Predictive modeling of response to repetitive transcranial magnetic stimulation in treatment-resistant depression

Predictive modeling of response to repetitive transcranial magnetic stimulation in treatment-resistant depression | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Predictive modeling of response to repetitive transcranial magnetic stimulation in treatment-resistant depression Lindsay Benster, Cory Weissman, Federico Suprani, Kamryn Toney, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4396926/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Apr, 2025 Read the published version in Translational Psychiatry → Version 1 posted 11 You are reading this latest preprint version Abstract Identifying predictors of treatment response to repetitive transcranial magnetic stimulation (rTMS) remain elusive in treatment-resistant depression (TRD). Leveraging electronic medical records (EMR), this retrospective cohort study applied supervised machine learning (ML) to sociodemographic, clinical, and treatment-related data to predict depressive symptom response (>50% reduction on PHQ-9) and remission (PHQ-9 < 5) following rTMS in 232 patients with TRD (mean age: 54.5, 63.4% women) treated at the University of California, San Diego Interventional Psychiatry Program between 2017 and 2023. ML models were internally validated using nested cross-validation and Shapley values were calculated to quantify contributions of each feature to response prediction. The best-fit models proved reasonably accurate at discriminating treatment responders (Area under the curve (AUC): 0.689 [0.638, 0.740], p < 0.01) and remitters (AUC 0.745 [0.692, 0.797], p < 0.01), though only the response model was well-calibrated. Both models were associated with significant net benefits, indicating their potential utility for clinical decision-making. Shapley values revealed that patients with comorbid anxiety, obesity, concurrent psychiatric medication use, and more chronic TRD were less likely to respond or remit following rTMS. Patients with trauma and former tobacco users were more likely to respond. Furthermore, delivery of intermittent theta burst stimulation and more rTMS sessions were associated with superior outcomes. These findings highlight the potential of ML-guided techniques to guide clinical decision-making for rTMS treatment in patients with TRD to optimize therapeutic outcomes. Health sciences/Diseases/Psychiatric disorders/Depression Biological sciences/Psychology/Human behaviour Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION Major depressive disorder (MDD) is a prevalent and debilitating condition characterized by persistent feelings of sadness, hopelessness, and loss of interest or pleasure in daily activities ( 1 ). An estimated 322 million individuals worldwide meet criteria for MDD, representing a substantial public health challenge impacting quality-of-life, productivity, and overall well-being ( 2 ). While various pharmacological, psychotherapeutic, and neuromodulatory treatments are available, 30–40% of individuals with MDD experience treatment-resistant depression (TRD), defined as non-response to two or more trials of first-line interventions ( 3 ). Because MDD is a highly heterogeneous disorder, with 277 possible symptom combinations meeting DSM-5 criteria ( 4 ), it is possible that different subtypes or symptom clusters of MDD may respond differentially to various treatments ( 5 ). Therefore, aligning optimal treatment strategies to symptom presentations may be valuable for addressing TRD. Repetitive transcranial magnetic stimulation (rTMS) is an established clinical intervention for the treatment of TRD, with response rates of approximately 40–60%, providing a valuable treatment option for patients not responding to first line psychotherapy and pharmacotherapy ( 6 , 7 ). rTMS involves the application of repetitive magnetic pulses to the cortex, with the dorsolateral prefrontal cortex (DLPFC) a common target for MDD due to its role in mood regulation, cognitive control, executive functioning, and emotion processing ( 8 ). The DLPFC is also implicated in the pathophysiology of depression due to its connections with the limbic system, while acute elevations in mood following rTMS provide further evidence of DLPFC dysfunction in MDD ( 9 – 12 ). Ultimately, rTMS is thought to exert its therapeutic effects through modulation of cortical excitability, neurotransmitter systems, and neuroplastic mechanisms ( 13 ), but understanding the predictors of treatment response is crucial for tailoring treatment strategies and enhancing clinical outcomes in individuals with TRD. In recent years, advances in machine learning (ML) have provided new opportunities for leveraging large-scale datasets, such as electronic medical records (EMR), to improve healthcare delivery and inform clinical decision-making ( 14 ). For instance, ML prediction models can be used as structured tools to more accurately stratify patients into subgroups or individualized treatment pathways ( 15 ). In psychiatry, particularly in TRD and rTMS, ML can reveal latent multifactorial relationships that conventional methods might overlook. This includes the interplay between depressive phenotypes, their causes, comorbidities, and rTMS parameters. Despite the rapid growth of ML in psychiatric research, the generalizability and clinical utility of many ML models published to date remain poor ( 16 ). Motivated by the potential of ML to enhance understanding of rTMS treatment for patients with TRD, this study performed a comprehensive retrospective analysis to investigate demographic, clinical, and treatment-related factors associated with treatment response and remission. This retrospective approach offers several strengths, including relatively large sample sizes, the ability to capture real-world clinical practices, and identify long-term treatment outcomes ( 17 ) without the resource demands of a clinical trial. Specifically, this study explores 1) demographic and clinical characteristics of TRD patients receiving rTMS treatment, 2) various protocols and parameters utilized in rTMS administration, and 3) pre-treatment and treatment-related factors predictive of treatment response or remission. This paper aims to address the following questions: whether ML can predict response or remission following rTMS based on pre-treatment patient features and treatment-related factors better than chance; which features distinguish individuals responding favorably to rTMS from those who do not; and whether the ML models may offer clinical benefit. In generating ML models to predict rTMS treatment response and identifying the features that drive both positive and negative outcomes, this study aims to advance understanding of TRD, optimize treatment strategies, and ultimately improve patient outcomes. MATERIALS AND METHODS Design. A retrospective study was conducted, including EMR (Epic, Verona, WI, USA) of TRD patients, collected from the clinical database of the UC San Diego Interventional Psychiatry Program (UCSD-IPP). All contributing procedures comply with the ethical standards of the relevant national and international committees on human subjects’ research and the Helsinki Declaration and were approved by the UCSD Human Research Protections Program. Patients. Included patients were prescribed rTMS treatment from UCSD psychiatrists credentialed in delivering rTMS. rTMS was initiated in outpatients who met criteria for TRD, did not present with any treatment contraindications, had adequate insurance coverage, and provided consent. As illustrated in Fig. 1 , EMR records for 419 patients seeking rTMS in 2017–2023 at UCSD were screened for the following inclusion criteria: (i) primary diagnosis of MDD; (ii) aged 18 years and older; (iii) completion of at least 15 sessions by the date of final PHQ-9; (iv) clinically meaningful depressive symptoms, as indicated by PHQ-9 total score > 4; (v) failure to respond to at least 2 antidepressant pharmacotherapy trials; and (vi) completion of clinical evaluations prior to and following the acute treatment phase. A total of 246 patients began a rTMS treatment course and 239 completed at least 15 sessions before their final PHQ-9 assessment. Two participants had baseline PHQ-9 scores < 5, indicating no clinically significant depressive symptoms, and were excluded from analyses. Five patients were excluded for not having received an MDD diagnosis, yielding a final analytical sample of 232 MDD patients. All patients had previously failed to respond to at least two antidepressant treatments and 90% reported having an MDD diagnosis > 10 years. Eighteen patients had baseline PHQ-9 scores < 10, indicative of minimal depressive symptoms, and response rates within this subgroup were lower than patients with PHQ-9 ≥ 10 at baseline (38.9% vs. 43.9%) but did not significantly differ (X 2 = 0.17, p = 0.68). Treatment procedure. All rTMS treatment was performed in the UCSD-IPP clinic by technicians trained in protocol administration with oversight by an attending psychiatrist. Overseeing psychiatrists reviewed the treatment plans after each session. rTMS was delivered using either a Magventure MagPro with a B70 coil or a BrainsWay H1-coil system. For all patients, the resting motor threshold (rMT) was identified by finding the motor hotspot and utilizing the lowest level of stimulation intensity to elicit movement in the opposite abductor pollicis brevis (APB) muscle on 50% of stimulation attempts. Once rMT was obtained, treatment session parameters were determined. For Magventure treatments, the coil position was localized using the Beam F3 method, and treatment was delivered per standard practice ( 18 ). Patients who were initiated on the iTBS protocol (at 120% rMT), involving 600 total pulses in bursts of 3 with 8-second inter-train-intervals, were transitioned to either BiTBS (90% rMT; 3600 total pulses, 1800 delivered via 3-iTBS and 1800 delivered cTBS 120% rMT protocol to the left and right hemispheres, respectively) or 3-iTBS (iTBS delivered three times consecutively, per Li et al 2014 ( 19 ) if there was deemed poor efficacy or issues with tolerability). Standard high-frequency left (HFL; 19 minutes) and low-frequency right (LFR; 10 minutes) protocols were used less commonly, both at 120% rMT. The obsessive-compulsive disorder (OCD) protocol uses a B70 coil and delivers 2000 pulses per session via 50, 2-second trains to dorsomedial prefrontal cortex (DMPFC), at 100% rMT of the foot with the Magventure B70 coil. For the BrainsWay H1-coil system either iTBS or the standard FDA-approved 18 Hz protocol was used at 120% rMT. The end of each patient’s acute treatment period was clinician-determined and coincided with reduced frequency of rTMS treatment to < 2 sessions per week. Sequencing of TMS protocols across all patients included in the analysis is depicted in Supplemental Fig. 1 . Clinical assessment. The PHQ-9 is a nine-item self-report scale used to assess the presence and severity of depressive symptoms based on the Diagnostic and Statistical Manual (DSM) criteria for a Major Depressive Episode (MDE). Questions are on a four-point scale related to the frequency of symptoms presented on each item, including “not at all,” “several days,” “more than half the days,” and “nearly every day.” Total scores are calculated by adding scores to all nine items (total range: 0–27), and previously established cutoff scores for severity are: 0–4 (no symptoms), 5–9 (mild symptoms), 10–14 (moderate symptoms), 15–19 (moderately severe symptoms), and 20–27 (severe symptoms). The PHQ-9 was administered pre-treatment and at rTMS visits throughout the acute treatment phase to monitor changes in depressive symptoms. Only pre-treatment and post-acute treatment phase scores were used in the present analyses. Predictive features. Pre-treatment features. A total of 32 EMR-derived features were extracted from patient records, including: age, gender, ethnicity, religious affiliation, educational attainment, employment status, relationship status, body mass index (BMI) category (obesity: \(\ge\) 30 kg/m 2 ; overweight: 25 \(\le\) kg/m 2 < 30; normal 18.5 < kg/m 2 < 25), alcohol, tobacco, and cannabis use (each operationalized as 3-level factor: current, former, or never), family history of mood disorders, 1st degree relatives with mood or psychiatric disorder diagnoses, prior ECT treatment, prior rTMS treatment, history of psychological trauma, non-suicidal self-injury (NSSI), history of suicidal ideation (SI) or suicide attempts, psychiatric hospitalizations, and duration of current MDE. Psychiatric comorbidities were also recorded, including anxiety disorders, trauma-related disorders, substance use disorders, OCD, somatic symptom and related disorders, personality disorders, autism spectrum disorders, eating disorders, and neurocognitive disorders. Comorbid diagnoses were extracted from physician notes. Psychiatric medication use, including benzodiazepines, non-benzodiazepine sedatives, anxiolytics, antipsychotics, and psychostimulants, was recorded, and antidepressant medication burden was translated into daily defined dose (DDD) equivalents. Treatment-related features. To determine the role of treatment-related factors in depression outcomes, the following five parameters were recorded: protocol type (deep TMS, iTBS, bilateral TBS, or others that included HFL, LFR, 3-iTBS, or OCD protocols); number of total rTMS sessions; cumulative number of pulses received throughout the treatment course; and, whether patients switched rTMS protocols during treatment. Statistical analysis. R statistical software (v.4.3.1) was used for data analysis. Independent sample t -tests and chi-square tests were used to compare demographic and clinical characteristics between rTMS non-responders and responders, defined as patients with ≥ 50% decrease in post-treatment PHQ-9 scores from pre-treatment scores. Pairwise post-hoc tests for factors with more than two levels were false discovery rate (FDR)-corrected. To mitigate low variance within categorical variables, variables were combined and/or operationalized as follows: rTMS protocols: 18 Hz deep TMS (dTMS), BiTBS, or iTBS, made up the majority (86.6%) of rTMS protocols in the sample; therefore, other protocols, including 3-iTBS, 3-iTBS/OCD, OCD, and HFL were combined into a single factor level (“Other”). Race/ethnicity: patients were coded as non-Hispanic, white, or “Other”. Psychiatric comorbidities: patients were coded with 1 if they had at least one comorbid diagnosis. Duration of current MDE: patients were coded as 12 months in duration. Patients were binarized indicating completion of a 4-year degree or not. Relationship status was coded as: currently single/unpartnered, partnered/married, or separated/divorced/widowed. Non- antidepressant psychiatric medication use was coded 1 or 0 for each drug