Mental Health Counseling From Conversational Content With Transformer-Based Machine Learning

Area of Science:

• Digital Health
• Machine Learning in Healthcare
• Behavioral Health

Background:

• Asynchronous text-based counseling is a growing, accessible behavioral health care modality.
• Assessing process and content in text-based therapy at scale remains a challenge.

Purpose of the Study:

• To employ machine learning (ML) for evaluating clinical content and client-reported outcomes in a large cohort of text-based counseling sessions.
• To identify associations between therapeutic interventions, content, and client outcomes in digital mental health.

Main Methods:

• A quality improvement study analyzed deidentified text messages from asynchronous text-based counseling provided via a mobile app between 2014-2019.
• A transformer-based deep learning model automatically categorized message content and interventions.
• Mixed-effects regressions assessed the association between interventions, content, and client outcomes (engagement, satisfaction, PHQ-8 symptom change).

Main Results:

• The study included 166,644 clients and 20,600,274 messages, with participants predominantly female, aged 26-35, and White.
• ML analysis revealed significant variability in therapist interventions and content.
• Therapeutic interventions and content were associated with client engagement (R=0.43), satisfaction (R=0.46), and symptom improvement (PHQ-8 change, R=0.13).

Conclusions:

• Supportive counseling interventions were associated with improved client outcomes, mirroring findings in traditional therapy.
• Machine learning-based content evaluation offers a scalable and specific method for psychotherapy research.
• This approach can enhance the understanding and quality of digital mental health services.