Published on 19 September 2026

Analysis · Depression · Neuromodulation

Can mismatch negativity predict response to iTBS in depression?

▬ Publication Journal of Affective Disorders · 2026;409:121908 · Anderson et al. DOI 10.1016/j.jad.2026.121908 PMID 42086176 Scientific 65 Editorial 70

Editorial collections A Study Under the Microscope PEB Academy

The essentials

This sub-analysis of a randomized, double-blind, sham-controlled trial tested whether a simple electroencephalographic marker, mismatch negativity, could serve as a biomarker in depression treated with intermittent theta-burst stimulation of the dorsomedial prefrontal cortex. The first two questions come back negative: the marker does not significantly distinguish the 46 patients from the 64 healthy controls and is not correlated with baseline symptoms, and no significant change is detected under stimulation, though the study is not powered to exclude an effect. The only signal is a correlation, in the active arm (23 patients), between baseline marker amplitude and depressive improvement (r = 0.48, p = 0.021), not found in the sham arm. But the comparison of the two correlations between arms is not significant (p = 0.091), and that comparison is the actual test of prediction. A correlation significant in one arm and not in the other does not demonstrate a difference between arms. The authors call for replication, and they are right to.

Context

Repetitive transcranial magnetic stimulation helps a portion of depressed patients (the authors cite response rates of 30 to 50%), but no biomarker currently predicts which patients will respond before treatment begins. A simple, inexpensive biomarker would change how patients are directed toward these treatments, which require repeated in-clinic sessions. On this site, the search for such a marker has already been discussed in relation to reward sensitivity as a predictor of response to rTMS.

The study therefore asks the following question. In depressed adults, does mismatch negativity distinguish patients from controls and reflect symptom severity, does it change under stimulation, and does its baseline amplitude predict response?

The mechanism

Why this marker

Mismatch negativity is a brain wave recorded when a rare sound interrupts a series of identical sounds: here, standard tones of 25 ms (85% of stimuli) and deviant tones of 50 ms (15%). It is automatic and does not require the subject’s attention. Drawing on the literature, the authors link it to NMDA receptor-dependent plasticity. The hypothesis is that a more plastic brain at baseline would respond better to a stimulation protocol intended to induce plasticity. A useful clarification: the study measured neither NMDA receptors nor plasticity itself, only the electrical wave recorded at a single electrode.

Element	What the study did
Duration mismatch negativity	Measured before and after stimulation in patients, once only in controls, on a single EEG channel
StatusDirect measure taken by the study
NMDA receptor-dependent plasticity	Not measured
StatusProposed interpretation of the marker, outside the scope of the study
Stimulation target	Dorsomedial prefrontal cortex, intermittent theta-burst stimulation
StatusTarget localization not uniform across patients: scalp landmark for the first 9, MRI neuronavigation for the following 37

The study at a glance

Question (PICO)
Population
Adults 18 to 54 years old with a current unipolar major depressive episode, recruited between 2015 and 2020, whose medication had been unchanged for one month (39 of 46 patients on an antidepressant). Exclusions included bipolar depression, substance use disorders and ongoing benzodiazepine treatment. 46 patients (25 randomized to active stimulation, 21 to sham) and 64 healthy controls aged 18 to 58
Intervention
Intermittent theta-burst stimulation of the dorsomedial prefrontal cortex at 90% of resting motor threshold, two sessions per day totaling 2,400 pulses per day, five days a week for two weeks
Comparator
Sham stimulation following an identical protocol: the placebo face of a mixed active/placebo coil, which blocks about 95% of the magnetic field, combined with transcutaneous electrical nerve stimulation synchronized to the pulses
Outcomes
Three questions: whether the marker differs between patients and controls and correlates with baseline symptoms, whether it changes under stimulation, and whether baseline amplitude is associated with improvement on the self-rated MADRS and the BPRS affective subscale. Follow-up assessment four weeks after enrollment
Design
Single-center sub-analysis (Uppsala University Hospital, Sweden) of a randomized, double-blind, sham-controlled trial. The parent trial was pre-registered (NCT02905604); pre-registration of the sub-analysis is not specified. 43 patients analyzed for change over time and prediction (23 and 20). Oxford level of evidence 2b for the predictive question

