Blood Horizon Predictor

Overview

Blood Horizon Predictor is a production-grade machine learning system deployed at Mayo Clinic to forecast blood utilization, reducing inventory waste to <1% and saving $600K+ annually. This full-stack healthcare analytics platform serves 500+ clinicians with real-time predictions, integrating multiple BigQuery data pipelines and cloud infrastructure.

Problem Statement

Blood product management is a critical challenge in healthcare:

• High waste costs: Expired blood products cost hospitals hundreds of thousands annually
• Supply chain complexity: Balancing inventory across multiple blood groups and product types
• Patient safety: Ensuring adequate supply for surgeries while minimizing waste
• Manual processes: Clinicians spending hours on manual assessment and forecasting

Solution Architecture

Machine Learning Pipeline

1. Data Integration
   • Integrated 3 BigQuery data pipelines for blood groups, inventory levels, and surgery schedules
   • Processed historical transfusion data across 7+ years
   • Feature engineering for temporal patterns, seasonal trends, and surgical demand
2. Predictive Models
   • Developed ensemble ML models for multi-horizon forecasting (1-day, 3-day, 7-day)
   • Implemented time series forecasting with ARIMA, Prophet, and LSTM architectures
   • Achieved <1% inventory waste through accurate demand prediction
3. Deployment Infrastructure
   • Built Flask application for model serving and API endpoints
   • Containerized with Docker for reproducible deployments
   • Deployed on Google Cloud Run for auto-scaling and high availability
   • Real-time predictions with <200ms latency

Full-Stack Dashboard

• Frontend: Interactive React-based dashboard with real-time data visualization
• Backend: Flask REST API integrating BigQuery, Cloud Storage, and ML models
• Database: BigQuery for data warehousing and analytics
• Cloud Platform: Google Cloud (Cloud Run, BigQuery, Cloud Storage, Vertex AI)

Key Features

✅ Real-time Forecasting: Predict blood demand 1-7 days in advance
✅ Multi-Product Support: Forecast across all blood groups (A+, A-, B+, B-, AB+, AB-, O+, O-)
✅ Surgical Integration: Incorporate upcoming surgery schedules for demand spikes
✅ Inventory Optimization: Automated alerts for reordering and waste prevention
✅ Clinician Dashboard: User-friendly interface for 500+ healthcare professionals
✅ Cloud-Native: Scalable, secure, and HIPAA-compliant infrastructure

Impact & Results

Technical Stack

Machine Learning: Python, Scikit-learn, TensorFlow, Prophet, ARIMA, LSTM
Backend: Flask, FastAPI, SQLAlchemy
Frontend: React, D3.js, Plotly
Cloud Infrastructure: Google Cloud Run, BigQuery, Cloud Storage, Vertex AI
DevOps: Docker, Kubernetes, CI/CD with GitHub Actions
Data Pipeline: Apache Airflow, BigQuery, Pandas

Workflow Optimization

Beyond prediction accuracy, the system reduced manual assessment time by >75% through:

• Automated data aggregation from multiple hospital systems
• Smart alerts for low inventory and expiration warnings
• Integration with existing Electronic Health Record (EHR) systems
• Mobile-responsive dashboard for on-the-go decision making

Collaboration

Worked closely with:

• Pathologists to understand blood usage patterns and clinical workflows
• IT Infrastructure Teams for secure cloud deployment
• Hospital Administrators for cost-benefit analysis and ROI tracking
• Clinicians for user experience testing and iterative improvements

Future Enhancements

• Integration with national blood bank networks for regional supply optimization
• Predictive models for rare blood types and emergency scenarios
• Mobile app for push notifications and on-call decision support
• Expansion to other Mayo Clinic locations nationwide

Status: In Production (May 2025 - Present)
Organization: Mayo Clinic, Rochester, MN
Role: Data Science Intern - ML Engineer