Imputation of confidential data sets with spatial locations using disease mapping models

Area of Science:

• Epidemiology
• Data Privacy
• Geographic Information Systems (GIS)

Background:

• Public use data files with fine geographic information (e.g., census tracts) pose disclosure risks.
• Identifying individuals is a concern for malicious data users when precise locations are available.
• Balancing data utility for spatial analysis with privacy protection is crucial.

Purpose of the Study:

• To propose and evaluate a method for releasing spatially explicit data while mitigating privacy risks.
• To enable spatial analyses using simulated geographies derived from attribute data.
• To assess the trade-offs between disclosure risk reduction and analytic validity.

Main Methods:

• Fitting disease mapping models to predict areal-level counts based on file attributes.
• Sampling new geographic locations using estimated models to create synthetic geographies.
• Applying the method to North Carolina mortality data for causes of death.
• Evaluating disclosure risks and analytic validity of the synthetic geographic data.

Main Results:

• The proposed method generates synthetic geographies that allow for spatial analysis.
• Evaluations indicate a reduction in disclosure risks compared to releasing fine geographic data.
• The analytic validity of spatial analyses performed on synthetic geographies is assessed.
• The approach demonstrates feasibility using real-world cause-of-death data.

Conclusions:

• Simulated geographies offer a viable solution for releasing sensitive spatial data.
• This method enhances data accessibility for spatial epidemiology and public health research.
• It provides a framework for balancing data utility and privacy in geographic data release.