Evaluation of a Multiagent LLM-Based Simulated Patient to Train Therapists in Written Exposure Therapy for PTSD

Project Summary There is a tremendous gap between the need for posttraumatic stress disorder (PTSD) treatments and access to care.Written Exposure Therapy (WET) is an efficient PTSD treatment developed to improve treatment access. There is high demand for capacity-building for WET, but the most common strategy to train therapists–having them attend a workshop–is insufficient on its own to achieve good clinical and implementation outcomes and the recommended approach of adding case consultation is not feasible for many systems due to lack of access to consultants, demands on provider time, and cost. Use of simulated patients (SPs) to an avenue for low-stakes experiential learning with timely feedback, but their use in psychotherapy has been limited by significant practical barriers. The goal of the proposed exploratory project is to refine and test a therapist-facing generative artificial intelligence (AI)-based Simulated Interactive Training tool (SIT) to learn WET. The SIT includes a diverse set of virtual Simulated Patients, along with a Simulated Consultant. The combination gives therapists opportunities to practice WET, receive real-time feedback, and correct behaviors via clinical scenarios of varying difficulty. We will partner with health systems and training programs that do not have capacity or funding for WET consultation to conduct a randomized controlled trial (RCT) comparing a traditional training workshop + SIT (W+SIT) to common existing training paradigms: workshop alone (W) and workshop + consultation (W+C), among a sample of 90 therapists and 270 patients. Aim 1: To evaluate the impact of SIT therapist training on patient PTSD, depression, and functioning. Aim 2: To test the impact of SIT training on therapist treatment fidelity, reach ( number/proportion of eligible patients treated with WET), and cultural responsiveness. Aim 3: to test the impact of SIT on therapist mechanisms of change including self-efficacy, perceptions of WET, and intentions to use WET. Examine active learning through the use of SIT as a way to engage these mechanisms. Aim 4: to examine Readiness Evaluation for AI Deployment and Implementation (READI) criteria and conduct comparisons of safety and privacy implications, equity, engagement, and potential for implementation (including costs)..