Patient Risk Stratification System
Impact87% prediction accuracy
Accuracy87%
Readmissions Cut22%
Patients Scored50K+
Overview
Developed a machine learning-based risk stratification system for a healthcare provider to predict patient readmission likelihood within 30 days. The system helps care coordinators prioritize high-risk patients for proactive intervention.
The Challenge
The healthcare provider had high readmission rates leading to penalties and poor patient outcomes. They needed early identification of at-risk patients.
The Solution
Built an ensemble ML model combining gradient boosting and logistic regression, integrated with their EHR system. Created interpretable risk scores with feature explanations for clinicians.
Key Results
87% accuracy in predicting 30-day readmissions
22% reduction in actual readmissions
Deployed for 50K+ patients annually
Tech Stack
PythonScikit-learnXGBoostSHAPPostgreSQLFastAPIFHIR API