A deep learning model has achieved human-level proficiency in three of five exercises, accurately categorizing verbal expressions from more than 33,000 talk-therapy patients who underwent their sessions online.
The model was developed and tested by researchers at Ieso Digital Health, a U.K.-based provider of internet-enabled cognitive behavioral therapy (CBT).
Michael Ewbank, PhD, and colleagues had their work published online July 3 in Psychotherapy Research.
The trained their algorithms on 340 manually annotated session transcripts. Using the system to automatically categorize patients’ utterances into one or more of five specified textual categories, they found it to be as good as humans at identifying talk in three.
“Understanding patient responses to psychotherapy is important in developing effective interventions,” the authors write in outlining their study objectives. “However, coding patient language is a resource-intensive exercise and difficult to perform at scale.”
Additionally, the team found associations between the patients’ language and their clinical outcomes.
The application of deep learning-facilitated automatic annotation “provides an effective means of obtaining categorization of patient utterances during text-based internet-enabled CBT, at a scale previously beyond the scope of psychotherapy research, providing evidence of both positive and negative associations between patient utterances categories and outcomes,” the authors conclude.
“Coupled with an automated understanding of therapist language,” they add, “deep learning can be used to enable a data-driven understanding of the relationship between therapeutic interventions, patient language and clinical outcomes.”
The study is available in full for free.