🏥 Predictive Patient Readmission Dashboard

Identifying High-Risk Patients using Machine Learning & SQL

📝 Project Overview

In the era of Value-Based Care, the Hospital Readmissions Reduction Program (HRRP) penalizes hospitals with higher-than-expected 30-day readmission rates. As a clinician, I developed this project to demonstrate how data science can be used for Risk Stratification to prevent these readmissions.

This project builds a predictive pipeline that extracts clinical data from a MySQL database and uses a Random Forest Classifier to predict which patients are at high risk of being readmitted within 30 days of discharge.

---

🚀 The Data Pipeline

1. Data Engineering (database_and_simulation.py)

• Simulation Logic: Developed a synthetic dataset of 1,000 patient encounters.
• Clinical Variables: Includes Age, Primary Diagnosis (CHF, COPD, Sepsis, etc.), Comorbidities, and Length of Stay (LOS).
• Relational Storage: The script automatically initializes a MySQL database (clinical_analytics_db) and handles secure ETL (Extract, Transform, Load) processes.

2. Machine Learning Workflow (readmission_analysis.py)

• Feature Engineering: Utilized One-Hot Encoding for categorical diagnosis data.
• Model Selection: Implemented a Random Forest Classifier to handle non-linear clinical relationships.
• Evaluation Metrics: Focused on AUC-ROC and Recall—crucial metrics in healthcare where missing a high-risk patient is more costly than a false alarm.

---

📊 Model Performance & Insights

The model evaluates factors that drive patient returns. By analyzing the "Feature Importance," the dashboard provides actionable insights for Discharge Planners.

Key Metrics Sample:

Metric	Score	Clinical Interpretation
AUC-ROC	0.78	Good ability to distinguish between high and low-risk patients.
Recall (1)	0.82	Successfully identified 82% of patients who were eventually readmitted.
Top Predictor	Age	Strongest correlation with post-discharge complications.

---

📂 Project Structure

├── .env                       # Environment variables (Credentials)
├── .gitignore                 # Securely hides sensitive files
├── requirements.txt           # Project dependencies
├── database_and_simulation.py # ETL: Populates MySQL database
├── readmission_analysis.py    # ML: Trains model and generates report
└── readmission_features.png   # Visualization of risk drivers

---

⚙️ Installation & Usage

1. Clone this repository and navigate to the project

git clone [https://github.com/azucena-m/predictive-patient-readmission-dashboard.git]

2. Setup Credentials: Create a .env file and add your MySQL details: DB_USER=root DB_PASSWORD=your_password DB_HOST=127.0.0.1 DB_NAME=clinical_analytics_db

3. Install dependencies pip install -r requirements.txt

4. Execute Pipeline

• First, populate the database: python database_and_simulation.py
• Second, run the analysis: python readmission_analysis.py

---

Clinical Relevance

This project addresses the "data-to-bedside" gap by identifying high-risk patients before they leave the hospital, Case Managers can prioritize them for:

• Home Health follow-ups
• Telehealth monitoring
• Post-discharge planning

Author: Azucena Marroquin - Nurse & Programmer Analyst