Designing and Testing a Large Language Model-Based Cognitive Processing Therapy Worksheet Helper

Stanford University, Stanford CA

Investigators

Abstract

Project Summary The goal of the proposed exploratory project is to refine and test a generative artificial intelligence (AI)-based Cognitive Processing Therapy (CPT) Worksheet Helper. The development phase follows a user-centered design approach to generate qualitative (e.g., user experience) and quantitative (e.g., feasibility, usability) feedback about iterative versions of the AI Worksheet Helper from patients (n=10), therapists (n=10), and CPT experts (n=3). In addition to refining the tool itself so that it can engage CPT mechanisms, feedback will be elicited about how the AI Worksheet Helper can be most effectively integrated into CPT to support homework. CPT+AI Worksheet Helper will then be compared to CPT+CPT Coach in 50 individuals with PTSD using a pilot effectiveness randomized controlled trial (RCT). Aim 1 (Discover/Design/Build): Develop a generative AI-based Worksheet Helper using a user-centered design approach that collects and iteratively integrates qualitative and quantitative feedback from patients, therapists, and CPT experts. Aim 1a Discover/Design: Develop and refine the AI Worksheet Helper via iterative patient feedback about the tool’s functionality, design, and usefulness via focus groups and in-depth individual user-design sessions. Aim 1b Discover/Design: Gather qualitative and quantitative therapist feedback about the AI Worksheet Helper’s functionality, design, and requirements to seamlessly integrate into and potentially enhance CPT’s effectiveness. Aim 1c Build: Integrate user feedback and work collaboratively with CPT experts to ensure the AI Worksheet Helper facilitates core therapeutic components. Aim 2 (Test): Evaluate the feasibility and acceptability of the AI Worksheet Helper in the context of treatment as part of a pilot RCT design (N=50). Aim 2a: Evaluate enrolled CPT + AI Worksheet Helper patients’ (n=25) satisfaction with the tool during and between sessions, and the perceived helpfulness for learning skills and reducing PTSD symptoms using a mixed-method approach. Aim 2b: Qualitatively and quantitatively evaluate study (routine care) therapists’ perceived usefulness of the tool and its ability to facilitate engagement, skill development, and worksheet completion. Aim 3 (Test): Determine the AI Worksheet Helper’s initial effectiveness for improving treatment completion and engage previously hypothesized treatment mechanisms (i.e., trauma cognition changes, homework completion, and self-efficacy) using a pilot RCT design (N=50). Aim 3a (Outcomes): Evaluate whether the CPT + AI Worksheet Helper produces greater treatment completion, session attendance. Aim 3b (Mechanisms): Determine whether the tool can facilitate active learning, and produce greater behavioral intention, homework completion and quality, trauma cognition, and self-efficacy change than CPT+CPT Coach. Aim 3c (Predictors): Test whether AI Worksheet Helper engagement is associated with outcomes and mechanisms engagement. Data will support a larger trial to determine the AI Worksheet Helper's ability to engage treatment mechanisms and reduce symptoms.