Objective Maintaining relevance in a psychodynamic dialogue is a nuanced task, requiring therapists to balance between following patients’ free associations while avoiding less effective interventions. Identifying less effective sequences of talk is especially challenging given the diversity of psychodynamic approaches and methodological barriers to analyzing session discourse. This study introduces a novel approach using the MATRIX coding system, an evidence-based tool, to differentiate content correlated with better session outcomes. Method Transcripts of 367 sessions were coded using the MATRIX. Therapist Out-of-MATRIX utterances, indicating a deviation from core therapeutic focus, were examined for their predictive value. Outcome measures included the next-session alliance and patient functioning scores. Two machine-learning-based models, using the Random Forest algorithm, predicted session-by-session changes in clinical outcomes based on MATRIX codes, and interpreted using the SHapley Additive exPlanations. Results Therapist Out-of-MATRIX utterances accurately predicted next-session changes in alliance and patient functioning scores. Our model also identified an optimal dose-effect relationship for the number of Out-of-MATRIX interventions needed for effective therapy session. Conclusion This study demonstrates the potential of using contemporary research tools to analyze therapeutic discourse, revealing how psychotherapy produces its benefits. Its scope extends beyond prediction, providing practical recommendations on how to enhance therapists’ performance and outcomes.
Read full abstract