ICU Mortality Prediction using Machine Learning
1. Clinical Problem
ICU mortality prediction can support triage, staffing, and resource allocation.
2. Clinical & Product Use Case
- User: ICU clinicians and care teams
- When: Early in ICU admission
- How: Risk score surfaced in an ICU dashboard or EHR widget
3. Dataset
- Source: MIMIC (ICU EHR dataset)
- Data: structured features such as vitals, labs, demographics, diagnosis codes
4. Modeling Approach
- Preprocess structured features from ICU admissions
- Train ML classification models (e.g., logistic regression / tree-based)
- Evaluate using AUC, precision/recall, calibration
5. Results & Insights
Summarize key performance and learnings here.
6. Integration & Safety Considerations
- Decision support, not automation
- Prospective validation & calibration
- UX design to avoid alarm fatigue
7. Future Roadmap
- Compare model families
- Add temporal features
- Incorporate clinician feedback