SNOMED/LOINC mapping

Hi,

One of the things that has been holding me up is the conundrum around
coded clinical documentation.

I would use OpenEHR exclusively but as I delve into the depths of
integration I am getting nervous that the archetypes have not yet been
mapped to LOINC or SNOMED.

As an example comparison take a patients (non-birth) weight for
example. In pediatric hospitals and for patients receiving chemo it
has always been critical to differentiate between a carefully measured
weight suitable to weight dose calculations versus a weight perhaps
measured adhoc at an outpatient clinic.

In SNOMED I find

27113001 : body weight - synonym of Body weight (observable entity)

In OpenEHR we have

openEHR-EHR-OBSERVATION.body_weight.v1.adl

which defines the weight, clothing worn, and device used.

In LOINC we have

LOINC_COMPONENT PROPERTY TIME_ASPCT SYSTEM SCALE_TYP METHOD_TYP
3141-9 Body weight Mass Pt ^Patient Qn Measured
3142-7 Body weight Mass Pt ^Patient Qn Stated
50064-5 Body weight Mass Pt ^Patient Qn Ideal
8335-2 Body weight Mass Pt ^Patient Qn Estimated

I expect the SNOMED term could be qualified - but I would have to
search around to figure out with which term would be appropriate
(illustrating the challenge).

So I guess ultimately the question boils down to what is
everyone/anyone else doing

- OpenEHR
- OpenEHR + LOINC
- OpenEHR + SNOMED
- All (tough especially as each system has its own set of terms and
concepts of what is what)

And perhaps what does your average EMR doc want?

I am not talking about ICD9/10 or CPT codes as I need to do those for
billing anyway.

thanks!

Greg

http://www.patientos.org
Open Source EMR in the making

Greg Caulton wrote:

So I guess ultimately the question boils down to what is
everyone/anyone else doing

- OpenEHR
- OpenEHR + LOINC
- OpenEHR + SNOMED
- All (tough especially as each system has its own set of terms and
concepts of what is what)

*the mapping work still largely has to be done; both Snomed and LOINC
will be mapped to various archetypes, LOINC obviously more around
results. You could get involved with the effort;-)

- thomas

Greg Caulton wrote:

So I guess ultimately the question boils down to what is

>everyone/anyone else doing

- OpenEHR
- OpenEHR + LOINC
- OpenEHR + SNOMED
- All (tough especially as each system has its own set of terms and
concepts of what is what)

>
*the mapping work still largely has to be done; both Snomed and LOINC
will be mapped to various archetypes, LOINC obviously more around
results. You could get involved with the effort;-)

- thomas

The problem with SNOMED mapping (and you wouldn't want me doing that
as I would be very bad at it) is that I assume (perhaps incorrectly)
that the OpenEHR foundation could not distribute the mapping without
the recipients first getting some kind of license for the SNOMED - not
insurmountable but that does complicate things.

I am not sure I fully understand the direction in OpenEHR for lab
tests - creating an archetype per lab test seems overkill but there
appear to be some lab test archetypes. Perhaps OpenEHR should just
define 'types' of lab tests and leave the distinct tests to be defined
as coded values.

As I discovered with looking at a CCR export LOINC has more than lab
tests and so there is definitely overlap - though it really needs
clinicians and a non-trivial effort. Likely archetypes would need to
be expanded to accommodate everything.

Greg

Greg Caulton wrote:

The problem with SNOMED mapping ([...]) is that I assume (perhaps
incorrectly) that the OpenEHR foundation could not distribute the
mapping without the recipients first getting some kind of license
for the SNOMED - not insurmountable but that does complicate things. [...]

A fare as I know there have not been any precedent about what is legal and
what is not to do with derivates of SNOMED CT without a license. But it
seems that people involved in IHTSDO’s work think it is ok to handle
references to SNOMED CT without a licence.

  Mikael Nyström
  Department of Biomedical Engineering
  Linköping University

  Member of IHTSDO’s technical committee

Hi Greg,

I think that to wait until all archetypes are mapped to terminology
reflects losing view of the value that the structured content found in
archetypes brings to health informatics. The structure inherent in an
archetype gives an excellent semantic handle on clinical content in the
large majority of cases, and the strategic and appropriate mapping of
terminology to these archetypes will (theoretically) enhance clarity and
minimise any potential ambiguity and make the models VERY powerful.
(Practical experience suggests that this is not as easy as some might
think - and has been alluded to by some others previously.)

Indeed, there are data points in archetypes that are very dependent on
mapping to terminology, such as lists of diagnoses or symptoms - the
free text alternative here is not pretty nor particularly useful for
querying. BUT THESE ARE VERY MUCH IN THE MINORITY, and will often have
more of a role in terminology subsets applied in templates. Archetypes
will only have terminology codes incorporated where it is universally
applicable - the archetype by definition is designed as a maximal data
set and for a universal usecase, so if there is doubt about the
universality of a given terminology code, it probably needs to be
applied within a template environment, rather than the archetype.

And remember that Snomed/LOINC is not the only terminology that needs to
be incorporated - there are a lot of places in the world that have no
mandated terminology; that do not have a relevant language version of
Snomed; that mandate an alternate terminology such as ICD-10 etc. So we
have a number of terminologies to manage.

