Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Jul 27.
Published in final edited form as: J Affect Disord. 2026 Jun 24;412:122167. doi: 10.1016/j.jad.2026.122167

Latent class modeling of repetitive transcranial magnetic stimulation response trajectories in treatment-resistant depression

Cory R Weissman a,*, Brian Cui a, Lindsay Benster a, Noah Stapper a, Jean Phillippe Miron a, Zafiris J Daskalakis a, Lawrence G Appelbaum a, Jordan N Kohn b
PMCID: PMC13401939  NIHMSID: NIHMS2194803  PMID: 42341943

Abstract

Over 30% of patients with major depressive disorder (MDD) develop treatment-resistant depression (TRD). Repetitive transcranial magnetic stimulation (rTMS) is effective for TRD, yet there is substantial heterogeneity in treatment response. The ability to characterize distinct response trajectories could improve clinical decision-making and enhance early triaging of patients receiving acute rTMS treatment. We applied latent class mixture modeling (LCMM) to Patient Health Questionnaire-9 (PHQ-9) scores collected weekly during acute rTMS treatment in 308 patients with TRD treated at the UCSD Interventional Psychiatry Program. We also conducted a sensitivity analysis with baseline-adjusted PHQ-9 scores. LASSO-penalized multinomial logistic regression with nested cross-validation identified baseline clinical predictors of trajectory class membership. A three-class quadratic model best fit raw PHQ-9 trajectories: High Baseline, Minimal Improvement (HBMI); Moderate Baseline, Moderate Improvement (MBMI); and High Baseline, Large Improvement (HBLI). The HBLI group achieved the greatest symptom reduction (65.5%) and highest response rate (84%). Baseline-adjusted modeling yielded a 2-class solution (moderate vs. large improvement). Predictive modeling of raw PHQ-9 achieved strong multi-class discrimination (AUC = 0.82). History of suicidal ideation was associated with higher odds of HBMI class membership, while tobacco use and lack of previous suicide attempt was associated with HBLI membership. In conclusion, patients with high baseline depression severity and suicidal ideation are at greatest risk for nonresponse and may benefit from early consideration of alternative interventions.

Keywords: Treatment-resistant depression, Response trajectories, Repetitive transcranial magnetic stimulation, Latent class modeling, Clinical predictors

1. Introduction

Over 30% of individuals diagnosed with major depressive disorder (MDD) develop treatment-resistant depression (TRD). Evidence shows that TRD is a highly debilitating, costly, and deadly condition (McIntyre et al., 2023). There is a need to develop effective treatments for TRD. Repetitive transcranial magnetic stimulation (rTMS) is effective in TRD, with response and remission rates approximating 58–69% and 28–36%, respectively, in real-world clinical settings (Sackeim et al., 2020). rTMS is more effective than standard pharmacotherapy (Papakostas et al., 2024). However, there remains considerable heterogeneity in treatment outcomes and in trajectories of treatment response across rTMS patients.

To better understand this heterogeneity, researchers have begun to statistically characterize trajectories of rTMS response. One advanced modeling technique is latent class mixture modeling (LCMM), in which subgroups of patients with similar response trajectories are grouped together and captured by a simple function. This technique offers the ability to unmask individual patient trajectories hidden in group averages, and aid in triaging individuals with differential response likelihoods. Only three studies to date have been published exploring latent trajectories of rTMS response based on standardized clinical depression scales in real-world clinical settings. In these studies four latent classes with unique trajectory profiles emerged with higher severity of baseline depression predicting worse clinical outcomes (Chen et al., 2024; Janssen-Aguilar et al., 2025; McInnes et al., 2024). Further research is needed to verify these findings and explore additional predictors of response trajectory. In this study, we utilized a similar LCMM approach in a large clinical cohort to confirm and extend past findings on latent treatment trajectories with the ultimate objective of guiding clinicians in predicting rTMS outcomes. We hypothesized that a four-class model would emerge, and that higher baseline severity of depression would predict assignment to a non-response class trajectory.

2. Methods

This retrospective cohort study was conducted using patient electronic medical records (EMR; Epic, Verona, WI, USA) from the UC San Diego Interventional Psychiatry Program (UCSD-IPP). All procedures complied with ethical standards of the relevant national and international committees on human subject research and the Helsinki Declaration and were approved by the UCSD Human Research Protections Program.

