Real-world evidence of SAINT® effectiveness in a heterogenous clinical population with psychiatric and medical comorbidities: A multicenter, open-label, community-based study

Abstract

Objective

SAINT®, a functional connectivity-guided, accelerated form of intermittent theta-burst stimulation, has shown high efficacy in depression in open-label and randomized controlled trials. To evaluate real-world effectiveness across diverse clinical practice settings, an open-label, multicenter study was conducted.

Methods

Adult patients with major depressive disorder who failed to achieve satisfactory improvement from prior antidepressant medication in the current episode were enrolled across 7 private psychiatric clinics in the U.S. All participants received a 5-day course of SAINT. The primary outcome was the Clinical Global Impressions Scale of Improvement (CGI-I) score following the final treatment session with CGI-S (measure of severity) collected as a secondary measure.

Results

A per protocol analysis of the acute outcomes data included 101 participants. All patients who initiated SAINT treatment completed the full 5-day protocol, with no treatment discontinuations due to adverse events or tolerability concerns. At the primary endpoint, treatment day 5, 68.3% of participants achieved CGI-I ≤ 2 (“much improved”) with a mean CGI-I of 2.32. Meaningful clinical benefit (≥1 point reduction in CGI-S from baseline through 5 days following treatment) was observed in 89.1% of participants. Improvement rates did not significantly differ by baseline severity or across clinical sites.

Conclusions

This analysis of the primary outcome includes the largest cohort of patients treated with SAINT to date. The results support previously reported rapid and significant improvements in depressive symptoms and extend these findings to a more clinically representative population.

Highlights

• •SAINT is a rapid-acting and highly effective acute treatment for depression.
• •A multisite trial was conducted to test the generalizability of SAINT.
• •This report includes the largest cohort of patients treated with SAINT to date.
• •The results support rapid and significant improvements in depressive symptoms in a real-world setting.

Introduction

Transcranial magnetic stimulation (TMS) is an FDA-cleared, noninvasive brain stimulation treatment for major depressive disorder (MDD) [1], [2]. Although TMS offers a viable alternative to psychotherapy and psychopharmacological options, conventional treatment schedules require daily in-office visits for 6–9 weeks. This prolonged treatment schedule delays the onset of symptom relief, with full benefits often emerging over an average period of 4–6 weeks [3]. This prolonged schedule also limits access to TMS due to the logistical burden of daily travel or potential temporary relocation for over 6 weeks.

SAINT® (Stanford Accelerated Intelligent Neuromodulation Therapy; also referred to as Stanford Neuromodulation Therapy [SNT]) was developed to address the limitations of conventional TMS. SAINT utilizes a higher daily treatment dose of 18,000 intermittent theta burst stimulation (iTBS) pulses compared to 3000 10 Hz pulses, a condensed treatment schedule (5 days vs. 6–9 weeks), and individualized targeting based on functional connectivity compared to targeting based on anatomical heuristics [4], [5], [6]. An initial open-label trial of SAINT and a subsequent randomized controlled trial (RCT) demonstrated high efficacy with rapid-acting onset of clinical benefit [4], [5]. These promising reports led to the U.S. Food and Drug Administration (FDA) clearance of SAINT for the treatment of MDD in adult patients in 2022 [7], which was recently supported by a larger RCT replicating prior rapid and robust antidepressant effects [8].

Building on prior controlled clinical trials, the current study initiated a large, naturalistic, multicenter clinical trial designed with three primary objectives: 1) evaluate the external validity of prior efficacy findings; 2) assess the feasibility of implementing SAINT in routine clinical practice; and 3) examine the generalizability by opening enrollment to a clinical population with complex and comorbid presentations. Here we report an analysis of acute treatment outcomes following the conclusion of enrollment.

Methods

Study design and participants

This study is part of an ongoing, multicenter clinical trial evaluating the real-world effectiveness of SAINT for the treatment of MDD. Participants were enrolled across 7 geographically diverse, community-based psychiatric clinics in the United States. Participating clinics were located in Bellevue, Washington; Brooklyn, New York; La Jolla, California; Los Angeles, California; Natick, Massachusetts; and Sunnyvale, California. This trial was preregistered in the U.S. Clinical Trials registry (NCT05819021). Primary and secondary outcome metrics were defined in that registration prior to the first participant enrolled. Participants were recruited through a variety of methods including online advertisement available on both the study sponsor and treatment providers’ websites and physician referral. The study enrolled adult participants with a confirmed primary diagnosis of MDD in a major depressive episode who had failed to receive satisfactory improvement from a prior antidepressant medication in the current episode and were determined to be a good candidate for TMS. Participants were excluded if they had any contraindications to TMS or magnetic resonance imaging (MRI) including ferromagnetic implants or a seizure disorder. Participants were also excluded if they had an abnormal brain MRI or were deemed to be at high risk of suicide attempt or had made a suicide attempt within 6 months. To reflect real-world clinical practice, no baseline severity requirements were imposed, and participants were not required to meet minimum scores on clinical assessments to be enrolled in this trial.

All research procedures were conducted in accordance with the ethical standards outlined in the Declaration of Helsinki. This study was approved by a fully accredited independent Institutional Review Board, Advarra IRB. All participants provided written informed consent following a full explanation of study procedures and prior to participating in any study activities. This was a postmarket trial in which participants paid for treatment, reflecting the current accessibility of SAINT in routine, non-research clinical practice.