category. To identify pre-treatment patient characteristics and treatment-associated features that best discriminate response and remission groups, a series of classification models were trained using the caret package: (i) regularized logistic regression (elastic net regression; ENet); (ii) support vector machine (SVM) with radial basis function; (iii) random forest (RF); and (iv) stochastic gradient boosting machines (GBM). All features were filtered for near-zero-variance (< 5%) and linear combinations. Nested cross-validation was used for hyperparameter tuning and model performance evaluation using nestedcv with 10 inner folds and 20 outer folds. RF parameters optimized in the nested grid search included mtry (number of variables to split), splitrule (quality of each node split), and min.node.size (minimum number of samples in a node), and GBM parameters included interaction.depth (highest interaction per tree), n.minobsinnode (minimum number of samples in a node to split), n.trees (total number of trees), and shrinkage (learning rate), whereas the default tuning grids in caret were applied for ENet and SVM. To mitigate class imbalance in the dataset, Synthetic Minority Over-sampling TEchnique (SMOTE) was applied to equalize the class ratio (1:1). Feature selection via a univariate filter was performed on each outer fold prior to hyperparameter tuning, and a Relief-based algorithm (ReliefFbestK in CORElearn package; k = 10) was applied in parallel for comparison with the univariate filter. Generally, performance was superior for models trained after univariate filtering ( Supplemental Fig. 3 ), with 10 features selected on average (range: 7–13) for all models. Optimization criteria for each model’s hyperparameter tuning procedure were receiver operating characteristic area under the curve (ROC AUC). Continuous features were standardized, and categorical features were one-hot encoded, with the first level dropped as reference. Missing variables were treated as missing completely at random and nonparametric random forest-based imputation was performed using missForest . Generalization performance and 95% confidence intervals were generated by averaging AUC across the best tune of all outer folds. Predicted probabilities across all outer folds were merged to generate additional performance metrics ( F -score, sensitivity, specificity) and models with the best performance for an outcome were selected for downstream analysis. Significance of AUC values were determined by permutation testing (B = 99) of class labels, re-running the procedure for each model, and calculating the proportion of permuted models with AUC values greater than the observed data. Alpha level was set at 0.05. The contributions of predictive features were determined using SHapley Additive exPlanations (SHAP) analysis in the fastshap R package, and directionality of associations between features and outcome variables was ascertained by visualizing individual SHAP values and their dependencies with the shapviz package in R. Using the best performing models, the predicted probabilities of treatment response and remission were derived. Model calibration was assessed with the Hosmer-Lemeshow (HL) goodness-of-fit test by binning predicted probabilities into deciles. Probability estimates were updated using beta regression, which has been shown to outperform other calibration methods ( 20 ), and recalibrated probabilities were reassessed using HL test. Decision curve analysis (DCA) was applied using dcurves to assess the clinical utility of prediction models through estimation of net benefit across thresholds of predicted risks using recalibrated probabilities from both “response” and “remission” classification models. Net benefit is equivalent to the percentage of individuals appropriately treated (“true positives”, responders and remitters) minus a weighted percentage of those inappropriately treated (“false positives”). RESULTS Patient and treatment-related characteristics Sociodemographic and clinical characteristics of patients, and treatment-related characteristics included in the analysis are outlined in Table 1 . Across all patients, post-treatment PHQ-9 scores were significantly reduced compared to pre-treatment values (mean difference = -7.56, t 231 = -17.6, d = 1.16, 95% CI = 0.99–1.32). In total, 101 patients (44%) responded to rTMS, (mean difference = -13.0, t 100 = -29.4, d = 2.94, 95% CI = 2.48–3.39; Fig. 2 ) and 53 patients (23%) remitted. The average patient age was 55.1±16.9 and 54.1±17.0 years in the responder and non-responder groups, respectively. Patient sex and ethnicity did not differ between groups; however, responders were more likely to have been former smokers than non-responders, and less likely to have never used alcohol. Most patients were naïve to rTMS (88.4%) and ECT (83.3%) treatment, and the majority (60%) had MDEs lasting longer than 12 months at treatment outset. Baseline PHQ-9 scores did not differ between responders (17.9±5.4) and non-responders (17.6±5.3). Responders were more likely to have a history of trauma and to be diagnosed with trauma disorder compared to non-responders, but were less likely to have an anxiety disorder or be prescribed benzodiazepines or antipsychotic medications. The ratio of patients receiving sequential iTBS-BiTBS to iTBS-only treatment was higher among non-responders than responders ( p adj = 0.04), and non-responders received fewer rTMS sessions on average compared to responders. Table 1 Sociodemographic and clinical characteristics of patient sample. All patients (n = 232) Responders (n = 101; 43.5%) Non-Responders (n = 131; 56.5%) Test Statistic Sociodemographic characteristics Age (yrs) 54.5 (16.9) 55.1 (16.9) 54.1 (17.0) t 198 = 1.13, p = 0.26 Sex (%F) 63.4% 62.4% 64.1% \({\chi }_{1}^{2}\) = 0.02, p = 0.89 Ethnicity (%NHW/B/H/A/P/MO) 76/1/6/4/1/12 75/1/4/5/2/13 76/2/7/3/0/12 \({\chi }_{6}^{2}\) = 5.43, p = 0.49 Religiously affiliated (%) 23.9% 18.2% 28.3% \({\chi }_{1}^{2}\) = 3.16, p = 0.08 Relationship status (S/P/SW) 35 / 48 / 17 39 / 46 / 15 31 / 50 / 19 \({\chi }_{1}^{2}\) = 1.59, p = 0.45 College graduate (%) 81.6% 79.8% 82.9% \({\chi }_{1}^{2}\) = 0.37, p = 0.54 BMI category (%Nm / Ow / Ob) 40 / 30 / 30 48 / 29 / 23 34 / 31 / 35 \({\chi }_{2}^{2}\) = 5.79, p = 0.06 1st -degree relative w/ psych dx (%) 77.4% 81.9% 74.0% \({\chi }_{1}^{2}\) = 1.92, p = 0.17 Tobacco use (% N/C/F) 75 / 7 / 18 63 / 9 / 28 85 / 5 / 10 \({\varvec{\chi }}_{2}^{2}\) = 14.7, p < 0.01 Cannabis use (% N/C/F) 67 / 22 / 11 63 / 25 / 12 69 / 20 / 11 \({\chi }_{1}^{2}\) = 1.01, p = 0.60 Alcohol use (% N/C/F) 37 / 46 / 17 26 / 52 / 22 45 / 41 / 14 \({\varvec{\chi }}_{1}^{2}\) = 10.1, p < 0.01 Clinical characteristics Psychiatric comorbidities (n) 0.23 (0.4) 0.23 (0.4) 0.24 (0.4) \({t}_{216}\) =0.16, p = 0.87 Antidepressant drug use (DDD) 1.75 (1.7) 1.65 (1.5) 1.83 (1.8) \({t}_{229}\) =0.79, p = 0.43 Prior rTMS treatment 11.6% 15.8% 8.4% \({\chi }_{1}^{2}\) = 3.07, p = 0.08 Baseline PHQ-9 score 17.8 (5.3) 17.9 (5.4) 17.6 (5.3) \({t}_{213}\) =0.42, p = 0.67 MDE duration (< 6/6–12/12 + mo) 21%/18%/60% 28%/19%/53% 16%/18%/66% \({\chi }_{2}^{2}\) = 4.76, p = 0.09 Prior ECT 17.7% 13.9% 20.6% \({\chi }_{1}^{2}\) = 1.79, p = 0.18 History of trauma 45.3% 54.5% 38.2% \({\varvec{\chi }}_{1}^{2}\) = 6.11, p = 0.01 Prior psych hospitalization 35.8% 41.6% 31.3% \({\chi }_{1}^{2}\) = 2.63, p = 0.11 Suicidal ideation history 65.1% 66.3% 64.1% \({\chi }_{1}^{2}\) = 0.12, p = 0.73 Non-suicidal self-injury history 11.2% 9.9% 12.2% \({\chi }_{1}^{2}\) = 0.31, p = 0.58 Anxiety disorder 72.8% 65.3% 78.6% \({\varvec{\chi }}_{1}^{2}\) = 5.08, p = 0.02 Prior suicide attempt 25.9% 29.7% 22.9% \({\chi }_{1}^{2}\) = 1.38, p = 0.24 Substance use disorder 14.2% 14.9% 13.7% \({\chi }_{1}^{2}\) = 0.06, p = 0.81 ADHD diagnosis 15.5% 19.8% 12.2% \({\chi }_{1}^{2}\) = 2.51, p = 0.11 Trauma disorder diagnosis 22.4% 31.7% 15.3% \({\varvec{\chi }}_{1}^{2}\) = 8.84, p < 0.01 Benzodiazepine use 29.0% 21.8% 34.4% \({\varvec{\chi }}_{1}^{2}\) = 4.20, p = 0.04 Antipsychotic medication use 24.2% 17.0% 29.8% \({\varvec{\chi }}_{1}^{2}\) = 5.04, p = 0.02 Psychostimulant medication use 19.5% 20.0% 19.1% \({\chi }_{1}^{2}\) = 0.03, p = 0.86 Non-benzodiazepine sedative use 25.5% 27.0% 24.4% \({\chi }_{1}^{2}\) = 0.20, p = 0.66 Mood stabilizer use 18.2% 14.0% 21.4% \({\chi }_{1}^{2}\) = 2.07, p = 0.15 Treatment-related characteristics TMS protocol (% BiTBS/dTMS /iTBS/iTBS-BiTBS/Other) 16/36/10/25/13 13/41/16/18/13 18/33/5/31/14 \({\varvec{\chi }}_{4}^{2}\) = 11.8, p = 0.02 Total number of rTMS sessions 37.4 (10.6) 39.1 (10.8) 36.2 (10.2) \({\varvec{t}}_{209}\) =2.08, p = 0.04 Cumulative rTMS pulses 88,804(44,862) 88,832(48,831) 88,782(41,738) \({t}_{196}\) =0.01, p = 0.99 rTMS protocol switch (%) 58.2% 53.5% 61.8% \({\chi }_{1}^{2}\) = 1.64, p = 0.20 Weeks from final patient in DB 136 (81) 146 (85) 128 (78) \({t}_{204}\) =1.62, p = 0.11 Resting motor threshold (%) 46.0% 46.3% 45.7% \({t}_{200}\) =0.44, p = 0.66 Length of treatment (days) 66.3 (19.9) 69.0 (19.5) 64.2 (19.9) \({t}_{217}\) =1.85, p = 0.07 Other rTMS protocols include 3-iTBS, 3-iTBS/OCD, HFL, and OCD. Ethnicities reported include Non-Hispanic White (NHW), Black (B), Hispanic (H), Asian (A), Persian (P), and Mixed or Other (MO). Substance use reported as never (N), current (C), or former (F). Relationship status reported as Single (S), Partnered/Married (P), or Widowed/Divorced (W). Psychiatric comorbidities included OCD, somatic symptom disorders, personality disorders, autism spectrum disorders, eating disorders, and neurocognitive disorders. BMI categories corresponded to kg/m 2 \(\ge\) 30 (Ob), 25 \(\le\) Ow < 30, and 18.5 < Nm < 25. Classification results Treatment Response. Across the four models tested, the SVM model predicted treatment response significantly above chance ( p < 0.01; Supplemental Fig. 2 ) with modest discrimination performance (mean AUC = 0.689, 95% CI = 0.638–0.740), outperforming RF (0.654, 0.594–0.715), ENet (0.668, 0.616–0.719), and GBM (0.617, 0.563–0.671) models ( Supplemental Fig. 3 ). Thus, the SVM model was evaluated further. The sensitivity and specificity of the model to discriminate treatment responders from non-responders were 72.3% and 53.4%, respectively, with an F-score of 0.64. Following recalibration, observed and predicted probabilities of treatment response were similar (HL test: \({\chi }_{8}^{2}\) =5.51, p =0.70, RMSE=1.73; Supplemental Fig. 4 ). In descending order, the most important variables in prediction of treatment response were anxiety disorder, former tobacco use, obesity, trauma disorder, antipsychotic use, benzodiazepine use, iTBS-BiTBS protocol, > 12-month duration of current MDE, history of trauma, number of rTMS sessions, and iTBS protocol (Fig. 3 A). Partial dependence plots illustrating the relationship between the top-3 features, the total number of rTMS sessions, and treatment response are shown in Fig. 3 B. DCA indicated that for threshold probabilities between 25% and 55%, corresponding to a number needed-to-treat (NNT) between 4 and 2 respectively, the net benefit of applying the model to aid in determining whether a patient should undergo rTMS is higher than either a “treat all” or “treat none” approach (Fig. 5 A). Depression remission. Across the four models tested, the GBM model predicted remission following rTMS treatment significantly above chance ( p < 0.01; Supplemental Fig. 2 ) with good discrimination performance (mean AUC = 0.745, 95% CI = 0.692–0.797), outperforming RF (0.735, 0.680–0.791), ENet (0.740, 0.668–0.813), and SVM (0.683, 0.627–0.738) models ( Supplemental Fig. 3 ). Thus, the GBM model was evaluated further. The sensitivity and specificity of the model to discriminate remitters from non-remitters were 58.5% and 77.7%, respectively, with an F-score of 0.50. Following recalibration, the observed and predicted probabilities of treatment response remained dissimilar (HL test: \({\chi }_{8}^{2}\) =19.1, p =0.01, RMSE=2.73; Supplemental Fig. 4 ). In descending order, the most important variables in the prediction of depression remission were pre-treatment PHQ-9 score, obesity, antipsychotic use, current MDE greater than 12 months, benzodiazepine use, anxiety, prior ECT, and iTBS-BiTBS protocol (Fig. 4 A). Partial dependence plots illustrating the relationship between the top-four features and treatment response are shown in Fig. 4 B. DCA indicated that for threshold probabilities between 15% and 40%, corresponding to a NNT between approximately 7 and 3, respectively, using the model to determine whether a patient should undergo rTMS yielded greater net benefit than employing either a “treat all” or “treat none” strategy (Fig. 5 ). DISCUSSION Findings presented in this study underscore the substantial variability in depression responses and remission, despite rTMS efficacy, and the promise of ML to predict treatment outcomes using EMR data alone. Importantly, both ML models for response and remission identified psychiatric comorbidities, depressive episode duration, and specific rTMS protocols as predictive features, with trauma history also associated with response outcomes. Predictive features of treatment outcomes : Past research investigating the relationship between psychiatric comorbidities and treatment response has yielded valuable insights into the complex interplay between mental health conditions and their respective treatments ( 21 ). This literature presents mixed results regarding impact of trauma history on rTMS. Experiences of childhood trauma and PTSD diagnosis have been reported to negatively affect rTMS outcomes in MDD patients ( 22 , 23 ) or to have no significant impact ( 24 ). Positive findings, as presented here, may point to a consequence of trauma-induced neuroplastic changes that sensitize the brain to neurostimulation ( 25 ). For example, the therapeutic impact of rTMS could be modulated by individual differences in neural circuitry affected by trauma, which are not always uniformly engaged by standard rTMS protocols, such as 10 Hz rTMS to the left DLPFC as done in Hu et al. (2012) and Yesavage et al. (2018). Former tobacco use was identified as a positive predictor of treatment response. Past literature on tobacco use in rTMS outcomes is mixed ( 26 , 27 ), but nicotine’s complex neurobiological effects may enhance efficacy. Nicotine potentiates dopaminergic neurotransmission via nicotinic acetylcholine receptor binding on dopaminergic neurons in the mesolimbic pathway associated with mood regulation ( 28 ). This dopamine surge may have a persistent effect that synergizes with rTMS mechanisms, inducing changes in brain plasticity ( 29 ). Coupled with trauma history, these associations suggest past experiences and lifestyle modifications may prime the brain's receptivity to rTMS. Features including comorbid anxiety disorder, concomitant benzodiazepine or