Transcranial Magnetic Stimulation Retreatment for Recurrent Depression: A Large Real-World Multisite Cohort Study

Abstract

Introduction:

When depressive symptoms return after an initial response to transcranial magnetic stimulation (TMS), evidence to guide use of a second acute course is strikingly limited. We aimed to estimate how often and how quickly patients with recurrent major depressive disorder benefit from a second course of transcranial magnetic stimulation after they previously improved with TMS.

Methods:

A retrospective cohort study across 20 outpatient centres included 259 patients who received ≥2 TMS courses between January 1, 2023 and October 1, 2025 and had clinically significant depressive symptoms at the start of course 2; 177 prior responders formed the primary cohort. Treatment used standard 10-Hz left dorsolateral prefrontal cortex TMS. Response and remission were assessed using the PHQ-9 and/or BDI-II, defined as a ≥50% reduction and PHQ-9 ≤4 and/or BDI-II ≤9; symptom change from baseline to end of course 2 and time to first response by session were compared between courses.

Results:

Among past responders, 146 of 177 patients (82.5%; 95% CI, 76.1%-87.8%) responded again during course 2 and 96 of 177 (54.2%; 95% CI, 46.6%-61.7%) achieved remission. PHQ-9 scores decreased by a median of 65.0% (P<.001), and BDI-II scores decreased by a mean of 12.0 points (95% CI, −14.0 to −10.0; P<.001). Among responders, the first session meeting response criteria occurred earlier during retreatment than during the initial course on both PHQ-9 (median session, 12 vs 16; hazard ratio=1.40, 95% CI, 1.14-1.72; P=.001) and BDI-II (13.5 vs 22.5; hazard ratio=1.46, 95% CI, 1.07-2.00;P=.018).

Conclusions:

Most patients who previously benefited from acute TMS improved again when symptoms recurred, and among responders, the first session meeting response criteria often occurred earlier during retreatment than during the initial course.

Introduction

Although repeat acute courses of transcranial magnetic stimulation (TMS) are increasingly routine in depression care, second-course response has not been characterised at scale in the kind of real-world clinical datasets that clinicians and patients most often rely on when making decisions^1–4. As a result, the prognosis for retreatment is commonly inferred from small, heterogeneous cohorts, leaving uncertainty about what patients can reasonably expect when symptoms recur after an initial favourable course^5–8. This evidence gap matters because returning for a second course is not uncommon^{1, 9–11}. It is a frequent part of longitudinal depression management, with implications for shared decision-making, timing of follow-up, continuity planning, and the clinical logic of initiating retreatment promptly when depressive symptoms begin to re-emerge^{1, 9–13}. The converse may also be clinically important: some patients who relapse after an initial benefit may delay or forgo retreatment, potentially prolonging symptomatic burden unnecessarily.

What limited evidence exists suggests that prior benefit may “carry forward” in a clinically meaningful way^5–8, yet two pragmatic questions remain insufficiently answered. First, among patients who previously responded, how often is response recaptured, and how often is full remission achieved, when a second acute course is delivered under routine conditions? Second, does improvement tend to occur earlier during retreatment, a possibility that would reshape patient counselling and mid-course decisions about whether a course is on track? Answering these questions requires more than end-point outcomes alone^{14, 15}. It also requires measurement approaches that reflect everyday practice, where symptom monitoring may involve more than one instrument and where different scales can classify the same clinical trajectory differently^{14, 16–19}. A large routine-care cohort with course-level symptom data across depression instruments therefore provides an opportunity to estimate second-course outcomes with substantially greater precision while also examining how quickly response emerges across treatment sessions¹⁴.