Quality control

Criterion	Status
Parent trial	Solid
FindingRandomized, double-blind, sham-controlled, pre-registered. According to the authors, the sub-analysis is underpowered for change in the marker and for prediction (parent trial: Bodén et al., 2021)
Power for the predictive question	Insufficient
FindingPost hoc power of 67% and 63% in the active arm, 5% and 9% in the sham arm; power for the between-arm comparison not reported
EEG measurement	Caveat
FindingSingle channel, variable interval between the last session and the follow-up measurement (median 15.5 days, interquartile range 8.5), duration of ongoing medication not accounted for
Transparency	Good
FindingOpen access, detailed methods, limitations disclosed, supplementary material available, no conflicts of interest declared; placebo coil loaned by the manufacturer

Results

r = 0.48Correlation, in the active arm only (n = 23), between baseline marker amplitude and improvement on the self-rated MADRS (95% CI 0.08 to 0.74; p = 0.021). The comparison with the sham arm is not significant (z = 1.69; p = 0.091).

Question	Result
Patients versus controls	t(108) = 0.56, p = 0.575, d = 0.11 (SE 0.19); post hoc power of 82% according to the authors, without specifying the targeted effect size
PEB readingNegative, a small difference not excluded the effect size targeted by the power calculation is not specified
Marker and baseline symptoms	r = −0.22 with the self-rated MADRS (p = 0.133); r = −0.08 with the BPRS affective subscale (p = 0.593), 46 patients
PEB readingNegative but underpowered post hoc power of 31% and 8%
Change under stimulation	Treatment × time interaction not significant: F(1,41) = 1.62, B = 0.211, SE 0.17, p = 0.210
PEB readingNegative but underpowered post hoc power of 53%
Prediction, active arm	n = 23: r = 0.48 with the self-rated MADRS (95% CI 0.08 to 0.74; p = 0.021); r = 0.46 with the BPRS affective subscale (95% CI 0.06 to 0.73; p = 0.026)
PEB readingSignal in one arm only two measures pointing in the same direction
Prediction, sham arm	n = 20: r = −0.03 with the self-rated MADRS (95% CI −0.47 to 0.41; p = 0.879); r = −0.14 with the BPRS affective subscale (95% CI −0.55 to 0.32; p = 0.555)
PEB readingSmall sample
Between-arm comparison	Fisher’s r-to-z test not significant: z = 1.69, p = 0.091 (self-rated MADRS); z = 1.94, p = 0.052 (BPRS affective)
PEB readingPrediction not demonstrated this is the test that matters

The methodological lesson extends beyond this study. To claim that a marker predicts response to an active treatment, one must show that its association with improvement differs under active treatment and under placebo. Finding a significant correlation in one arm and a non-significant correlation in the other is not enough: with about twenty patients per arm, two correlations can fall on either side of the significance threshold without being truly different. Here, the direct comparison is not significant, even though it comes close for the BPRS affective subscale (p = 0.052), which the authors describe as a trend. The title asserts a prediction; the abstract refers to potential predictive properties without mentioning this between-arm comparison. The test that would actually put the prediction to the test does not confirm it.

An imbalance existed between arms: ADHD was more common in the sham arm (7 of 21 patients versus 2 of 25). In a supplementary analysis excluding these patients, the active-arm correlations remain significant (r = 0.48, p = 0.029, and r = 0.44, p = 0.044, on 21 patients) and the sham-arm correlations remain close to zero (14 patients); the between-arm comparison is not reported for this analysis.

