There have been approximately 8,351 clinical drugs in the SNOMED International release for over a year now. In 2021, there were 7,077.
For over 7 years, NLM has been instantiating RxNorm generics into these releases on a monthly basis in support of alignments and the RxNorm extension.
My question is simple, and definitely not to offend:
Are the number of SNOMED clinical drugs expected to rise, or have we plateaued?
Hi Robert,
First, I am speaking from my personal experience and in no way representing anyone else. But most of my current work supporting others depends on continuing improvements to the clinical drugs model and the content coverage in the International release. To improve the coverage we need to coordinate efforts among many parties to submit validated content to SI. In addition to the 8,351 clinical drugs in the core, a quick lookup shows that several extensions have additional clinical drugs in their module: Argentina has 4331, Canada 2146, Ireland 3155, Norway 4813, and I guess that the US might have 4-5 thousand (looking only at the current prescribable, public domain subset of RXNorm for simplicity - but understimating the real gap for now). Uruguay has its own VMP layer, but it would depend on the addition of about 1,200 clinical drugs to improve alignment. There are many overlaps between those extensions that should be promoted to the core, and there are also additional implied layers of clinical drugs (less granular clinical drugs to handle alignment with more general ingredients and dose forms, more granular clinical drugs for liquid dose forms that in the current model require concentration and presentation strength representation - most of the Argentina content falls in this category as it is currently out of scope for the IE content).
So we have overlapping content among extensions that we should identify and then collaborate with SI for their promotion. The low-hanging fruit here are solid dose forms, particularly tablets and capsules that are easy to reconcile across extensions and usually cover 40-50% of the content.
Looking at examples from several extensions, there are variations in how clinical drugs are represented with oral and parenteral liquid dose forms. Some of them depend on country-specific prescribing, dispensing, usage patterns, and local regulations. For example, the original model has unit of presentation size at the CD level, but some countries need to represent that at the packaging level and some guidance recommended the later approach that leaded to some variability that needs to be harmonized to provide flexibility for implementations. In many cases, IE dose forms are too specific but it would be possible to create higher level dose forms (e.g. injectable) that are used in many countries.
Another pattern that contributes to modeling diversity and difficult to recognize equivalencies is the usage of presentation strength in single dose liquid dose form units of presentation. An ampoule of 80 mg of gentamicin could be considered sufficiently defined using only presentation strength, a simplification reflecting local usage in many countries. On the other hand, based on other scenarios the IE has concentration groupers (e.g. gentamicin 40 mg/ml) defined only with concentration strength (the presentation strength-only definition would not classify infer it even if the unit of presentation size is 2 ml. So countries would need, for those cases, to create a gentamicin 40 mg/ml, 2 ml (80 mg) etc. with both representations (that accounts for most of Argentina CDs specializations - but in the early stages of the model circa 2017-2018 it used the presentation strength like Ireland and others are using now. So now we have significantly more experience and can identify areas of the model or the content that could be improved through collaboration, respecting the local needs for adopting variants but probably agreeing in higher levels of alignment between the broad gap between MP-only (form?) and CDs.
Another interesting pattern to discuss is the use of the unit of presentation as the denominator for presentation strength. It is required, but it is too specific sometimes. In RXNORM it follows the UCUM way (1, per unit of presentation) rather than specifying the specific unit of presentation (that is represented elsewhere in the same CD). At the generic level, that denominator unit might produce a difference at the generic level that might not be relevant (e.g. whether the denominator is a tablet or a caplet or a capsule, the per unit approach might be sufficient to help unify differences that are relevant at the real product level rather and add too much detail at the generic CD level.
Finally, many countries would expect more granular dose forms at the real product level, and some grouper dose forms at the CD (or SCD) level. That is somewhat heterogeneous, but the point is that perhaps the international CD doesnât have to specify the most specific dose form present in the real product that motivated the creation of the virtual level, perhaps a grouper dose form that better accomodates controlled variability for non-significant differences might be good. Recognizing that defining what is significant and non-significant is subjective and use case specific.
So the number of SNOMED clinical drugs is expected to rise significantly as we improve collaboration and enhance the model to facilitate alignment. We have plateaued because of the significant effort if in isolation.
I have requested an affiliate license of Ireland and Norway extensions, so I can merge them with a few national CD extensions and classify them together and identify part of the overlap and get back to your question with more precisions. On the other hand, I believe that RXNORM (and dm+d, the original base for the validation and population of the CD content in the past decade) would be extremely valuable to validate and benchmark content coverage. However, as you know, there are significant gaps that need to be addressed.
