Published on 19 September 2026
High-frequency rTMS for negative symptoms in schizophrenia: a small effect, and two sets of numbers
The essentials
This systematic review pools 17 randomized trials against sham stimulation on high-frequency repetitive transcranial magnetic stimulation, added to antipsychotic treatment. Fifteen of them, that is 821 patients recounted from the data supplied in the appendix, provide a PANSS negative subscore: the standardized mean difference there is -0.31 (95% CI -0.49 to -0.14), with 32% heterogeneity. The publication’s abstract instead announces -0.22 with 9% heterogeneity. We redid the calculation from the individual trial-level data the authors attached: it reproduces exactly the figures in the Results section, not those in the abstract. Three subgroups stand out, analyzed separately and never crossed: 20 Hz (-0.48), more than fifteen sessions (-0.40), dorsolateral prefrontal cortex (-0.34). The result on the SANS scale shows 91% heterogeneity and cannot be interpreted. The attached PRISMA checklist states that no protocol was prepared, so nothing establishes that these subgroups were decided before the analysis. No conflict of interest or funding declared.
Context
Negative symptoms are what antipsychotics treat least well, and what weighs most heavily on functional outcome. A patient whose delusions have resolved but who no longer leaves the house remains a treatment failure. No drug has demonstrated a clear-cut effect on this dimension, which explains the enduring interest in neuromodulation.
rTMS raises a difficulty that pharmacology does not: the dose is not a single figure but a set of settings. Frequency, anatomical target, intensity relative to motor threshold, number of sessions, targeting method. Two trials that both call themselves rTMS trials can have little in common. This review illustrates the problem, and adds a second one: reporting.
The study at a glance
| Element | Content |
|---|---|
| Population | Adults with schizophrenia or schizoaffective disorder, on antipsychotic treatment |
| DetailDiagnosis established using DSM, ICD, or the MINI interview, without the authors specifying which version. Antipsychotic dosage is reported in chlorpromazine equivalents for some of the trials. The authors report 911 patients across 17 trials; summing the sample sizes in their Table 1 gives 1,021, or 981 counting only once the control group shared by the two arms of Xiu et al. The figure of 911 cannot be reproduced from any table supplied. | |
| Intervention | High-frequency rTMS as an adjunct, 5 to 20 Hz across the included trials |
| DetailThe inclusion criterion accepted any frequency above 1 Hz. Of the 18 arms described, 9 are at 10 Hz, 6 at 20 Hz, 2 at 15 Hz, and 1 between 5 and 7 Hz. The target is the dorsolateral prefrontal cortex for 16 of the 18 arms, the bilateral prefrontal cortex for one, the cerebellum for one. The number of sessions ranges from 10 to 40, at a rate of five per week. | |
| Comparator | Sham stimulation |
| DetailThe authors describe the whole set as double-blind trials, the procedures differing only by the orientation of the coil or the use of a sham coil. Yet their own reference list describes the Garg et al. 2016 trial as a rater-blind study. Double blinding is therefore not established for the entire set of trials. | |
| Criteria | Negative symptoms, measured with the PANSS or the SANS |
| DetailThe authors designate PANSS-negative and the SANS as the two measures of the primary outcome, not the SANS as a secondary outcome. Secondary outcomes: PANSS-positive subscore and the Calgary Depression Scale. No GRADE certainty assessment. The PRISMA checklist item devoted to reporting bias assessment is marked “Na”. | |
| Design | Systematic review with random-effects meta-analysis |
| DetailPROSPERO registration CRD42023450243, but the attached PRISMA checklist states that no protocol was prepared. Search across six databases, cut off in July 2023 according to the Methods section and March 2024 according to the abstract. 59 references screened, 17 trials retained, four subgroup analyses, one leave-one-out sensitivity analysis. Oxford level of evidence 1a by design, to be weighed against the quality of reporting. | |
Results
| Result | Value |
|---|---|
| PANSS-negative, all analyzable trials | SMD -0.31 (95% CI -0.49 to -0.14), I² = 32%, 15 trials, 821 patients |
