Stanford neuromodulation therapy for treatment‐resistant depression: a randomized controlled trial confirming efficacy, and an EEG study providing insight into mechanism of action and a potentially predictive biomarker of efficacy

Abstract

Stanford neuromodulation therapy (SNT) is a rapid‐acting, high‐dose, intermittent theta‐burst stimulation protocol. Although it has previously demonstrated efficacy for treatment‐resistant depression (TRD) in a randomized controlled trial (RCT), replication in a larger sample is needed. Additionally, the electrophysiological effects of SNT remain unknown. Here we report results from a new double‐blind, sham‐controlled RCT along with electroencephalography (EEG) findings from the initial and current trials. In the current RCT, 53 participants with TRD were enrolled, and 48 who continued to meet entry criteria were randomized to receive active (N=24) or sham (N=24) SNT. At 1‐month, remission (primary outcome) was achieved in 50.0% of active vs. 20.8% of sham participants (χ² _1,48=4.5, p=0.035), and response (secondary outcome) similarly favored active treatment (54.2% vs. 25.0%; χ² _1,48=4.3, p=0.039). Beta band EEG findings converged across trials: frontal beta power decreased significantly following active but not sham SNT in both the initial pilot study and the current trial. Additionally, beta baseline activity and post‐SNT changes related to treatment efficacy in the current study. Specifically, greater post‐SNT reduction in left anterior cingulate cortex (L‐ACC) beta power correlated with greater clinical improvement immediately (rho=0.48, p=0.019) and 1‐month after (rho=0.51, p=0.012) active SNT. Moreover, higher pre‐treatment L‐ACC beta power predicted greater subsequent clinical benefit from active SNT (immediate‐post: β=‐10.26, p=0.0042; 1‐month after: β=‐9.00, p=0.024). Neither of these L‐ACC beta power findings was observed with sham stimulation. In sum, this study replicates SNT’s therapeutic efficacy, identifies left frontal beta suppression as a potential mechanism of action, and highlights baseline L‐ACC beta power as a candidate scalable pre‐treatment biomarker of efficacy.

Major depressive disorder (MDD) is a leading contributor to global disability, affecting over 280 million people and being responsible for more than 5% of all years lived with disability worldwide ¹ . In the US, the economic burden of MDD exceeds $326 billion annually ² .

Approximately 30% of individuals diagnosed with MDD experience treatment‐resistant depression (TRD), usually defined as an insufficient response to at least two adequate trials of antidepressant medications ³ . TRD contributes to additional societal costs (more than $90 billion annually) and disability, including lost productivity, unemployment, and further health care costs. It is also associated with an increased all‐cause mortality risk compared with MDD that is not treatment‐resistant ⁴ .

When patients with TRD do not respond to first‐line pharmacological and psychotherapeutic treatments, brain stimulation interventions are often considered. Repetitive transcranial magnetic stimulation (rTMS) is one of a growing number of brain stimulation treatments with proven efficacy for TRD ⁵ , ⁶ , ⁷ . Advances in our understanding of the neurobiology of depression and the mechanisms underlying the clinical effects of rTMS have guided the development of innovative brain stimulation methods for TRD ⁸ , ⁹ , ¹⁰ , ¹¹ . We recently developed an accelerated version of rTMS called Stanford neuromodulation therapy (SNT), based on converging lines of evidence related to the antidepressant effects of rTMS ¹² , ¹³ .

SNT is an accelerated, high‐dose, patterned, functional connectivity magnetic resonance imaging (fcMRI)‐guided rTMS protocol that utilizes intermittent theta‐burst stimulation ¹² , ¹³ . In an initial randomized controlled trial (RCT), treatment with a 5‐day course of active SNT resulted in remission rates of 57.1% immediately post‐SNT and 46.2% 1‐month post‐SNT in participants with moderate to severe TRD ¹³ . Although this initial trial showed robust clinical effects (Cohen’s d of 1.7 and 1.4 immediately post‐SNT and 1‐month post‐SNT, respectively), its limited sample size (N=29) and the ongoing challenge of research reproducibility in psychiatry ¹⁴ highlighted the need for replication in a larger trial.

Additionally, our understanding of the therapeutic mechanisms underlying SNT is limited. Increased knowledge of SNT’s neurophysiological effects may guide future development of the treatment, potentially allowing for optimization of patient selection and further personalization of treatment to enhance clinical outcomes.

Thus far, functional neuroimaging has yielded fruitful initial insights into the mechanism of action of SNT. Specifically, imaging data from the above‐mentioned RCT ¹³ demonstrated that functional connectivity between the amygdala and the default mode network (DMN) significantly increased following active SNT and was correlated with better clinical outcomes ¹⁵ . Further, a seed‐based approach identified increased anticorrelation between the left dorsolateral prefrontal cortex (L‐DLPFC) and the DMN following active as compared to sham SNT ¹⁶ . Finally, a spatiotemporal analysis revealed that SNT induced signaling shifts in the L‐DLPFC and bilateral anterior cingulate cortex (ACC), that directional shifts in the ACC predicted improvement in depression symptoms post‐treatment, and that pre‐treatment ACC signaling predicted the likelihood of subsequent SNT treatment response ¹⁷ .

