; RFQu, Reflective Functioning Questionnaire – Uncertainty; WSAS, Work and Social Adjustment Scale.

There were no significant fixed effects for condition (waitlist, intervention) across the majority of models. The only exception to this was when examining psychosocial functioning in which there was a significant effect of condition (intervention vs waitlist), F(1, 127.12) = 4.75, p = 0.03 whereby psychosocial dysfunction was significantly higher in the intervention condition compared to waitlist at each timepoint. However, further analysis adding an interaction between time x condition was not statistically significant F(2, 136.07) = 1.09, p = 0.34, suggesting that the change over time did not differ between those who were originally assigned to the waitlist compared to intervention.

9.4. Additional Analyses: Clinically Significant Change

Additional analyses were conducted to explore CSC on the primary variable of borderline symptom severity (BSL‐23). We examined all available posttreatment data (n = 63) and found that 24 (38%) participants were classified as having demonstrated CSC followed by 16 (25%) who were considered improved, but the change was not clinically significant. It was apparent that 19 (30%) participants were unchanged on this measure, while 4 (6%) participants' symptom severity worsened. At 6‐month follow‐up (n = 37) it was apparent that compared to baseline 14 (38%) participants achieved CSC and 6 (16%) participant's borderline symptom severity had improved. In contrast, 14 (38%) participants' scores were unchanged from baseline and 3 (8%) participants had scores that reflected deterioration.

GEE models were utilized to examine changes in participants probability of engaging in each of the specific behavioral items of the BSL‐supp over time. We found that the odds of having an uncontrolled angry outburst decreased by 77% over time (Exp(B) = 0.23, p < 0.001). More specifically, examination of the EMMs showed that the likelihood of ‘not’ having angry outbursts were 0.51 (95% CI: 0.42, 0.61) at baseline and significantly increased to 0.82 (95% CI: 0.68, 0.91) at 6‐month follow‐up (p < 0.001). There was no significant difference between posttreatment and 6‐month follow up (p = 0.31) suggesting that the change occurred between baseline and posttreatment and was maintained at 6‐month follow‐up. There was no statistically significant difference between participants who commenced the research trial in the waitlist condition compared to intervention (Exp(B) = 0.98, p = 0.95).

Similar overall trends were observed for the odds of engaging in sexual activities that the person later felt ashamed about which decreased by 85% over time (Exp(B) = 0.15, p = 0.02); the odds of engaging in high risk behaviour (e.g., knowingly driving too fast, running around on the roofs of high buildings, balancing on bridges, etc.) decreased by 73% over time (Exp(B) = 0.27, p = 0.007); the odds people told someone they wanted to kill themselves decreased by 65% (Exp(B) = 0.35, p = 0.048); and the odds of binge‐eating decreased 42% over time (Exp(B) = 0.48, p = 0.04). Finally, the odds that people had used drugs in the past week reduced by 42% over time (Exp(B) = 0.58, p = 0.03) though in this instance the significant difference in EMMs occurred between posttreatment = 0.64 (95% CI: 0.52, 0.74) and 6‐month follow‐up = 0.75 (95% CI: 0.62, 0.85). Full statistical results including pairwise comparisons and EMMs are presented in Table S2.

10. Discussion

This study evaluated the effectiveness of a 12‐session therapy group for adults with a diagnosis of BPD. We utilized a non‐randomized control trial design to compare the short‐term group intervention to a naturalistic waitlist control. The findings suggested that compared to waitlist, participation in the short‐term therapy group resulted in significant reductions in the primary outcomes of borderline symptom severity and core features of personality dysfunction (i.e., self and interpersonal). These findings were maintained for those who participated in the 6‐month follow‐up interview. Further, the reduction in borderline symptom severity was clinically significant for a third of the sample at the end of the intervention and 6‐month follow‐up. A recent study by Azevedo et al. (2024) provided benchmark data from DBT skills training groups and proposed a critical value of 0.62 on the BSL‐23 as an indicator of effectiveness. This critical value was exceeded in the current study. Our findings support a growing body of research indicating that short‐term psychological interventions may be beneficial for some people with a diagnosis of BPD (Spong et al. 2021).

We also examined whether engagement in specific self‐destructive/impulsive behaviors common to BPD changed over‐time. The odds of participants having uncontrollable angry outbursts, telling others they were going to kill themselves, binge‐eating, high risk behaviour (such as knowingly driving too fast etc.), and sexual activities they later felt ashamed of, were all significantly decreased at posttreatment and maintained at 6‐month follow‐up. The odds of engaging in illicit drug use significantly decreased by 6‐month follow‐up. Road Maps aimed to increase understanding of connections between thoughts, physiological sensations, emotions, behaviours, and interpersonal experiences through functional analysis as the primary exploratory intervention. These findings lend support the use of this approach, as a potential facilitator of behavior change.