antipsychotic medications, and a longer duration of current MDE were linked to lower probabilities of treatment response and remission. Anxiety is highly comorbid with depression, and this specific comorbidity may reflect distinct neurobiological underpinnings, affecting the response to rTMS. For instance, hyperactivity in the amygdala and its connectivity with the PFC, can be more pronounced in individuals with comorbid anxiety ( 30 ). An extensive body of research supports that comorbid anxiety can complicate the course of depression, often leading to a more chronic and treatment-resistant condition ( 31 ). Concurrent use of benzodiazepines or antipsychotics was associated with less favorable responses. These medications exert sedative effects and reduce cortical excitability, potentially impeding operating mechanisms of rTMS. Benzodiazepines, for instance, enhance GABAergic inhibition, which may diminish excitatory processes ( 32 ). Antipsychotics, particularly those blocking dopamine receptors, have been shown to attenuate rTMS outcomes ( 33 ), potentially disrupting the dopaminergic modulation crucial for rTMS-induced neuroplasticity. These findings suggest patients prescribed benzodiazepines or antipsychotics may require adjusted protocols, perhaps with increased intensity or frequency to overcome medication-induced reductions in cortical excitability, or timing treatment sessions relative to medication dosing. The observation that longer depressive episodes correlated with poorer outcomes is consistent research indicating depression chronicity can negatively impact the efficacy of various treatments. Chronic depression is associated with more severe neurobiological changes, including alterations in the hippocampus and PFC ( 34 ). Earlier intervention might enhance response by addressing such changes before they become entrenched. Fava (2003) highlighted the predictive role of untreated depression duration in treatment resistance ( 35 ) and underscoring the possibility that adjusting standard protocols, or combining with other interventions, may increase efficacy for these patients. The observed absence of sociodemographic variables, such as age, sex, and ethnicity, impacting outcomes aligns with past studies ( 36 ), yet diverges from others. For example, Hanlon and colleagues (2022) found younger age and female sex were associated with better rTMS response rates, suggesting demographic characteristics may influence treatment success ( 37 ). The contrasting evidence indicates a complex relationship between sociodemographic and rTMS efficacy, warranting further investigation to clarify their roles in treatment personalization and optimization. Accrued effects from patient history, such as trauma and anxiety, and concurrent benzodiazepine use may influence rTMS outcomes. The potential cumulative effect of these factors on response could be linked to their disparate impact on brain neuroplasticity and neurotransmission. For instance, a history of trauma and relatively lower anxiety might enhance receptivity to rTMS via induced neuroplastic changes, whereas the suppressive effect of benzodiazepines on cortical excitability could counteract these benefits, requiring adjustments in treatment protocols to optimize efficacy. Further investigation into the interaction of these factors may provide deeper insights into personalized rTMS strategies for patients with complex clinical profiles. Regarding treatment-related factors, the number of rTMS sessions was a positive predictor, with SHAP values indicating a positive linear dose-response relationship. This relationship aligns with notions that additional sessions sustain antidepressant effects, possibly through cumulative neuroplastic changes. However, this is not consistently upheld, as a meta-analysis reported higher doses of rTMS, measured by total pulses, were not always associated with depression improvement ( 38 ). Adherence to the iTBS protocol was also among the strongest predictors of response. Notably, iTBS demonstrated superiority over iTBS-BiTBS in our sample, challenging some of the existing literature that suggests bilateral approaches can be advantageous for certain patient populations ( 39 ). Despite this, the current findings likely relate to iTBS-BiTBS patients experiencing greater treatment-resistance at baseline, with providers transitioning protocols from unilateral to bilateral in cases of inadequate symptom relief. Machine learning, electronic medical records, and clinical decision-making Recent reviews of ML prediction models in psychiatric research have found that despite relatively ‘good’ discrimination performance (AUC ~ 0.65–0.80), most studies had methodological weaknesses that limited their generalizability and utility in clinical decision-making ( 16 , 40 ). The authors recommended more comprehensive reporting of ML methods and results, more robust validation methods, model calibration, and formal assessment of clinical utility, such as DCA ( 40 , 41 ). In our analyses, DCA demonstrated that the response and remission models outperformed "treat all" and “treat none” strategies within probability threshold ranges of 25–55% and 15–40%, or NNT between 2 and 4 patients, and 3 and 7 patients, respectively. Both models indicate net clinical benefit within their respective threshold ranges, though more refined models incorporating additional patient features, such as neurobiological markers, polygenic risk scores, and psychosocial determinants, may confer additional benefit. The implications of these findings are multi-fold. Firstly, they re-affirm the clinical utility of rTMS in managing TRD, especially when considering individual patient histories and comorbidities. Secondly, these results lend support for the predictive power of ML algorithms derived from EMR for refining treatment approaches, suggesting that these tools could become integral to psychiatric precision medicine. Future research should aim to validate these predictive models in diverse clinical settings, ensuring that findings are replicable and robust. Limitations. The retrospective study design provides a robust dataset but holds inherent limitations. Firstly, these ML models were not externally validated in an independent sample. However, nested cross-validation, a robust internal validation procedure, provides a strong estimation of model generalizability. Secondly, this study is susceptible to selection bias since it involves patients who initiated and completed treatment, which may exclude those with different response trajectories, possibly skewing findings towards patients more likely to complete treatment. Moreover, the sample in this study was predominantly highly educated and non-Hispanic White, limiting generalization to other demographics and pointing more towards the need to address disparities in access to TMS ( 42 ). Third, information bias may arise from EMR data availability, accuracy, and completeness, with additional variability in the timing of data collection vis-à-vis treatment, possibly creating discrepancies between recorded and actual patient conditions. Future research should address these limitations by implementing prospective designs, standardizing rTMS protocols, timely data collection, and diverse populations. CONCLUSION This study elucidates predictors of treatment response to rTMS in TRD patients and highlights the importance of treatment approaches tailored to individual characteristics. By leveraging EMR and ML techniques, this study identified demographic, clinical, and treatment-related factors associated with outcomes, providing valuable insights for clinical practice and research. Declarations Conflict of interest: All authors declare that they have no conflict of interest related to the research presented in this manuscript. Disclosures: The authors have no financial ties to disclose. Funding: This research was funded by the University of California San Diego Health Sciences Research Award #RG114131 to author LGA, R25 MH101072 to author KT, and generous support from the Kreutzkamp Family Foundation. References American Psychiatric Association, American Psychiatric Association, editors. Diagnostic and statistical manual of mental disorders: DSM-5. 5th ed. Washington, D.C: American Psychiatric Association; 2013. 947 p. (WHO) WHO. Mental disorders [Internet]. [cited 2023 Jun 5]. Available from: https://www.who.int/news-room/fact-sheets/detail/mental-disorders Zhdanava M, Pilon D, Ghelerter I, Chow W, Joshi K, Lefebvre P, et al. The Prevalence and National Burden of Treatment-Resistant Depression and Major Depressive Disorder in the United States. J Clin Psychiatry. 2021;82(2):0–0. Zimmerman M, Ellison W, Young D, Chelminski I, Dalrymple K. How many different ways do patients meet the diagnostic criteria for major depressive disorder? Compr Psychiatry. 2015;56:29–34. Lynch CJ, Gunning FM, Liston C. Causes and Consequences of Diagnostic Heterogeneity in Depression: Paths to Discovering Novel Biological Depression Subtypes. Biol Psychiatry. 2020;88(1):83–94. Benster LL, Weissman CR, Stolz LA, Daskalakis ZJ, Appelbaum LG. Pre-clinical indications of brain stimulation treatments for non-affective psychiatric disorders, a status update. Transl Psychiatry. 2023;13(1):1–12. Aaronson S. Overview of a large transcranial magnetic stimulation patient outcomes database. Brain Stimul Basic Transl Clin Res Neuromodulation. 2021;14(6):1717. Fitzgerald PB, Oxley TJ, Laird AR, Kulkarni J, Egan GF, Daskalakis ZJ. An analysis of functional neuroimaging studies of dorsolateral prefrontal cortical activity in depression. Psychiatry Res Neuroimaging. 2006;148(1):33–45. Mayberg HS. Limbic-cortical dysregulation: a proposed model of depression. J Neuropsychiatry Clin Neurosci. 1997;9(3):471–81. George MS, Lisanby SH, Avery D, McDonald WM, Durkalski V, Pavlicova M, et al. Daily left prefrontal transcranial magnetic stimulation therapy for major depressive disorder: a sham-controlled randomized trial. Arch Gen Psychiatry. 2010;67(5):507–16. Cole EJ, Stimpson KH, Bentzley BS, Gulser M, Cherian K, Tischler C, et al. Stanford Accelerated Intelligent Neuromodulation Therapy for Treatment-Resistant Depression. Am J Psychiatry. 2020;177(8):716–26. Chung SW, Hoy KE, Fitzgerald PB. Theta-burst stimulation: a new form of TMS treatment for depression? Depress Anxiety. 2015;32(3):182–92. Peng Z, ZHOU C, XUE S, BAI J, YU S, LI X, et al. Mechanism of Repetitive Transcranial Magnetic Stimulation for Depression. Shanghai Arch Psychiatry. 2018;30(2):84–92. Beam AL, Kohane IS. Big Data and Machine Learning in Health Care. JAMA. 2018;319(13):1317–8. Rajkomar A, Dean J, Kohane I. Machine Learning in Medicine. N Engl J Med. 2019;380(14):1347–58. Meehan AJ, Lewis SJ, Fazel S, Fusar-Poli P, Steyerberg EW, Stahl D, et al. Clinical prediction models in psychiatry: a systematic review of two decades of progress and challenges. Mol Psychiatry. 2022;27(6):2700–8. Rassen JA, Bartels DB, Schneeweiss S, Patrick AR, Murk W. Measuring prevalence and incidence of chronic conditions in claims and electronic health record databases. Clin Epidemiol. 2018;11:1–15. Beam W, Borckardt JJ, Reeves ST, George MS. An efficient and accurate new method for locating the F3 position for prefrontal TMS applications. Brain Stimulat. 2009;2(1):50–4. Li CT, Chen MH, Juan CH, Huang HH, Chen LF, Hsieh JC, et al. Efficacy of prefrontal theta-burst stimulation in refractory depression: a randomized sham-controlled study. Brain J Neurol. 2014;137(Pt 7):2088–98. Ojeda FM, Jansen ML, Thiéry A, Blankenberg S, Weimar C, Schmid M, et al. Calibrating machine learning approaches for probability estimation: A comprehensive comparison. Stat Med. 2023;42(29):5451–78. Krishnan KRR. Comorbidity and depression treatment. Biol Psychiatry. 2003;53(8):701–6. Yesavage JA, Fairchild JK, Mi Z, Biswas K, Davis-Karim A, Phibbs CS, et al. Effect of Repetitive Transcranial Magnetic Stimulation on Treatment-Resistant Major Depression in US Veterans. JAMA Psychiatry. 2018;75(9):884–93. Hu YT, Hu XW, Han JF, Zhang JF, Wang YY, Wolff A, et al. Childhood trauma mediates repetitive transcranial magnetic stimulation efficacy in major depressive disorder. Eur Arch Psychiatry Clin Neurosci. 2021;271(7):1255–63. Ng E, Wong EHY, Lipsman N, Nestor SM, Giacobbe P. Adverse childhood experiences and repetitive transcranial magnetic stimulation outcomes for depression. J Affect Disord. 2023;320:716–24. Karsen EF, Watts BV, Holtzheimer PE. Review of the Effectiveness of Transcranial Magnetic Stimulation for Post-traumatic Stress Disorder. Brain Stimul Basic Transl Clin Res Neuromodulation. 2014;7(2):151–7. Abdelrahman AA, Noaman M, Fawzy M, Moheb A, Karim AA, Khedr EM. A double-blind randomized clinical trial of high frequency rTMS over the DLPFC on nicotine dependence, anxiety and depression. Sci Rep. 2021;11(1):1640. Poulet E, Galvao F, Haffen E, Szekely D, Brault C, Haesebaert F, et al. Effects of smoking status and MADRS retardation factor on response to low frequency repetitive transcranial magnetic stimulation for depression. Eur Psychiatry. 2016;38:40–4. Benowitz NL. Pharmacology of Nicotine: Addiction, Smoking-Induced Disease, and Therapeutics. Annu Rev Pharmacol Toxicol. 2009;49:57–71. Belujon P, Grace AA. Dopamine System Dysregulation in Major Depressive Disorders. Int J Neuropsychopharmacol. 2017;20(12):1036–46. Shin LM, Liberzon I. The neurocircuitry of fear, stress, and anxiety disorders. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2010;35(1):169–91. Comorbid anxiety and depression in adults: Epidemiology, clinical manifestations, and diagnosis - UpToDate [Internet]. [cited 2024 Mar 3]. Available from: https://www.uptodate.com/contents/comorbid-anxiety-and-depression-in-adults-epidemiology-clinical-manifestations-and-diagnosis Wilson S, Olsen S, Sullivan C, Cooper D, Somez I, Widge AS, et al. Concurrent Benzodiazepine Use and TMS Clinical Outcomes. Biol Psychiatry. 2021;89(9):S288. Hebel T, Abdelnaim M, Deppe M, Langguth B, Schecklmann M. Attenuation of antidepressive effects of transcranial magnetic stimulation in patients whose medication includes drugs for psychosis. J Psychopharmacol Oxf Engl. 2020;34(10):1119–24. Sheline YI, Gado MH, Kraemer HC. Untreated depression and hippocampal volume loss. Am J Psychiatry. 2003;160(8):1516–8. Fava M. Diagnosis and definition of treatment-resistant depression. Biol Psychiatry. 2003;53(8):649–59. Leuchter M, Citrenbaum C, Wilson A, Tibbe T, Jackson N, Krantz D, et al. Holding Steady: Age-Related Effects on Treatment Response During rTMS Treatment of Depression. Am J Geriatr Psychiatry. 2023;31(3, Supplement):S119–20. Hanlon CA, McCalley DM. Sex/Gender as a Factor That Influences Transcranial Magnetic Stimulation Treatment Outcome: Three Potential Biological Explanations. Front Psychiatry. 