Published on 16 September 2026
Intranasal esketamine or rTMS for treatment-resistant depression: a reconstructed comparison
In brief
No trial has ever randomised patients between intranasal esketamine and repetitive transcranial magnetic stimulation. This work reconstructs the comparison from the individual participant data of two existing trials, THREE-D for rTMS and TRANSFORM-2 for esketamine, matched on the propensity score one-to-one-to-one into three groups of 94: 282 participants out of the 615 available, and out of the 554 retained after restriction. The primary outcome is the 17-item Hamilton score, analysed as a continuous measure at four weeks. rTMS does better than the medication arm by 5.35 points, esketamine by 2.89 points. Between the two techniques, the difference is 2.46 points in favour of rTMS, with an interval running from a 5.82-point advantage to a 0.89-point disadvantage: the unfavourable bound stays below the threshold of clinical relevance chosen by the authors, set at 3 points. This is not evidence of equivalence. It is the rejection of the hypothesis that esketamine is superior to rTMS by a clinically relevant amount. The major limitations are that the comparison between the two techniques is not randomised, that the rTMS trial had no sham stimulation arm, and that the ranking in favour of rTMS appears only after adjustment: on crude response rates, rTMS is the weakest of the three arms.
The context
The question comes up in the consulting room as soon as a patient has failed two lines of antidepressant treatment and the next step has to be chosen. Both options exist, and they differ in how patients reach them, in cost and in tolerability profile. Until now, nothing allowed them to be compared other than by setting placebo-controlled trials side by side, which amounts to comparing two measurements taken with two instruments in two different rooms. The authors write that, “To the best of our knowledge, this is the first analysis directly comparing once daily rTMS treatment with intranasal esketamine”. The wording is cautious and deserves to be kept exactly as it stands: they themselves cite an earlier naturalistic study that compared an accelerated rTMS protocol with esketamine, set aside because its stimulation protocol was experimental.
The work presented here does not remove the problem of indirect comparison, it reduces it. By going back to individual participant data rather than aggregate results, it can match patients on their baseline severity and analyse the three groups with a single model. This is the best level attainable in the absence of a dedicated trial, and it remains, by construction, an observational comparison. The authors explicitly present it as exploratory and reported it according to the STROBE guidelines, which are intended for observational studies.
The study at a glance
| Question, population, measure | |
|---|---|
| Population | |
| Adults with treatment-resistant depression, defined here as non-response to at least one evidence-based treatment, with the analysis restricted to patients who had experienced one to three antidepressant failures. Mean baseline Hamilton score of 23.5 in the rTMS arm, 28.7 in the esketamine arm and 28.9 in the medication arm before matching. | |
| Interventions | |
| Repetitive transcranial magnetic stimulation following the protocol of the THREE-D trial, a non-inferiority trial conducted in Canada from September 2013 to October 2016, whose two active arms, 10 Hz stimulation and intermittent theta-burst stimulation, were pooled into a single arm. Intranasal esketamine following the protocol of the TRANSFORM-2 trial, a superiority trial conducted from August 2015 to November 2017 at 39 sites in five countries, the United States, the Czech Republic, Germany, Poland and Spain, funded by the manufacturer of the product. | |
| Comparator | |
| The control arm of TRANSFORM-2: initiation of a new antidepressant combined with a saline placebo nasal spray, administered twice a week. No augmentation strategy is described in the publication. | |
| Primary outcome | |
| The 17-item Hamilton Depression Rating Scale score, treated as a continuous variable, after four weeks of treatment. Response, defined as a reduction of at least 50%, and remission, defined as a score below 8, as secondary outcomes. | |
| Design | |
| Secondary analysis of individual participant data from two randomised trials, combining restriction, one-to-one-to-one propensity score matching into three groups of 94 patients, and regression adjustment. The propensity score was built on three baseline severity measures only: the Hamilton scale, the MADRS and the Inventory of Depressive Symptomatology. |
The quality check
| Criterion | Judgement |
|---|---|
| Access to individual participant data | Yes |
| FindingTRANSFORM-2 data obtained through the Yale University data-sharing platform, under project number 2024-0540, THREE-D data obtained from the investigators. Matching is therefore done patient by patient, not by comparing published means. | |
