Dose form Intended sites vs. Routes of administration

[dmcmurtrie (Dion McMurtrie)]

Following this discussion, I want to return to Matt’s core question: does distinguishing between dose form intended site and route of administration actually support SNOMED CT’s clinical use cases? Or does the current split create more barriers than benefits?

The “intended site” concepts are really routes
As Matt has shown, many of the concepts in the intended site hierarchy aren’t sites at all. “Oral” doesn’t mean “pertaining to the mouth” - it means “taken by mouth for systemic effect via the enteral route.”

“Parenteral” is a route concept. “Cutaneous” and “Transdermal” describe routes, not anatomical sites. Meanwhile “Oromucosal” genuinely pertains to the mouth but is not subsumed by “Oral.” The hierarchy is inconsistently aligned with both site and route semantics - it’s neither one nor the other.

As Stuart has pointed out, SNOMED has already deviated from EDQM by building a hierarchy of intended sites where EDQM maintains a flat list - and in some cases SNOMED’s choices already differ from EDQM’s (e.g., “intraocular” in SNOMED where EDQM uses “ocular”). So the argument that we must maintain intended site for EDQM alignment is already compromised. We have translated a regulatory data model into a concept model, and that translation has introduced its own inconsistencies.

The practical barrier
Dose forms are defined using intended site, but clinical workflows, FHIR resources (MedicationRequest.dosageInstruction.route, MedicinalProductDefinition.route), and prescribing systems all use route. Because the two hierarchies do not align, implementers cannot determine the applicable route from a dose form via ECL. Systems end up maintaining local mappings between routes and dose forms - a shared burden borne by every national extension and every implementing system. This is a problem that should be solved once, centrally, in the terminology.

Julie raised the important point that dose forms embody pharmaceutical quality standards - sterility, pH, osmolarity. This is a real consideration - but it argues for retaining the information, not the specific attribute. Replacing Has dose form intended site = Parenteral with Has applicable route = Parenteral route carries the same implication about sterility requirements. The quality inference follows from where the product is administered, regardless of whether we call that property “intended site” or “route.”

This can (and must) be completely separate from the concept of licensed route from the regulator, which is different again and also may change over time for the same product.

I’d love someone to clearly articulate a use case in scope for SNOMED CT that requires both intended site and route as separate concepts. All clinical use cases I can identify could be supported by route alone.

A path forward
The proposal isn’t to remove an axis from dose form definitions - it’s to replace intended site with route, which serves both the classification purpose and the clinical workflow purpose.

• Has basic dose form = Tablet + Has applicable route = Vaginal route

achieves the same classification outcome as

• Has basic dose form = Tablet + Has dose form intended site = Vaginal

but uses concepts that align with the rest of the terminology and the broader standards ecosystem. The ontological benefit is identical. The practical benefit is better.

Specifically:

1. Replace Has dose form intended site with Has applicable route on dose form concepts, using concepts from the existing route of administration hierarchy. The formulation characteristics (basic dose form, state of matter, release characteristics, administration method) remain as they are - route replaces only the intended site axis.
2. Make route inferrable on products via GCIs. Rather than stating route on every product concept, use necessarily-true axioms to assert that products with a given manufactured dose form necessarily have applicable route(s). For example: “Product with manufactured dose form = Vaginal tablet → Has applicable route = Vaginal route.” This maintains the route information in one place, applies it automatically through classification, and handles products with multiple applicable routes naturally - Stuart’s vancomycin example, licensed for both IV and oral, would simply have multiple GCIs. But again this would be applicable not licensed routes which would need to be handled separately by the national extensions dealing with specific regulatory data.
3. Create a mapping between route and intended site for EDQM compatibility so that it is possible to transit that direction. Although really all that is needed is a mapping between EDQM forms and SNOMED CT forms, and the rest flows. It is easier to maintain this than everyone else having to maintain a way to navigate between intended site and route.

This directly supports prescribing and decision support use cases, aligns with how FHIR and IPS already handle route, and resolves the mapping burden that every implementing nation currently bears independently.

[krees (Karen Rees)]

Thanks for the comments Dion.

Two key issues I see in taking this approach:

1. Further significant deviation from IDMP and EDQM

2. Clinical safety issues - Will changes bring about any potential clinical safety concerns that need to be considered?

Dose form intended site and route of administration serve different use cases as Julie has pointed out previously. (I agree the name ‘dose form intended site’ is not the best use of terminology!)

