Project Summary: Heart Failure Readmission & Resource Optimization

Predict readmission risk and forecast staffing resources to improve outcomes and reduce costs.

We built a healthcare-ready pipeline using MIMIC-IV to cohort heart failure patients, extract features, and train classification models (Logistic Regression, Random Forest, XGBoost) for 30-day readmission. A downstream regression estimates nursing hours by care level and length of stay to support staffing plans.

Highlights

• Clear cohorting and preprocessing strategy for reproducibility.
• Calibrated classification with AUROC/AUPRC tracking; regression uses MAE/RMSE.
• Operational framing with personas, KPIs, and pilot-to-scale roadmap.

Explore

• Overview: Risk-Based Clinical Decision Support
• Case study structure: Problem, approach, architecture, operational considerations, and organizational impact.

References

• Repository: cyranothebard/heart_failure_readmission
• Slide deck: Google Slides

This post summarizes key takeaways; see linked pages for details.