Features of dissociation differentially predict antidepressant response to ketamine in treatment-resistant depression

Ketamine, an FDA-approved anaesthetic with dissociative effects at subanesthetic doses, produces rapid antidepressant effects in patients with major depressive disorder (MDD) and bipolar depression. Dissociation here refers to transient detachment from self or surroundings and can include depersonalization, derealization and amnesia; at typical clinical ketamine doses these phenomena are usually mild and short-lived. Prior studies have produced mixed findings about whether ketamine-induced dissociation is associated with antidepressant response: some small studies found no relationship, others found correlations with psychotomimetic symptoms, and an earlier report from the investigators observed that increased dissociative symptoms correlated with antidepressant improvement at 230 minutes and seven days post-infusion. These inconsistencies left unresolved whether specific dimensions of dissociation might uniquely predict clinical benefit. Niciu and colleagues designed the present follow-up study to test whether empirically derived subdimensions of the Clinician-Administered Dissociative States Scale (CADSS)—derealization, depersonalization, and amnesia—are differentially associated with subsequent antidepressant response to a single subanesthetic ketamine infusion. Using a larger pooled sample (n = 126) drawn from three clinical studies, the investigators hypothesised that greater dissociative symptoms would predict greater reductions in depressive symptoms and that particular CADSS subscales would show unique predictive relationships.

The analysis pooled data from 126 treatment-resistant depressed inpatients (ages 18–65; 84 with MDD, 42 with bipolar disorder) drawn from three related protocols conducted at the NIMH Clinical Research Center: Ketamine-Bipolar (Ket-BD, n = 39), Ketamine-Riluzole (Ket-Riluzole, n = 52), and Ketamine-MOA (Ket-MOA, n = 35). All participants were experiencing a current major depressive episode of at least moderate severity (screen thresholds of ≥20 on the 10-item MADRS or ≥18 on the 17-item HAM-D) and met treatment-resistance criteria by a modified Antidepressant Treatment History Form. Subjects were medication-free for at least two weeks prior to infusion (five weeks for fluoxetine), except bipolar patients who were maintained on therapeutic lithium or valproate. A single subanesthetic ketamine infusion (0.5 mg/kg over 40 minutes) was given; Ket-MOA and Ket-BD used double-blind saline-controlled designs while Ket-Riluzole was open-label. Depressive symptoms were assessed with the 17-item Hamilton Depression Rating Scale (HAM-D). Dissociative symptoms were measured with the 19-item CADSS, scored 0–4 per item. The CADSS has been previously partitioned into three subscales: amnesia (2 items), depersonalization (5 items), and derealization (12 items); CADSS change scores were computed from baseline (60 minutes pre-infusion) to 40 minutes post-infusion, the time of maximal dissociative effects. Antidepressant outcome was operationalised as percent change in HAM-D at 230 minutes, Day 1 and Day 7 post-infusion. To test the psychometric structure of the CADSS subscales, a confirmatory factor analysis for ordered categorical indicators was performed and evaluated using fit indices including the root mean square error of approximation (RMSEA), comparative fit index (CFI), and Tucker-Lewis index (TLI). For clinical prediction, the investigators fitted a general linear model with maximum likelihood estimation and repeated measures (compound symmetry covariance structure) predicting HAM-D percent change across the three timepoints. Fixed effects included time, study, CADSS score (total or subscale), and their interactions; significant interactions were probed via simple-slope comparisons by study and timepoint. Analyses were conducted in Mplus v7 and SAS v9.4, with two-tailed significance set at p ≤ .05. Because demographic features differed across the three study samples, study was included as a covariate in analyses.

