JMIR Publications

ABSTRACT

Background:

Graduate speech-language pathology (SLP) education must prepare learners to integrate reasoning, assessment, intervention, counseling, communication, and documentation in dynamic encounters. Yet clinical placements provide inconsistent exposure, while conventional simulations may be difficult to scale or overly constrained. Generative AI enables more interactive, repeatable training. CaseyMD is an end-to-end multiagent AI clinical simulation platform designed for broader health professions education, with SLP and aphasia serving as an initial testbed for complex, communication-intensive workflows.

Objective:

This pilot study evaluated the feasibility and acceptability of CaseyMD for graduate SLP clinical training using aphasia-focused simulations as a clinical testbed and explored preliminary patterns in learner performance and perceived preparedness.

Methods:

A single-arm pilot feasibility study was conducted with graduate SLP students enrolled in coursework addressing adult neurogenic communication disorders. Participants remotely completed instructor-assigned simulations spanning case history, assessment, treatment planning, treatment delivery, counseling, and documentation. CaseyMD used dynamically responsive patient, caregiver, and clinical supervisor agents, captured learners’ unconstrained interactions and clinical work products, and generated structured post-session feedback that was available for human clinical instructor review. Feasibility was evaluated through participant retention, simulation completion, technical functioning, and engagement. Outcomes included clinical self-efficacy, perceived readiness for clinical placement, simulation design ratings, clinical instructor evaluations of performance, and qualitative analysis of platform-generated performance feedback.

Results:

Eight graduate SLP students enrolled, and 7 completed all study procedures, yielding a retention rate of 87.5%. All 7 completers completed each of the 5 assigned simulations, resulting in 35 of 35 prescribed sessions completed (100%). The median recorded simulation time was 112.6 minutes per session, and minor technical issues were resolved without preventing completion. Mean clinical self-efficacy increased from 8.17 (SD 0.97) before the intervention to 8.46 (SD 0.55) after the intervention, with 4 of 7 participants (57.1%) showing overall increase. Mean clinical placement readiness increased from 3.62 (SD 0.47) to 3.90 (SD 0.39), with 6 of 7 participants (85.7%) improving overall. The mean Simulation Design Scale rating was 3.75 (SD 0.64) of 5, with Feedback/Guided Reflection (mean 4.18, SD 0.66) and Problem Solving (mean 4.06, SD 0.61) rated highest. Clinical instructors rated overall performance at a mean of 4.00 (SD 0.82) of 5, with interpersonal skills receiving the highest rating (mean 4.43, SD 0.79). Qualitative analysis identified recurring strengths in counseling, caregiver education, and patient-centered communication, while systematic assessment, treatment implementation, and objective documentation remained areas for development.

Conclusions:

CaseyMD demonstrated feasible, acceptable end-to-end clinical simulation using coordinated generative AI agents, enabling learners to engage in dynamic clinical conversations while preserving auditable workflows and human clinician authority over interpretation and feedback. Larger controlled studies should evaluate effectiveness and broader applications.