LOINC and SNOMED CT: LOINC Extension – What this means for NHS and Unified Test List
This the second part of our `LOINC and SNOMED CT: LOINC Extension` series. So if you haven’t read the first part of this, you can read it here – it covers the broader vision, why this work matters and the way this might work for different users. In this article, we cover the specifics of how the LOINC Extension might have a special significance for the UK and the pathology standards used in the NHS.
For those of you who have followed some of our previous posts, you might be aware that the UK did not have significant LOINC usage. So when NHS England looked at the suitable standards for sharing pathology (lab) results, they decided to use SNOMED CT as the reference terminology. This is not a SNOMED CT vs LOINC decision, but rather a case of using a single reference terminology (SNOMED CT) for all of health and care. So NHS Digital created the Unified Test List (#UTL), a subset of SNOMED CT concepts that cover lab results and requests. You can read more about the UTL here. It is important to note that the interoperability standards for England are not just `codes`, but a combination of the UTL (for codes), HL7 FHIR (for messaging) and UCUM (for units).
Following the initial work on the three standards and some pilot tests, the DAPB approved these pathology standards for use in England. Based on the history of how similar standards approvals resulted in adoption of SNOMED CT in primary care, we can assume that the plan is for these standards to be adopted in laboratory medicine (pathology). As an aside, but a useful analogy, it is useful to note that the UK uses NICIP for radiology/imaging codes – radiology being part of the broader diagnostic medicine space. It is worth saying that the NICIP is SNOMED based for those who aren’t familiar with it. The biggest difference being that the UTL doesn’t have the metadata baggage of the NICIP or the short codes e.g. ‘CAAAG’ for systems which can’t handle a 64 bit integer.
The LOINC Extension – LOINC in SNOMED CT
The recently announced LOINC Extension (aka LOINC content expressed as SNOMED CT) promises to make the entirety of LOINC (Observables) content available to all SNOMED CT users. We have unpacked this statement in the previous article and the most important aspect here is the question:
What does having all of LOINC content accessible via SNOMED CT as an extension mean for existing Observable content in SNOMED CT?
What happens if someone were to combine the LOINC extension with the Unified Test List content, given they are both in a sense extension built on the International Edition?
In order to answer these questions, we have to understand some aspects of the underlying common `model` that exists between LOINC and SNOMED CT. Then we have to understand what happens when we try to express/create LOINC concepts as fully defined SNOMED CT concepts (in the LOINC extension or the UTL).
A common model for Lab Medicine
We are a team that have worked in this space for a while – having been involved in the original design of the UTL. For a while, we have recognised a `common underlying` model that seems to be shared across LOINC, SNOMED CT and even the NPU (lab coding standard used in Scandinavia). At the risk of oversimplifying things (since nothing is simple in terminologies/ontologies), there is a fundamental `set of attributes` to lab results:
- Component: the thing you are analysing/measuring etc – most often what you report on (e.g. substance like Sodium). LOINC sometimes call this the `analyte`
- Property: the property of the component you are trying to measure
- Specimen: the specimen or substrate where that thing you are measuring exists. LOINC calls this `system`
- Technique: the method/process by which you are analysing
You need to note that `what a lab test is` is a philosophically different question. So the attributes above do not describe the entirety of what a lab specialist would call a `test`. This is why you see other `attributes` like `units`, `scale`, `pre-condition/adjustments` etc added in various coding systems.
An ontology is a conceptualisation of a world/model as defined by the ontology author (usually for a use case). So if you change the definition of a `lab test` the attributes and corresponding lab ontology models would change!
LOINC Concept in SNOMED CT
Now that you understand the above model, we can look at what a LOINC concept (lab result) would look like in SNOMED CT. It goes without saying, that since a lab result is what you observe/report, they belong to the `Observable entity` hierarchy in SNOMED CT. The machine readable concept model (MRCM) of SNOMED CT specifies a large number of valid attributes for an Observable Entity concept. However, for the purposes of this conversation, the following attributes are the most relevant:
- Inheres in
- There is a much longer list of 23 sanctioned attributes here
So if we look at how LOINC models a test `Amiodarone [Moles/volume] in Serum or Plasma`, it would appear like this (screenshots from Pathnexus – our lab data harmonisation platform).
So if the LOINC extension were to `fully define/model` LOINC concepts in the SNOMED CT world/model, they would be transformed into their SNOMED CT equivalent attributes as we see below in the UTL. Please note that `inheres in` adds some complexity, where `specimens` are not always the where the thing you are measuring that exists.
Inheres in – if you measure `folate in a red blood cell (RBC)`, the `folate` you are measuring is technically in the RBC, even if the specimen you process is blood!
Unified Test List Concept Model
The UTL which is a subset of SNOMED CT of course relies on the underlying SNOMED CT model to represent lab requests as Observable entities. So to complete the comparison, let us look at what a UTL concept would look like in SNOMED CT – for a broadly equivalent concept `Mass concentration of amiodarone in plasma (observable entity)`.
