I just realized you wrote “de-anonymization”. We don’t do that, but we do a risk analysis that simulates an attack, making some pessimistic assumptions about wh

I just realized you wrote “de-anonymization”. We don’t do that, but we do a risk analysis that simulates an attack, making some pessimistic assumptions about what data an attacker may have for linkage, to estimate how risky data is to share, essentially as a benchmark. TLDR; more unique combinations of quasi-identifiers make data more susceptible to a linkage attack. We also pitch our system as a risk analysis only tool, though no one is using it for that currently. If you’re interested in methods, I’d recommend El Emam’s book, “Guide to the De-Identification of Personal Health Information”, which covers a lot in detail. Our attack simulation is based on his work.