Procedures

Imaging and fMRI target generation

Prior to treatment, all participants underwent a structural and functional MRI for individualized left dorsolateral prefrontal cortex (L-DLPFC) target generation. All scanners were 3 T with ≥ 20 channel head coils. Detailed imaging protocol and target generation procedures have been described previously [4], [5]. MRI scans were conducted at locally validated imaging centers, and the cost of MRI acquisition was included as part of the treatment cost.

SAINT

SAINT® was implemented using the Magnus Neuromodulation System (Magnus Medical, Inc.), an FDA-cleared integrated platform that combines stimulation hardware with proprietary software specifically configured for SAINT delivery [4], [6], [9]. The treatment sessions consisted of 50 iTBS sessions delivered over 5 consecutive days (10 sessions per day with a 50-minute intersession interval). Stimulation parameters utilized iTBS in 60 cycles of 10 bursts of three pulses at 50 Hz in 2-second trains (5 Hz) with an 8-second inter-train interval. Ten sessions were delivered per day (18,000 pulses/day) for a total of 90,000 pulses at 90% resting motor threshold (rMT), corrected to adjust for target depth [10] and not exceeding 120% rMT for safety or as limited by tolerability. Target selection was derived using a proprietary functional connectivity–based targeting algorithm originally developed at Stanford University and deployed through SNT, a secure cloud-based platform maintained by Magnus Medical. SNT generated individualized stimulation targets that were subsequently used for neuronavigation-guided treatment delivery.

All study site personnel completed a standardized training and certification program developed by experienced SAINT research staff prior to participant enrollment. Training included MRI-guided neuronavigation, protocol-specific treatment delivery, participant safety, and device operation, with competency assessments conducted during both pre-enrollment training and supervised treatment of initial study participants. Protocol adherence and participant safety were further supported through ongoing onsite and remote monitoring conducted by trained clinical research associates in accordance with GCP, FDA, and ICH guidelines, including review of treatment delivery, eligibility criteria, source documentation, and regulatory compliance throughout the study.

Clinical assessments

The primary outcome was the Clinical Global Impressions Scale of Improvement (CGI-I) score following the final treatment session on day 5 and was rated by a study clinician conducted in accordance with established administration guidelines and anchor-based approaches described by Busner and Targum [11], [12]. The CGI-I was selected as the primary outcome because it reflects the clinician’s global assessment of change, incorporating symptom improvement, patient functioning, and overall clinical presentation. The CGI-I was also assessed 5 days posttreatment. The CGI-S, a measure of symptom severity [13], was assessed by a study physician at baseline, following the final treatment (day 5) and 5 days posttreatment. The Montgomery–Åsberg Depression Rating Scale (MADRS) was included as an exploratory measure to provide detailed symptom-level context [14]. The MADRS was assessed at baseline and 5 days posttreatment. To ensure interrater reliability and minimize bias, the MADRS was conducted by third-party raters specializing in psychiatric evaluations with no financial conflict of interest in the outcomes. For all clinician-rated assessments, instruction was given to maintain the same rater for individual participants whenever feasible to enhance longitudinal consistency. The self-report version of the MADRS (MADRS-S) was used as a complementary measure to the clinician-assessed MADRS and collected at baseline and 5 days posttreatment [15]. The Young Mania Rating Scale (YMRS) was administered at baseline and each day of treatment to assess for treatment emergent mania (defined as an increase of ≥ 2 points on 2 or more items [16].

Cognitive changes were assessed at baseline and following the final treatment session on day 5 using Creyos neurocognitive online assessments: Double Trouble (response inhibition), Feature Match (attention), Grammatical Reasoning (verbal reasoning), Monkey Ladder (visuospatial working memory), Odd One Out (deductive reasoning), Token Search (working memory) [17], [18]. Assessments were administered either in-person (60.6%) or via a provided hyperlink (39.4%) and completed on either a desktop (99%) or tablet (1%).

Demographic and socioeconomic information (Tables 1 and S1, respectively) was self-reported by participants. Concomitant medication (Table S2) and comorbidities (Table S3) were recorded by the investigator or designee using participant medical records and self-report. Depression history was assessed using years since diagnosis and length of current depressive episode.

Clinical outcome analysis

Following closure of enrollment, an analysis was conducted to evaluate acute treatment outcomes in line with the preregistered protocol. The intention-to-treat (ITT) cohort included all participants who enrolled in the study and initiated at least one treatment session. The per-protocol (PP) cohort included participants from the ITT sample who completed the full SAINT treatment as defined by the study protocol (i.e. at least 48/50 SAINT sessions within 5 consecutive days, completed the primary outcome assessment, the CGI-I, on treatment day 5 or the day following and had no documented major protocol deviations). Fidelity across sites to the intervention was assessed from reported protocol deviations. All analyses were done in R (version 4.1.2) [19], [20], [21], [22], [23].

