Bridging the gap: a practical step-by-step approach to warrant safe implementation of large language models in healthcare
Jessica D. Workum, Davy van de Sande, Diederik Gommers, Michel E. van Genderen
Abstract
Large Language Models (LLMs) offer considerable potential to enhance various aspects of healthcare, from aiding with administrative tasks to clinical decision support. However, despite the growing use of LLMs in healthcare, a critical gap persists in clear, actionable guidelines available to healthcare organizations and providers to ensure their responsible and safe implementation. In this paper, we propose a practical step-by-step approach to bridge this gap and support healthcare organizations and providers in warranting the responsible and safe implementation of LLMs into healthcare. The recommendations in this manuscript include protecting patient privacy, adapting models to healthcare-specific needs, adjusting hyperparameters appropriately, ensuring proper medical prompt engineering, distinguishing between clinical decision support (CDS) and non-CDS applications, systematically evaluating LLM outputs using a structured approach, and implementing a solid model governance structure. We furthermore propose the ACUTE mnemonic; a structured approach for assessing LLM responses based on Accuracy, Consistency, semantically Unaltered outputs, Traceability, and Ethical considerations. Together, these recommendations aim to provide healthcare organizations and providers with a clear pathway for the responsible and safe implementation of LLMs into clinical practice.