Critical appraisal

Domain	Judgment
Randomization and blinding	Inherited from the trial
FindingThe sub-analysis benefits from the parent trial’s design. Three of 46 patients excluded from the analyses after randomization, two of them after unblinding; ADHD more frequent in the sham arm
Status of the question	Exploratory
FindingA predictive question addressed within a sub-study, with correlations computed on small samples
Marker measurement	Caveat
FindingSingle-channel EEG and variable measurement conditions
Statistical analysis	Fragile inference
FindingThe predictive conclusion rests on a difference in significance, not a significant difference. No correction for multiple comparisons is reported
Adequacy of the conclusion	Mixed
FindingCautious discussion and a call for replication, but a title more assertive than the results

Level of evidence

Scientific65/100

Editorial70/100

PEB assessment: reasonable confidence in the absence of a difference between patients and controls, low confidence in the predictive signal. The most solid result of the study is the absence of a detected difference between depressed patients and healthy controls, with this paradigm and this measure; a small difference is not excluded. The absence of change in the marker under stimulation remains indeterminate, for lack of power. What the study suggests is a possible link between baseline amplitude and response to active stimulation. What remains a hypothesis is that this link is specific to active treatment and that it reflects plasticity.

The colleague test

What an experienced colleague would say if shown this study in two minutes, between two consultations.

Nothing to change in practice yet. The idea is appealing, a more plastic brain would respond better to stimulation, and that is not absurd. But significant in one arm and not in the other, with a between-arm comparison that does not hold up, I cannot call that a prediction.

Translation for practice: no test currently allows patients to be selected for iTBS. This study mainly teaches how to read a claim of a predictive biomarker.

What you can do with this

Continue referring patients for magnetic stimulation based on the usual clinical criteria: no routine EEG test currently predicts response.
When a study announces a predictive biomarker, look for the interaction test between the marker and the treatment arm. If it is missing or not significant, prediction has not been demonstrated.
When a patient asks whether a test exists to find out in advance whether stimulation will work: not yet, and the current leads remain exploratory.

Frequently asked questions

Is mismatch negativity abnormal in depression?

In this study, no significant difference was detected compared with healthy controls (d = 0.11), and amplitude was not correlated with baseline symptom severity. The authors report a post hoc power of 82% without specifying for which effect size: a small difference is not excluded. A meta-analysis they cite had found a difference; they point to the role of tone duration, shorter here (25 and 50 ms) than in positive studies (50 and 100 ms, or 50 and 150 ms). The result holds only for this paradigm, in this population.

Why isn’t the correlation in the active arm enough?

Because a treatment-specific prediction requires the association to be stronger under active stimulation than under sham. The direct comparison of the two correlations is not significant (p = 0.091 for the self-rated MADRS, p = 0.052 for the BPRS affective subscale).

What should be taken from the negative results?

The first, the comparison with controls, is the most informative, without excluding a small difference. The second, on change in the marker over time, is underpowered (53%): as the authors themselves acknowledge, it does not allow one to conclude there is no change.

Annotated bibliography

Source study. Anderson M, Bodén R, Wass CE, Zora H, Bengtsson J, Persson J. Auditory mismatch-negativity predicts response to dorsomedial prefrontal intermittent theta-burst stimulation in major depressive disorder. Journal of Affective Disorders. 2026;409 (September 15, 2026): article 121908. DOI 10.1016/j.jad.2026.121908 · PMID 42086176, record established in the PubMed registry on September 10, 2026. Open access under a Creative Commons licence. Funding: unrestricted grants from the Swedish Research Council (2016-02362), the Swedish Society of Medicine and the Märta and Nicke Nasvell Foundation; postdoctoral fellowship from the Swedish Brain Foundation (J. Persson); Petrus & Augusta Hedlunds Foundation (M-2024-2287, C. E. Wass). Conflicts of interest: the authors declare no conflict of interest; the acknowledgments mention that the placebo coil was loaned by MagVenture. Registration: parent trial registered on clinicaltrials.gov (NCT02905604). Supplementary material consulted: participant flow diagram (Figure S1) and Tables S1 to S3.

Parent trial. Bodén R, Bengtsson J, Thörnblom E, Struckmann W, Persson J. Dorsomedial prefrontal theta burst stimulation to treat anhedonia, avolition, and blunted affect in schizophrenia or depression, a randomized controlled trial. Journal of Affective Disorders, 2021;290:308 to 315. DOI 10.1016/j.jad.2021.04.053. The randomized trial from which the sub-analysis is drawn, and which supplies its stimulation procedures; it included patients with schizophrenia or depression, whereas the sub-analysis covers depression only.

Editorial collections

A Study Under the Microscope PEB Academy