Leveraging Machine Learning to Train the Workforce in Cognitive Behavioral Techniques for Psychosis
Recent evidence suggests that fewer than 1% of Americans who are diagnosed with a psychotic disorder have access to Cognitive Behavioral Therapy for psychosis (Kopelovich, Nutting, Blank, Buckland, & Spigner, 2021: [Full article: Preliminary point prevalence of Cognitive Behavioral Therapy for psychosis (CBTp) training in the U.S. and Canada (tandfonline.com)). Training in this intervention is notoriously difficult to access, expensive, and time-intensive. Moreover, individuals who experience psychotic symptoms benefit from interacting with practitioners who have an understanding of psychosis, what works and what doesn’t, and who can reinforce high-yield, evidence-based strategies for addressing psychotic symptoms.
Led by Dr. Sarah Kopelovich, a board certified clinical psychologist and co-director of the SPIRIT Center, a unique partnership between the University of Washington and a team of behavioral health and machine learning experts at Lyssn.io, Inc has set out to change the mental health training game. Funded by the National Institute of Mental Health (R42MH123215), the team combined their expertise in evidence-based clinical training, CBT for psychosis, natural language processing, and user-centered design to develop, pilot, and test a fully automated CBT for psychosis online training tool.
What does the CBTp Training Tool Do?
The CBTp Training Tool is a novel online training that introduces learners to “bite-sized” training modules covering foundational cognitive behavioral competencies [Kopelovich et al., In Press(a)]. After a brief video and text-based educational content, learners “meet” their four clients. Learners can select a client, which will then unlock a brief video that will loko and feel like a real telehealth session. After listening to the client’s concern, the learner records their response. Using cutting-edge spoken language technology, the tool analyzes their response and provides real-time feedback. The training tool provides learners with 24/7, on-demand access to microlearning modules, expert demonstrations, and practice opportunities with expert-vetted feedback. Learners can decide whether to try the skill again with the same client, try the skill with a new client, or move along to another skill module. They can practice as often as they would like, when they would like, in the comfort of their own space.
How does it work?
Using natural language processing and trained machine learning algorithms, the tool analyzes learners responses and compares it to roughly 4,000 previously-recorded and rated responses. In a validation study [Kopelovich et al., In Press(b)] machine learning model demonstrated strong correlations between human and machine ratings across learning modules, with all models exceeded 80% of human agreement. This finding indicates that the machine learning model functioned similarly to a human rater and similarly to other ML-based solutions for the evaluation of psychotherapy skills and Cognitive Behavioral Therapy.
What are the benefits of using the tool?
The study team has sought feedback from users during the initial development phase, during an iterative user centered design stage, at the conclusion of a 2-week field trial, and following exposure to the tool for a 3-month trial period. In addition, the study team is currently concluding a large Randomized Clinical Trial (RCT) across a dozen mental health clinics across the US, which tested use of the training tool among 100 behavioral health practitioners and nearly 300 of their patients with a schizophrenia spectrum disorder.
Practitioners have found the training tool easy to use, have called out the helpfulness of the clinical handouts and psychoeducation content, and expressed a desire for more trainings of this nature. Learners also reported an increased sense of confidence after completing the training and the hope that “every clinicians needs to complete this training!”
For systems of care, automated training tools like the CBTp Training Tool will provide a rapid means of scaling and sustaining evidence-based interventions in routine care settings across the US, resulting in more mental health providers across the country providing higher-quality care to individuals with psychosis.
This work was supported by a National Institute of Mental Health (NIMH) Fast-Track Small Business Technology Transfer grant (R42MH123215). For more information about this project, please contact the Principal Investigator, Sarah Kopelovich, PhD, ABPP (skopelov@uw.edu).
Selected References
Kopelovich, S. L., Slevin, R., Brian, R. M., Shepard, V., Baldwin, S. A., Ben-Zeev, D., Tanana, M., & Imel, Z. (2025). Preliminary investigation of an artificial intelligence-based cognitive behavioral therapy training tool. Psychotherapy, 62(1), 12–21. https://doi.org/10.1037/pst0000550
Kopelovich, S. L., Brian, R. M., Tanana, M., Slevin, R., Pace, B., Stewart, S. K., Shepard, V., Ben-Zeev, D., Baldwin, S. A., Soma, C. S., Stanco, S., & Imel, Z. (2025). Development and validation of a cognitive behavioral therapy for psychosis online training with automated feedback. Psychotherapy, 62(1), 1–11. https://doi.org/10.1037/pst0000548
Zilcha-Mano, S. (2025). A glance into the future of artificial intelligence-enhanced scalable personalized training: A response to Kopelovich, Brian, et al. (2025) and Kopelovich, Slevin, et al. (2025). Psychotherapy, 62(1), 22-27.