In this retrospective cohort study of TMS delivered during routine clinical care for major depression, we evaluate outcomes after a second treatment course among patients with clinically significant baseline symptoms and a favourable outcome in a prior course. Using commonly implemented depression measures, the Patient Health Questionnaire-9 (PHQ-9) and the Beck Depression Inventory-II (BDI-II), and both scale-specific and clinically pragmatic either-scale endpoints, we estimate end-of-course response and remission and quantify baseline-to-end-of-course symptom change during retreatment^16–19. To assess whether retreatment response emerges earlier, we compare time to first response by treatment session between the first and second courses. Finally, we examine whether baseline symptom severity and the interval between courses are associated with second-course response and remission, thereby providing concrete, patient-relevant estimates of the likelihood and timing of benefit during a second acute course of TMS.

Aims of the Study.

1. To clarify the clinical value of a second acute course of transcranial magnetic stimulation for patients whose depressive symptoms returned after prior benefit.

2. To estimate how often meaningful improvement and remission are recaptured in routine care, and how much symptoms change during retreatment.

3. To examine whether improvement appears earlier during retreatment than during the initial course in eventual responders, and whether starting symptom burden or time between courses helps explain retreatment outcomes.

Methods

Study Design and Objectives

We conducted a retrospective cohort study of patients who received TMS for major depressive disorder during routine clinical care between January 1, 2023, and October 1, 2025, across all 20 outpatient TMS centres within Discovery Behavioral Health, spanning five US states (Washington, California, Texas, Kansas, and Maryland), where measurement-based care protocols included routine daily collection of depression scale scores. The Mass General Brigham Institutional Review Board approved this study (Protocol #2023P001692) and granted a waiver of informed consent due to the retrospective nature of the study and minimal risk to participants.

The primary objectives were to assess how often response and remission were recaptured during a second acute TMS course in patients who returned with clinically significant depressive symptoms after prior benefit, and whether response emerged earlier during retreatment than during the initial course. Secondary objectives were to quantify symptom change during the second course and to examine whether baseline symptom severity and the interval between courses were associated with second-course response and remission.

Treatment Protocol and Treatment Courses Definition

All TMS treatments used the standard FDA-approved 37-minute 10 Hz protocol targeting the left dorsolateral prefrontal cortex; full parameters are provided in the Supplement. Treatments were typically administered five days per week, and courses generally extended to approximately 36 sessions.

The analytic unit was a TMS treatment course. For patients who received more than one course during the study window, the earlier course was designated course 1 and the subsequent course was designated course 2. The retreatment analyses focused on course 2. Course 1 established prior benefit and served as the within-patient comparator for response-timing analyses.

Depression Measures and Clinical Outcome Definitions

Depressive symptoms were assessed using PHQ-9²⁰ and BDI-II²¹. Clinically significant baseline depression was defined as PHQ-9 ≥10²⁰ and/or BDI-II ≥14²¹. Within each course, baseline for each instrument was defined as its earliest scored day, typically the first treatment day or within 1–2 days of course initiation in this daily-assessment setting. Response on a given scale was defined as a ≥50% reduction from that course’s baseline score. Remission was defined using absolute end-of-course thresholds of PHQ-9 ≤4²² and BDI-II ≤9²³ Response and remission were evaluated independently. Because response is a proportional change metric whereas remission is an absolute threshold metric, remission did not necessarily imply response, particularly on BDI-II when baseline severity was near the inclusion threshold.

The prespecified primary outcome definitions were either-scale endpoints, selected to reflect routine clinical practice in which symptom monitoring may involve more than one depression instrument^24–26. In this setting, improvement identified on either instrument is typically considered clinically meaningful when informing treatment decisions. Accordingly, response was defined as meeting response criteria on either PHQ-9 or BDI-II, and remission as meeting remission criteria on either scale. Scale-specific outcomes were also estimated, but only among patients who met the corresponding baseline severity threshold on that scale.

Construction of the Analytic Cohorts

Patients entered the course 2 depression cohort if they received a second course for depression during the study period and met the baseline depression threshold on at least one instrument at the start of course 2 (PHQ-9 ≥10 and/or BDI-II ≥14). Within this course 2 depression cohort, the prespecified primary analytic cohort comprised past responders. A patient was classified as a past responder if response had been achieved during course 1 on PHQ-9 and/or BDI-II. Secondary descriptive analyses focused on past remitters, defined as patients within the course 2 depression cohort who had achieved remission during course 1 on PHQ-9 and/or BDI-II.

