Published on 16 September 2026
Magnetic seizure therapy versus ultra-brief ECT: non-inferiority at the edge of its margin
In brief
The CREST-MST trial compared magnetic seizure therapy (MST) with right unilateral ultra-brief pulse electroconvulsive therapy (ECT) in three North American academic centres, one in Canada and two in the United States, in adults with non-psychotic major depression, 82% of whom had already failed two or more lines of treatment. The trial carries two primary outcomes, tested in sequence: remission on the HRSD-24, and a worsening of at least 25% on an autobiographical memory test. On remission, at the end of the course, 22.5% of patients reached it on MST against 27.8% on ECT, a difference of 5.3 points in favour of ECT. Non-inferiority is declared at p = 0.048 for a margin of 15 points, with an upper interval bound of 14.9: it clears the margin by a tenth of a point. The second primary outcome is far more clear-cut: worsening of autobiographical memory affected 2.7% of patients on MST against 17.3% on ECT (p = 0.0003). An analysis planned in the statistical analysis plan, with the margin tightened to 8 points to reflect an observed remission rate well below the 50% expected, is inconclusive (p = 0.32). One point needs stating at the outset: the authors themselves note that the spread of MST will depend on the regulatory approvals specific to each country. This trial alone does not turn it into an option for routine care.
The context
In the consulting room the question takes a very concrete form. A patient with treatment-resistant depression who is offered a course of ECT often gives the same answer: they are afraid for their memory. That fear is not irrational. It corresponds to a documented, sometimes lasting, effect of electrical stimulation, particularly on autobiographical memory. It leads to refusals, to interrupted courses, to indications postponed until the clinical state leaves no choice. The authors in fact open with a figure that sums up the problem: among the patients who could benefit from it, about 1% accept ECT.
Part of the answer has already come from changes in technique: the move to right unilateral stimulation and to ultra-brief pulses reduced the cognitive burden, at the cost of a persistent debate about efficacy. It is precisely this modality, the most cognitively sparing of those in actual clinical use, that serves as the comparator here. The trial therefore does not compare a new device with an outdated technique, it compares it with the best that is done on the cognitive front. That is a demanding methodological choice, and it makes the result more interesting.
Regulatory framework, as of 13 August 2026. The authors write that the availability of MST will depend on the regulatory approvals specific to each jurisdiction and on the establishment of practitioner training programmes. The trial itself was conducted under an investigational device exemption granted by the US Food and Drug Administration and under a clinical trial authorisation issued by Health Canada, which places the equipment within a research framework rather than within routine care. Device authorisation, commercial availability and reimbursement are three distinct questions, and this trial settles none of them. This reminder belongs to the context, not to the results of the study.
The mechanism
What hypothesis magnetic seizure therapy rests on
| Element | What it covers |
|---|---|
| The pathophysiological hypothesis | A seizure triggered from a more circumscribed cortical focus |
| FindingThe electrical current of ECT passes through the skull and spreads widely. A magnetic field, by contrast, is not impeded by bone and induces a current whose penetration remains more superficial. Citing earlier work, the authors recall that the magnetic seizure would be 5 to 10 times more focal than right unilateral ultra-brief ECT and would spare the medial temporal structures involved in episodic and autobiographical memory. This claim is the starting point of the trial, not one of its conclusions. | |
| What the trial measured | Remission, response, suicidal ideation, autobiographical memory |
| FindingClinical outcomes, an autobiographical memory test and an exploratory cognitive battery. These are measures of effect, not measures of mechanism: they say what happens to the patient, not where the seizure spread. | |
| What is documented about the seizure itself | Stimulation parameters, seizure durations, number of sessions |
| FindingMST is delivered with a MagPro XP device and a twin coil, in a midline frontal position over the Fz electrode, at 100% of machine output and 100 Hz, with trains increasing from 2 seconds up to a maximum of 10 seconds. ECT is delivered at six times the seizure threshold. Seizure durations are close: 36.0 seconds against 37.4 seconds for motor duration, 47.5 against 50.5 seconds for electroencephalographic duration. The mean number of sessions, on the other hand, differs markedly, 14.2 on MST against 11.4 on ECT. No imaging and no marker of intracerebral propagation is reported: the focality hypothesis is not tested here. | |
The study at a glance
| Population, intervention, comparator, outcomes | |
