{
  "abstract": "Introduction Artificial Intelligence (AI) technology in healthcare is expanding rapidly as a tool in both clinical diagnostic and educational settings. While its role in communication training is growing, its potential in simulation-based Emergency Medicine (EM) education remains underexplored. Simulation is central to EM training, but it is time and resource intensive to design. This study aimed to investigate whether a large language model (LLM) can generate EM simulation scenarios of comparable quality to established cases.Methods Ten peer-reviewed EM simulation cases were selected from an open-access database. Two experienced simulation educators independently scored each case using a modified version of the Simulation Scenario Quality Instrument (SSQI), adapted to a 24-item Yes/No checklist. Case summaries were then provided to ChatGPT (LLM-based AI) with a standardised prompt to generate full EM scenarios for a doctor with four years’ EM experience. Each AI-generated scenario was required to include three clinical progression steps, stepwise vital signs, pre- and post-resuscitation ABGs/VBGs, and relevant bloods, ECGs and/or imaging. AI-generated cases were scored using the same checklist and compared with the original scenarios.Results Original EM simulation cases achieved an average completeness score of 79.6%, compared with 55.8% for AI-generated scenarios. Both formats scored 100% for core clinical flow elements, including learning objectives, arrival vitals, and progression realism. AI scenarios frequently omitted contextual details such as location, equipment lists, EMS history, scenario timing, and conditional progression logic. Additionally, AI cases described ECG findings but did not provide full tracings.Abstract #249 Figure 1Completeness scores for original versus AI-generated emergency medicine cases, assessed using a modified simulation scenario quality instrument (SSQI)Conclusion AI can produce clinically coherent EM simulation scenarios with strong coverage of key clinical elements. However, omissions in contextual detail and branching progression limit their current educational value. Improved prompt design and structured asset requests may enhance the completeness and validity of AI-generated scenarios. *presenting author",
  "authors": [
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "E Lait"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "W Al Attas"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "J Burton"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "P Kane"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "E Tivnan"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "M Bourke"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "D Breslin"
    },
    {
      "affiliations": [
        "Mater Misericordiae University Hospital, Dublin"
      ],
      "name": "G O’Connor"
    }
  ],
  "title": "#249 Artificial intelligence (AI) generated simulation cases in emergency medicine – a pilot study of validity using a modified version of the simulation scenario quality instrument (SSQI)",
  "uid": "57e0dad3-5a31-5920-b976-e166a7e1e85d"
}