Healthcare AI with Readmission Risk Prediction

• 

  Michael Lee Question: Why is this patient considered high risk?

• 

  Michael Lee Question: Should this patient be prioritized for immediate follow-up?

• 

  Michael Lee Question: Should this patient be prioritized for immediate follow-up?

• 

  Michael Lee Question: What interventions would you recommend to reduce readmission risk?

• 

  Sarah Johnson Q: Why is this patient considered high risk? consult with myexterbot

• 

  Sarah Johnson Q: Why is this patient considered high risk? consult with myexterbot

• 

  Sarah Johnson Q: What interventions would you recommend to reduce readmission risk?

• 

  Sarah Johnson Q: What interventions would you recommend to reduce readmission risk?

• 

  Sarah Johnson Q: What is the 30-day readmission risk for this patient? (Metrics lost within API connection mishap)

• 

  Sarah Johnson Q: What is the 30-day readmission risk for this patient?

• 

  Terminal Honcho Output with JsonRPC with the Question: What is the 30-day readmission risk for this patient?

• 

  deployment proof

Inspiration

Hospitals struggle to predict which patients are likely to be readmitted within 30 days. We wanted to explore how AI agents can use structured medical data to support early risk detection and improve clinical decision-making.

What it does

This project is an AI healthcare agent that analyzes patient FHIR data and predicts 30-day readmission risk as LOW, MEDIUM, or HIGH, with clear reasoning based on medical conditions, medications, and vital signs.

How I built it

I built an A2A-compatible agent using Python and the Google ADK framework. It connects to a FHIR R4 server, retrieves patient data (conditions, medications, observations), and applies a rule-based risk scoring model to generate predictions.

Challenges I ran into

I faced issues with A2A message formatting, API key authentication, and FHIR context injection between external agents and the main system. Debugging tool communication was a key challenge.

Accomplishments that im proud of

A2A agent communication FHIR healthcare data access Automated risk scoring with explanation output Multi-tool agent workflow

What I learned

I learned how to build multi-agent systems, handle structured healthcare data with FHIR, and manage real-world integration issues between external AI agents and APIs.

What's next for Healthcare AI with Readmission Risk Prediction

Next, I plan to improve the model using machine learning, add real-time hospital data streaming, and enhance explainability using more advanced clinical reasoning models.

Built With

• a2a
• de-identified
• fastapi
• gadk
• gemini
• httpx
• json-rpc
• litellm
• ngrok
• patient
• python