Published on 15 September 2026
Neuromodulation for depression: what does a ranking of fourteen protocols prove?
In brief
A frequentist network meta-analysis brings together 129 randomised controlled trials, 272 treatment arms and 7,667 patients with a depressive episode, unipolar or bipolar, to compare fourteen non-invasive neuromodulation strategies with one another and with sham stimulation. Two primary outcomes structure the network: response for efficacy, all-cause discontinuation for tolerability. Ten strategies out of fourteen do significantly better than sham on response. Transcranial focused ultrasound stimulation holds the top rank with an odds ratio of 7.24 (95% CI 1.35 to 38.47), on two trials and 42 patients in total. Then come bilateral repetitive transcranial magnetic stimulation, 5.75 (3.29 to 10.07) on 13 trials, and bilateral theta burst stimulation, 5.37 (2.51 to 11.36) on 5 trials. High-frequency rTMS over the left dorsolateral prefrontal cortex, 3.42 (2.51 to 4.66), remains the best documented strategy of the network, present in 63 of the 129 trials, and its interval is the narrowest of all. The single point estimate below 1 belongs to low-frequency rTMS applied over the left dorsolateral prefrontal cortex, 0.35, but with an interval running from 0.04 to 3.46 and a p value of 0.356: an imprecise estimate, not an established unfavourable result. On tolerability, no active strategy separates itself from sham stimulation. Funding is public and no competing interest is declared. The limits lie in what the network contains: branches carried by one or two trials, 87.6% of trials at low or unclear risk of bias, and stimulation parameters more conservative than those used in practice today. A ranking of fourteen techniques is not fourteen demonstrations.
The context
The question this work asks is the one every psychiatrist meets at the moment of referral. A patient with depression, often after two or three insufficient pharmacological lines, is sent to a neuromodulation service. The patient comes back with a protocol. Which one, and on what reasoning? The answer depends mostly on what the centre offers and on its habits, and the referring clinician rarely has an argument to hand for asking for something else.
The literature does not help much, because it is fragmented. Each technique has its trials, its meta-analyses, its advocates. Head-to-head comparisons between two active protocols are rare, which the authors identify from the outset as the major limitation of the field. Sham stimulation remains the dominant comparator, which produces a collection of placebo-controlled results that are hard to set against one another. Add to that the inflation of acronyms: high-frequency or low-frequency rTMS, left, right or bilateral, continuous, intermittent or bilateral theta burst stimulation, accelerated, deep, priming or synchronised protocols, transcranial direct current stimulation, and now focused ultrasound. A clinician who does not work in a stimulation service has no reason to know how to rank that list.
A network meta-analysis is the tool designed for this situation: it links trials that never compared themselves directly, through a shared comparator, and produces a relative estimate for each pair of treatments. The authors claim two gains over the previous network synthesis in the field: the exclusion of trials that started a pharmacological or psychotherapeutic treatment at the same time, and analyses stratified on resistance and on the setting of administration. It is precisely because the tool answers the clinical demand so well that its result has to be read with method. A ranking reads quickly. It is overread more quickly still.
The methodological point
What a rank measures, and what it does not
The ranking produced by a network meta-analysis, here the surface under the cumulative ranking curve, is a probability of position, not an effect size. It answers the question: given the estimated effects and their uncertainty, what is the probability that this technique occupies the best ranks? Two consequences follow.
The first is that a high rank can rest on a very imprecise estimate. A technique assessed in two small trials can display a spectacular point odds ratio, with a confidence interval so wide that it covers both a major effect and an effect barely above the comparator. The rank does not show that width: it condenses the distribution into a single figure that has the appearance of a conclusion. In the network presented here, the contrast is explicit. Transcranial focused ultrasound stimulation holds the top rank on two trials and 42 patients, with an interval running from 1.35 to 38.47. High-frequency rTMS over the left prefrontal cortex holds a middle position on sixty-three trials, with an interval running from 2.51 to 4.66. No ranking tool flags that difference by itself: a rank built on two trials and a rank built on sixty-three trials look alike on the screen, and are not alike in what they permit.
