Cross-document PHI Linking

The fundamental process of determining whether disparate PHI mentions refer to the same real-world entity. This involves probabilistic matching algorithms that weigh the similarity of quasi-identifiers like names, dates of birth, and addresses.

• Uses blocking techniques to reduce the quadratic comparison space.
• Applies fuzzy string matching (Levenshtein distance, Jaro-Winkler) to handle typographical errors.
• Critical for ensuring a single patient's records are consistently linked for longitudinal analysis.

A privacy-preserving mechanism that ensures every instance of a specific individual receives the same pseudonym across all documents. This is distinct from simple redaction.

• Maintains longitudinal data integrity for research without exposing real identities.
• Relies on a secure, deterministic hashing function applied to a master patient index.
• Prevents data fragmentation where a single patient's records become disconnected after de-identification.

The statistical linkage of records using indirect identifiers that, when combined, can uniquely identify an individual. Examples include {gender, ZIP code, date of birth}.

• Calculates uniqueness metrics to assess re-identification risk.
• Must account for data sparsity and missing values in clinical narratives.
• A core challenge in satisfying the Expert Determination method under HIPAA.

The process of using chronological information to validate or reject a potential PHI link. A mention of a 30-year-old patient in 2024 cannot refer to the same individual as a mention of a 70-year-old in the same year.

• Parses and normalizes relative dates (e.g., 'three weeks ago').
• Uses temporal reasoning to build consistent patient timelines.
• Essential for disambiguating family members who share addresses or last names.

An advanced technique that models PHI mentions as nodes in a knowledge graph, with edges representing probabilistic matches. Community detection algorithms then cluster nodes into distinct identities.

• Resolves complex transitive closure problems (A matches B, B matches C, therefore A matches C).
• Allows for global optimization across an entire corpus rather than pairwise decisions.
• Handles conflicting evidence gracefully by weighing edge confidence scores.

The specific linguistic task of resolving anaphoric references that span document boundaries. For example, linking 'the patient' in a discharge summary to a specific named individual in an admission note.

• Extends traditional within-document coreference resolution.
• Requires a global entity store to maintain referential context.
• Prevents the creation of orphaned PHI mentions that escape redaction because they lack an explicit name.