Traditional data classification often treats individual Medicare claims as non-sensitive once direct identifiers like names are removed. However, this logic fails in small communities. When a researcher joins a seemingly anonymous claims dataset with a local voter registration list or a census tract, the math changes. A specific age, zip code, and a rare diagnosis can act as a fingerprint. In a tiny town, that fingerprint points to a specific neighbor.
To fix this, we must shift from a static classification to a context-aware model. Instead of labeling data based solely on its intrinsic properties, we should classify it based on its linkability. This means assessing how a dataset behaves when it meets others. We should implement a
