Neural complexity EEG biomarkers of rapid and post-rapid ketamine effects in late-life treatment-resistant depression: a randomized control trial

Treatment-resistant depression (TRD), typically defined as failure to respond to two or more adequate antidepressant trials, remains a major clinical problem across the lifespan. Ketamine, an N-methyl-D-aspartate receptor (NMDAR) antagonist, has emerged as an effective non-monoaminergic treatment for TRD, with proposed antidepressant mechanisms that include a transient glutamatergic surge, modulation of excitatory/inhibitory (E/I) balance, release of brain-derived neurotrophic factor (BDNF) and activation of mTOR pathways. Prior electroencephalogram (EEG) studies have used gamma-band oscillations as a proxy for ketamine’s glutamatergic effects, but gamma is a linear measure and may not capture broader, system-level changes in neural communication that could underpin antidepressant response in late-life TRD (LL-TRD). Murphy and colleagues sought to evaluate non-linear EEG biomarkers of ketamine’s rapid and post-rapid effects in LL-TRD. Specifically, they performed a secondary analysis of a randomised controlled trial comparing a single 40-minute intravenous ketamine infusion (multiple dose arms) with an active psychoactive control (midazolam), using two complementary complexity measures—Lempel–Ziv complexity (LZC), which indexes signal randomness, and multiscale entropy (MSE), which quantifies signal regularity across temporal scales. The primary aim was to test whether subanesthetic ketamine increases complexity in rapid (baseline to 240 min) and post-rapid (24 h and 7 days) windows relative to midazolam, and to examine relationships between complexity changes and clinical outcome (MADRS change at day 7).

This work is a secondary analysis of a registered, institutional review board–approved clinical trial conducted at the Michael E. DeBakey VA Medical Center. Participants were adults with late-life TRD enrolled in a trial that randomised patients, using a Bayesian adaptive strategy, to one of four dose groups: ketamine 0.1 mg/kg, 0.25 mg/kg, 0.5 mg/kg or midazolam 0.03 mg/kg. Neurophysiological and clinical assessments were collected across ten visits; on the infusion day EEG and other measures were obtained at pre-infusion baseline, during the 40-minute infusion (30 min), and at 60, 120 and 240 minutes post-infusion, with additional follow-ups at 24 hours and 7 days post-infusion. Resting-state EEG was recorded with a 64-channel system while subjects completed 2-minute eyes-closed and eyes-open blocks; this analysis used the eyes-closed recordings to reduce visual and attentional confounds. Signals were digitised at 1000 Hz and downsampled to 250 Hz for analysis; preprocessing included 1–50 Hz band filters, removal of line noise harmonics, artifact subspace reconstruction (ASR), spherical spline interpolation for bad channels, conversion to common average reference, ICA-based component rejection (ocular, muscular, electrode noise), and Laplacian spatial filtering. To simplify inference about global cortical complexity, LZC and MSE were first computed per channel and then averaged across channels to yield a single value per subject/timepoint. LZC was estimated from a 100 s eyes-closed segment (first 110 s with first 10 s discarded), following Hilbert transform and binarisation around the mean rectified signal; values were normalised using surrogate shuffled signals to yield a 0–1 scale. MSE was computed using non-overlapping 4 s epochs averaged across the recording, with 20 scale factors, m = 2 and r = 0.5, and then summarised across scales or binned (scales 1–5, 6–10, 11–15, 16–20) to reduce comparisons. Statistical analysis used linear mixed models (LMMs) with random intercepts for subjects to test two contexts: rapid effects (pre-infusion to 240 min) and post-rapid effects (pre-infusion, 24 h, 7 days). Models began full factorial (time, drug, and for MSE, scale) and were refined to minimise the Bayesian Information Criterion (BIC). Kendall’s tau correlations (Bonferroni-corrected within test families) assessed associations between baseline and potentiated complexity measures and day 7 MADRS change; potentiation was expressed as percentage change from baseline.