Eligibility for each analysis was defined to match the outcome being estimated. For either-scale course 2 outcomes, patients were eligible if they met the course 2 baseline depression threshold on at least one instrument. For scale-specific outcomes, regression, and time-to-response analyses, eligibility was defined separately by instrument and course among patients who met that scale’s baseline threshold and had evaluable baseline and follow-up data.

End-of-Course Ascertainment

Because the aim was to estimate retreatment outcomes as they occur in routine practice rather than only among patients who completed a fixed number of sessions, end-of-course status was not tied to protocol completion. Instead, for each instrument within each course, end-of-course status was defined a priori using the last available assessment in that course, typically recorded on the final treatment day or within 1–2 days of it because symptom scales were generally collected daily during treatment visits. The same baseline and end-of-course rules were used to determine course 1 responder and remitter status. The distribution of second-course treatment length is shown in Supplementary Figure S1. This approach was intended to align outcome ascertainment with the routine-care estimand; requiring completion of a fixed number of sessions would have excluded patients whose courses ended earlier for clinical or administrative reasons.

Statistical Analysis

The primary end-of-course analyses estimated the proportion of past responders who achieved response and remission during course 2, reported with exact (Clopper-Pearson) 95% confidence intervals²⁷. Symptom change during course 2 among past responders was evaluated using scale-specific change scores, defined as end-of-course score minus baseline score. Because PHQ-9 and BDI-II change scores differed in distribution, inferential tests were chosen accordingly (paired t-test for BDI-II and Wilcoxon signed-rank test for PHQ-9)²⁸. Distributional diagnostics, including histograms and Q-Q plots, are provided in the Supplement.

To examine predictors of second-course outcomes among past responders, we fitted logistic regression models separately for PHQ-9 and BDI-II, with course 2 end-of-course response and remission as the dependent variables. The prespecified predictors were baseline symptom severity, defined as the scale-specific baseline score, and time between courses in years, measured from the end of course 1 to the start of course 2. Because dates of depressive relapse or recurrence were not systematically captured, this interval should not be interpreted as time to relapse or recurrence, nor as a measure of durability of benefit following course 1. Results are reported as odds ratios with 95% confidence intervals and p-values. Additional candidate covariates, including age, gender, site, comorbid diagnoses, concomitant psychotropic medication classes, and course length, were considered but not included because the limited number of outcome events was insufficient to support the resulting parameter burden. Site and medication categories were sparse, medication data were incomplete, and course length was measured after treatment began and could itself be influenced by clinical response. Including these variables would therefore increase the risk of sparse-data bias, separation, overfitting, and unstable estimates. Medication-class changes between courses were summarised descriptively in the Supplement.

To evaluate whether response emerged earlier during retreatment, we compared session number at first response between course 1 and course 2 separately for PHQ-9 and BDI-II. For each course and scale, event time was defined as the first treatment session at which response criteria were met relative to that course’s baseline. These analyses were designed to estimate response kinetics conditional on eventual response and were therefore restricted to patients who achieved response in the course being analysed. The primary comparisons of response timing assessed the distribution of response-session times between courses using log-rank tests²⁹. Results were displayed using Kaplan-Meier cumulative response-by-session curves³⁰.

To estimate the relative rate of reaching response while accounting for within-patient correlation, we additionally fitted Cox proportional hazards models with course indicator (course 2 versus course 1) as the predictor and robust standard errors clustered by patient^{31, 32}. Proportional hazards assumptions were assessed using Schoenfeld residual tests^{33, 34}. As a within-person sensitivity analysis, we also compared session number at first response between courses among patients who responded in both courses using the Wilcoxon signed-rank test²⁸. These additional analyses are reported in the Supplement.