Thank you for putting an interesting wake up call topic on the table. I am looking forward to continuing our discussions and consensus building to decrease alignment barriers and duplicate or diverging efforts.
Sounds about right, Guillermo.
RxNorm has 12,128 generics in the Prescribable subset.
For the purposes of instantiating generics on the SNOMED IDMP, we apply all generics from the RxNorm Active subset. That number is 17,587.
As always, you can find this monthly update (discussed at various Expos) here:
Yes, similarly AMT has ~8,000 - although with different granularity again.
Thanks for raising this - itâs an interesting question, but I think it risks focusing on a symptom rather than the underlying issue.
The number of Clinical Drug (CD) concepts isnât particularly meaningful in isolation. The more important question is whether CD is the right level for stable international content.
From practical experience across Australia, the UK, and Ireland - and experimenting with these alongside RxNorm - thereâs a consistent pattern:
This is largely because CD sits in the zone of real jurisdictional variation - strength representation, BoSS choices, dose form granularity, available products, local regulatory rules, and differing prescribing/reimbursement rules/frameworks. Even with perfectly aligned modelling, these wonât align cleanly across countries globally.
There may be enough consistency within parts of Europe (e.g. under EMA) for this to work more effectively, and the SNOMED International drug model clearly leans in that direction through close alignment with IDMP/EDQM. But aligning tightly to those regulatory standards makes interoperability harder for those of us working in different regulatory environments. And itâs not clear this fully resolves the problem even within Europe.
So CD doesnât consistently function as a stable shared layer in the way it is currently positioned. Even if the model were further generalised, real product variation would still limit alignment. And even with a perfectly harmonised model, the union of all CDs would include many products that donât exist in a given jurisdiction - creating additional complexity rather than value.
We already see this in Australia, where questions come up about IE CD concepts that donât map to the local reality and how to avoid them.
In that context, asking âhow many CDs should there be?â is difficult to answer meaningfully
Fortunately, interoperability approaches already account for this using decomposed data (ingredient, strength, form), and CDS/analytics largely operate at MP/MPF.
So in practice, CD becomes more about local implementation (generic prescribing) than international interoperability.
A more useful framing internationally might be
At the same time, we need to be clear about what international CDs are intended for, and how they can be safely used with or without national extensions across SNOMED International members.
There are also real risks and costs here - relying on externally managed CD identifiers in core systems (e.g. prescribing or reimbursement) can create significant disruption if those concepts are remodelled, as the UK experienced. This has led some national extensions to deliberately try to insulate themselves from this international content - I believe we need to find ways to enable extension builders to do this while still interoperating with and gaining and delivering value to the international layer. If not, it is a nuisance not a benefit.
CDs are clearly important for prescribing systems, but this suggests that trying to optimise or standardise the number of CDs internationally may not be the right goal.
More useful questions might be which use cases require a genuinely globally shared CD layer?
Or viewed another way, what level of alignment is realistically achievable, and which use cases would that enable?
Thanks for your comments here. From a historical perspective, when SI first embarked on the development of a drug model back in 2016, support for the prescribing use case was explicitly out of scope, for the very reasons you have listed in your comments (i.e. jurisdictional variability); however, we had some pressure from some members (those without their own drug extension) to support the CD level, which has led to the situation we have now.
I think your suggestion about determining what level should be supported by the international reelase is good, and the discussion should be had to avoid going down a road where complexity overwhelms utility.
Different countries and regions might benefit from diverse alignment profiles. I understand the Australian alignment strategy is sound considering that AMT has been around for about 15 years, that its model has evolved over 4 major versions, using SNOMED CT RF2 mechanisms (that works on standard terminology servers/FHIR), and that it has broad adoption, works well, and already has an adapted CD-like layer. Limited or negative benefit in changing that, and aligning at the MPF level seems logical. If at some point there is a use case for alignment at a different level, that could be achieved with additional axioms (like the Canadian implementation, which models the locally registered strength (e.g. X 1% cream) and eventually adds axioms aligned with IE editorial policies (e.g. X 10 mg/g cream) to infer core CDs when present, or to create those CDs locally otherwise).
