How do insomnia and anxiety change separately after ketamine in treatment-resistant depression?

Three symptom domains measured at one acute time point

The paper asks whether ketamine's short-term reach extends beyond mood, and whether anxiety and insomnia improve as one process or as partly separate ones. Seventy-one inpatients with treatment-resistant depression (TRD) — 46 with major depressive disorder and 25 with bipolar depression — were rated at baseline and again 24 hours after a single intravenous infusion of 0.5 mg/kg ketamine. All post-treatment values therefore describe acute post-infusion ratings, not sustained outcomes.

Each domain had its own instrument, and the scales are not interchangeable. Depression was scored with the Montgomery–Åsberg Depression Rating Scale (MADRS, 10 items, total 0–60). Anxiety was scored with the Hamilton Anxiety Rating Scale (HAM-A, 14 items, total 0–56). Insomnia was not measured with a dedicated insomnia scale: it was indexed by the three sleep items of the 17-item Hamilton Depression Rating Scale (early, middle and late insomnia, each 0–2, summed 0–6). That 0–6 range matters for reading every insomnia number below, because a change of about one point is a large share of a very short scale.

All three domains fell, but insomnia by a small absolute margin

Separate linear mixed-effects models, each with time as a fixed effect and a random intercept per participant, showed significant pre–post reductions in every domain. MADRS fell by an estimated 11.72 points (SE 1.26; 95% CI –14.21 to –9.23; t(140) = –9.30) from a mean baseline of 33.07. HAM-A fell by 7.87 points (SE 0.82; 95% CI –9.49 to –6.26; t(140) = –9.64) from a baseline of 21.76. Insomnia fell by 0.89 points (SE 0.24; 95% CI –1.36 to –0.42; t(140) = –3.74) from a baseline of 2.58. All three survived Holm correction across the three outcomes.

The individual counts show the same ordering. HAM-A decreased in 63 of 71 patients (88.7%), MADRS in 57 of 71 (80.3%), and insomnia in 42 of 71 (59.2%); 17 patients showed no insomnia change and 12 worsened. The authors describe the depression and anxiety improvements as large and the insomnia improvement as modest. Part of that modesty is structural: six participants entered with an insomnia score of 0 and 17 with a score of 1, leaving little room to fall on a 0–6 scale. A diagnosis-adjusted sensitivity analysis found no significant diagnosis or time-by-diagnosis effect for any outcome, so the models gave no evidence that MDD and bipolar participants changed differently.

Anxiety change tracked depression change closely

Anxiety and depression were associated at both time points and moved together over the infusion. Baseline anxiety correlated with depression at Pearson r = 0.39, and the post-ketamine correlation was r = 0.57 (Holm-adjusted P < 0.001 for both). The change scores were more tightly coupled still: ΔHAM-A versus ΔMADRS had Spearman ρ = 0.61 (Holm-adjusted P < 0.001). In a simple linear regression, ΔMADRS predicted ΔHAM-A with β = 0.41 (SE 0.06; 95% CI 0.29–0.53; R² = 0.40, P < 0.001).

The authors note that anxiety and depression commonly covary across pharmacological and non-pharmacological treatments, so this coupling should not be read as specific to ketamine. What the data support is a description: in this sample, patients whose mood scores dropped most also tended to show the largest anxiety drops.

Insomnia stayed linked to anxiety but not to its change

Insomnia behaved differently in the two kinds of analysis. Cross-sectionally, insomnia severity correlated with anxiety at baseline (Spearman ρ = 0.36, Holm-adjusted P = 0.014) and again post-ketamine (ρ = 0.36, Holm-adjusted P = 0.014), but not with depression at baseline (ρ = 0.20, Holm-adjusted P = 0.36). Partial correlations controlling for depressive symptoms kept the insomnia–anxiety link significant at both points (baseline ρ = 0.32, P = 0.008; post-ketamine ρ = 0.29, P = 0.014).

