Asclepios AI is a data-driven exploration into how machine learning, domain knowledge, and thoughtful analysis can help improve outcomes for individuals undergoing Substance Use Disorder (SUD) treatment. Using real admissions and treatment episode data, our goal is to identify patterns that can help facilities personalize treatment duration, improve completion rates, and reduce relapse/readmission.
This project is being developed by a collaborative team:
- Caesar Ghazi (@CaesarGhazi)
- Moe Alwathiq (Moe-phantom)
- Rafaa Ali (@RafaaAli)
- Wuor Bhang (@WuorBhang)
How accurately can patient demographics, history, and treatment factors predict an optimal treatment duration that minimizes relapse risk in Substance Use Disorder treatment?
Despite the critical importance of completing adequate treatment for sustained recovery, SUD facilities lack systematic approaches to personalizing treatment length based on patient characteristics and predicted outcomes. Approximately 30% of patients are readmitted within one year, suggesting many individuals either receive insufficient treatment or discontinue prematurely.
At the same time, treatment facilities struggle with capacity planning. They often cannot effectively match the timing and volume of new admissions with available beds and staff. This mismatch leads to under-used treatment slots, prolonged waitlists, and reduced access to care.
Asclepios AI aims to use data-driven insights to address these challenges.
We're creating a smart system that helps treatment centers:
- Predict how long each patient needs treatment for the best chance of recovery
- Reduce the number of patients who return after treatment (readmissions)
- Make the most of their available resources (beds, staff, time)
Think of it like a personalized recommendation system - just like how Netflix recommends movies you might like, our system recommends the optimal treatment plan for each patient based on their unique situation.
Substance use disorders affect millions of people worldwide and cost society billions of dollars each year. Unfortunately:
- About 30% of patients return to treatment within a year
- Treatment facilities struggle to match the right treatment length to each patient
- Resources are often wasted or underused due to poor planning
Our AI system aims to solve these problems by using data to make better predictions about what each patient needs.
We use publicly available SUD admissions and treatment episode datasets, which include variables such as:
- Patient demographics
- Substance use patterns
- Co-occurring issues
- Treatment history
- Treatment length
- Completion status
- Readmission / relapse factors
- 📁 Raw Data:
1_datasets/raw/ - 📁 Processed Data:
1_datasets/processed/ - 📁 Sample Data:
1_datasets/sample/ - 📄 Data Documentation:
1_datasets/README.md
This step includes:
- Cleaning and formatting raw datasets
- Handling missing or inconsistent data
- Encoding categorical variables
- Normalizing and transforming features
- Structuring datasets for modeling
📁 Folder:
2_data_preparation/
We explore:
- Relationships between treatment factors and outcomes
- Duration patterns across substance types
- Demographic and behavioral trends
- Variables most correlated with relapse or completion
- Facility-level patterns affecting patient success
📁 Folder:
3_data_exploration/
We will build models to:
- Predict relapse likelihood
- Identify high-risk patients
- Predict facility resource demand
- Evaluate model fairness and robustness
Analysis includes:
- Feature engineering
- Training and validation
- Model comparison
- Interpretability
📁 Folder:
4_data_analysis/
Our communication will include:
- Insight summaries
- Visual dashboards or graphs
- Interpretations for clinicians and decision-makers
- Recommendations based on model findings
- Limitations and ethical considerations
📁 Folder:
5_communication_strategy/
Our final deliverables will recap the entire analysis with:
- A structured summary of findings
- Visualizations
- Model results
- Recommendations for treatment facilities
- Reflections and next steps
📁 Folder:
6_final_presentation/
Below is the full layout of the project repository:
Asclepios_Ai/
│
├── 0_domain_study/
├── 1_datasets/
│ ├── processed/
│ ├── sample/
│ └── raw/
├── 2_data_preparation/
├── 3_data_exploration/
├── 4_data_analysis/
├── 5_communication_strategy/
└── 6_final_presentation/
This project is licensed under the MIT License. 📄 View License