Reproduce “MIMIC-\RNum{4}-Ext-22MCTS: A 22 Millions-Event Temporal Clinical Time-Series Dataset with Relative Timestamp for Risk Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jing Wang, Xing Niu, Tong Zhang, Jie Shen, Juyong Kim, Jeremy C. Weiss
MIMIC-IV-Ext-22MCTS is a 22 million-event temporal clinical time-series dataset, created by applying an LLM-based framework to unstructured discharge summaries from MIMIC-IV-Note. Models fine-tuned on this resource showed improved performance, including up to a 10% accuracy increase in medical question answering and up to a 3% gain in clinical trial matching.
Key insights
- Combining contextual retrieval with specifically designed LLM prompts offers a robust and scalable method for extracting detailed, temporally-aware clinical events from complex, unstructured clinical text while mitigating hallucination risks.
- Anchoring all extracted clinical event timestamps to a major reference event, such as hospital admission, provides a standardized and interpretable temporal context essential for understanding patient trajectories and disease progression.
- Clinical datasets enriched with fine-grained temporal information significantly enhance the performance of deep learning models in various downstream healthcare applications, demonstrating the value of explicit temporal context.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.