Ambiguous Concept

I do not understand what is meant by "incoming historical associations" - when promoting an extension to International?  -when adding new International associations?  These logical queries would be useful in managing a SNOMED CT INSTANCE-LEVEL datamart but I fail to see their relevance for an managing a CONCEPT-LEVEL reference ontology auch as the international release.  If this is meant to be advice to extension managers or National release centers, then I think this should be explained.

Looks like there is a typo in the second example. should be 69878008 |Polycystic ovaries (disorder)| (as per the resolution).

(you could arguably include 830049001| polycystic bilateral ovaries (disorder) | as an association too…)

The third example also has a typo in the “conjunction inactivation step”, I expect it should be “POSSIBLY_EQUIVALENT_TO”.
As per the combinatorial logic described

(A) POSSIBLY_EQUIVALENT_TO (B OR C) and (B) POSSIBLY_EQUIVALENT_TO (D OR E) implies A POSSIBLY_EQUIVALENT_TO (C OR D OR E)

Hi Matt,

Many thanks for the feedback and for identifying the typo in the first example, now corrected.

With respect to the use of "EQUIVALENT_TO" in the final example, please see my response to the same query raised on the Confluence page for Classification Concept.

Additional notes section: ‘Share difficult ambiguous inactivations in the internal editor's meetings’. This guidance will be used by those managing extensions, working in the community edition and the undertaking the authoring course. While it would be understood this relates our internal content team, consider some generic advice for others - perhaps the help desk?

Final example - suggest a new example.
• Is delivered equivalent to a delivery procedure?
• Occipitoanterior presentation is the “normal’’ presentation. The concept 199779004 |Persistent occipitoposterior or occipitoanterior position - delivered (disorder)| is more likely to represent the persistent OP position or the OA (without persistent) position. Does persistent OA even exist from a clinical recording perspective? Maria Braithwaite
• The concept 37235006 |Occipitoposterior position (finding)| in the EQUIVALENT_TO section is incorrect. It should be 70068004 |Persistent occipitoposterior position (disorder)|.

I agree with Cathy Richardson that a new example might be better instead of the third one. Which has a lot going on...
The more I look at it, I'm not excited about the association to "delivery procedure". It's really a finding about fetal presentation...
If the baby hasn't been delivered, then it's a  current finding relative/relevant to the ongoing obstetric care.
If the baby has "now been delivered" it's a historical comment, most likely attempting to describe the labour...

I Believe EQUIVALENT TO is now PARTIALLY EQUIVALENT TO - is that correct?

- no EQUIVALENT TO is stated on the parent page Draft Editorial Guidance in Historical associations section

Hi Ian Spiers, many thanks, both pages updated.

Thanks for highlighting the PARTIALLY EQUIVALENT TO element of the draft guidance - it troubles me.

Both the 'summary definition for the author' and the 'implications for the vendor/user' sections in the parent page guidance state that the PARTIALLY EQUIVALENT TO mechanism will be invoked where a freshly-inactivated concept represents a conjunction, and that the set of active targets chosen "...ensure exact semantic equivalence between the inactivated concept and the sum of its replacements...".

Conjunction has been a feature of SNOMED's DL since the mid 90's. Given that the in-scope concepts you wish to inactivate are conjunctions that can be fully represented by a set of active targets (that is, they are not themselves classification-derived content that feature residual/closure category problems), why not just leave them active (with the set of PARTIALLY EQUIVALENT TO targets as their set of asserted parents, or by virtue of their asserted role modelling, somewhere in their ancestry)?

There will always be a need for plenty of clinically pragmatic conjunctions (be they disorders, syndromes, multi-part procedures or whatever), some of which may have been coined during the development of classifications. I also know they pose an editorial boundary challenge. However I don't see how this proposal helps - notably how the international community will reproducibly understand (both as users and editors) which ones get inactivated and which do not.