Participants with TRD were prescribed once daily rTMS treatment by psychiatrists credentialed in delivering rTMS. Three hundred eight patients satisfied inclusion criteria for the analysis. To ensure that analyses included only the acute phase of rTMS treatment, we excluded observations that occurred during or after prolonged interruption of treatment or during a maintenance phase of rTMS. Patient depressive symptom severity was assessed at least once per week with the Patient Health Questionnaire-9 (PHQ-9). Please see Supplemental Material for further details of inclusion criteria and rTMS treatment information.

All statistical analyses were conducted in RStudio using R (v4.5.0). To classify patients into subgroups based on longitudinal PHQ-9 scores, LCMMs were implemented using lcmm (Proust-Lima et al., 2017). The optimal number of latent trajectories (1–5) and polynomial degrees (1–3) was determined using Bayesian information criterion (BIC), and relative improvement in model fit was determined using the BIC log Bayes factor approximation (2×ΔBIC, where ΔBIC is the difference in BIC between a model and its less complex counterpart). Best practices for LCMM implementation were applied according to previously published guidelines (Van De Schoot et al., 2017) and model selection criteria included BIC, mean posterior probability of class membership, and minimum class size. Due to known influences of baseline symptom severity on class membership, a secondary analysis was performed in which each patient’s baseline PHQ-9 value was subtracted from each observed PHQ-9 score (McInnes et al., 2024). To identify whether patient characteristics were predictive of trajectory group, LASSO-penalized multinomial logistic regression was implemented using glmnet in a nested cross-validation procedure (inner folds = 5, outer folds = 5). See Supplemental Material for further details of the statistical approach.

3. Results

3.1. Patient characteristics

Sociodemographic and clinical characteristics of all 308 patients are provided in Table 1. The proportion of our cohort with mild, moderate, moderately severe, and severe depression, according to conventional PHQ-9 total score cutoffs, was 7.8%, 20.1%, 34.7%, and 37.3%, respectively. On average, PHQ-9 scores were 17.5 ± 5.2 pre-treatment and 10.8 ± 6.6 at the end of treatment (t307 = 19.3; Cohen’s d = 1.10, 95% CI: [0.96, 1.24]), reflecting a 38.1% decrease from baseline. Nearly half (48%; N = 117) of patients achieved ≥50% reduction in PHQ-9 scores (i.e., treatment response) and 19% (N = 58) achieved full remission (PHQ-9 < 5). The average number of rTMS sessions received was 32.02 ± 4.67 over 7.49 ± 1.74 weeks, with PHQ-9 scores assessed 7.25 ± 1.41 times during treatment.

Table 1.

Sociodemographic and clinical characteristics for all patients grouped together and for each trajectory class.

