Sustaining therapist motivational interviewing skills with a feedback system augmented by machine learning

Following initial training, Motivational Interviewing (MI) skills tend to decline over time without ongoing feedback or coaching. However, providing ongoing support is challenging, as practitioners have had to rely on time- and cost-intensive approaches, such as human-coded fidelity feedback, to assess therapist skills. One promising solution to resource limitations is the development of machine learning systems that automatically evaluate therapy sessions and provide therapists with customized feedback.

This study examined temporal trends in therapist MI skill use in a setting that integrated human and machine-learning-based feedback, using a retrospective observational design. The analysis evaluated 5210 recorded therapy sessions for MI fidelity across 15 therapists in a publicly funded behavioral health setting using multilevel growth curve models.

Model results indicated small but significant increases in ratings of empathy, collaboration, the reflection-to-question ratio, and MI proficiency over time, with no change in trend for the percentage of open-ended questions. The total number of recorded sessions and years of experience were not consistently associated with the use of MI skills over time.

Study findings suggest that a combined machine-learning and human feedback-and-coaching model may support the sustained use of therapeutic skills over time. These results highlight the potential of machine learning systems to supplement and support human-based feedback and coaching systems.