A linear programming model for preserving privacy when disclosing patient spatial information for secondary purposes

Area of Science:

• Health Informatics
• Geospatial Data Analysis
• Privacy-Preserving Techniques

Background:

• A linear programming (LP) model was developed for de-identifying patient spatial data using the HIPAA Expert Determination method.
• The model aimed to maximize spatial detail while minimizing re-identification risk.
• A limitation was identified for areas with small populations, leading to improbable results.

Purpose of the Study:

• To extend the existing LP model to effectively handle small population areas.
• To ensure de-identified spatial datasets maintain a very small risk of re-identification.
• To improve the accuracy and applicability of spatial data de-identification methods.

Main Methods:

• An extended linear programming (LP) model was developed to accommodate locations with smaller populations.
• The model calculates transition probabilities for randomizing original patient locations to new ones.
• The extended model was designed to maintain a very low risk of re-identification.

Main Results:

• The enhanced LP model was applied to a dataset of 11,740 postal codes in Ottawa, Canada.
• The study demonstrated the limitations of the previous LP model with small population areas.
• The extended model successfully de-identified the entire dataset, overcoming previous limitations.

Conclusions:

• The enhanced LP model effectively de-identifies geospatial information in small population areas with minimal postal code distortion.
• The model adheres to privacy regulations like the HIPAA Privacy Rule.
• Future extensions can incorporate additional data like age and gender for comprehensive de-identification.