Electroencephalography (EEG) offers a cost‐effective tool for measuring the effects of SNT on neural activity with sub‐millisecond resolution. EEG measures of oscillatory activity within the ~13‐30Hz frequency range (i.e., beta power) have been linked to depressive states in humans. For example, an anhedonia symptom subtype in a large transdiagnostic sample (N=420) was specifically associated with elevated frontal beta activity ¹⁸ , and abnormally high frontal beta power was linked to depressive symptoms in several cross‐sectional resting‐state EEG studies ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ . Similarly, intracranial EEG measurements in the ACC have shown that beta power is related to depressive symptoms in TRD ²⁵ and tracks changes in depressive symptoms after deep brain stimulation ²⁶ , ²⁷ . Pre‐treatment measures of beta power have also been identified as potential predictors of rTMS depressive symptom outcomes ²⁸ , ²⁹ , ³⁰ . Yet, the effects of SNT on beta activity have not been investigated.

In this double‐blind, randomized, sham‐controlled trial, we assessed SNT’s antidepressant efficacy in a newly recruited sample of patients with TRD. We also analyzed the neurophysiological effects of SNT via EEG recorded pre‐ and post‐treatment, examining data both from a subset of the above‐mentioned initial RCT ¹³ as well as from the current trial.

We specifically investigated the effects of active SNT on left frontal beta activity. We further investigated how these effects and pre‐treatment beta activity relate to individual treatment outcomes by investigating that activity in cortical regions previously implicated in SNT’s therapeutic benefits – L‐DLPFC and left ACC (L‐ACC) – using EEG co‐registered with MRI for source localization. We hypothesized that active SNT would again demonstrate significantly greater antidepressant efficacy than sham SNT, and that only active treatment would modulate left frontal beta activity.

METHODS

Study design

We conducted a double‐blind, randomized, sham‐controlled trial, prospectively registered in the US Clinical Trials registry (NCT04739969). All procedures were carried out in accordance with the ethical standards outlined in the Declaration of Helsinki. The study was approved by the Stanford University Institutional Review Board. All participants provided written consent before taking part in any study procedures.

Participants

The study was carried out at the Department of Psychiatry of Stanford University from June 17, 2021 to June 6, 2024. We recruited individuals from the community using media advertisements and from referring clinics both inside and outside of Stanford University.

Included participants had a primary diagnosis of MDD according to the DSM‐5, were currently experiencing a moderate to severe depressive episode (score ≥20 on the Montgomery‐Åsberg Depression Rating Scale, MADRS ³¹ ), were between 22 and 65 years old, and had moderate‐to‐severe levels of treatment resistance as measured by the Maudsley Staging Method ³² (score ≥7 required), confirmed by medical records. Adequate treatment failures were defined according to the Antidepressant Treatment History Form (ATHF) ³³ . Participants were required to maintain a stable antidepressant regimen for 6 weeks prior to treatment and to remain on this regimen throughout the study (including all follow‐up assessments after the 5‐day treatment protocol).

Key exclusion criteria were any primary psychiatric diagnosis other than MDD (except for stable comorbid anxiety disorder), unstable symptoms between screening and baseline as defined by a >30% change in MADRS score, intellectual disability, autism spectrum disorder, moderate or severe substance use disorder, active suicidal ideation (defined as a score ≥8 on the Modified Scale for Suicidal Ideation, MSSI ³⁴ ), contraindication to MRI ³⁵ , or any condition that would increase the risk associated with receiving intermittent theta‐burst stimulation ³⁶ . Individuals with prior exposure to rTMS, non‐response to electroconvulsive therapy (ECT), or a depth‐adjusted SNT treatment dose >65% maximum stimulator output (MSO) were also excluded.

Participants were instructed to continue usual intake patterns of caffeine‐ or xanthine‐containing products (e.g., coffee, tea, cola drinks, and chocolate) without significant change for the duration of the study, and to abstain from alcohol for at least 24 hours before the start of each rTMS session.

Randomization

Participants were randomized to active vs. sham SNT in a 1:1 ratio in randomly selected block sizes of 2, 4 or 6 as based on a random number generator and without stratification.

Intervention

The SNT paradigm has been previously described ¹² . Briefly, a baseline structural and resting‐state functional‐connectivity MRI (rs‐fcMRI) was obtained for each participant. Custom scripts were used to identify the portion of the L‐DLPFC maximally anticorrelated with the subgenual ACC, which served as the target for both active and sham stimulation. Stimulation was delivered via a MagVenture MagPro X100 (Skovlunde, Denmark) TMS device equipped with a Cool‐B65 A/P coil. The personalized L‐DLPFC functional target was localized for each participant using a Localite TMS Navigator (Localite, Bonn, Germany).