Longitudinally, that link did not carry into change. ΔINS versus ΔMADRS had Spearman ρ = 0.13 (Holm-adjusted P = 0.69) and ΔINS versus ΔHAM-A had ρ = 0.14 (Holm-adjusted P = 0.69). The regression of ΔINS on ΔMADRS was likewise non-significant (β = 0.03; 95% CI –0.02 to 0.07; R² = 0.02, P = 0.26). Results held when using Kendall's tau-b, and sensitivity analyses removing sleep-related items from the MADRS and HAM-A did not change direction, magnitude or significance. The distinction is the paper's central point: insomnia and anxiety were related at a given moment, yet their short-term movements after ketamine were largely independent.

Baseline insomnia severity predicted larger absolute drops

Greater baseline insomnia severity was associated with larger absolute reductions in insomnia scores. The authors flag this as clinically interesting but caution that it may partly reflect floor effects, regression to the mean, or simply more room to improve, since participants were recruited on depressive rather than insomnia severity. They also treat it as a hypothesis for stratified replication rather than a validated predictor.

Exploratory polysomnography, recorded on the night before and the night after infusion in all 71 participants, suggested short-term increases in total sleep time, slow-wave sleep and sleep efficiency, plus reduced sleep latency. The authors report that improvements in objective sleep measures appeared mainly in patients with subjective insomnia improvement, and they interpret the objective changes as possibly reflecting glutamatergic modulation and synaptic plasticity rather than as proof of a mechanism. These sleep-physiology results are exploratory and are not the basis for the correlation findings above.

Why insomnia may follow its own course

The authors offer an interpretation rather than a demonstrated mechanism: insomnia in some patients with TRD may reflect a comorbid condition with partly distinct neurobiological substrates — chronic hyperarousal, HPA-axis dysregulation, heightened sympathetic or orexin activity, and altered prefrontal-limbic circuitry — instead of being only a symptom of depression. On that view, a drug acting on mood and anxiety need not move sleep by the same amount or on the same timetable.

They also point to sample heterogeneity as a possible explanation for the weak coupling. Patients were screened for obstructive sleep apnea with the STOP-Bang questionnaire, but no diagnostic polysomnography was performed to confirm or exclude apnea, and comorbid sleep disorders such as circadian rhythm disorders, restless legs syndrome or periodic limb movement disorder were not systematically screened. The authors suggest stronger insomnia–mood coupling might appear in samples where insomnia is more tightly tied to depressive pathology, and greater dissociation in samples including primary sleep disorders.

What the 24-hour window does and does not support

Several boundaries sit directly on these results. The insomnia measure is a three-item proxy drawn from a depression scale; it does not capture hypersomnia, sleep quality or clinically diagnosed insomnia, so the findings describe HAM-D insomnia symptoms rather than the full spectrum of sleep disturbance. The design is a single-arm, pre–post secondary analysis, so the changes are associations with ketamine treatment, not evidence of ketamine-specific or mechanistically distinct effects. Insomnia subtypes could not be stratified because nearly all patients had overlapping symptoms.

The timing is equally decisive. Clinical ratings were taken 24 hours after infusion, so all outcomes are acute; the authors explicitly frame the sleep findings as acute changes in clinician-rated insomnia symptoms rather than evidence of a sustained insomnia phenotype. The exploratory improver-versus-non-improver classification, which used linear discriminant analysis and logistic regression, stratified patients by post-treatment change and is presented as a descriptive comparison, not as an independent predictive subgroup.

The practical split is between coupling and co-occurrence

For readers deciding what to take from this paper, the useful distinction is not whether insomnia improved — it did, by 0.89 points on a 0–6 scale in 42 of 71 patients — but how that improvement related to everything else. Anxiety and depression changed as a coupled pair (Δ correlation ρ = 0.61; ΔMADRS explaining 40% of ΔHAM-A variance), while insomnia change was statistically unrelated to change in either domain (ρ = 0.13 and 0.14, both Holm-adjusted P = 0.69).

That makes baseline insomnia severity, not the size of the mood response, the variable this paper points to for identifying who shows the largest absolute sleep-score reductions — with the explicit caveat that the association may be an artifact of limited room to improve. The paper's own priority is therefore a stratified follow-up: test whether patients selected for insomnia severity show the same dissociation, and whether objective sleep gains track subjective ones outside a 24-hour window.

Sources used in this answer