The primary outcome for this study was the CGI-I score in the per protocol (PP) study cohort at the end of treatment day 5. Specifically, we tested whether the group mean was significantly less than a score of 4, where a score of 4 reflects no change in symptoms following treatment and lower scores indicate greater improvement. This analysis was performed using a one-sample, two-tailed test with an alpha level of 0.05. The Shapiro-Wilk test was used in conjunction with visual inspection of quantile-quantile-plots (QQ-plots) to determine if assumptions of normality were violated. In such an event, a Wilcoxon signed rank test with continuity correction was performed using the wilcox.test function from the {stats} package in R. Biserial rank correlation coefficient is reported for the effect size of this analysis. This analysis was also performed on both the PP and ITT populations.

To evaluate whether the primary outcome (mean CGI-I score at the fifth day of treatment) remained significantly below 4 while accounting for covariates, each covariate was entered individually into a linear model predicting CGI-I score. The marginal means across all observations were estimated with the {emmeans} package in R [24]. The marginal mean CGI-I was tested if it remained significantly different from 4 using two-sided t-tests derived from these models. Covariates tested included, age, sex, gender, study site, years in current depressive episode, and baseline CGI-S score. These covariates were chosen based on prior evidence demonstrating association between these factors and antidepressant treatment outcomes in TMS and ECT research [25], [26], [27], [28], [29], [30]. Because our dataset did not include the number of failed antidepressant trials, episode duration served as the most appropriate proxy for treatment resistance. Finally, because this study was conducted across 7 community clinics, site was included as a covariate to rule out potential site differences.

Aligning with previous reports in real-world trial designs and to best capture improvements in an MDD population with high rates of medical and psychiatric comorbidities, we report the percentage of participants rated as a score of ≤ 2 on the CGI-I (“much improved”) within 1 week posttreatment (the lower value at either the end of treatment day 5 or 5 days posttreatment) [28]. This threshold was selected a priori to provide a baseline-independent measure of global improvement and to capture clinically meaningful change in a sample with no imposed minimum baseline severity and with comorbid conditions. Improvement rates by study site were compared using a Pearson’s Chi-square test with the chisq.test function from the {stats} package in R.

Covariates associated with improvement were assessed with independent generalized linear models (GLMs) using the glm function from the {stats} package in R within the per-protocol population using the CGI-I score ≤ 2 status as a binary outcome [27], [30], [31]. Each independent model was compared to a null model with the likelihood ratio test (LRT). If any significant (p_FDR < 0.05) univariate predictors were found, a follow-up multivariate analysis was performed to explore joint contributions to primary and secondary outcomes. Covariates tested were consistent with those tested for the primary outcome.

Clinically meaningful benefit was defined as a reduction in the CGI-S by 1 or more points [32], [33]. For results categorizing by treatment severity at baseline using the CGI-S, severity was categorized as: asymptomatic: 1, mild: 2–3, moderate: 4–5, severe: 6–7. Frequencies were compared using a Pearson’s Chi-square test with the chisq.test function from the {stats} package in R.

For posttreatment CGI-S differences, baseline values were compared to values collected within 1 week of completing treatment (the lower value of either the end of treatment day 5 or 5 days posttreatment). For posttreatment MADRS and MADRS-S differences, baseline values were compared to the 5-day posttreatment values. Imputing missing MADRS and MADRS-S values was performed using the {mice} package in R [34]. One hundred complete datasets were generated using non-missing participant demographics (age, sex assigned at birth, gender, ethnicity, race, education level, employment status, and income), and convergence was confirmed by visual inspection. Estimated differences between baseline and posttreatment values were pooled and summarized across all datasets. For posttreatment YRMS differences, baseline values were compared to the end of treatment day 5 values. For all assessments, a two-sided, paired t-test using t.test from the {stats} package in R with an alpha level of 0.05 was used.

Because MDD presents with substantial heterogeneity across symptomatic domains, we explored whether SAINT reduced specific symptoms relevant to our sample. Four MADRS sub-scale items (anxiety/inner tension, sleep disturbance, concentration difficulties, and suicidal thoughts) were chosen because they represent distinct depressive symptom factors [35] and were highly prevalent within our cohort (approximately 42% with anxiety, 25% with ADHD, and nearly 23% with insomnia; Table S3). Suicidality was also included given its clinical importance and prior evidence of its rapid improvement with SAINT [4], [36], [37]. Participants reporting 2 or greater on these MADRS items at baseline were included. This criteria was decided a priori and chosen to exclude participants not endorsing symptoms or with low severity symptoms where no or minimal effect of treatment would be expected. Differences between baseline and 5 days posttreatment were determined with a two-sided, paired t-test using t.test from the {stats} package in R with alpha set to 0.05. Bonferroni correction was used to adjust p-values for multiple comparisons.

For Creyos neurocognitive assessments, differences in raw scores between baseline and 5 days posttreatment were determined with a two-sided, paired t-test using t.test from the {stats} package in R with alpha set to 0.05. Bonferroni correction was used to adjust p-values for multiple comparisons.

Concomitant medications were coded to their respective RxNorm Concept Unique Identifier (RxCUI) using RxTerms [38]. Anatomical Therapeutic Chemical (ATC) classifications for each medication were obtained from the RxClass API. Psychostimulants were classified using the ATC level 3 class name: PSYCHOSTIMULANTS, AGENTS USED FOR ADHD AND NOOTROPICS (class ID: N06B). Benzodiazepines were classified using the ATC level 4 class names: Benzodiazepine derivatives and Benzodiazepine related drugs (class IDs: N03AE, N05BA, N05CD, N05CF). Medications were included if reported to be taken during enrollment and not posttreatment.