Participants were treated with 1,800 pulses of intermittent theta‐burst stimulation (3‐pulse 50‐Hz bursts at 5‐Hz for 2‐sec trains, with trains every 10 sec) per session at 90% resting motor threshold depth‐adjusted to the personalized functional target. Ten sessions were delivered per day (18,000 pulses/day) for 5 consecutive days (90,000 total pulses). The inter‐session interval between same‐day treatment sessions was 50 min.

To ensure adequate sleep each evening prior to SNT intervention (a safety requirement), study physicians had the option to prescribe certain sleep medications (e.g., zolpidem, zaleplon, eszopiclone, quetiapine) immediately prior to or during the course of SNT. In these cases, participants were instructed to take the medication the night prior to the stimulation session but not during the morning of the sessions or at any time during the intervention day. The use of alternative hypnotics or the short‐term use of anxiolytic medications (e.g., hydroxyzine, propranolol) during an intervention day required prior approval by a study physician.

If a participant met the remission criterion according to the MADRS (see below) at the immediate post‐SNT visit or a weekly post assessment, but then failed to do so at a subsequent weekly post assessment, he/she was offered additional SNT treatment in blocks of 2 days (6 additional treatments in total maximum allowed) until either reaching the prior lowest total MADRS score or undergoing the 1‐month follow‐up visit. A minimum of 2 days must have passed between additional SNT treatment and the 1‐month follow‐up visit. The purpose of these additional days was to return participants who already had achieved remission from the acute SNT intervention to euthymia before transitioning to long‐term follow‐up, as inspired by the fixed taper schedule utilized in the pivotal ECT Prolonging Remission in Depressed Elderly (PRIDE) study ³⁷ . Three participants in the active group and one in the sham group received additional SNT treatment.

Blinding

All participants, clinical assessors, SNT operators, study physicians, and other study staff were blinded to treatment assignments. Clinical assessors, EEG technicians, and SNT operators were separate individuals.

Clinical outcomes

Assessments were performed at screening, baseline, immediate post‐SNT (3‐4 days after the final treatment), 1‐, 2‐ and 3‐weeks after day 5 of SNT, and 1‐month post‐SNT, by evaluators blinded to treatment condition. Evaluators also met biweekly with a licensed psychologist to ensure consensus and prevent rater drift. The 1‐month visit had to occur between 25 and 35 days after the day 5 of SNT.

The primary outcome was rate of remission 1‐month after treatment, as defined by a MADRS score ≤10 ³⁸ . The secondary outcome was rate of response, defined as a MADRS score reduction of ≥50% from baseline, at 1‐month. As an exploratory outcome, a generalized linear mixed effects model was used to assess the effects of group and time from baseline through 1‐month post‐treatment on the total MADRS score.

Safety

Adverse events (AEs) were systematically screened for with a series of standardized questions asked at each visit. Any positive response was documented and reported to a study physician for further evaluation and determination of whether or not an AE had occurred. Per protocol, stimulation‐based site discomfort was not considered an AE, given that it is expected and normally occurs with rTMS, but pain persisting beyond stimulation would be coded as an AE.

History of suicidal ideation and behavior was assessed with the Columbia‐Suicide Severity Rating Scale (C‐SSRS) ³⁹ , and suicidality was monitored during the trial via the suicidal thoughts item of the MADRS and the MSSI. The Young Mania Rating Scale (YMRS) ⁴⁰ was administered at the end of each treatment day and at select follow‐up visits to monitor for any potential treatment‐emergent mania. An independent data safety monitoring board (DSMB) reviewed study progress and safety at regular intervals.

Clinical statistics

A power analysis based on clinical efficacy indicated that 18 participants were required for each arm. However, we initially aimed to enroll a larger sample of 100. Unfortunately, study initiation was substantially delayed by the COVID‐19 pandemic, resulting in the study being behind on its milestones. Accordingly, after enrollment of approximately half of the intended sample size, discussion with the DSMB led to a decision to perform an interim analysis. The pre‐specified stopping rules were to end the trial if both the primary and secondary outcomes were positive, or, conversely, if the difference between active and sham SNT outcomes was minimal (or favoring sham) such that the trial appeared to be futile. As the primary outcome was found to be positive, the study was completed.

Unless otherwise noted, categorical data are presented as percentages and compared using the Pearson chi square test of independence, and Fisher’s exact test where necessary. Baseline continuous data are presented as means with standard deviations (SDs) and compared using ANOVA. As no baseline characteristics displayed imbalance between groups or an association with the dependent variable of the primary outcome, no covariates were included in final models.

The primary and secondary outcomes were assessed according to the intent‐to‐treat principle, and the last observation was carried forward in cases of missing data with no interpolation. For the exploratory analysis, we used a general linear mixed‐effects model with group and time as interacting fixed effects and a random intercept for each participant. Because residuals of standard linear mixed‐effects models were not normally distributed, exploratory analyses were conducted using a generalized linear mixed‐effects model implemented in SAS (PROC GLIMMIX, version 9.4), which applied empirical (sandwich) standard errors (SEs) to ensure robust inference. Fixed effects included treatment group, time, and their interaction. A random intercept was included for each participant. Participant‐level autocorrelations were modeled with an autoregressive covariance structure of order 1. Models were estimated using restricted maximum likelihood (REML) with the Satterthwaite approximation for denominator degrees of freedom. Least‐squares means were estimated for each group at each time point.