2022;13:869070. Yu CL, Kao YC, Thompson T, Brunoni AR, Hsu CW, Carvalho AF, et al. The association of total pulses with the efficacy of repetitive transcranial magnetic stimulation for treatment-resistant major depression: A dose-response meta-analysis. Asian J Psychiatry. 2024;92:103891. Blumberger DM, Vila-Rodriguez F, Thorpe KE, Feffer K, Noda Y, Giacobbe P, et al. Effectiveness of theta burst versus high-frequency repetitive transcranial magnetic stimulation in patients with depression (THREE-D): a randomised non-inferiority trial. The Lancet. 2018;391(10131):1683–92. Grzenda A, Widge AS. Electronic health records and stratified psychiatry: bridge to precision treatment? Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2024;49(1):285–90. Vickers AJ, Calster BV, Steyerberg EW. Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests. BMJ. 2016;352:i6. Nagra DS, Stolz LA, Weissman CR, Appelbaum LG. Access to interventional psychiatric treatments in the United States: Disparities and proposed solutions. Glob Health Econ Sustain. 2024;2(1):2456. Additional Declarations The authors have declared there is NO conflict of interest to disclose Supplementary Files SupplementaryMaterials.docx Cite Share Download PDF Status: Published Journal Publication published 27 Apr, 2025 Read the published version in Translational Psychiatry → Version 1 posted Editorial decision: revise 18 Sep, 2024 Review # 4 received at journal 22 Aug, 2024 Reviewer # 4 agreed at journal 08 Aug, 2024 Review # 3 received at journal 09 Jun, 2024 Reviewer # 3 agreed at journal 03 Jun, 2024 Reviewer # 2 agreed at journal 26 May, 2024 Reviewer # 1 agreed at journal 13 May, 2024 Reviewers invited by journal 12 May, 2024 Submission checks completed at journal 10 May, 2024 Editor assigned by journal 09 May, 2024 First submitted to journal 09 May, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4396926","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":301555539,"identity":"80a3feed-d4a6-4dc5-bfd6-8e72d589b38a","order_by":0,"name":"Lindsay Benster","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyElEQVRIiWNgGAWjYBACPnYeKIu9AUiwAfEBAlrYmMFaDBgYeA6QrEUigWgtvAcfF/z5I2c+843hxx9lDHJ8NxIIaeFLNp7ZZmAsczvHWJrnHIOxJGEtPGbSvA0GiTOkc8yYGdsYEjcQocX8N88fg/oZkmfMGH+2MdQTo8WMmYfNIEFCgseMgbeNIcGAGL9I87YZG87gSSsG+kXCcOaZB/i18LP3HvzM80dOXoL98EZgiNnI8x0nYAs6kCBN+SgYBaNgFIwC7AAAjTo2unt9dxEAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0003-0510-6448","institution":"SDSU/UC San Diego","correspondingAuthor":true,"prefix":"","firstName":"Lindsay","middleName":"","lastName":"Benster","suffix":""},{"id":301555540,"identity":"b38b72e7-9768-4646-8727-c7d0fa24f927","order_by":1,"name":"Cory Weissman","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Cory","middleName":"","lastName":"Weissman","suffix":""},{"id":301555541,"identity":"ce75dd9c-5a98-4e8d-970d-7ce8eb2e31e8","order_by":2,"name":"Federico Suprani","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Federico","middleName":"","lastName":"Suprani","suffix":""},{"id":301555542,"identity":"3c094079-6a6b-4531-b952-fda3a65ac53d","order_by":3,"name":"Kamryn Toney","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Kamryn","middleName":"","lastName":"Toney","suffix":""},{"id":301555543,"identity":"a578b838-63fc-418d-899e-0c625fc08c31","order_by":4,"name":"Houtan Afshar","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Houtan","middleName":"","lastName":"Afshar","suffix":""},{"id":301555544,"identity":"fab06f22-746d-46ba-82a8-c6ae626c6bd5","order_by":5,"name":"Noah Stapper","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Noah","middleName":"","lastName":"Stapper","suffix":""},{"id":301555545,"identity":"5eb5c642-b9ab-4e3f-a874-5f028bbc8c9c","order_by":6,"name":"Vanessa Tello","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Vanessa","middleName":"","lastName":"Tello","suffix":""},{"id":301555546,"identity":"80899127-b0b5-43bf-8bab-613ddf5c2cdc","order_by":7,"name":"Louise Stolz","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Louise","middleName":"","lastName":"Stolz","suffix":""},{"id":301555547,"identity":"70686a8f-cd1b-4f4f-92e2-e0e2097e3bb1","order_by":8,"name":"Mohsen Poorganji","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Mohsen","middleName":"","lastName":"Poorganji","suffix":""},{"id":301555548,"identity":"4de253c1-638e-4770-b472-8fd4c296fa71","order_by":9,"name":"Zafiris Daskalakis","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Zafiris","middleName":"","lastName":"Daskalakis","suffix":""},{"id":301555549,"identity":"ed12e116-cd5a-46be-930e-e98926c5cd7c","order_by":10,"name":"Lawrence Appelbaum","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Lawrence","middleName":"","lastName":"Appelbaum","suffix":""},{"id":301555550,"identity":"f0cb7aca-70f7-4bd6-ae79-eb0ee3d5b318","order_by":11,"name":"Jordan Kohn","email":"","orcid":"","institution":"University of California San Diego","correspondingAuthor":false,"prefix":"","firstName":"Jordan","middleName":"","lastName":"Kohn","suffix":""}],"badges":[],"createdAt":"2024-05-09 19:30:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4396926/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4396926/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41398-025-03380-w","type":"published","date":"2025-04-27T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":57037421,"identity":"804afff0-c098-4af1-bcfe-16fe5a125dc4","added_by":"auto","created_at":"2024-05-23 18:49:14","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":64689,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient inclusion\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/42f08a29d6b6bbff187e3622.png"},{"id":57036922,"identity":"46f6c871-390c-43c9-b47e-2054f70ab5f6","added_by":"auto","created_at":"2024-05-23 18:41:14","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":65614,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eDepression scores pre- and post-rTMS treatment course. Each point represents an individual patient’s PHQ-9 score at baseline and the conclusion of their acute treatment phase, grouped by response status. Horizontal dashed line at PHQ-9 = 5 indicates the threshold for depression remission.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/fd6b337d4e8ec15905da2091.png"},{"id":57036924,"identity":"cf69d9f0-c35c-46e1-afb1-f40995c205d2","added_by":"auto","created_at":"2024-05-23 18:41:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":154710,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eA) Ranked feature importance for treatment response classification model. Grey bars indicate negative association with treatment response (worse outcome), and black bars indicate positive association (better outcome). B) Partial dependence plots illustrating directional associations between selected features and treatment response. Larger positive SHAP values indicate greater probabilities of treatment response, whereas more negative values indicate lower probabilities of response.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/794cf045a4b47e7e7bbcd46f.png"},{"id":57036926,"identity":"aa046554-353c-4924-b5bb-3b4f063dd424","added_by":"auto","created_at":"2024-05-23 18:41:14","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":131668,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eA) Ranked feature importance for depression remission classification model. Grey bars indicate negative association with remission (worse outcome). B) Partial dependence plots illustrating directional associations between selected features and depression remission, indicated by SHAP values.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/3878d3b82c53ea531ace6cb9.png"},{"id":57036927,"identity":"6b9990d3-b572-42cb-a632-e95ab75d1b74","added_by":"auto","created_at":"2024-05-23 18:41:14","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":119388,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eDecision curve analysis for recalibrated A) treatment response and B) depression remission models. Threshold probability represents the point at which a positive response is equally valued to avoiding unnecessary treatment. Decision curves for each model are shown in blue, compared to a “treat all” approach shown in red and a “treat none” approach shown in green.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/de2be4358e73359bc81597bd.png"},{"id":81515729,"identity":"9574706e-7f2f-4794-ab60-c39cc1c886cc","added_by":"auto","created_at":"2025-04-28 07:07:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1840988,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/5f948a52-11c2-4b46-a325-e2ecae7b210c.pdf"},{"id":57036925,"identity":"b3b1fd13-7b24-4ca6-9d3d-3bd3257d5a94","added_by":"auto","created_at":"2024-05-23 18:41:14","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":332433,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-4396926/v1/ffb9bcdb456df71bc9df8cbb.docx"}],"financialInterests":"The authors have declared there is \u003cb\u003eNO\u003c/b\u003e conflict of interest to disclose","formattedTitle":"Predictive modeling of response to repetitive transcranial magnetic stimulation in treatment-resistant depression","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eMajor depressive disorder (MDD) is a prevalent and debilitating condition characterized by persistent feelings of sadness, hopelessness, and loss of interest or pleasure in daily activities (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). An estimated 322\u0026nbsp;million individuals worldwide meet criteria for MDD, representing a substantial public health challenge impacting quality-of-life, productivity, and overall well-being (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). While various pharmacological, psychotherapeutic, and neuromodulatory treatments are available, 30\u0026ndash;40% of individuals with MDD experience treatment-resistant depression (TRD), defined as non-response to two or more trials of first-line interventions (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Because MDD is a highly heterogeneous disorder, with 277 possible symptom combinations meeting DSM-5 criteria (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e), it is possible that different subtypes or symptom clusters of MDD may respond differentially to various treatments (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). Therefore, aligning optimal treatment strategies to symptom presentations may be valuable for addressing TRD.\u003c/p\u003e \u003cp\u003eRepetitive transcranial magnetic stimulation (rTMS) is an established clinical intervention for the treatment of TRD, with response rates of approximately 40\u0026ndash;60%, providing a valuable treatment option for patients not responding to first line psychotherapy and pharmacotherapy (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). rTMS involves the application of repetitive magnetic pulses to the cortex, with the dorsolateral prefrontal cortex (DLPFC) a common target for MDD due to its role in mood regulation, cognitive control, executive functioning, and emotion processing (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). The DLPFC is also implicated in the pathophysiology of depression due to its connections with the limbic system, while acute elevations in mood following rTMS provide further evidence of DLPFC dysfunction in MDD (\u003cspan additionalcitationids=\"CR10 CR11\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Ultimately, rTMS is thought to exert its therapeutic effects through modulation of cortical excitability, neurotransmitter systems, and neuroplastic mechanisms (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e), but understanding the predictors of treatment response is crucial for tailoring treatment strategies and enhancing clinical outcomes in individuals with TRD.\u003c/p\u003e \u003cp\u003eIn recent years, advances in machine learning (ML) have provided new opportunities for leveraging large-scale datasets, such as electronic medical records (EMR), to improve healthcare delivery and inform clinical decision-making (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). For instance, ML prediction models can be used as structured tools to more accurately stratify patients into subgroups or individualized treatment pathways (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). In psychiatry, particularly in TRD and rTMS, ML can reveal latent multifactorial relationships that conventional methods might overlook. This includes the interplay between depressive phenotypes, their causes, comorbidities, and rTMS parameters. Despite the rapid growth of ML in psychiatric research, the generalizability and clinical utility of many ML models published to date remain poor (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Motivated by the potential of ML to enhance understanding of rTMS treatment for patients with TRD, this study performed a comprehensive retrospective analysis to investigate demographic, clinical, and treatment-related factors associated with treatment response and remission.\u003c/p\u003e \u003cp\u003eThis retrospective approach offers several strengths, including relatively large sample sizes, the ability to capture real-world clinical practices, and identify long-term treatment outcomes (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e) without the resource demands of a clinical trial. Specifically, this study explores 1) demographic and clinical characteristics of TRD patients receiving rTMS treatment, 2) various protocols and parameters utilized in rTMS administration, and 3) pre-treatment and treatment-related factors predictive of treatment response or remission. This paper aims to address the following questions: whether ML can predict response or remission following rTMS based on pre-treatment patient features and treatment-related factors better than chance; which features distinguish individuals responding favorably to rTMS from those who do not; and whether the ML models may offer clinical benefit. In generating ML models to predict rTMS treatment response and identifying the features that drive both positive and negative outcomes, this study aims to advance understanding of TRD, optimize treatment strategies, and ultimately improve patient outcomes.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cp\u003e \u003cb\u003eDesign.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA retrospective study was conducted, including EMR (Epic, Verona, WI, USA) of TRD patients, collected from the clinical database of the UC San Diego Interventional Psychiatry Program (UCSD-IPP). All contributing procedures comply with the ethical standards of the relevant national and international committees on human subjects\u0026rsquo; research and the Helsinki Declaration and were approved by the UCSD Human Research Protections Program.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePatients.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIncluded patients were prescribed rTMS treatment from UCSD psychiatrists credentialed in delivering rTMS. rTMS was initiated in outpatients who met criteria for TRD, did not present with any treatment contraindications, had adequate insurance coverage, and provided consent. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, EMR records for 419 patients seeking rTMS in 2017\u0026ndash;2023 at UCSD were screened for the following inclusion criteria: (i) primary diagnosis of MDD; (ii) aged 18 years and older; (iii) completion of at least 15 sessions by the date of final PHQ-9; (iv) clinically meaningful depressive symptoms, as indicated by PHQ-9 total score\u0026thinsp;\u0026gt;\u0026thinsp;4; (v) failure to respond to at least 2 antidepressant pharmacotherapy trials; and (vi) completion of clinical evaluations prior to and following the acute treatment phase.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eA total of 246 patients began a rTMS treatment course and 239 completed at least 15 sessions before their final PHQ-9 assessment. Two participants had baseline PHQ-9 scores\u0026thinsp;\u0026lt;\u0026thinsp;5, indicating no clinically significant depressive symptoms, and were excluded from analyses. Five patients were excluded for not having received an MDD diagnosis, yielding a final analytical sample of 232 MDD patients. All patients had previously failed to respond to at least two antidepressant treatments and 90% reported having an MDD diagnosis\u0026thinsp;\u0026gt;\u0026thinsp;10 years. Eighteen patients had baseline PHQ-9 scores\u0026thinsp;\u0026lt;\u0026thinsp;10, indicative of minimal depressive symptoms, and response rates within this subgroup were lower than patients with PHQ-9\u0026thinsp;\u0026ge;\u0026thinsp;10 at baseline (38.9% vs. 43.9%) but did not significantly differ (X\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.17, p\u0026thinsp;=\u0026thinsp;0.68).\u003c/p\u003e \u003cp\u003e \u003cb\u003eTreatment procedure.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAll rTMS treatment was performed in the UCSD-IPP clinic by technicians trained in protocol administration with oversight by an attending psychiatrist. Overseeing psychiatrists reviewed the treatment plans after each session. rTMS was delivered using either a Magventure MagPro with a B70 coil or a BrainsWay H1-coil system. For all patients, the resting motor threshold (rMT) was identified by finding the motor hotspot and utilizing the lowest level of stimulation intensity to elicit movement in the opposite abductor pollicis brevis (APB) muscle on 50% of stimulation attempts. Once rMT was obtained, treatment session parameters were determined. For Magventure treatments, the coil position was localized using the Beam F3 method, and treatment was delivered per standard practice (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Patients who were initiated on the iTBS protocol (at 120% rMT), involving 600 total pulses in bursts of 3 with 8-second inter-train-intervals, were transitioned to either BiTBS (90% rMT; 3600 total pulses, 1800 delivered via 3-iTBS and 1800 delivered cTBS 120% rMT protocol to the left and right hemispheres, respectively) or 3-iTBS (iTBS delivered three times consecutively, per Li et al 2014 (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e) if there was deemed poor efficacy or issues with tolerability). Standard high-frequency left (HFL; 19 minutes) and low-frequency right (LFR; 10 minutes) protocols were used less commonly, both at 120% rMT. The obsessive-compulsive disorder (OCD) protocol uses a B70 coil and delivers 2000 pulses per session via 50, 2-second trains to dorsomedial prefrontal cortex (DMPFC), at 100% rMT of the foot with the Magventure B70 coil. For the BrainsWay H1-coil system either iTBS or the standard FDA-approved 18 Hz protocol was used at 120% rMT. The end of each patient\u0026rsquo;s acute treatment period was clinician-determined and coincided with reduced frequency of rTMS treatment to \u0026lt;\u0026thinsp;2 sessions per week. Sequencing of TMS protocols across all patients included in the analysis is depicted in \u003cb\u003eSupplemental Fig.\u0026nbsp;1\u003c/b\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003eClinical assessment.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe PHQ-9 is a nine-item self-report scale used to assess the presence and severity of depressive symptoms based on the Diagnostic and Statistical Manual (DSM) criteria for a Major Depressive Episode (MDE). Questions are on a four-point scale related to the frequency of symptoms presented on each item, including \u0026ldquo;not at all,\u0026rdquo; \u0026ldquo;several days,\u0026rdquo; \u0026ldquo;more than half the days,\u0026rdquo; and \u0026ldquo;nearly every day.\u0026rdquo; Total scores are calculated by adding scores to all nine items (total range: 0\u0026ndash;27), and previously established cutoff scores for severity are: 0\u0026ndash;4 (no symptoms), 5\u0026ndash;9 (mild symptoms), 10\u0026ndash;14 (moderate symptoms), 15\u0026ndash;19 (moderately severe symptoms), and 20\u0026ndash;27 (severe symptoms). The PHQ-9 was administered pre-treatment and at rTMS visits throughout the acute treatment phase to monitor changes in depressive symptoms. Only pre-treatment and post-acute treatment phase scores were used in the present analyses.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePredictive features.\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003ePre-treatment features.\u003c/em\u003e A total of 32 EMR-derived features were extracted from patient records, including: age, gender, ethnicity, religious affiliation, educational attainment, employment status, relationship status, body mass index (BMI) category (obesity: \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\ge\\)\u003c/span\u003e\u003c/span\u003e 30 kg/m\u003csup\u003e2\u003c/sup\u003e; overweight: 25 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e kg/m\u003csup\u003e2\u003c/sup\u003e \u0026lt; 30; normal 18.5 \u0026lt; kg/m\u003csup\u003e2\u003c/sup\u003e \u0026lt; 25), alcohol, tobacco, and cannabis use (each operationalized as 3-level factor: current, former, or never), family history of mood disorders, 1st degree relatives with mood or psychiatric disorder diagnoses, prior ECT treatment, prior rTMS treatment, history of psychological trauma, non-suicidal self-injury (NSSI), history of suicidal ideation (SI) or suicide attempts, psychiatric hospitalizations, and duration of current MDE. Psychiatric comorbidities were also recorded, including anxiety disorders, trauma-related disorders, substance use disorders, OCD, somatic symptom and related disorders, personality disorders, autism spectrum disorders, eating disorders, and neurocognitive disorders. Comorbid diagnoses were extracted from physician notes. Psychiatric medication use, including benzodiazepines, non-benzodiazepine sedatives, anxiolytics, antipsychotics, and psychostimulants, was recorded, and antidepressant medication burden was translated into daily defined dose (DDD) equivalents.\u003c/p\u003e \u003cp\u003e \u003cem\u003eTreatment-related features.\u003c/em\u003e To determine the role of treatment-related factors in depression outcomes, the following five parameters were recorded: protocol type (deep TMS, iTBS, bilateral TBS, or others that included HFL, LFR, 3-iTBS, or OCD protocols); number of total rTMS sessions; cumulative number of pulses received throughout the treatment course; and, whether patients switched rTMS protocols during treatment.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis.\u003c/h2\u003e \u003cp\u003eR statistical software (v.4.3.1) was used for data analysis. Independent sample \u003cem\u003et\u003c/em\u003e-tests and chi-square tests were used to compare demographic and clinical characteristics between rTMS non-responders and responders, defined as patients with \u0026ge;\u0026thinsp;50% decrease in post-treatment PHQ-9 scores from pre-treatment scores. Pairwise post-hoc tests for factors with more than two levels were false discovery rate (FDR)-corrected. To mitigate low variance within categorical variables, variables were combined and/or operationalized as follows: rTMS protocols: 18 Hz deep TMS (dTMS), BiTBS, or iTBS, made up the majority (86.6%) of rTMS protocols in the sample; therefore, other protocols, including 3-iTBS, 3-iTBS/OCD, OCD, and HFL were combined into a single factor level (\u0026ldquo;Other\u0026rdquo;). Race/ethnicity: patients were coded as non-Hispanic, white, or \u0026ldquo;Other\u0026rdquo;. Psychiatric comorbidities: patients were coded with 1 if they had at least one comorbid diagnosis. Duration of current MDE: patients were coded as \u0026lt;\u0026thinsp;6 months, 6\u0026ndash;12 months, or \u0026gt;\u0026thinsp;12 months in duration. Patients were binarized indicating completion of a 4-year degree or not. Relationship status was coded as: currently single/unpartnered, partnered/married, or separated/divorced/widowed. Non- antidepressant psychiatric medication use was coded 1 or 0 for each drug category.\u003c/p\u003e \u003cp\u003eTo identify pre-treatment patient characteristics and treatment-associated features that best discriminate response and remission groups, a series of classification models were trained using the \u003cem\u003ecaret\u003c/em\u003e package: (i) regularized logistic regression (elastic net regression; ENet); (ii) support vector machine (SVM) with radial basis function; (iii) random forest (RF); and (iv) stochastic gradient boosting machines (GBM). All features were filtered for near-zero-variance (\u0026lt;\u0026thinsp;5%) and linear combinations. Nested cross-validation was used for hyperparameter tuning and model performance evaluation using \u003cem\u003enestedcv\u003c/em\u003e with 10 inner folds and 20 outer folds. RF parameters optimized in the nested grid search included \u003cem\u003emtry\u003c/em\u003e (number of variables to split), \u003cem\u003esplitrule\u003c/em\u003e (quality of each node split), and \u003cem\u003emin.node.size\u003c/em\u003e (minimum number of samples in a node), and GBM parameters included \u003cem\u003einteraction.depth\u003c/em\u003e (highest interaction per tree), \u003cem\u003en.minobsinnode\u003c/em\u003e (minimum number of samples in a node to split), \u003cem\u003en.trees\u003c/em\u003e (total number of trees), and \u003cem\u003eshrinkage\u003c/em\u003e (learning rate), whereas the default tuning grids in \u003cem\u003ecaret\u003c/em\u003e were applied for ENet and SVM. To mitigate class imbalance in the dataset, Synthetic Minority Over-sampling TEchnique (SMOTE) was applied to equalize the class ratio (1:1).\u003c/p\u003e \u003cp\u003eFeature selection via a univariate filter was performed on each outer fold prior to hyperparameter tuning, and a Relief-based algorithm (ReliefFbestK in \u003cem\u003eCORElearn\u003c/em\u003e package; k\u0026thinsp;=\u0026thinsp;10) was applied in parallel for comparison with the univariate filter. Generally, performance was superior for models trained after univariate filtering (\u003cb\u003eSupplemental Fig.\u0026nbsp;3\u003c/b\u003e), with 10 features selected on average (range: 7\u0026ndash;13) for all models. Optimization criteria for each model\u0026rsquo;s hyperparameter tuning procedure were receiver operating characteristic area under the curve (ROC AUC). Continuous features were standardized, and categorical features were one-hot encoded, with the first level dropped as reference. Missing variables were treated as missing completely at random and nonparametric random forest-based imputation was performed using \u003cem\u003emissForest\u003c/em\u003e.\u003c/p\u003e \u003cp\u003eGeneralization performance and 95% confidence intervals were generated by averaging AUC across the best tune of all outer folds. Predicted probabilities across all outer folds were merged to generate additional performance metrics (\u003cem\u003eF\u003c/em\u003e-score, sensitivity, specificity) and models with the best performance for an outcome were selected for downstream analysis. Significance of AUC values were determined by permutation testing (B\u0026thinsp;=\u0026thinsp;99) of class labels, re-running the procedure for each model, and calculating the proportion of permuted models with AUC values greater than the observed data. Alpha level was set at 0.05. The contributions of predictive features were determined using SHapley Additive exPlanations (SHAP) analysis in the \u003cem\u003efastshap\u003c/em\u003e R package, and directionality of associations between features and outcome variables was ascertained by visualizing individual SHAP values and their dependencies with the \u003cem\u003eshapviz\u003c/em\u003e package in R.\u003c/p\u003e \u003cp\u003eUsing the best performing models, the predicted probabilities of treatment response and remission were derived. Model calibration was assessed with the Hosmer-Lemeshow (HL) goodness-of-fit test by binning predicted probabilities into deciles. Probability estimates were updated using beta regression, which has been shown to outperform other calibration methods (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e), and recalibrated probabilities were reassessed using HL test. Decision curve analysis (DCA) was applied using \u003cem\u003edcurves\u003c/em\u003e to assess the clinical utility of prediction models through estimation of net benefit across thresholds of predicted risks using recalibrated probabilities from both \u0026ldquo;response\u0026rdquo; and \u0026ldquo;remission\u0026rdquo; classification models. Net benefit is equivalent to the percentage of individuals appropriately treated (\u0026ldquo;true positives\u0026rdquo;, responders and remitters) minus a weighted percentage of those inappropriately treated (\u0026ldquo;false positives\u0026rdquo;).