A randomized, sham-controlled trial of individualized accelerated intermittent theta-burst stimulation for treatment-resistant depression, published September 16 in the Journal of Affective Disorders, found no statistically significant difference between active stimulation and sham on either its primary or secondary outcome. Fifty-one patients were randomized across three arms: aiTBS individualized to each patient’s resting-state brain connectivity, aiTBS individualized to both connectivity and an EEG-derived oscillatory frequency, and sham. Forty-three completed the protocol. Neither active arm separated from sham on the Montgomery-Åsberg Depression Rating Scale at one week, the trial’s primary endpoint, or on the Dimensional Anhedonia Rating Scale, its secondary measure.
That result runs against the direction the field’s most-discussed 2026 neuromodulation trial pointed. In June, a Mass General Brigham-led team published a JAMA Psychiatry trial finding that individually targeted, connectivity-guided accelerated TMS produced an 80 percent response rate against 60 percent for standard scalp-landmark targeting, a result read at the time as strong evidence that personalization earns its added cost and complexity. The two trials are not testing the same claim, and treating this new result as a straightforward contradiction would overstate what either study shows. The distinction between what each trial compared is the whole story.
Two different questions, not one repeated experiment
The Mass General Brigham trial compared two active delivery methods against each other: imaging-guided targeting against scalp-based targeting. Both arms received real stimulation; the trial answered whether personalizing where the stimulation goes changes the outcome, without needing to establish that stimulation itself beats placebo. That design has a clear advantage: a head-to-head comparison of two active treatments sidesteps the sham-response and expectancy problems that have dogged device trials generally, because both groups know they are being treated.
The UC San Diego trial asked a different, harder question: does an individualized aiTBS protocol, personalized twice over, to a patient’s connectivity and to their own EEG frequency, beat a sham condition designed to feel indistinguishable from active treatment. That is the comparison a device has to win to demonstrate it does anything at all beyond the expectation of treatment. On this trial’s evidence, individualized aiTBS did not win it. The two trials therefore answer different parts of the same larger question, personalization improves outcomes relative to a worse active comparator in one study, and personalization does not clearly beat placebo in another. One trial did not confirm what the other refuted; they tested different claims.
What the null result does and does not establish
A negative sham-controlled trial is a meaningful data point, not proof that stimulation individualization has no effect. Fifty-one randomized patients split across three arms is a modest sample for detecting a moderate effect size, and a trial this size can fail to detect a smaller true benefit as easily as it can correctly find no benefit at all. The paper does not claim otherwise, and neither does this piece. What the result does establish is narrower and still useful: the specific individualization approach tested here, in this population, at this dose and duration, produced no measurable antidepressant signal beyond sham over a one-week endpoint. That is a fact about this trial, not a verdict on personalized neuromodulation as a category.
It also reopens a question the Mass General Brigham design was structured to avoid. If the clearest recent evidence for personalization comes from a comparison that never included a placebo arm, and the trial that did include one found nothing, the field does not yet have a sham-controlled demonstration that connectivity- or frequency-guided individualization produces an antidepressant effect on its own. It has evidence that personalized targeting can outperform a weaker active comparator, and separately, one negative test of whether personalization outperforms placebo. Both can be true simultaneously, and neither settles the underlying mechanistic question.
Who ran this trial, and what they have a stake in
Lawrence Appelbaum, the paper’s lead author and director of UC San Diego’s Human Performance Optimization Lab, discloses that he co-founded Hybrid Harmony LLC, a neurotechnology company, and has consulted for Wave Neuroscience, a company that markets individualized, EEG-guided neuromodulation protocols. Senior author Zafiris Daskalakis discloses research and equipment support from MagVenture and a seat on BrainsWay’s scientific advisory board. Co-author Paul Fitzgerald discloses equipment support from BrainsWay and founder or board roles at TMS Clinics Australia and Resonance Therapeutics. None of those relationships is concealed; they appear in the paper’s own disclosure statement. They are worth naming here because a negative trial published by authors with active financial ties to individualized-stimulation and device companies is not the outcome those ties would predict, which if anything strengthens confidence that the result reflects the data and not an incentive to report a positive one.
What to watch
A larger, adequately powered sham-controlled trial of connectivity- or frequency-guided aiTBS would be the next test of whether this trial’s null result holds up or reflects sample size. Watch also for whether the Mass General Brigham group, or others working from the active-versus-active design, eventually run a version of their comparison with a sham arm included; that would close the gap between the two 2026 results directly, instead of leaving them to answer adjacent questions. Until then, the accurate summary of where personalized neuromodulation stands is narrower than either trial alone suggests: it can beat a worse active comparator, and it has not yet beaten placebo in the one trial built to test that specifically.