Gender-specific effectiveness of dialectic-behavioral therapy for patients with complex post-traumatic stress disorder (DBT-PTSD) – results of an observational single center study

ABSTRACT

Background: Complex post-traumatic stress disorder (cPTSD) was recently included in the ICD-11, extending the PTSD symptom profile to encompass disturbances in self-organization (DSO). Trauma-focused Dialectical Behavior Therapy (DBT-PTSD) is an effective psychotherapeutic treatment, particularly among women. However, empirical data on DBT-PTSD's gender-specific effectiveness in naturalistic settings remains limited.

Objective: The present study examined whether DBT-PTSD is similarly effective in reducing symptoms of cPTSD among women, men, and gender minority (GM) individuals under routine clinical conditions.

Method: This observational single-center study included 215 patients (women = 78.1%, men = 18.6%, GM = 3.3%; 44.6+/−11.5 years) with cPTSD treated with DBT-PTSD at a psychosomatic inpatient clinic in Austria. Primary outcome was mean change in symptom severity assessed by the International Trauma Questionnaire (ITQ) before (T1) and after treatment (T2) and secondary assessment included anxiety and depression, somatization, dissociation, assessment of functioning, personality traits, emotion regulation, locus of control and borderline symptomatology. Repeated measures ANOVAs were used to evaluate treatment outcomes. Baseline predictors (T1) for ITQ score at T2 were identified by machine learning models and validated with mixed linear models.

Results: While DBT-PTSD overall led to significant reductions in cPTSD symptoms with large effect size (p < .001, η²= 0.17), no significant difference was observed between men and women neither for the ITQ (p = .90, η² < 0.01), nor for any of the secondary outcomes (p = .08–.86; η² ≤ 0.01). Secondary outcomes showed statistically significant improvements in depression, anxiety, somatization emotion regulation and the locus of control (p < .001–.012). Higher negative affectivity was associated with worse outcomes in both men and women, while higher antagonism was associated with better outcomes in women and worse outcomes in men.

Conclusions: DBT-PTSD is effective in reducing cPTSD symptoms among men and women in real-world clinical settings. However, larger samples of men are needed to validate the findings of this study.

HIGHLIGHTS

• Trauma-focused Dialectical Behavior Therapy (DBT-PTSD) significantly reduces symptoms of complex post-traumatic stress disorder (cPTSD) in routine inpatient care.

• Men and women equally benefitted from the intervention.

• Gender-specific predictors of treatment outcomes were identified.

1. Background

In the 11th revision of the International Classification of Diseases (ICD-11) complex post-traumatic stress disorder (cPTSD) was introduced as a distinct diagnosis. To meet the diagnostic criteria for cPTSD, individuals must exhibit a significant ‘disturbance in self-organization’ (DSO) in addition to the core symptoms of post-traumatic stress disorder (PTSD). DSO is defined by (a) difficulties in emotion regulation, (b) a negative self-concept, and (c) unstable or dysfunctional interpersonal relationships (Maercker et al., 2013). Epidemiological studies estimate cPTSD prevalence in the general population to range between 0.5% and 3.8% (Cloitre et al., 2019; Maercker et al., 2018; Riedl et al., 2024) with most studies reporting no significant gender differences (Lonnen & and Paskell, 2024).

Although there is a substantial body of literature on gender-specific differences in trauma-related disorders, it remains unclear to what extent these findings apply to cPTSD. Prior research highlights gender-based differences in the type and context of trauma exposure. Women are disproportionately affected by interpersonal trauma, such as sexual assault, childhood abuse, or domestic violence (Olff, 2017; Tolin & Foa, 2006), whereas men more frequently experience trauma related to accidents, physical violence, or combat (Ho et al., 2021; Tolin & Foa, 2006). Additionally, several studies indicate that women are twice as likely as men to meet PTSD diagnostic criteria (Hapke et al., 2006; Kilpatrick et al., 2013; Tolin & Foa, 2006). These differences are shaped by a complex interplay of biological, psychological, and social factors (Christiansen & Berke, 2020; Christiansen & Hansen, 2015; Kimerling et al., 2018).

Women more often develop internalizing symptoms such as anxiety, dissociation, depression, and panic attacks (Fullerton et al., 2001; Olff, 2017), and are also more likely to seek psychological support increasing the chances of recognition and treatment (Karatzias et al., 2007; Tarrier et al., 2000). In contrast, men are more prone to suppress emotions and exhibit externalizing behaviors such as aggression or impulsivity, which may contribute to underdiagnosis (Black & Flynn, 2021), higher dropout rates during therapy (van Minnen et al., 2002) and generally less favorable treatment outcomes (Wade et al., 2016). Emerging evidence also points to elevated rates of trauma exposure and distress among gender minorities (GM) (Feil et al., 2023; Henry et al., 2021) – including transgender, non-binary, and gender-diverse individuals – who remain underrepresented in trauma research.

While evidence supports the effectiveness of established psychotherapeutic interventions in reducing PTSD symptoms, as well as associated depression and anxiety, data on their impact on cPTSD and DSO symptoms remain limited (Ford, 2021). Historically, individuals meeting cPTSD criteria – especially those with childhood trauma histories – have shown poorer treatment outcomes compared to those with adult-onset trauma (Karatzias et al., 2019).

Among disorder-specific treatments, Dialectical Behavior Therapy for PTSD (DBT-PTSD) has emerged as a promising approach, particularly for cPTSD stemming from childhood sexual abuse (Bohus et al., 2013). DBT-PTSD integrates cognitive behavioral techniques with mindfulness, acceptance strategies, and skills training in emotion regulation, interpersonal effectiveness, stress tolerance, and self-esteem enhancement (Bohus et al., 2020). Two randomized controlled trials (RCTs) support its efficacy: a 12-week inpatient study (n = 74) showed significantly greater symptom reduction than the control group (CAPS: 33.2 vs. 2.1) (Bohus et al., 2013), and a 15-month outpatient study (n = 194) demonstrated its superiority over cognitive processing therapy, with higher effect sizes (d = 1.35 vs. 0.98) and lower dropout rates (25.5% vs. 40.7%) (Bohus et al., 2020).

Despite these encouraging results, it remains unclear whether DBT-PTSD is equally effective for survivors of other trauma types. To date, no studies have examined its gender-specific effectiveness, and most evidence is based on RCTs. While RCTs are the gold standard in psychotherapy research, their narrow inclusion criteria and standardized protocols often lead to homogenous samples and may not reflect the diversity of real-world patients or clinical settings. In contrast, naturalistic observational studies can offer valuable insights into treatment effectiveness in routine care.

The present study aims to assess the gender-specific effectiveness of DBT-PTSD in a naturalistic sample of individuals with cPTSD. Specifically, we hypothesize that (a) DBT-PTSD will lead to significant symptom reduction across all genders, (b) women will report greater difficulties in emotion regulation and higher depression and anxiety symptoms than men, and (c) women will show greater symptom improvement. We also explore gender-specific predictors of treatment outcomes.

2. Methods

2.1. Patient sample and procedure

This single-center, retrospective study analyzed routine clinical data collected between April 2021 and July 2024 at the Waldviertel Psychosomatic Center (PSZW) in Eggenburg (Austria). This residential mental health facility, operating within the Austrian public healthcare system, provides treatment covered by public health insurance. It serves adults (18+ years), excluding those with acute suicidal behavior or active substance abuse, though patients with past suicidal or self-injurious behavior are eligible. Clinical outcomes are routinely assessed using the Computer-based Health Evaluation System (CHES) (Holzner et al., 2012) at preadmission, treatment start (T1), and end of treatment (T2, typically within the last four days of the inpatient stay). Inclusion criteria for this study were: (a) treatment at Waldviertel Psychosomatic Center (PSZW), (b) minimum treatment duration of 4 weeks, (c) cPTSD diagnosis per ITQ algorithm, and (d) sufficient German proficiency to participate in treatment and complete assessments, as judged by the clinical team. For patients who discontinued treatment early, T2 assessments were used if available.

