Accelerated transcranial magnetic stimulation for depression: optimizing dose, spacing, targeting, and durability rather than speed alone

Repetitive transcranial magnetic stimulation (rTMS) offers an established treatment for major depressive disorder (MDD), but repeated attendance creates substantial burden (1,2). Accelerated transcranial magnetic stimulation (aTMS) encompasses different interventions. We propose evaluating them as a constrained optimization problem: choose the least burdensome protocol that meets prespecified clinical, durability and safety requirements. Our contribution is an operational decision framework, not a claim that multidimensional treatment optimization is itself new.This Opinion focuses on adults with nonpsychotic unipolar MDD, often treatment-resistant depression (TRD), which is commonly defined as failure of at least two adequate antidepressant trials, although definitions and degrees of resistance vary. Bipolar depression, psychotic depression, severe melancholic depression, older adults with substantial comorbidity and people at acute suicidal risk require separate evidence.For this revision, we supplemented the original bibliography with targeted searches on aTMS, intermittent theta-burst stimulation (iTBS), Stanford Neuromodulation Therapy (SNT), intersession intervals, safety, suicidality and durability, checking PubMed records, publisher pages and accessible full texts up to 14 September 2026. We prioritized randomized comparisons, including null findings, and used systematic reviews to describe heterogeneity and observational studies to inform feasibility and follow-up. Preprints, conference abstracts and unpublished data were excluded. This selective narrative approach is neither systematic nor exhaustive. Table 1 provides qualitative, outcomespecific certainty judgments based on design, precision, masking, consistency and applicability; these are not formal GRADE assessments.The term "accelerated" covers three different changes: calendar acceleration compresses treatment into fewer days, session shortening reduces session duration, and dose intensification increases daily sessions or pulses. None of these by itself guarantees faster clinical improvement, and shorter sessions do not necessarily reduce overall resource use, because accelerated schedules may require longer clinic stays, higher pulse doses or neuronavigation.We therefore describe each protocol across four domains. Stimulation parameters include pattern, intensity, pulse number and intersession spacing. Circuit delivery includes how the target is selected, whether it can be located reproducibly and whether stimulation engages the intended circuit; these are distinct properties, and success on one does not establish the others. Clinical scheduling covers treatment days, sessions and continuation; implementation burden covers patient, staff, device and imaging resources. These distinctions matter for interpreting the evidence. Early accelerated studies and a brief research report suggested feasibility without showing that protocols are interchangeable (3)(4)(5), and high-frequency rTMS (HF-rTMS) and theta-burst protocols need separate appraisal.Reviews comparing accelerated with conventional treatment report heterogeneous findings, and a statistically nonsignificant difference does not demonstrate non-inferiority (6,7). By contrast, THREE-D specifically demonstrated non-inferiority of once-daily iTBS to its 10-Hz comparator (8). Its approximately 3-minute sessions, however, represent session shortening, not calendar acceleration, and its 37.5-minute comparator should not stand for all conventional protocols.For accelerated iTBS (aiTBS), a meta-analysis of five double-blind randomized trials (239 participants) favored active stimulation for study-defined response, but it combined unipolar and bipolar depression and rated certainty as low to moderate, limiting diagnosis-specific conclusions (9). The Stanford sequence also needs to be read step by step: an early letter on high-dose stimulation (10), the open-label Stanford Accelerated Intelligent Neuromodulation Therapy (SAINT) study (11), and then randomized trials of SNT. The first sham-controlled trial analyzed 29 participants and favored active treatment on the four-week change in Montgomery-Åsberg Depression Rating Scale (MADRS) score (12). Open-label remission rates should not be treated as randomized efficacy estimates. Two recent trials extend this picture. Ramos et al. randomized 100 outpatients to scalptargeted aiTBS or sham over 15 weekdays; the primary week-5 outcome on the 17-item Hamilton Depression Rating Scale (HDRS-17) favored active treatment (d=0.65, 95% CI 0.29-1.00) (13). This supports that specific regimen, not equivalence to SNT or a guaranteed response within days. The SNT replication randomized 48 participants, and its one-month remission advantage must be interpreted with caution because of imprecision, interim stopping, a short (one-month) primary follow-up