TER076

Synopsis

This study has two aims. First, it will explore how musculoskeletal physiotherapists use—or would like to use—chat-based language models to support assessment, diagnosis, and treatment of whiplash. A single 90-minute in-person focus group with a stratified sample of five to six clinicians will surface real-world prompts, scenarios where models help or fall short, patient specificity in queries, and strategies for handling unreliable responses. Draft prompts derived from this discussion will then be emailed back to participants for feedback and refinement. Second, the finalized prompts (approximately 25–30 items ranging from simple yes/no checks to complex case scenarios) will be submitted to leading language models. Two independent physiotherapists will score each response as either “guideline-concordant or not contradictory” versus “contradictory,” with a third rater resolving any disagreements. Expected Outcomes: The project will yield a validated library of clinician-authored prompts organized by assessment, diagnosis, and treatment tasks; qualitative insights into factors driving physiotherapist engagement with language models; and quantitative measures of model alignment with 2024 whiplash guidelines. Findings will guide best practices for safe, effective integration of conversational language models into musculoskeletal care.