Finally, to clarify the practical implications of using more than one depression instrument, we conducted descriptive sensitivity analyses in past responders that compared either-scale outcomes with single-scale outcomes and quantified concordance and non-overlap between PHQ-9 and BDI-II classifications for end-of-course response and remission. We also summarised directional consistency by calculating the percent reduction on the alternate scale among single-scale responders when the alternate scale was evaluable, defined as having available baseline and end-of-course scores. These analyses were intended to show how often apparent cross-scale disagreement still reflected clinically meaningful improvement on the alternate instrument. We also performed exploratory descriptive comparisons of course 2 responders and nonresponders among patients with course 1 response, and separately among those with course 1 remission. Full details are provided in the Supplement.

All analyses were conducted in Python (3.10.16) using numpy (2.2.6), pandas (2.3.3), scipy (1.15.3), and statsmodels (0.14.4).

Results

Analysis Cohort and Baseline Characteristics

The course 2 depression cohort included 259 patients whose second course was delivered for depression and who met the prespecified baseline severity threshold at the start of course 2 (PHQ-9 ≥10 and/or BDI-II ≥14) (Figure 1). Of these, 177 met response criteria during course 1 and formed the prespecified primary analytic cohort, and 108 met remission criteria during course 1 and formed the secondary descriptive subgroup. The remaining 82 patients did not meet course 1 response criteria; their course 2 outcomes are presented in Supplementary Table S8. Past responders and past remitters were similar at course 2 baseline; detailed characteristics are presented in Table 1.

Figure 1.

Cohort derivation and analytic groups. Flow diagram of the routine-care depression-indicated TMS across 20 centres from January 1, 2023, to October 1, 2025. During the study period, 259 patients received a second course; 177 met second-course depression inclusion criteria and had responded in course 1, forming the prespecified primary cohort, and 108 were past remitters, forming the secondary descriptive subgroup. Course 1 denotes the earlier treatment course and course 2 the retreatment course. Response was defined as a ≥50% reduction from course baseline on PHQ-9 and/or BDI-II, and remission as PHQ-9 ≤4 and/or BDI-II ≤9.

Table 1.

Baseline characteristics of past responders and past remitters

Characteristic	Past responders (n=177)	Past remitters (n=108)
Age, years, median (IQR)	39.1 (31.4–50.0)	38.5 (31.5–50.0)
Gender, n (%)
— Female	127 (71.8%)	75 (69.4%)
— Male	50 (28.2%)	33 (30.6%)
Baseline depression severity and measurement (course 2)
— PHQ-9 baseline score, mean (SD)	18.0 (5.5)	17.6 (5.8)
— BDI-II baseline score, mean (SD)	23.2 (12.1), n=175	20.2 (13.0), n=105
— Meets PHQ-9 baseline threshold (≥10), n (%)	160 (90.4%)	94 (87.0%)
— Meets BDI-II baseline threshold (≥14), n (%)	136 (77.7%), n=175	69 (65.7%), n=105
Comorbid diagnoses (as available)
— Anxiety disorder diagnosis, n (%)	140 (84.3%), n=166	88 (86.3%), n=102
— PTSD diagnosis, n (%)	29 (17.5%), n=166	19 (18.6%), n=102
— Substance use disorder diagnosis, n (%)	16 (9.6%), n=166	9 (8.8%), n=102
Review of systems (as available)
— Anxiety symptoms, n (%)	94 (83.2%), n=113	56 (81.2%), n=69
— Trauma/stress symptoms, n (%)	34 (25.0%), n=136	21 (25.0%), n=84
— Substance use symptoms, n (%)	16 (22.5%), n=71	12 (26.1%), n=46
Past history (as available)
— Psychiatric hospitalization history, n (%)	17 (16.2%), n=105	11 (16.9%), n=65
— Substance use history, n (%)	24 (41.4%), n=58	15 (41.7%), n=36
Concomitant medications (as available)
— Antidepressant	158 (96.9%), n=163	98 (97.0%), n=101
— Antipsychotic (augmentation)	58 (35.6%), n=163	31 (30.7%), n=101
— Mood stabiliser	30 (18.4%), n=163	17 (16.8%), n=101
— Benzodiazepine	31 (19.0%), n=163	20 (19.8%), n=101
— Stimulant	26 (16.0%), n=163	12 (11.9%), n=101

Baseline characteristics were assessed at the start of course 2. Past responders and past remitters are defined by course 1 outcomes.

