Rapid advances in artificial intelligence (AI), particularly large language models (LLMs), are transforming how clinicians and researchers understand, assess, and treat anxiety and depression. Yet the field is navigating critical questions regarding validity, interpretability, real-world impact, and clinical safety. This Scientific Research Symposium brings together leading investigators working across the translational pipeline, from mechanistic discovery to clinical deployment, to highlight how AI can be used responsibly to advance mental health care.
The session will begin with Roy Perlis, MD, MSc (Harvard Medical School), who has developed methods for leveraging LLMs to quantify depression severity, stratify suicide risk, and capture symptom expression directly from patient language. His work demonstrates how AI-driven phenotyping can augment traditional assessments, improve risk prediction, and refine our understanding of transdiagnostic symptom structure.
Next, Millard Brown, MD (Spring Health) will describe the integration of AI into real-world care delivery. Spring Health’s large-scale clinical data infrastructure provides a unique testbed for evaluating how LLM-based tools can support care navigation, enhance therapeutic engagement, and assist clinicians in making data-informed treatment decisions. This talk will emphasize implementation challenges and opportunities in diverse patient populations and employer-based benefit systems.
The symposium will conclude with Karthik Sarma, MD, PhD (UCSF), who is focused on developing guided reasoning frameworks for LLM systems in diagnosis, triage, and patient interaction. His research examines how to evaluate the mental health impact and safety of AI-based conversational agents, and how LLMs can interact with electronic health records to support clinicians while minimizing cognitive and documentation burden.
Together, these talks will provide a cohesive overview of how AI can help: (1) uncover latent symptom mechanisms; (2) enable precision in assessment and risk stratification; and (3) responsibly guide treatment delivery. The session will discuss ethical considerations, clinical guardrails, and future directions for designing AI systems that enhance, rather than replace, human therapeutic judgment.
@inproceedings{sarma2026adaa_science,
author = {Sarma, K. V.},
title = {Accelerating the Science of Large Language Models for Behavioral Health},
booktitle = {ADAA Annual Meeting},
year = {2026},
note = {Invited Guest Speaker, 2026 ADAA Scientific Research Symposium},
}