Mining Time-Series Data from Clinical Reports
Using LLMs to reconstruct temporal sepsis trajectories from text
This research introduces a novel approach that extracts temporally-structured clinical data from unstructured text reports using Large Language Models.
Key Innovations:
- Automated reconstruction of detailed sepsis progression timelines from clinical narratives
- Creation of a new temporal clinical data corpus without manual annotation
- LLM-based pipeline for medical phenotyping, extraction, and temporal annotation
- Transformation of retrospective summaries into structured time-series data
Business Impact: This approach addresses a critical healthcare data challenge by leveraging abundant narrative reports to create structured temporal datasets that can improve clinical decision support, predictive modeling, and training data availability for medical AI systems.