The second consequence concerns the transitivity assumption. Comparing A and B indirectly through sham assumes that the A against sham trials and the B against sham trials were similar enough to be set in relation: the same populations, the same definitions of response, the same blinding conditions. In neuromodulation this assumption is under strain, and the reason is the placebo condition itself. Sham stimulation does not make blinding watertight, and it is not a stable entity from one trial to the next. The authors note that about half the trials use a coil tilted at 90 degrees, the others dedicated sham coils, tilts of 30 to 45 degrees or non-magnetic rotating devices. They recall that a tilt of 45 degrees can still induce motor evoked potentials, where a tilt of 90 degrees is considered neurologically inert. These conditions produce neither the same sensory experience nor the same expectation, and expectation weighs heavily on the depressive response. A patient who feels a scalp twitch and a patient who feels nothing are not blinded to the same degree, whatever the protocol says. The formal assessment of network coherence remains reassuring, with no significant incoherence except for one loop on efficacy, but that test has limited power when the branches are thin.
A third point governs any clinical reading: a point estimate below 1 against sham, such as the one reported here for low-frequency rTMS on the left, only becomes a result if its interval excludes the null value. That is not the case here. With an interval from 0.04 to 3.46, the datum is compatible with a frankly unfavourable effect and with a moderately favourable one. This is neither a signal of harm nor a demonstration of inefficacy. It is a branch of the network where we do not know.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Adults aged 18 years or over with a unipolar or bipolar depressive episode, diagnosed according to DSM or ICD criteria, included in randomised controlled trials of parallel-group or cross-over design. 129 trials, 272 arms, 7,667 patients, mean age 43.5 years, 54.4% women. Trials that started a pharmacological or psychotherapeutic treatment at the same time are excluded. | |
| Intervention | |
| Fourteen active strategies: high-frequency rTMS on the left and on the right, low-frequency rTMS on the left and on the right, bilateral rTMS, deep TMS, accelerated TMS, priming TMS and synchronised TMS, continuous, intermittent and bilateral theta burst stimulation, transcranial direct current stimulation, transcranial focused ultrasound stimulation. Electroconvulsive therapy does not enter this network. | |
| Comparator | |
| Sham stimulation, the reference comparator of the network, and direct comparisons between active strategies where they exist. | |
| Outcomes | |
| Efficacy: response rate on a clinician-administered scale, the 17-item HDRS preferred, or the MADRS, on an intention-to-treat basis. Tolerability: all-cause discontinuation rate. Secondary outcome: remission. Ranking by SUCRA. Analyses stratified on treatment-resistant depression and on the setting of administration, monotherapy or add-on therapy. | |
| Design | |
| Systematic review and random-effects network meta-analysis, frequentist model, Stata MP 14.0, PRISMA reporting, prospective registration of the protocol. Embase, PubMed/MEDLINE and Web of Science searched from inception to 27 August 2025. Level of evidence CEBM 1a. |
Quality control
| Criterion | Status |
|---|---|
| Size and coverage of the network | Sound |
| Finding129 randomised trials, 272 arms and 7,667 patients for fourteen active strategies and sham stimulation. Response is available on 208 arms from 98 trials, all-cause discontinuation on all 272 arms, remission on 122 arms from 58 trials only. | |
| Search and selection | Sound |
| FindingThree databases searched to 27 August 2025, 10,486 references identified and 43 from other sources, 342 full texts assessed, 129 retained. The search was run in duplicate, with disagreements settled by a third reviewer. Trials that started a drug or a psychotherapy at the same time are set aside, which isolates the effect of the neuromodulation itself. | |
| Risk of bias and certainty | Reservation |
| FindingCochrane tool across six domains: 42.6% of trials at low risk, 45.0% unclear, 12.4% high, that is 113 trials out of 129 at low or unclear risk. The reasons for uncertainty are insufficient reporting of randomisation and allocation, and blinding. Under GRADE, certainty is moderate to high for 13 comparisons out of 14, and low for high-frequency rTMS on the right because of an incoherence. The limitations section of the publication also assigns to those 58 trials at unclear risk a figure of 42.6%, which contradicts the 45.0% given in the results section. | |
| Coherence of the network | Verified |
| FindingLoop-based incoherence analyses are non-significant for efficacy as for tolerability, with the exception of the loop linking high-frequency rTMS on the left, low-frequency rTMS on the right and bilateral rTMS on efficacy. This test loses its power exactly where the branches are thin. | |
| Small-study effects | Tested |