To assess differential changes from baseline between treatment groups, we specified linear contrast statements comparing pre‐post change scores across groups at each follow‐up time point. Resulting p values were Bonferroni‐corrected for five comparisons to control the family‐wise error rate. SEs and confidence intervals (CIs) were extracted from the model‐estimated covariance structure. All analyses were conducted in SAS Studio (v9.4).

EEG data collection and pre‐processing

In both the initial ¹³ and current RCTs, eyes‐open resting state EEG (rs‐EEG) recordings were performed at baseline and at the immediate post‐SNT visit. We recorded 5 min of rs‐EEG per session in the initial trial, and 6 min of rs‐EEG (two 3‐min recordings) per session in the current trial.

All EEG data were collected using a 64‐channel actiCap slim EEG cap with the acti64Champ amplifier (Brain Products GmbH, Germany). During data collection, the EEG signal was referenced to Cz (vertex reference) and sampled at a rate of 10kHz. Channels were positioned according to the 10‐10 system. Participants were seated upright during the recording and were instructed to look at a fixation cross while letting their minds wander.

EEG data were pre‐processed in MATLAB using a semi‐automated approach and open‐source functions from EEGLAB ⁴¹ and ARTIST ⁴² toolboxes. All pre‐processing was performed by an experienced rater who was blinded to the intervention group and clinical scores associated with each recording (see supplementary information for a detailed list of pre‐processing steps).

The 1‐50Hz power spectral density was computed using Welch’s method with 2‐sec Hamming windows (50% overlap). Epochs with 1‐50Hz power deviating from the mean by 50 dB or a z‐score >1.96 were rejected. The average length of the rs‐EEG recordings post‐processing was 4.57±0.46 min (range: 2.97‐4.97) for the initial RCT pilot data and 5.34±0.58 min (range: 2.97‐5.87) for the current trial. Beta power (13‐30Hz) was averaged in a left frontal cortical region of interest (Fz, F3, F1, AF3, and AFz).

Source localization was also performed using the open‐source Brainstorm toolbox ⁴³ to extract beta power in two frontal cortical regions previously implicated in the antidepressant efficacy of rTMS: L‐DLPFC and L‐ACC ⁸ , ⁹ , ¹¹ , ¹² , ¹⁷ . EEG sensor data were first co‐registered to individual structural T1‐weighted images, and a realistic boundary element method (BEM) forward model was computed from 15,000 cortical vertices (i.e., sources) with unconstrained dipole orientations. To compute the inverse model, we used standardized low‐resolution electromagnetic tomography (sLORETA) and the diagonal noise covariance. Beta power in source space was then computed using Welch’s method (2‐sec windows with 50% overlap) and averaged within L‐DLPFC and L‐ACC regions using the Desikan‐Killiany atlas left rostral middle frontal and left rostral anterior cingulate regions, respectively.

EEG statistics

To account for non‐normally distributed beta power values, a log‐transform was applied to all power values. To assess whether effects of SNT on beta power differed between active and sham groups, linear mixed‐effects models were performed using the R lme4 package ⁴⁴ . Beta power was defined as the outcome; time point (pre‐ vs. post‐SNT), treatment group (active vs. sham), and their interaction as fixed effects; age as a covariate; and participant as a random effect.

Post‐hoc simple‐effects analyses were performed by deriving estimated marginal means for each treatment group (active and sham) at each time point and comparing pre‐ vs. post‐treatment within each treatment group. All statistical analyses were performed using R software ⁴⁵ , version 4.3.1 or newer (in RStudio v.2023.06.0 or newer ⁴⁶ ).

The relationship between changes in L‐DLPFC and L‐ACC beta power and improvements in depressive symptoms after treatment was then assessed using Spearman’s correlations. Additionally, linear regression models were fitted to the data, adjusted for age, to assess whether baseline L‐DLPFC and L‐ACC beta power was predictive of SNT clinical outcomes. Histograms and Q‐Q plots of all model residuals were checked to assess normality. The variance inflation factor was used to assess for collinearity in the fitted model.

RESULTS

Demographics

From June 17, 2021 to June 6, 2024 we screened 199 participants, of whom 73 (36.7%) were preliminarily eligible, 53 (26.6%) were enrolled, and 48 (24.1%) were ultimately randomized and received active SNT (N=24) or sham SNT (N=24). One participant in the sham group withdrew early (after one day of treatment). All other participants (N=47) completed the study. The CONSORT diagram is presented in Figure 1.

Figure 1.