Parameter All patients (N = 308) (N = 61, 19.8%; high baseline, minimal improvement) (N = 174, 56.5%; moderate baseline, moderate improvement) (N = 73, 23.7%; high baseline, large improvement) Test statistic
Mean (SD), or N (%)
Age (years) 54.2 (17.1) 50.7 (17.5) 55.5 (16.9) 54.2 (17.3) 3.20
Gender (% female) 196 (64%) 38 (62%) 117 (68%) 41 (56%) 3.00
Non-Hispanic White 230 (75%) 40 (66%) 137 (80%) 53 (73%) 5.54
College degree 247 (82%) 49 (82%) 144 (85%) 54 (75%) 3.20
BMI (kg/m2) 27.5 (6.0) 28.4 (7.4) 27.3 (5.3) 27.3 (6.4) 0.92
Relationship status (partnered/separated/single) 150 (49%)/64 (21%)/93 (30%) 28 (46%)/14 (23%)/19 (31%) 95 (55%)/32 (18%)/46 (27%) 27 (37%)/18 (25%)/28 (38%) 6.97
Currently employed 122 (40%) 26 (43%) 68 (40%) 28 (38%) 0.27
History of rTMS 36 (12%) 5 (8%) 23 (13%) 8 (11%) 1.22
History of ECT 52 (17%) 19 (31%)a 20 (12%) 13 (18%) 12.2**
History of SI 195 (64%) 49 (80%)a 96 (56%) 50 (68%) 12.6**
History of SA 78 (25%) 25 (41%)a 36 (21%) 17 (23%) 9.78**
Prior psychiatric hospitalization 115 (38%) 25 (41%) 56 (33%) 34 (47%) 4.67
Comorbid anxiety disorder 235 (77%) 51 (84%) 132 (77%) 52 (71%) 2.86
Psychological trauma 146 (48%) 32 (52%) 80 (47%) 34 (47%) 0.69
Comorbid OCD 24 (7.8%) 4 (6.6%) 11 (6.4%) 9 (12%) p = 0.30
Comorbid PD 15 (4.9%) 3 (4.9%) 6 (3.5%) 6 (8.2%) p = 0.30
Comorbid AUD/SUD 49 (16%) 6 (9.8%) 31 (18%) 12 (16%) 2.26
Comorbid ADHD 58 (19%) 8 (13%) 34 (20%) 16 (22%) 1.85
Current MDE (>1 yr) 157 (63%) 37 (77%) 85 (60%) 35 (59%) 5.19
Tobacco use (N/C/F) 237 (77%)/24 (8%)/45 (15%) 53 (87%)/3 (5%)/5 (8%) 132 (77%)/13 (8%)/27 (16%) 52 (71%)/8 (11%)/13 (18%) p = 0.30
Alcohol use (N/C/F) 123 (40%)/130 (42%)/53 (17%) 32 (52%)/21 (34%)/8 (13%) 63 (37%)/77 (45%)/32 (19%) 28 (38%)/32 (44%)/13 (18%) 4.86
Cannabis use (N/C/F) 201 (66%)/72 (24%)/33 (11%) 50 (82%)/6 (10%)/5 (8%)b 110 (64%)/40 (23%)/22 (13%) 41 (56%)/26 (36%)/6 (8%) 14.4**
Family psychiatric history 215 (70%) 41 (67%) 118 (69%) 56 (77%) 1.85
rTMS protocol (BiTMS/dTMS/iTBS/other) 48 (16%)/86 (28%)/136 (44%)/38 (12%) 9 (15%)/18 (30%)/25 (41%)/9 (15%) 26 (15%)/47 (27%)/79 (45%)/22 (13%) 13 (18%)/21 (29%)/32 (44%)/7 (10%) 1.38
Pre-treatment PHQ-9 17.5 (5.2) 22.5 (2.6)a,b 14.1 (3.9)b 21.5 (3.2) 186**
Mid-treatment PHQ-9 13.2 (6.0) 20.3 (3.7)a,b 10.4 (4.6)b 14.0 (5.4) 123**
Post-treatment PHQ-9 10.8 (6.6) 20.4 (3.8)a,b 8.8 (4.8)b 7.5 (4.6) 135**
Pre-treatment SI (item 9) 0.83 (1.0) 1.6 (1.1)a,b 0.52 (0.8)b 0.93 (1.0) 45.2**
Post-treatment SI (item 9) 0.52 (0.9) 1.4 (1.3)a,b 0.28 (0.5) 0.35 (0.6) 48.4**
Treatment responder 117 (48%) 6 (13%)a,b 59 (44%)b 52 (84%) 56.3**
AD medication use 244 (81%) 49 (80%) 140 (82%) 55 (77%) 0.62
AD medication burden 1.83 (1.75) 1.93 (2.28) 1.81 (1.58) 1.78 (1.65) 0.19
BZ medication use 82 (27%) 16 (26%) 50 (29%) 16 (23%) 1.23
BZ medication burden 0.14 (0.38) 0.22 (0.62) 0.12 (0.29) 0.11 (0.29) 0.97
AP medication use 78 (26%) 18 (30%) 43 (25%) 17 (24%) 0.61
AP medication burden 0.17 (0.40) 0.16 (0.35) 0.16 (0.35) 0.19 (0.52) 0.33
Number of rTMS sessions 32.0 (4.7) 31.6 (5.1) 32.1 (4.7) 32.3 (4.3) 0.67
Open in a new tab

