DataTerminology.comNEW
Processes that modify data to reduce the identifiability of individuals by removing, transforming, or generalizing identifiers and quasi-identifiers.
DESCRIPTION
Data de-identification aims to lower re-identification risk while preserving analytical utility. Techniques include direct identifier removal, generalization, suppression, and noise addition, often guided by k-anonymity, l-diversity, or t-closeness. Differential privacy provides formal guarantees for aggregate analyses. Legal frameworks such as HIPAA and guidance from NIST set expectations for acceptable risk. De-identification is not equivalent to full anonymization; residual risks remain and should be evaluated continuously.
ALTERNATIVE DEFINITIONS
U.S. HHS
De-identification is a process of removing identifying information from data so that it cannot be used to identify an individual.
Regulatory guidance focused on health information privacy.
NISTIR 8053
De-identification is the general term for processes used to remove or obscure identifying information to protect the privacy of individuals.
Highlights a family of techniques and risk-based approach.
RELATED TERMS
K-anonymity Differential Privacy Data Masking Tokenization HIPAA Safe Harbor GDPR Re-identification Risk Privacy Enhancing Technologies
COMPETING TERMS
Anonymization Pseudonymization
RELATIONS
- Broader: De-identification
KNOWLEDGE AREAS
Data Privacy & Compliance AI Governance Data Management Data Science
TARGET AUDIENCE
Data Scientist Data Engineer Data Steward Data Governance Manager AI Engineer Chief Data Officer
EXTERNAL SOURCES
- Wikipedia - De-identification
- U.S. HHS - HIPAA de-identification guidance
- NIST - NIST report on de-identification
- UK ICO - Personal data and anonymisation guidance
TOP QUESTIONS
How does Data De-identification differ from anonymization?
De-identification reduces identifiability but may retain residual risk. Anonymization aims for irreversibility. Many laws treat anonymized data as out of scope, but few datasets are perfectly anonymous.
Category: Misunderstanding
What are common techniques in Data De-identification?
Direct identifier removal, generalization, suppression, sampling, perturbation, hashing, and synthetic data. Selection depends on risk tolerance and analytical needs.
Category: Example
What is k-anonymity’s role in Data De-identification?
K-anonymity ensures each record is indistinguishable from at least k−1 others by quasi-identifiers. It helps bound re-identification risk from linkage.
Category: Concept
How is Data De-identification assessed for risk?
Use attacker models, quasi-identifiers, external data availability, and metrics like k-anonymity and re-identification tests. Regular reassessment is needed as data ecosystems evolve.
Category: Risk
When is differential privacy appropriate for Data De-identification?
For sharing aggregate statistics or training models while bounding privacy loss. It provides formal guarantees with privacy budgets, at the cost of accuracy.
Category: BestPractice
What are pitfalls in Data De-identification?
Underestimating quasi-identifiers, over-generalizing and losing utility, and failing to document transformations. Attackers can leverage auxiliary datasets for linkage.
Category: Mistakes
How do regulations view Data De-identification?
HIPAA provides Safe Harbor and expert determination methods. GDPR favors pseudonymization and risk-based anonymization; de-identified but re-identifiable data may remain in scope.
Category: Regulation
Is synthetic data a Data De-identification technique?
Yes, when generated from real data to mimic distributions. However, it requires disclosure risk assessment to prevent memorization and leakage.
Category: Opinion
Where should Data De-identification be implemented in pipelines?
Apply at ingestion or before data sharing, as close to the source as feasible. Integrate with catalogs and lineage for transparency and auditability.
Category: Management