tcmhwd/bsi-mortality-demo
3
BSI 30-Day Mortality Prediction — Research Demo
⚠ Research Demonstration Only — Not for clinical use.
Reviewer demo for a comparative study of structured ML and RAG-LLM approaches to 30-day mortality prediction in bloodstream infection (BSI).
Modes
Data
- External validation cohort only (Hospital 20, de-identified)
- 10 representative patients spanning all risk bands and both outcomes
- No raw patient data beyond the 10 curated demo cases is included
Models
- TabM (early arm): deep learning tabular model, Optuna-tuned
- XGBoost (AST-informed arm): gradient boosting with microbiological features
- RAG-LLM: OpenAI embeddings + FAISS retrieval + GPT inference (cached)