Note of course that this concept is specific to `plasma` but you understand the broader premise. We’ll come to the difference in specimen later.
In case you are curious how the corresponding `LOINC Test: Amiodarone [Moles/volume] in Serum or Plasma` could be created, here is a brief explanation. Given the power of SNOMED CT and its hierarchical structure, this allows us to create new tests and to validate them on the fly – thanks to the machine readable concept model. Of course, if this exists LOINC, the LOINC Extension will likely create this concept for us using a similar approach.
If you want to create standardised tests in your LIMS or as part of your Pathology standardisation then check out Pathnexus – you can standardise your tests (and data) to either LOINC or SNOMED CT!
Bringing LOINC and UTL together
Now that we have covered those things, we can see that the concepts in the UTL might overlap or be identical with ones that might be added in the LOINC extension. For example, let us look at what content exists for `amiodarone` in LOINC and UTL respectively.
Here are matches for Amiodarone in LOINC – we have 15 matches and you can see they are from across many systems/specimens, techniques etc.
In the UTL, there are 2 matches for Amiodarone, see below.
So the questions here are:
- How do we detect which concepts in the UTL that are semantically equivalent to ones in the LOINC extension (when they are created)?
- How could we associate existing UTL concepts with LOINC extension content, if they are not equivalent?
Here comes the Description Logic Classifier – yes, I did bring up that `description logic` word that most people try to avoid (or mention in whispers). Suffice to say, that the DL classifier/reasoner is our little friend that’ll solve this problem for us. But first a little digression.
Since everyone says AI is so cool (and gets excited) how Word2Vec can infer seemingly magical associations, let’s bring in the `semantic engine` that does the same thing. This is the DL Classifier – short for Description Logic Classifier. For some reason, when I first trained, I learned to refer to this as the `DL Reasoner` and not as the `classifier`, cos in effect it `reasons` over the logical statements in a given ontology (axioms about the world) and tells us stuff about them. As an oversimplified example, if you tell an ontology (assert via axioms) that:
- Bacterial Pneumonia is a Pneumonia caused by a `Bacterium`
- Pneumococcus pneumoniae is a type of `Bacterium`.
Then you create a concept like:
- Pneumococcus pneumoniae: `Pneumonia` `caused by` `Pneumococcus pneumoniae`
You could ask the DL Reasoner what type of thing it is and it would tell you that the following relationship exists between them.
|__ Pneumococcus pneumoniae
If you then add another concept deliberately called `PneumococcAL pneumoniae` like:
- PneumococcAL pneumoniae: `Pneumonia` `caused by` `Pneumococcus pneumoniae`
And you ask the DL reasoner what it knows about `PneumococcAL pneumoniae`, it will tell you the following things:
|__ PneumococcAL pneumoniae = Pneumococcus pneumoniae
Yes, thats right the DL reasoner does not care that we named them different, so it would tell us (dumb humans) that logically `Pneumococcus pneumoniae` and `PneumococcAL pneumoniae` are one and the same thing!!
Relationship between UTL and LOINC concepts
Now if you applied the same logic to content in the LOINC extension and the UTL extension, assuming they are modeled using the same SNOMED CT International concepts, then the DL reasoner would identify:
- Concepts that are exactly the same as each other.
- Concepts that are sub/super types of each other.
We don’t have to do this work, before making the following observations
- It is possible that a significant fraction of the ~3,800 concepts in the UTL might also exist in the future LOINC extension.
- This is because the UTL content was derived from empirical usage of lab results from large trusts in the UK. Since LOINC uses the same approach, there is likely a good overlap between this content.
- There will likely be many cases where the UTL content might end up being a `sub type` of LOINC concepts.
- This is because within LOINC, there are a lot of what we called `grouper` concepts. These are tests that do not necessarily make the distinction between where it was measured – blood or serum or plasma. However, the UTL tries to create separate concepts for such tests – which by the way also can exist in the UTL.
- However, we’d think in many cases (given the modeling is correct), these `more specific` tests will likely become subtypes of the LOINC grouper (or less specific) concepts.
The more interesting questions here are:
- Should we use the LOINC extension alongside the UTL content in the UK?
- Should existing UTL content be promoted to the SNOMED CT International edition, so they can be aligned with the LOINC extension content first?
- If the UK NRC continues to add UTL concepts which also exist in the LOINC extension, what if any is the impact when say a supplier tries to combine them?
- If the LOINC Extension work results in content changes to the SNOMED CT International hierarchies, how would these impact the UTL extension, which indirectly depends on the SNOMED CT International Edition?
- What are the other bits of LOINC that we could now take advantage of as mentioned in our previous post.
Lots of food for thought here, but this article is already a lot longer than it was intended to be. So we will pick some of these questions in our next article and also share early metrics on the possible extent of UTL – LOINC content overlap. If you are interested in the upcoming article, please follow us on LinkedIn or subscribe to our newsletter on our website. Also, are you as excited about the potential of this announcement? Please comment if you agree with what we are saying or if you have questions. We’d love to hear from you!