Personalized functional target coordinates defined in each participant’s native T1-weighted space were transformed to MNI152 standard space using Advanced Normalization Tools (ANTs; antsApplyTransformsToPoints) to enable group-level localization [39]. Beam F3 (−40.6, 41.7, 34.3) was used as a conventional dorsolateral prefrontal cortex reference target [40]. Euclidean distance between each target coordinate and Beam F3 was calculated in MNI space, and mean and standard deviation were computed across participants. Spearman correlation was performed to assess the association between distance to Beam F3 and minimum CGI-I score within 1 week posttreatment.

Role of funding source

This multisite trial was sponsored by Magnus Medical Inc. The clinician site investigators participated in study design, and all authors had full access to the data and retained final responsibility for clinical oversight, interpretation of the data, and the decision to submit the manuscript for publication. Study conduct, data collection, statistical analyses, and interpretation were performed in accordance with prespecified protocols and procedures intended to preserve scientific rigor and data integrity.

Results

Participant characteristics

A total of 119 participants were assessed for eligibility. Prior to enrollment, 11 participants were excluded; 108 participants were allocated to receive intervention and were included in the ITT cohort (mean years since diagnosis = 19.4, mean years in current depressive episode = 3.8; Table 1). A total of 107 participants were treated with a full course of SAINT, and 101 participants were included in the per protocol analysis (Figure S1) for this analysis of acute clinical outcomes. Only 3 treatment-related protocol deviations were reported indicating that sites delivered the intervention consistently and as intended. Baseline demographic and clinical characteristics are summarized in Table 1, socioeconomic demographics are presented in Table S1, concomitant medication information is summarized in Table S2, and psychiatric comorbidities are presented in Table S3. Socioeconomic demographics was included in the data collection after enrollment began, resulting in missing data for approximately 24% of the ITT cohort and disproportionally affecting participant data from sites that enrolled participants earlier in the study. However, participants without socioeconomic data did not differ from those with socioeconomic data in both mean baseline MADRS (t(35.0) = 1.36, p = 0.18) or mean posttreatment MADRS (t(32.3) = 0.57, p = 0.57). The most prevalent psychiatric comorbidities observed in this trial, anxiety disorders and ADHD, are consistent with rates reported in epidemiological studies [41], [42], which similarly identify these conditions as among the most common co-occurring disorders in individuals with depression (Table S3). In addition, of the participants who reported a co-occurring thyroid disorder, 71% (10/14) were taking a medication used to manage this condition at baseline.

Table 1.

Participant demographics and clinical characteristics.

			ITT	PP
Age at Consent			N = 108	N = 101
	Mean (SD)		43.2 (18.8)	42.96 (18.5)
	Median [min, max]		40.5 [18, 85]	41 [18, 85]
Sex Assigned at Birth
	Female		48 (44.4%)	45 (44.6%)
	Male		60 (55.60%)	56 (55.4%)
Gender
	Man		56 (51.9%)	52 (51.5%)
	Woman		46 (42.6%)	43 (42.6%)
	Non-Binary		3 (2.8%)	3 (3%)
	Transgender Woman		3 (2.8%)	3 (3%)
Race
	White		89 (82.4%)	83 (82.2%)
	Asian		15 (13.9%)	15 (14.9%)
	Asian | White		2 (1.9%)	1 (1%)
	Black or African American		1 (0.9%)	1 (1%)
	Native Hawaiian or Other Pacific Islander | White		1 (0.9%)	1 (1%)
Ethnicity
	Not Hispanic or Latino		102 (94.4%)	95 (94.1%)
	Hispanic or Latino		4 (3.7%)	4 (4%)
	Not reported		2 (1.9%)	2 (2%)
CGI-S Baseline Severity
	Mild (2−3)		7 (6.5%)	6 (5.9%)
	Moderate (4−5)		58 (53.7%)	55 (54.5%)
	Severe (6−7)		43 (39.8%)	40 (39.6%)
Years Since Diagnosis
	Mean (SD)		19.4 (14.5)	19.0 (14.2)
	Median [min, max]		16.2 [0.0, 60.8]	16.15 [0.0,60.8]
Years in Current Episode
	Mean (SD)		3.8 (5.9)	3.5 (5.2)
	Median [min, max]		1.7 [0.1, 34.6]	1.74 [0.1,34.1]
MADRS Baseline Score			N = 105	N = 98
	Mean (SD)		22.0 (7.3)	21.9 (7.3)
	Median [min, max]		21 [9, 41]	21 [9, 41]

Depression symptoms

The primary outcome, mean CGI-I at the end of treatment day 5 in the per protocol cohort, was 2.32 (SD = 0.92, n = 101, V = 16, p < 0.0001, r = −0.99), significantly lower than a score of 4 which would have indicated no change in symptoms (Fig. 1A). Similar results were observed in the ITT cohort (mean = 2.33, SD = 0.92, n = 107, V = 17, p < 0.0001, r = −0.99). No covariates were found to impact the primary outcome (Table S4).

Fig. 1.