CONSORT diagram

Baseline demographic and clinical characteristics of participants are presented in Table 1. They were similar between the two groups. Among the entire cohort (mean±SD), participants’ current depressive episode had lasted for 4.9±5.2 years; they had a history of 3.2±2.6 past depressive episodes; they had failed a total of 4.5±1.8 lifetime adequate antidepressant treatments (2.5±1.5 in the current episode) with a Maudsley staging method score of 8.5±1.4 (moderately treatment resistant), and they presented with a moderately severe current symptom burden (mean MADRS score of 27.9±5.7). Personalized, fcMRI‐derived SNT brain targets were obtained for each participant and qualitatively similarly located throughout the L‐DLPFC in both groups.

Table 1.

Baseline demographic and clinical characteristics of participants in the trial

	Active SNT (N=24)	Sham SNT (N=24)
Sex, N (%)
Male	10 (41.6)	15 (62.5)
Female	14 (58.3)	9 (37.5)
Gender, N (%)
Male	11 (45.8)	15 (62.5)
Female	13 (54.2)	9 (37.5)
Age (years), mean±SD	44.8±11.8	47.3±12.1
Race, N (%)
White	22 (91.6)	17 (70.8)
Asian	1 (4.2)	6 (25.0)
American Indian/Alaska Native	0	1 (4.2)
Other	2 (8.3)	1 (4.2)
Ethnicity, N (%)
Hispanic or Latino	1 (4.2)	4 (16.7)
Non‐Hispanic or Latino	23 (95.8)	20 (83.3)
N. past episodes, mean±SD	3.1±2.8	3.3±2.5
Duration of current episode (months), mean±SD	64.8±63.7	53.4±60.9
N. adequate antidepressant failures (lifetime), mean±SD	4.5±1.7	4.5±1.9
N. adequate antidepressant trials (current episode), mean ±SD	2.3±1.6	2.6±1.4
Maudsley Staging Method score, mean±SD	8.5±1.4	8.5±1.5
Baseline MADRS score, mean±SD	28.5±5.8	27.3±5.5

SNT – Stanford modulation therapy, MADRS – Montgomery‐Åsberg Depression Rating Scale

Primary outcome

The pre‐defined primary outcome was the rate of remission 1‐month after treatment using the MADRS score (remission defined as MADRS ≤10). The remission rates were 50.0% and 20.8% in the active and sham SNT groups, respectively (X² _1,48=4.5, p=0.035, odds ratio, OR=3.8, number needed to treat, NNT=3.4) (see Figure 2).

Figure 2.

One‐month remission rate in individuals receiving active and sham Stanford neuromodulation therapy

Secondary outcome

The pre‐defined secondary outcome was the rate of response (defined as ≥50.0% improvement from baseline) 1‐month after treatment using the MADRS. The response rates for active and sham SNT were 54.2% and 25.0%, respectively (X² _1,48=4.3, p=0.039, OR=3.5, NNT=3.4) (see Figure 3).

Figure 3.

One‐month response rate in individuals receiving active and sham Stanford neuromodulation therapy

Safety

There were no serious adverse events (SAEs). No AE was significantly more common with active SNT as compared to sham SNT (see supplementary information).

Exploratory clinical outcomes

We assessed whether there were any baseline demographic or clinical differences between participants who responded to or remitted with treatment vs. those who did not. Females were more likely to be responders than males 1‐month after treatment (X² _1,48=5.3, p=0.021). There were no other statistically significant demographic or clinical differences between responders and non‐responders, nor were there any differences for remitters vs. non‐remitters.

Additionally, we assessed the effects of group and time from baseline through 1‐month post‐treatment on the total MADRS score (see Figure 4). We identified statistically significant effects of group (F_1,46=5.24, p=0.0267), time (F_5,201=15.3, p<0.001), and group by time interaction (F_5,201=3.5, p=0.0047). In post‐hoc testing, the change in total MADRS score from baseline was significantly different between the two groups at each time point (see supplementary information).

Figure 4.

Longitudinal clinical outcomes by mean score on the Montgomery‐Åsberg Depression Rating Scale (MADRS) in participants receiving active and sham Stanford neuromodulation therapy. Solid circles represent generalized mixed‐effects model‐estimated marginal means. Error bars represent model‐based standard errors.

SNT effects on beta activity

In the initial trial ¹³ , EEG was recorded in a subset of 16 participants (7 active, 9 sham; mean age: 49.6±15.9 years; 12 male). At one month, MADRS scores improved by 57.3% in the active group and 25.9% in the sham group. We used this subset as an initial test of our hypothesis that active and sham treatment would differentially modulate left frontal beta activity. Although the small sample limited statistical power and the overall group x time interaction did not reach significance (β=0.44, t₁₅=1.95, p=0.071), post‐hoc analysis showed that left frontal beta power significantly decreased after treatment in the active group (t₁₅=2.41, p=0.030, d=0.62) (see Figure 5).

Figure 5.