The extracted text reports that the sample comprised participants from the original trial (the abstract notes N = 33 military veterans with LL-TRD). Rapid- and post-rapid-window analyses were conducted for both LZC and MSE. LZC rapid effects: The LMM for the rapid window showed significant fixed effects of drug (F = 6.01, p = 0.018) and a time × drug interaction (F = 3.84, p = 0.006). Estimates indicated that LZC was broadly increased by ketamine relative to midazolam, with the strongest increase occurring at the 30 min post-infusion measurement. LZC post-rapid effects: In the post-rapid window (pre-infusion, 24 h, 7 days), the LZC model revealed no significant fixed factors or interactions (drug: F = 0.095, p = 0.76; time: F = 1.12, p = 0.34; drug × time: F = 0.23, p = 0.80), indicating no detectable ketamine versus midazolam differences in LZC at these later time points. MSE rapid effects: The MSE rapid-window model returned multiple significant terms: drug (F = 9.2, p = 0.003), time (F = 11.68, p < 0.001), scale (F = 170.2, p < 0.001), drug × time (F = 13.85, p < 0.001), and drug × scale (F = 3.4, p < 0.001). Both groups showed MSE curves that rose toward the end of the first quarter of total timescales and then declined toward scale 20. The authors note that ketamine produced a broad reduction of MSE during day 1 relative to midazolam, but interaction terms denote scale- and time-sensitive effects with the greatest drug-related differences at 30 min post-infusion and primarily affecting timescales 6–20. MSE post-rapid effects: For the post-rapid model, significant fixed effects were observed for day (F = 5.67, p = 0.004), drug (F = 4.77, p = 0.03), and scale (F = 144.98, p < 0.001), plus a day × drug interaction (F = 7.85, p < 0.001). MSE was broadly reduced at 24 hours compared with day 1 (p = 0.009) and compared with day 7 (p = 0.013), while day 1 and day 7 entropy values were not distinct. Across the post-rapid window, ketamine increased overall entropy relative to midazolam (p = 0.031), and simple effects analysis indicated that the distinctions between day 1 and 24 h, and between 24 h and day 7, were observed primarily in the ketamine group. Correlations with clinical outcome: No significant correlations were found between baseline or potentiated complexity measures (LZC or MSE) and MADRS change at day 7 after correction (Kendall’s tau; Bonferroni adjustment). Thus, complexity as measured here did not predict antidepressant response at the primary clinical endpoint.

Murphy and colleagues interpret their findings as evidence that a single subanesthetic ketamine infusion produces time-varying effects on non-linear EEG complexity in individuals with late-life TRD. Rapid effects included increased LZC and changes in MSE around 30 minutes after infusion start, with MSE exhibiting scale- and time-dependent patterns rather than effects confined to a single timescale. In the post-rapid window, drug-related effects were restricted to MSE, with a reduction in entropy at 24 hours that reverted to baseline by day 7. The authors note that these complexity changes were observable outside the temporal window previously associated with gamma-band potentiation in this sample, suggesting that complexity measures capture different or broader aspects of neural dynamics than linear oscillatory metrics. The discussion emphasises mechanistic distinctions between the two complexity metrics: LZC quantifies the degree of randomness or the number of recurring binary patterns in a time series, offering a broad indication of deviation from uniformity, whereas MSE assesses regularity across temporal scales and thus can reflect subsystem-specific dynamics. The study team suggests that complexity complements gamma measures by indexing system-wide information processing and functional adaptability, but their analyses did not find evidence that baseline complexity or ketamine-induced complexity changes mediated antidepressant response (MADRS reduction) at day 7. The authors therefore caution that complexity may act as a complementary physiological descriptor rather than a direct moderator or mediator of clinical benefit and recommend adequately powered mediation analyses in future work. Several limitations are acknowledged. The analysis was secondary to a trial designed for dose optimisation and employed Bayesian adaptive randomisation, resulting in few participants in some lower-dose ketamine arms and potential dose-mixing that reduced power for dose-specific inferences. The sample had a narrow age range and limited size, constraining subgroup analyses such as biological sex effects. Methodological constraints include the continuous nature of complexity measures without established clinical cutoffs, the use of eyes-closed recordings (when eyes-open may influence complexity), lack of accurate EEG source localisation with 64 channels, and reliance on an active psychoactive control (midazolam) rather than healthy controls or inert placebo. The authors also note that psychological correlates such as dissociation were not assessed here and might relate to acute complexity changes. They recommend replication in larger, age-diverse samples, inclusion of healthy controls and non-active placebo arms, exploration of sex differences, and combined analyses of complexity with oscillatory properties (power and cross-frequency interactions) to better characterise ketamine’s physiological signature.

The authors conclude that neural complexity measures provide preliminary evidence as biomarkers of rapid and post-rapid ketamine effects in late-life treatment-resistant depression. Their results indicate time-varying, non-linear alterations in EEG complexity following a single subanesthetic ketamine infusion that are distinct from and extend beyond previously reported gamma-band effects. Although complexity changes were observed independently of depressive symptom reduction at day 7, the investigators argue that complexity—being non-linear and amplitude-independent—could offer complementary information about system-wide neural dynamics. They recommend replication in larger, multi-cohort studies with age-matched healthy controls to establish clinical utility and to enable well-powered multivariate investigations into the physiological correlates of ketamine-induced antidepressant response.