This was previously discussed in #2840. At that point, the decision was to use Location. However, this doesn't work.

Every system I've ever seen that tracks Patient addresses has had the ability to distinguish physical addresses (those you can visit) from mailing addresses (e.g. PO Boxes and care-of addresses). To suggest that in order to convey that simple structure requires including an extension (which we don't define) to link the patient to a contained Location having only an address is precisely the sort of craziness we're trying to avoid with FHIR.

Saying the Patient is associated with a Location (other than when admitted to a hospital) is extremely rare. There's no obvious benefit to doing this and a huge number of drawbacks. Furthermore, it gets even messier if the patient has multiple physical addresses (e.g. home address, work address, vacation home). That's easily handled within the address data type. If we use Location, it would mean we'd have to "type" the association to Location and possibly even associated time-periods with it.