Both IDMP and EDQM make the distinction - one enables modelling of a dose form and supports regulatory processes (it is not directly applicable to a patient) and the other (ROA) pertains to a clinical situation and the patient, generally at the time of prescribing.

I will be interested in any other feedback regarding the two issues and how any changes, such as those suggested above by Dion, may impact (positively and negatively) other users.

[dmcmurtrie (Dion McMurtrie)]

Thanks Karen.

On the two issues you’ve raised:

1. “Further significant deviation from IDMP and EDQM”

I’d push back on the framing here. Compatibility and alignment with IDMP/EDQM does not require identical representation - and in fact, identical representation transplanted from IDMP or EDQM into SNOMED CT would be problematic. IDMP and EDQM are data models designed primarily for regulatory use cases. SNOMED CT is an ontology with a concept model, designed for clinical use cases. These are fundamentally different semantic frameworks.

What is required for interoperability is that each side - IDMP/EDQM on one hand, SNOMED CT on the other - has sets of concepts with equivalent meanings, represented using each system’s own semantics. These form anchor points of direct equivalence across the two systems. SNOMED CT’s own Medicinal Product Model specification acknowledges this explicitly - compatibility with the IDMP model

“does not entail full and exact compliance; there would be little value in exact duplication.”

Each system needs to achieve what it needs to achieve for its own use cases within its own semantic framework, while providing those anchors of equivalence for smooth interoperability.

It’s also worth looking at how IDMP is actually implemented in practice. FHIR’s AdministrableProductDefinition - which the spec explicitly states corresponds to the IDMP Pharmaceutical Product (ISO 11615/11616) - has routeOfAdministration (1..*) as a required element, bound to SNOMED CT Route Codes. There is no “intended site” element anywhere in the resource. The FHIR IDMP mapping page confirms that this maps to “Pharmaceutical Product.Route of Administration” in IDMP. Intended site is a characteristic used internally by EDQM (ISO 11239) to classify dose forms - it is baked into the dose form code itself, not a separate concept transmitted at the product level. When HL7’s Biomedical Research and Regulation work group built FHIR resources specifically to implement IDMP, they used route. And the same SNOMED CT Route Codes value set is used consistently across the entire FHIR medication model - from MedicinalProductDefinition.route through to MedicationRequest.dosageInstruction.route and MedicationAdministration.dosage.route. SNOMED CT dose forms using route of administration would therefore align more closely with how IDMP is actually realised in the interoperability standards everyone is implementing, not less.

The main interoperability surface between SNOMED CT and EDQM is on the dose forms themselves. We already have the EDQM-to-SNOMED CT dose form map with ~310 mappings, and many of these are straightforward because the dose form concepts often share identical names. If the dose forms are mapped (and they are), does the attribute used to classify those dose forms on the SNOMED CT side materially affect that interoperability? I’d genuinely like to understand: what specific problem does replacing intended site with route on SNOMED CT dose forms cause for IDMP/EDQM alignment? What does this “further significant deviation” actually break in practice?

As Stuart has already demonstrated, we’ve already deviated from EDQM in this space - building a hierarchy where EDQM has a flat list, using “intraocular” where EDQM uses “ocular”, and so on. I’m not aware that this has caused a significant problem. The question should be: does the proposed change make the dose form map harder to produce or use? I don’t think it does - but I’m open to hearing a concrete scenario where it would.

2. Clinical safety

This is an important consideration. Are you aware of any specific clinical safety concerns you can articulate? I ask because I think the clinical safety argument actually runs in favour of alignment with route. Clinical systems, FHIR resources, and prescribing workflows all use route of administration. The current disconnect between intended site on dose forms and route on prescriptions is itself a source of implementation complexity - and implementation complexity is where clinical safety risks tend to hide. Every national extension independently maintaining mappings between these two hierarchies is duplicated effort and duplicated risk.

On your point about different use cases

You’ve noted that intended site supports regulatory processes and route of administration pertains to clinical situations. I think this actually supports the proposal rather than argues against it. SNOMED CT is designed for clinical purposes. It has to work with standards like FHIR that are serving clinical use cases - MedicationRequest.dosageInstruction.route, MedicinalProductDefinition.route, AdministrableProductDefinition.routeOfAdministration. If intended site serves regulatory uses and route serves clinical uses, then SNOMED CT - as a clinical terminology - should be optimising for the clinical representation. We won’t be successful as a clinical terminology if we’re in lockstep with regulatory modelling at the expense of the clinical use cases and interoperability standards that are squarely in SNOMED CT’s wheelhouse.

