All projectsView source
Classification
Patient Readmission Analysis
Unplanned hospital readmissions within 30 days indicate potential gaps in care quality and drive up costs. This project identified risk factors for readmission among diabetic patients.
Top 5 readmission risk factors identified
2025PythonPandasScikit-learnPower BI
Dataset
UCI: Diabetes 130-US Hospitals
Key Questions
- Which clinical factors are most predictive of 30-day readmission?
- Do medication changes at discharge affect readmission risk?
- Can we build a risk score to flag high-risk patients before discharge?
Methods
- Data cleaning and feature engineering from 50+ clinical variables
- Gradient boosted trees and logistic regression comparison
- SHAP value analysis for model interpretability
- Risk score development with operational threshold setting
Results
The top 5 readmission risk factors were: number of inpatient visits in the prior year, number of diagnoses, length of stay, discharge to home (vs. skilled nursing), and whether diabetes medication was changed. The model achieved an AUC of 0.68.
Readmission Rate by Risk Factor
Recommendations
- Flag patients with 3+ prior inpatient visits for enhanced discharge planning
- Ensure medication reconciliation is completed for all patients with medication changes
- Coordinate post-discharge follow-up within 7 days for high-risk patients
- Track readmission rates by discharge disposition to identify systemic gaps
Limitations
The dataset spans 1999-2008, so clinical practices may have evolved significantly. The AUC of 0.68 reflects the inherent difficulty of readmission prediction. Social determinants of health were not available in the dataset.