A method for managing re-identification risk from small geographic areas in Canada

Area of Science:

• Health data privacy
• Statistical disclosure control
• Geographic information systems

Background:

• Common practice involves suppressing or aggregating data from small geographic areas to protect privacy.
• The uniqueness criterion suggests an area is "small" if unique individuals (based on quasi-identifiers) approach a certain threshold.
• Existing thresholds for health data uniqueness range from 0% to 20%.

Purpose of the Study:

• To develop and validate predictive models for determining when geographic areas exceed specific uniqueness thresholds (5% and 20%).
• To provide data custodians with guidance on selecting appropriate uniqueness thresholds based on data characteristics and recipient capabilities.
• To manage the risk of re-identification in health datasets originating from small geographic areas.

Main Methods:

• Estimated uniqueness for urban Forward Sortation Areas (FSAs) using Canadian census data (20% population).
• Constructed logistic regression models to predict uniqueness above 5% and 20% thresholds.
• Validated model accuracy using 10-fold cross-validation, with population size and quasi-identifier equivalence classes as predictors.

Main Results:

• Models demonstrated high prediction accuracy (specificity > 0.9) and significant parameters.
• Sensitivity was 0.87 for the 5% threshold and 0.74 for the 20% threshold.
• Higher thresholds identified fewer records as "small," reducing disclosure control actions; guidance provided for threshold selection.

Conclusions:

• Developed models effectively manage re-identification risks associated with small geographic areas.
• Offers flexibility for data custodians to define "small geographic areas" based on data type and recipient.
• Enables tailored disclosure control strategies balancing privacy protection and data utility.