New Ieso Digital Health study uses deep learning to reveal the features of psychotherapy that are related to patient improvement
27 Aug 2019
- First of its kind research paper uses a deep learning model to automatically identify the content of therapist language during a psychotherapy session
- Provides previously unavailable insights into what aspects of psychotherapy are most effective
- Valuable knowledge extracted from 90,000 hours of Internet-Enabled Cognitive Behavioural Therapy (ie-CBT) transcripts, that is impossible for humans to do alone
- Findings confirm that using CBT methods is associated with patient improvement
- Conversely, the study found non-CBT related conversation is negatively correlated with improvement (e.g. more small talk means patients are less likely to get better)
- This represents a first step towards practical, quality controlled mental healthcare, with huge potential for monitoring therapy to improve clinical practice in CBT
Ieso Digital Health, a leader in internet-enabled cognitive behavioural therapy (ie-CBT), today announced the publication of new research which applied deep learning to large-scale clinical data to understand what aspects of psychotherapy content are associated with clinical outcomes. The paper ‘Quantifying the association between psychotherapy content and clinical outcomes using deep learning’, is published in JAMA Psychiatry, a peer-reviewed and highly-cited journal globally with more than 4.8 million article views and downloads annually.
Today 1 in 4 adults  experience at least one diagnosable mental health problem in any one year, but compared with the treatment of physical conditions, the quality of care of mental health disorders remains poor, as does the rate of improvement in treatment. Unlike other medical treatments, psychotherapy is comprised of a series of one-to-one discussions, which means there is a lack of systematic methods for measuring the treatment delivered. However, with ie-CBT, a patient communicates with a therapist using real-time instant messaging which means conversations can be captured as transcripts.
With unique access to 90,000 hours of anonymised recorded therapy transcripts from its ie-CBT platform, Ieso has trained a deep learning model (a type of machine learning), to automatically recognise the content of the language used by therapists during patient CBT sessions. The Ieso research team then used this model to measure the treatment delivered to determine which features are associated with an improvement in patient symptoms.
Having been impossible for humans to analyse such a large data set, the research provided valuable insights into the relationship between therapy content and clinical outcomes that have previously been unavailable. The findings showed that when treatment contained a greater quantity of CBT change methods, patients are more likely to show an improvement in symptoms. Patients were less likely to improve when sessions had increased quantity of ‘non-therapy’ content (i.e. conversations not related to treatment).
Michael Ewbank, a senior scientist at Ieso and lead author on the paper, commented: “With our deep learning model, we can extract knowledge accumulated across thousands of hours of CBT in a way that would be impossible for a human to do. What is exciting about this study is that it demonstrates the potential of Ieso’s data set, where we can understand more about what the active ingredients of therapy are, what works for whom, and develop new and more effective treatments for mental health disorders. Our work represents a first step towards a practicable approach for quality controlled behavioural health care with the goal of improving the efficacy of psychotherapy.”
Access the full research paper, ‘Quantifying the association between psychotherapy content and clinical outcomes using deep learning’ at JAMA Psychiatry.