\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003ePatient and treatment-related characteristics\u003c/h2\u003e \u003cp\u003eSociodemographic and clinical characteristics of patients, and treatment-related characteristics included in the analysis are outlined in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Across all patients, post-treatment PHQ-9 scores were significantly reduced compared to pre-treatment values (mean difference = -7.56, \u003cem\u003et\u003c/em\u003e\u003csub\u003e231\u003c/sub\u003e = -17.6, \u003cem\u003ed\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.16, 95% CI\u0026thinsp;=\u0026thinsp;0.99\u0026ndash;1.32). In total, 101 patients (44%) responded to rTMS, (mean difference = -13.0, \u003cem\u003et\u003c/em\u003e\u003csub\u003e100\u003c/sub\u003e = -29.4, \u003cem\u003ed\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.94, 95% CI\u0026thinsp;=\u0026thinsp;2.48\u0026ndash;3.39; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) and 53 patients (23%) remitted. The average patient age was 55.1\u0026plusmn;16.9 and 54.1\u0026plusmn;17.0 years in the responder and non-responder groups, respectively. Patient sex and ethnicity did not differ between groups; however, responders were more likely to have been former smokers than non-responders, and less likely to have never used alcohol. Most patients were na\u0026iuml;ve to rTMS (88.4%) and ECT (83.3%) treatment, and the majority (60%) had MDEs lasting longer than 12 months at treatment outset. Baseline PHQ-9 scores did not differ between responders (17.9\u0026plusmn;5.4) and non-responders (17.6\u0026plusmn;5.3). Responders were more likely to have a history of trauma and to be diagnosed with trauma disorder compared to non-responders, but were less likely to have an anxiety disorder or be prescribed benzodiazepines or antipsychotic medications. The ratio of patients receiving sequential iTBS-BiTBS to iTBS-only treatment was higher among non-responders than responders (\u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003eadj\u003c/em\u003e\u003c/sub\u003e = 0.04), and non-responders received fewer rTMS sessions on average compared to responders.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSociodemographic and clinical characteristics of patient sample.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAll patients (n\u0026thinsp;=\u0026thinsp;232)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eResponders (n\u0026thinsp;=\u0026thinsp;101; 43.5%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-Responders (n\u0026thinsp;=\u0026thinsp;131; 56.5%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTest Statistic\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSociodemographic characteristics\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (yrs)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e54.5 (16.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e55.1 (16.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e54.1 (17.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003et\u003csub\u003e198\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;1.13, p\u0026thinsp;=\u0026thinsp;0.26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex (%F)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e64.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.02, p = 0.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEthnicity (%NHW/B/H/A/P/MO)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e76/1/6/4/1/12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75/1/4/5/2/13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e76/2/7/3/0/12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{6}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 5.43, p = 0.49\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReligiously affiliated (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 3.16, p = 0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelationship status (S/P/SW)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e35 / 48 / 17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39 / 46 / 15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31 / 50 / 19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.59, p = 0.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCollege graduate (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e81.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e79.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.37, p = 0.54\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI category (%Nm / Ow / Ob)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40 / 30 / 30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 / 29 / 23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34 / 31 / 35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{2}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 5.79, p = 0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1st -degree relative w/ psych dx (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e81.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.92, p = 0.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTobacco use (% N/C/F)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e75 / 7 / 18\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e63 / 9 / 28\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e85 / 5 / 10\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{2}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 14.7, p \u0026lt; 0.01\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCannabis use (% N/C/F)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e67 / 22 / 11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 / 25 / 12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69 / 20 / 11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.01, p = 0.60\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAlcohol use (% N/C/F)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e37 / 46 / 17\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e26 / 52 / 22\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e45 / 41 / 14\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 10.1, p \u0026lt; 0.01\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eClinical characteristics\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePsychiatric comorbidities (n)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.23 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.23 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.24 (0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{216}\\)\u003c/span\u003e\u003c/span\u003e=0.16, p\u0026thinsp;=\u0026thinsp;0.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAntidepressant drug use (DDD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.75 (1.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.65 (1.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.83 (1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{229}\\)\u003c/span\u003e\u003c/span\u003e=0.79, p\u0026thinsp;=\u0026thinsp;0.43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrior rTMS treatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 3.07, p = 0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaseline PHQ-9 score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17.8 (5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17.9 (5.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17.6 (5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{213}\\)\u003c/span\u003e\u003c/span\u003e=0.42, p\u0026thinsp;=\u0026thinsp;0.67\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMDE duration (\u0026lt;\u0026thinsp;6/6\u0026ndash;12/12\u0026thinsp;+\u0026thinsp;mo)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21%/18%/60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28%/19%/53%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16%/18%/66%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{2}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 4.76, p = 0.09\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrior ECT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.79, p = 0.18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistory of trauma\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e45.3%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e54.5%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e38.2%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 6.11, p = 0.01\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrior psych hospitalization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e35.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 2.63, p = 0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSuicidal ideation history\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e64.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.12, p = 0.73\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-suicidal self-injury history\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.31, p = 0.58\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAnxiety disorder\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e72.8%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e65.3%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e78.6%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 5.08, p = 0.02\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrior suicide attempt\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.38, p = 0.24\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubstance use disorder\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e14.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.06, p = 0.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eADHD diagnosis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 2.51, p = 0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTrauma disorder diagnosis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e22.4%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e31.7%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e15.3%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 8.84, p \u0026lt; 0.01\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBenzodiazepine use\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e29.0%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e21.8%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e34.4%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 4.20, p = 0.04\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAntipsychotic medication use\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e24.2%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e17.0%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e29.8%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 5.04, p = 0.02\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePsychostimulant medication use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.03, p = 0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-benzodiazepine sedative use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 0.20, p = 0.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMood stabilizer use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 2.07, p = 0.15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTreatment-related characteristics\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTMS protocol (% BiTBS/dTMS /iTBS/iTBS-BiTBS/Other)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e16/36/10/25/13\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e13/41/16/18/13\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e18/33/5/31/14\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{\\chi }}_{4}^{2}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 11.8, p = 0.02\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTotal number of rTMS sessions\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e37.4 (10.6)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e39.1 (10.8)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e36.2 (10.2)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varvec{t}}_{209}\\)\u003c/span\u003e\u003c/span\u003e\u003cb\u003e=2.08, p\u0026thinsp;=\u0026thinsp;0.04\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCumulative rTMS pulses\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e88,804(44,862)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e88,832(48,831)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e88,782(41,738)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{196}\\)\u003c/span\u003e\u003c/span\u003e=0.01, p\u0026thinsp;=\u0026thinsp;0.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003erTMS protocol switch (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{1}^{2}\\)\u003c/span\u003e\u003c/span\u003e = 1.64, p = 0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeeks from final patient in DB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e136 (81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e146 (85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e128 (78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{204}\\)\u003c/span\u003e\u003c/span\u003e=1.62, p\u0026thinsp;=\u0026thinsp;0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResting motor threshold (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e46.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{200}\\)\u003c/span\u003e\u003c/span\u003e=0.44, p\u0026thinsp;=\u0026thinsp;0.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLength of treatment (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.3 (19.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69.0 (19.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e64.2 (19.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{217}\\)\u003c/span\u003e\u003c/span\u003e=1.85, p\u0026thinsp;=\u0026thinsp;0.07\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eOther rTMS protocols include 3-iTBS, 3-iTBS/OCD, HFL, and OCD. Ethnicities reported include Non-Hispanic White (NHW), Black (B), Hispanic (H), Asian (A), Persian (P), and Mixed or Other (MO). Substance use reported as never (N), current (C), or former (F). Relationship status reported as Single (S), Partnered/Married (P), or Widowed/Divorced (W). Psychiatric comorbidities included OCD, somatic symptom disorders, personality disorders, autism spectrum disorders, eating disorders, and neurocognitive disorders. BMI categories corresponded to kg/m\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\ge\\)\u003c/span\u003e\u003c/span\u003e \u003cem\u003e30 (Ob), 25\u003c/em\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e \u003cem\u003eOw \u0026lt; 30, and 18.5 \u0026lt; Nm \u0026lt; 25.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eClassification results\u003c/h2\u003e \u003cp\u003e \u003cem\u003eTreatment Response.\u003c/em\u003e Across the four models tested, the SVM model predicted treatment response significantly above chance (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01; \u003cb\u003eSupplemental Fig.\u0026nbsp;2\u003c/b\u003e) with modest discrimination performance (mean AUC\u0026thinsp;=\u0026thinsp;0.689, 95% CI\u0026thinsp;=\u0026thinsp;0.638\u0026ndash;0.740), outperforming RF (0.654, 0.594\u0026ndash;0.715), ENet (0.668, 0.616\u0026ndash;0.719), and GBM (0.617, 0.563\u0026ndash;0.671) models (\u003cb\u003eSupplemental Fig.\u0026nbsp;3\u003c/b\u003e). Thus, the SVM model was evaluated further. The sensitivity and specificity of the model to discriminate treatment responders from non-responders were 72.3% and 53.4%, respectively, with an F-score of 0.64. Following recalibration, observed and predicted probabilities of treatment response were similar (HL test: \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{8}^{2}\\)\u003c/span\u003e\u003c/span\u003e=5.51, \u003cem\u003ep\u003c/em\u003e=0.70, RMSE=1.73; \u003cb\u003eSupplemental Fig.\u0026nbsp;4\u003c/b\u003e). In descending order, the most important variables in prediction of treatment response were anxiety disorder, former tobacco use, obesity, trauma disorder, antipsychotic use, benzodiazepine use, iTBS-BiTBS protocol, \u0026gt;\u0026thinsp;12-month duration of current MDE, history of trauma, number of rTMS sessions, and iTBS protocol (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Partial dependence plots illustrating the relationship between the top-3 features, the total number of rTMS sessions, and treatment response are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB. DCA indicated that for threshold probabilities between 25% and 55%, corresponding to a number needed-to-treat (NNT) between 4 and 2 respectively, the net benefit of applying the model to aid in determining whether a patient should undergo rTMS is higher than either a \u0026ldquo;treat all\u0026rdquo; or \u0026ldquo;treat none\u0026rdquo; approach (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003e \u003cem\u003eDepression remission.\u003c/em\u003e Across the four models tested, the GBM model predicted remission following rTMS treatment significantly above chance (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01; \u003cb\u003eSupplemental Fig.\u0026nbsp;2\u003c/b\u003e) with good discrimination performance (mean AUC\u0026thinsp;=\u0026thinsp;0.745, 95% CI\u0026thinsp;=\u0026thinsp;0.692\u0026ndash;0.797), outperforming RF (0.735, 0.680\u0026ndash;0.791), ENet (0.740, 0.668\u0026ndash;0.813), and SVM (0.683, 0.627\u0026ndash;0.738) models (\u003cb\u003eSupplemental Fig.\u0026nbsp;3\u003c/b\u003e). Thus, the GBM model was evaluated further. The sensitivity and specificity of the model to discriminate remitters from non-remitters were 58.5% and 77.7%, respectively, with an F-score of 0.50. Following recalibration, the observed and predicted probabilities of treatment response remained dissimilar (HL test: \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\chi }_{8}^{2}\\)\u003c/span\u003e\u003c/span\u003e=19.1, \u003cem\u003ep\u003c/em\u003e=0.01, RMSE=2.73; \u003cb\u003eSupplemental Fig.\u0026nbsp;4\u003c/b\u003e). In descending order, the most important variables in the prediction of depression remission were pre-treatment PHQ-9 score, obesity, antipsychotic use, current MDE greater than 12 months, benzodiazepine use, anxiety, prior ECT, and iTBS-BiTBS protocol (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Partial dependence plots illustrating the relationship between the top-four features and treatment response are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eB. DCA indicated that for threshold probabilities between 15% and 40%, corresponding to a NNT between approximately 7 and 3, respectively, using the model to determine whether a patient should undergo rTMS yielded greater net benefit than employing either a \u0026ldquo;treat all\u0026rdquo; or \u0026ldquo;treat none\u0026rdquo; strategy (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eFindings presented in this study underscore the substantial variability in depression responses and remission, despite rTMS efficacy, and the promise of ML to predict treatment outcomes using EMR data alone. Importantly, both ML models for response and remission identified psychiatric comorbidities, depressive episode duration, and specific rTMS protocols as predictive features, with trauma history also associated with response outcomes.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ePredictive features of treatment outcomes\u003c/span\u003e:\u003c/h2\u003e \u003cp\u003ePast research investigating the relationship between psychiatric comorbidities and treatment response has yielded valuable insights into the complex interplay between mental health conditions and their respective treatments (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). This literature presents mixed results regarding impact of trauma history on rTMS. Experiences of childhood trauma and PTSD diagnosis have been reported to negatively affect rTMS outcomes in MDD patients (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e) or to have no significant impact (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). Positive findings, as presented here, may point to a consequence of trauma-induced neuroplastic changes that sensitize the brain to neurostimulation (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). For example, the therapeutic impact of rTMS could be modulated by individual differences in neural circuitry affected by trauma, which are not always uniformly engaged by standard rTMS protocols, such as 10 Hz rTMS to the left DLPFC as done in Hu et al. (2012) and Yesavage et al. (2018).\u003c/p\u003e \u003cp\u003eFormer tobacco use was identified as a positive predictor of treatment response. Past literature on tobacco use in rTMS outcomes is mixed (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e), but nicotine\u0026rsquo;s complex neurobiological effects may enhance efficacy. Nicotine potentiates dopaminergic neurotransmission via nicotinic acetylcholine receptor binding on dopaminergic neurons in the mesolimbic pathway associated with mood regulation (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). This dopamine surge may have a persistent effect that synergizes with rTMS mechanisms, inducing changes in brain plasticity (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). Coupled with trauma history, these associations suggest past experiences and lifestyle modifications may prime the brain's receptivity to rTMS.\u003c/p\u003e \u003cp\u003eFeatures including comorbid anxiety disorder, concomitant benzodiazepine or antipsychotic medications, and a longer duration of current MDE were linked to lower probabilities of treatment response and remission. Anxiety is highly comorbid with depression, and this specific comorbidity may reflect distinct neurobiological underpinnings, affecting the response to rTMS. For instance, hyperactivity in the amygdala and its connectivity with the PFC, can be more pronounced in individuals with comorbid anxiety (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). An extensive body of research supports that comorbid anxiety can complicate the course of depression, often leading to a more chronic and treatment-resistant condition (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eConcurrent use of benzodiazepines or antipsychotics was associated with less favorable responses. These medications exert sedative effects and reduce cortical excitability, potentially impeding operating mechanisms of rTMS. Benzodiazepines, for instance, enhance GABAergic inhibition, which may diminish excitatory processes (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). Antipsychotics, particularly those blocking dopamine receptors, have been shown to attenuate rTMS outcomes (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e), potentially disrupting the dopaminergic modulation crucial for rTMS-induced neuroplasticity. These findings suggest patients prescribed benzodiazepines or antipsychotics may require adjusted protocols, perhaps with increased intensity or frequency to overcome medication-induced reductions in cortical excitability, or timing treatment sessions relative to medication dosing.\u003c/p\u003e \u003cp\u003eThe observation that longer depressive episodes correlated with poorer outcomes is consistent research indicating depression chronicity can negatively impact the efficacy of various treatments. Chronic depression is associated with more severe neurobiological changes, including alterations in the hippocampus and PFC (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e). Earlier intervention might enhance response by addressing such changes before they become entrenched. Fava (2003) highlighted the predictive role of untreated depression duration in treatment resistance (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e) and underscoring the possibility that adjusting standard protocols, or combining with other interventions, may increase efficacy for these patients.\u003c/p\u003e \u003cp\u003eThe observed absence of sociodemographic variables, such as age, sex, and ethnicity, impacting outcomes aligns with past studies (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e), yet diverges from others. For example, Hanlon and colleagues (2022) found younger age and female sex were associated with better rTMS response rates, suggesting demographic characteristics may influence treatment success (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e). The contrasting evidence indicates a complex relationship between sociodemographic and rTMS efficacy, warranting further investigation to clarify their roles in treatment personalization and optimization.\u003c/p\u003e \u003cp\u003eAccrued effects from patient history, such as trauma and anxiety, and concurrent benzodiazepine use may influence rTMS outcomes. The potential cumulative effect of these factors on response could be linked to their disparate impact on brain neuroplasticity and neurotransmission. For instance, a history of trauma and relatively lower anxiety might enhance receptivity to rTMS via induced neuroplastic changes, whereas the suppressive effect of benzodiazepines on cortical excitability could counteract these benefits, requiring adjustments in treatment protocols to optimize efficacy. Further investigation into the interaction of these factors may provide deeper insights into personalized rTMS strategies for patients with complex clinical profiles.\u003c/p\u003e \u003cp\u003eRegarding treatment-related factors, the number of rTMS sessions was a positive predictor, with SHAP values indicating a positive linear dose-response relationship. This relationship aligns with notions that additional sessions sustain antidepressant effects, possibly through cumulative neuroplastic changes. However, this is not consistently upheld, as a meta-analysis reported higher doses of rTMS, measured by total pulses, were not always associated with depression improvement (\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e). Adherence to the iTBS protocol was also among the strongest predictors of response. Notably, iTBS demonstrated superiority over iTBS-BiTBS in our sample, challenging some of the existing literature that suggests bilateral approaches can be advantageous for certain patient populations (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e). Despite this, the current findings likely relate to iTBS-BiTBS patients experiencing greater treatment-resistance at baseline, with providers transitioning protocols from unilateral to bilateral in cases of inadequate symptom relief.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eMachine learning, electronic medical records, and clinical decision-making\u003c/h2\u003e \u003cp\u003eRecent reviews of ML prediction models in psychiatric research have found that despite relatively \u0026lsquo;good\u0026rsquo; discrimination performance (AUC\u0026thinsp;~\u0026thinsp;0.65\u0026ndash;0.80), most studies had methodological weaknesses that limited their generalizability and utility in clinical decision-making (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e). The authors recommended more comprehensive reporting of ML methods and results, more robust validation methods, model calibration, and formal assessment of clinical utility, such as DCA (\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e). In our analyses, DCA demonstrated that the response and remission models outperformed \"treat all\" and \u0026ldquo;treat none\u0026rdquo; strategies within probability threshold ranges of 25\u0026ndash;55% and 15\u0026ndash;40%, or NNT between 2 and 4 patients, and 3 and 7 patients, respectively. Both models indicate net clinical benefit within their respective threshold ranges, though more refined models incorporating additional patient features, such as neurobiological markers, polygenic risk scores, and psychosocial determinants, may confer additional benefit.\u003c/p\u003e \u003cp\u003eThe implications of these findings are multi-fold. Firstly, they re-affirm the clinical utility of rTMS in managing TRD, especially when considering individual patient histories and comorbidities. Secondly, these results lend support for the predictive power of ML algorithms derived from EMR for refining treatment approaches, suggesting that these tools could become integral to psychiatric precision medicine. Future research should aim to validate these predictive models in diverse clinical settings, ensuring that findings are replicable and robust.\u003c/p\u003e \u003cp\u003e \u003cb\u003eLimitations.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe retrospective study design provides a robust dataset but holds inherent limitations. Firstly, these ML models were not externally validated in an independent sample. However, nested cross-validation, a robust internal validation procedure, provides a strong estimation of model generalizability. Secondly, this study is susceptible to selection bias since it involves patients who initiated and completed treatment, which may exclude those with different response trajectories, possibly skewing findings towards patients more likely to complete treatment. Moreover, the sample in this study was predominantly highly educated and non-Hispanic White, limiting generalization to other demographics and pointing more towards the need to address disparities in access to TMS (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e). Third, information bias may arise from EMR data availability, accuracy, and completeness, with additional variability in the timing of data collection vis-\u0026agrave;-vis treatment, possibly creating discrepancies between recorded and actual patient conditions. Future research should address these limitations by implementing prospective designs, standardizing rTMS protocols, timely data collection, and diverse populations.\u003c/p\u003e \u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study elucidates predictors of treatment response to rTMS in TRD patients and highlights the importance of treatment approaches tailored to individual characteristics. By leveraging EMR and ML techniques, this study identified demographic, clinical, and treatment-related factors associated with outcomes, providing valuable insights for clinical practice and research.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflict of interest: \u003c/h2\u003e\n\u003cp\u003eAll authors declare that they have no conflict of interest related to the research presented in this manuscript.\u003c/p\u003e\n\u003ch2\u003eDisclosures: \u003c/h2\u003e\n\u003cp\u003eThe authors have no financial ties to disclose.\u003c/p\u003e\n\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis research was funded by the University of California San Diego Health Sciences Research Award #RG114131 to author LGA, R25 MH101072 to author KT, and generous support from the Kreutzkamp Family Foundation.