Percentages are based on available data; where shown, n= indicates the number with data, otherwise all patients are included.

Recaptured Response and Remission During Course 2

Retreatment outcomes were favourable among past responders (Table 2; Figure 2A). On the prespecified either-scale endpoint, 146 of 177 patients (82.5%; 95% CI, 76.1%-87.8%) recaptured response during course 2 and 96 of 177 (54.2%; 95% CI, 46.6%-61.7%) achieved remission. For descriptive context only, either-scale response and remission in first-course treatment across the same health system were 66.8% (95% CI, 64.6%-69.0%) and 39.0% (95% CI, 36.8%-41.3%), respectively (Supplementary Table S1). These benchmarks are not directly comparable because the retreatment cohort was restricted to prior responders. Scale-specific estimates on PHQ-9 and BDI-II are shown in Table 2.

Table 2.

Course 2 response and remission among past responders

>Endpoint (Course 2)	>n/N	%	Exact 95% CI
Either-scale response (PHQ-9 or BDI-II)	146/177	82.5	76.1–87.8
PHQ-9 response (among PHQ-9 baseline ≥10)	120/160	75.0	67.6–81.5
BDI-II response (among BDI-II baseline ≥14)	86/136	63.2	54.5–71.3
Either-scale remission (PHQ-9 or BDI-II)	96/177	54.2	46.6–61.7
PHQ-9 remission (among PHQ-9 baseline ≥10)	56/160	35.0	27.6–42.9
BDI-II remission (among BDI-II baseline ≥14)	64/136	47.1	38.4–55.8

Figure 2A.

Second-course response and remission outcomes among past responders. Bars show the proportion of patients who recaptured response or achieved remission during retreatment, with exact 95% confidence intervals. The prespecified either-scale endpoint is shown first, followed by scale-specific estimates for context, as reported in Table 2.

Beyond categorical response and remission, past responders also showed substantial symptom improvement during course 2 (Figure 2B). Mean BDI-II scores decreased by 12.0 points (standard deviation [SD], 13.4; 95% CI, −14.0 to −10.0; P < .001; Cohen d = −0.894), corresponding to a median 52.9% reduction from baseline (interquartile range [IQR], 22.2%-80.6%). PHQ-9 scores also improved substantially, with a median reduction of 65.0% from baseline (IQR, 43.8%-80.0%; Wilcoxon W = 103.5; P < .001; rank-biserial r = −0.986).

Figure 2B.

Second-course symptom change from baseline to end-of-course among past responders. Violin plots with individual patient observations and within-person trajectories comparing baseline versus end-of-course depressive symptom scores during the second course among past responders. For each panel, n reflects paired score availability (past responders with non-missing baseline and end-of-course scores on the instrument shown; PHQ-9 n=177; BDI-II n=175, as two participants lacked course 2 BDI-II assessments). This differs from Table 2, where scale-specific response and remission denominators are restricted to patients meeting that scale’s baseline clinical threshold at course 2 baseline (PHQ-9 ≥10; BDI-II ≥14). Panels show PHQ-9 and BDI-II scores; lines connect each patient’s baseline and end-of-course values. Distributional diagnostics supporting inferential test selection are provided in Supplementary Figures S2–S3.