Pre‐post changes in left frontal beta power in patients receiving active and sham Stanford neuromodulation therapy (SNT) in the initial and current trial. The group x time interaction was not significant (p=0.071) in the initial trial (N=16), whereas it was significant (p<0.001) in the current trial (N=44). In post‐hoc tests, frontal beta power was significantly reduced after active SNT compared to baseline in the initial (p=0.030) and current (p=0.005) trial. Error bars reflect standard error of the mean.

In the current RCT, EEG was obtained in 45 participants at baseline (age: 45.4±11.8 years; 24 male; N=24 active, N=21 sham) and in 44 participants at the immediate post‐SNT visit (N=23 active, N=21 sham; see supplementary information for data exclusion details). The interaction effect between group and time on frontal beta band power was significant (β=0.58, t₄₂=3.69, p<0.001). Replicating the initial trial, post‐hoc comparisons revealed a significant decrease in beta power after treatment in the active group (t=2.96, p=0.005, d=0.56) (see Figure 5). Findings persisted in sensitivity analyses excluding participants with concurrent benzodiazepine use (see supplementary information).

Associations between beta activity and SNT outcomes

To investigate how beta activity might be related to clinical outcomes in the current trial, we localized beta activity to two key regions previously implicated in rTMS efficacy: L‐ACC and L‐DLPFC. A pre‐post treatment reduction in beta activity was observed in the L‐ACC among responders to active treatment, but not in non‐responders (see supplementary information). To further identify whether variance in beta power in L‐ACC and L‐DLPFC was meaningfully related to clinical improvement, we assessed correlations between treatment‐related changes in beta power and reductions in MADRS scores, and used linear regression to model the relationship between baseline beta activity and symptom improvement.

In the active SNT group, greater reductions in L‐ACC beta power correlated with greater reductions in MADRS scores both immediately post‐SNT (rho=0.48, p=0.019) and 1‐month post‐SNT (rho=0.51, p=0.012) (see Figure 6). By contrast, in the sham group, changes in L‐ACC beta power were not significantly correlated with MADRS score improvements at either time point (immediately post‐SNT: rho=0.13, p=0.57; 1‐month post‐SNT: rho=0.17, p=0.45) (see Figure 6). Changes in L‐DLPFC beta power were not correlated with MADRS score improvements in either group.

Figure 6.

Scatterplots of individual data showing that greater reductions in left anterior cingulate cortex (L‐ACC) beta power after Stanford neuromodulation therapy (SNT) were associated with greater improvements in depressive symptoms both at the immediate post‐SNT visit (rho=0.48, p=0.019) and 1‐month (rho=0.51, p=0.012) after active SNT, but not sham treatment (immediate‐post: rho=0.13, p=0.57; one‐month: rho=0.17, p=0.45). MADRS – Montgomery‐Åsberg Depression Rating Scale.

In the active SNT group, higher pre‐treatment L‐ACC beta power was a significant predictor of greater reductions in MADRS scores immediately post‐SNT (R²=0.40; β=−10.26, 95% CI: −17.60 to −3.76, p=0.0042) and 1‐month post‐SNT (R²=0.30; β=−9.00, 95% CI: −17.37 to −1.38, p=0.024) (see Figure 7). In contrast, greater pre‐treatment L‐ACC beta power was not a significant predictor of reductions in MADRS scores after sham treatment (immediate post‐SNT, p=0.20; 1‐month post‐SNT, p=0.22) (see Figure 7). Associations between baseline L‐DLPFC beta power and clinical outcomes were weaker and not persistent (see supplementary information).

Figure 7.

Scatterplots of individual data showing an association between higher left anterior cingulate cortex (L‐ACC) beta power before Stanford neuromodulation therapy (SNT) and greater improvement in depressive symptoms both at the immediate post‐SNT visit (β=−10.26, p=0.0042) and 1‐month (β=−9.00, p=0.024) after active SNT, but not sham treatment (immediate‐post: p=0.20; 1‐month: p=0.22). MADRS – Montgomery‐Åsberg Depression Rating Scale.

DISCUSSION

The present study examined the antidepressant effects of SNT and investigated its underlying neurophysiological mechanisms. Clinically, SNT produced significantly greater antidepressant effects in participants with TRD than sham treatment, with a particularly notable 50% rate of remission with active SNT at 1‐month.

Using rs‐EEG to probe neurophysiological changes, we found that SNT consistently reduced frontal beta power. Greater reductions in L‐ACC beta activity following active SNT were associated with greater improvements in depressive symptoms, both at the immediate post‐SNT visit and 1‐month post‐treatment. Additionally, higher baseline beta power in the L‐ACC predicted greater symptom improvement at both time points. These findings suggest that L‐ACC beta activity may play a mechanistic role in the therapeutic benefit of SNT, and that baseline beta power could serve as a scalable neurophysiological predictor of treatment response.

This trial marks the largest SNT RCT to date, and the 50% remission rate with active treatment is notable for several reasons. First, it is very similar to the 46.2% one‐month remission rate from the initial RCT ¹³ . Research into the development of more effective antidepressants faces a number of challenges ⁴⁷ , which may contribute to difficulties with reproducibility. According to one report, out of 43 highly cited psychiatric studies, only 16 (37%) replicated ¹⁴ . The fact that both SNT RCTs demonstrated significantly greater antidepressant efficacy with active as compared to sham treatment and a similar remission rate lends confidence to the veracity of these results in a double‐blind trial setting.