Non-parametric Kruskal-Wallis rank sum test statistic shown for continuous variables, Pearson’s chi-square test or Fisher’s Exact test (p-value shown) for categorical variables. Holm-corrected Conover-Iman and pairwise post hoc chi-square tests were implemented for continuous and categorical variables, respectively, where indicated. ECT: electroconvulsive therapy. SI: suicidal ideation. SA: suicide attempt. OCD: obsessive-compulsive disorder. PD: personality disorder. AUD: alcohol use disorder. SUD: substance use disorder. ADHD: attention deficit hyperactivity disorder. MDE: major depressive episode. N: never. C: current. F: former. BiTMS: bilateral rTMS. dTMS: deep rTMS. iTBS: intermittent theta burst stimulation. AD: antidepressant. BZ: benzodiazepine. AP: antipsychotic. Cumulative medication burden calculated using WHO-defined daily dose equivalents within medication class.

**

p < 0.01.

a

Differs from MBMI.

b

Differs from HBLI.

3.2. Trajectory modeling and class differences

For trajectory modeling of raw PHQ-9 scores, the selected model was a 3-class quadratic (LQ), based on BIC, minimum posterior probability (81.2%), and minimum class size (19.8% of total sample) (see Supplementary Table 1). Although the 2-class solution had a slightly more favorable BIC than the 3-class solution, the latter was selected for improved clinical relevance and interpretability (e.g., larger minimum class size, more distinct patient classes). The trajectories were labeled as High Baseline, Minimal Improvement (HBMI; n = 61), Moderate Baseline, Moderate Improvement (MBMI; n = 174), and High Baseline, Large Improvement (HBLI; n = 73) (Fig. 1A). At baseline, MBMI had the lowest PHQ-9 scores (14.1, SD = 3.9), whereas HBMI (22.5, SD = 2.6) and HBLI (21.5, SD = 3.2; p = 0.08) did not differ. All 3 classes differed by PHQ-9 score at mid- and post-treatment timepoints. By the end of treatment, HBLI (7.5, SD = 4.6; 65.5% reduction) had lower scores than MBMI (8.8, SD = 4.8; 36.8% reduction; p = 0.04) and HBMI (20.4, SD = 3.8; 9.0% reduction; p < 0.001), as well as the highest proportion of treatment responders (84%) relative to HBMI (13%) or MBMI (44%). The largest pre- to post-treatment decreases in PHQ-9 and SI scores were observed in HBLI (PHQ-9: Cohen’s d = 4.47 [4.06, 4.88]; SI: OR = 11.3 [4.48, 28.4]), followed by the MBMI (d = 1.69 [1.46, 1.93]; OR = 3.82 [2.04, 7.18]) and HBMI classes (d = 0.68 [0.32, 1.04]; OR = 1.92 [0.84, 4.37]), with changes in SI among HBLI patients 83% higher than HBMI, and 66% higher than MBMI, an expected result given that classes were defined by outcome trajectories. Sensitivity analysis for patients with baseline PHQ-9 scores >10 revealed similar results to the primary analysis (see Supplementary Tables 2 and 3). Class differences were also observed by depression-related history. HBMI had higher proportions of patients with prior ECT, SI, and SA (Table 1). Further sensitivity analysis exploring class membership for the 3-class and 2-class LCMMs with rTMS protocols as covariate revealed no differential probability of class assignment across rTMS protocols (see Supplementary Table 4).

Fig. 1.

Fig. 1.

Open in a new tab

Latent class mixed model-derived trajectories for raw PHQ-9 (left; 3-class quadratic) and baseline-adjusted PHQ-9 (right; 2-class quadratic) scores. Predicted values and 95% median percentiles (indicated by shaded ribbons) derived from 2000 draws of model posterior distributions.

A secondary trajectory analysis using within-patient baseline-adjusted PHQ-9 scores indicated a 2-class quadratic as the optimal fit, labeled as Moderate Improvement (MI; n = 236, 76.6%) and Large Improvement (LI; n = 72; 23.4%). The shift from a 3-class to 2-class solution is expected, as subtracting pre-treatment PHQ-9 scores reduces inter-individual variance that can drive class assignment when modeling raw values. Notably, pre-treatment PHQ-9 scores were higher in the LI (20.6, SD = 3.8) than MI group (16.6, SD = 5.3; p < 0.001), but 87% of LI patients became treatment responders, compared to only 35% in the MI group (X2 = 47.8, p < 0.001), as post-treatment scores were far lower in LI (6.9, SD = 4.6) versus MI (12.0, SD = 6.7) patients (Fig. 1B). Most of the LI group (N = 50; 69.4%) consisted of HBLI patients, followed by MBMI patients (N = 21; 29.2%); only 1 HBMI patient was also in the LI group.