As data were collected in a naturalistic setting, information on treatment fidelity was not available. All patients provided written informed consent for the use of their clinical data for quality assurance and research. Study participation did not affect the clinical treatment. The study was approved by the Ethics Committee of the University of Innsbruck (No. 108/2022) and conducted in accordance with the Declaration of Helsinki.

2.2. Treatment

DBT-TF is a structured psychological intervention specifically developed for individuals with a history of childhood sexual abuse and emotion dysregulation. DBT-PTSD is designed in phases, integrating core principles of standard DBT with additional components from trauma-focused cognitive–behavioral therapy, compassion-focused therapy, and acceptance and commitment therapy. This approach consists of distinct treatment phases (commitment; planning and motivation; addressing trauma-related escape strategies; exposure – reducing re-experiencing of trauma-related emotions; developing a life worth living), each tailored to address the complex symptom profiles of individuals with PTSD stemming from severe sexual or physical childhood abuse (Steil et al., 2015).

Each phase includes essential and optional modules, guided by manualized ‘if–then’ rules to tailor treatment to individual needs. Psychotherapists delivering DBT-PTSD at the PSZW received specialized training at the Central Institute in Mannheim under Martin Bohus and ongoing supervision consistent with DBT-PTSD protocols. The intervention was delivered within a multidisciplinary framework and included complementary therapies such as structured physical activity (45 min, twice weekly), group biofeedback (80 min, weekly), art therapy (90 min, weekly), and music therapy (60 min, weekly).

2.3. Measures

2.3.1. Primary outcome: International Trauma Questionnaire (ITQ)

The ITQ is a self-report tool used to assess symptoms of PTSD and cPTSD, in line with ICD-11 criteria. The PTSD subscale includes six items covering re-experiencing, avoidance, and heightened threat (two items each), plus three items assessing impairment of personal functioning, working ability and role functioning by the PTSD symptoms. cPTSD includes an additional six items measuring disturbances in self-organization (DSO), comprising affective dysregulation, negative self-concept, and relational difficulties (two items each), along with three DSO-related impairment items. Each item is rated for the last month on a five-point Likert scale (0 = not at all; 4 = very much). To be consistent and comparable with prior research, PTSD and DSO symptom subscales were calculated separately, excluding functional impairment items. Subscale scores range from 0 to 24; the total ITQ score ranges from 0 to 48. A symptom was considered present if rated ≥2 (i.e. at least ‘moderately’ bothered) (Cloitre et al., 2018). The categorial PTSD label required endorsement of at least one symptom in each PTSD cluster and one associated impairment item and for cPTSD the PTSD criteria had to be met, along with at least one symptom present (i.e. rated ≥2) in each DSO symptoms and one DSO impairment item. Per ICD-11, a person could be diagnosed with either PTSD or cPTSD. Good to excellent internal consistency (α = .88 – .91) and robust construct validity has been observed for the English original version of the ITQ (Cloitre et al., 2018; Hyland et al., 2017). An item-response analysis of the German ITQ showed good model fit and acceptable to good alignment with the original English items (Christen et al., 2021). In our sample, acceptable to good internal consistency was observed for the ITQ total score (α = .80; ω = .79), while the internal consistency for the PTSD subscale was acceptable (α = .73; ω = .71) and for the DSO subscale questionable (α = .67; ω = .63).

2.3.2. Secondary outcomes

The Posttraumatic Stress Disorder Checklist for DSM-5 (PCL-5) is a 20-item measure assessing PTSD symptoms across four clusters in the last month with higher scores indicating greater symptom severity (α = .86). The Dissociation-Tension Scale acute (DSS-4) consists of four items measuring depersonalization, derealization, somatoform dissociation, and analgesia (α = .83; ω = .83). The World Health Organization Disability Assessment Scale (WHODAS 2.0, 12-item version) evaluates functional impairment across six domains – mobility, life activities, communication, social participation, self-care, and social functioning – along with a total disability score. Scores range from 0 to 100, with higher scores indicating greater disability (α = .82; ω = .79). The Personality Inventory for DSM-5 – Short Form (PID-5-SF) assesses five dysfunctional trait domains, namely negative affectivity (α = .64; ω = .76), detachment (α = .70; ω  = .70), antagonism (α = .67; ω = .79), disinhibition (α = .74; ω = .78), and psychoticism (α = .82; ω = .76). The Emotion Regulation Questionnaire (ERQ) measures two strategies of emotion regulation: cognitive reappraisal (α = .92; ω = .89) and expressive suppression (α = .76; ω = .76). The Generalized Anxiety Disorder-7 (GAD-7) assesses symptoms of anxiety such as nervousness and excessive worrying (α = .86; ω = .78), while the Patient Health Questionnaire-9 (PHQ-9) evaluates depressive symptoms with scores ≥10 indicating moderate depression (α = .84; ω = .72). The Patient Health Questionnaire-15 (PHQ-15) measures somatic symptom severity, with scores ≥5 indicating mild, ≥10 moderate, and ≥15 severe symptoms (α = .82; ω = .79). The Borderline Symptom List 23 (BSL-23) assesses borderline symptomatology (α = .95; ω = .93). Sleep impairment was assessed with the Jenkins Sleep Scale (JSS) (α = .78; ω = .74), locus of control with the Internal–External Locus of Control Short Scale–4 (IE-4) and level of current pain was assessed with a numerical rating scale ranging from 0–10 with higher values indicating higher pain. Sociodemographic data and data on previous treatment were assessed using specifically designed self-report questionnaires.

2.4. Statistics

Only participants with complete ITQ data at both baseline (T1) and end of treatment (T2) were included in the primary analyses. Independent sample t-tests were used to compare T1 scores for all primary and secondary outcomes between included and excluded participants, as well as between male and female participants. Group differences in this study were quantified using Hedge’s g (interpreted as: g<0.2 = negligible, 0.2–0.5 = small, 0.5–0.8 = moderate, >0.8 = large) (Ellis, 2010).

The primary aim was to examine gender differences in symptom change from T1 to T2 on the ITQ total score and its subscales (PTSD and DSO). These were analyzed using repeated measures ANOVAs with gender as a between-subjects factor. An a priori power analysis indicated that a sample size of n = 160 would be sufficient to detect a small treatment-by-gender interaction (effect size f = 0.25; α = .05; power = .80; correlation = .60). Given the lack of prior data on gender-specific DBT-PTSD outcomes, a small effect size was chosen to detect subtle differences. To assess potential bias due to attrition, two intention-to-treat (ITT) sensitivity analyses were conducted: (a) Last Observation Carried Forward (LOCF) and (b) Expectation-Maximization (EM), assuming data were Missing Completely At Random (MCAR), tested via Little's test. We additionally analyzed the data using linear mixed-effects models (LMMs) to address potential limitations of repeated-measures ANOVA, namely unequal group sizes and missing data, which LMMs can accommodate more flexibly. Because participants were measured at only two time points, we specified random intercept–only models, as only t − 1 random effects can be estimated, and with t = 2 this limits the model to a single random effect (the intercept) (McCoach & Kaniskan, 2010). This specification accounts for subject-specific baseline differences while yielding robust estimates of the fixed effects. To further evaluate the robustness of the results we calculated the Bayes factors to quantify the strength of evidence for or against the tested effects. A Bayes factor greater than 1 indicates evidence in favor of the alternative hypothesis, whereas values below 1 indicate evidence for the null, with larger deviations reflecting stronger evidence. As DBT-PTSD was developed for individuals with histories of severe childhood sexual or physical abuse, the primary analyses were also repeated in a subsample reporting either physical or sexual violence or unspecified childhood abuse. Effect sizes were classified as negligible (η²<0.01), small (η²= 0.01–0.06), medium (η²= 0.06–0.14), or large (η²≥0.14) (Ellis, 2010).

