Evidence Trace, complete trial record

In plain terms

In plain terms: Transcranial Magnetic Stimulation (TMS) for Depression has strong evidence from human clinical trials. The benefit shown in those trials is if that describes your situation.

Transcranial Magnetic Stimulation (TMS) for Depression

A Replicated: replicated trials on a qualifying endpoint

Category Devices

Evidence Randomized controlled trials + meta-analysis of randomized trials

Reviewed September 2026

PMIDs 17573044 · 20439832 · 25655160 · 29946106 · 23507264 · 28030740 · 30917990 · 31901449 · 23399312 · 29955803 · 25683231 · 35044731 · 41997374 · 42383744 · 42418203

The endpoint this grade is based on: Validated scale scored by a clinician blinded to treatment

The evidence as the record states it

Evidence record

PMID 17573044 PMID 20439832 PMID 25655160 PMID 29946106 PMID 23507264 PMID 28030740 PMID 30917990 PMID 31901449 PMID 23399312 PMID 29955803 PMID 25683231 PMID 35044731 PMID 41997374 PMID 42383744 PMID 42418203

George et al. (Arch Gen Psychiatry, 2010; PMID 20439832): NIH-funded trial with an active sham that reproduced scalp sensation; remission 14.1% versus 5.1%, number needed to treat 12, masking held. O'Reardon et al. (Biol Psychiatry, 2007; PMID 17573044): 301 patients, sham-controlled; remission about double with active treatment, with the primary MADRS comparison significant only after a baseline adjustment. Levkovitz et al. (World Psychiatry, 2015; PMID 25655160): 212 patients, deep TMS with the H1 coil at 18 Hz, a different coil and frequency from the figure-8 trials; remission 32.6% versus 14.6%; sponsored by the coil's maker (Brainsway), whose developers are among the authors. Kaster et al. (Neuropsychopharmacology, 2018; PMID 29946106): 52 older adults aged 60 to 85 with late-life depression, H1-coil deep TMS; remission 40.0% versus 14.8%; funded by the Canadian Institutes of Health Research, with in-kind equipment support from the coil's maker declared. Berlim et al. (Psychol Med, 2014; PMID 23507264): meta-analysis of 29 trials, 1,371 patients; remission 18.6% versus 5%, number needed to treat 8. Brunoni et al. (JAMA Psychiatry, 2017; PMID 28030740) and Mutz et al. (BMJ, 2019; PMID 30917990): network meta-analyses of 81 and 113 trials; high-frequency left prefrontal stimulation response odds ratio about 3. Lefaucheur et al. (Clin Neurophysiol, 2020; PMID 31901449): European guideline, Level A. Berlim et al. (Int J Neuropsychopharmacol, 2013; PMID 23399312): blinding held in the trials that tested it. Kept in the record: Yesavage et al. (JAMA Psychiatry, 2018; PMID 29955803), 164 veterans, remission 40.7% versus 37.4%, no difference; Kedzior et al. (Depress Anxiety, 2015; PMID 25683231), effect smaller 8 to 16 weeks after treatment ends; Mehta et al. (J Clin Psychiatry, 2022; PMID 35044731), a non-significant reduction in suicidal ideation in randomized trials (g 0.16, confidence interval crossing zero) and a significant one only in uncontrolled studies. Pooled analyses through 2026: Eren et al. (J Affect Disord, 2026; PMID 41997374), 30 sham-controlled comparisons and 1,850 patients, response odds ratio 4.78 and remission 3.55, with larger effects in single-blind than double-blind trials; Zhang et al. (J Evid Based Med, 2026; PMID 42383744), network meta-analysis of 141 trials and 10,587 patients, high-frequency left stimulation response odds ratio 3.34, with only 3 trials at low risk of bias and most comparisons rated low or very low confidence. Orbo et al. (JAMA Netw Open, 2026; PMID 42418203): 73 patients in North Norway, 10 once-daily sessions of theta-burst stimulation against a validated sham coil; better than sham on blinded MADRS at day 10 (3.6 points), not on the self-rated BDI-II, and not sustained at 4 weeks.

Why Grade A and not higher

Grade A requires a hard clinical endpoint or objectively measured functional outcome, independently replicated across at least two trials. Transcranial Magnetic Stimulation (TMS) for Depression meets that standard based on the trials cited above. A downgrade to Grade B would follow if replication attempts produced null results, if the effect size proved substantially smaller in a broader population, or if the studied dose proved impractical to sustain.

What this grade covers and what it does not

Grade A under the rule as revised in September 2026, which counts a validated outcome scale scored by a clinician blinded to treatment alongside hard outcomes and instrument-measured function. The Hamilton and Montgomery-Asberg scales are scored by trained raters who do not know which treatment the patient received; they are not questionnaires the patient fills in. On those scales the effect has been replicated across more than 100 randomized trials in many countries, with consistent odds ratios near 3 for response, a European guideline rating of Level A, and an active sham in the NIH trial that held the blind. Two of the supporting trials used H1-coil deep TMS at 18 Hz (Levkovitz et al., 2015; PMID 25655160; Kaster et al., 2018; PMID 29946106), a different coil and frequency from the figure-8 trials; the letter rests on the figure-8 and theta-burst trials and their pooled analyses, with the deep TMS trials as supporting context, and the deep TMS pivotal trial was sponsored by the coil's maker. WHAT THIS GRADE DOES NOT COVER: no randomized trial shows fewer suicide attempts, hospitalizations or relapses, and a pooled analysis found a non-significant reduction in suicidal ideation in randomized trials (Mehta et al., 2022; PMID 35044731). Trial quality is uneven: in a 2026 network meta-analysis of 141 trials only 3 were at low risk of bias and most comparisons were rated low or very low confidence (Zhang et al., 2026; PMID 42383744), and effects are larger in single-blind than double-blind trials (Eren et al., 2026; PMID 41997374). The absolute gain is modest, about one extra remission for every 8 to 12 people treated, and it fades in the months after a course ends. A 164-patient Veterans Affairs trial found no difference from sham. Accelerated protocols (SAINT and similar) rest on far fewer trials and are not what this letter describes. It covers adults with major depression that did not respond to at least one antidepressant, treated with high-frequency or theta-burst stimulation of the left prefrontal cortex.

What would move this grade

A null result in a well-powered independent replication trial, or evidence that the effect does not hold at a practically sustainable dose, would prompt downgrade to Grade B.

How this was studied

Form: High-frequency (10 Hz) or intermittent theta-burst stimulation of the left dorsolateral prefrontal cortex with a figure-8 coil; H1-coil deep TMS at 18 Hz in two supporting trials · Dose: One session a day, 3 to 40 minutes depending on protocol, 5 days a week · Population: Adults with major depression that did not respond to at least one adequate antidepressant trial · Duration: 4 to 6 weeks (20 to 30 sessions) in the acute course · What is sold differs: Accelerated multi-session-per-day protocols are marketed as faster versions of the same treatment; they rest on far fewer randomized trials and this grade does not describe them.

How to track whether this is working A clinician-rated depression scale (HAM-D or MADRS) at baseline, weekly during the course, and at 1, 3 and 6 months after it ends, because the trials found the effect fades without maintenance.