3.3. Baseline predictors of class membership

Multinomial logistic regression (LASSO) was implemented to identify patient characteristics at baseline that were predictive of class membership. Holdout test set performance was considerably higher than chance (mean multiclass AUC across imputations = 0.817, 95% CI: 0.805–0.827; mean Cohen’s kappa = 0.524, 0.490–0.556). Examination of class-vs-class discrimination performance indicated that the model effectively classified patients into HBMI versus MBMI (mean AUC = 0.952, 0.941–0.962) and HBLI versus MBMI (mean AUC = 0.913, 0.902–0.924); however, only modestly discriminated between HBMI and HBLI (mean AUC = 0.674, 0.643–0.707). Significant patient characteristics associated with HBMI was history of SI (OR = 1.19, 1.14–1.28) and current or former cannabis use (OR = 0.763, 0.734–0.790), whereas for HBLI, current or former tobacco use (OR = 1.17, 1.13–1.22), higher BMI (OR = 0.897, 0.868–0.921), prior SA (OR = 0.913, 0.890–0.933), comorbid anxiety (OR = 0.93, 0.908–0.947), and college degree (OR = 0.937, 0.916–0.953) were associated. For MBMI, lower pre-treatment PHQ-9 score was the only predictor (OR = 0.093, 0.078–0.114).

When adjusting for pre-treatment PHQ-9 scores, binomial logistic regression models failed to distinguish between MI and LI classes better than chance (AUC = 0.521, 95% CI: 0.506, 0.543). In a secondary analysis, the inclusion of pre-treatment PHQ-9 scores improved classification accuracy (AUC = 0.72 [0.69, 0.74]); however, baseline symptom severity dominated feature selection, and no other patient characteristics demonstrated independent predictive value.

4. Discussion

In this study we utilized LCMM to identify trajectories of treatment response, as measured by the PHQ-9, in a naturalistic sample of patients receiving rTMS for TRD. Results revealed that a 3-class quadratic model best explained trajectories of response. We replicated findings demonstrating rapid early response in the group with the lowest baseline severity (MBMI) with deceleration of improvement later in their trajectory. Unique to our sample, we characterized a group with high baseline PHQ-9 that experienced large improvement. We also identified that history of SI and cannabis use were associated with less improvement in patients with high baseline PHQ-9. A history of SA was also associated with lower odds of a HBLI trajectory.

Past studies have reported on latent trajectories of rTMS response in TRD in naturalistic samples. Chen and colleagues utilized a combination of cubic and quadratic equations to identify group trajectories of patients receiving an extended rTMS course of 51 sessions into: rapid response-standard treatment length, slowed response, linear response-extended course, and nonresponse groups (Chen et al., 2024). They found that high baseline depression scores were associated with nonresponse. Jannsen-Aguilar and colleagues analyzed trajectories of response for patients receiving iTBS for TRD and grouped patients into: lower baseline-rapid improvement, lower baseline-gradual improvement, higher baseline-rapid improvement and highest baseline-minimal improvement groups (Janssen-Aguilar et al., 2025). Again, a high baseline PHQ-9 predicted poor response. Also, early response was found to be a strong predictor of later response and remission. Similar analyses have been completed in clinical trial data. LCMM analysis with cubic response trajectories of modified patient global impression of changes (PGIC) scores from participants treated with rTMS in the BRIGHtMIND trial yielded the following groups: strong improver, delayed-moderate improver, plateau-moderate improver, non-improver and worsening (Briley et al., 2024). This is the only study to demonstrate a worsening rTMS trajectory. In another analysis that combined clinical trials comparing standard rTMS to accelerated rTMS, Kaster and colleagues utilized quadratic and linear response trajectories to describe the following groupings: nonresponse, minimal response, lower symptoms response, and higher symptoms response (Kaster et al., 2020). Accelerated treatment did not impact response profiles. Another analysis by Kaster and colleagues of the THREE-D study combined cubic, quadratic, and linear response profiles to produce the following groups: rapid response, lower baseline symptoms-linear response, higher baseline symptoms-linear response, and nonresponse (Kaster et al., 2019). Most recently, this group also conducted a group-based trajectory modeling approach on the FOUR-D study (Blumberger et al., 2022), and found similar response groupings: nonresponse, partial response, linear response and rapid response, in a late-life depression sample (Adler et al., 2025). Researchers have also attempted to explore other nonlinear modeling approaches, such as exponential decay functions, which may outperform linear approaches (Berlow et al., 2023). In summary, results from the aforementioned studies demonstrate that rTMS treatment trajectories are most often discretized into 3–5 groups, and that rates of improvement are often non-linear and vary between groups. Our 3-class model is simpler than those with 4–5 classes, which may be easier to implement in clinical settings than more complex models.

