Understanding the relationship between patient language and outcomes in internet-enabled cognitive behavioural...

Area of Science:

• Psychology
• Computer Science
• Artificial Intelligence

Background:

• Understanding patient language in psychotherapy is crucial for effective interventions but manual coding is resource-intensive.
• Automating the analysis of patient utterances in text-based internet-enabled Cognitive Behavioural Therapy (iCBT) is needed for large-scale studies.
• Identifying linguistic markers associated with therapeutic success can enhance intervention development.

Purpose of the Study:

• To develop and validate a deep learning model for automatic identification and categorization of patient utterances in iCBT.
• To assess the association between specific patient language patterns (utterances) and clinical outcomes like reliable improvement and engagement.
• To enable scalable, data-driven analysis of therapeutic conversations.

Main Methods:

• Trained a deep learning model on 340 manually annotated transcripts of text-based iCBT sessions.
• Utilized the model to automatically code utterances from approximately 34,000 patient transcripts.
• Employed logistic regression to analyze the relationship between coded utterances and clinical outcomes (reliable improvement, engagement).

Main Results:

• The deep learning model achieved human-level agreement in categorizing three out of five patient utterance types.
• Increased 'change-talk' (indicating movement toward change or self-exploration) was significantly associated with higher odds of reliable improvement and engagement.
• Increased 'counter change-talk' (indicating movement away from change) was associated with lower odds of reliable improvement and engagement.

Conclusions:

• Deep learning offers an effective method for large-scale, automated coding of patient language in digital mental health interventions.
• Patient language, specifically 'change-talk,' is a significant predictor of treatment success in iCBT.
• This automated approach facilitates a deeper, data-driven understanding of therapist-patient interactions and therapeutic processes.