Resting-State Electroencephalogram Complexity Is Associated With Oral Ketamine Treatment Response: A Bayesian Analysis of Lempel-Ziv Complexity and Multiscale Entropy

Abstract

Introduction: Subanesthetic doses of ketamine are a promising novel treatment for suicidality; however, the evidence for predictive biomarkers is sparse. Recently, measures of complexity, including Lempel-Ziv complexity (LZC) and multiscale entropy (MSE), have been implicated in ketamine's therapeutic action. We evaluated electroencephalogram (EEG)-derived LZC and MSE differences between responders and nonresponders to oral ketamine treatment.

Methods: A total of 31 participants received six single, weekly (titrated) doses of oral racemic ketamine (0.5-3 mg/kg) and underwent EEG scans at baseline (Week 0), post-treatment (Week 6), and follow-up (Week 10). Resting-state (eyes closed and open) recordings were processed in EEGLAB, and complexity metrics were extracted using the Neurokit2 package. Participants were designated responders or nonresponders by clinical response (Beck Suicide Scale [BSS] score reduction of ≥ 50% from baseline to the respective timepoint or score ≤ 6) and then compared in terms of complexity across resting-state conditions and time.

Results: Employing a Bayesian mixed effects model, we found strong evidence that LZC was higher in the eyes-open compared to eyes-closed condition, as were MSE scales 1-3. At a global level, responders displayed elevated eyes-open baseline complexity compared to nonresponders, with these values decreasing from baseline to post-treatment (Week 6) in responders only. Exploratory analyses revealed that the elevated baseline eyes-open LZC in responders was spatially localized to the left frontal lobe (F1, AF3, FC1, and F3).

Conclusion: EEG-complexity metrics may be sensitive biomarkers for evaluating and predicting oral ketamine treatment response, with the left prefrontal cortex bein a possible treatment response region.

Keywords: complexity; electroencephalogram; ketamine; suicidality.

FIGURE 1

Estimated marginal means for multiscale entropy showing condition‐by‐scale interaction across timepoint and response status. Bars represent 95% highest posterior density. BAS, baseline; EC, eyes closed; EO, eyes open; FUP, follow‐up; NR, nonresponder; POST, posttreatment; RESP, responder.

FIGURE 2

Estimated marginal means for multiscale entropy showing response‐by‐scale interaction across timepoint and condition. Bars represent 95% highest posterior density. BAS, baseline; EC, eyes closed; EO, eyes open; FUP, follow‐up; NR, nonresponder; POST, posttreatment; RESP, responder.

FIGURE 3

Estimated marginal means for Lempel–Ziv complexity demonstrating condition‐by‐timepoint interaction across response status. Bars represent 95% highest posterior densities. BAS, baseline; EC, eyes closed; EO, eyes open; FUP, follow‐up; NR, nonresponder; POST, posttreatment; RESP, responder.

FIGURE 4

Estimated marginal means for Lempel–Ziv complexity showing timepoint‐by‐condition‐by‐response status interaction. Bars represent 95% highest posterior densities. BAS, baseline; EC, eyes closed; EO, eyes open; FUP, follow‐up; NR, nonresponder; POST, posttreatment; RESP, responder.

FIGURE 5

Topographical plot of by‐channel eyes‐open Lempel–Ziv complexity values at baseline for (A) responders, (B) nonresponders, and (C) the evidence (Bayes factors) for a difference between the groups.

FIGURE 6

Estimated marginal means for Lempel–Ziv complexity at channels Fp1, AF3, FC1, and F3, showing timepoint‐by‐condition‐response status interaction. Bars represent 95% highest posterior density. AF3, anterior‐frontal 3; BAS, baseline; EC, eyes closed; EO, eyes open; F3, frontal 3; FC1, fronto‐central 1; Fp1, frontal pole 1; FUP, follow‐up; NR, nonresponder; POST, posttreatment; RESP, responder.