Our findings should be considered in the context of other known predictors of rTMS response in depression. Patients taking benzodiazepines (Hunter and Leuchter, 2020) and with comorbid anxiety tend to have worse outcomes (Benster et al., 2024), and those on stimulant medications experience improved outcomes (Hunter et al., 2019). Curiously, we found that tobacco use history was associated with positive rTMS outcomes, yet it is unclear if this represents a true effect or reflects the small sample size in our HBLI group. Future work should explore the interaction of these characteristics with neurophysiological predictors of response in the context of trajectory modeling.

Results from our analysis emphasize the importance of close clinical monitoring for patients with high baseline PHQ-9 scores and comorbid suicidality. They should be carefully considered for alternative interventional treatment approaches, such as electroconvulsive therapy (ECT) or ketamine, early in their rTMS course. Evidence suggests that IV ketamine and ECT exert rapid action on reduction of suicidality and depression, even for patients with severe baseline depression (Kellner et al., 2005; Wilkinson et al., 2018). Experimental methods such as accelerated TMS paradigms also demonstrate promise in these clinical scenarios.

Limitations of this study include the retrospective nature of this analysis, hampering causal inference, and the moderate sample size, which could affect our ability to detect the appropriate number of classes in the data. Heterogeneous rTMS treatment protocols and parameters used could also impact generalizability, and although no between-class differences in rTMS protocol were found, confounding by indication remains a possibility outside of a randomized trial. Lastly, although cost-sensitive weights were incorporated into prediction models (i.e., adjusting for class membership uncertainty, down-weighting the majority class), their performance may be somewhat optimistic, thereby necessitating replication and external validation.

In conclusion, our results provide a robust and meaningful addition to the growing field of rTMS response trajectory modeling. Future studies should aim to replicate the above key findings and fuse trajectory modeling with parallel forms of predictive science, such as neurophysiological biomarkers of response.

Supplementary Material

Supplement

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jad.2026.122167.

Footnotes

CRediT authorship contribution statement

Cory R. Weissman: Writing – review & editing, Writing – original draft, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Brian Cui: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Lindsay Benster: Writing – review & editing, Project administration, Methodology. Noah Stapper: Writing – review & editing, Methodology, Investigation, Formal analysis, Data curation. Jean Phillippe Miron: Writing – review & editing, Supervision, Project administration, Methodology, Conceptualization. Zafiris J. Daskalakis: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Conceptualization. Lawrence G. Appelbaum: Writing – review & editing, Supervision, Software, Resources, Project administration, Methodology, Investigation, Data curation, Conceptualization. Jordan N. Kohn: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Cory R Weissman reports financial support was provided by NIMH. Lawrence G. Appelbaum reports a relationship with Hybrid Harmony LLC that includes: equity or stocks. Lawrence G. Appelbaum reports a relationship with Wave Neuroscience LLC that includes: consulting or advisory. Lawrence G. Appelbaum reports a relationship with Eiro LLC that includes: funding grants. Lawrence G. Appelbaum reports a relationship with Wu-Tsai Human Performance Alliance that includes: funding grants. Zafiris J. Daskalakis reports a relationship with Brainsway that includes: board membership. Zafiris J. Daskalakis reports a relationship with Kreutzkamp Family Foundation that includes: funding grants. Zafiris J. Daskalakis reports a relationship with Wellcome Foundation that includes: funding grants. Lawrence G. Appelbaum reports a relationship with Wellcome Foundation that includes: funding grants. Cory R Weissman reports a relationship with Wellcome Foundation that includes: funding grants. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

