Artificial intelligence, the ability of a computer program or a machine to think and learn, is increasingly being used in many areas, including mental health treatment. A new study uses artificial intelligence to measure psychotherapy treatment and to examine links between specific aspects of the therapy and outcomes. This information could potentially lead to improvements in psychotherapy.

Because psychotherapy involves a series of private discussions, it is challenging to measure the content of a therapy discussion. The study authors sought to address the lack of systematic methods for measuring psychotherapy and to better understand the impact of specific components of psychotherapy.

They reviewed transcripts of nearly 90,000 hours of therapy from almost 15,000 patients receiving internet-enabled cognitive behavioral therapy (CBT) for a mental health disorder between 2012 and 2018. During internet-enabled CBT, patients communicate with a therapist using a real-time text-based message system. Internet-enabled CBT has been shown to be effective for treatment of depression.

The researchers, led by Michael P. Ewbank, Ph.D., with the University of Cambridge, U.K., developed a method to objectively quantify psychotherapy. They defined 24 categories of psychotherapy features. A psychologist then tagged the therapist words from transcripts of nearly 300 therapy sessions based on those categories. They then trained a computer model (a deep learning model, a type of artificial intelligence) to automatically categorize other therapy session text.

They found that increased use of some specific session features was associated with more likely improvements in symptoms and greater patient engagement. For example, use of a greater number of change methods—cognitive and behavioral techniques used in CBT—was associated with greater improvement in symptoms. Change methods include such things as teaching coping skills; reattribution (a technique that challenges automatic thoughts and assumptions by considering alternative causes of events); conceptualization (process where the therapist and client work collaboratively to describe and explain the client’s issues); and psychoeducation (teaching families and patients about their psychiatric disease).

Other therapy components associated with improved symptoms included discussion related to reviewing homework, setting goals at the first therapy session and regularly setting an agenda for therapy (the therapist and patient deciding on topics for discussion during therapy). Homework is used in CBT to practice skills learned in therapy and to apply them more generally in daily life.

A greater quantity of non-therapy-related content was associated with lower likelihood of symptom improvement and engagement.

The authors conclude that “the application of deep learning to large clinical data sets can provide valuable insights into psychotherapy, informing the development of new treatments and helping standardize clinical practice.”