Alluvial (Sankey) visualization comparing observed diabetes prescribing patterns in NHANES 2017-2018 against both 2017 and 2025 ADA Standards of Care. Patients are classified by comorbidity-based risk profile using a hierarchical schema, then linked to their observed medication regimen. Flow color indicates guideline concordance: green for concordant treatment, red for treatment gaps, gray for indeterminate.
The ADA Standards of Care have evolved substantially between 2017 and 2025, with current guidelines emphasizing SGLT2 inhibitors and GLP-1 receptor agonists for patients with specific cardiometabolic risk profiles (heart failure, CKD, ASCVD, obesity). This analysis evaluates the same NHANES cohort against both guideline vintages to disentangle two sources of treatment gaps: (1) gaps that existed even against contemporary 2017 standards, and (2) gaps attributable to the subsequent evolution of evidence-based recommendations. The result is a more nuanced picture of where implementation efforts should focus — not just "practice doesn't match guidelines," but which gaps reflect systemic undertreatment versus retrospective guideline shifts.
- 2025 guideline concordance: ~6.6% of patients with guideline-specified indications were receiving recommended therapy
- 2017 guideline concordance: ~39.1% — substantial gaps existed even against the standards of the time
- The largest treatment gaps appear in patients with heart failure and CKD, where SGLT2 inhibitors are now first-line but were rarely prescribed in 2017-2018
The analysis produces two Sankey diagrams:
2025 guidelines applied to 2017-2018 practice — Hierarchical classification (Heart Failure > CKD > ASCVD > Obesity > No major comorbidities) maps patients to the most clinically significant condition with a specific drug class recommendation.
2017 guidelines applied to 2017-2018 practice — Simpler classification reflecting the fewer condition-specific drug class recommendations available at the time (ASCVD, Marked Hyperglycemia, No major comorbidities).
Data source: NHANES 2017-2018 (publicly available from CDC)
Language: R
Key packages: ggalluvial, ggplot2, haven, tidyverse, gt
Sample: 815 adults with diagnosed diabetes after exclusions for missing data
# 1. Clone repository
git clone https://github.com/coby-dulitzki/ada-guideline-care-cascade.git
cd ada-guideline-care-cascade
# 2. Download NHANES data (see data/README.md for direct links)
# 3. Run analysis
Rscript analysis/diabetes_guideline_concordance.R
Rscript analysis/variable_mapping_table.Rada-guideline-care-cascade/
├── README.md
├── .gitignore
├── analysis/
│ ├── diabetes_guideline_concordance.R # Main analysis and Sankey figures
│ └── variable_mapping_table.R # NHANES variable mapping table
├── figures/ # Output visualizations
│ ├── sankey_2025_guidelines.png
│ ├── sankey_2017_guidelines.png
│ └── nhanes_variable_mapping.png
└── data/
└── README.md # NHANES download instructions
Coby Dulitzki, MS — Doctoral Researcher, Health Systems & Policy, OHSU-PSU School of Public Health
Analysis: C. Dulitzki, 2025