Minimally Important Differences (MID) were estimated using both anchor-based and distribution-based methods. WHODAS-12 item 5 (‘How much have you been emotionally affected by your health problems?’) served as the anchor, considered valid if its correlation with outcome measures exceeded r > .30 (Revicki et al., 2007). Participants were grouped by changes in anchor responses (‘improved,’ ‘no change,’ ‘deteriorated’), and mean ITQ scores (95% CI) were calculated per group. The lower CI bound of the ‘improved’ group defined the anchor-based cut-off. SEM-based distributional cut-offs were calculated from baseline and change score standard deviations (SDs), based on ITQ internal consistency. The MID most closely aligned across both approaches was selected.

Repeated measures ANOVAs were used to evaluate changes from T1 to T2 on secondary outcomes, with gender as the between-subjects factor. Bonferroni–Holm corrections were applied to adjust for multiple comparisons (Holm, 1979).

To identify robust baseline predictors of post-treatment ITQ scores, we used eXtreme Gradient Boosting (XGBoost), a machine learning method suited for complex clinical data and missing values. Separate models were trained for male and female participants to explore gender-specific patterns; gender minority participants were excluded due to small sample size. Variables with <40% missingness in the male sample (also met in females) were included. XGBoost's native handling of missing data was used, as imputation offered no benefit. Sixty-two baseline variables across cognitive, emotional, personality, social, functional, healthcare, and sociodemographic domains were included. The full list of included variables is presented in appendix D. Model performance was assessed via repeated 5-fold cross-validation (10 repetitions, 50 splits total). This approach reduces variance in performance estimates and improves stability assessment. Performance was measured using MAE and R², with standard errors computed across test folds. While models were optimized using MSE, we report MAE due to its clinical relevance and interpretability. Predictor importance was determined using SHapley Additive exPlanations (SHAP), aggregated across folds.

As validation, Linear Mixed Models (LMMs) with Bayesian Model Averaging (BMA) were calculated for men and women separately using the top 10 SHAP-ranked predictors for each group, which resulted in n = 15 overlapping included predictors. LMMs account for repeated measures (time within participants) and random intercepts, capturing individual baseline heterogeneity. BMA incorporates model uncertainty by averaging coefficients across all candidate models, weighted by Bayesian Information Criterion (BIC)-derived probabilities. This yields posterior inclusion probabilities (PIP), indicating the likelihood a predictor belongs in the true model. Baseline predictors were always included as grouped main effects. Analyses used only converged models to preserve parameter uncertainty, with all predictors z-standardized except for a single binary predictor (Sick Leave). We report PIP, weighted coefficient means, standard deviations, 95% credible intervals, and the number of contributing models.

Analyses used IBM SPSS v29.0 and Python 3.9.13, executed in PyCharm 2023.2.8 (Community Edition).

3. Results

3.1. Sample

A total of n = 220 patients who were categorized as cPTSD cases based on the ITQ scoring were eligible. Of these, five patients were excluded from the study since they had a treatment duration below 4 weeks. The remaining n = 215 patients were included in the analyses. Of the initially included n = 215 patients, n = 41 (19.1%) discontinued therapy and another n = 12 (5.5%) did not complete the final assessment. No statistically significant difference in treatment discontinuation was found in relation to gender (men =  17.5%; women =  19.0%; GM =  28.6%; χ²= 0.473; p = .79). The remaining n = 162 (75.3%) with complete data at the end of treatment were included for the per protocol analyses. Patients with and without incomplete data at T2 did not statistically differ in any of the primary or secondary outcomes at baseline (T1) (p = .18–.92).

As shown in Table 1, the majority of the sample were women (n = 168, 78.1%), while the remaining patients identified as men (n = 40, 18.6%) or GM (n = 7, 3.3%). GM individuals were significantly younger than male and female patients (p = .02) and had been treated in inpatient facilities more frequently (p  = .02). No significant differences between women and men were found regarding treatment duration, relationship status, parenthood, education or household income (all p > .05).

Table 1.

Sociodemographic data for study sample stratified by gender.

Variable	Total Sample (n = 215)	Men (n = 40)	Women (n = 168)	Gender minority (n = 7)	Test (df)	p-value
	mean (SD)	mean (SD)	mean (SD)	mean (SD)
Age in years	44.6 (11.5)	45.7 (10.7)	44.9 (11.5)	32.9 (11.2)	F = 3.98	.020
	n (%)	n (%)	n (%)	n (%)
Marital Status	192 (89.3)	36 (90.0)	151 (89.9)	5 (71.4)	χ2 (6) = 5.89	.44
Single	103 (53.6)	23 (63.9)	76 (50.3)	4 (80.0)
Married	33 (17.2)	7 (19.4)	26 (17.2)	0 (0.0)
Divorced / separated	50 (26.0)	6 (16.7)	43 (28.5)	1 (20.0)
Widowed	6 (3.1)	0 (0.0)	6 (4.0)	0 (0.0)
Missing values	23 (10.7)	4 (10.0)	17 (10.1)	2 (28.6)
Number of children					χ2 (10) = 7.67	.71
0	132 (61.4)	26 (65.0)	100 (59.5)	6 (85.7)
1	24 (28.9)	4 (28.6)	20 (29.4)	0 (0.0)
2	32 (38.6)	3 (21.4)	28 (41.2)	1 (100.0)
3	16 (19.3)	4 (28.6)	12 (17.6)	0 (0.0)
>3	11 (13.3)	3 (21.4)	8 (11.8)	0 (0.0)
Highest level of education	214 (99.5)	40 (100.0)	167 (99.4)	7 (100.0)	χ2 (10) = 12.37	.26
No formal education	4 (1.9)	2 (5.0)	2 (1.2)	0 (0.0)
Lower secondary education	35 (16.4)	7 (17.5)	25 (15.0)	3 (42.9)
Apprenticeship	65 (30.4)	17 (42.5)	46 (27.5)	2 (28.6)
Vocational secondary education	28 (13.1)	3 (7.5)	24 (14.4)	1 (14.3)
University entrance	69 (32.2)	9 (22.5)	59 (35.3)	1 (14.3)
Other	13 (6.0)	2 (5.0)	11 (6.9)	0 (0.0)
Missing values	1 (0.5)	0 (0.0)	1 (0.6)	0 (0.0)
Monthly household income	213 (99.1)	40 (100.0)	166 (98.8)	7 (100.0)	χ2 (6) = 6.57	.36
<1000€	54 (25.4)	11 (27.5)	41 (24.7)	2 (28.6)
1000€–2000€	113 (53.1)	21 (52.5)	89 (53.6)	3 (42.9)
2000€–3000€	27 (12.7)	3 (7.5)	24 (14.5)	0 (0.0)
>3000€	19 (8.9)	5 (12.5)	12 (7.2)	2 (28.6)
Missing values	2 (0.9)	0 (0.0)	2 (1.2)	0 (0.0)

Note. χ² = Chi-square-test; df = Degrees of Freedom; SD = Standard Deviation; F = F-statistic from One-way Analysis of Variance (ANOVA).

