It would work in many cases, though not all. You would not hash everything together. Instead, you hash normalized identifiers independently, such as email address, phone number, or physical address. An incoming dataset would only need to match one of these to be excluded.

> physical address

Not unique to a person

> email address, phone number

Also often not unique to a person, although email addresses probably tend to have much longer lifespans as identifiers than phone numbers.

If the idea is to have a true opt-out system, it's really really difficult to implement given how these systems work.

If you look at the data provided by services like accurint, you'll frequently see the same SSNs used for decades by multiple different individuals, often with IDs from different states with the same name and DoB despite obviously being different people. With how the system works in the US, it can often be impossible for anyone to determine which physical person the SSN was actually originally assigned to.

Same obviously applies to other identifiers you suggested, but even the seemingly good ones are not very good at uniquely identifying people.