Using spatiotemporal models to generate synthetic data for public use.

Area of Science:

• Public Health Data Science
• Biostatistics
• Geospatial Epidemiology

Background:

• Public-use data release faces disclosure risks, especially in disease mapping with small counts.
• Integrating statistical disclosure limitation with small area estimation and disease mapping is underexplored.
• Small counts in disease mapping present inferential challenges and privacy concerns.

Purpose of the Study:

• To propose and evaluate spatiotemporal data analysis for generating privacy-preserving synthetic public-use data.
• To address the intersection of disease mapping, small area estimation, and statistical disclosure limitation.
• To demonstrate the utility of synthetic data for analyzing health disparities.

Main Methods:

• Utilized a Bayesian spatiotemporal model to analyze ten years of county-level heart disease death counts across multiple age-groups.
• The model accounts for spatial, temporal, and inter-age-group dependencies.
• Generated synthetic data from the posterior predictive distribution of the fitted model.

Main Results:

• Demonstrated the synthetic data's capacity to preserve key data dependencies (spatial, temporal, age-group).
• Confirmed the privacy-preserving attributes of the generated synthetic datasets.
• Showcased comparable estimates of urban/rural disparities between synthetic and suppressed real data.

Conclusions:

• Spatiotemporal Bayesian modeling offers a robust method for generating synthetic disease mapping data.
• Synthetic data can effectively balance data utility and privacy protection for public release.
• This approach facilitates the analysis of health disparities without compromising individual privacy.