Secondary outcomes in the current study revealed large and significant improvement in emotion regulation capacity in the intervention group relative to the waitlist, which was maintained at 6‐month follow up. This is an important therapeutic outcome as postulated by theories informing evidence‐based interventions such as dialectical behaviour therapy (DBT; Linehan ¹⁹⁹³). Road Maps taught emotion‐focussed coping strategies such as self‐compassion and mindfulness which may have contributed to improvement. Further research is needed to elucidate factors mediating change within the intervention.

The RFQ‐8 proposes that shifts towards zero reflect more genuine mentalizing (Fonagy et al. 2016). In the current study, there was a significant reduction in uncertainty about mental states suggesting improved understanding and interpretation of self or others' internal mental states. In contrast, certainty about mental states significantly increased. We compared this pattern of results to the average score of a nonclinical sample on the same measure (Cucchi et al. 2018) and found that the ratings of certainty in our study were shifting towards the average score observed in the nonclinical sample. One explanation may be that some degree of certainty is required to mentalize well. However, recent studies have scrutinized the original scoring method proposed for the RFQ‐8 (Müller et al. 2022; Woźniak‐Prus et al. ²⁰²²). These studies suggest that the tool targets uncertainty of mental states as a unidimensional construct and that the subscale relating to certainty is questionable. As such, the shifts in ‘certainty about mental states’ should be interpreted with caution.

Generally, psychotherapeutic interventions for BPD have been associated with slight functional improvement at follow‐up (Álvarez‐Tomás et al. ²⁰¹⁹). However, it was apparent from the secondary outcomes of the current study that there was a moderate and significant reduction in psychosocial dysfunction that was maintained at follow‐up. While Road Maps does not specifically target psychosocial functioning, it does aim to assist participants to build agency and be active in their own lives. Further research is needed to explore interventions which address important life domains such as study, employment, and activities of daily living.

There were several limitations to this study. First, we did not randomly allocate participants to conditions as this was not appropriate or feasible in this clinical setting. Therefore, the findings should be considered with caution given research has shown that effect sizes may be larger in studies that compare interventions to a non‐manualized control (Spong et al. 2021). As common in clinical research settings, there was attrition over the course of the intervention and 6‐month follow‐up which resulted in a smaller number of participants assessed at the final timepoint (n = 37). Many of these participants had completed the treatment (87%) and therefore the results may be skewed towards those had a more positive experience of the intervention. While the statistical approach utilized in this study is robust to missing data and uneven sample sizes, findings should be interpreted with this in mind.

Another limitation was that assessments were self‐report which are more vulnerable to social desirability bias. Further research is needed including multiple methods of assessment to verify the findings. Finally, the timing of this evaluation coincided with the global pandemic which affected the delivery of group at times, particularly where local lockdowns and density limits were enforced. Unexpected changes to the delivery of groups or fear of contagion may have impacted engagement, accessibility, and opportunities for generalisation of key concepts.

Despite these limitations, a major strength of this study was its ecological validity. It demonstrated the potential effectiveness of a short‐term therapy group being delivered in resource‐constrained environments such as a publicly funded tertiary mental health service. At baseline, average symptom severity was high to very high and psychosocial functioning was moderately to severely impaired, suggesting that even a short intervention can have an impact for people experiencing significant levels of distress and impaired everyday functioning. The intervention was developed with the view to broader implementation by a multidisciplinary mental health workforce, without the need for intensive training in a specialist evidence‐based treatment. However, whilst it aligns with a generalist treatment approach, consideration of access to adequate training and supervision is still needed to ensure safe delivery of the intervention.

Future research is required to replicate the findings of this study within a more rigorous research design. It would also be advantageous to further explore how short‐term interventions are situated within a broader system of stepped care and whether the availability of an intermediate intervention facilitates greater access to intensive evidence‐based psychotherapies for those in need.

In sum, the findings of this study are encouraging and challenge assumptions that short‐term interventions are not beneficial for people with a diagnosis of BPD. Indeed, brief interventions can have a significant impact on functioning and symptomatology for some people and are an integral part of a broader stepped model of care.

Ethics Statement

Approved by the Southern Adelaide Clinical Human Research Ethics Committee (reference number HREC/20/SAC/92).

Consent

All participants provided electronic or written informed consent.

Conflicts of Interest

The authors undertook this study as part of their paid employment at the Borderline Personality Disorder Collaborative.

Supporting information

RoadMaps JCLP SuppTable1 290125.

RoadMaps JCLP SuppTable2 290125.

Data Availability Statement

The data that supports the findings of this study are available on request from the corresponding author. Data are not publicly available due to privacy or ethical restrictions.