The majority of patients had received multiple psychiatric inpatient treatments in the past, with substantially higher numbers for women and GM individuals (p = .020). Overall, 84.5% of patients had received at least one year of previous outpatient treatment, with no significant gender difference (p = .65). In terms of reported traumatic experiences, women reported the highest proportion of experienced sexual violence (p = .006). Details are shown in Table 2.

Table 2.

Clinical data for study sample stratified by gender.

Variable	Total Sample (n = 215)	Men (n = 40)	Women (n = 168)	Gender minority (n = 7)
	mean (SD)	mean (SD)	mean (SD)	mean (SD)	Test (df)	p-value
Treatment duration (weeks)	9.68 (2.2)	9.09 (1.8)	9.85 (2.3)	9.1 (1.9)	F = 2.18	.116
	n (%)	n (%)	n (%)	n (%)
Number of previous psychiatric inpatient treatments	170 (79.1)	30 (75.0)	135 (80.4)	5 (71.4)	χ2 (10) = 21.62	.020
1	25 (14.7)	4 (13.3)	21 (15.6)	0 (0.0)
2	28 (16.5)	12 (40.0)	15 (11.1)	1 (20.0)
3	24 (14.1)	3 (10.0)	21 (15.6)	0 (0.0)
>3	93 (75.3)	11 (36.7)	78 (57.8)	4 (80.0)
Missing values	45 (20.9)	10 (25.0)	33 (19.6)	2 (28.6)
Duration of previous psychiatric inpatient treatments	162 (75.3)	30 (75.0)	127 (75.6)	5 (71.4)	χ2 (4) = 4.79	.31
<2 weeks	9 (5.6)	4 (13.3)	5 (3.9)	0 (0.0)
2–6 weeks	40 (24.7)	8 (26.7)	31 (24.4)	1 (20.0)
>6 weeks	113 (69.8)	18 (60.0)	91 (71.7)	4 (80.0)
Missing values	53 (24.7)	10 (25.0)	41 (24.4)	2 (28.6)
Duration of previous psychotherapeutic outpatient treatment	187 (87.0)	34 (85.0)	147 (87.5)	6 (85.7)	χ2 (4) = 2.48	.65
<3 months	12 (6.4)	1 (2.9)	10 (6.8)	1 (16.7)
3–12 months	17 (9.1)	4 (11.8)	13 (8.8)	0 (0.0)
>12 months	158 (84.5)	29 (85.3)	124 (84.4)	5 (83.3)
Missing values	28 (13.0)	6 (15.0)	21 (12.5)	1 (14.3)
Number of treatments in this clinic	215 (100.0)	40 (100.0)	168 (100.0)	7 (100.0)	χ2 (8) = 14.45	.07
1	123 (57.2)	29 (72.5)	93 (55.4)	1 (14.3)
2	53 (24.7)	9 (22.5)	40 (23.8)	4 (57.1)
3	24 (11.2)	1 (2.5)	22 (13.1)	1 (14.3)
>3	15 (6.9)	1 (2.5)	13 (7.7)	1 (14.3)
Missing values	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Type of trauma
Accident	5 (2.6)	2 (5.7)	3 (2.0)	0 (0.0)	χ2 (2) = 1.72	.42
Loss	21 (10.9)	5 (14.3)	15 (9.9)	1 (16.7)	χ2 (2) = 0.76	.68
Emotional abuse	27 (14.1)	6 (17.1)	20 (13.2)	1 (16.7)	χ2 (2) = 0.39	.82
Physical abuse / violence	62 (32.3)	10 (28.6)	51 (33.8)	1 (16.7)	χ2 (2) = 1.04	.59
Sexual abuse	45 (23.4)	2 (5.7)	43 (28.5)	0 (0.0)	χ2 (2) = 10.1	.006
Abuse not specified	26 (13.5)	3 (8.6)	22 (14.6)	1 (16.7)	χ2 (2) =0.93	.63
Childhood not specified	31 (16.1)	3 (8.6)	27 (17.9)	1 (16.7)	χ2 (2) = 1.82	.40
Neglect	9 (4.7)	0 (0.0)	9 (6.0)	0 (0.0)	χ2 (2) = 2.56	.28
Hospital experiences	4 (2.1)	1 (2.9)	2 (1.3)	1 (16.7)	χ2 (2) = 6.78	.034
Mobbing	12 (6.3)	3 (8.6)	9 (6.0)	0 (0.0)	χ2 (2) = 0.74	.69
Natural disaster	1 (0.5)	1 (2.9)	0 (0.0)	0 (0.0)	χ2 (2) = 4.51	.11
Other	28 (14.6)	9 (25.7)	18 (11.9)	1 (16.7)	χ2 (2) = 4.36	.113
Missing values	23 (10.7)	7 (17.5)	20 (11.9)	1 (14.3)

Note. χ² = Chi-square-test; df = Degrees of Freedom; SD = Standard Deviation; F = F-statistic from One-way Analysis of Variance (ANOVA); p < .05 indicates statistical significance.

3.2. Baseline comparison

At baseline, women reported significantly higher dissociation scores (4.0 vs. 3.1 points; p = .047; g = 0.24) as well lower depression scores (18.3 vs. 20.0 points; p = .031; g = 0.39) and less impaired social functioning (4.8 vs. 5.5 points; p = .042; g = 0.39) than men. In addition, women showed a higher internal locus of control (2.8 vs. 2.4 points; p = .031; g = 0.43), while men reported a higher external locus of control (3.8 vs. 3.5 points; p = .040; g = 0.41). All statistically significant differences were of small effect size. No statistically significant differences were observed for the ITQ total and subscales scores, the PID-5 subscale scores, the PCL-5, BSL-23, PHQ-15, GAD-7, ERQ subscales, and the other WHODAS-subscales. Details can be found in Appendix A.

3.3. Primary outcome

In general, patients reported statistically significant symptom improvements (all p<.001) with large effect sizes in the ITQ total score (η² = 0.17) and the DSO subscale (η² = 0.14) and medium effect size for the PTSD subscale (η² = 0.12), with a mean reduction of 12.2% in the overall symptom levels. Neither for the ITQ total score, nor for the PTSD or DSO subscales statistically significant time * gender effects were observed (p = .38–.91). Details are shown in Table 3. Results of the ITT analyses did not indicate any significant time * gender effects and thus confirmed the main results. In addition, across all outcomes the LMM results closely mirrored the ANOVA findings, confirming a significant main effect of Time, no main effect of Gender, and no Time × Gender interaction. This convergence provides reassurance that the observed treatment effects are robust to analytic approach despite group imbalance and missingness. The Bayes factor further validated the main results for the time effect (BF₁₀ = 1.44×10¹¹), whereas there was no evidence for a main effect of gender (BF₁₀ = 0.06) or time * gender interaction (BF₁₀ = 0.07). Details for the sensitivity analyses are presented in Appendix B.

Table 3.

Analysis of variance (ANOVAs) for changes of the International Trauma Questionnaire (ITQ) scores during treatment (T1-T2) stratified for men and women (per protocol analysis).

	Group	T1		T2		Time				Time * Gender
		M	SD	M	SD	delta	%	pa	η2	pa	η2
ITQ total	Men (n = 31)	37.7	(6.2)	33.3	(10.5)	4.4	11.7%	.006	0.06
	Women (n = 131)	37.4	(5.5)	32.8	(8.7)	4.6	12.3%	<.001	0.22
	Total (n = 162)	37.5	(5.6)	32.9	(9.0)	4.6	12.2%	<.001	0.17	.90	<0.01
ITQ PTSD	Men (n = 31)	19.1	(3.6)	17.2	(5.0)	2.0	10.3%	.024	0.04
	Women (n = 131)	19.3	(3.2)	17.3	(4.9)	1.9	10.1%	<.001	0.15
	Total (n = 162)	19.2	(3.3)	17.3	(4.9)	1.9	10.1%	<.001	0.12	.97	0.00
ITQ DSO	Men (n = 31)	18.5	(3.9)	16.6	(5.7)	1.9	10.1%	.024	0.03
	Women (n = 131)	18.2	(3.5)	15.5	(4.8)	2.7	14.7%	<.001	0.22
	Total (n = 162)	18.2	(3.6)	15.7	(5.0)	2.5	13.8%	<.001	0.14	.38	0.01

Note. ITQ = International Trauma Questionnaire; PTSD = post-traumatic stress disorder; DSO = disturbances in self-organization; M = mean score; SD = standard deviation; T1 = baseline score; T2 = end of treatment; delta = mean difference between baseline and end of treatment; % = percentage in distress levels (i.e. (T2-T1/T1)*100); effect sizes were classified as negligible (η² < 0.01), small (η² = 0.01–0.05), medium (η² = 0.06–0.14), or large (η² ≥ 0.14); ^aBonferroni-Holms corrected p-value.

To determine the optimal minimally important difference (MID) in this sample, both anchor-based and distribution-based methods were used. Anchor-based results showed mean ITQ total change scores of 9.3 points (95% CI: 6.9–11.7) for improved patients, 3.9 (95% CI: 1.8–6.1) for those with no change, and 1.2 (95% CI: –2.1–4.5) for those who worsened. Distribution-based estimates yielded SDC values between 5.7–6.8 and RCI values between 8.0–9.6, with higher scores consistently observed when based on the SD of change scores. To balance sensitivity and specificity, a cut-off of 7.0 points was selected as the MID, aligning with the lower bound of anchor-based estimates. Overall, 32.7% of participants showed symptom improvement exceeding the MID, with slightly higher rates in women (32.3%) than men (30.0%). The highest rate was observed in gender-diverse patients (60.0%). Symptom deterioration above the MID was reported by 3.1% of women, 10.0% of men, and 20.0% of gender-diverse participants (total sample: 4.9%). However, these gender differences were not statistically significant (χ²= 7.54, p = .114). Distribution-based MID results are detailed in Appendix B.

3.4. Secondary outcome

Overall, patients reported statistically significant symptom improvements with large effect sizes for trauma-related symptoms as measured by the PCL-5 (p<.001, η²= 0.17), depressive symptoms (p<.001, η²= 0.39), and anxiety (p<.001, η²= 0.22). Additionally, patients showed a statistically significant improvement in their ability to engage in emotional reappraisal, with a moderate effect size (p = .002, η²= 0.07). No significant changes were observed for dissociative symptoms (p = .29) or emotional suppression (p = .93). No significant time*gender effects were observed for any of the variables. For details, see Table 4.

Table 4.

Analysis of variance (ANOVAS) for changes of the PCL-5, DSS-4, PHQ-9, GAD-7 and ERQ scores during treatment (T1-T2) stratified for men and women (per protocol analysis).

	Group	T1		T2		Time				Time * Group
		M	SD	M	SD	delta	%	pa	η2	pa	η2
PCL-5	Men (n = 31)	58.9	13.6	49.1	18.9	9.8	16.7%	.004	0.06
	Women (n = 131)	56.0	11.0	47.6	16.2	8.4	15.0%	<.001	0.20
	Total (n = 162)	56.5	11.5	47.9	16.6	8.6	15.3%	<.001	0.17	.7	0.00
DSS-4	Men (n = 31)	3.1	2.6	2.9	2.6	0.2	5.6%	.72	0.00
	Women (n = 131)	3.9	2.4	3.6	2.6	0.4	9.7%	.086	0.02
	Total (n = 162)	3.8	2.4	3.4	2.6	0.3	9.1%	.29	0.01	.69	0.00
PHQ-9	Men (n = 31)	20.2	4.4	15.8	6.2	4.5	22.1%	<.001	0.17
	Women (n = 131)	18.2	4.4	13.9	5.1	4.3	23.7%	<.001	0.44
	Total (n = 162)	18.6	4.4	14.3	5.3	4.3	23.4%	<.001	0.39	.86	0.00
PHQ-15	Men (n = 31)	16.3	5.7	13.8	6.9	2.5	15.3%	<.001	0.07
	Women (n = 131)	17.1	5.2	15.2	5.5	1.9	11.1%	<.001	0.15
	Total (n = 162)	17.0	5.3	14.9	5.8	2.1	12.4%	<.001	0.16	.45	0.00
GAD-7	Men (n = 31)	16.5	3.5	13.7	5.0	2.8	17.0%	<.001	0.07
	Women (n = 131)	14.7	3.9	11.4	4.7	3.3	22.6%	<.001	0.31
	Total (n = 162)	15.1	3.8	11.8	4.8	3.2	21.4%	<.001	0.22	.57	0.00
ERQ reapr.	Men (n = 31)	3.2	1.3	3.4	1.4	0.2	7.0%	.44	0.00
	Women (n = 131)	3.5	1.3	4.2	1.3	0.8	21.8%	<.001	0.22
	Total (n = 162)	3.4	1.3	4.1	1.4	0.7	19.7%	.002	0.07	.08	0.02
ERQ suppr.	Men (n = 31)	5.1	1.2	5.3	1.2	0.2	3.8%	.48	0.00
	Women (n = 131)	4.7	1.4	4.5	1.4	−0.2	3.6%	.16	0.01
	Total (n = 162)	4.8	1.4	4.6	1.4	−0.1	2.3%	.93	0.00	.23	0.01
IE-4 IC	Men (n = 31)	2.5	0.9	2.6	1.3	0.1	4.0%	.52	0.00
	Women (n = 131)	2.8	1.0	3.2	1.0	0.4	14.3%	<.001	0.14
	Total (n = 162)	2.7	1.0	3.1	1.0	0.4	14.8%	.012	0.05	.16	0.01
IE-4 EC	Men (n = 31)	3.7	0.9	3.5	1.7	0.2	5.4%	.15	0.02
	Women (n = 131)	3.5	0.9	3.1	0.9	0.4	11.4%	<.001	0.09
	Total (n = 162)	3.5	0.9	3.2	0.9	0.3	8.6%	.007	0.05	.76	0.00

Note. PCL-5 = Posttraumatic Stress Disorder Checklist for DSM-5; DSS-4 = Dissociation-Tension Scale; PHQ-9 = Patient Health Questionnaire 9; GAD-7 = Generalized Anxiety Disorder-7; ERQ reapr = Emotion Regulation Questionnaire – Reappraisal Subscale; ERQ suppr. = Emotion Regulation Questionnaire – Suppression Subscale; IE-4 IC = Internal–External Locus of Control Short Scale–4 – Internal Control Subscale; IE-4 EC = Internal–External Locus of Control Short Scale–4 – External Control Subscale; M = mean score; SD = standard deviation; T1 = baseline score; T2 = end of treatment; % = percentage in distress levels (i.e. (T2–T1/T1)*100); effect sizes were classified as negligible (η² < 0.01), small (η² = 0.01–0.05), medium (η² = 0.06–0.14), or large (η² ≥ 0.14); ^a Bonferroni-Holms corrected p-value.