Second, these remission rates allow for meaningful comparisons of efficacy between SNT and other common interventions for TRD. Conventional rTMS, for example, is approved by the US Food and Drug Administration (FDA) for TRD, commonly utilized in this population, and demonstrates relatively modest efficacy. In RCTs, rTMS typically yields remission rates of around 15% ⁷ , with somewhat higher rates – up to 37% – reported in open‐label studies ⁴⁸ , ⁴⁹ , ⁵⁰ .

Similarly, esketamine, another intervention approved by the FDA for TRD, has demonstrated moderate efficacy, with response rates between 50% and 70%, and remission rates ranging from 36% to 52.5% ⁵¹ , ⁵² . However, esketamine requires frequent in‐person visits for administration and monitoring, especially during the acute phase, and patients are unable to drive themselves home afterward due to transient side effects. These logistical challenges, along with the potential need for ongoing maintenance treatments, can pose significant barriers to access and adherence.

SNT, by comparison, offers a non‐pharmacological alternative with a favorable side effect profile and a shorter course of treatment. Indeed, its tolerability appears to be comparable to that of conventional rTMS ³⁶ . Perhaps most striking is that the antidepressant outcomes reported in SNT trials appear comparable to those achieved with ECT ⁵³ , which remains the gold standard for TRD. In a meta‐analysis of ECT outcomes in TRD, the acute remission rate was found to be approximately 48% ⁵³ , suggesting that SNT may offer similar therapeutic benefits but with far fewer associated risks.

Notably, in a recent trial comparing intravenous ketamine with ECT for non‐psychotic TRD, the ketamine arm achieved a remission rate of 37.9%, while the ECT arm only reached 21.8% remission ⁵⁴ . Both these rates are lower than those observed in the initial SNT trial ¹³ and in the current RCT. Unlike ECT ⁵⁴ , SNT does not carry risks of cognitive side effects, nor is there the potential for anesthesia‐related complications.

Taken together, these comparisons highlight SNT’s potential as a highly effective and better‐tolerated treatment option for individuals with TRD, particularly since re‐treatment appears to be highly effective for previous treatment responders who experience a subsequent relapse ⁵⁵ .

Third, the efficacy of SNT reinforces the potential of neuroscience‐informed principles to guide the development of novel treatment paradigms and further refine those that exist already. Indeed, rTMS has been described as perhaps the most important advancement in TRD management, due to both its demonstrated efficacy and its potential for further optimization ⁵⁶ . Regarding the latter, a number of different treatment paradigms have been described in recent years ⁵⁷ , ⁵⁸ , ⁵⁹ . For SNT, the relative therapeutic contribution of each of the modified elements remains unknown and requires further study. In the case of functional imaging‐derived targeting, however, such evidence is beginning to emerge. A recent analysis of a large dataset demonstrated that individualized rTMS targets show stronger association with clinical efficacy than TMS targets based on group functional connectivity profiles ⁶⁰ . Furthermore, a prospective randomized trial observed that the use of functional connectivity‐guided targeting in an accelerated course of rTMS resulted in superior outcomes as compared to scalp‐based targeting, with a large effect size ⁶¹ .

These developments underscore how a deeper understanding of the neural mechanisms underlying the therapeutic effects of SNT may facilitate the further optimization of treatment outcomes. Towards that end, this study is the first to investigate the effects of SNT on EEG‐based neurophysiological measures. We validate prior conventional rTMS EEG findings through both replication in two independent samples and the use of high‐density EEG with source reconstruction in a double‐blind, randomized, sham‐controlled design. While earlier studies using fMRI have linked SNT efficacy to changes in functional connectivity in frontal cortical networks, our EEG findings provide complementary mechanistic insight by capturing neural dynamics at distinct oscillatory frequencies with high temporal resolution. The current EEG findings provide novel evidence that L‐ACC beta activity is involved in SNT’s underlying mechanism of action, and identify baseline beta activity in targeted prefrontal regions as a potential predictive biomarker of subsequent SNT efficacy.

Specifically, we found that active SNT reduced left frontal beta power, with greater reductions in the L‐ACC associated with greater symptom improvement both immediately and 1‐month after treatment – a relationship that was observed for active but not sham treatment. These results replicate and extend findings from conventional 5Hz rTMS studies linking frontocentral beta power reductions to symptom improvement ⁶² , ⁶³ , and parallel evidence from deep brain stimulation trials in TRD showing acute beta power reductions in the anterior cingulate alongside clinical response ²⁶ , ²⁷ . Additionally, the inhibition of ACC activity has been shown to induce antidepressant‐like effects in mice, indicating potential causality of this mechanism ⁶⁴ , ⁶⁵ . These findings also complement fMRI evidence that SNT alters directed ACC connectivity, as beta oscillations have previously been implicated in the large‐scale integration of prefrontal network interactions ⁶⁶ . L‐ACC beta power may therefore represent a complementary, scalable neurophysiological measure of SNT’s mechanism of action, with the potential to support real‐time monitoring of neural engagement and individualized dosing.

