[nradov]

This is a solved problem from a modeling standpoint. The HL7 Reference Information Model allows any entity (such as a person) to have multiple names. Each name can be tagged with a type (legal, maiden, alias, etc) and validity date range. A name can contain multiple parts in any order, optionally tagged as prefix / suffix / family / given. Names can also be explicitly marked as null if unknown or not assigned. There are open source RIM implementations in several languages.

[deathanatos]

Interesting; I didn't think I'd see an HL7 reference in this thread. I work somewhat with FHIR, which also has a HumanName[1] data type, and I think it handles most of the cases in this thread.

For those not familiar: FHIR is a standard that covers health and patient data. IMO, it's a pretty good model. (HL7 is the organization, and there are few other standards under it.)

I'm less familiar with RIM; could you link to it's definition of a name? (The best I could find suggested that it was nothing more than an unconstrained piece of text.)

[1]: https://www.hl7.org/fhir/datatypes.html#humanname

[nradov]

Unfortunately due to the way the HL7 web site is structured there's no way to give a direct link. Go to the Normative Edition, Foundation, Data Types, Basic Types.

http://www.hl7.org/implement/standards/product_brief.cfm?pro...

The FHIR data model is a little simpler to allow for easier implementation. In the vast majority of real world healthcare use cases it works well. But from a modeling standpoint if you need to cover odd edge cases it sometimes helps to look back at the old RIM.

[a-priori]

What's the purpose of structured modelling of names? Why does it matter which parts are family vs given vs whatever else? Lots of the W3C International Examples they give use a `<text>` field. Why not just use that?

https://www.hl7.org/fhir/datatypes-examples.html#2.24.1.13.1

[nradov]

It matters for collation order, if you want to show a list of patients sorted by family name.

[Piskvorrr]

Is this a current use case, or did that just make sense 100 years ago? (As in these assumptions: "tracking actual relationships in data is Hard, and family name correlates strongly with real-world relationships") I mean, it sort of works even today, but neither of the assumptions are as strong any more.

[nradov]

It's still a current use case for healthcare provider workflows. Also helps a lot when doing automated record linkage between different systems.