|---|---|
| Population | |
| Adults aged 18 and over, inpatients or outpatients, in a non-psychotic major depressive episode, with an HRSD-24 score of at least 21 at baseline. Depression with psychotic features is an exclusion criterion. Treatment resistance was not required for inclusion, but 82% of participants in each arm had already failed at least two treatments and about a third had received a course of repetitive transcranial magnetic stimulation. Three academic centres: Toronto in Canada, Dallas and San Diego in the United States. | |
| Intervention | |
| Magnetic seizure therapy, midline frontal transcranial magnetic stimulation at 100 Hz and 100% output, inducing a seizure under general anaesthesia, two to three times a week, until remission or a maximum of 21 sessions. | |
| Comparator | |
| Right unilateral ultra-brief pulse electroconvulsive therapy, at six times the seizure threshold, that is, the reference modality for cognitive preservation. The protocol prohibited switching to a bitemporal montage in case of non-response, unlike common practice. | |
| Primary outcomes | |
| Two primary outcomes, tested in a closed sequence. First, non-inferiority on remission, defined as an HRSD-24 score of 10 or less together with a reduction of at least 60% from baseline, on two consecutive assessments. Then, and only if non-inferiority is established, superiority on autobiographical memory, with worsening defined as a reduction of at least 25% in the total score of the Autobiographical Memory Test, about one standard deviation. | |
| Design | |
| Randomised non-inferiority trial, parallel groups, margin pre-specified at 15 percentage points, randomisation in permuted blocks of 2 and 4 prepared by the study statistician, patients and independent raters blinded, team administering the treatments not blinded. Prospectively registered on ClinicalTrials.gov under NCT03191058. Enrolment from 26 June 2018 to 1 March 2024. | |
| Numbers | |
| 292 patients enrolled, 239 randomised, 3 withdrawn before the first session. The modified intention-to-treat sample includes 236 patients, 119 on ECT and 117 on MST. The primary analysis is based on the sample of patients who received an adequate course, eight sessions or earlier remission, that is 219 patients, 108 on ECT and 111 on MST. Enrolment was stopped before reaching the target of 260 patients with an adequate course, a decision endorsed by the data monitoring committee and the funder. |
The quality check
| Item checked | Judgement |
|---|---|
| Pre-specified non-inferiority design | Pre-specified and registered |
| FindingThe 15-point margin, the definition of remission and the two-step testing procedure appear in the protocol and the statistical analysis plan published as supplementary material, with prospective registration. This is the minimum condition for a non-inferiority conclusion to mean anything. | |
| Blinding | Blinding measured, imperfect among cognitive raters |
| FindingPatients and independent raters are blinded, and both devices were triggered simultaneously at every session to scramble the cues. Blinding was measured: the blinding index is −0.02 (95% CI −0.14 to 0.10) among clinical raters, which is excellent, but 0.23 (0.10 to 0.35) among cognitive raters and 0.17 (0.04 to 0.28) among patients, two values that depart significantly from zero. The authors conclude that blinding was well preserved; the more cautious reading is that blinding holds on the clinical outcome and loosens on the cognitive outcome. | |
| Primary analysis population | Completers, not intention to treat |
| FindingThe analysis plan explicitly designates the completer sample as the primary analysis, on the grounds that intention to treat introduces undesirable conservatism into a non-inferiority trial. It is a reasoned choice, but it is also the least protective one. Worth noting: here, the modified intention-to-treat analysis is more favourable to the conclusion sought than the primary analysis, with a difference of 3.8 points and p = 0.021, probably because early dropouts were more frequent on ECT. | |
| Precision on the remission outcome | Upper bound 14.9 for a margin of 15 |
| FindingThe bound on which the decision rests lies at 14.9 points, for a margin set at 15. The conclusion holds by a tenth of a percentage point. This point is developed further below: it is not trivial. | |
| Funding and supply of equipment | Public funding, devices supplied by the manufacturer |
| FindingThe trial is funded by a grant from the US National Institute of Mental Health, reference R01MH112815, and the authors state that the funder played no part in the design, the analysis or the writing. The acknowledgements, however, indicate that Magventure supplied the MST-XP devices and twin coils used at the three sites, and two of the senior authors declare having received in-kind equipment support from the same manufacturer for studies they initiated. This is not industry funding of the trial, but nor is it complete independence from the manufacturer of the device under evaluation. | |
| External validity | Three expert centres, two countries |
