Question: FHIR and un-semantic interoperability


 I did not understand the blog post about un-semantic interoperability.  Can you elaborate?  Will FHIR provide any of this un-semantic interoperability?


Well, the original post on unsemantic interoperability is just pointing out that many people mis-understand the nature of what semantic interoperability is trying to achieve:

We’ve had semantic interoperability in healthcare since we started having healthcare. Since the beginning of healthcare (by whatever definition you can use), healthcare practitioners have exchanged data using spoken and written words, and the semantic meaning has been clear (well, as clear as it can be given that human knowledge is limited).

So whatever it is that we are doing, it’s not introducing semantic interoperability. In fact, what we are doing is introducing a new player into the mix: computers. And not, in actual fact, computers, but the notion that there is something to be gained by processing healthcare information by persons or devices who don’t properly understand it. So, in fact, what we are actually doing is seeking for unsemantic interoperability.

It’s a matter of perspective. Perhaps, one day, we’ll really be working on true semantic inteoperability. But right now, we can afford to chase a lesser goal, which is exchanging data that can be used usefully in some limited pre-ordained ways.

And FHIR provides lots of this – that’s what it’s good at – getting data that is passably well self-described to be exchanged as easily as possible.

For systems that is really working towards genuine semantic interoperability, FHIR is actually a step backwards (though I’d argue that the easy availability of information that is passably well described is a huge improvement over the non-availability of information that is well described).


  1. Lloyd McKenzie says:

    Semantic interoperability means interoperability where the receiver of the information can not only parse it, they can understand what it means. Computable semantic interoperability means that the receiver that needs to understand is a machine or piece of software rather than a human. Humans can bring a lot more context and inference to bear when attempting understanding than computers can, therefore the bar is lower when attempting to achieve human-to-human interoperability than it is when attempting to achieve human to computer or computer to computer interoperability.

    Computer semantic interoperability means that the combination of the meaning of the elements and any context needed to firmly establish that meaning must be either well defined and fixed or clearly identified in the instance. It also means that humans need to properly populate the elements with data the computer expects to be there.

    Computer semantic interoperability falls down in to places:
    1. Specifications that aren’t clear about what data is supposed to go where (meaning that the receiving computer can’t be confident about what the data in a given place actually means)

    2. User interfaces and human operator tendencies that result in inappropriate data getting stuck in places regardless of how rigorously the specifications are defined – due to misunderstanding, laziness, no appropriate place for the data, etc.

    These two things are often in tension. Being extremely precise about meaning generally involves breaking information into lots of narrowly defined pieces. That tends to be a pain for both interface designers and human data enterers.

    FHIR makes different “default” choices from those of v3 and CDA in the balance between ease of interface and precision of data elements, though it provides the ability via extensions for those implementers who want to to go as discrete and precise as they wish.

    In that sense, I’m not sure that FHIR *is* a step backwards. One could even argue it’s a step forward.

    Really robust semantic interoperability is going to be dependent on machine learning and other technologies that will allow the computer to be at least as good at reasoning and inferring context as humans are now – at which point we won’t need discrete data at all. Computers will be able to take a big blob of text/dictation/images/etc. and extract all the relevant information needed for their functions without it being neatly divided up for them.

    Until then, I think FHIR’s a decent step in the right direction.

  2. Karim Keshavjee says:

    Unfortunately in all the discussion about semantic and/or non-semantic interoperability, we keep forgetting the purpose of achieving them. If we really took time to understand the purpose of interoperability, we’d have achieved a lot more than we have to date.

    Sometimes I think we’re pursuing a Holy Grail when a plain chalice would do fine, or even a cup.

Leave a Reply

Your email address will not be published. Required fields are marked *

question razz sad evil exclaim smile redface biggrin surprised eek confused cool lol mad twisted rolleyes wink idea arrow neutral cry mrgreen


%d bloggers like this: