Study design and participants

For an extensive overview of the study design, we refer to previous publications.^{Reference Leuzinger-Bohleber, Hautzinger, Fiedler, Keller, Bahrke and Kallenbach25–Reference Beutel, Krakau, Kaufhold, Bahrke, Grabhorn and Hautzinger28} The study included N = 252 participants with chronic depression. Participants had to be between 21 and 60 years old, score ≥17 on the Beck Depression Inventory – II (BDI-II),^{Reference Hautzinger, Keller and Kühner32} receive a score ≥9 points on the Quick Inventory of Depressive Symptomatology Clinician Rating (QIDS-C)^{Reference Rush, Trivedi, Ibrahim, Carmody, Arnow and Klein33} and meet the diagnostic criteria for major depression or dysthymia. If participants used antidepressant medication, they had to have been on a stable dosage for ≥4 weeks. Exclusion criteria comprised psychotic illness, substance misuse, dementia, borderline, schizoid, paranoid and antisocial personality disorder, acute and/or chronic physical illness and acute suicidality. Written informed consent was obtained from all participants. They were informed about the trial treatments (long-term CBT or PAT) and could choose to be randomised to CBT or PAT or to select their preferred treatment. The study's four-arm design included randomised or preferred CBT or PAT respectively. Participants were randomised by an independent statistics centre, generating separate random allocation sequences for the four study sites. The study sites were university-affiliated research institutions with out-patient clinics (university hospitals in Frankfurt, Mainz, Berlin and Hamburg, Germany) that coordinated diagnostics, training workshops and patient referral. Session frequency and treatment duration were not predetermined. For sample size calculation, see Beutel et al^{Reference Beutel, Leuzinger-Bohleber, Rüger, Bahrke, Negele and Haselbacher26} and Leuzinger-Bohleber et al.^{Reference Leuzinger-Bohleber, Hautzinger, Fiedler, Keller, Bahrke and Kallenbach25} The study was registered (Clinical Trial Register ISRCTN91956346). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human patients were approved by the Ethics Committee of the Physician Board of Rhineland-Palatinate, Mainz, Germany (Ref: 837.124.075659).

Interventions

Psychoanalytic therapy (PAT) for depression has been well described.^{Reference Leuzinger-Bohleber, Manfred, Fischmann and Beutel15} Study therapists were trained in the empirically validated manual for the treatment of chronic depression.^{Reference Taylor29} The study manual integrates the treatment of people with chronic depression with research on embodied memories and early trauma.^{Reference Leuzinger-Bohleber, Manfred, Fischmann and Beutel15} It is recognised that childhood trauma overwhelms the individual's capacity to maintain a minimal sense of safety and destroys trust in the availability of reliable and empathic others as well as one's own self-agency. Procedural, embodied memories developed before full maturation of memory structures are understood to unconsciously determine the individual's emotions, thoughts and actions in relationships in the present. Unprocessed, unconscious conflicts of the past can be repeated and understood in the transference and successively transferred into a healing process.

Cognitive–behavioural therapy (CBT) for depression is based on the work of Beck, Lewinsohn and others, and is integrated into a widely used and well-accepted treatment protocol in Germany.^{Reference Hautzinger30} CBT therapists used five modules (problem analysis, goals, psychoeducation and rationale for treatment; behavioural activation and increasing pleasant activities; cognitive interventions to restructure basic assumptions and schemata; social skill training, problem-solving and stress management; maintenance and relapse prevention). From a cognitive–behavioural view, threat-related information processing, heightened emotional reactivity and disrupted reward processing have been formulated as treatment targets in the context of chronic depression and childhood trauma.^{Reference McLaughlin, Colich, Rodman and Weissman34}

Compared with CBT, PAT scored higher on items characteristic of psychodynamic interpersonal therapy (e.g. feelings and perceptions linked to past experiences; focus on the patient–therapist relationship) with moderate effect sizes (0.35–0.74), whereas CBT was rated higher (0.30–0.84) on cognitive–behavioural items (e.g. focus on irrational belief systems; teaching specific techniques). The interrater reliability was high (intraclass correlation coefficient >0.85).^{Reference Leuzinger-Bohleber, Hautzinger, Fiedler, Keller, Bahrke and Kallenbach25} A difference in total session numbers between CBT and PAT may be considered as an intrinsic feature of the treatment approaches^{Reference Benecke, Huber, Staats, Zimmermann, Henkel and Deserno35} regulated by German guidelines, with CBT comprising up to 80 and PAT up to 300 sessions covered by insurance.^{Reference Dieckmann, Dahm, Streeck-Fischer, Neher and Schmidt36}