Earlier Response Among Responders During Course 2

In the responder-only timing analysis, the first session meeting response criteria occurred earlier during retreatment on both scales (Figure 3; Table 3). On PHQ-9, the median session to first response was 12 (IQR, 5.5-22.0) in course 2 vs 16 (IQR, 9.5-27.0) in course 1, with different response-time distributions by log-rank testing (χ² = 7.937; P = .005) and a higher rate of reaching response in the clustered Cox model (hazard ratio [HR], 1.40; 95% CI, 1.14-1.72; P = .001). On BDI-II, the median session to first response was 13.5 (IQR, 5.0-24.2) in course 2 vs 22.5 (IQR, 13.0-30.0) in course 1, with similar evidence from log-rank testing (χ² = 5.343; P = .021) and the clustered Cox model (HR, 1.46; 95% CI, 1.07-2.00; P = .018). The paired within-patient sensitivity analyses, restricted to patients who responded in both courses, were consistent with the primary findings, showing earlier first response in course 2 on both PHQ-9 and BDI-II (Supplementary Table S4).

Figure 3.

Session to first response among responders during course 2 versus course 1. Kaplan–Meier curves show the distribution of session to first response among patients who ultimately achieved response in that course. A left-shifted course 2 curve indicates earlier first response among responders during retreatment. Median sessions to first response and statistical comparisons are reported in Table 3.

Table 3.

Session to first response: course 1 versus course 2

>Scale	Responders N (C1 vs C2)	Median Session (IQR), C1	Median Session (IQR), C2	Log-Rank χ2 (p)	HR (C2 vs C1), Robust 95% CI; p
PHQ-9	131 vs 123	16 (9.5–27.0)	12 (5.5–22.0)	7.937 (0.005)	1.40 (1.14–1.72; 0.001)
BDI-II	64 vs 70	22.5 (13.0–30.0)	13.5 (5.0–24.2)	5.343 (0.021)	1.46 (1.07–2.00; 0.018)

Secondary and Exploratory Findings

Among past remitters, retreatment outcomes were even more favourable: 93 of 108 patients (86.1%; 95% CI, 78.1%-92.0%) achieved response again during course 2 and 77 of 108 (71.3%; 95% CI, 61.8%-79.6%) achieved remission again. Scale-specific results for past remitters and outcomes by past-responder subtype are provided in the Supplementary Results (Supplementary Tables S2a–S3b).

Baseline severity was associated with higher odds of PHQ-9 response, while the corresponding BDI-II association did not reach statistical significance and time between courses was not significant in any model (Supplementary Table S6). For PHQ-9 response, the odds ratio [OR] was 1.16 (95% CI, 1.06-1.27; P = .001). For BDI-II response, the OR was 1.04 (95% CI, 1.00-1.08; P = .064). Neither baseline severity nor time between courses was associated with remission on either scale (Supplementary Table S6).

Exploratory descriptive comparisons indicated that course 2 responders had greater symptom reduction during course 1 than course 2 nonresponders (Supplementary Table S9).

Because PHQ-9 and BDI-II did not always classify the same individuals as responders or remitters, either-scale outcomes were higher than single-scale outcomes by design. Among either-scale responders, 79 of 146 patients (54.1%) met response criteria on both scales, 49 (33.6%) met PHQ-9 response criteria only, and 18 (12.3%) met BDI-II response criteria only, underscoring that instrument choice can influence categorical outcome labelling even when overall improvement is clinically meaningful. Among either-scale remitters, 60 of 96 patients (62.5%) met remission criteria on both scales, 7 (7.3%) met PHQ-9 remission criteria only, and 29 (30.2%) met BDI-II remission criteria only. In directional consistency analyses among single-scale responders, improvement on the alternate scale was common, even when the threshold for categorical response was not met. Cross-scale overlap is shown in Supplementary Figure S4, with full concordance and directional consistency in Supplementary Tables S7a–S7c.

Discussion

In this large routine-care TMS retreatment cohort, most patients who previously benefited from an acute course again achieved clinically meaningful improvement when symptoms recurred. Using prespecified, pragmatic endpoints that reflect real-world measurement practices, more than four in five past responders recaptured response during the second course and more than half achieved remission, with substantial within-person symptom reductions on both PHQ-9 and BDI-II. These retreatment outcomes also compare favourably with first-course response and remission rates observed in the same health system, although direct comparison is limited by the restriction of the retreatment cohort to prior responders. Within this selected group, however, the findings indicate a particularly high likelihood of renewed benefit when symptoms re-emerge¹¹. Clinically, this reduces the uncertainty that has often surrounded decisions about whether to restart TMS and supports retreatment as a dependable component of longitudinal depression care.

