Beyond PHI Redaction: Automating k-Anonymity for Regulatory-Grade De-Identification

Redacting PHI fields is only the start of a de-identification data pipeline. A dataset can have every personal name, birthdate, and address removed and still expose patients: if only one person in the dataset is a 42-year-old in zip code 10001, that combination identifies them as surely as a name would. These quasi-identifiers – unique combinations of ordinary fields like age, zip code, gender, and diagnosis codes – are why redaction-only tools cannot deliver regulatory-grade de-identification.

This webinar shows how to measure and resolve quasi-identifier risk automatically with John Snow Labs’ de-identification libraries and models. Pipelines calculate k-anonymity, l-diversity, and t-closeness, then generalize fields only where the metrics require it. For example, “Zip code: 10001, Age: 42” may become “Zip code: 100xx, Age: 40–45”, and diagnosis code E11.36 (diabetic cataract) may roll up to E11.3 (type 2 diabetes with eye complications). Every transformation is automatic, logged, and traceable. You’ll leave knowing:

• Why redacted datasets can still be re-identifiable
• What k-anonymity, l-diversity, and t-closeness measure, and how to pick targets
• How to build a pipeline that applies these metrics and generalizations automatically, in an auditable workflow

This webinar is intended for data architects, engineers, and scientists looking to learn how to deliver de-identified medical data suitable for HIPAA Expert Determination.