Posttreatment CGI-I, percent clinically meaningful benefit and percent CGI-I ≤ 2. A.) Count of participant CGI-I scores at end of treatment day 5. Visual inspection of QQ-plots and the Shapiro-Wilk normality test indicated a deviation from normality (W = 0.88, p < 0.0001). Wilcoxon signed rank test with a continuity correction indicated that the mean CGI-I was less than 4 (mean = 2.32, SD = 0.92, n = 101, V = 16, p < 0.0001, r = −0.99). B.) Percent clinically meaningful difference defined as a reduction of 1 point in CGI-S within one week of treatment. All severities (n = 101), mild severity (n = 6), moderate severity (n = 55), severe (n = 40). C.) Percent CGI-I ≤ 2 within 1 week of treatment. All severities (n = 101), mild severity (n = 6), moderate severity (n = 55), severe (n = 40). Baseline severity for B.) and C.) defined by CGI-S as described in the Methods section.

In line with these results, 89.1% of participants in the per protocol exploratory analysis achieved a clinically meaningful benefit following treatment (Fig. 1B). Categorizing participants by baseline symptom severity according to CGI-S scores revealed a clinically meaningful benefit in 83.3% of participants with mild severity, 87.3% with moderate severity, and 92.5% with severe symptoms; these rates did not statistically differ ($\chi$²(2) = 0.87, p = 0.65). Rates of clinically meaningful benefit also did not differ between study sites ($\chi$²(6) = 4.46, p = 0.61).

An additional exploratory analysis demonstrated that across the entire per protocol cohort, 68.3% of participants scored ≤ 2 on the CGI-I within 1 week posttreatment (Fig. 1C). This criterion was met by 83.3% of participants with mild severity, 70.9% with moderate severity, and 62.5% with severe symptoms. No statistical difference in improvement rates (CGI-I ≤2) between baseline severities was observed ($\chi$²(2) = 1.42, p = 0.49). These rates also did not statistically differ between study sites ($\chi$²(6) = 5.70, p = 0.46), and no covariates remained significant after multiple comparisons.

A consistent reduction in the CGI-S, MADRS and MADRS-S was observed following treatment. The average decrease in the CGI-S from baseline to within 1 week of treatment was 2.21 points, a 43.7% reduction (Fig. 2A). Likewise, the average decrease in the MADRS from baseline to 5 days posttreatment was 9.02 points, a 41.3% reduction (Fig. 2B). Finally, the average decrease in the MADRS-S from baseline to 5 days posttreatment was 4.97 points, a 37.8% reduction (Fig. 2C). Taken together, these results demonstrate a consistent reduction in symptom severity as assessed by site clinicians (CGI-S), by third party raters (MADRS), and by participants (MADRS-S).

Fig. 2.

Change from baseline to posttreatment for CGI-S, MADRS, and MADRS-S. A.) CGI-S within 1 week of treatment was reduced compared to baseline (mean baseline/posttreatment difference = −2.21, SD = 1.47, 95% CI [−1.92, −2.50], t(100) = 15.146, p = 2.2e−16, Cohen’s d = 1.53, n = 101). CGI-S baseline (mean = 5.05, SD = 1.07), posttreatment (mean = 2.84, SD = 1.55). B.) Posttreatment MADRS was reduced compared to baseline (mean baseline/posttreatment difference = −9.02, SD = 9.22, 95% CI [−7.08, −10.97], t(88) = 9.23, p = 1.388e−14, Cohen’s d = 0.98, n = 89). MADRS baseline (mean = 21.9, SD = 7.54), posttreatment (mean = 12.8, SD = 10.2). C.) Posttreatment MADRS-S was reduced compared to baseline (mean baseline/posttreatment difference = −4.97, SD = 5.38, 95% CI [−3.73, −6.20], t(74) = 8.0, p = 1.295e−11, Cohen’s d = 0.92, n = 75). MADRS-S baseline (mean = 13.1, SD = 4.59), posttreatment (mean = 8.16, SD = 5.99). *** indicates p < 0.001 comparing posttreatment to baseline. Error bars represent ± 1 SD.

Notably, a fraction of MADRS and MADRS-S were missing compared to the CGI-S (11.9% and 25.7%, respectively). For missing MADRS, three participants were missing a baseline MADRS, five participants were missing a posttreatment MADRS, and four participants had MADRS assessments outside the time window (>3 days after the five day posttreatment visit). Imputing missing baseline and posttreatment MADRS total scores resulted in a similar estimated treatment effect compared to the estimated treatment effect for participants without missing MADRS assessments (mean baseline/posttreatment difference = −9.06, SE = 0.95, t = −9.56, df = 90.73, p = 2.21e−15). For missing MADRS-S, four participants were missing baseline MADRS-S, twenty participants were missing posttreatment MADRS-S, and two participants had MADRS-S assessments outside the time window (>3 days after the five day posttreatment visit). Imputing missing baseline and posttreatment MADRS-S total scores resulted in a similar estimated treatment effect compared to the estimated treatment effect for participants without missing MADRS-S assessments (mean baseline/posttreatment difference = −5.49, SE = 0.68, t = −8.10, df = 61.13, p = 2.95e−11).