These results align with the limited existing evidence base on second-course TMS among initial responders, but smaller naturalistic studies with limited samples and heterogeneous designs have left imprecise estimates and uneven guidance for counselling patients considering retreatment^5–8. By leveraging a multi-site real-world dataset with course-level symptom tracking on two widely used instruments, the present study strengthens confidence that recaptured benefit is a robust clinical phenomenon while also addressing the key gap that motivated this work: quantifying retreatment prognosis at a scale that better reflects contemporary clinical practice and yields tighter confidence intervals for response and remission. In that way, the present data do not simply reiterate earlier observations but consolidate them, strengthening the empirical basis for retreatment.

Beyond establishing the likelihood of renewed benefit, this study also clarifies how quickly that benefit tends to emerge during retreatment³⁵. Among responders, the first session meeting response criteria occurred earlier during the second course than during the first on both PHQ-9 and BDI-II, suggesting that symptom relief during retreatment may emerge sooner for many eventual responders. In practice, this may refine how clinicians and patients interpret early-course trajectories, particularly when deciding whether a course is “on track” midstream. While the mechanisms cannot be determined here, earlier first response among responders during retreatment is consistent with a model in which prior benefit reflects a treatment-responsive neurobiological phenotype and/or where re-engagement with a previously effective intervention accelerates clinical change^{36, 37}. The observed cross-instrument non-overlap also shows why pragmatic, measurement-based endpoints matter. Approximately half of either-scale responders met response criteria on both instruments, while the remainder met criteria on PHQ-9 or BDI-II alone; among these single-scale responders, a substantial number also showed meaningful improvement on the alternate scale despite not crossing the categorical response threshold. This pattern suggests that categorical “nonresponse” on one instrument can coexist with meaningful improvement on another, supporting both the use of either-scale definitions as a pragmatic approximation of routine-care decision-making^{16, 17, 19}, and the value of multi-instrument monitoring when feasible to reduce misclassification of improvement. These findings therefore support not only retreatment itself, but also more nuanced interpretation of symptom change during clinical care.

Although this study has notable strengths, including the size of the retreatment cohort, the use of course-level symptom data from routine clinical care, two widely implemented symptom measures, and session-level analyses of response kinetics that account for within-patient clustering, several limitations warrant careful consideration. First, the retrospective observational design precludes causal attribution and is vulnerable to selection effects, including the possibility that patients who returned for a second course differ systematically from those who did not, and that clinicians’ decisions to recommend retreatment were informed by factors not captured in the dataset. In addition, the absence of a nonretreated comparator group prevents direct assessment of whether observed outcomes were attributable to retreatment itself. Second, symptom measurement was pragmatic rather than protocolised, so missingness, variable assessment timing, and differences in instrument use could influence classification, even as the either-scale endpoint was intended to reflect this real-world context. In addition, end-of-course outcomes were based on the last available assessment rather than protocol completion, and reasons for abbreviated treatment courses were not systematically captured. Third, concomitant medications and adjunctive psychotherapies may have varied across courses and over time, and these factors could contribute to both overall outcomes and apparent differences in response timing. Finally, although the cohort spanned 20 clinics across five US states, all data came from a single health system and a single TMS protocol, so unmeasured system-specific care characteristics may limit generalisability. These limitations also highlight a key positive: despite the inherent heterogeneity of routine care³⁸, the consistency of favourable retreatment outcomes across instruments and analytic approaches suggests that the signal is not fragile and provides a foundation for prospective studies to clarify the mechanisms underlying earlier retreatment response^{36, 37}.