| FindingThree academic centres well versed in the technique, patients mostly outpatients and younger than those in usual ECT cohorts, psychotic depression excluded, and switching to a bitemporal montage prohibited in case of failure. The authors acknowledge that the included population is probably not representative of the one that receives ECT today. Transfer to an ordinary care network remains to be shown. | |
The findings
| Outcome | Result |
|---|---|
| Remission at the end of the course, first primary outcome | 22.5% against 27.8%, difference of 5.3 points in favour of ECT, interval −4.4 to 14.9, non-inferiority p = 0.048 |
| PEB readingThe non-inferiority conclusion is the one defined by the protocol. It does not say that the two treatments do equally well: it says that the trial could not rule out a loss of efficacy of up to the tolerated margin. Both rates are also lower than those reported by older ECT trials, around 50%. The authors put forward several explanations: exclusion of psychotic depression, a stricter remission criterion, a lower mean age, a high proportion of resistant patients, benzodiazepine use in 55% of participants, and selection, through the blinded randomised design, of patients with few expectations of either technique. | |
| Worsening of autobiographical memory, second primary outcome | 2.7% against 17.3%, p = 0.0003 |
| PEB readingThis is the clearest difference in the trial, and it goes in the direction of the starting hypothesis. It is indeed a second primary outcome, tested in a closed procedure after non-inferiority: its statistical validity therefore does not depend on a correction for multiple comparisons. The sensitivity analyses in the supplementary material all point the same way, with the worsening threshold lowered to 20% or with missing data counted as worsening. The result nonetheless concerns one specific cognitive domain, autobiographical memory, not cognition as a whole. | |
| Analysis with the margin tightened to 8 points | p = 0.32, inconclusive result |
| PEB readingThe statistical analysis plan provided that, if the observed remission rate fell below the expected 50%, the test would be rerun with a proportionally equivalent margin, that is 30% of the rate observed on ECT. With this margin of 8 points, the interval covers both zero and the margin, and non-inferiority is not established. The authors report it in the supplementary material and mention it among the limitations. It is the most informative analysis in the trial, and it clearly tempers the main conclusion. | |
| Symptomatic response, reduction of at least 50% on the HRSD-24 | 47% (52 of 111) against 48% (52 of 108), non-inferiority established, p = 0.021 |
| PEB readingThe two arms do not differ on this outcome, and non-inferiority is established with more room than on remission. This does not demonstrate their equivalence: the test performed is a non-inferiority test, not an equivalence test, and the absence of a significant difference is not evidence of the absence of a difference. The remark holds both ways, for the experimental device as for ECT. | |
| Remission of suicidal ideation | 48% (44 of 91) against 48% (36 of 75), non-inferiority established, p = 0.033 |
| PEB readingAnalysis restricted to participants whose suicidal ideation score was not zero at baseline, with a non-inferiority margin constructed differently, at 14 points. A result consistent with the primary outcome, but still a secondary outcome in a subgroup. | |
| Tolerability and adverse events | Profile broadly in favour of MST, except for breathing |
| PEB reading2658 adverse events in total, 1447 on ECT and 1211 on MST, the great majority mild. Memory complaints, confusion, headache and jaw pain are more frequent on ECT, and reorientation time after the session is shorter on MST, 6.2 minutes against 15.3. One signal points the other way: breathing difficulties after the session are markedly more frequent on MST, in the order of 21 to 27% against 3 to 4% (p < 0.0001), which the authors link to the speed of emergence and which calls for particular anaesthetic vigilance. Thirteen serious events were reported in 10 participants, including one death from myocardial infarction about six months after the last session, during the naturalistic follow-up phase. | |
What the non-inferiority margin really commits you to
A non-inferiority trial does not ask “is the new treatment better”, but “can we rule out that it is worse by more than so much”. That “so much” is the margin, here 15 percentage points of remission. The number is not a measured quantity, it is a convention set by the investigators before the trial, and the protocol gives its justification: it was calibrated on an expected remission rate of 50% on ECT, so that a rate of 35% on MST was still judged clinically acceptable. Readers have to ask themselves whether they accept that convention: would you accept, for a patient who trusts you with an ECT indication, losing up to fifteen chances of remission in a hundred in exchange for a better memory profile? The question has no universal answer. It has individual answers, and it is the clinician with the patient who gives them, not the protocol.