| FindingComparison-adjusted funnel plots, symmetrical for all-cause discontinuation and for remission, slightly asymmetrical for response because of small trials of intermittent theta burst stimulation and of high-frequency rTMS on the left with a marked effect. Egger test non-significant for every outcome. That test stays weakly powered in a field where small positive trials abound. | |
| Balance of the network across branches | Reservation |
| Finding63 trials cover high-frequency rTMS on the left and 28 cover transcranial direct current stimulation, whereas bilateral theta burst stimulation has 5, low-frequency rTMS on the left has 4, focused ultrasound and priming TMS have 2 each, and high-frequency rTMS on the right has a single one. The ranking therefore mixes very unequal levels of information. | |
| Homogeneity of stimulation parameters | Reservation |
| FindingAcross the 123 active TMS arms, the most frequent settings are 10 Hz (36%), 110% of the motor threshold (29%), two weeks of treatment (29%) and ten sessions in total (35%). 91% of trials last six weeks or less and 92% deliver fewer than twenty sessions. The authors themselves point out that these parameters fall short of contemporary practice, which runs closer to 120% of the motor threshold and twenty to thirty sessions over four to six weeks. | |
| Time horizon | No data |
| FindingThe analysis covers the acute phase only. Neither maintenance of the effect nor the continuation and maintenance phases are assessed, a limitation the authors acknowledge. | |
| Independence and competing interests | Sound |
| FindingChinese public funding, from the National Natural Science Foundation of China, the China Postdoctoral Science Foundation and the Natural Science Foundation of Sichuan Province. The authors declare no competing interests. In a field where the device industry is very present, this point is not decorative. | |
The findings
The table below carries the estimates reported against sham stimulation on the response outcome. The column giving the volume of trials is as informative as the column giving the odds ratio, and should be read first.
| Strategy | Odds ratio against sham (95% CI) | Basis in the network | What the figure allows |
|---|---|---|---|
| Transcranial focused ultrasound stimulation | 7.24 (1.35 to 38.47) | 2 trials, 42 patients | Preliminary signal. The highest rank, the thinnest information in the network. It supports no referral outside a research protocol. |
| Bilateral rTMS | 5.75 (3.29 to 10.07) | 13 trials | High and significant estimate, consistent in the resistant subgroup. A serious candidate, to be confirmed by direct comparison. |
| Bilateral theta burst stimulation | 5.37 (2.51 to 11.36) | 5 trials | The same reading as bilateral rTMS, on a narrower base. Shorter sessions, which is a logistical argument and not an efficacy argument. |
| Priming TMS | 4.48 (1.60 to 12.68) | 2 trials | A protocol little used in routine practice. Significant but very imprecise estimate. |
| High-frequency rTMS on the left | 3.42 (2.51 to 4.66) | 63 trials | The best supported and the most precise estimate of the network. A middle rank, with more confidence than the ranks placed above it. |
| Low-frequency rTMS on the right | 3.32 (1.86 to 5.93) | 10 trials | A significant result, often forgotten in discussions centred on high-frequency stimulation on the left. |
| Accelerated TMS | 3.32 (1.21 to 9.21) | 3 trials | Significant by a narrow margin, wide interval. The accelerated format is not to be judged on this estimate alone. |
| Intermittent theta burst stimulation | 3.16 (1.86 to 5.37) | 15 trials | Efficacy comparable to that of high-frequency rTMS on the left, on a solid base for a recent protocol. |
| Deep TMS | 2.83 (1.23 to 6.42) | 3 trials | Significant, imprecise. The device differs, and so does the comparability of the sham arms. |
| Transcranial direct current stimulation | 1.99 (1.31 to 3.03) | 28 trials | A more modest effect but a well documented one, and the most accessible technique of the network. |
| Synchronised TMS | 2.20 (0.75 to 6.42), p = 0.150 | 2 trials | Non-significant. Absence of demonstration, not demonstration of absence. |
| High-frequency rTMS on the right | 1.75 (0.36 to 8.58), p = 0.489 | 1 trial | Non-significant, GRADE certainty low because of an incoherence. Nothing conclusive. |
| Continuous theta burst stimulation | 1.22 (0.30 to 5.00), p = 0.782 | 2 trials | Non-significant. The ranking places it among the least effective, on two trials. |
| Low-frequency rTMS on the left | 0.35 (0.04 to 3.46), p = 0.356 | 4 trials | The only point estimate below 1, and non-significant. A branch where uncertainty dominates, neither a signal of harm nor proof of inefficacy. |
In the subgroup of treatment-resistant depression, seven strategies do better than sham stimulation: bilateral rTMS 5.70 (3.32 to 9.68), bilateral theta burst stimulation 4.90 (2.44 to 9.97), intermittent theta burst stimulation 4.76 (2.66 to 8.41), priming TMS 4.62 (1.95 to 10.80), high-frequency rTMS on the left 3.53 (2.56 to 4.90), low-frequency rTMS on the right 3.42 (1.90 to 6.23) and deep TMS 2.77 (1.40 to 5.42). The bilateral protocols therefore stay in front when the network is restricted to the population that actually generates the demand for neuromodulation. That consistency is an argument for internal robustness. It remains internal to the synthesis and does not replace a head-to-head trial between active protocols.