| Randomisation between the two techniques | Absent |
| FindingPatients were never randomised between rTMS and esketamine. Only the comparison between esketamine and the medication arm retains randomisation, since both arms come from the same trial. Matching corrects for measured differences, not for unmeasured ones. | |
| Balance after matching | Partial |
| FindingThe authors report balance achieved on baseline severity, but persistent imbalances between the two trials in age, employment status, age at onset, anxiety comorbidity, benzodiazepine use and treatment history. These variables are handled by adjustment in the model, not by matching. | |
| Blinding in the rTMS trial | No sham arm |
| FindingThe THREE-D trial compared two active stimulation protocols, with no sham stimulation. The authors themselves acknowledge this limitation and state that it may have inflated the estimated effect of rTMS compared with a sham-controlled trial. | |
| Consistency of rating scales | Conversion |
| FindingThe primary outcome of TRANSFORM-2 was the MADRS. Scores were converted to the Hamilton scale using a published correspondence table based on percentile matching. The conversion introduces individual measurement error. A post hoc analysis running the calculation the other way, with everything converted to the MADRS, gives similar results. | |
| Length of observation | Four weeks |
| FindingA short window for rTMS, whose standard protocols last at least six weeks according to the authors. Two reasons are given: the esketamine data stop at four weeks, and weeks five and six of THREE-D were missing not at random. | |
| Analysis plan | Exploratory |
| FindingThe authors report that there was no formal non-inferiority testing framework and no pre-registered statistical analysis plan, the registration covering only the data access request. No correction for multiple comparisons was applied, on the grounds that the work is exploratory. | |
| Source of the esketamine data | Industry trial |
| FindingTRANSFORM-2 was funded by the manufacturer of the product. The present analysis is funded by a private foundation with no link to that product, and several of the authors declare support from manufacturers of stimulation equipment. The details are given in the annotated bibliography. | |
The findings
| Outcome | Value |
|---|---|
| rTMS versus medication | −5.35 points (95% CI −8.77 to −1.93; p = 0.002) |
| PEB readingThe only comparison whose point estimate exceeds the 3-point threshold, but whose interval also includes values below that threshold. A comparison between two different trials, to be interpreted bearing in mind the absence of a sham arm in THREE-D. | |
| Esketamine versus medication | −2.89 points (95% CI −5.38 to −0.40; p = 0.023) |
| PEB readingThe only comparison that retains the original randomisation, both arms coming from TRANSFORM-2. The point estimate, 2.89 points, falls just short of the 3-point threshold, and the interval contains both negligible differences and differences well above the threshold. | |
| rTMS versus esketamine | −2.46 points (95% CI −5.82 to +0.89), no p value reported |
| PEB readingA non-significant difference. The authors’ exact wording is that their findings “reject the hypothesis that esketamine is superior to rTMS by a clinically meaningful amount”, and that “rTMS is either similarly effective or superior to intranasal esketamine”. This is not a demonstration of equivalence. | |
| Adjusted mean scores at four weeks | rTMS 14.4 (11.9 to 16.9); esketamine 16.9 (14.4 to 19.3); medication 19.7 (17.4 to 22.1) |
| PEB readingEstimated marginal means on the Hamilton scale. The intervals for rTMS and esketamine overlap widely. | |
| Crude response rates | esketamine 62.8%, medication 50.0%, rTMS 47.9% |
| PEB readingOn the unadjusted data, rTMS is the weakest of the three arms, just below the medication arm. The rate of 50% on a new antidepressant combined with a placebo spray is high for a resistant population, which the authors link to the possibility that the placebo set-up of TRANSFORM-2 inflated the effect of starting an antidepressant. | |
| Crude remission rates | esketamine 33.0%, medication 22.3%, rTMS 6.4% |
| PEB readingNeither technique shows a significant increase in the probability of remission compared with medication, either in the main analysis or in the weighting sensitivity analysis. The result in favour of rTMS is therefore not found on the most demanding outcome. | |
| Odds ratios for response | rTMS 3.41 (95% CI 1.41 to 8.57); esketamine 2.08 (1.09 to 4.04), versus medication |
| PEB readingEstimates from a logistic regression adjusted for baseline covariates. The gap between the crude response rate for rTMS, the lowest of the three, and its adjusted odds ratio, the highest of the three, is carried entirely by the adjustment model. It cannot be read in the raw data. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Construction of the comparison | Observational |