  1. Adler E, Blumberger DM, Chen X, Lee HH, Nestor SM, Downar J, Mulsant B, Rajji TK, Noda Y, Daskalakis ZJ, Kaster TS, 2025. Trajectories of response to bilateral rTMS in late-life depression. Brain Stimul. 18, 2067–2074. 10.1016/j.brs.2025.10.020. [DOI] [PubMed] [Google Scholar]
  2. Benster L, Weissman C, Suprani F, Toney K, Afshar H, Stapper N, Tello V, Stolz L, Poorganji M, Daskalakis Z, Appelbaum L, Kohn J, 2024. Predictive Modeling of Response to Repetitive Transcranial Magnetic Stimulation in Treatment-resistant Depression. 10.21203/rs.3.rs-4396926/v1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Berlow YA, Zandvakili A, Brennan MC, Williams LM, Price LH, Philip NS, 2023. Modeling the antidepressant treatment response to transcranial magnetic stimulation using an exponential decay function. Sci. Rep 13, 7138. 10.1038/s41598-023-33599-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Blumberger DM, Mulsant BH, Thorpe KE, McClintock SM, Konstantinou GN, Lee HH, Nestor SM, Noda Y, Rajji TK, Trevizol AP, Vila-Rodriguez F, Daskalakis ZJ, Downar J, 2022. Effectiveness of standard sequential bilateral repetitive transcranial magnetic stimulation vs bilateral theta burst stimulation in older adults with depression: the FOUR-D randomized noninferiority clinical trial. JAMA Psychiatry 79, 1065. 10.1001/jamapsychiatry.2022.2862. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Briley PM, Webster L, Lankappa S, Pszczolkowski S, McAllister-Williams RH, Liddle PF, Auer DP, Morriss R, 2024. Trajectories of improvement with repetitive transcranial magnetic stimulation for treatment-resistant major depression in the BRIGhTMIND trial. Npj Ment. Health Res 3, 32. 10.1038/s44184-024-00077-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chen X, Blumberger DM, Downar J, Middleton VJ, Monira N, Bowman J, Kriske, Joseph, Kriske, John, Donachie N, Kaster TS, 2024. Depressive symptom trajectories with prolonged rTMS treatment. Brain Stimul. 17, 525–532. 10.1016/j.brs.2024.04.010. [DOI] [PubMed] [Google Scholar]
  7. Hunter AM, Leuchter AF, 2020. Benzodiazepine use and rTMS outcome. Am. J. Psychiatry 177, 172. 10.1176/appi.ajp.2019.19060603. [DOI] [PubMed] [Google Scholar]
  8. Hunter AM, Minzenberg MJ, Cook IA, Krantz DE, Levitt JG, Rotstein NM, Chawla SA, Leuchter AF, 2019. Concomitant medication use and clinical outcome of repetitive Transcranial Magnetic Stimulation (rTMS) treatment of Major Depressive Disorder. Brain Behav. 9, e01275. 10.1002/brb3.1275. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Janssen-Aguilar R, Al Hassan M, Al-Shamali HF, Menzies P, Dunlop K, Jha M, Lin Q, Lou W, Blumberger DM, Kaster T, Bhat V, 2025. Predictors of response and remission in treatment-resistant depression: outcomes and trajectories of repetitive transcranial magnetic stimulation. Brain Stimul. 18, 1187–1189. 10.1016/j.brs.2025.06.008. [DOI] [PubMed] [Google Scholar]
  10. Kaster TS, Downar J, Vila-Rodriguez F, Thorpe KE, Feffer K, Noda Y, Giacobbe P, Knyahnytska Y, Kennedy SH, Lam RW, Daskalakis ZJ, Blumberger DM, 2019. Trajectories of response to dorsolateral prefrontal rTMS in major depression: a THREE-D study. Am. J. Psychiatry 176, 367–375. 10.1176/appi.ajp.2018.18091096. [DOI] [PubMed] [Google Scholar]