On the other side of the spectrum is Argentina drug extension. No previous structured central distribution of local drug dictionaries, implemented national extension in 2018, fully aligned from scratch, 67,000 concepts, and getting similar benefits to Australian AMT in a few releases leveraging core CDs and extending them as appropriate. Modeling clinical drugs requires significant resources, so for countries that would benefit from that level of alignment there is no point in reinventing the wheel many times so everyone has its own layer. Also, in our case, other countries in the region would benefit from a common CD layer. This does not contradict the Australian strategy; it simply points out that it is not the only strategy that could be generalized to suggest that CDs should not be maintained in the International release. It is true that there are regional variations, different granularities, and a complex set of challenges. We need to resolve or improve them, and countries would define their alignment strategy based on local requirements.
It is true, though, that the international release population of representative CDs (or any regional effort working at that level of abstraction) is difficult to maintain without discovering and validating the abstract levels based on the real products in specific markets. It is also true that the model is flexible, but it leaves room for different modeling patterns (for example, single-dose continuous-dose forms). As mentioned in my previous post, we classified the drug extensions of Norway, Canada, Ireland, and Argentina together and found only ~600 equivalencies, most of which were CDs or MPOnly/MPOnly forms. Most of the equivalencies were found in solid dose forms because of the continuous-dose-form modeling patterns.
The suggestion that the International Release should stop supporting CDs is a significant concern that probably cannot be resolved in the short term. And as suggested by the original poster, CD coverage is evolving slowly, in our view because countries that would want to align at the CD level should contribute to the coverage based on their content. In the meantime, there are areas where the International Release layer can benefit all alignment strategies: the building blocks we all need to be updated, and with near complete coverage: substances, MP contains, MPOnly, and MPOnlyForm levels. Even if the CDs are not in the core, if there is proof from the FDA, EMA, Australia, and many other countries that those higher-level classes are useful abstractions, we should fast-track coverage and stability.
Based on my understanding, there is no CRS with an ability to create full-fledged CDs in batch. Our SLA prevents any kind of contribution over 50 concepts per year or quarter.
Anyone savvy enough with the OWLAPI (and serializing RF2 with their own namespace) could consider our release as a contribution, with caveats being non-existent MPs, MPFs, and even Product classes to support specific RxNorm CDs (and of course the file not being reviewed by pharmacists ((USP dose forms to EDQM/MDFs)) ).
This would be a great project for anyone looking to contribute content from RxNorm. Based on internal discussions at NLM, no one has asked for a US Drug Extension. But we keep producing the RxNorm-in-OWL each month.
If SI were to accept a contribution of fully modeled clinical drugs aligned with the rest of the International SCDM, that could be handled as a RF2 delta submission in the context of a collaborative editing project, and the potential size of the contribution would likely exceed 2,000 clinical drugs. It would not be a typical CRS NRC submission.
Now, for SI to consider such a contribution, we would need to validate the content across several extensions OR authoritative sources even if they are not currently published as formal extensions.
That is were RXNorm content could help the validation effort towards ensuring the proposed coverage increase is beneficial for countries aligned or intending to align with the international model at the CD level. It would also help validate the need for MPOnly and MPcontains superclasses.
My team started with the prescribable subset of RXnorm because we are not sure if there are source-specific IP/licensing restrictions on content exceeding the public domain prescribable subset.
As an example, looking at the RRF there are about 1750 SCD with âinjectionâ dose forms. So we will use the OWL API to understand how you managed those dose forms mappings. I understand you have made the mappings, assumptions, known issues, etc. used for the RRF to OWL transformations in a previous communication, but if you could make that available again that would be much appreciated.
Our approach would be to transform RRF or OWL to an RF2 extension format to make it easy to compare it with other extensions. We plan to undertake that project in June 2026, as a learning and research exercise.
Great! I would encourage the use of the RxNav Prescribable API. Anything returned with restrictions would be returned âPROPRIETARYâ.
Start with REST method getAllConceptsByTTY . Scope is Active: concepts in the current RxNorm data set that have an atom with SAB=RXNORM and SUPPRESS=N (no worries about proprietary content here). And go from there.
The map-rxnorm-to-snomed source out on GitHub is circa 2023, and documented the best we could given time constraints. It employs the public RxNorm REST APIs. When tested from an outside IP in on an older cloud architecture we found the ExponentialBackoff solution for failed calls to be finicky. (Hopefully this has been resolved as we migrated to a better cloud solution.) You will find the configured dose form mappings in there. Also, the deployment run documentation has a typo (swap options -jar and -Xmx âŚ-Xmx shouldnât even be necessary these days). Weâll think about updating this source if the community finds it helpful.
Unrelated: Weâre looking for Expo abstract ideas. (will an extension be made?). 