5_communication_strategy

🏥 Asclepios Treatment Intelligence

AI-Powered Substance Use Treatment Prediction App

Built with Streamlit, Plotly, Pandas, and Joblib

---

📌 Overview

The Asclepios Treatment Intelligence app is an interactive Streamlit application designed to provide AI-driven predictions for substance-use treatment patients.

The system loads pre-trained ML models (High-Risk Prediction + Length-of-Stay Predictors) and generates:

• High-risk admission likelihood
• Detox length-of-stay (LOS) prediction
• Rehab length-of-stay (LOS) prediction
• Patient risk insights & feature-engineered scores
• Interactive UI with professional medical-themed styling

This application is ideal for clinical researchers, treatment facilities, and data-science practitioners exploring applied predictive healthcare analytics.

---

🚀 Key Features

🔮 Predictive Modeling

• High-risk treatment prediction
• Estimated Length-of-Stay (Detox + Rehab)
• Advanced feature engineering:
• Years using substance
• SDOH Score
• Risk synergy indicators
• Acuity scoring
• Legal mandate flag
• MAT (methadone) maintenance flag

🎨 Custom UI/UX

• Fully customized Streamlit theme
• Styled headers, cards, tabs, buttons
• Sidebar navigation
• Risk badges & visual cues
• Gradient backgrounds and shadow effects

---

🧠 Model Architecture

The application loads saved models via:

../models/model_high_risk.pkl
../models/asclepios_los_detox.pkl
../models/asclepios_los_rehab.pkl
../models/model_features_high_risk.pkl
../models/model_features_los.pkl

---

📁 Project Structure

5_communication_strategy/
|
|── Streamlit/
│     ├── app.py                    # Streamlit app
│── models/
│     ├── model_high_risk.pkl
│     ├── asclepios_los_detox.pkl
│     ├── asclepios_los_rehab.pkl
│     ├── model_features_high_risk.pkl
│     └── model_features_los.pkl
│── README.md

---

🧩 How It Works

1️⃣ Model Loading

The app uses @st.cache_resource to efficiently load trained models:

models = {
    "high risk": joblib.load("../models/model_high_risk.pkl"),
    "detox los": joblib.load("../models/asclepios_los_detox.pkl"),
    "rehab los": joblib.load("../models/asclepios_los_rehab.pkl"),
    "features high risk": joblib.load("../models/model_features_high_risk.pkl"),
    "features los": joblib.load("../models/model_features_los.pkl"),
}

2️⃣ Feature Engineering

The app builds clinical-grade engineered features, including:

• Years_Using_Substance
• SDOH_Score
• Risk synergy features
• Acuity scoring
• MAT maintenance
• Legal mandate flag

These are dynamically computed from user inputs.

3️⃣ Prediction Workflow

The render_prediction_interface() function:

• Collects patient inputs via Streamlit UI
• Generates patient profile badges
• Builds a prediction dataframe
• Runs ML models
• Displays results with styled metrics and insight cards

---

🖼️ User Interface Modules

🎯 Patient Predictions

Main interactive page:

• Demographics
• Clinical input
• Substance history
• Social factors
• Prediction results

📊 Model Insights

displays analytic charts or model information

📚 About

Project description and usage details.

---

🛠️ Installation & Run

1. Install dependencies

pip install streamlit pandas plotly joblib

2. Run the app

streamlit run app.py

Ensure the models/ directory exists and contains the .pkl model files.

---

🧪 Model Inputs Used

Demographics

• Age
• Sex
• Race
• Education
• Marital Status

Clinical Factors

• Treatment Service Type
• Psychiatric Problems
• Methadone Usage
• Primary Substance + Route + Frequency

Substance History

• Age First Use
• Secondary/Tertiary Substances
• Prior Treatments
• DSM Criteria

Social Determinants

• Employment
• Living Conditions
• Arrests
• Referral Source
• State FIPS
• Insurance

---

🎨 Custom Styling

The app uses 200+ lines of handcrafted CSS to transform Streamlit UI components:

• Cards
• Buttons
• Tabs
• Metrics
• Badges
• Progress bars
• Sidebar
• Input fields

Brand colors used:

#764e7e
#1A2A4F
#F87B1B
#F7A5A5

---

📜 License

This project is for research and educational purposes.
Modify freely but ensure dataset and model usage comply with privacy and ethical guidelines.

---

👨‍💻 Author

Bhang (Wuor Bhang)
Rafaa Ali
Caesar Ghazi
Moe Alwathiq

MIT Emerging Talent • Data Scientist
AI/ML Healthcare & Treatment Analytics Researcher