To be clear on what’s being proposed: the replacement attribute on the dose form would represent the route(s) that the dose form is formulated for - what the form is designed and quality-controlled to support. This is exactly as fixed and definitional as the current intended site. An oral tablet doesn’t change its formulation-intended route any more than it changes its intended site. The pharmaceutical quality implications that Julie rightly highlighted - sterility, pH, osmolarity - are just as implied by the route as by the intended site. “Has applicable route = Parenteral route” carries the same sterility implication as “Has dose form intended site = Parenteral”. No information is lost. What varies clinically - the actual route used for a particular patient, including off-label use - is a separate concept that belongs on the prescription or administration record, not on the dose form definition. The proposal is to align the dose form’s fixed attribute with the terminology used everywhere else, not to introduce clinical variability into the dose form model.

Having our cake and eating it too

But I think we can go further than just replacing one attribute with another. Rather than framing this as “SNOMED CT deviates from IDMP”, we should be asking: can we infer the EDQM intended site from the stated route of administration? If so, the EDQM intended site link is preserved - it’s just derived rather than directly stated.

This seems very achievable. The EDQM intended sites are a small, flat list (fewer than 30 concepts, many of which are veterinary-only or edge cases). The mapping from route to intended site is largely systematic - oral route maps to oral intended site, parenteral routes (IV, IM, SC) map to parenteral intended site, ocular route maps to ocular intended site, and so on through vaginal, nasal, rectal, auricular, cutaneous, transdermal, pulmonary, dental, sublingual, buccal, and the rest. Some edge cases would need manual attention - oromucosal vs oral being the most obvious, and a few of the more specialised sites like gastric, gastroenteral, and intestinal (which relate to device-administered access) - but the bulk of the mapping is mechanical.

This could be implemented a few ways:

(a) A maintained reference set mapping each SNOMED CT route concept to its corresponding EDQM intended site. This is a small, relatively static artefact - the EDQM intended site list rarely changes, and the mapping is largely obvious from the semantics. This reference set could be maintained centrally.

(b) Hierarchy organisation - if the route of administration hierarchy were organised so that grouper concepts correspond to EDQM intended sites (e.g., a “Parenteral route” grouper subsuming IV, IM, SC etc.), then the intended site becomes inferrable via subsumption. This is arguably the most ontologically clean approach, though it would require creating some grouper concepts that don’t currently exist (notably “Parenteral route”).

(c) Computable rules / GCIs - a set of axioms asserting that dose forms with a given route necessarily correspond to a given intended site for the purposes of EDQM mapping. This keeps the mapping logic inside the terminology.

The fundamental question

But before we spend time designing inference mechanisms, I think the more fundamental question is: if SNOMED CT dose forms had route of administration instead of intended site, and there was no inferred intended site at all - given that the EDQM-to-SNOMED CT dose form map already exists and maps dose form to dose form directly - would that actually cause a problem for IDMP/EDQM compatibility or interoperability? The dose form code carries the intended site information implicitly (that’s how EDQM classifies its own dose forms internally). FHIR’s IDMP resources don’t transmit intended site as a separate element. So what concrete use case requires intended site to be separately computable on the SNOMED CT side?

If the group can identify such a use case, the inference approaches above show how it could be addressed. If no such use case exists, then we’re maintaining a parallel hierarchy that serves no interoperability purpose while creating real barriers to the clinical use cases that SNOMED CT was designed for.

I’d welcome concrete examples from anyone where this approach would fall short so I can understand the issue better.

[greynoso (Guillermo)]

Most local drug dictionaries require an appropriate interpretation of the source context and the corresponding representation in SNOMED. Many drug dictionaries have general dose forms (e.g., solution) that need to be combined with a route of administration to map to an SNOMED dose form; others have multiple routes of administration, most of them non-definitional and with a different meaning as discussed before, so there is a need to define the intended site or the intended route. I fully agree with Julie and Karen, but I also agree with some of Dion and Matt points. It is true that we need more granular and more general dose forms to enable different implementations to use the appropriate level for each case. There are countries that have dose form = Injectable and others that want a more precise DF like Intramuscular, and all are valid use cases. We need to find a non-destructive, backward compatible way to enable flexibility without chaos. While I would not merge routes and intended sites, enriching the intended sites with more granular “intended routes” dose form hierarchy is a good option. It is true that GCIs or additional axioms are also possible, but we should find a compromise that avoids radical changes or rigid solutions. Echologically friendly evolution. We can preserve the IDMP compatibility and extend the dose form hierarchy if needed. The challenge is that the current dose form model is very much aligned with EDQM, but it could be refined using the same principles.