Demographic and clinical characteristics differed across the three study samples, prompting inclusion of study as a covariate in all models. The confirmatory factor analysis of CADSS items at 40 minutes post-infusion (n = 126) supported the proposed three-factor structure: RMSEA = .06 (95% CI .042–.077), CFI = .97, and TLI = .96, indicating acceptable to good model fit by conventional criteria. Four mixed models were run predicting percent change in HAM-D using CADSS total score and each of the three CADSS subscales (derealization, depersonalization, amnesia). For CADSS total score there was a significant three-way interaction of CADSS total × study × time (F(4,204) = 2.70, p = .03). Follow-up simple-slope testing showed that greater CADSS total score predicted greater reduction in HAM-D at Day 7 within the Ket-MOA study (B = -0.62, SE = 0.30, t(204) = -2.09, p = .04), and this slope differed from the Ket-BD study (comparison t(204) = -1.99, p = .048). Across all studies, the CADSS total score–HAM-D relationship was negative but not statistically significant at 230 minutes (B = -0.25, SE = 0.14, t = -1.70, p = .09) or Day 1 (B = -0.10, SE = 0.14, t = -0.72, p = .47). The Derealization subscale produced a very similar pattern: a significant three-way interaction (F(4,204) = 2.73, p = .03) and a significant negative association with HAM-D percent change at Day 7 in the Ket-MOA study (B = -1.55, SE = 0.65, t(204) = -2.37, p = .019). That Ket-MOA Day 7 slope differed significantly from the Ket-BD and Ket-Riluzole slopes in pairwise comparisons. Like the total score, derealization was not significantly related to HAM-D percent change at 230 minutes (B = -0.52, SE = 0.28, t = -1.89, p = .06) or Day 1 (B = -0.29, SE = 0.27, t = -1.04, p = .30) when pooling across studies. In contrast, the Depersonalization subscale showed a consistent relationship across studies and timepoints: higher depersonalization predicted greater antidepressant response (i.e. larger reductions in HAM-D percent change) with B = -0.79, SE = 0.38, t(118) = -2.07, p = .04; this effect did not interact with study or time. The Amnesia subscale was unrelated to HAM-D percent change in all analyses.

Niciu and colleagues interpret these findings as an extension of their earlier report: specific dissociative experiences during ketamine infusion—particularly depersonalization and, to a more limited extent, derealization—are modest predictors of subsequent antidepressant response. Depersonalization showed a consistent negative association with HAM-D percent change across all studies and timepoints, making it less likely to be a spurious result. Derealization's association was observed only at Day 7 and only in the Ket-MOA sample, and the investigators acknowledge the possibility that this isolated finding could represent a Type I error. The authors discuss methodological differences that might account for between-study variation. For example, Ket-BD participants were maintained on mood stabilisers (lithium or valproate) whereas Ket-MOA participants were unmedicated, and Ket-BD subjects tended to have longer and more severe courses of illness. A recent meta-analysis cited by the authors suggests bipolar disorder patients may be less prone to dissociation than patients with other disorders, although the present pooled analysis did not find significant diagnostic (MDD vs BD) differences in dissociation, implying diagnosis alone does not fully explain the study-to-study variability. Neurobiological interpretation is offered cautiously. The investigators propose that depersonalization may index neurobiological processes related to ketamine's antidepressant mechanism—potentially glutamate receptor modulation leading to altered glutamatergic tone, downstream circuit-level changes, and shifts in default mode network connectivity—while acknowledging that off-target effects cannot be dismissed. They also note that this study provides the first reported validation of the CADSS subscales proposed by Bremner and colleagues in an actively depressed MDD/BD population, supporting the instrument's psychometric structure for capturing distinct aspects of ketamine-induced dissociation. Key limitations highlighted by the authors include the potential for inadequate blinding—patients and staff may detect active ketamine due to its psychoactive and sympathomimetic effects, which could inflate observed antidepressant responses—use of combined MDD and BD datasets which might underrepresent ketamine's efficacy in MDD alone, and the possibility of Type I errors in secondary or exploratory analyses. The investigators conclude that future a priori studies are required to confirm these observations. Pragmatically, they suggest that if time is limited, clinicians and researchers administering ketamine might focus on the Depersonalization subscale of the CADSS as a predictor of subsequent antidepressant response, while continuing to investigate underlying mechanisms and potential confounds.