We need to understand that “knowing someone’s identity” is not coextensive with “knowing their name”, and that in fact knowing someone has a rare medical condition may be more identifying[0] than knowing their name.
[0] Or k-deanonymising, for anyone who’s pedantic about identity being an absolute.
This is a good example of the law and popular conception of a concept being badly out of date, to the advantage of industry and disadvantage of regular people. Therefore industry has a vested interest in keeping the public perception focused on "de-identified" with a narrow definition of PII.
For the kind of data in this particular database (mostly insurance claims data), it's highly unlikely that you could learn much through re-identification of the data.