In conclusion, in a real-world cohort of patients returning for a second course of TMS after a prior favourable outcome, recaptured response was common, remission was frequent, and among responders, the first session meeting response criteria often occurred earlier during retreatment than during the initial course. These findings address a persistent evidence gap in longitudinal depression management by providing more precise, practice-aligned estimates of what patients and clinicians can expect from retreatment when symptoms recur after initial benefit. More broadly, they support retreatment as an empirically supported component of ongoing care planning rather than an exceptional case, while reinforcing the value of measurement-based monitoring strategies that reflect how outcomes are tracked outside of trials.

Supplementary Material

Supplement

Supplementary Figure S1. Distribution of second-course treatment length. Histogram of treatment length, defined as the total number of treatment sessions delivered during the second TMS course among past responders (n=177). The second-course length varied across patients (median, 36 sessions; IQR, 26–37; mean, 30.2; range, 5–40). Vertical lines denote the mean and median, and the shaded band indicates the interquartile range.

Supplementary Figure S2. Histograms of PHQ-9 and BDI-II second-course change scores. Histograms of change scores computed as end-of-course minus baseline score for PHQ-9 and BDI-II among past responders; negative values indicate symptom improvement.

Supplementary Figure S3. Normal Q–Q plots of PHQ-9 and BDI-II second-course change scores. Theoretical normal quantiles are plotted against observed change scores. The red line represents the expected values under a normal distribution. Points lying close to the reference line indicate approximate normality of the change score distributions.

Supplementary Figure S4. PHQ-9 and BDI-II scale overlap for second-course response and remission. Stacked bars show, among either-scale responders (n=146) and either-scale remitters (n=96), the percentage classified by PHQ-9 only, by BDI-II only, or by both instruments. This figure illustrates how instrument choice can influence categorical outcome labelling in routine measurement-based care; corresponding counts are provided in Supplementary Table S7a.

Supplementary Table S1. End-of-course outcomes in the first course.

Supplementary Table S2a. End-of-course outcomes among past remitters (n=108).

Supplementary Table S2b. End-of-course outcomes among past non-remitters (n=75).

Supplementary Table S3a. End-of-course outcomes by past-responder subtype: Course 1 PHQ-9 responders (subset n=153).

Supplementary Table S3b. End-of-course outcomes by past-responder subtype: Course 1 BDI-II responders (subset n=129).

Supplementary Table S4. Within-person paired sensitivity analysis: session to first response.

Supplementary Table S5. Proportional hazards diagnostics (Schoenfeld residual tests).

Supplementary Table S6. Predictors of end-of-course outcomes among past responders.

Supplementary Table S7a. Cross-scale overlap among either-scale endpoint achievers.

Supplementary Table S7b. Directional consistency among single-scale responders.

Supplementary Table S7c. Exploratory characteristics of both-scale and single-scale responders.

Supplementary Table S8. Course 2 outcomes by course 1 response status.

Supplementary Table S9. Exploratory comparison of course 2 responders and nonresponders.

Supplementary Table S10. Major medication classes across course 1 and course 2.

Significant Outcomes.

1. In patients who had previously benefited from TMS and returned with recurrent depressive symptoms, a second acute course was highly likely to restore clinically meaningful improvement, with more than four in five recapturing response and more than half reaching remission.

2. Among patients who responded, improvement tended to appear earlier during retreatment than during the initial course, suggesting that prior TMS benefit may help identify a subgroup likely to improve more quickly when retreated.

3. Using PHQ-9 and BDI-II together captured meaningful improvement that would have been missed by relying on either scale alone, supporting pragmatic multi-instrument monitoring in routine TMS care.

Limitations.

1. The retrospective observational design limits causal inference and introduces potential selection bias, because patients who returned for retreatment may have differed systematically from those who did not.

2. Symptom assessment was pragmatic rather than fully protocolised, so missing data, variable assessment timing, and differences in instrument use could have affected response and remission classification.

3. The cohort came from one health system and was treated with a single TMS protocol, which may limit generalisability, while changes in concomitant medications or psychotherapies across courses could have influenced estimates of retreatment benefit and timing.

Data Sharing Statement:

No additional data are available.