{
  "abstract": "Background Structured genomic information in combination with other multimodal data can provide a holistic view of a patient’s cancer, thus helping in guiding targeted therapies. Genomic test reports present unstructured data that includes results based on methodologies like Next Generation Sequencing (NGS), and classification of variants made by clinical scientists based on established guidelines and evidence (e.g., pathogenic or of uncertain significance). Additionally, these reports offer a clinical interpretation that provides a comprehensive explanation of the findings.Method In this study, we investigate on utilising generative AI models in a responsible way, specifically Large Language Models (LLMs) that are compact and lightweight, often referred to as ‘Small Language Models’ for extracting the information into a standardised format for better interpretability and analysis. First, we use a targeted, evidence-based search to identify relevant information within the reports, focusing on specific details like results from next-generation sequencing (NGS) panels. Second, we feed this information into an open-source clinical LLM, grounding it with iteratively improved guidelines to ensure it generates a standard, structured output for a particular target. We discard any results that don’t meet our output requirements. Finally, we perform a quality check on the output using logic-based rules. Any data flagged as potentially inaccurate is then verified by a human expert to ensure its validity.Conclusion Based on our study, we can conclude that (1) evidence-based search provides the right input and ensures the computational overhead is reduced, (2) choosing the right LLM ensures that proper guidelines can produce high-quality outputs and (3) LLMs have the capability to identify information that is not always reported in a semi-structured format (e.g., duplication in gene detected by a panel).Acknowledgements for Funding or Support This activity is part of a collaborative working agreement between Great Ormond Street Hospital NHS Foundation Trust and Roche Products Ltd. M-GB-00025149 | December 2025",
  "authors": [
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Sebin Sebu"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Nick James"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Jaskaran Kawatra"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Avish Vijayaraghavan"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Shiren Patel"
    },
    {
      "affiliations": [
        "Roche"
      ],
      "name": "Rebecca Pope"
    },
    {
      "affiliations": [
        "Roche"
      ],
      "name": "Alexandros Zenonos"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Neil Sebire"
    },
    {
      "affiliations": [
        "Great Ormond Street Hospital for Children NHS Foundation Trust, UK"
      ],
      "name": "Pavithra Rajendran"
    }
  ],
  "title": "58 How trustworthy are clinically trained Small Language Models for extracting genomic information from reports?",
  "uid": "7e518697-1755-59d3-85ad-1375a9022327"
}