The problem is that the remission rate observed on ECT was not 50% but 27.8%. A margin of 15 points calibrated on 50% represents a relative loss of 30%; the same 15-point margin applied to 27.8% would allow a relative loss of more than half. The authors had anticipated this case and provided in the analysis plan for a test with a proportional margin, that is 8 points: that test is inconclusive, p = 0.32. They acknowledge it among the limitations, adding that a more conservative margin would require an unattainable sample size. In other words, the non-inferiority of this trial is established in the strict sense of the protocol, but it cannot be separated from a margin that has become very wide in relation to the efficacy actually observed.
A word on the statistics themselves, because the presentation lends itself to confusion. The non-inferiority test is one-sided. The authors state that all non-inferiority analyses rest on the upper bound of a one-sided 95% interval, and that the intervals presented in the results, whose lower bound corresponds to a two-sided 90% interval, have an upper bound identical to that one-sided bound. The reported interval, −4.4 to 14.9, should therefore be read as follows: the bound that decides is 14.9, and it falls below the margin of 15 by a very small amount. The lower bound, −4.4, means only that the trial does not rule out a slight advantage for MST either. Put plainly, a p of 0.048 and a bound of 14.9 for a margin of 15 tell the story of a borderline non-inferiority, not of a demonstrated equivalence.
Critical appraisal
| Domain | Judgement |
|---|---|
| Randomisation process | Permuted blocks, list not disclosed |
| FindingRandomisation in permuted blocks of size 2 and 4, a schedule prepared by the study statistician and never communicated to the team during the trial. The groups are comparable at baseline on age, severity and treatment history. Two imbalances deserve mention: overall clinical severity is slightly higher in the MST arm, and benzodiazepine use during treatment is more frequent there, 61% against 49%, which the authors flag as a potential confounder working against the experimental device. | |
| Deviations from intended interventions | Treating team not blinded |
| FindingThis is the main limitation. A team that knows which treatment it is administering may, without intending to, adjust the dose, the number of sessions or the seizure threshold. The fact can be observed here: patients in the MST arm received on average 14.2 sessions against 11.4 in the ECT arm, nearly three sessions more. This gap may reflect legitimate escalation in the absence of response, but it also constitutes differential exposure to treatment and to anaesthesia, whose effect the trial cannot disentangle. | |
| Missing data | 17 patients fall short of an adequate course |
| FindingOf 236 treated patients, 219 received an adequate course, with asymmetric attrition, 11 on ECT against 6 on MST. Withdrawals for non-serious adverse events follow the same direction, 12 on ECT against 3 on MST, and withdrawals for intolerance 9 against 4. Missing data on the primary outcomes remain below the 5% threshold that would have triggered the imputation procedures planned in the protocol. The trial was also stopped before the target sample size, 219 adequate courses against 260 planned, which reduces precision without, according to the authors, changing the conclusion. | |
| Measurement of the outcome | Blinded independent raters |
| FindingPatients and independent raters blinded on outcomes that are by nature sensitive to expectation, remission as much as memory complaints, with simultaneous triggering of both devices to preserve blinding. This is the methodological strength of the trial. The reservation lies in the blinding index of the cognitive raters, significantly different from zero, which leaves room for some guessing on the very outcome where the gap is largest. | |
| Selection of the reported result | Primary outcomes in a closed testing procedure |
| FindingPrimary outcomes, margin and testing sequence are pre-specified in the protocol and analysis plan, available as supplementary material. The closed procedure, which allows the cognitive test only if non-inferiority is established, controls the type I error rate for these two outcomes without further correction. The additional cognitive measures, on the other hand, are explicitly exploratory and uncorrected for multiple comparisons: their convergence is consistent, it does not amount to an independent demonstration. | |
| Interpretation of the remission outcome | Borderline non-inferiority |
| FindingA conclusion reached a tenth of a point from the margin, with remission rates far lower than expected in both arms and a tightened-margin analysis that is inconclusive. The authors describe their result accurately in the body of the paper and in the discussion. Their summary conclusion goes further than our reading does, since they consider that the overall risk-benefit profile supports considering MST as a first-line convulsive therapy, particularly in patients who refuse ECT. | |
| Competing interests | Industry ties declared by senior authors |
| FindingThe trial is publicly funded, but the declaration of interests is provided and it is not empty: the first and last authors declare in-kind equipment support from Magventure, manufacturer of the device under evaluation, as well as ties with other neuromodulation companies, and Magventure supplied the devices at the three sites. Added to this is allegiance bias, specific to teams that develop a technique and evaluate it, which is not financial in nature but is real. | |
Level of evidence
A single, well-conducted randomised trial, level 1b on the Oxford Centre for Evidence-Based Medicine scale. Confidence is high that a difference in impact on autobiographical memory exists between the two techniques, and that it is large. It is clearly lower on comparative efficacy, on durability beyond the course, on relapse rates, and on transfer beyond three expert centres in two countries.
