does tms work for depression — what the trials actually show

two antidepressants in, maybe three, and someone says the word tms. the clinic website says seventy percent of patients respond. your prescriber says it's worth a look. nobody hands you the trials.

here they are. what the largest comparison ever run actually found, why the number on the brochure is not the number in the trials, what the new accelerated protocols have and have not shown, who profits from the most exciting one, and who tms has not been shown to help.

what tms is, in one paragraph

transcranial magnetic stimulation is a non-invasive treatment^[9]: a coil held against the head delivers magnetic pulses to a region of the brain, usually the left dorsolateral prefrontal cortex. the standard protocol is one session a day, five days a week, over several weeks. the two main forms are high-frequency rtms and intermittent theta burst stimulation, which delivers more pulses in less time^[11]. the “accelerated” protocols you may have read about pack many sessions into a few days^[10],[11].

the biggest comparison: 113 trials, 6,750 people

in 2019 the bmj published a network meta-analysis of every randomised trial of non-surgical brain stimulation for a major depressive episode: 113 trials, 262 treatment arms, 6,750 adults with major depressive disorder or bipolar depression^[1]. ten of the eighteen treatment strategies beat sham on response. the ones you will be offered were among them. high-frequency left rtms had an odds ratio of 3.17 against sham (95% confidence interval 2.29 to 4.37). intermittent theta burst was 3.20 (1.45 to 7.08). bilateral rtms was 4.92 (2.93 to 8.25)^[1].

an odds ratio of about three means the odds of responding were roughly three times those on sham. odds are not the same as a share of people, and the paper reports odds, so that is what this page reports. every strategy was at least as acceptable as sham, meaning people were no more likely to drop out^[1].

why the clinic number is not the trial number

there is no single tms success rate, and the trials do not report one. they report odds against sham. where response rates were actually tabulated, in the systematic review of trials in older adults, they ranged from 6.7% to 54.3% across trials, with large differences in dose and protocol^[8]. a 2026 network meta-analysis of 141 randomised trials described the effect sizes as modest, with confidence intervals that overlapped across most comparisons^[2].

the high percentages you see on clinic websites tend to come from studies with no sham arm. that is not a sneer at clinics; it is how the literature is shaped. in adolescents, for example, the pooled response figure of about 41% comes mainly from uncontrolled studies, while the only large sham-controlled trial found no benefit over sham^[9]. a sham arm is the difference between measuring the treatment and measuring the experience of being treated.

one more thing the trials say about quality. of the 141 studies in that 2026 network, 87 were rated high risk of bias, 51 had some concerns, and only three were low risk. most of the rankings were graded low or very low confidence, and the authors ask for head-to-head trials to confirm them^[2]. the evidence that tms beats sham is solid. the evidence about which protocol is best is not.

how many tms sessions does it take? the dose question

added 12 november 2026: a dose-response meta-analysis in bmc medicine pooled 108 sham-controlled trials of rtms for depression, 134 active arms and 5,621 adults, and asked how the benefit over sham changes with the dose^[15]. for the short-term result, the benefit rose and then flattened. the highest predicted benefit came at about 30,000 to 39,000 pulses in total, 1,800 to 2,200 pulses a session, 14 to 16 sessions and 2.7 to 3.1 weeks of treatment, with little extra benefit from more^[15]. at follow-up, more than a week after treatment ended, the links to dose were smaller and mostly straight lines rather than curves, with the best estimates at 26,000 to 33,000 pulses and 10 to 14 sessions^[15].

what it does not tell you is just as important. each part of the dose was modelled on its own, even though pulses, sessions and weeks rise together, and the figures describe the average across groups within the doses the trials used^[15]. the authors say plainly that these are reference points, not a dosing rule for any one person and not a single best protocol, and that lasting benefit may depend more on a maintenance plan than on a bigger dose^[15]. it cannot say whether the 90,000 pulse accelerated course below is better or worse than a standard one; that needs head-to-head trials. a question it does make reasonable: if a clinic proposes far more sessions than the trials found useful, ask what the extra is expected to add.

theta burst and the accelerated protocols

theta burst is the form most likely to be offered now, because it is faster. a 2024 network meta-analysis of 23 randomised trials in 960 people found intermittent theta burst to the left prefrontal cortex beat sham on response and on remission, and a combined left-and-right protocol did too^[3]. on safety it found no difference from sham in dropouts, in headache or discomfort at the site, or in switches to mania^[3]. the 2026 network gave theta burst the highest probability of being the most effective modality, especially in treatment-resistant depression, with the same overlapping-intervals caveat^[2].

