Mithras Privacy & de-identification · 8 field notes
Privacy & de-identification
Privacy in health data is not a policy paragraph — it is machinery: k-anonymity floors, suppression rules, federated queries, audit trails that cannot be edited. These pieces explain the apparatus, including where it fails and what honest defenses look like.
8 of 50 field notes · Privacy & de-identification · plain language
The journal holds 50 field notes across 7 sections: Patient rights (8), Privacy & de-identification (8), Standards & interoperability (8), Regulation worldwide (8), Research methods (8), Chronic conditions (6), Principles (4). You are viewing the Privacy & de-identification section.
Why small numbers are dangerous: small-cell suppression in health statistics
Small cell suppression in health data, explained: why tiny counts identify real people, where thresholds like 5 and 11 come from, and how k floors extend it.
№ 14 Privacy & de-identificationDe-identified, anonymized, pseudonymized: what these words actually mean
De-identified vs anonymized vs pseudonymized data: three terms, three legal regimes under HIPAA and GDPR, and a clear table of standard, reversibility, and scope.
№ 18 Privacy & de-identificationFederated analytics: research the data without moving the data
How federated analytics lets researchers study health data without moving it: in-region queries, aggregate-only results, fewer transfer problems.
№ 29 Privacy & de-identificationDifferential privacy in healthcare: when noise protects patients
Differential privacy in healthcare, explained in plain language: what epsilon means, how calibrated noise protects patients, and where it fits.
№ 38 Privacy & de-identificationRe-identification attacks: how 'anonymous' health data gets unmasked
Real examples of re-identification attacks on anonymized data — the Weld case, AOL, Netflix, Strava — and the defenses that assume linkage exists.
№ 02 Privacy & de-identificationk-anonymity, explained with a hospital waiting room
What k-anonymity means, why k ≥ 5 is a common floor in health data, and how the waiting-room test makes a 2002 idea easy to reason about.
№ 43 Privacy & de-identificationAppend-only, tamper-evident: what a real audit trail looks like
What makes an audit log tamper evident in healthcare: append-only writes, hash chains, cryptographic anchoring, and HIPAA's six-year retention rule.
№ 49 Privacy & de-identificationHIPAA Safe Harbor vs. Expert Determination: two legal paths to de-identify data
HIPAA Safe Harbor vs Expert Determination: the two paths to de-identify data, the 18 identifiers, a statistical risk standard, and the research trade-offs explained.