Instruments

Depressive symptoms were measured yearly over the course of 5 years using the Beck Depression Inventory – II (BDI-II).^{Reference Hautzinger, Keller and Kühner32} The instrument incorporates 21 statements rating the severity of depressive symptoms during the past 2 weeks on a 4-point Likert scale (0–3). Higher values denote higher symptom load. The German language version of the BDI-II has shown good reliability, sensitivity to change and validity.^{Reference Kühner, Keller, Schricker, Beddig, Huffziger and Timm37} Good internal consistency was replicated in our sample (ω_baseline = 0.82).

Childhood trauma was measured with the Childhood Trauma Questionnaire (CTQ).^{Reference Bernstein, Ahluvalia, Pogge and Handelsman38,Reference Klinitzke, Romppel, Häuser, Brähler and Glaesmer39} A global scale and six subscales assess self-rated emotional abuse and neglect, physical abuse and neglect, sexual abuse, and family inconsistency, specific to the German version^{Reference Maercker and Bromberger40}. Family inconsistency refers to a subjective appraisal of instability, unsafety and unpredictability in the home and the relationship with the main family members more generally (e.g. ‘I was scared that my family could break apart at any time’ or ‘[My caregivers] or other family members were unpredictable’). Three additional items assess minimising responses on a seventh scale (e.g. ‘I had the best family in the world’). Participants are asked about the time of their upbringing and to rate each of the 31 items (e.g. ‘I had enough to eat’) on a 5-point Likert scale (ranging from 1 = never to 5 = very often). The German version of the abuse and neglect scales have been validated.^{Reference Klinitzke, Romppel, Häuser, Brähler and Glaesmer39} We found good internal consistency for the total score (ω = 0.95), emotional abuse (ω = 0.89), emotional neglect (ω = 0.92), physical abuse (ω = 0.88), sexual abuse (ω = 0.92), inconsistency experiences (ω = 0.85) and minimisation (ω = 0.80) but low internal consistency for physical neglect (ω = 0.65), which is in line with previous findings.^{Reference Klinitzke, Romppel, Häuser, Brähler and Glaesmer39,Reference Ernst, Beutel, Zwerenz and Krakau41} Since the family inconsistency subscale is not frequently used, we examined the proposed seven-factor structure using confirmatory factor analysis, resulting in acceptable fit (comparative fit index CFI = 0.98, Tucker–Lewis index TLI = 0.98, root mean square error of approximation RMSEA = 0.05, standardised root mean squared residual SRMR = 0.08) with all factor loadings ≥0.40 on the intended factors except for the physical neglect item ‘Someone brought me to a physician when I needed it’. We provide prevalence rates for individuals reporting at least moderate levels of childhood abuse and neglect, using the cut-offs initially proposed by Bernstein et al,^{Reference Bernstein, Ahluvalia, Pogge and Handelsman38} which have been widely adopted.^{Reference Häuser, Schmutzer, Brähler and Glaesmer42}

Statistical analyses

All analyses were conducted in R (R Core Team, 2023, version 4.3.3 for Windows). Average levels of the reported CTQ values were compared between therapy types using Student's t-tests or Welch's t-tests, as appropriate. Proportions of at least moderate abuse and neglect experiences were compared using χ²-tests or Fisher's exact tests, as appropriate. Our hypotheses on the effect of the interaction between therapy type and childhood trauma on changes in depressive symptoms were tested using a linear mixed-effects model approach to account for repeated measurements of each participant. The model included a participant-specific Gaussian random intercept. We accounted for initial depression severity and varying treatment doses by including the sample mean-centred baseline depression (BDI-II) score and total therapy dose (session number) as fixed covariates. Therapy type, childhood trauma, time and their respective interactions were incorporated as fixed effects. A three-way interaction between therapy type, childhood trauma and time was included to evaluate our central hypothesis. Time was assessed as years after treatment start. Therapy type was coded as a factor variable (0 = CBT; 1 = PAT). The modelling approach uses all available data (i.e. not only completers) and is valid under the missing at random (MAR) assumption. Following recommendations,^{Reference Pek and Flora43} we report the unstandardised regression coefficients and report 95% confidence intervals alongside P-values. We conducted graphical post hoc analyses of significant interactions to interpret the slopes.