and additional treatment before the endpoint (14; Table 1). The H1-coil trial randomized 104 participants and met its week-6 non-inferiority criterion (15). Because stimulation pattern, intensity, session count and continuation also differed, however, it cannot isolate the effect of calendar compression. Its companion patient-reported-outcome publication describes the same trial, not an independent replication (16).A meta-regression and a dose-response meta-analysis have linked pulse dose and spacing with outcomes (17,18). The larger synthesis included 59 studies and 2,908 participants with different designs and stratified its analyses by comparator. Its association with intervals of approximately 30-75 minutes does not establish a therapeutic window, because associations across studies can be confounded by ecological bias, correlated pulse and session counts, protocol family, treatment resistance, baseline severity, intensity, targeting, sham methods, selective reporting and small-study effects (18). Direct randomized evidence does not support a single optimal interval. Goodman et al. compared 0-, 30-and 60-minute intervals in 182 participants receiving twice-daily bilateral dorsomedial prefrontal iTBS with total pulses held constant, and found no significant differences in improvement, response or remission (19). This does not establish equivalence or settle spacing for high-dose left dorsolateral prefrontal cortex (DLPFC) protocols, but it challenges any universal optimal interval. Adequately powered parameter trials should hold other components constant and prespecify superiority or non-inferiority hypotheses.Targeting raises similar questions. Taylor et al. randomized 40 participants to connectivity-or scalpbased targeting (20). On the primary outcome, baseline-adjusted MADRS at one month, median reductions were 24 versus 18 points (standardized effect 0.8, 95% CI 0.26-1.54), and binary outcomes pointed in the same direction. Targets were derived prospectively from a 41-minute multiecho scan and a convergent depression circuit, which differs from the SNT subgenualanticorrelation procedure, and split-half reproducibility does not establish between-day test-retest reliability or portability across scanners. This small estimation trial therefore supports further testing of this pipeline, not of "precision targeting" in general. Confirmatory studies should specify acquisition, preprocessing, automated selection and human quality control, report processing failures and test whether the added benefit justifies imaging and navigation costs.Safety cannot be inferred from pulse totals alone (21), and trials recorded adverse events differently.In the accelerated HF-rTMS comparison, site discomfort affected 11/58 versus 2/57 participants, and three accelerated-arm withdrawals were due to discomfort or migraine (4). Ramos et al. reported week-1 scalp pain in 17.4% versus 4.4%, no hypomanic or manic switches and 89/100 completers (13). The SNT replication reported no serious adverse events and one sham-arm withdrawal, but did not record expected stimulation-site discomfort as an adverse event (14). In the H1-coil trial, treatment-site discomfort and headache were more common with the accelerated protocol (15). These differently ascertained results should not be pooled casually. Small trials cannot exclude rare seizures or mood switches. Intensive courses therefore require explicit assessment, with standardized denominators, of headache, scalp pain, fatigue, anxiety, syncope, cognitive changes and discontinuation, together with screening and daily reassessment for sleep deprivation, alcohol use or withdrawal, major caffeine changes, seizure-threshold-altering medications, neurological disease and medical instability. Evidence from selected unipolar samples does not establish safety in bipolar depression or complex neurological or medical comorbidity (21). Moreover, safety observed in specialist centers with strict eligibility criteria and close monitoring may not transfer to routine practice, where screening, staffing and adverse-event surveillance differ. Patient selection limits generalizability in a similar way. Trials should report age, inpatient or outpatient setting, degree of treatment resistance, exclusion of psychosis, comorbidities and concomitant pharmacotherapy. Ramos et al., for example, enrolled medicated outpatients aged 18-65 years with low suicide risk, so their results cannot be extended to emergency settings (13). Masking deserves equal attention. Sham must reproduce the sound and somatic sensations of active stimulation, because scalp discomfort, twitching and early improvement can reveal allocation. Trials should measure pretreatment expectancy and record participant, operator and rater guesses with confidence ratings, ideally before substantial improvement occurs; staff contact should be balanced between arms. In Ramos et al., sham included electrical scalp stimulation, and participant and rater guesses were close to chance, but