Abstract:

Clinical simulation is a core component of nursing education, but existing approaches remain limited when the objective is to train free-form communication, clinical reasoning, and structured handover in complex geriatric situations. Manikin-based simulation is well suited to procedural training but provides limited conversational realism, while standardized patients are costly and difficult to scale. Recent advances in large language models (LLMs) create new opportunities for virtual patient interactions. This paper presents CareConvers, a conversational AI-based clinical simulation platform designed to support the training of nursing students in geriatric assessment, adapted communication, and clinical decision-making.

CareConvers is a web-based platform built with a Vue.js frontend, a Node.js/Express backend, Google Gemini as the conversational engine, and PostgreSQL/Supabase for session persistence and interaction logging. The simulation is based on a clinically grounded scenario set in a long-term care facility. The learner, acting as a student nurse, encounters Madame Aubry, an older patient presenting with moderate neurocognitive disorder, bilateral coxarthrosis with acute pain exacerbation, and uncompensated hearing impairment. Through a free-text interaction, the student must introduce themself appropriately, adapt communication to the patient’s cognitive and sensory limitations, identify unusual behavior, conduct a pain assessment using OPQRST, measure vital signs, use the Algoplus behavioral pain scale when self-report is unreliable, decide to escalate the situation, and deliver an ISBAR handover to a supervising nurse.

A major contribution of the platform lies in its hybrid validation architecture. CareConvers combines the flexibility of an LLM with deterministic server-side rules to assess ten learning objectives in real time. The system tracks progress on OPQRST and ISBAR, provides real-time visual indicators, delivers brief formative feedback during the interaction, and generates a structured summative feedback report at the end of the simulation. Additional safeguards were implemented to improve clinical fidelity and response consistency.

The tool is being evaluated with 25 first-year Bachelor of Science in Nursing students through a mixed-methods study informed by the Technology Acceptance Model. Participants complete the simulation individually after a short briefing on the scenario and patient record. After the session, they review AI-generated summative feedback and complete a post-simulation evaluation. The evaluation combines the System Usability Scale with a purpose-designed questionnaire covering clinical realism, pedagogical value, perceived learning gains, usefulness of real-time guidance, and intention to reuse the tool for exam preparation. Open-ended questions collect qualitative feedback on strengths and areas for improvement. Learning analytics extracted from interaction logs include objective completion rate, OPQRST coverage, ISBAR completeness, number of conversational turns, and simulation duration.

Keywords:

Nursing education, remote learning, LLM, evaluation.