An Integration Problem — or a Data Model Problem?
Hospitals believe they have an integration problem and build more and more interfaces. Why that raises complexity instead of resolving it.
Hospitals believe they have an integration problem. So they build more and more interfaces. The problem does not get smaller — it gets more expensive.
The supposed integration problem
Every new connection is treated as an interface task: system A talks to system B, so we build the bridge between them. HL7 solves the transport reliably — messages arrive, fields are mapped, the data flow stands. But transport was never the real problem.
What HL7 does not solve
Four things remain open, however clean the transport is:
- a shared clinical data model — what are we even talking about;
- consistent semantics — does the same field mean the same thing in both systems;
- terminology governance — who maintains codes and mappings over time;
- secondary use — can the data later be evaluated reliably for research and management.
No interface closes any of these gaps, because they all lie below the interface.
Why more interfaces make it worse
Connect n systems pairwise and you get a quadratic number of translations — each with its own, often implicit, semantics. With every bridge, not only the effort grows but also the number of places where meaning is silently lost. Complexity rises and responsibility frays: no one sees the whole any more.
Reversing the perspective
Modern approaches reverse the order. First a shared data basis is established — a clinical data layer, a canonical model that all systems map against. Then integration becomes a technical detail: no longer n-to-n, but each system once against the centre. The interface is still necessary — but it is no longer the place where meaning is negotiated.
The uncomfortable answer
Hospitals rarely have an integration problem. They have a data model and governance problem disguised as an integration problem. As long as no one owns the data model — with a name, not a role — every integration becomes the next layer of complexity. The right first question is therefore not “how do we connect the systems?” but “on what basis do we describe what these data are?”.
We support hospitals and their IT in exactly this shift — from the next interface towards an owned, shared data basis. If you want to know whether your integration problem is in truth a data model problem: get in touch.
The formal foundation — the AION model and the CAIRN reference implementation — is documented at aion-clinical.eu.