  11. Kaster TS, Chen L, Daskalakis ZJ, Hoy KE, Blumberger DM, Fitzgerald PB, 2020. Depressive symptom trajectories associated with standard and accelerated rTMS. Brain Stimul. 13, 850–857. 10.1016/j.brs.2020.02.021. [DOI] [PubMed] [Google Scholar]
  12. Kellner CH, Fink M, Knapp R, Petrides G, Husain M, Rummans T, Mueller M, Bernstein H, Rasmussen K, O’Connor K, Smith G, Rush AJ, Biggs M, McClintock S, Bailine S, Malur C, 2005. Relief of expressed suicidal intent by ECT: a consortium for research in ECT study. Am. J. Psychiatry 162, 977–982. 10.1176/appi.ajp.162.5.977. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. McInnes AN, Olsen ST, Sullivan CRP, Cooper DC, Wilson S, Irem Sonmez A, Sophia Albott C, Olson SC, Peterson CB, Rittberg BR, Herman A, Bajzer M, Nahas Z, Widge AS, 2024. Trajectory modeling and response prediction in transcranial magnetic stimulation for depression. Pers. Med. Psychiatry 47–48, 100135. 10.1016/j.pmip.2024.100135. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. McIntyre RS, Alsuwaidan M, Baune BT, Berk M, Demyttenaere K, Goldberg JF, Gorwood P, Ho R, Kasper S, Kennedy SH, Ly-Uson J, Mansur RB, McAllister-Williams RH, Murrough JW, Nemeroff CB, Nierenberg AA, Rosenblat JD, Sanacora G, Schatzberg AF, Shelton R, Stahl SM, Trivedi MH, Vieta E, Vinberg M, Williams N, Young AH, Maj M, 2023. Treatment-resistant depression: definition, prevalence, detection, management, and investigational interventions. World Psychiatry 22, 394–412. 10.1002/wps.21120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Papakostas GI, Trivedi MH, Shelton RC, Iosifescu DV, Thase ME, Jha MK, Mathew SJ, DeBattista C, Dokucu ME, Brawman-Mintzer O, Currier GW, McCall WV, Modirrousta M, Macaluso M, Bystritsky A, Rodriguez FV, Nelson EB, Yeung AS, Feeney A, MacGregor LC, Carmody T, Fava M, 2024. Comparative effectiveness research trial for antidepressant incomplete and non-responders with treatment resistant depression (ASCERTAIN-TRD) a randomized clinical trial. Mol. Psychiatry 29, 2287–2295. 10.1038/s41380-024-02468-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Proust-Lima C, Philipps V, Liquet B, 2017. Estimation of extended mixed models using latent classes and latent processes: the R package lcmm. J. Stat. Softw 78. 10.18637/jss.v078.i02. [DOI] [Google Scholar]
  17. Sackeim HA, Aaronson ST, Carpenter LL, Hutton TM, Mina M, Pages K, Verdoliva S, West WS, 2020. Clinical outcomes in a large registry of patients with major depressive disorder treated with Transcranial Magnetic Stimulation. J. Affect. Disord 277, 65–74. 10.1016/j.jad.2020.08.005. [DOI] [PubMed] [Google Scholar]
  18. Van De Schoot R, Sijbrandij M, Winter SD, Depaoli S, Vermunt JK, 2017. The GRoLTS-Checklist: guidelines for reporting on latent trajectory studies. Struct. Equ. Model. Multidiscip. J 24, 451–467. 10.1080/10705511.2016.1247646. [DOI] [Google Scholar]
  19. Wilkinson ST, Ballard ED, Bloch MH, Mathew SJ, Murrough JW, Feder A, Sos P, Wang G, Zarate CA, Sanacora G, 2018. The effect of a single dose of intravenous ketamine on suicidal ideation: a systematic review and individual participant data meta-analysis. Am. J. Psychiatry 175, 150–158. 10.1176/appi.ajp.2017.17040472. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement
NIHMS2194803-supplement-Supplement.docx (33.2KB, docx)

RESOURCES