Why do archetypes need a suitable reference model?

People, particularly those in working in the standards arena, are arguing that we need a formalism for describing archetypes that is independent of any reference model. This is actually an oxymoron.   Here are some thoughts.

The Detailed Clinical Models group was formed independently of HL7 to try and find a methodology for describing clinical data specifications. Some members of this group have committed to the openEHR methodology, others have continued to seek a 'reference model independent' approach. For the past couple of years the group has been looking at UML and a recent email from a leader in this space has suggested that perhaps UML is not the best approach.

The openEHR methodology and Archetype Definition Language (ADL) requires a reference model and the openEHR reference model has been specifically designed to support clinical archetypes. ADL can work with the HL7 RIM or ISO13606 as well and is part of the ISO standard. Early work with ADL drew attention to a range of issues that had to be addressed before using ADL with the RIM and the HL7 community moved away from this path about 4 years ago. One can understand that people seek a level playing field and do not want to give openEHR an advantage, but we need to face the issues head on; we need the best reference model available to provide the basis for archetypes.

ISO 13606 is the alternative. It is more generic and based on an earlier work. To proceed with 13606 a set of 'proto-archetypes' are required which mirror or offer the organising capacity of the openEHR entry classes, then specialise these prototypes to produce the actual clinical archetypes. This has advantages - the proto-archetypes can change without changing the reference model. But it also raises new problems: it forces the tooling to be generic and difficult for clinicians to understand; places a larger overhead on the queries; makes software much harder to write and maintain; and does not guarantee the outcome desired (ie people can do what they like). Also specialisation is pretty rudimentary at present and requires features in ADL 1.5 will make this much more robust but are not part of the standard. Specifically, those building archetypes will have to think about how to do timing, time series, averages, maximums, totals, the state of the patient at the time of measurement rather than the core data (as this is almost unlimited in different clinical and research environments), information about the measurement, the device, where and how it was measured....

So the question that needs addressing is, "How suitable is a particular reference model for using with archetypes?" The NHS, Sweden and Denmark and soon probably Brazil and Scotland - have taken a big step in the openEHR direction. People working with openEHR have sorted out how to make this work in a way that will take years to reproduce in another space. The politics is yet to be resolved.

I think the openEHR community have to stare down this idea of a 'reference model independent' archetype formalism - that is what ADL is! It is, after all, the reference model that makes archetypes work. At present it is the openEHR reference model that makes archetypes work in healthcare - it was design specifically for that purpose.

As this argument unfolds the openEHR community will grow and get stronger. We will understand the next generation of issues - such as specialisation - while others are struggling to find something vaguely equivalent. Also, the archetypes that can be used in different systems will work natively in real working systems (a much safer proposition than a theoretical solution).