Exploratory analyses of individual MADRS items demonstrated a consistent reduction across select symptoms. Participants reporting 2 or greater at baseline on these items showed, on average, a 1.2- to 1.5-point decrease at the 5-day posttreatment visit. Specifically, treatment reduced symptom severity for anxiety/inner tension (item 3, mean difference = −1.24, Fig. 3A), sleep (item 4, mean difference = −1.59, Fig. 3B), concentration (item 6, mean difference = −1.23, Fig. 3C), and suicidality (item 10, mean difference = −1.14, Fig. 3D).

Fig. 3.

Changes in individual MADRS items for anxiety, sleep, concentration and suicidality. Select individual MADRS items decreased from baseline to 5 days posttreatment for participants reporting 2 or greater for a given item: A.) Anxiety (item 3; mean baseline/posttreatment difference = −1.24, SD = 1.18, t(66) = 8.58, p = 2.43e−12, p_corrected = 9.72e−12, 95% CI [−0.95, −1.53], Cohen’s d = 1.05, n = 67). Anxiety baseline (mean = 3.07, SD = 0.89), posttreatment (mean = 1.84, SD = 1.31). B.) Sleep disturbances (item 4; mean baseline/posttreatment difference = −1.59, SD = 1.50, t(38) = 6.62, p = 8.043e−8, p_corrected = 3.22e−7, 95% CI [−1.10, −2.08], Cohen’s d = 1.06, n = 39). Sleep baseline (mean = 3.00, SD = 0.86), posttreatment (mean = 1.41, SD = 1.39). C.) Concentration (item 6; mean baseline/posttreatment difference = −1.23, SD = 1.46, t(55) = 6.30, p = 5.34e−8, p_corrected = 2.14e−7, 95% CI [−0.84, −1.62], Cohen’s d = 0.84, n = 56), Concentration baseline (mean = 3.05, SD = 1.00), posttreatment (mean = 1.82, SD = 1.59). D.) Suicidality (item 10; mean baseline/posttreatment difference = −1.14, SD = 1.25, t(49) = 6.47, p = 4.37e−8, p_corrected = 1.75e−7, 95% CI [−0.79, −1.49], Cohen’s d = 0.92, n = 50). Suicidality baseline (mean = 2.46, SD = 0.71), posttreatment (mean = 1.32, SD = 1.17). p_corrected are corrected p-values using Bonferroni correction for multiple comparisons. *** indicates p_corrected < 0.001 comparing posttreatment to baseline. Error bars represent ± 1 SD.

Exploratory analyses investigated the modulatory effects of concomitant medications previously reported to influence treatment outcomes [23], [43]. The reported use of psychostimulants was associated with an additional 4.31-point decrease in the MADRS, on average, at 5-days posttreatment compared to baseline (p = 0.038, Figure S2A). The MADRS decrease from baseline to 5-days posttreatment was similar between participants reporting benzodiazepine use and those not reporting benzodiazepine use (Figure S2B).

Treatment targets were mapped to MNI space to assess proximity to Beam F3. The mean Euclidean distance between targets and Beam F3 was 15.5 mm (SD = 6.4). There was no association between distance to Beam F3 and minimum CGI-I score within 1 week posttreatment (Spearman ρ = −0.06, p = 0.58; Fig. 4), indicating that clinical outcomes were not systematically related to Beam F3 proximity.

Fig. 4.

Target distance to Beam F3 and posttreatment CGI-I. Minimum CGI-I score within 1 week posttreatment in relation to Euclidean target distance from Beam F3. Gray bands represent 95% confidence intervals.

Safety

Adverse events (AEs) reported were in line with reports from previous SAINT trials and other TMS studies. The most frequently reported AEs included headache, fatigue, discomfort at the stimulation site, difficulty sleeping, dizziness, and tinnitus (Table S5). These most frequently reported AEs were resolved in an average of 2.6 days with more than half (50.7%) resolving the same day.

Two serious adverse events (SAE) occurred during this reporting period. One SAE involved a psychiatric hospitalization for suicidal ideation followed by nonsuicidal self-injury with the participant denying suicide attempt. The event occurred 2 days after SAINT treatment. Based on the timing of the event, clinical review, and absence of (hypo)manic symptoms during the treatment period, the SAE was determined by the site principal investigator to be unrelated to treatment. An additional SAE was reported for a participant who underwent a planned 2-day hospitalization. This admission was arranged after an incidental finding on the study MRI prompted the participant’s physician to recommend a clinical MRI. Inpatient evaluation was pursued to expedite the imaging workup. The hospitalization was deemed not medically necessary and undertaken primarily for logistical convenience. No other SAEs were reported in the acute study period.

There were no reports of treatment-emergent mania as measured by the YMRS. In addition, YMRS total scores did not differ between baseline and treatment day 5 (mean difference = 0.22, SD = 1.43, 95% CI [−0.065, 0.50], t(100) = 1.53, p = 0.13, Cohen’s d = 0.152, baseline mean (SD) = 0.406 (1.01), treatment day 5 mean (SD) = 0.624 (1.24), n = 101).

Creyos neurocognitive assessments revealed no decline in performance from baseline to posttreatment, and any increase in performance from baseline is likely attributed to a practice effect in accordance with Creyos Report Interpretation Guide (Table S6) [44].