“accelerated” means more sessions in fewer days. a meta-analysis of four randomised and ten before-and-after studies found accelerated tms improved symptoms quickly but was not significantly more effective than standard tms (standardised mean difference −0.67, 95% confidence interval −1.62 to 0.27)^[10]. there was a hint that symptoms kept improving after the accelerated course ended, which the authors say needs longer follow-up to confirm^[10]. faster, then. not clearly better.

the stanford protocol, and who profits from it

the protocol you have most likely heard about is stanford neuromodulation therapy, saint or snt: a high-dose intermittent theta burst course of ten sessions a day for five consecutive days, about 90,000 pulses, to the left dorsolateral prefrontal cortex^[5]. the replication trial, published in 2026, randomised 48 people with treatment-resistant depression to active or sham. at one month, 50.0% of the active group were in remission versus 20.8% on sham; response was 54.2% versus 25.0%^[4]. the fda cleared a device built around the protocol in september 2022, through the 510(k) route, which establishes that a device is substantially equivalent to one already on the market rather than judging how many people respond^[6].

the clearance letter is worth reading, because the regulator wrote its own caveats into it. the clinical data came from 70 patients in four trials, one of them randomised and sham-controlled with 29 people, the other three open-label, all “conducted in close geographical proximity”, so “generalizability to the broader United States population has not been evaluated”. effectiveness was measured at four weeks and “has not been established beyond the timepoints evaluated”. and only 19 of the 29 people in the sham-controlled trial completed the check on whether they could tell which arm they were in, so “uncertainty remains as to the adequacy of the blinding for the remainder”^[6]. the same record shows why the protocol is called accelerated: 1,800 pulses a session, ten sessions a day, 90,000 pulses in five days, against the predicate device's 600 a session, once a day, 18,000 over six weeks^[6].

who tms has not been shown to help

teenagers. the 2022 review found only two randomised trials in adolescents. the only large sham-controlled one found tms no more effective than sham, and the encouraging pooled figures come mainly from uncontrolled studies^[9]. added 27 october 2026: a trial of 61 adolescents in china, published in july 2026, found lower depression scores (cohen's d 0.71) and lower suicidal-ideation scores (d 0.67) than sham after ten daily sessions of high-dose theta burst, with the gap still there at one month; anxiety scores did not differ^[12]. it was single-blind, single-site and registered after it started, so it is one trial pointing the other way, not yet replicated.

people whose depression comes with insomnia. added 28 october 2026: a meta-analysis of seven randomised trials in 709 people (367 given high-frequency rtms to the left prefrontal cortex, 342 in the comparison groups) found better sleep quality scores (standardised mean difference 0.97, 95% confidence interval 0.23 to 1.71) and lower depression scores (0.61, 0.38 to 0.84) than the comparison groups, with no significant difference in adverse reactions (risk ratio 1.09, 0.80 to 1.48). sleep latency did not reach significance (0.50, from negative 0.17 to 1.18)^[13]. the limits are large: heterogeneity for the sleep scores was 91%, the authors rate the certainty as very low to low, and they call the findings exploratory and in need of confirmation by higher quality trials^[13]. a pooled signal in this group, not a settled answer.

people starting tms and a new antidepressant at the same time. added 3 november 2026: a double-blind trial in prague randomised 60 people with major depression who were all starting an ssri to one of three arms of 20 sessions: rtms over the dorsomedial prefrontal cortex, standard high-frequency rtms over the left prefrontal cortex, or sham. at six weeks depression scores had fallen in every arm (by 14.3, 8.2 and 12.6 points on the madrs), and no active arm beat sham. remission was 20%, 15% and 45%, with sham numerically highest, and 45% of the dorsomedial arm dropped out^[14]. the authors call it exploratory: too small to detect what rtms adds on top of a new ssri, with a high sham response and uneven dropout^[14]. one small trial pointing the other way from the pooled networks above, not yet replicated, and it says nothing about people whose depression has already not responded to medication.

older adults. seven randomised trials, 260 people in total, plus seven uncontrolled ones. response ranged from 6.7% to 54.3%, and the review's own conclusion is that efficacy in older adults is still unclear, with dose and protocol varying so much between trials that they are hard to add up^[8].

does the brand of machine matter

it does not appear to. across 31 randomised sham-controlled trials of high-frequency rtms, a meta-analysis found no significant differences in response, remission, or dropout among the three commercially available devices it compared^[7]. with the caveat that no trial randomised people between machines; this is a subgroup analysis of trials designed to test rtms, not the devices^[7].

what to ask before you sign up

these are questions, not advice. they are the ones the trials make it reasonable to ask.

• which protocol, exactly: standard high-frequency, theta burst, or accelerated, and how many sessions over how many days?
• the response and remission rates you quote: are they from sham-controlled trials, or from your own patients without a sham comparison?
• what counts as treatment-resistant for coverage, and how many prior medications do i need to have tried?
• what happens at three and six months, and is there a maintenance plan? the accelerated data on that is thin^[10].
• given my age, is there trial evidence for people like me, or is this extrapolated from younger adults?^[8],[9]
• if tms does not work, what is the next step, and was ect discussed?^[1]

post your question anonymously and hear from people who have been there