If the judgement is based on the presence of English language conjunction words (which the guidance seems to suggest), what happens to something like 230492006 |Excessive daytime sleepiness with sleep paralysis (disorder)|? I have chosen this one because (a) it contains 'with', (b) it is fully-defined, (c) it has three stated supertypes which seem to cover the semantics of the implied conjunction and I imagine would become the targets of its PARTIALLY EQUIVALENT TO relationships if inactivated, and (d) I don't know whether it has any association with an externally-maintained 'classification':

EquivalentClasses(
	:230492006 |Excessive daytime sleepiness with sleep paralysis (disorder)|
	ObjectIntersectionOf(
		:141000119100 |Daytime somnolence (finding)|
		:277180005 |Sleep paralysis (disorder)|
		:372947007 |Excessive somnolence (finding)|...

For several reasons this would seem to be in-scope for inactivation (and users could be told they now had to enter "...ALL of the PARTIALLY_EQUIVALENT_TO targets..."), but something (a very un-reproducible 'hunch') makes me think this would be spared. How will an author know what to do, and how will users know what is likely fate  of each of their familiar, useful (and used) conjunctions?

Kind regards

Ed

>  I feel a duty to ask

194733006 |Mitral and aortic stenosis (disorder)| is the projection into SNOMED of the old READ code [G130.]
This code was I think, in turn, the projection into READ2 of D396.0 MITRAL VALVE STENOSIS AND AORTIC VALVE STENOSIS from ICD9

Questions about whether the scope of content in scope is entirely clearly demarcated, and whether those who make active use of the existing content are aware and supportive of the change are all appropriate. But whether this particular representational equivalence problem exists in theory and, if it does, where and how said equivalences might become explicitly represented is orthogonal to what we either might - or should - do with that new expressivity.

As you know very well, the underlying representational issues run rapidly into some longstanding and exceptionally thorny informatics questions and dubious past practices, and a "large" change management endeavour. My own suggestion would be to begin by removing the more egregious content before anybody makes significant use of it, leaving much of the equally dubious but actively used content mostly alone for now. This approach more or less aligns with your final recommendations (a) and (b), except side stepping (b) by exploiting wherever robust code usage statistics indicate near-zero usage and so no impact (yet).

> I know your issue is with procedures and this response considers disorders

The motivation for proposing the PARTIALLY_EQUIVALENT_TO mechanism wasn't originally grounded purely in procedures, nor strictly in co-occurrence.

Rather, the specific use cases that most directly led to it being proposed were I think the pending inactivations of certain disorder codes derived directly from ICD-9-CM , such as 17593008 Gastric ulcer with hemorrhage, with perforation AND with obstruction (disorder) or 2066005 Gastric ulcer with hemorrhage AND perforation but without obstruction (disorder), and the question of what historical associations such codes should be assigned if they are inactivated. The case for inactivating this specific tranche of codes had been put to and was supported by the Edition Committee.

Such "narratively summative" codes only become valid once the constituent states finally become simultaneously co-occurrent. But, clinically, some of the constituent states will in some patients have come into being long before the rest: the ulcer could (would) have been present for days or even weeks. Of course, some patients - such as your pancytopoenic pancreatitis - will have presented from the beginning already with a "full house" of all relevant co-occurring phenomena. It is, however, undeniably entirely plausible that the record of some fraction of all individual patients who eventually become deserving of one or other of the example ICD-9-CM summative codes  above could in fact already have had each of the following added to their cumulative EPR earlier, potentially at different times and on different and earlier dates:

397825006 Gastric ulcer (disorder)
61401005 Gastric hemorrhage (disorder)
235674005 Perforation of stomach (disorder)
126765001 Gastrointestinal obstruction (disorder)

Its therefore very likely that some patients will ONLY have some combination of the above discrete codes present in their EPR, with no summative/aggregational code having ever been recorded at all.

The underlying problem is that "conjunction" in the terminology, both overt and hidden, is being used as a proxy for statement composition in the EPR information model. This of course poses a significant data quality, retrieval and analysis problem if those summative codes are in truth not reliably recorded. Worse, precisely because we've chosen the option of torturing the terminology, we typically lack the information model querying capabilities that would be required to assess how unreliable recording actually is, by looking for those "lost" cases that were only ever recorded as separately coded items. If we had that capability, we wouldn't need to confuse matters by also building a parallel solution into the terminology.