What is demonstrated. In this trial, in adults with non-psychotic major depression, worsening of autobiographical memory occurred far less often on MST than on right unilateral ultra-brief ECT, 2.7% against 17.3%, on a pre-specified primary outcome, tested in a closed procedure, measured by independent raters and robust to every sensitivity analysis reported. The size of the gap and the soundness of data collection make this finding hard to attribute to chance.
What is suggested. That MST offers antidepressant efficacy close to that of right unilateral ultra-brief ECT. “Suggested” is indeed the right word: non-inferiority is concluded a tenth of a point from the margin, on a completer analysis, in a population whose remission rates are far lower than those on which the margin had been calibrated, and the proportional-margin analysis planned for this scenario is inconclusive. Also suggested: that the memory sparing is explained by a more limited spread of the seizure, a hypothesis consistent with the literature cited but not tested here.
What is opinion. That this profile would be enough to make MST a first-line convulsive therapy. That is the authors’ position, and it is defensible; it cannot be demonstrated with these data alone, since it requires weighing a cognitive advantage against an efficacy outcome met by a narrow margin, a trade-off that belongs to the clinician and the patient. Our reading, and it is an opinion we stand by, is that the cognitive result justifies continuing the evaluation with replication trials and longer follow-up, but that it is not enough today to disqualify ECT, whose efficacy in severe treatment-resistant depression remains the best established in the therapeutic arsenal, particularly in the psychotic forms that this trial specifically excluded.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Fine, memory comes out of it much better, that I can see and it interests me, all the more so because it is a genuine primary outcome and not a rescue analysis. But non-inferiority passes at 0.048 with a bound of 14.9 for a margin of 15, and when they redo the calculation with a margin adapted to the rate actually observed, it no longer passes. Both arms get fewer remissions than I see in my own unit, and the magnetic arm received three more sessions. And anyway, I don’t have the machine. So it changes what I say to a patient who is afraid for their memory, it doesn’t change what I offer them. ”
What this means in practice: this trial enriches the clinical conversation about the cognitive impact of convulsive therapies and confirms that it depends heavily on the modality used. It changes no indication and, on its own, opens no new option in routine practice.
What you can do with this
- What you can tell a patient who fears ECT for their memory. That the fear concerns a real, measurable effect, and that it depends heavily on the technique used: among the ECT modalities in current use, right unilateral ultra-brief pulse stimulation is the one that best spares autobiographical memory. It is a point for discussion, not a guarantee.
- What you should not imply. That MST would be an accessible alternative. The publication does not document availability outside the trial setting, and the authors themselves make its wider use conditional on regulatory approvals specific to each country. If the question comes from the patient, the answer is that it is a device under investigation, used in this trial within a research setting.
- What you can measure. That memory complaints can be documented, before and after a course. This trial uses an autobiographical memory test as an outcome measure: the same principle applies in clinical practice, where a structured assessment is worth more than a retrospective impression, and often reassures the patient more effectively than a general explanation.
- What you watch for. Replication outside these three centres, follow-up beyond the course, relapse rates, and the behaviour of the technique in the populations excluded here, psychotic depression and older patients. These are what will tell whether the observed cognitive gap is confirmed as a lasting advantage.
- What you keep as a point of method. Non-inferiority is not equivalence, and the margin that defines it is a human choice, calibrated on an expected efficacy. When the observed efficacy is half the one used for calibration, the margin mechanically becomes more permissive, and that is exactly what the tightened-margin analysis shows here.