Regarding patients’ Health-Related Quality of Life (HRQOL), a statistically significant improvement in the WHODAS-12 total score was observed in the overall sample, with a large effect size (p<.001, η²= 0.24). When examining specific HRQOL domains, large effect sizes were found for increased participation in daily activities (p<.001, η²= 0.26) and social participation (p<.001, η²= 0.25). Medium effect sizes were observed for improvements in social functioning (p<.001, η²= 0.10), mobility (p<.001, η²= 0.08), and communication (p<.001, η²= 0.09), while no statistically significant improvement was found for self-care (p = .29). In terms of gender differences, no significant time-by-gender effects were found for any of the variables. For detailed results, see Table 5.

Table 5.

Analysis of variance (ANOVAS) for changes of WHODAS scores during treatment (T1-T2) stratified for men and women (per protocol analysis).

		T1		T2		Time				Time * Group
	Group	M	SD	M	SD	delta	%	pa	η2	pa	η2
Total score	Men (n = 31)	26.6	9.8	21.9	11.4	4.7	17.6%	<.001	0.08
	Women (n = 131)	25.7	7.1	20.1	8.6	5.5	21.5%	<.001	0.32
	Total (n = 162)	25.8	7.7	20.5	9.2	5.4	20.7%	<.001	0.24	.56	0.002
Mobility	Men (n = 31)	3.3	2.4	2.4	2.7	0.9	27.1%	.006	0.04
	Women (n = 131)	3.6	2.4	3.1	2.3	0.5	13.7%	.003	0.05
	Total (n = 162)	3.5	2.4	3.0	2.4	0.6	16.1%	<.001	0.08	.25	0.01
Life activities	Men (n = 31)	5.0	2.3	3.8	2.6	1.2	23.8%	.002	0.06
	Women (n = 131)	5.1	1.8	3.1	2.1	2.0	38.9%	<.001	0.41
	Total (n = 162)	5.1	1.9	3.2	2.2	1.8	36.0%	<.001	0.26	.06	0.02
Communcation	Men (n = 31)	4.7	2.2	4.2	2.3	0.5	11.3%	.08	0.02
	Women (n = 131)	4.5	1.7	3.7	2.1	0.8	18.5%	<.001	0.16
	Total (n = 162)	4.6	1.8	3.8	2.1	0.8	17.1%	<.001	0.09	.37	0.01
Soc participation	Men (n = 31)	6.5	1.7	5.2	2.0	1.3	20.3%	<.001	0.11
	Women (n = 131)	5.9	1.5	4.9	1.6	1.1	18.1%	<.001	0.26
	Total (n = 162)	6.0	1.5	4.9	1.7	1.1	18.5%	<.001	0.25	.46	0.00
Selfcare	Men (n = 31)	1.5	2.1	1.4	2.1	0.0	2.1%	.92	0.00
	Women (n = 131)	1.8	1.9	1.5	1.8	0.3	18.0%	.029	0.03
	Total (n = 162)	1.7	1.9	1.5	1.8	0.3	15.5%	.29	0.01	.39	0.01
Social functioning	Men (n = 31)	5.6	1.8	4.9	2.2	0.7	12.8%	.029	0.03
	Women (n = 131)	4.7	1.8	3.9	1.9	0.8	16.6%	<.001	0.13
	Total (n = 162)	4.9	1.8	4.1	2.0	0.8	15.8%	<.001	0.10	.86	0.00

3.5. Predictors of therapy outcome

To benchmark model performance, we used a naïve baseline predicting the mean ITQ score without covariates. This yielded MAEs of 8.37 (SD = 4.98) for men and 6.99 (SD = 5.33) for women, serving as lower-bound comparisons. XGBoost models improved upon these baselines, especially for women, where MAE dropped by 25% (to 5.24) and R² reached 0.40, with low error variability (SE < 0.10). For men, performance gains were more modest – MAE dropped by 21% (to 6.60), but R² remained low (0.10) with high variability (SD = 0.53), suggesting less stable predictions. For details, see Figure 1.

Figure 1.

Distribution of model performance metrics across 50 repeated cross-validation folds for the male and female subgroup. Left column: Mean absolute error (MAE); right column: coefficient of determination (R²). Dashed vertical lines indicate baseline performance (mean-only prediction). On average, the model outperforms the baseline in terms of MAE across both subgroups, with the largest and most consistent improvements in the female group. R² values are highest and most consistent in the female group and variable in the male subgroup, where several folds yield negative explained variance.

3.6. Feature importance

SHAP analyses revealed subgroup-specific patterns of predictor importance. In men, top predictors included sleep impairment, negative affectivity, somatization, inability to work, and self-efficacy. While broader trait psychopathology (e.g. disinhibition, depression) also appeared among the top features, their relative importance was lower compared to the female subgroup. In women, negative affectivity, psychoticism, baseline ITQ, age, dissociation, and social functioning emerged as the most influential predictors. SHAP values showed more consistent importance patterns across folds in women than in men. For details, see Figure 2.

Figure 2.

Cross-validated SHAP feature importance distributions across 50 repeated train-test splits for the male subgroup (top), and female subgroup (bottom). Boxplots display the distribution of mean absolute SHAP values per feature across all validation folds, indicating both the average magnitude and variability of each predictor's contribution to model output. Higher mean SHAP values reflect greater overall influence on the prediction of post-treatment ITQ scores. Results highlight consistent contributions of PID-5 trait domains (e.g. Psychoticism, Negative Affectivity) and trauma-related symptom measures (e.g. ITQ, DSS-4), with subgroup-specific variations such as the prominence of JSS – Jenkins Sleep Scale (T1), somatic symptoms and work disability in the male sample. These distributions provide a stability-aware estimate of feature relevance across resampling iterations.

3.7. Validation of SHAP predictors

To assess the predictive strength of the features identified via XGBoost and SHAP, gender-specific mixed linear models were calculated. Baseline ITQ was forced into every model to ensure that individual differences in initial symptom severity were always controlled for. This prevents spurious associations that could arise if predictors were simply capturing baseline level differences, making the predictor × time interactions interpretable as genuine mechanisms of change. All possible model combinations of 15 predictors (2¹⁵ = 32,768) were fitted. Of these, 4,902 converged for males and 3,156 for females; the remainder failed to converge. For both men and women, baseline ITQ was fixed at 1.00 across all converged models, while all other main effects averaged to zero with 95% CrIs spanning zero. This validates that baseline ITQ accounts for initial symptom level and confirms that interpretation should focus on predictor × time interactions, which capture differential change over time.

For male patients, several predictors showed consistent associations with change over time. Notably, External Control (IE-4) exhibited a robust negative effect (β = –0.96, SD = 0.33, 95% CrI [−1.34, –0.14], PIP = 1.00), indicating stronger external control was linked to stronger improvement trajectories. Conversely, several psychopathology dimensions were associated with worsening symptoms across time: Negative Affectivity (PID-5) (β = 3.79, SD = 0.93, 95% CrI [2.30, 4.55], PIP = 0.65), Antagonism (PID-5) (β = 4.10, SD = 1.70, 95% CrI [2.09, 8.11], PIP = 0.58), and Sleep Problems (β = 3.00, SD = 1.03, 95% CrI [0.80, 3.97], PIP = 0.10). By contrast, some predictors showed large estimated effects but low inclusion probabilities, such as Sick Leave (β = 8.13, PIP = 0.17), PHQ-9 Depression (β = 5.10, PIP = 0.12), and PHQ-15 Somatic Symptoms (β = –7.54, PIP = 0.11), indicating that although strong effects occasionally emerged, they were not consistently supported across models. Similarly, higher Age (β = –2.55, PIP = 0.11) and WHODAS – Getting Along with People (β = –1.59, PIP = 0.07) predicted more favorable changes in ITQ scores, yet their low inclusion probabilities temper confidence in these findings. Other predictors, including Pain (NAS), DSS-4, RS-13 Resilience, Psychoticism, and Disinhibition, had both small effects and low PIPs (0.01–0.08) with wide intervals spanning zero, suggesting practical irrelevance in our sample.

