FHIR R4 is the specification most payer and patient-access work now targets, and almost none of the difficulty lives in the specification itself. It lives in what the other end actually returns.
FHIR R4 is the specification most payer and patient-access work now targets, and almost none of the difficulty lives in the specification itself. It lives in what the other end actually returns.
The failures I see are consistent. An Implementation Guide is declared but the resource profiles inside it are not enforced, so a Patient resource arrives with a date of birth as a string in one system and an ISO instant in another. A $everything operation is implemented but paginated with a link that the client cannot follow without a bearer token it does not have. A Bundle contains a search.mode of include that points at a resource the responding server has already deleted. None of these break a demo. All of them break a production interface on the third week.
The work is therefore mostly conformance and evidence. I build against the patient-access and provider-directory requirements, capability statements first, then a mapping sheet that names every field, its cardinality, its terminology binding and the system that owns it. Where the remote FHIR endpoint cannot be trusted, the integration keeps a normalised internal model and reconciles, rather than passing unvalidated resources through to whatever consumes them.
Two things I insist on before any build starts: a sandbox with the exact version of the remote FHIR server that production will speak, and a written list of the resources the patient actually needs. Without the first, the interface is tested against a fiction. Without the second, the project integrates everything and satisfies nobody.
Delivered from a Nevada base: remote by default, on site across the Wasatch Front and to the rural counties when the work needs presence. Scoped in writing, fixed in fee where the boundary is clear, and measured against the interface or control it was meant to produce.
Which FHIR version should a new integration target? R4 for anything involving the patient-access and payer rules today; R4B or R5 only when the counterpart server states support in its CapabilityStatement and the consuming application is ready for it.
Do we need a FHIR facade over our existing database? Often yes, and it should be thin. A facade that models the whole clinical record is a second product. One that exposes the resources the counterpart asks for, validated against the IG, is a few weeks of work and can be replaced later.
How do you test without a live counterpart? An IG-derived fixture set plus a replay of real responses captured from the sandbox. The fixture proves conformance; the replay proves the interface survives what the counterpart actually sends.
Working with organisations across Salt Lake City and the surrounding area — on site, remote, or both.
Serving clients worldwide — remote delivery from Las Vegas, Nevada, USA · Open 24 hours · 702-559-3489 · Las Vegas, Nevada 89101