masking remains protocol-specific (13).An antidepressant effect does not demonstrate a specific antisuicidal effect. A suicide item, a dedicated suicidal-ideation scale, suicidal behavior, suicide attempts and suicide mortality are different endpoints. In a sham-controlled aiTBS crossover study of 50 participants, analyses of the 32 with suicidal ideation at baseline showed no superiority over sham (22). aTMS should not delay established crisis assessment and treatment, and suicide-prevention claims require adequately designed studies. Durability needs the same care, because immediate improvement, one-month remission and relapse-free survival are distinct outcomes. In the SNT replication, three active-arm participants and one sham-arm participant received additional assigned treatment before one month, so the endpoint does not reflect uniformly rescue-free persistence after five treatment days (14). In an uncontrolled SNT cohort of 46 participants, 32 remitted initially and 15 remained in remission at 12 weeks: 33% of all treated participants and 47% of initial remitters (23). Different scales, selection and lack of randomization limit comparison with conventional rTMS, and beyond three to six months-controlled evidence is insufficient to define an optimal maintenance schedule. Future studies should report sustained response and remission at 1, 3 and 6 months, relapse-free survival among initial remitters, time to rescue, cumulative retreatment, medication changes and exposure to continuation or maintenance treatment. Scheduled maintenance and unplanned rescue should be reported separately, and treatment must never be withheld merely to preserve a rescue-free endpoint.For non-emergency outpatient research, we propose testing three benchmarks: response within 14 calendar days from the first stimulation, remission by weeks 4-6, and maintenance of benefit for three months after the acute course, confirmed at six months. These are author-proposed research targets, not validated thresholds, guarantees or universal clinical deadlines. Response should require at least 50% improvement, and remission a prespecified scale threshold such as MADRS ≤10 or HDRS-17 ≤7, alongside measures of functional recovery and patient-valued benefit. Benefit is sustained only without intervening relapse or unplanned rescue; scheduled continuation is acceptable if prespecified and reported. Relapse criteria and the handling of missing follow-up should also be prespecified. Trials should report the proportion of participants meeting each target and time-to-event distributions, counting nonresponders and discontinuations rather than analyzing responders alone, with comparable assessment schedules across arms. Before recruitment, investigators, patients and services should agree on acceptable response and remission rates, adverse-event limits, maximum daily and total patient time, staff and device time, and total cost. These limits depend on setting and stakeholder perspective; current evidence does not justify a universal monetary threshold.Our decision rule is to choose the least burdensome protocol among those that satisfy these prespecified requirements. An expedited pathway that reduces travel but exceeds tolerability or durability limits does not meet the framework, whereas a slightly longer acute course may be preferable if it achieves comparable benefit with fewer rescue visits. Comparisons between protocols require uncertainty estimates, because the absence of a significant difference cannot establish preserved efficacy.Accelerated protocols distribute costs and benefits differently among patients, providers and payers, and economic assessments of once-daily iTBS cannot automatically be applied to high-dose aTMS (24). For patients, fewer travel days may reduce disruption to work and caregiving, while long treatment days can increase lodging needs and fatigue (25,26). Providers must account for clinic occupancy, imaging, processing, navigation, staff coordination, device capacity and extended opening hours, and payers for the entire acute and maintenance pathway. A protocol may thus be efficient for patients but resource-intensive for providers, or vice versa, so efficiency claims should state the perspective adopted. Waiting time also matters: five treatment days after a long wait for access may offer little practical acceleration. Comparative evaluations should report costs per sustained responder or remitter and, where appropriate, quality-adjusted life-years, separately from patients' out-of-pocket costs. Targeting reproducibility, imaging access and throughput also need to be demonstrated outside specialist centers. In our view, the next advance is not a faster protocol but a testable, protocol-specific balance between clinical benefit and implementation burden. aTMS should be adopted according to the evidence for its actual parameters and population, while uncertainty about component necessity, long-term prevention and scalability remains explicit.