For female patients, several predictors consistently related to change over time. PID-5 Traits such as Negative Affectivity (β = 3.09, SD = 0.12, 95% CrI [2.93, 3.17], PIP = 1.00) and Psychoticism (β = 3.07, SD = 0.07, 95% CrI [2.97, 3.09], PIP = 1.00) were strongly linked with worse outcomes, while Antagonism (β = –0.50, SD = 0.06, 95% CrI [–0.54, –0.43], PIP = 1.00) and Disinhibition (β = –0.44, SD = 0.04, 95% CrI [–0.45, –0.30], PIP = 1.00) consistently indicate that higher baseline scores predicted more improvement across time. Older Age (β = 2.74, SD = 0.22, 95% CrI [2.39, 2.88], PIP = 1.00) and WHODAS Getting Along with People (β = 1.92, SD = 0.05, 95% CrI [1.89, 2.02], PIP = 0.98) as well as Dissociation (DSS-4; β = 1.80, SD = 0.04, 95% CrI [1.67, 1.81], PIP = 0.73) also predicted worse outcomes. Others, like Pain (NAS) (β = 1.73, PIP = 0.29), PHQ-15 Somatic Symptoms (β = –0.74, PIP = 0.03) and PHQ-9 Depression (β = 0.66, PIP = 0.01), had weak or inconsistent evidence despite sizable coefficients. Finally, predictors including Sick Leave, RS-13 Resilience, and External Control (IE-4) had both very low PIPs (<0.01) and wide intervals overlapping zero, indicating practical irrelevance in this sample.

Details are shown in Table 6.

Table 6.

Validation of predictors for outcome for male patients.

	Predictor (T1) × Time	PIP	β (SD)	95% CrI (coef.)	n models	% of 4,902 converged models
Men	ITQ – Total Score	1.00	−0.30 (0.13)	−0.51 to 0.05	4902	100.0
	IE-4 – External Control (T1)	1.00	−0.96 (0.33)	−1.34 to 0.14	4161	84.88
	PID-5: Negative Affectivity	0.65	3.79 (0.93)	2.30–4.55	2337	47.67
	PID-5: Antagonism	0.58	4.10 (1.70)	2.09–8.11	2269	46.29
	ASKU – General Self-Efficacy	0.45	−3.58 (1.59)	−4.63 to 1.50	2408	49.12
	Sick Leave (Unable to Work)	0.17	8.13 (3.40)	3.19–11.93	2455	50.08
	PHQ-9 Depression Severity	0.12	5.10 (2.74)	−0.31 to 7.30	2472	50.43
	PHQ-15 Somatic Symptom Severity	0.11	−7.54 (3.84)	−10.54 to 0.35	2489	50.78
	Age	0.11	−2.55 (0.59)	−3.06 to −1.02	2494	50.88
	JSS – Jenkins Sleep Scale	0.10	3.00 (1.03)	0.80–3.97	2490	50.80
	PID-5: Psychoticism	0.08	0.35 (1.33)	−1.07–2.48	2336	47.65
	WHODAS – Getting Along with People	0.07	−1.59 (0.70)	−3.22 to −0.26	2479	50.57
	RS-13 Resilience Scale	0.03	−0.76 (3.26)	−4.29 to 7.85	2479	50.57
	PID-5: Disinhibition	0.02	1.05 (0.75)	−0.61 to 2.23	2256	46.02
	DSS-4 – Dissociative Symptoms Scale	0.02	−0.44 (0.49)	−1.46 to 0.65	2449	49.96
	Pain – NAS Overall	0.01	1.37 (1.33)	−0.91 to 5.11	2474	50.47
	Predictor × Time	PIP	β (SD)	95% CrI (coef.)	n models	% of 3,156 converged models
Women	ITQ – Total Score	1.00	−0.72 (0.04)	−0.75 to −0.67	3156	100.0
	PID-5: Negative Affectivity	1.00	3.09 (0.12)	2.93–3.17	1361	43.1
	JSS – Jenkins Sleep Scale	1.00	−0.35 (0.26)	−0.75 to −0.19	1731	54.8
	PID-5: Antagonism	1.00	−0.50 (0.06)	−0.54 to −0.43	1190	37.7
	Age	1.00	2.74 (0.22)	2.39–2.88	1593	50.5
	PID-5: Psychoticism	1.00	3.07 (0.07)	2.97–3.09	1624	51.5
	PID-5: Disinhibition	1.00	−0.44 (0.04)	−0.45 to −0.30	1509	47.8
	WHODAS – Getting Along with People	0.98	1.92 (0.05)	1.89–2.02	1534	48.6
	DSS-4 – Dissociative Symptoms	0.73	1.80 (0.04)	1.67–1.81	1557	49.3
	Pain – NAS Overall	0.29	1.73 (0.08)	1.70–1.97	1548	49.0
	PHQ-15 Somatic Symptom Severity	0.03	−0.74 (0.13)	−0.76 to −0.75	1616	51.2
	PHQ-9 Depression Severity	0.01	0.66 (0.32)	0.40–1.07	1615	51.2
	ASKU – General Self-Efficacy	0.00	−0.36 (0.03)	−0.49 to −0.36	1576	50.0
	Sick Leave (Unable to Work)	0.00	0.07 (0.15)	0.04–0.73	1583	50.2
	RS-13 Resilience	0.00	0.10 (0.15)	−0.15 to 0.18	1559	49.4
	IE-4 – External Control	0.00	0.41 (0.08)	0.32–0.60	1410	44.7

4. Discussion

This study investigated the effectiveness of Dialectical Behavior Therapy for PTSD (DBT-PTSD) for women and men in a naturalistic sample of inpatients with complex post-traumatic stress disorder (cPTSD). Overall, the results demonstrated that DBT-PTSD significantly reduced both PTSD and DSO symptoms with large effect sizes, thus confirming its applicability in routine clinical settings. Importantly, no significant gender differences were observed in treatment outcomes across primary or secondary measures, indicating that DBT-PTSD is a gender-fair treatment approach for patients with trauma-related disorders.

At baseline, men and women showed similar symptom severity, with only small differences. In line with previous studies, women more often reported sexual violence (Olff, 2017; Tolin & Foa, 2006) and higher dissociation (Fullerton et al., 2001; Olff, 2017), while men showed greater social functioning impairment. However, contrary to the literature, men in our sample also reported higher depressive symptom severity than women. Additionally, contrary to our hypotheses and previous meta-analyses suggesting superior outcomes for women in trauma-focused therapies (Wade et al., 2016), our data revealed no statistically significant difference in mean symptom change between men and women during the intervention. In addition, there were no statistically significant gender differences in premature discontinuation of treatment. This challenges assumptions about gender-based responsiveness and suggests DBT-PTSD may be equally effective for men and women in structured inpatient settings.

One potential explanation for this discrepancy lies in differences in study design. While the structured and standardized nature of randomized controlled trials (RCTs) allows for causal interpretation and the evaluation of treatment efficacy, their narrow inclusion and exclusion criteria – designed to ensure relatively homogeneous samples – may not accurately reflect the diversity of individuals seeking psychotherapy in real-world practice (Smith, 2024). Furthermore, the highly controlled settings and standardized protocols used in RCTs may not reflect how treatments are implemented in everyday clinical practice and these discrepancies may limit the generalizability of findings in RCT to real-world settings (Fernainy et al., 2024). In contrast, naturalistic studies like ours better reflect clinical practice and enhance ecological validity, offering valuable insights into how treatments perform in real-world settings.

