Hidden Markov model detection of interpersonal interaction dynamics in predicting patient depression improvement in psychotherapy: Proof-of-concept study

Previous human ethology studies have demonstrated that the interpersonal interactions displayed in therapy by both patients and therapists influences a patient's depression improvement. Pairing novel statistical techniques such as the hidden Markov model (HMM), interpersonal interaction dynamics can be uncovered by partitioning time into empirically-derived nonverbal behavioral states. This approach allows for better patient-therapist behavioral dynamics distinctions in predicting depression improvement and, subsequently, for the processes behind depression improvement. For the 39 participating patients, the first 15 min of the first or second therapy session was recorded on video to examine the interpersonal interaction behaviors of patients and therapists. The video recordings were encoded for vocalization, looking and leg movement behavior events at a 1 s frequency. A Bayesian multivariate multilevel HMM was fitted on the behavioral event data. It is demonstrated that patients that show improvement in the depression score are characterized by interpersonal interaction dynamics of hyperfocus when listening to their therapist in psychotherapy when compared to non-improving patients. The data supports evidence for the emergence of differences in interpersonal interaction dynamics through changed durations of the patient hyper focused listening states, but not through changed state-switching dynamics over time. Due to our relatively small sample size we could not fit multilevel HMMs composed of more than three hidden states. We suggest that applying HMMs will aid human ethological behavior studies in uncovering interpersonal interaction dynamics that occur in therapy and be able to use these dynamics to predict patient depression symptom improvement.