The stratification on the setting of administration produces a result the referral letter can use. As an add-on to ongoing treatment, efficacy is better for bilateral rTMS 5.37 (2.66 to 10.80), priming TMS 4.95 (1.32 to 18.54), high-frequency rTMS on the left 3.49 (2.20 to 5.47) and intermittent theta burst stimulation 3.06 (1.67 to 5.70). As monotherapy, the highest response rates go to bilateral theta burst stimulation 9.87 (3.19 to 30.27), low-frequency rTMS on the right 7.77 (2.01 to 29.96) and transcranial direct current stimulation 3.46 (2.03 to 5.87). The abstract states the same asymmetry in the authors’ own words: “Bilateral TBS showed the highest response rate when administered as monotherapy, whereas bilateral rTMS was most effective as add-on therapy”. These contrasts rest on subsets of trials and on indirect comparisons: they orient a hypothesis, they do not fix a rule of indication.
On tolerability the result is simple and reassuring: no active strategy separates itself from sham stimulation on the all-cause discontinuation rate, and the authors place bilateral theta burst stimulation at the lowest discontinuation rate. One strategy alone, low-frequency rTMS on the right, is ranked less well tolerated than sham, with no significant difference in discontinuation rates between the other interventions. This result should be read for what it is: all-cause discontinuation is a coarse indicator of acceptability, which says nothing about headache, nothing about discomfort at the stimulation site, and nothing about rare events. Remission, the secondary outcome, is available in only 58 trials out of 129, with the best ranking probabilities going to bilateral rTMS, priming TMS and bilateral theta burst stimulation. The convergence between response and remission supports the bilateral protocols, on a thinner base.
The result concerning low-frequency rTMS on the left deserves a clarification, because it lends itself to misreading. It does not concern low-frequency stimulation in general: applied over the right dorsolateral prefrontal cortex, the usual target of that protocol, low-frequency stimulation does significantly better than sham. It is low-frequency stimulation applied on the left, four trials in this network, that produces a point estimate of 0.35 with an interval running from 0.04 to 3.46. That branch concludes nothing, in either direction. The neurobiological rationale the authors invoke, left prefrontal hypoactivity and right hyperactivity, would in fact predict that inhibition on the left is ineffective, but a coherent pathophysiological hypothesis does not replace an interval that excludes the null value.
A last point of reading concerns the best documented strategy. The authors note that the parameters used in the trials of high-frequency rTMS on the left are more conservative than those of current practice, and suggest that its observed effect may underestimate its real clinical efficacy. This is an authors’ hypothesis, stated as such, and not a result of the synthesis. It nevertheless has a practical consequence: comparing a protocol delivered as ten sessions over two weeks with recent protocols delivered more intensively is a comparison partly biased by dose.