While a recent naturalistic study of inpatient DBT-PTSD reported lower effect sizes than earlier RCTs (Oppenauer et al., 2023), our sample showed large improvements in trauma symptoms and secondary outcomes (anxiety, depression, somatization, and borderline symptoms). These effects were slightly smaller than those reported in observational studies of trauma-focused psychodynamic therapy (PDT-TF) for inpatients with cPTSD (Lampe et al., 2024; Riedl et al., 2025), possibly due to higher baseline symptom severity in our sample. Unlike PDT-TF studies, which showed greater change on the DSO scale than on PTSD symptoms, our results were more balanced – likely reflecting the routine inclusion of exposure techniques in DBT-PTSD. Notably, DBT-PTSD was designed for individuals with PTSD and severe emotion regulation difficulties–symptoms particularly prevalent in survivors of sexual or physical assault or childhood abuse (Hyland et al., 2017; Yehuda & LeDoux, 2007). In our sample, this subgroup showed particularly strong outcomes. This indicates that while DBT-PTSD is effective for patients with various forms of traumatic experiences, it appears to be particularly effective for survivors of childhood abuse or sexual and physical assault. This aligns with recent developments such as Structured Exposure DBT (SE-DBT), an adaptation of DBT-PTSD aimed at individuals with complex trauma and emotional dysregulation who may not meet full PTSD criteria. SE-DBT combines DBT with trauma-focused and exposure-based elements and further incorporates compassion-focused therapy (Gilbert, 2014), and acceptance and commitment therapy (Hayes et al., 2006), offering a promising approach for treating complex trauma beyond traditional PTSD frameworks.

Due to the small number of gender minority (GM) participants (n = 7), they were excluded from main analyses. However, descriptive data suggested comparable treatment benefits to those seen in men and women. This is noteworthy given that GM individuals are disproportionately affected by trauma-related mental health issues (Feil et al., 2023; Ramos & Marr, 2023). These findings should be interpreted with caution, as the sample was small and lacked detailed subgroup identification. Moreover, traditional psychotherapies may reflect heteronormative assumptions and fail to address issues specific to GM individuals. Future research should prioritize inclusive recruitment and critically assess the suitability of standard treatments for this population.

Machine learning informed Bayesian Model Averaging revealed gender-specific predictors of treatment outcomes. For men, the clearest protective factor was external control, which consistently predicted more favorable treatment trajectories. This suggests that interventions enhancing external orientation–such as structured, goal-directed approaches or external sources of accountability–may play a particularly supportive role in men's recovery. At the same time, psychopathology-related traits such as Negative Affectivity and Antagonism robustly predicted symptom worsening in men, indicating that emotional lability, anxiousness and fear of separation (i.e. PID-5 negative affectivity subscale) as well as grandiosity, dishonestness and manipulative behavior (i.e. PID-5 antagonism subscale) may represent critical therapeutic targets.

In contrast, the predictive profile in women was more multifaceted. Negative Affectivity and Psychoticism were consistently linked to unfavorable outcomes, aligning with prior research emphasizing the detrimental role of high emotional reactivity and psychotic-like features in trauma-related treatment courses (Frost et al., 2025). Unexpectedly, Antagonism and Disinhibition–traits typically associated with maladaptive outcomes–were associated with improvement, raising questions about whether such features might facilitate engagement in trauma processing or align with specific DBT-PTSD mechanisms in women. Additional factors like age, social functioning, and dissociation also contributed to worse trajectories, supported by strong posterior inclusion probabilities. By contrast, some predictors that stood out for males (e.g. external control) were essentially irrelevant in females.

These gender-specific patterns underscore the need for gender-sensitive research in assessment and intervention. Since men are still underrepresented in trauma-focused therapy research there is a need to investigate gender-specific interventions for traumatized men, such as all-male stabilization groups with more focus on male-specific topics (Røberg et al., 2018). Taken together, these findings emphasize that structural and contextual factors – such as gender roles, socialization processes, and access to support – shape post-traumatic adaptation in ways that extend beyond symptom severity alone. Future research should aim to clarify mechanisms underlying these gendered pathways and test whether targeted, gender-sensitive adaptations can enhance treatment outcomes in DBT-PTSD and related interventions.

Several limitations must be acknowledged. First, the reliance on self-report measures may introduce response bias. Recent studies for example have found a trend for false positive cases, which was met by introducing clinical checks to the questionnaire (Shevlin et al., 2025). Since this development only took place after this study was conducted, it was not possible to include the newly introduced clinical checks and thus a potential risk for false positive inclusion of patients was given. Second, the small sample of male participants likely reduced the power to detect gender differences. Additionally, due to the very small number of GM individuals this subgroup had to be excluded from the main analyses. Third, the lack of follow-up data precludes conclusions about long-term treatment effects. Fourth, the internal consistency of the DSO subscale was lower than typically reported in prior research. While the ITQ generally demonstrates acceptable to good psychometric properties, our findings suggest that in this sample the DSO construct was less consistently endorsed and thus measurement precision for the DSO subscale may have been limited. Finally, the exploratory nature of the machine learning analysis limits its generalizability, though internal validation was rigorous. A key strength of this study is its two-step analytic strategy. First, feature selection with gradient-boosted trees (XGBoost) and SHAP values provided a transparent, data-driven way to identify candidate predictors, capturing potential non-linearities and interactions. Second, these predictors were validated in linear mixed models, with Bayesian Model Averaging accounting for model uncertainty and yielding interpretable posterior inclusion probabilities (PIPs) and credible intervals. This approach allows robust identification of relevant predictors and insights into mechanisms of change. Nevertheless, limitations remain. Collinearity may cause predictors to ‘compete’ for inclusion, inflating uncertainty around correlated constructs (e.g. Antagonism and Negative Affectivity). In addition, the models assume linear effects and may not capture more complex dynamics, while weaker predictors with low PIPs may be sample-specific, limiting generalizability. Nonetheless, the findings offer a promising first step toward understanding gender-differentiated mechanisms in post-traumatic symptom trajectories and lay the groundwork for future research in larger and more diverse clinical cohorts.

Despite these limitations, this study is the first to assess gender-specific effectiveness of DBT-PTSD in individuals with cPTSD under routine conditions. The integration of predictive modeling offers novel insights into differential treatment mechanisms and highlights important targets for gender-sensitive clinical tailoring.

In conclusion, this study provides evidence that DBT-PTSD is effective in reducing cPTSD symptoms in both men and women under routine inpatient conditions. The absence of significant gender differences in symptom reduction suggests that DBT-PTSD, when delivered as part of a structured multimodal inpatient program can be broadly effective. Future studies should compare the effectiveness and efficacy of recent developments such as SE-DBT compared to DBT-PTSD in treating traumatized patients who do not fulfill the diagnostic criteria for cPTSD. Additionally, our findings emphasize the need for gender-specific clinical adaptations. For male patients, who often present with more externalizing symptoms and less emotional disclosure, trauma-focused interventions might benefit from greater integration of functional, somatic, and occupational dimensions. The use of all-male stabilization groups, a stronger emphasis on psychoeducation about emotional processing, and attention to masculinity-related treatment barriers may increase engagement and efficacy. The descriptive improvements seen in gender-diverse patients also underline the importance of inclusive treatment frameworks, although conclusive evidence remains lacking due to small sample sizes.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The datasets presented in this article are not readily available because of the vulnerability of the study sample. Participants of this study did not agree for their data to be shared publicly, so supporting data is not available.

Supplemental Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/20008066.2025.2580210