AI-enabled conversational agent increases engagement with cognitive-behavioral therapy: A randomized controlled trial
Key Takeaways
Limbic Care, a generative AI care companion app, is 2.4 times more engaging than digital workbooks delivering CBT content
Limbic Care increases engagement duration (minutes) by 3.8 times more than digital workbooks
Limbic Care is as safe and effective as digital workbooks, with no increase in adverse events over 6 weeks of self-directed treatment
Background
Managing symptoms of mental health conditions can be challenging, especially when patients have limited access to support between therapy sessions or while waiting for treatment. Research shows that engaging with therapeutic activities during these gaps, such as completing homework exercises in cognitive-behavioural therapy (CBT), can significantly enhance treatment outcomes. However, patients often struggle with traditional resources like static worksheets or general online content, which can feel impersonal or difficult to connect with.
To address this, we explored whether our generative AI (genAI) care companion app, Limbic Care, could provide a more interactive and personalised experience. With the advent of large language models (LLMs), Limbic Care is designed to deliver CBT materials through a conversational, adaptable interface, potentially increasing engagement unlike traditional interventions. Our primary goals were to evaluate: 1) whether a genAI-enabled app increases patient engagement with therapeutic activities, 2) its effectiveness in reducing symptoms of anxiety, depression, and sleep disturbances, and improving overall well-being, and 3) how it compares in safety to the standard of care.
Methods
The randomised control trial ran across six weeks among 540 young adults in the United States. Participants were randomly assigned to one of two groups: one used the Limbic Care app, while the other used a static PDF version of the CBT exercises.
Results
Limbic Care app users demonstrated significantly higher engagement than PDF users, with 2.4 times more frequent sessions overall, and 3.8 times total usage duration, underscoring the app’s effectiveness in sustaining user interaction over longer periods of time. Maintaining engagement is essential, as extended interaction can enhance the retention of therapeutic content and allow for closer patient monitoring.
Limbic Care had a comparable clinical effect to standard care demonstrating that a genAI-enabled app can be safely used in clinical settings when combined with domain-specific safety guardrails.
While both groups showed similar symptom reduction with PDFs and the app, there was a significant reduction in anxiety and depression symptoms with the LLM guided sessions feature (particularly with WSAS scores). This feature uses genAI-powered clinical personalisation to tailor exercises to a participant’s specific problems. Most interestingly, the benefits of increased engagement magnified over the study period, which suggests that the usage of genAI-enabled interventions is advantageous for engaging patients over long periods of time on waitlists.
2.4x
More sessions in Limbic Care vs. PDFs
3.8x
Longer usage in Limbic Care vs. PDFs
80%
Complete homework in the app
Conclusions
Our findings suggest that Limbic Care, a clinical AI tool, can safely enhance patient engagement and improve clinical outcomes through clinically personalised interventions. Compared to traditional CBT methods, patients are more engaged with a genAI-enabled care companion app without any adverse effects. This higher engagement may reduce therapy drop-out rates, a persistent challenge in mental health interventions. Moreover, by enabling ongoing, automatic crisis detection during periods without human supervision, such as on waiting lists or between therapy sessions, Limbic Care provides safe, effective support in the absence of a therapist. This demonstrates how clinical AI can elevate care standards and offer an effective, scalable solution for delivering personalised mental healthcare.