| ReadingFigure from the Results section, which our independent recalculation on the individual data in the appendix reproduces to the third decimal. The publication’s abstract announces -0.22 with 9% heterogeneity: the same data do not reproduce it. The authors elsewhere count 791 patients where their own table gives 821. It is the Results section’s figure that should be retained, not the abstract’s. | |
| 20 Hz trial subgroup | SMD -0.48 (95% CI -0.71 to -0.25), I² = 0%, 6 trials, 301 patients |
| ReadingThe most favorable subgroup, and the only one perfectly homogeneous. It concerns frequency alone. At 10 Hz, eight trials and 480 patients, the value drops to -0.26 (95% CI -0.49 to -0.02) with 33% heterogeneity. The authors themselves note that the two intervals overlap and that the difference between frequencies is probably not significant. | |
| Subgroup of more than fifteen sessions | SMD -0.40 (95% CI -0.57 to -0.24), I² = 7%, 12 trials, 635 patients |
| ReadingTo be handled with caution: in their discussion, the authors write that lengthening treatment duration produced only slight improvements and did not strengthen the effect, which contradicts the recommendation in their own conclusion. The number of participants in this subgroup is given as 605 in the text and as 635 in Table 3; it is 635 that the recount confirms. | |
| SANS scale | SMD -0.99 (95% CI -1.64 to -0.33), I² = 91%, 10 trials, 533 patients |
| ReadingHeterogeneity of this magnitude means the pooled trials are not measuring the same thing. This figure should not be cited as a result, and we report it here only to explain why. It should be noted that this is not a secondary outcome but one of the two measures of the primary outcome, which makes the contradiction with the PANSS result all the more troubling. | |
| PANSS-positive and depressive symptoms | SMD -0.02 (p = 0.81) and -0.09 (p = 0.43), not significant |
| ReadingFourteen trials and 801 patients for PANSS-positive, six trials and 361 patients for the Calgary scale. No demonstrated difference, which does not establish an absence of effect. The included trials were not designed for these outcomes. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Selection and conduct of the review | Sound, reporting flawed |
| FindingThe work is traceable: the 59 references screened are listed with their reason for exclusion, and the extracted data are supplied trial by trial, which allows everything to be recalculated. Risk of bias is reported and it is not good: adequate sequence generation for 14 of the 18 arms, every other domain rated unclear for the entire set of arms, and 41.67% of judgements rated high risk. Reporting itself is flawed: the abstract, the Results section, and the discussion give three different sets of figures. | |
| Heterogeneity of the primary outcome | Moderate, 32% |
| FindingThe abstract announces 9%, the Results section 32%. Our recalculation on the individual data supplied reproduces 32%. The argument for homogeneity put forward in the abstract therefore does not hold: the trials agree moderately, not very well. | |
| The SANS result | Not interpretable |
| FindingPooling trials whose heterogeneity reaches 91% produces a figure, not a result. The authors report it, and go so far as to write that the SANS shows a stronger effect than the PANSS, without pointing out that this comparison has no value. It is the main weakness of their presentation. | |
| Status of the subgroups | Pre-specification not established |
| FindingThe review is registered on PROSPERO, but the attached PRISMA checklist explicitly states that no protocol was prepared. Nothing therefore allows one to affirm that these four subgroups were decided before the analysis. A subgroup remains, in any case, an observational comparison within a set of trials: it can guide the choice of a protocol, it does not demonstrate one. | |
| Independence | No conflict declared |
| FindingNo specific funding and no competing interests declared. Open-access publication under a Creative Commons Attribution licence. The authors conclude there is no publication bias based on a funnel plot judged symmetrical and a nonsignificant Egger’s test, p = 0.9605: with fifteen trials, this argument is weak. | |
Level of evidence
Confidence is reasonable that an effect exists and that it is small: fifteen randomized trials against sham stimulation, a confidence interval excluding zero, moderate heterogeneity, and raw data that reproduce the result announced in the Results section. It is low regarding the exact magnitude within subgroups, because a subgroup rests on fewer trials, is not established as pre-specified, and inherits its authors’ choices. It is nil regarding the SANS.