The Clinical Knowledge Manager (CKM - the openEHR archetype repository)
is being designed from the following point of view - lets model the
clinical content as per clinician requirements in the archetypes, and
get them reviewed, published and out there and being used - as they are
increasingly being used in real implementations!
And by all means, let's incorporate terminology as fast and as sensibly
as we can. The reality is that terminology mapping is possibly a slower
process than we think, yet one that can happen in parallel or following
on from archetype publication. In the pending CKM, archetypes will be
formally regarded as publishable based on agreement on both clinical
content and design, by a combination of clinician teams and openEHR
geeks. Terminology mapping can be commenced as soon as the content has
been finalised (otherwise we don't know what content needs a code), but
in the majority of archetypes can safely happen at a later date without
major impact on semantic representation of clinical content. For the
minority of archetypes that are identified as having immediate
terminology requirements, then this process needs to happen simultaneously.
Similarly, translations can only commence once the content is finalised
and published, and will no doubt be an iterative process, with
translations being dependant on need.

The CKM is not far away and looking good - currently undergoing beta
testing... Will keep you posted.

Cheers

Heather

Greg Caulton wrote:

Hi Greg

I am going to take a clinical view in this response. A problem that we have noted is that it is relatively easy to get clinicians to agree about the data that they might want to collect and while structuring it can be a challenge, it is proving tractable. The content from terminologies at various points is a relatively straight forward problem and we have a rapid subsetting tool now that is based on queries against terminologies.

Using terminologies to label values (as is used in HL7 v2 and v3) is the domain of LOINC largely, and we have a generic laboratory archetype that allows labels to be coded. This does mean that the queries need to search for LOINC codes as well as the archetypes which is more challenging.

Any way, more in line…

Greg Caulton wrote:

Hi,

One of the things that has been holding me up is the conundrum around
coded clinical documentation.

It is important to differentiate codes used to label content (LOINC) from the content itself. SNOMED is attempting to do both but it is not clear if it is taking the right approach.

I would use OpenEHR exclusively but as I delve into the depths of
integration I am getting nervous that the archetypes have not yet been
mapped to LOINC or SNOMED.

Mapping is complex - as a single code might relate to three or four variables in an archetype (as the code is precoordinated). Average body temperature in the humicrib over 24 hours is an example. LOINC might have a 24hr average, a 10 and an 8 hour as well - separate codes.

SNOMED has a syntax for post-coordinating such associated terms : the archetype is a more formal expression of the same thing.

As an example comparison take a patients (non-birth) weight for
example.  In pediatric hospitals and for patients receiving chemo it
has always been critical to differentiate between a carefully measured
weight suitable to weight dose calculations versus a weight perhaps
measured adhoc at an outpatient clinic.

This would be a protocol statement - perhaps the device would be stated, the precision would be to 3 decimal places and the weight could be labelled ‘pre-chemotherapy for dose calculation’. It is still a weight…

In SNOMED I find

27113001 : body weight - synonym of Body weight (observable entity)

SNOMED struggles through its scope. Any measureable thing of the body can have an ‘observable entity’, an ‘observation of’ and a ‘procedure to measure’. This is very patchy - one specific problem is that not everything is there as an observable entity. Things like blood pressure have been taken by the terminologists (probably not clinicians) to mean any pressure of blood (central venous, capillary etc) further adding to difficulties.

In OpenEHR we have

openEHR-EHR-OBSERVATION.body_weight.v1.adl

Eric Browne has suggested that we should think about what an archetype records - for that its purpose.

openEHR-EHR-OBSERVATION.body_weight.v1.adl  RECORDS Body weight (observable entity)

It also records information about the state of the person when the measurement was taken - for variables that are deemed important (STATE) and information about the recording (PROTOCOL) - these are patterns in the openEHR archetypes. 

which defines the weight, clothing worn, and device used.

In LOINC we have

LOINC_COMPONENT	PROPERTY	TIME_ASPCT	SYSTEM	SCALE_TYP	METHOD_TYP
3141-9	Body weight	Mass	Pt	^Patient	Qn	Measured
3142-7	Body weight	Mass	Pt	^Patient	Qn	Stated
50064-5	Body weight	Mass	Pt	^Patient	Qn	Ideal
8335-2	Body weight	Mass	Pt	^Patient	Qn	Estimated

At present the archetype for body weight does not have an idea of estimation - but the value can have the idea of approximatation…clearly it is not appropriate to estimate some things - but body weight is one that people might do. It is common for the foetus for instance.

In SNOMED and openEHR a patient can be the information source for any information (this provides the idea of stated).

Ideal is perhaps the idea of GOAL - a separate archetype with Targets. But this also relates to the idea of an expected Peak Flow - and we have been discussing adding this idea to the quantity so that we can talk about % of expected. More work to do here.

You are going to be useful when we come to do the body weight review! :slight_smile:

I expect the SNOMED term could be qualified - but I would have to
search around to figure out with which term would be appropriate
(illustrating the challenge).

Yes - it is challenging to get the label codes from SNOMED.

So I guess ultimately the question boils down to what is
everyone/anyone else doing

- OpenEHR
- OpenEHR + LOINC
- OpenEHR + SNOMED
- All (tough especially as each system has its own set of terms and
concepts of what is what)

Well, interoperability only has a chance if we get a foundation on which to build. LOINC will be used for labs in many places and will map to SNOMED at some point in the future. It is needed for HL7 v2 which is the main communication platform at the moment.
openEHR is designed to work with multiple terminologies - have a look at the DV_TEXT in the data types model and you will see the mapping class. This does mean that coded terms can be associated with data over time.

We have to have a pathway that can bring people along. With archetypes, at least we can lock it down more and more as people see the benefit of semantic interoperability.

And perhaps what does your average EMR doc want?

Well, something simple that can be used effectively by a large range of clinicians in different settings. We can agree the archetype relatively easily if we work at it - it does take some time to see how important features of the reference model bring the data forms into alignment.

I am not talking about ICD9/10 or CPT codes as I need to do those for
billing anyway.

Well, you can have them on a diagnosis or reason for encounter as well.

Cheers, Sam

(attachments)

OceanInformaticsl.JPG