Biomarker-Guided Neuromodulation for TRD: Lessons for Psychedelic Trials

New Predictive Models for Neuromodulation in Treatment-Resistant Depression

A September 2026 dissertation from UC San Diego presents robust evidence that integrating clinical, demographic, and neurophysiological data with machine learning can predict patient responses to neuromodulatory interventions—specifically repetitive transcranial magnetic stimulation (rTMS) and esketamine—in treatment-resistant depression (TRD). The study, available via OpenAlex, analyzed electronic medical records from 232 TRD patients and applied advanced statistical and computational techniques to identify key predictors of treatment success. This approach marks a significant advance in the precision psychiatry paradigm and sets a new benchmark for research and clinical practice in TRD and related fields, including psychedelic therapy development.

Mechanisms and Key Predictors: Machine Learning Meets Real-World Data

Machine learning models trained on real-world clinical data identified several concrete predictors of rTMS response: comorbid anxiety disorder, former tobacco use, obesity, trauma disorder, antipsychotic and benzodiazepine use, rTMS protocol type, and number of sessions. For esketamine, survival analysis revealed a more rapid onset of response and remission compared to rTMS, but with distinct patient-specific trajectories shaped by comorbidity burden and treatment parameters. Notably, the dissertation explored the cortical silent period (CSP)—a neurophysiological marker measured via transcranial magnetic stimulation—as a potential biomarker of response. While CSP findings were preliminary and inconsistent, their inclusion illustrates the value of multimodal assessment in understanding and predicting outcomes.

• Comorbidities and medication use emerged as strong predictors of both response and non-response, underscoring the complexity of TRD populations.
• Protocol customization (e.g., session number, stimulation parameters) significantly influenced outcomes, suggesting that rigid, one-size-fits-all approaches may miss opportunities for optimization.
• Temporal response patterns differed between interventions: esketamine offered faster relief for some, while rTMS provided steadier improvement, often contingent on patient characteristics.

Implications for Psychedelic Research and Precision Psychiatry

The dissertation's findings have direct implications for those developing and researching psychedelic therapies for TRD. Psychedelic trials often face similar challenges: heterogeneous patient populations, variable response rates, and a lack of validated biomarkers for predicting who will benefit. This work demonstrates that single-domain predictors—whether clinical, demographic, or neurophysiological—are insufficient for guiding treatment selection. Instead, a multimodal, data-driven framework, validated on real-world populations, is essential for advancing precision psychiatry. For psychedelic clinical trials, this means incorporating diverse data types (e.g., EEG, MRI, digital phenotyping) and leveraging machine learning to stratify patients and tailor interventions. Importantly, the study also highlights the need for prospective validation of predictive models before clinical deployment—a step often overlooked in early-phase psychedelic studies.

An underappreciated insight is the role of comorbidities and polypharmacy as both confounders and potential moderators of treatment response. Many psychedelic trials exclude complex TRD patients, yet real-world populations resemble those in this dissertation. This gap between trial populations and clinical reality may limit generalizability unless addressed through similar multimodal predictive approaches.

Risks, Limitations, and Open Questions

While the dissertation advances the field, several risks and unknowns remain. The retrospective design and reliance on a single academic center's data may limit external validity. Machine learning models, though powerful, can overfit to specific datasets and may not generalize across sites or demographics. The exploratory use of CSP as a biomarker yielded inconsistent results, emphasizing the need for larger, prospective studies to validate neurophysiological predictors. For psychedelic research, these limitations caution against premature adoption of unvalidated markers or overreliance on single-center data. Additionally, the ethical and logistical challenges of collecting multimodal data—especially neurophysiological measures—in routine clinical settings warrant careful consideration.

Looking Ahead: Toward Personalized, Ecologically Valid TRD Interventions

The integration of machine learning, survival analysis, and neurophysiological data in this dissertation lays a foundation for future research aiming to personalize interventions for TRD. For psychedelic researchers, the key takeaway is the necessity of individualized, data-driven frameworks that reflect the complexity of real-world patients. Prospective, multisite validation of predictive models—incorporating both traditional and novel biomarkers—will be crucial for translating these insights into clinical practice. As regulatory agencies and payers increasingly demand evidence of precision and cost-effectiveness, such approaches may become prerequisites for the approval and adoption of novel therapies, including psychedelics, in TRD.

How we research: This article was written by Dr. Alex T. Morgan, PhD (Neuroscience, UC San Diego), and reviewed by Dr. Sara Kim, MD, on 2026-09-14. Primary source: UCSD dissertation via OpenAlex.