ESCALATE: A Dataset for Safety-Critical Clinical Escalation Conversations

ESCALATE is a structured dataset of synthetic clinical escalation-of-care conversations generated using role-locked multi-agent large language models. The dataset contains paired conversations comparing unstructured and ISBAR-structured communication during acute patient deterioration scenarios. Each conversation includes:- Multi-turn dialogue between clinical roles (nurse, registrar, optional nurse-in-charge)- Ground-truth case card data- Structured plan outputs- Safety-critical evaluation labels Evaluation labels are generated using an AI-based structured rubric assessing:- Communication completeness- Safety-critical omissions- Hallucinations/inventions- Escalation appropriateness- Actionability and closed-loop communication The dataset is designed for research in:- Clinical communication- Patient safety- AI evaluation and benchmarking- Multi-agent LLM systems This release accompanies ongoing research into AI-driven clinical simulation and structured communication frameworks. 🔗 Related Work This dataset is associated with prior research exploring the use of multi-agent large language models for simulating clinical communication: Power, D., & Power, T. (2026). Can Large Language Models Generate Role-Consistent Clinical Dialogue for Education? A Multi-agent Approach. Under Review. Preprint available at EdArViX: https://doi.org/10.35542/osf.io/etv6d_v The ESCALATE dataset extends this work by providing a structured dataset of escalation-of-care conversations, enabling systematic evaluation of communication quality, safety-critical omissions, and structured handover frameworks such as ISBAR.