Dissociation predicts a flatter response to trauma-focused PTSD therapy. It does not disqualify the patient
- Hypothesis-driven post hoc analysis of a German multicentre randomised trial (n = 193 women with childhood-abuse-related PTSD; 98 assigned to DBT-PTSD, 95 to cognitive processing therapy). With baseline PTSD severity in the model, pretreatment dissociation predicted smaller improvement on the clinician-administered CAPS-5: duration F(1, 186) = 4.15, p = 0.043; peak intensity F(1, 186) = 5.65, p = 0.018.
- Both treatment by dissociation interactions were null (duration F(1, 186) = 1.34, p = 0.249; intensity F(1, 186) = 0.61, p = 0.436). That is no evidence of a protocol-specific effect, which at these values is not the same as evidence that none exists. Partial correlations with baseline severity controlled: CPT r = -0.23 (p = 0.028) and r = -0.24 (p = 0.019); DBT-PTSD r = -0.07 (p = 0.501, not significant) and r = -0.20 (p = 0.049).
- Dissociation correlated 0.44 with baseline PTSD severity (p < 0.0001), and higher baseline severity predicted greater gain. That collinearity can mask the dissociation effect in any analysis that does not adjust for severity, which is the authors' reading of why an earlier meta-analysis (Hoeboer et al., 2020) found no significant correlation.
- Design limit: patients were randomised to therapy, not to dissociation level, so the dissociation-outcome link is observational and cannot establish cause. The authors say so explicitly. Dissociation was captured by the 21-item self-report DSS only, and no participant was assessed for a dissociative disorder diagnosis.
A patient depersonalises mid-exposure. She is in the room and she is not. Whether that costs her anything measurable has been argued for a decade without a well-powered answer, and this trial supplies one from a sample large enough and severe enough to matter.
What the model found once severity was in it
The data come from the multicentre trial of Bohus and colleagues comparing DBT-PTSD with cognitive processing therapy in 193 women whose index trauma was childhood sexual or physical abuse. Diagnosis and outcome ran through the CAPS-5, administered by experienced clinical psychologists blinded to allocation. Dissociation was self-reported on the 21-item Dissociation Tension Scale, scored both as mean duration across the past week and as peak intensity. The sample was severe: mean baseline CAPS-5 40.4 ± 9.94, mean DSS duration 24.0 ± 15.85, and 48.2% carried a co-occurring borderline personality disorder diagnosis on the IPDE.
Both therapies worked. CAPS-5 fell from 40.96 ± 8.95 to 26.41 ± 16.04 under CPT and from 39.93 ± 10.84 to 20.56 ± 15.81 under DBT-PTSD. Inside that overall gain, the general linear model (overall F(6, 186) = 4.30, p = 0.0004) showed both dissociation terms pulling against improvement once CAPSpre was entered. The mechanism the authors propose for the earlier null results is worth reading twice: dissociation and baseline severity correlate at 0.44, and severity is itself a positive predictor of change, so in unadjusted analyses the two effects cancel. On the authors' reading, the apparent contradiction in this literature is plausibly a modelling artefact rather than a disagreement about reality.
The second model is the clinically suggestive one, and the authors label it exploratory. A reliable early drop in dissociation over the first three months, defined by the Jacobson and Truax criterion as at least 12.43 points of duration and 1.30 points of intensity, predicted further CAPS-5 improvement across the following year (F(1, 185) = 12.18, p = 0.0006), holding baseline severity, baseline dissociation and the early PTSD change constant. The patients who achieved that drop had started far higher than those who did not: duration 49.3 ± 18.8 versus 24.7 ± 14.7 (t = 4.46, p < 0.001), intensity 4.12 ± 1.63 versus 2.86 ± 1.63 (t = 3.05, p = 0.003). Most of them reached symptomatic remission.
For your practice
Read the direction of this finding carefully, because it is easy to misread in both directions. It is not a licence to withhold trauma-focused therapy from dissociative patients. Not one of the 193 participants deteriorated reliably, the heaviest dissociators who shifted early did best of all, and the authors state plainly that attenuated efficacy is no argument for exclusion. Equally, it is not a reason to treat a high DSS score as noise. A statistically significant negative predictor with a partial correlation near -0.2 to -0.3 is a real cost to a real patient's outcome.
The practical move is measurement at three months rather than triage at intake. Baseline dissociation tells you the size of the headwind; the change in dissociation over the first quarter of treatment tells you whether the therapy is going to land. If depersonalisation, derealisation or memory gaps are still running at intake levels after twelve weeks of exposure or cognitive work, that is signal, and the exploratory model suggests the following year will bring less than you hoped. What to do about it is the question this trial does not answer: it never manipulated dissociation, and the authors name treating it directly as the open question that rigorously designed trials still have to settle.
Note also what was not measured. The DSS is a self-report symptom scale, not a diagnosis, and this trial never established how many participants met criteria for a dissociative disorder. A high score names a treatment obstacle. It does not name a condition, and nothing here supports converting a questionnaire number into a diagnostic label.
In this post hoc analysis baseline dissociation predicted a smaller gain, not a failed treatment: nobody in the trial reliably worsened, and the heaviest dissociators whose symptoms shifted early gained the most.
Randomisation was to therapy, not to dissociation level, so the association reported here is observational and cannot establish cause, as the authors state. Dissociation was measured only by the self-report DSS, with no diagnostic interview for dissociative disorders, so the proportion of participants carrying such a diagnosis is unknown. External validity is narrow: cis-women with childhood abuse as index trauma, unusually high dissociation scores, two specific face-to-face protocols, and 32.1% dropout.