๐ค AI Summary
This study addresses the scarcity of authentic nurseโpatient dialogue data in emergency triage research, which has largely relied on structured electronic health records. To bridge this gap, the authors propose the first framework capable of generating medically plausible, multimodal triage conversations from structured clinical notes. By incorporating role-specific settings and strategic prompting mechanisms, the system explicitly controls linguistic disfluencies and decision-making behaviors reflective of real-world triage scenarios. The pipeline integrates natural language generation, speech synthesis, and automatic speech recognition, complemented by human evaluation to establish an end-to-end dialogue simulation and validation workflow. Approximately 800 dialogues and corresponding audio recordings were generated, demonstrating high consistency in triage urgency levels across text, ASR transcripts, and audio modalities, thereby validating both the efficacy and multimodal coherence of the proposed approach.
๐ Abstract
Research in emergency triage is restricted to structured electronic health records (EHR) due to regulatory constraints on nurse-patient interactions. We introduce TriageSim, a simulation framework for generating persona-conditioned triage conversations from structured records. TriageSim enables multi-turn nurse-patient interactions with explicit control over disfluency and decision behaviour, producing a corpus of ~800 synthetic transcripts and corresponding audio. We use a combination of automated analysis for linguistic, behavioural and acoustic fidelity alongside manual evaluation for medical fidelity using a random subset of 50 conversations. The utility of the generated corpus is examined via conversational triage classification. We observe modest agreement for acuity levels across three modalities: generated synthetic text, ASR transcripts, and direct audio inputs. The code, persona schemata and triage policy prompts for TriageSim will be available upon acceptance.