Quantifying ACO Revenue Leakage Through Healthcare Network Graph Analysis
This project analyzes Medicare specialty referral patterns using network graph theory to:
- Identify key opinion leaders in healthcare referral networks
- Quantify revenue leakage from out-of-network referrals
- Prioritize partnership opportunities for Accountable Care Organizations (ACOs)
- Visualize care coordination pathways
Key Results:
- 93-specialty network analyzed (536K providers, 9.6M service records)
- $18.4B annual revenue leakage quantified (52.8% out-of-network rate)
- Ophthalmology and Dermatology identified as $7.6B partnership opportunity
cms-physician-network-analysis/
├── data/
│ ├── Medicare_Physician_..._2023.csv # Raw CMS data (not in repo)
│ ├── specialty_summary.csv # Aggregated specialty metrics
│ ├── specialty_network.graphml # Network graph structure
│ ├── centrality_analysis.csv # Node centrality scores
│ └── leakage_analysis_summary.csv # ACO leakage metrics
├── notebooks/
│ ├── 01_data_exploration.ipynb
│ ├── 02_specialty_network_construction.ipynb
│ ├── 03_network_analysis.ipynb
│ └── 04_visualization_dashboard.ipynb
├── src/ # Modular Python source files
├── visualizations/
│ ├── network_graph.html # Force-directed network
│ ├── influence_ranking.html # Top specialties bar chart
│ ├── leakage_sankey.html # Referral flow diagram
│ └── opportunity_matrix.html # Partnership prioritization
├── docs/
│ └── Executive_Summary.md # Full case study report
├── .gitattributes
├── requirements.txt
└── README.md
- Python 3.10+
- pip or conda package manager
# Clone repository
git clone https://github.com/SaeMind/cms-physician-network-analysis.git
cd cms-physician-network-analysis
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt- Download CMS Medicare Provider Utilization dataset from data.cms.gov
- Place CSV file in
data/directory - Update filepath in notebook
02_specialty_network_construction.ipynb
jupyter notebook
# Execute notebooks in sequence: 01 → 02 → 03 → 04- 93 specialties representing Primary Care, Specialists, Surgical, Diagnostic, and Hospital-Based categories
- 147 referral pathways based on empirical care patterns
- Network density: 0.0172 (sparse, realistic healthcare referral structure)
| Rank | Specialty | Influence Score |
|---|---|---|
| 1 | Diagnostic Radiology | 0.500 |
| 2 | Diagnostic Pathology | 0.460 |
| 3 | Surgical Oncology | 0.395 |
- 52.8% referral leakage rate
- $18.4B estimated annual revenue loss
- Primary Care drives 86% of leakage
| Specialty | Revenue Opportunity |
|---|---|
| Ophthalmology | $4.3B |
| Dermatology | $3.3B |
| Urology | $1.9B |
- 10% stratified sample (966K records) from 9.6M total
- Specialty standardization and categorization
- Provider-level aggregation to specialty metrics
- Nodes: Medical specialties (filtered for minimum 10 providers)
- Edges: Directed referral pathways based on care patterns
- Weights: Estimated referral volume (service correlation × 20% referral rate)
- PageRank: Overall network importance
- Betweenness Centrality: Care coordination roles
- Degree Centrality: Direct referral connections
- Composite Influence Score: Weighted combination of metrics
- Simulated 10-specialty ACO network
- Calculated in-network vs. out-of-network referral volume
- Revenue loss estimation ($2,500 per referral)
All visualizations are interactive HTML files rendered with Plotly:
- Network Graph: Force-directed layout — node size = influence score, node color = specialty category
- Influence Ranking: Horizontal bar chart of top 15 specialties by composite influence score
- Leakage Sankey: Flow diagram showing ACO source → external specialty destination
- Opportunity Matrix: Scatter plot of network influence vs. revenue opportunity
| Component | Technology |
|---|---|
| Language | Python 3.10+ |
| Graph Analysis | NetworkX 3.0+ |
| Data Manipulation | Pandas, NumPy |
| Visualization | Plotly, Matplotlib, Seaborn |
| Notebooks | Jupyter |
| Network Storage | GraphML format |
For ACO Executives: Quantifies revenue leakage with precision, providing an objective framework for partnership prioritization and data-driven network expansion decisions.
For Clinical Operations: Identifies high-leakage referral sources, maps care coordination pathways, and supports EMR referral management optimization.
For Strategic Planning: Competitive intelligence on specialty referral patterns, ROI projections for practice acquisitions, and network gap analysis.
- Physician-level analysis (if CMS shared patient data available)
- Geographic clustering (state and MSA-level patterns)
- Temporal trends (multi-year longitudinal analysis)
- Predictive modeling (ML for referral behavior forecasting)
- Cost integration (episode-based cost data for ROI precision)
Andrew Lee Clinical Data Science | Healthcare Network Analytics
- Data Source: Centers for Medicare & Medicaid Services (CMS)
- Inspiration: ACO network optimization challenges in value-based care
- Tools: NetworkX community for graph analysis algorithms
MIT License — see LICENSE file for details
Lee, A. (2026). Medicare Specialty Referral Network Analysis.
GitHub repository: https://github.com/SaeMind/cms-physician-network-analysis