It must be clear about what the review does not say. Stimulation frequency and anatomical target were analyzed in two separate subgroups. No analysis combines the two. Writing that “20 Hz over the dorsolateral prefrontal cortex” gives -0.48 would be a misreading: -0.48 is the value for the 20 Hz trial subgroup, whatever the target, and the dorsolateral target taken alone gives -0.34. The authors’ conclusion, which recommends 20 Hz, more than fifteen sessions, and the dorsolateral prefrontal cortex, stacks three separate analyses as though they described a single protocol tested as such. That is not the case.
The colleague test
What an experienced colleague would say if shown this study in two minutes, between two consultations.
An effect of minus 0.31 is modest, and an abstract that does not match its own results section does not inspire confidence. If the unit already offers rTMS, the clearest signal favors 20 Hz. But this study cannot be said to have tested a complete protocol.
Translation for practice: the useful debate is not whether to use rTMS, but how. This review suggests that the settings matter. It does not say which to adopt, and it does not say so because it never compared them against each other.
What you can do with this
- Keep the order of magnitude in mind. A standardized mean difference of -0.31 corresponds to roughly 1.9 points on the PANSS negative subscore, using a within-group standard deviation of 6.0 points recalculated from these trials’ data. That is not nothing on a dimension where almost nothing works, but it is not a change of clinical state.
- Do not present “20 Hz over the dorsolateral prefrontal cortex” as the protocol validated by this work. Frequency and target were analyzed separately, never crossed, and neither analysis is established as pre-specified.
- Never cite the SANS result. Its 91% heterogeneity makes it unusable, including as an argument in favor of the technique.
- Do not rely on this review to settle the number of sessions. The long-protocol subgroup gives -0.40, but the authors themselves write in their discussion that lengthening duration did not strengthen the effect. They contradict themselves, and nothing allows the question to be settled.
- Tell the patient about a possible, modest benefit on negative symptoms, not a clear-cut improvement. And do not conclude from the absence of a signal on positive or depressive symptoms that there is none: the included trials targeted something else.
- The course of action is set out in the NICE decision tree for schizophrenia in adults.
Frequently asked questions
What does an SMD of -0.31 mean for a patient?
About 1.9 points on the PANSS negative subscore, calculated using the within-group standard deviation of 6.0 points observed in these trials. On a scale whose negative subscore ranges from 7 to 49, and whose end-of-study values here hover around 20, this is a small effect. Perceptible on average across a group, rarely at the bedside.
Why dismiss the SANS figure when it looks far more favorable?
Precisely because it is more favorable. A result with 91% heterogeneity can take almost any value depending on which trials are pooled. Citing it, even with a caveat, still puts it into circulation.
Is the 20 Hz subgroup enough to fix a protocol?
No. It is enough to guide a choice when the protocol is not already fixed, and even then with the caveat that, with no pre-registered protocol, nothing shows this subgroup was planned before the analysis. The authors themselves note that the confidence intervals of the 10 Hz and 20 Hz subgroups overlap. Only a trial directly comparing two settings would settle the question.
Should rTMS be offered first-line for negative symptoms?
Nothing in this work suggests it. This is a treatment given as an adjunct to patients already receiving an antipsychotic, and it is within that framework that the effect was measured. Access to and reimbursement of rTMS fall outside this publication and should be checked locally.
Annotated bibliography
Source study. Wang B, Zhu X, Chen L, Liu S, Wang C. The effects of high-frequency repetitive transcranial magnetic stimulation on negative symptoms in schizophrenia patients: A systemic review and meta-analysis. PLOS One. 2025;20(12):e0337847. DOI 10.1371/journal.pone.0337847. Received August 9, 2024, accepted November 11, 2025, published December 11, 2025. PROSPERO registration CRD42023450243. The authors declare they received no specific funding for this work and have no competing interests. PMID 41379812, established in the PubMed registry on September 1, 2026. The supplementary material was reviewed: the PRISMA 2020 checklist, the table of the 59 references screened with the inclusion or exclusion reason for each, and the table of trial-by-trial extracted data, from which every estimate reported here was independently recalculated.
Editorial collections
Tags
Verified on September 1, 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on September 19, 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.