How can AI be used to strengthen the therapeutic alliance in psychotherapy?
Thursday May 22, 2025 13:20 - 13:40 F1
Speaker: Henrik Jahren
Presenters: Karin Hammerfald, Fabian Schmidt
Track: Future Health and Care
Therapeutic alliance is a well-established predictor of treatment outcomes, with microcounseling skills playing a critical role in its development. These skills enable therapists to build rapport and communicate effectively using techniques such as reflective listening, open-ended questions, and affirmations. Regular feedback, particularly when immediate, is essential for therapists to refine these skills and facilitate therapeutic change. However, traditional methods for coding therapist behaviors are often resource-intensive and costly. This study explored the potential of Large Language Models (LLMs), specifically fine-tuned models, to automatically identify microcounseling skills in psychotherapy sessions.
The approach involved fine-tuning GPT-4o-mini on psychotherapy transcripts annotated by human experts. The process included transcript preprocessing, manual annotation, segmentation into dialogue chunks, and training on the annotated data. The model was designed to classify therapist statements, provide explanations for its classifications, and suggest alternative responses.
The results were encouraging, with the model achieving strong performance metrics (Accuracy: 0.74, Precision: 0.81, Recall: 0.74, F1: 0.75, Cohen’s κ: 0.65). It excelled at identifying common and linguistically distinct counseling skills but struggled with more subtle skills requiring an understanding of implicit relational dynamics.
These findings underscore the potential of LLMs to revolutionize clinical practice by providing scalable, automated feedback on therapist skills. While limitations remain, the study demonstrates the feasibility of using digital tools to enhance psychotherapy effectiveness, supporting the broader objectives of digital transformation in mental health care.
Topic
Future Health and Care
Seminar type
Live + On site
Lecture type
Presentation
Objective of lecture
Tools for implementation
Level of knowledge
Introductory
Target audience
Researchers
Students
Healthcare professionals
Patient/user organizations
Keyword
Benefits/effects
Education (verification)
Patient centration
Innovation/research
Test/validation
Lecturers
Henrik Jahren Speaker
Chief Psychologist & Founder
Braive AS
Henrik Haaland Jahren is a Norwegian, Australia-trained Clinical Psychologist and entrepreneur, co-founder and Chief Psychologist of Braive.
Braive is a digital mental health platform for mental health treatment to support both patients and clinicians in their work to improve mental health. Together with his wife, Braive's CEO Hermine Bonde Jahren, they established Braive with the mission to improve the access to and quality of mental health care worldwide.
Henrik's dedication to integrating technology with mental health care and through the ongoing collaboration with leading research institutions from the two disciplines, including Department of Psychology at University of Oslo and Department of Computer science at KTH, has positioned Braive at the forefront of the digital health sector, striving to make high-quality mental health care accessible on a global scale.
Karin Hammerfald Presenter
Postdoctoral Researcher
University of Oslo
Karin is a postdoctoral researcher at the University of Oslo, contributing to the interdisciplinary project ALEC2 that integrates clinical psychology, computer science, and industry. With over 20 years of experience in psychiatric institutions, academia, and the health-tech sector, she has a strong background in evidence-based psychotherapy, clinical supervision, and quality management in mental health care. Her current work focuses on developing personalized diagnostics, monitoring tools, and tailored feedback systems for both patients and clinicians by leveraging machine learning to enhance internet-based cognitive-behavioral therapy (iCBT).
Fabian Schmidt Presenter
M.Sc.
Kungliga Tekniska Högskolan, KTH