k-anonymity

k-anonymity applies k-anonymity principles to HyperLogLog (HLL) sketches to quantify and control re-identification risk while enabling cardinality estimation in federated clinical data queries across electronic medical records.

Key Features:

• Theoretical and Empirical Privacy Control: Proves theoretical bounds on k-anonymity risk for HyperLogLog sketches and provides code to compute these bounds for empirical analysis.
• Federated Query Support: Operates on federated queries across multiple clinical data repositories to produce aggregate counts without sharing individual-level records.
• Efficient Data Estimation: Uses HyperLogLog probabilistic sketches to estimate large cardinalities with low memory overhead and to balance accuracy against privacy leakage.

Scientific Applications:

• Clinical Research: Enables aggregate patient-count queries (e.g., number of patients with diabetes) across connected hospitals' electronic medical records while reducing re-identification risk.
• Data Privacy Studies: Provides a framework for analyzing privacy–utility tradeoffs and evaluating k-anonymity risks of HLL-based federated analytics.

Methodology:

Employs HyperLogLog (HLL) probabilistic sketches for cardinality estimation, derives and computes theoretical bounds on k-anonymity risk for HLL outputs, and integrates k-anonymity by ensuring individual entries are indistinguishable from at least k-1 others; includes code to compute the bounds.