Regionalization with Self-Organizing Maps for Sharing Higher Resolution Protected Health Information

Area of Science:

• Geographic Information Systems (GIS)
• Health Informatics
• Spatial Analysis

Background:

• Sharing Protected Health Information (PHI) at finer scales presents privacy challenges.
• Existing regionalization methods may not optimally balance data utility and patient privacy.
• HIPAA guidelines necessitate minimum population thresholds for sharing PHI.

Purpose of the Study:

• To compare four regionalization approaches for creating HIPAA-compliant geographical aggregations.
• To assess the fitness of max-p-regions, REDCAP, and their Self-Organizing Map (SOM) variants for analysis and display.
• To identify methods that maximize spatial units and intra-unit homogeneity while meeting privacy requirements.

Main Methods:

• Implemented max-p-regions, REDCAP, and SOM variants for geographical regionalization.
• Aligned regional configurations with census boundaries and optimized for homogeneity.
• Evaluated regional configurations using measures of model-fit, compactness, homogeneity, and resolution.

Main Results:

• The SOM procedure significantly improved assessment measures when applied to max-p-regions (MSOM).
• Applying SOM to REDCAP generally degraded assessment measures.
• REDCAP outperformed other methods on most evaluation metrics.

Conclusions:

• REDCAP is recommended for its superior performance in balancing privacy and analytical utility.
• The SOM variant of max-p-regions (MSOM) offers high resolution with suitable performance.
• Understanding the impact of SOM on top-down vs. bottom-up regionalization is crucial.