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAmerican Psychiatric Association, American Psychiatric Association, editors. Diagnostic and statistical manual of mental disorders: DSM-5. 5th ed. Washington, D.C: American Psychiatric Association; 2013. 947 p.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e(WHO) WHO. Mental disorders [Internet]. [cited 2023 Jun 5]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/mental-disorders\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/mental-disorders\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhdanava M, Pilon D, Ghelerter I, Chow W, Joshi K, Lefebvre P, et al. The Prevalence and National Burden of Treatment-Resistant Depression and Major Depressive Disorder in the United States. J Clin Psychiatry. 2021;82(2):0\u0026ndash;0.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZimmerman M, Ellison W, Young D, Chelminski I, Dalrymple K. How many different ways do patients meet the diagnostic criteria for major depressive disorder? Compr Psychiatry. 2015;56:29\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLynch CJ, Gunning FM, Liston C. Causes and Consequences of Diagnostic Heterogeneity in Depression: Paths to Discovering Novel Biological Depression Subtypes. Biol Psychiatry. 2020;88(1):83\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenster LL, Weissman CR, Stolz LA, Daskalakis ZJ, Appelbaum LG. Pre-clinical indications of brain stimulation treatments for non-affective psychiatric disorders, a status update. Transl Psychiatry. 2023;13(1):1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAaronson S. Overview of a large transcranial magnetic stimulation patient outcomes database. Brain Stimul Basic Transl Clin Res Neuromodulation. 2021;14(6):1717.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFitzgerald PB, Oxley TJ, Laird AR, Kulkarni J, Egan GF, Daskalakis ZJ. An analysis of functional neuroimaging studies of dorsolateral prefrontal cortical activity in depression. Psychiatry Res Neuroimaging. 2006;148(1):33\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayberg HS. Limbic-cortical dysregulation: a proposed model of depression. J Neuropsychiatry Clin Neurosci. 1997;9(3):471\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGeorge MS, Lisanby SH, Avery D, McDonald WM, Durkalski V, Pavlicova M, et al. Daily left prefrontal transcranial magnetic stimulation therapy for major depressive disorder: a sham-controlled randomized trial. Arch Gen Psychiatry. 2010;67(5):507\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCole EJ, Stimpson KH, Bentzley BS, Gulser M, Cherian K, Tischler C, et al. Stanford Accelerated Intelligent Neuromodulation Therapy for Treatment-Resistant Depression. Am J Psychiatry. 2020;177(8):716\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung SW, Hoy KE, Fitzgerald PB. Theta-burst stimulation: a new form of TMS treatment for depression? Depress Anxiety. 2015;32(3):182\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng Z, ZHOU C, XUE S, BAI J, YU S, LI X, et al. Mechanism of Repetitive Transcranial Magnetic Stimulation for Depression. Shanghai Arch Psychiatry. 2018;30(2):84\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeam AL, Kohane IS. Big Data and Machine Learning in Health Care. JAMA. 2018;319(13):1317\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajkomar A, Dean J, Kohane I. Machine Learning in Medicine. N Engl J Med. 2019;380(14):1347\u0026ndash;58.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeehan AJ, Lewis SJ, Fazel S, Fusar-Poli P, Steyerberg EW, Stahl D, et al. Clinical prediction models in psychiatry: a systematic review of two decades of progress and challenges. Mol Psychiatry. 2022;27(6):2700\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRassen JA, Bartels DB, Schneeweiss S, Patrick AR, Murk W. Measuring prevalence and incidence of chronic conditions in claims and electronic health record databases. Clin Epidemiol. 2018;11:1\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeam W, Borckardt JJ, Reeves ST, George MS. An efficient and accurate new method for locating the F3 position for prefrontal TMS applications. Brain Stimulat. 2009;2(1):50\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi CT, Chen MH, Juan CH, Huang HH, Chen LF, Hsieh JC, et al. Efficacy of prefrontal theta-burst stimulation in refractory depression: a randomized sham-controlled study. Brain J Neurol. 2014;137(Pt 7):2088\u0026ndash;98.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOjeda FM, Jansen ML, Thi\u0026eacute;ry A, Blankenberg S, Weimar C, Schmid M, et al. Calibrating machine learning approaches for probability estimation: A comprehensive comparison. Stat Med. 2023;42(29):5451\u0026ndash;78.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrishnan KRR. Comorbidity and depression treatment. Biol Psychiatry. 2003;53(8):701\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYesavage JA, Fairchild JK, Mi Z, Biswas K, Davis-Karim A, Phibbs CS, et al. Effect of Repetitive Transcranial Magnetic Stimulation on Treatment-Resistant Major Depression in US Veterans. JAMA Psychiatry. 2018;75(9):884\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu YT, Hu XW, Han JF, Zhang JF, Wang YY, Wolff A, et al. Childhood trauma mediates repetitive transcranial magnetic stimulation efficacy in major depressive disorder. Eur Arch Psychiatry Clin Neurosci. 2021;271(7):1255\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNg E, Wong EHY, Lipsman N, Nestor SM, Giacobbe P. Adverse childhood experiences and repetitive transcranial magnetic stimulation outcomes for depression. J Affect Disord. 2023;320:716\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKarsen EF, Watts BV, Holtzheimer PE. Review of the Effectiveness of Transcranial Magnetic Stimulation for Post-traumatic Stress Disorder. Brain Stimul Basic Transl Clin Res Neuromodulation. 2014;7(2):151\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdelrahman AA, Noaman M, Fawzy M, Moheb A, Karim AA, Khedr EM. A double-blind randomized clinical trial of high frequency rTMS over the DLPFC on nicotine dependence, anxiety and depression. Sci Rep. 2021;11(1):1640.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoulet E, Galvao F, Haffen E, Szekely D, Brault C, Haesebaert F, et al. Effects of smoking status and MADRS retardation factor on response to low frequency repetitive transcranial magnetic stimulation for depression. Eur Psychiatry. 2016;38:40\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenowitz NL. Pharmacology of Nicotine: Addiction, Smoking-Induced Disease, and Therapeutics. Annu Rev Pharmacol Toxicol. 2009;49:57\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBelujon P, Grace AA. Dopamine System Dysregulation in Major Depressive Disorders. Int J Neuropsychopharmacol. 2017;20(12):1036\u0026ndash;46.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShin LM, Liberzon I. The neurocircuitry of fear, stress, and anxiety disorders. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2010;35(1):169\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eComorbid anxiety and depression in adults: Epidemiology, clinical manifestations, and diagnosis - UpToDate [Internet]. [cited 2024 Mar 3]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.uptodate.com/contents/comorbid-anxiety-and-depression-in-adults-epidemiology-clinical-manifestations-and-diagnosis\u003c/span\u003e\u003cspan address=\"https://www.uptodate.com/contents/comorbid-anxiety-and-depression-in-adults-epidemiology-clinical-manifestations-and-diagnosis\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilson S, Olsen S, Sullivan C, Cooper D, Somez I, Widge AS, et al. Concurrent Benzodiazepine Use and TMS Clinical Outcomes. Biol Psychiatry. 2021;89(9):S288.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHebel T, Abdelnaim M, Deppe M, Langguth B, Schecklmann M. Attenuation of antidepressive effects of transcranial magnetic stimulation in patients whose medication includes drugs for psychosis. J Psychopharmacol Oxf Engl. 2020;34(10):1119\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheline YI, Gado MH, Kraemer HC. Untreated depression and hippocampal volume loss. Am J Psychiatry. 2003;160(8):1516\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFava M. Diagnosis and definition of treatment-resistant depression. Biol Psychiatry. 2003;53(8):649\u0026ndash;59.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeuchter M, Citrenbaum C, Wilson A, Tibbe T, Jackson N, Krantz D, et al. Holding Steady: Age-Related Effects on Treatment Response During rTMS Treatment of Depression. Am J Geriatr Psychiatry. 2023;31(3, Supplement):S119\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHanlon CA, McCalley DM. Sex/Gender as a Factor That Influences Transcranial Magnetic Stimulation Treatment Outcome: Three Potential Biological Explanations. Front Psychiatry. 2022;13:869070.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu CL, Kao YC, Thompson T, Brunoni AR, Hsu CW, Carvalho AF, et al. The association of total pulses with the efficacy of repetitive transcranial magnetic stimulation for treatment-resistant major depression: A dose-response meta-analysis. Asian J Psychiatry. 2024;92:103891.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlumberger DM, Vila-Rodriguez F, Thorpe KE, Feffer K, Noda Y, Giacobbe P, et al. Effectiveness of theta burst versus high-frequency repetitive transcranial magnetic stimulation in patients with depression (THREE-D): a randomised non-inferiority trial. The Lancet. 2018;391(10131):1683\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrzenda A, Widge AS. Electronic health records and stratified psychiatry: bridge to precision treatment? Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2024;49(1):285\u0026ndash;90.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVickers AJ, Calster BV, Steyerberg EW. Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests. BMJ. 2016;352:i6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNagra DS, Stolz LA, Weissman CR, Appelbaum LG. Access to interventional psychiatric treatments in the United States: Disparities and proposed solutions. Glob Health Econ Sustain. 2024;2(1):2456.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"translational-psychiatry","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"tp","sideBox":"Learn more about [Translational Psychiatry](http://www.nature.com/tp/)","snPcode":"41398","submissionUrl":"https://mts-tp.nature.com/cgi-bin/main.plex","title":"Translational Psychiatry","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-4396926/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4396926/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Identifying predictors of treatment response to repetitive transcranial magnetic stimulation (rTMS) remain elusive in treatment-resistant depression (TRD). Leveraging electronic medical records (EMR), this retrospective cohort study applied supervised machine learning (ML) to sociodemographic, clinical, and treatment-related data to predict depressive symptom response (\u003e50% reduction on PHQ-9) and remission (PHQ-9 \u003c 5) following rTMS in 232 patients with TRD (mean age: 54.5, 63.4% women) treated at the University of California, San Diego Interventional Psychiatry Program between 2017 and 2023. ML models were internally validated using nested cross-validation and Shapley values were calculated to quantify contributions of each feature to response prediction. The best-fit models proved reasonably accurate at discriminating treatment responders (Area under the curve (AUC): 0.689 [0.638, 0.740], p \u003c 0.01) and remitters (AUC 0.745 [0.692, 0.797], p \u003c 0.01), though only the response model was well-calibrated. Both models were associated with significant net benefits, indicating their potential utility for clinical decision-making. Shapley values revealed that patients with comorbid anxiety, obesity, concurrent psychiatric medication use, and more chronic TRD were less likely to respond or remit following rTMS. Patients with trauma and former tobacco users were more likely to respond. Furthermore, delivery of intermittent theta burst stimulation and more rTMS sessions were associated with superior outcomes. These findings highlight the potential of ML-guided techniques to guide clinical decision-making for rTMS treatment in patients with TRD to optimize therapeutic outcomes.","manuscriptTitle":"Predictive modeling of response to repetitive transcranial magnetic stimulation in treatment-resistant depression","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-23 18:41:09","doi":"10.21203/rs.3.rs-4396926/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2024-09-18T14:14:15+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2024-08-22T20:18:36+00:00","index":4,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-08-09T01:42:51+00:00","index":4,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2024-06-09T23:25:42+00:00","index":3,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-06-03T19:43:51+00:00","index":3,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-05-26T21:45:53+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2024-05-13T18:17:47+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2024-05-12T14:56:06+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-10T10:47:07+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-09T19:26:47+00:00","index":"","fulltext":""},{"type":"submitted","content":"Translational Psychiatry","date":"2024-05-09T19:26:47+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"translational-psychiatry","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"tp","sideBox":"Learn more about [Translational Psychiatry](http://www.nature.com/tp/)","snPcode":"41398","submissionUrl":"https://mts-tp.nature.com/cgi-bin/main.plex","title":"Translational Psychiatry","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"129cd16d-d344-4adc-9bf9-bd0a66607c5e","owner":[],"postedDate":"May 23rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":31816779,"name":"Health sciences/Diseases/Psychiatric disorders/Depression"},{"id":31816780,"name":"Biological sciences/Psychology/Human behaviour"}],"tags":[],"updatedAt":"2025-04-28T07:07:18+00:00","versionOfRecord":{"articleIdentity":"rs-4396926","link":"https://doi.org/10.1038/s41398-025-03380-w","journal":{"identity":"translational-psychiatry","isVorOnly":false,"title":"Translational Psychiatry"},"publishedOn":"2025-04-27 04:00:00","publishedOnDateReadable":"April 27th, 2025"},"versionCreatedAt":"2024-05-23 18:41:09","video":"","vorDoi":"10.1038/s41398-025-03380-w","vorDoiUrl":"https://doi.org/10.1038/s41398-025-03380-w","workflowStages":[]},"version":"v1","identity":"rs-4396926","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4396926","identity":"rs-4396926","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}