Discussion

Although controlled clinical trials are essential for establishing efficacy under defined conditions, complementary evaluation in routine clinical practice provides important insight into real-world effectiveness and implementation. Accordingly, this study evaluated SAINT in a naturalistic clinical setting, thereby extending existing evidence into broader, community-based practice contexts. Outcomes of this study demonstrate a significant acute reduction in symptoms as evidenced by an average CGI-I of 2.32 (SD = 0.92) at the end of treatment day 5, the primary endpoint. Moreover, an exploratory analysis demonstrated that 89.1% of participants experienced a clinically meaningful benefit and 68.3% achieved much improvement within 1-week posttreatment. This effectiveness was observed even with a clinically complex population inclusive of comorbid conditions typically excluded from controlled trials such as obsessive-compulsive disorder, Parkinson’s disease, and Alzheimer’s disease. Participants also had co-morbid diagnoses of endocrine, metabolic, and autoimmune disorders, further increasing cohort heterogeneity. Such medical comorbidities, particularly those associated with inflammatory and endocrine dysregulation, are well documented to reduce antidepressant treatment response and increase risk of treatment-resistant depression [45], [46], [47], [48], [49].

The present study was designed as a real-world, naturalistic investigation, and our primary objective was to capture clinical improvement as judged by treating clinicians without adding undue burden to the clinical workflow. The CGI-I is a global assessment of patient function, including comorbid conditions. Because psychiatric and medical comorbidities were common in our sample, we chose CGI-I ≤ 2 as our threshold for demonstrating a substantial improvement a priori to account for potential residual symptoms from comorbidities even if depression symptoms improved.

SAINT includes multiple adjustments that differentiate it from conventional TMS or iTBS protocols, such as adapting clinical workflows to accommodate functional MRI scans for target generation and 10-hour treatment days. This study supports that community-based clinics can successfully implement SAINT. In addition to logistical feasibility, high fidelity to the intervention was demonstrated with 94% of participants treated per protocol. Additionally, clinical outcomes did not significantly differ across sites.

Despite a limited number of assessments in this trial, individual items from the MADRS can be used to investigate the effect of treatment on endorsed symptomatology. Data from this study showed significant and robust improvements in patient-reported measures of anxiety, concentration, and sleep disturbances, consistent with results from prior trials [4], [5]. Notably, in the first open-label study of SAINT, 100% remission from suicidality following treatment was measured by the Columbia-Suicide Severity Rating Scale, with sustained remission at the 1-month follow-up [4]. Although a dedicated scale to measure suicidality was not included in this study, the MADRS item 10 measures suicidality and showed a significant decrease. Although these symptom improvements are encouraging, it should be noted that these results are from a small, overlapping sample of participants, limiting their interpretability, and a more comprehensive evaluation with domain-specific assessments is needed to understand how these symptoms are affected by treatment.

The absence of a relationship between Euclidean distance from Beam F3 and acute clinical improvement indicates that therapeutic response was not systematically related to proximity to this conventional scalp-based target within the present image-guided treatment cohort. Importantly, this finding alone does not establish the superiority of individualized fMRI-guided targeting over anatomical heuristics, as all participants received personalized image-guided stimulation and no direct Beam F3 comparison arm was included. Rather, the observed variability in individualized target locations relative to Beam F3, together with the lack of a systematic relationship between Beam F3 proximity and outcome, is consistent with the broader hypothesis that optimal stimulation sites may vary across individuals due to inter-individual variability in functional neuroanatomy [50]. These findings therefore align with, but do not independently prove, the rationale for individualized image-guided targeting approaches proposed in prior work [51], [52], [53], [54]. Although the open-label nature of the present study precludes a direct causal inference, a direct comparison between fMRI-guided and anatomical targeting in a randomized, controlled trial design would further support this finding.

More attention has shifted toward increasing neuroplasticity by concurrently administered psychostimulants and other medications prior to therapeutic interventions to improve outcomes [9], [55], [56]. This naturalistic trial included a sizable cohort of participants taking prescribed psychostimulants, allowing us to evaluate whether these medications improved treatment response. In line with previous reports [23], [43], we found that participants taking psychostimulants showed greater symptomatic improvement compared to participants who were not taking psychostimulants. Additional studies specifically investigating the use of psychostimulates and outcomes following treatment are needed to confirm this exploratory finding. Conversely, some studies have reported a decreased response to neuromodulation interventions in patients taking benzodiazepines [43], [57], while other studies report no effect [58]. We identified a similar reduction in symptoms in our cohort of participants taking benzodiazepines as compared to those not taking benzodiazepines. Continued collection of detailed concomitant medications is required to further understand the interaction of co-administered pharmacotherapies and neuromodulation interventions.

Limitations

This study included an MDD population with a wider range of depressive symptom severity and disease chronicity than prior SAINT/SNT trials which focused on highly treatment-resistant cohorts. This study design may improve understanding of SAINT in routine clinical practice and address an important gap in the literature regarding the effectiveness of SAINT outside of narrowly defined TRD populations. However, because formal treatment-resistance criteria and detailed antidepressant failure histories were not systematically collected, direct comparison with prior SAINT studies in rigorously characterized TRD cohorts is limited.