Frequently asked questions
Is magnetic seizure therapy available in routine practice?
At the date of our verification, 13 August 2026, the publication did not document any availability outside the trial setting. The trial’s authors themselves state that the availability of MST will depend on the regulatory approvals specific to each country and on practitioner training. Device authorisation, actual presence in treatment centres and reimbursement are three distinct questions.
Does non-inferior mean just as effective?
No. It means that the trial was able to rule out a loss of efficacy greater than a threshold decided in advance, here fifteen points of remission. A non-inferiority concluded by a narrow margin leaves open the possibility of slightly lower efficacy. Demonstrating equivalence would require a different design and larger numbers.
Is the cognitive advantage of MST over ECT demonstrated?
For autobiographical memory, yes, in the sense that it is a pre-specified primary outcome, tested after non-inferiority by a procedure that controls the risk of statistical error, measured by independent raters, with a large gap that is robust to the sensitivity analyses. Two reservations remain. The result concerns one specific cognitive domain, not cognition as a whole, and the blinding of the cognitive raters, measured in the trial, is less well preserved than that of the clinical raters.
Is magnetic seizure therapy safer than ECT overall?
Not uniformly. The profile is broadly more favourable, with fewer memory complaints, less confusion, fewer headaches and faster recovery. One signal points the other way: breathing difficulties after the session were markedly more frequent on MST, which the authors attribute to the speed with which consciousness returns while neuromuscular blockade has not fully worn off. It is a point of anaesthetic vigilance, not a point of doctrine.
Why were remission rates so low in both arms of the CREST-MST trial?
22.5% and 27.8% are below the 50% on which the sample size calculation had been built. The authors put forward several converging explanations: exclusion of psychotic depression, which responds better to ECT, a stricter remission criterion than in older trials, a lower mean age, a high degree of resistance with a third of patients already treated with repetitive transcranial magnetic stimulation, benzodiazepine use in 55% of participants, the protocol ban on switching to a bitemporal montage, and selection, through the blinded design, of patients with few expectations of either technique. None of these explanations is tested, they are put forward in the discussion.
Annotated bibliography
Source study. Blumberger DM, McClintock SM, Thorpe KE, et al. Confirmatory efficacy and safety trial of magnetic seizure therapy versus right unilateral ultra-brief electroconvulsive therapy in depression (CREST-MST): a randomised, double-blind, non-inferiority trial in Canada and the USA. The Lancet Psychiatry 2026; 13 (5): 376-386. DOI 10.1016/S2215-0366(26)00060-X. PMID 41997695. Randomised non-inferiority trial, three centres in Canada and the United States, 292 patients enrolled and 239 randomised, MST against right unilateral ultra-brief pulse ECT. Contribution: the largest randomised comparison of these two techniques to date, with two primary outcomes including a structured cognitive outcome, and a protocol and statistical analysis plan published as supplementary material. Limitations: non-inferiority conclusion reached a tenth of a point from the margin, primary analysis in completers, target sample size not reached, team administering the treatments not blinded, differential exposure to treatment, psychotic depression excluded. The online version was the subject of two publisher corrections, the corrected version appearing on 16 April and then on 7 July 2026; that is the version that was read.
Trial registration. ClinicalTrials.gov, NCT03191058. Contribution: makes it possible to check that the primary outcomes and the margin were not changed after the fact, which reading the protocol and statistical analysis plan appended to the publication confirms. Limitation: registration guarantees neither the quality of the trial’s conduct nor the absence of selection among the secondary outcomes reported.
Accompanying editorial comment. Kayser S, Fallgatter AJ, Plewnia C. Magnetic seizure therapy: balancing efficacy and cognition. The Lancet Psychiatry 2026; 13 (5): 366-367. Contribution: the comment the journal published in the same issue, which weighs precisely the cognitive gain against the level of efficacy achieved. Limitation: it is an editorial, with no data of its own.
Published correspondence. Stubberud J. Cognitive safety after magnetic seizure therapy: from group differences to patient-level risk. The Lancet Psychiatry 2026; 13 (8): 643. Contribution: discusses the move from a group difference in autobiographical memory to an estimate of risk for an individual patient, a central question for use in consultation. Limitation: a letter, with no original analysis of the trial data.