| FindingTwo trials conducted in different countries, years and health systems, one in Canada between 2013 and 2016, the other in five countries between 2015 and 2017. Propensity score matching reduces measured differences, it does not recreate randomisation. The authors say so plainly. | |
| Selection of the rTMS group | Most severe subgroup |
| Finding94 patients were retained out of the 358 candidates in the rTMS arm, against 94 out of 102 for esketamine and all 94 patients in the medication arm. Since matching was on baseline severity, and the rTMS arm was on average less severe, the rTMS patients retained are the most severe in their original trial. The bearing of this point on the result is not analysed in the publication. | |
| Gap between raw data and adjusted estimates | Marked |
| FindingThe ranking of rTMS depends on adjustment: it comes last on crude response and remission rates, first on the adjusted estimates. That is mechanically possible, since several unfavourable covariates are more frequent in this arm, but it concentrates the conclusion on the validity of the model. This reading is our own appraisal; the publication does not put it this way. | |
| Asymmetry of blinding conditions | Unfavourable to esketamine |
| FindingrTMS is estimated here without subtraction of a sham arm, whereas esketamine is estimated in a trial where the comparator received a placebo spray. The authors flag both effects, which run in opposite directions: the absence of a sham arm may inflate the effect of rTMS, and the placebo set-up may inflate the effect of the medication arm. The direction of the net result is suggested, not quantified. | |
| Power | Limited |
| Finding282 patients in total. The interval of the direct comparison covers both a substantial advantage for rTMS and a modest inferiority. The authors state that the analysis of binary outcomes suffered from insufficient power due to the dichotomisation of a continuous variable. | |
| Tolerability and safety | Not analysed |
| FindingThis is not a negative result but an absence of analysis: no tolerability outcome was examined in this work. The discussion recalls the known profiles of each technique by citing other sources, which does not amount to a comparison. | |
| Sensitivity analyses | Consistent |
| FindingOverlap weighting, reverse conversion to the MADRS and restriction to patients who had failed at least two treatments all point in the same direction. In that last subgroup, esketamine gains 3.71 points and rTMS 5.28 points over the medication arm, the difference between the two techniques being 1.56 points with an interval running from 5.55 points in favour of rTMS to 2.42 points in favour of esketamine. The authors specify that balance after matching is notably worse there. | |
Level of evidence
A secondary analysis of individual participant data from randomised trials, which the authors themselves describe as exploratory and report as an observational study. The level of evidence is not graded in the publication: what follows is our editorial appraisal. Confidence is high that esketamine combined with a new antidepressant does better than that new antidepressant alone, because this comparison retains the randomisation of TRANSFORM-2 and reproduces the result of the original publication. It is moderate for the advantage of rTMS over medication, which rests on a comparison between two trials. It is moderate to low for the relative ranking of the two techniques, because of the absence of randomisation, the asymmetry of blinding conditions and the dependence of the result on adjustment. It is low for any extrapolation to longer durations, to remission outcomes or to patients more resistant than those included in the two original trials.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ So we still have no trial that randomises between the two. We have a matching exercise, which is better than nothing. What it tells us is that there is no efficacy argument for putting esketamine first. That is not the same as saying they are equivalent, and it is not an argument for putting rTMS first either. ”
What this means in practice: for a resistant patient to whom both options are accessible, expected efficacy is not enough to decide between them. The decision rests on real-world access, tolerability, organisational constraints and the patient’s preference.
What you can do with this
- Do not present esketamine as more effective than rTMS: this analysis does not support that, and this is its main contribution.
- Do not conclude the opposite either. The numerical advantage of rTMS is not significant, it disappears on crude response and remission rates, and the trial it comes from had no sham stimulation arm.
- Frame the choice for the patient in terms of constraints: in the protocols studied, rTMS is given daily over four to six weeks, with a theta-burst stimulation session lasting about three minutes, whereas esketamine is administered twice a week for four weeks with two hours of medical monitoring after each administration. Add travel and cost to that, rather than a hierarchy of efficacy.
