F. Akkawi, M. Schumar, S. Kakade, S. Popuri
Northwestern University,
United States
Keywords: GenAI, Clinical, AI in Healthcare, LLMs
Summary:
This presentation introduces a novel solution leveraging generative AI methodologies to create synthetic clinical data in FHIR format across multiple specialties and Diagnosis-Related Groups (DRGs). Our approach addresses the persistent challenge of data access in healthcare's highly regulated environment with proprietary software systems. Initial results demonstrate promising outcomes, with synthetic datasets showing strong relevance to predefined generation criteria and close alignment with real-world evidence. We comprehensively evaluated various Large Language Models (LLMs), assessing their accuracy, performance, cost-effectiveness, and environmental impact in synthetic data generation tasks. The solution represents a robust alternative to traditional data access methods, enabling AI innovation in clinical workflows while maintaining compliance with regulatory requirements. We will present our methodology and evaluation metrics and discuss practical implementation strategies for healthcare organizations seeking to accelerate AI adoption without compromising data security or patient privacy. Authors: Faisal Akkawi- F-akkawi@northwestern.edu Mary Schumar- mary.schumar@u.northwestern.edu Sunilkumar Kakade- sunilkumar.kakade@northwestern.edu Srilakshmi Popuri- srilakshmi.popuri@northwestern.edu (Presenting Author)