Stream: openehr
Topic: Transforming openEHR observations to FHIR profiles
Diego Bosca (Jul 12 2019 at 07:40):
I have finally something to show :)
https://standardsmania.blogspot.com/2019/07/openehr-observations-to-fhir-profiles.html
Grahame Grieve (Jul 12 2019 at 10:27):
@Ian McNicoll
Ian McNicoll (Jul 12 2019 at 11:07):
Ian McNicoll
Yes - I have been helping Diego (slightly) on this. 90% is his good works. The body weight example is pretty good as it is a sort of minimal example of the archetyping approach vs. profiling. Key question is how to handle things like our comments and Confounding factors ( I think they actually map pretty well to notes) but more tricky is what to do with Stat of dress - is that a component or a separate profiled observation.
https://www.hl7.org/fhir/bodyweight.html
or
https://fhir.nhs.uk/STU3/StructureDefinition/CareConnect-DCH-Weight-Observation-1/_history/1.3
or
https://fhir.nhs.uk/STU3/StructureDefinition/CareConnect-EMS-PDS-BirthWeight-Observation-1
Kevin Mayfield (Jul 12 2019 at 13:11):
or could be neither???
(trying this as an experiment at the moment but it's proving quick and useful)
ObsDef.PNG
Ian McNicoll (Jul 12 2019 at 13:25):
We are looking at this too - it is quick and useful once you know what your requirements are, so a good match for e.g labs. Not so sure about potentially more complex associations of values + 'metadata' body-weight/ birth-weight/ state-of-dress' is a good minimal example.
David Hay (Jul 12 2019 at 18:43):
@Kevin Mayfield - is that a tool to browse conformance resources?
Kevin Mayfield (Jul 12 2019 at 20:00):
That’s what I’m aiming for, a work in progress at the moment.
Patrick Werner (Jul 24 2019 at 14:19):
I have finally something to show :)
https://standardsmania.blogspot.com/2019/07/openehr-observations-to-fhir-profiles.html
interesting work. Just to let you know: These are deprecated profiles, the current status of clinical genomics on FHIR can be found here:
http://build.fhir.org/ig/HL7/genomics-reporting
Diego Bosca (Jul 25 2019 at 21:20):
yes @Patrick Werner found out afterwards :)
I have to come back to the FHIR->openEHR transformation, but I'm currently focused in the openEHR->FHIR one (which is providing quite a few challenges). Maybe I'll write a follow up post soon
Georg Fette (Sep 04 2019 at 08:38):
Hi @Diego Bosca , why did you restrict your transformation approach to an Observation profile and its extensions and not arbitrary profiles ? As I am also working on a project in which I transform archetypes to FHIR custom profiles, I wonder where the limitations of those transformations lie.
Diego Bosca (Sep 05 2019 at 08:15):
Well, I had to start somewhere :D
Now more seriously, most resources probably would need a source equivalent archetype (probably adhoc) and the genericity is lost. With observation archetypes (and maybe Compositions), it seems to me that the idea is more useful as there are quite a few observations already defined as openEHR archetypes that can be turned into profiles if you know where to put things.
The process I've defined doesn't require itself that the source archetype class is an Observation (you can take CLUSTER archetypes or even full OPT profiles).
The idea is to choose the data points from the archetype that would end in the profile. You can choose to leave data points out, or decide that something is the main value or the observation, one of the components, or even an extension to that profile.
Data types transformation part could potentially be reused in other archetypes-> resource transformations, but should be done carefully as not all data types are allowed as value[x] (or whatever[x]) in different profiles (this is one of the main reasons I decided to support the automatic generation of extensions in the first place...)
Grahame Grieve (Sep 05 2019 at 08:19):
right. OpenEHR Observation and FHIR Observation don't quite have the same semantic range. So some openEHR archetypes no on Observation are relevant.
Grahame Grieve (Sep 05 2019 at 08:19):
it's on my todo list to define an abstract Observation such that profiles on it (whether stated as ADL or structure definition or whatever) can be understand and tested in v2, CDA, and FHIR
Grahame Grieve (Sep 05 2019 at 08:20):
but whether I do that or not, I've long wondered why openEHR isn't working on turning openEHR observation archetypes into profiles on HL7 observation things.
Diego Bosca (Sep 05 2019 at 08:47):
I would say one of the reasons is the granularity, as a single archetype will most likely end in a set of profiles. Just this "split" would probably end giving different sets of profiles depending on the person doing it. This means that probably a process similar to the archetype governance would need to be put in place to agree in that split (in order to also achieve better profile reuse), but including more FHIR domain experts.
Having said that, I know several projects that are producing new archetypes that are already trying to generate the equivalent profiles, so I would say is more a matter of resources than anything else.
Diego Bosca (Sep 05 2019 at 08:50):
Then (and just talking from my personal experience) are the things which are difficult to be translated, such as alternatives in monovaluated attributes, some node constraints or lack of data types allowed in a given FHIR profile attribute, which means that there should be a human deciding how to deal with these things
Grahame Grieve (Sep 05 2019 at 10:12):
yes it doesn't seem easily automatable, but I've always thought that we could move towards that point faster than we are. But it's never seemed to be a priority
Diego Bosca (Oct 30 2019 at 10:54):
Second blog post https://standardsmania.blogspot.com/2019/10/using-linkehr-to-generate-fhir.html
I've put the transform functionality into LinkEHR, so everyone can try it. Should I make some kind of announcement in the #tooling stream?
Grahame Grieve (Oct 30 2019 at 11:20):
sure
Thomas Beale (Nov 12 2019 at 23:39):
@Grahame Grieve : I've long wondered why openEHR isn't working on turning openEHR observation archetypes into profiles on HL7 observation things.
That's easy enough - most of what openEHR Observation has isn't in FHIR Observation. This is not unexpected; FHIR Observation was clearly designed to pluck a single data item from a source system. openEHR Observation is about putting data in, including timing structures, complex data structures, body state data, protocol (another structure), guideline ids, and so on. It's a different thing.
Lloyd McKenzie (Nov 13 2019 at 00:44):
Most of those things are part of Observation, though complex data timing generally applies to the orders rather than to an individual result.
Diego Bosca (Nov 15 2019 at 12:05):
Just released a LinkEHR version with Bundle and Composition profile generation from openEHR archetypes. Now also a set of FHIR Observation profiles can be generated from a single openEHR archetype.
Last updated: Apr 12 2022 at 19:14 UTC