Critical appraisal
| Domain | Judgement |
|---|---|
| Overall level of evidence | Established |
| PEB readingA synthesis of randomised trials, PRISMA reporting, prospective registration, risk of bias and GRADE certainty documented, independent funding. The methodological frame is the highest available for this type of question. | |
| Hierarchy between techniques | Suggested |
| PEB readingThe ranking orients a hypothesis of superiority for the bilateral protocols, and the authors themselves speak of preliminary evidence. It does not demonstrate that superiority, for want of sufficient direct comparisons between active arms. | |
| Top rank of the network | Not conclusive |
| PEB readingTwo trials and 42 patients do not found a rank. The position of focused ultrasound must be treated as a research signal, with the novelty effect that attaches to emerging and rarely replicated techniques. The authors say so explicitly and do not harden their conclusion. | |
| Estimate for low-frequency stimulation on the left | Not conclusive |
| PEB readingAn interval from 0.04 to 3.46 on four trials: this branch allows neither a recommendation against nor a recommendation for. Presenting it as demonstrated inferiority would be a common reading error on this type of table. | |
| Comparative tolerability | Analysed, not discriminating |
| PEB readingAll-cause discontinuation is indeed a primary outcome of the synthesis, available on the 272 arms, and it does not separate the active strategies from sham stimulation. Absence of difference is not proof of equivalence, above all with wide intervals on thin branches, and this outcome does not capture adverse events that do not lead to discontinuation. | |
| Remission | Partial data |
| PEB readingReported in 58 trials out of 129 and for 12 strategies out of 14. The selection of trials that report remission is not random, which exposes this secondary outcome to an availability bias. | |
| External validity | Reservation |
| PEB readingAdults only, trial populations with their restrictive inclusion criteria, unipolar and bipolar depression pooled without a separate hierarchy by diagnosis. The stimulation parameters studied fall short of those used today, which complicates direct transposition to the patient in the consulting room. | |
| Outcome measure | Reservation |
| PEB readingResponse structures the network. It is a threshold on a scale, not a clinical state. Functioning, quality of life and cognition are assessed in five trials only, and in a heterogeneous way. | |
| Durability of the benefit | No data |
| PEB readingAcute phase only, with 91% of trials lasting six weeks or less. Nothing here informs on maintenance of the response or on the arrangements for continuation, a central question for the patient who relapses. | |
| Transparency and independence | Sound |
| PEB readingPublic funding, no declared conflict, data available on request from the corresponding author. Few device comparison studies present this profile. | |
From the consulting room to the stimulation service
The first question a reader will ask is the question of access. It calls for an honest answer: the protocols cited here are not equally available from one country and one centre to the next, and their funding status is not established by this article. Before referring a patient on the strength of this ranking, check with the receiving centre which modalities it actually delivers and on what financial terms. Nothing about availability or cover follows from this publication.
The second question is how to word the referral letter. The referring psychiatrist can state what the evidence shows and ask a question, rather than prescribe a protocol they will not be delivering themselves. A request for an opinion that mentions the existence of data favouring the bilateral protocols, and asks whether they are available locally, respects the competence of the team while enriching the discussion.
The third question bears on dose rather than on technique. Since the trials of the network mostly rest on ten sessions over two weeks at 110% of the motor threshold, whereas contemporary protocols run closer to twenty to thirty sessions over four to six weeks, the useful question to the centre is not only which protocol, but how many sessions and at what intensity. One reminder that falls outside the scope of this synthesis: the contraindications specific to magnetic stimulation, in particular intracranial ferromagnetic devices and a history of epilepsy, belong to the assessment made by the centre and are not addressed by this publication.
Level of evidence
PEB appraisal. Confidence is high that most non-invasive neuromodulation protocols do better than sham stimulation on response in the acute phase, and that high-frequency rTMS on the left is the best supported modality of the field, present in 63 of the 129 trials with the narrowest interval of the network. It is moderate on the idea that the bilateral protocols, rTMS and theta burst stimulation, do better than the unilateral ones: the convergence between the overall analysis, the resistant subgroup and remission is encouraging, it remains indirect, and the authors themselves describe their finding as “preliminary evidence that bilateral stimulation over DLPFC is more beneficial than unilateral stimulation”. Confidence is low on the top rank of the ranking, carried by two trials and 42 patients, and on any reading of that rank that goes beyond a research hypothesis. It is low as well on the position of low-frequency rTMS on the left, whose interval is too wide to conclude anything. On tolerability, the synthesis finds no difference between active strategies and sham stimulation on all-cause discontinuation: that is an absence of detected difference, not proof of equivalence, and it says nothing about adverse events that do not lead to discontinuation. Confidence is finally nil on maintenance of the benefit beyond the acute phase, for want of any analysis: the question was not asked, which is not the same thing as a negative answer.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ One hundred and twenty-nine trials, that is the biggest network I have seen on this question, and it says three useful things. First, almost everything works better than sham and the bilateral protocols come out in front, including in resistant patients. That I can put in my letter. Second, tolerability separates nothing, so the argument for choosing will not come from there. Third, the top of the ranking rests on two studies and forty-two patients, so I do not mention it to my patients. And what I mainly take away is that the trials in the network deliver ten sessions where our centres deliver twenty-five. ”
What this means in practice: this work improves the quality of the question put to the neuromodulation service. It does not supply a protocol to prescribe from the consulting room.
What you can do with this on Monday morning.
- In a referral letter for treatment-resistant depression, mention the existence of data favouring the bilateral protocols, bilateral rTMS and bilateral theta burst stimulation, and ask which of them the receiving centre delivers.
- Treat high-frequency rTMS on the left as the reference option when the choice is constrained: it is the best documented modality of the network, and a middle rank on sixty-three trials is worth more than a top rank on two.
- Ask the question of dose as much as the question of technique: the number of sessions planned and the intensity relative to the motor threshold, since the trials in the network fall short of current practice.
- Do not conclude from an odds ratio below 1 on a tiny branch that a protocol is to be avoided: here the interval for low-frequency stimulation on the left runs from 0.04 to 3.46 and settles nothing.
- Do not raise transcranial focused ultrasound stimulation with a patient as an available option, except within a research protocol.
- Check with the centre before referral which modalities are actually available and on what terms, since neither follows from this publication.
Frequently asked questions
Should low-frequency rTMS over the left prefrontal cortex no longer be used?
This work does not allow that conclusion. It reports, on four trials and by indirect comparison, an odds ratio of 0.35 with a confidence interval running from 0.04 to 3.46 and a p value of 0.356. An estimate that imprecise is compatible with an unfavourable effect and with a moderately favourable one. Worth noting: low-frequency stimulation applied on the right, its usual target, does significantly better than sham stimulation in this same network.
Why not refer patients to the technique ranked first?
Because that rank rests on two trials and 42 patients in total, with an interval running from 1.35 to 38.47. A high point estimate built on very little data produces exactly this kind of rank, and the history of medical devices contains many top ranks that did not survive replication. The authors themselves recommend caution on this point, and write that focused ultrasound is a novel intervention “warranting further investigation”.
Does this ranking apply to bipolar depression?
The depressive episodes included come from major depressive disorder and from bipolar disorder alike, pooled under the term depressive episode, and the publication gives no separate hierarchy by diagnosis. The respective proportion of the two populations is not reported in the text of the article. The question of manic switch is not addressed by this synthesis.
Is a network meta-analysis worth a head-to-head trial?
No. It produces comparisons that nobody has actually run, under an assumption of similarity between trials. It is a valuable instrument for ranking hypotheses. Superiority between two active protocols still has to be demonstrated by a trial that sets them against each other.
What does this work say about tolerability and adverse events?
Tolerability is one of the two primary outcomes, measured by the all-cause discontinuation rate across the 272 arms of the network, and no active strategy departs from sham stimulation. That information is useful but partial: a patient can put up with headache or local discomfort without stopping treatment, and this outcome does not count them. Remission, for its part, is available in only 58 trials out of 129.
Does this work say anything about cognition?
Little, and the authors present it as an exploratory observation. Five trials only in the network assessed cognitive domains, mainly executive function and working memory, with heterogeneous measures. No ranking of the techniques on that dimension is possible from these data.
Annotated bibliography
Wang P, Gao Y, Li H, Tian J, Chai S, Zhou Z, Huang X, Bao W, Hu X, Zhang L, Xing H, Li B, Gong Q, Huang X. Comparison of multiple non-invasive neuromodulation strategies for depressive episodes in major depressive disorder and bipolar disorder: A systematic review and network meta-analysis of randomized controlled trials. Psychiatry and Clinical Neurosciences 2026; 80 (2): 106-120, published online 11 November 2025. DOI 10.1111/pcn.13918 · PMID 41217021. The single source of this analysis. Systematic review and random-effects network meta-analysis, frequentist model, of 129 randomised controlled trials, 272 arms and 7,667 patients, fourteen non-invasive neuromodulation strategies compared with one another and with sham stimulation, with SUCRA ranking and analyses stratified on resistance and on the setting of administration. Chinese public funding, no competing interest declared.
Mutz J, Vipulananthan V, Carter B, Hurlemann R, Fu CHY, Young AH. Comparative efficacy and acceptability of non-surgical brain stimulation for the acute treatment of major depressive episodes in adults: systematic review and network meta-analysis. BMJ 2019; 364: l1079. A reference taken from the bibliography of the publication analysed, where it serves as a point of comparison: an earlier network synthesis, centred on electroconvulsive therapy and not covering focused ultrasound. The citation is reproduced as printed by the authors; the article itself was not consulted for this analysis.