JMIR Publications

ABSTRACT

Background:

The high prevalence of mental illness represents a serious societal crisis. One factor aggravating this crisis is the existence of barriers that limit access to mental health care services. To mitigate this problem, cognitive behavior therapy has been automated using virtual agents to extend the availability of such services. A pivotal process of cognitive behavior therapy is the correction of automatic thoughts. In this process, a therapist uses Socratic questioning to guide patients in exploring alternative perspectives and correcting their thoughts. The effectiveness of this process hinges on the therapist's ability to adapt the number and nature of questions based on the patient's changing mood and distress level. Superficial or inadequate correction of an automatic thought might reduce the therapy’s effectiveness. Since existing virtual agents lack this adaptive questioning capability, their effectiveness is limited. This situation raises the need for virtual agents that can automatically detect psychological distress and adapt question selection to an individual patient's psychological distress level.

Objective:

This study develops and evaluates a method that automatically adapts the number of Socratic questions based on the level of detected psychological distress during a cognitive behavior therapy session with a virtual agent. We hypothesize that this adaptive approach to question selection will lower psychological distress, reduce negative emotional states, and produce more substantial cognitive change compared to a random number of questions.

Methods:

In this study, which envisions healthcare support in daily life, we recruited participants from 18 to 65 for an experiment that involved two different conditions: 1) a virtual agent that asked a number of questions to prompt automatic thought correction based on psychological distress detection; or 2) a virtual agent that just asked a random number of questions. The participants were assigned to one of the two groups and experienced a single cognitive behavior therapy session with a virtual agent and completed questionnaires before and after

Results:

Participants completed the experiment. There were slight differences in sex, age, or pre-experiment psychological distress levels between the two groups. The adapted number of questions group showed a significantly greater reduction of psychological distress than the random number of questions group. We also found that there was a significant difference in cognitive change when the number of questions closely aligned with the model's estimate compared to when the number deviated from it.

Conclusions:

Adapting question selection based on the level of psychological distress improved the effectiveness of cognitive behavior therapy provided by virtual agents.