MediAgent

Autonomous Clinical Decision Support System

---

Overview

MediAgent is an end-to-end autonomous clinical decision support system that predicts hospital readmission risk for diabetic patients using a 4-agent AI pipeline. It combines machine learning, explainable AI, and large language models to generate physician-ready PDF reports — bridging the gap between data science and clinical practice.

Built on 101,766 real-world patient records from 130 US hospitals (1999-2008).

---

Problem Statement

Hospital readmissions within 30 days cost the US healthcare system $26 billion annually. Early identification of high-risk patients enables targeted interventions that reduce readmissions and improve patient outcomes.

MediAgent addresses this by providing physicians with instant, explainable, AI-generated risk assessments at the point of care.

---

System Architecture

+----------------------------------------------------------+
|                    MEDIAGENT PIPELINE                    |
|                                                          |
|   Patient Data                                           |
|       |                                                  |
|       v                                                  |
|   +---------------+                                      |
|   |  Data Agent   |  Clean, encode, preprocess           |
|   +-------+-------+                                      |
|           |                                              |
|           v                                              |
|   +---------------+                                      |
|   |  Risk Agent   |  XGBoost prediction + SHAP           |
|   +-------+-------+                                      |
|           |                                              |
|           v                                              |
|   +---------------+                                      |
|   |   LLM Agent   |  Groq LLaMA 3.3-70B summary         |
|   +-------+-------+                                      |
|           |                                              |
|           v                                              |
|   +---------------+                                      |
|   | Report Agent  |  Auto-generated physician PDF        |
|   +---------------+                                      |
+----------------------------------------------------------+

---

Key Features

Feature	Description
Multi-Agent Architecture	4 independent agents each with single responsibility
XGBoost Prediction	Trained on 101,766 patient records, 89% accuracy
SHAP Explainability	Top risk factors identified per patient
LLM Clinical Summary	Groq LLaMA 3.3-70B generates physician-ready summaries
Auto PDF Reports	Color-coded, professional reports downloadable instantly
Streamlit Interface	Interactive web app, no technical knowledge required
Secure by Design	API keys in .env, never hardcoded

---

Model Performance

Metric	Value
Dataset Size	101,766 records
Records After Cleaning	99,492
Model Accuracy	89%
Readmission Rate	11.2%
Top Risk Factor	number_inpatient
Training Algorithm	XGBoost (100 estimators)

---

SHAP — Top Risk Factors Identified

1. number_inpatient — Previous inpatient visits
2. time_in_hospital — Length of current stay
3. number_emergency — Emergency visit history
4. number_diagnoses — Complexity of medical profile
5. age — Patient age group

---

Tech Stack

Machine Learning: XGBoost, scikit-learn, SHAP, pandas, NumPy

AI / LLM: Groq API (LLaMA 3.3-70B-Versatile)

Web and Deployment: Streamlit, Hugging Face Spaces

Report Generation: fpdf2

Dev Tools: Python 3.14, Git, VS Code, python-dotenv

---

Project Structure

MediAgent/
├── agents/
│   ├── data_agent.py       # Agent 1: Data cleaning and preprocessing
│   ├── risk_agent.py       # Agent 2: XGBoost prediction + SHAP
│   ├── llm_agent.py        # Agent 3: LLM clinical summary
│   └── report_agent.py     # Agent 4: PDF report generation
├── data/
│   └── clean_data.csv      # Preprocessed dataset
├── models/
│   ├── risk_model.pkl      # Trained XGBoost model
│   └── feature_names.pkl   # Feature alignment
├── utils/
│   └── data_cleaner.py     # Preprocessing pipeline
├── reports/                # Generated PDF reports
├── app.py                  # Streamlit web interface
├── main.py                 # Pipeline orchestrator
├── requirements.txt        # Dependencies
└── .env.example            # Environment variables template

---

Setup and Installation

# 1. Clone the repository
git clone https://github.com/AreeshaM/MediAgent.git
cd MediAgent

# 2. Create virtual environment
python -m venv venv
venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Set up environment variables
# Create .env file and add your GROQ_API_KEY

# 5. Download dataset
# Place diabetic_data.csv in data/ folder
# Dataset: https://www.kaggle.com/datasets/brandao/diabetes

# 6. Train the model
python models/risk_model.py

# 7. Run the app
streamlit run app.py

---

Environment Variables

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_api_key_here

Get your free API key at: https://console.groq.com

---

Author

Areesha Mubeen Data Analyst and ML Engineer BS Computer Systems Engineering — Riphah International University areeshamubeen85@gmail.com GitHub: https://github.com/AreeshaM

---

License

This project is licensed under the MIT License.

---

Disclaimer: MediAgent is an AI-assisted tool intended to support, not replace, clinical judgment. Always consult a qualified physician before making medical decisions.