SCH: Personalized AI-Driven Models for Supporting User Engagement and Adherence in Health Interventions: Validation in Cognitive Behavioral Therapy for Anxiety

Untreated anxiety undermines long-term physical and emotional wellbeing, especially among college students, with rates worsening since the onset of the COVID-19 pandemic. Cognitive Behavioral Therapy (CBT) is the leading evidence-based intervention for anxiety, but many students fail to complete exercises

between CBT sessions, reducing its effectiveness. Socially assistive robots (SARs) help promote adherence to home-based practice in the context of elder care, social skill learning, and physical therapy, but it is unknown how SARs can enhance CBT. The specific objective of this research is to develop

personalized CBT SARs that can support CBT compliance for college students with anxiety. To meet the goals of the proposed work, we will conduct eight collaborative design sessions and three user studies and data collections and evaluations: Specifically, studies will determine how SAR personalization based

on implicit and explicit feedback can help promote greater CBT compliance and anxiety reduction outcomes for students. Specific Aim 1 will develop machine learning models to personalize a CBT SAR with implicit personalization–using only visual and auditory cues and no user input. Specific Aim 2 will

develop machine learning models to enhance SAR engagement based on explicit user feedback–using direct input from the user to change the SAR behaviors. Specific Aim 3a will test the efficacy of personalized CBT SARs on key outcomes of a 6-week CBT for anxiety intervention: robot-student alliance, CBT engagement, CBT adherence, and anxiety symptom reduction. In Study 3a, n=60 students

with anxiety will be randomly assigned to either a CBT SAR that performs implicit personalization (n=30) or a CBT SAR with no personalization (control, n=30). In Aim 3b, a separate sample of n=60 students will be randomly assigned to either complete a 6-week CBT SAR intervention that performs explicit

personalization (n=30) or a CBT SAR with no personalization (control, n=30). We predict that implicit and explicit CBT SAR personalization will enhance pre- versus post-intervention SAR-user alliance, engagement in CBT, and lower anxiety outcomes over the course of a 6-week daily CBT home-based intervention for anxiety compared to the non-personalized control CBT SAR.

RELEVANCE (See instructions): The proposed research is relevant to public health, as it will assess whether personalized SARs impact engagement and outcomes in CBT exercises for anxiety, which is key to developing effective, scalable treatments for mood disorders such as anxiety. This research aligns with the NIMH mission of leveraging

novel methods to intuitively and intelligently collect, sense, connect, analyze and interpret data from individuals, devices and systems to enable discovery and optimize health.