We also found that higher pre‐treatment beta activity in the L‐ACC predicted greater symptom improvement following active SNT. In contrast, pre‐treatment beta activity was unrelated to outcomes in the sham group, suggesting a specific association with SNT’s therapeutic mechanism. While previous open‐label studies have linked pre‐treatment beta power to clinical response following conventional rTMS protocols ²⁹ , ⁶⁷ , ⁶⁸ , ⁶⁹ , this is, to our knowledge, the first demonstration that pre‐treatment beta power is a predictor of response to an intermittent theta‐burst stimulation treatment protocol. This finding aligns with neuroimaging studies demonstrating that higher pre‐treatment activity in the anterior cingulate predicts more favorable outcomes with conventional high‐frequency rTMS ⁷⁰ , ⁷¹ , ⁷² , ⁷³ . Notably, simultaneous EEG‐fMRI research has shown that higher EEG beta power correlates with greater fMRI activity in the anterior cingulate ⁷⁴ , supporting the use of EEG beta power as a proxy for underlying regional activity.

As an accessible, cost‐effective and temporally precise tool, EEG is well‐suited for translation into the clinic, and may offer a practical supplement to fMRI for applying biomarkers to personalize SNT. For example, patients with elevated baseline L‐ACC beta power could be prioritized for SNT earlier in a treatment algorithm for TRD ⁵⁶ , potentially improving treatment outcomes and cost‐effectiveness. In addition, emerging evidence from studies of motor circuits suggests that higher pre‐stimulation frontocentral beta power facilitates TMS target engagement, as measured by increased propagation of neural signals from targeted cortical regions to subcortical regions ⁷⁵ . Given fMRI findings that greater modulation of downstream functional connections following SNT is associated with clinical response ¹⁵ , ¹⁶ , ¹⁷ , beta‐related facilitation of downstream target engagement may translate into improved clinical outcomes.

In line with this idea, closed‐loop deep brain stimulation protocols that stimulate based on real‐time beta activity may improve treatment outcomes for Parkinson’s disease, as compared to standard open‐loop, biomarker‐naive stimulation protocols ⁷⁶ . These results raise the possibility that EEG beta activity could similarly inform the development of adaptive or closed‐loop SNT paradigms. In addition to personalizing where we stimulate with SNT, such approaches could use real‐time EEG to optimize when we stimulate ⁷⁷ , potentially enhancing therapeutic outcomes by targeting brain states characterized by elevated beta activity.

Altogether, these findings identify L‐ACC beta power as a candidate biomarker for both patient stratification and SNT treatment optimization. Future studies should explore the application of EEG‐based biomarkers in prospective, biomarker‐guided SNT trials.

This study has some limitations. First, although this is the largest RCT of SNT to date, further replication with a greater sample size would be of benefit. Additionally, as in the prior RCT, this study was performed at a single site with participants who were primarily highly educated, White or Asian, and non‐Hispanic or Latino persons. Additional study in other demographic groups is needed to further assess the generalizability of these findings, although we note that a SNT study in bipolar depression at two other academic sites also reported positive findings ⁷⁸ .

All participants in this trial had a primary diagnosis of MDD (although a stable, co‐primary anxiety disorder was allowed), and so the efficacy of SNT in patients in whom depression is co‐primary or secondary remains unknown. While the remission rates achieved by SNT in the initial and current RCTs compare favorably to those reported in traditional rTMS RCTs ⁴ , ⁵ , SNT remains to be tested against rTMS (or another active comparator) to address comparative efficacy. Similarly, although a recent study found that 47% of participants who achieved remission with SNT were still in remission 12 weeks after treatment ⁷⁹ , the longer‐term durability of its antidepressant effect remains unknown.

Additionally, comparisons with healthy controls are needed to determine whether the observed reductions in beta activity following active SNT reflect normalization of neural activity or the engagement of compensatory mechanisms. The current work also focused specifically on beta activity in frontal cortical regions, based on extensive prior evidence linking frontal beta rhythms to depression and neuromodulation outcomes. While this hypothesis‐driven approach helped minimize the issue of multiple comparisons, future work should explore whether other EEG measures and cortical regions also play a role in the therapeutic effects of SNT.

In conclusion, active SNT was more effective than sham in achieving remission of TRD in a double‐blind RCT, replicating the initial RCT in an independent, larger sample. Further, we provide the first report of the electrophysiological effects of SNT, including the identification of a potential pre‐treatment biomarker. The EEG data reported here, in combination with previously reported neuroimaging findings ¹⁵ , ¹⁶ , ¹⁷ , provide an opportunity to build on our understanding of SNT’s therapeutic mechanism of action and the pathophysiology of TRD more generally.