Syndromic surveillance systems are essential for early detection of disease outbreaks and emerging health threats \cite{Mandl2004SyndromicGuide}. Traditional manual curation of clinical data is time-consuming and resource-intensive, creating critical delays in public health response. Large Language Models (LLMs) offer a promising solution to \textbf{accelerate medical data processing while maintaining accuracy}. However, deploying LLMs in healthcare settings requires addressing challenges of reliability, interpretability, and statistical validity. Active Learning (AL) frameworks can optimize the annotation process by strategically selecting the most informative samples for human review, reducing labeling effort while improving model performance. This work addresses these challenges by combining LLM-based classification with conformal prediction and active learning to create a robust, scalable system for syndromic surveillance in real-world clinical settings.

Poster