Abstract: This study evaluates GPT-4’s effectiveness in annotating medical notes to identify COVID-19 admissions, leveraging a dataset from the Consortium for Clinical Characterization of COVID-19 (4CE). Across multiple institutions and languages, GPT-4 demonstrated a high concordance with clinicians’ validations, showing 77% agreement in a variety of clinical questions. The research highlights GPT-4’s potential in accurately identifying patients for study enrollment, showcasing its proficiency in extracting explicit information and inferring implicit details from medical notes. Despite some challenges in deeper inference tasks, the findings advocate for the promising role of LLMs in streamlining clinical documentation and supporting healthcare research, indicating significant avenues for future advancements in AI applications in medicine.
Bio: Dr. Amelia Li Min Tan, PhD, leads research at Harvard Medical School, focusing on the intersection of biomedical informatics and personalized medicine across various medical topics. Her leadership in global consortia, such as the 4CE, reflects her commitment to transforming health insights into clinical actions. Holding a PhD from the National University of Singapore, Dr. Tan mentors students and drives innovation in healthcare through data science. Her work, featured in numerous publications, demonstrates her dedication to leveraging computational tools for health advancements.