- Before any referral, check the local conditions of access, initial prescribing and coverage for both techniques where you practise; they change over time and cannot be inferred from this study.
Frequently asked questions
Are esketamine and rTMS equivalent for treatment-resistant depression?
No. The study detects no clinically relevant difference, which is not a demonstration of equivalence. The authors write that their findings “reject the hypothesis that esketamine is superior to rTMS by a clinically meaningful amount”, and they state that they did not use “a formal non-inferiority testing framework”.
Does rTMS outperform esketamine on every outcome?
No. It comes out ahead on the adjusted primary outcome and on the adjusted odds ratio for response. On crude rates, it is the weakest of the three arms, for response as for remission. The ranking therefore depends on statistical adjustment, which calls for caution.
Does this result apply to all patients with treatment-resistant depression?
It applies to patients who had failed one to three lines of treatment, as recruited in the two original trials. The authors specify that “our results may not apply to more refractory cases”. Beyond that, there are no data here.
What does the study show about tolerability of esketamine versus rTMS?
Nothing. No tolerability outcome was analysed. This is not an absence of signal, it is an absence of analysis. Choosing on tolerability means going back to the safety data for each technique, not to this analysis.
Is a head-to-head randomised trial of esketamine and rTMS needed?
It would be decisive, and the authors explicitly call for one. In its absence, this work is the best evidence available, provided it is read as an observational and exploratory comparison, not as a trial.
Annotated bibliography
Source study. Kaster TS, Dai Y, Vila-Rodriguez F, Downar J, Daskalakis ZJ, Blumberger DM, Rhee TG. Efficacy of intranasal esketamine versus rTMS for treatment-resistant depression: analysis of individual participant data from two clinical trials. eClinicalMedicine, 2025, volume 90, article 103609, published online on 30 October 2025. DOI 10.1016/j.eclinm.2025.103609, PMID 41245530, PMCID PMC12615302. The journal does not number its issues, so the full reference comprises a volume and an article number, with no issue number or continuous pagination.
Funding. Support for this work is declared as coming from the Delaney Family Foundation. The authors state that no funder, including those of the original trials, had any role in study design, data collection, analysis, interpretation or writing. The analysis was conducted under project number 2024-0540 of the Yale University data-sharing platform, which has an agreement with Janssen Research and Development.
Competing interests. The declaration is detailed in the publication. The first author declares public institutional research support. Several co-authors declare research support and in-kind equipment support from manufacturers of stimulation devices, notably MagVenture, BrainsWay and Magnus, scientific advisory roles with companies in the sector, consulting fees, and the co-founding of a company in the mental health field. One author declares no competing interests. These links concern neuromodulation, that is, one of the two techniques being compared.
Original trials. For rTMS: Blumberger DM, Vila-Rodriguez F, Thorpe KE, et al. Effectiveness of theta burst versus high-frequency repetitive transcranial magnetic stimulation in patients with depression (THREE-D): a randomised non-inferiority trial. Lancet, 2018, volume 391, pages 1683 to 1692, DOI 10.1016/S0140-6736(18)30295-2, registration NCT01887782, funded by the Canadian Institutes of Health Research. For esketamine: Popova V, Daly EJ, Trivedi M, et al. Efficacy and safety of flexibly dosed esketamine nasal spray combined with a newly initiated oral antidepressant in treatment-resistant depression: a randomized double-blind active-controlled study. Am J Psychiatry, 2019, volume 176, pages 428 to 438, DOI 10.1176/appi.ajp.2019.19020172, registration NCT02418585, funded by Johnson and Johnson. These two references are taken from the bibliography of the publication analysed; their full texts were not reread for this analysis.
What was consulted. Verification carried out on the full text of the publication, 41 pages, including Tables 1 and 2, Figures 1 to 3 and their legends, the funding and competing interests statements and the reference list. The DOI and PMID were checked against the publication record and the publisher’s bibliographic record. The supplementary material, which contains several sensitivity analyses and the detailed results tables, was not supplied and was therefore not consulted: elements that appear only there are reported as summarised in the body of the text. The full texts of the two original trials were not reread. Article subjected to an independent double reading.