Digital Coach to Support Exposure Therapy Homework for Anxious Youth

Design. Open‐label, within‐participant, exposure‐instance‐level randomized trial embeddedin routine outpatient exposure‐based CBT at Mass General Brigham clinics. The unit ofrandomization is the individual clinician‐assigned, "BraveBot‐eligible" exposure homeworkinstance. Each time a youth opens the link for a BraveBot‐eligible exposure delivered viaSMS reminder, the backend system immediately randomizes that single instance 1:1 toBraveBot‐assisted vs self‐guided completion.Intervention. BraveBot is a scope‐limited, clinician‐supervised, large language model(GPT)‐powered audio‐to‐audio conversational assistant (provided by TheraLoop AI, Inc.)that supports completion of clinician‐assigned exposure homework. For BraveBot‐assistedinstances it clarifies the assignment and rationale, collects Subjective Units ofDistress (SUDS) ratings and expectations, encourages persistence when anxiety rises,prompts about safety behaviors and "white‐knuckling" at mid‐exposure check‐ins, andguides brief end‐of‐exposure reflection. It does not diagnose, generate exposurehierarchies, or make treatment decisions. For self‐guided instances, the youth completesthe same assigned exposure independently and then completes the same post‐exposuresurvey.Aims.Aim 1 (primary): whether BraveBot improves exposure‐homework QUALITY, assessed by fiveyouth‐reported 0‐100 ratings per completed instance (Effort, Perceived Capability,Understanding, Ability to Continue, Perceived Helpfulness).Aim 2 (secondary): whether BraveBot increases exposure‐homework ADHERENCE (probabilitythat a randomized instance is completed).Aims 3‐5 (exploratory): preliminary effects on clinical outcomes (RCADS anxiety, CALIS‐Clife interference, IUSC intolerance of uncertainty) and associations withexposure‐related processes; acceptability/usability/burden among youth, clinicians, andparents; and the performance and burden of BraveBot's automated risk‐detection andclinician‐notification (safety) workflow.Participants and schedule. Up to 40 youth ages 12‐22 with a clinical or subclinicalanxiety or related disorder for which exposure‐based CBT is indicated. BraveBot accessremains active for up to 45 homework‐reminder days (about six weeks), with a 12‐weekbackstop (i.e., 84 calendar days). Standardized self‐report measures are collected atbaseline (pre‐BraveBot), at the end of the active BraveBot period, and at a 3‐weekfollow‐up. A separate population of up to 10 clinicians may voluntarily enroll to use theBraveBot system with their patients and complete surveys about their experience.Analysis. Primary outcomes are analyzed with linear mixed‐effects models (fixed effect ofcondition; random intercepts for youth, and clinician/site as needed), withBenjamini‐Hochberg false‐discovery‐rate correction across the five co‐primary outcomes.Adherence is analyzed with a binomial‐logit generalized linear mixed‐effects model.