This open-label study was conducted in a self-pay population during a period when access to SAINT and insurance coverage by payors were limited; therefore, the cohort is representative of a subset of patients able to access treatment at that time. In addition, the absence of a sham comparator and the self-pay structure may have contributed to expectancy effects, as both participants and investigators were aware of the intervention and its perceived value [59]. Nevertheless, sham remission rates reported in prior SAINT sham-controlled trials have been relatively low, including 13% in the initial randomized controlled trial and 20% in a subsequent replication study, suggesting that placebo response alone is unlikely to fully account for the magnitude of clinical improvement observed in this cohort [5], [8]. The self-pay model may also have limited the demographic diversity of the sample. The sample was predominantly White, non-Hispanic, and highly educated, which limits the demographic and socioeconomic representativeness of the findings. Although the study included a highly comorbid psychiatric patient population, the results may not generalize to broader socioeconomic and demographically diverse populations.

The CGI-I was selected as the primary outcome measure due to its use in standard clinical practices and good interrater reliability [13]. However, the use of the CGI-I may be limited as it is less sensitive to capturing subtle, yet clinically relevant, changes following intervention. Additionally, given the absence of a control group and the limited validation of Creyos in intensive neuromodulation populations, observed cognitive changes, including improvements on response inhibition tasks, may reflect a combination of practice effects and potential treatment-related influences, and warrant cautious interpretation.

Finally, as this is an ongoing clinical trial with participants still active in the long-term follow-up phase of the study, data collection to evaluate participants over the 12 months following acute treatment is ongoing.

In conclusion, this study supports the feasibility, safety, and clinically meaningful symptom improvement associated with real-world outpatient implementation of SAINT in an MDD population with marked psychiatric comorbidity, medical comorbidity, symptom severity, and illness chronicity.

Declaration of Competing Interest

Dr. Feyder, Mr. Geoly, Ms. Ford, Dr. Lutkenhoff, Dr. Maron-Katz, and Dr. Stimpson currently receive a salary from and have equity/stock options in Magnus Medical. Dr. Bota has no disclosures to report. Dr. Carreon currently receives a salary from and has equity/stock options in Acacia Clinics. Drs. Muir and MacMillan are married and thus share disclosures: Co-Founders (with equity/stock options) in Radial Health, and have equity/options in Acacia Clincs; Ampa Health; Videra Health; Outro Health, Inc; Osmind PBC; Breakthrough6, Inc; First Tracks, PBC; Graymatters Health, Inc.; Brainify.ai; iRxReminder; Neurolief, Inc.; EmoBot, Inc; Psyrin, Inc.; hold stock in Brainsway, Inc., and have served as compensated advisors to Magnus Medical and Spring Health. Dr. Cook is Founder/Owner/Employee of the Los Angeles TMS Institute; he serves on DSMBs for NIH grants concerning TMS to Stanford University and the University of Southern California; has equity/options in NeuroSigma Inc and HeartCloud Inc; and is on-leave from a leadership role at NeuroSigma. His patents have been assigned to the Regents of the University of California, NeuroSigma, Monarch Biosciences, or HeartCloud. He is Professor Emeritus of Psychiatry and Bioengineering at UCLA. Dr. Solvason is Professor Emeritus Stanford Department of Psychiatry, Founder and Medical Director of IntegralTMS. Dr. Allen is a Partner/owner of Seattle Neuropsychiatric Treatment Center, and has served as a site Principal Investigator for trials with: Compass Pathways; Usona Institute; Janssen Research & Development; Wave Neuroscience; Alto Neuroscience; LivaNova; Kernel; Transcend Therapeutics; atai Life Sciences; and Magnus Medical; and has served on the Advisory Boards of LivaNova and Osmind. Dr. Pottanat is an owner of and receives a salary from Boston Precision Neurotherapeutics. Dr. Barbour receives salary from and has equity in Boston Precision Neurotherapeutics. Dr. Williams is a named inventor on Stanford-owned intellectual property relating to accelerated TMS pulse pattern sequences, neuroimaging-based TMS targeting, and novel psychedelic intervention for neuropsychiatric disorders; he is serving on scientific advisory boards for Soneira and Salma Health as a paid advisor; he also has equity/stock options in Soneira and Salma. Dr. Bentzley is a co-founder of, receives a salary from, and has equity/stock options in Magnus Medical and is an employee of Salma Health. He is also a named inventor on intellectual property relating to symptom tracking.

This multisite trial was sponsored and designed by Magnus Medical Inc. Magnus Medical, Inc. holds the exclusive rights to issued U.S. patents assigned to Stanford University covering the systems and methods that enable accelerated theta-burst stimulation protocols as well as precision neuronavigation and individualized targeting approaches. These patents include, among others, U.S. Patent Nos. 10,595,735 (accelerated protocols) and 11,839,455 (precision neuronavigation and targeting methods). The clinical sites were compensated by Magnus Medical for study participation. Seattle Neuropsychiatric Treatment Center (Dr. Allen), Los Angeles TMS (Dr. Cook), Brain Health Solutions (Dr. Bota), Fermata (currently known as Radial, Inc.; Dr. McMillan and Dr. Muir), Acacia Clinics (Dr. Carreon) are now SAINT commercial providers.

Appendix A. Supplementary material

Supplementary material