Not all diagnostic codes that correspond to a clinically completed narrative are statements of co-occurrence of the relevant phenomena: some require patient states to have occurred in a particular sequence. For example:

234522008 Disease relapse in transplant marrow (disorder)
85857008 Secondary syphilis, relapse (treated) (disorder)
200418007 Breast engorgement in pregnancy, the puerperium or lactation - delivered (finding)
38261000119106 Genital tract infection due to incomplete induced termination of pregnancy (disorder)

On the procedures side, we see similar potential for recording the same clinical narrative end point either as a sequence of individually coded steps or by means of a single summative code that only becomes valid at the point the "clinical story" eventually completes. For example:

2970004 Metacarpal lengthening and transfer of local flap (procedure)
7709002 Transfer of finger to opposite hand with amputation (procedure)
19697009 Debridement and suture (procedure)
308110009 Direct fundoscopy following mydriatic (procedure)

Indeed, most of the procedure codes potentially concerned do not seek to represent specifically the co-occurrence of their multiple phenomena as the defining characteristic of when the relevant clinical narrative has, in fact, completed. Rather, most require constituent subevents that could never co-occur, but instead are required to have occurred in a particular sequence over time:

1147961000000102 Breech delivery following termination of pregnancy (procedure)
56620000 Delivery of placenta following delivery of infant outside of hospital (procedure)
84901000000108 Admission to high dependency unit or intensive therapy unit following anaesthetic adverse event (procedure)
51901001 Irrigation following insertion of cannula (procedure)

370774005 Administration of prescribed medications based on arterial blood gas results (procedure)
473369002 Drainage of suprachoroidal fluid after choroidal detachment (procedure)
719864002 Monitoring following treatment for cancer (regime/therapy)
736777007 Reconstruction of breast using myocutaneous flap with microvascular anastomosis following mastectomy (procedure)
426877004 Removal of releasable suture following glaucoma surgery (procedure)
178112005 Re-excision of ganglion of wrist (procedure)
419139005 Fluoroscopic angiography of renal transplant with contrast (procedure)

And then there are those codes that combine a particular intervention with the prior diagnosis for which it was indicated:

473382005 Post exposure prophylaxis with antiviral therapy for potential sexual exposure to human immunodeficiency virus (procedure)
700214004 Antipsychotic drug therapy for dementia (procedure)
275931000 Aspirin prophylaxis for ischemic heart disease (procedure)
718026005 Cognitive behavioral therapy for psychosis (regime/therapy)

From all of the above, it is I hope clear that each of these compound statement-compositions-masquerading-as-precoordinated-codes could in principle be universally deprecated as codes and replaced by (mapped to) a semantically equivalent collection or sequence of multiple coded entries in the EPR. The PARTIALLY_EQUIVALENT_TO semantics offers one mechanism by which that mapping might be defined.

It should however also be obvious that - if that explicit encoding of representational equivalence turns out to be sufficient - then we don't actually necessarily always need to inactivate the "single summative code" permutation...though I would personally advocate against using that as an argument for maintaining in perpetuity the significant levels of representational overlap and optionality that characterises clinical informatics today.

I'll confess I'm a bit confused by what content/why is in scope of being deemed ambiguous now.
I've no concerns about the "OR" concepts being considered ambiguous, as their association with classifications is obvious.

But most (if not all) the terms with conjunctions like "and" or "with"... Seem legitimate.
It's one thing to introduce an editorial rule to say "we're not going to create precoordinated combinations of things anymore", but retiring content that users are happily using has impacts.

For example, would something like 249519007|Diarrhoea and vomiting| be retired and replaced by a couple of "PARTIALLY_EQUIVALENT_TO" associations?

Does Ed's example of 127034005 |Pancytopenia (disorder)| avoid inactivation, through the fortune of English having a single word for "reduced red cells AND reduced white cells AND reduced platelets"?

I remember looking at 200416006|Breast engorgement in pregnancy, the puerperium or lactation| a couple of years ago, and thinking the "or" wasn't great, but decided it was an artefact of us not having a single word to describe that period of time. "Perinatal" is close, but not quite... Ideally, anyone recording such a diagnosis should be able to be more specific e.g, "Breast engorgement in pregnancy".

Rarely are any procedures performed in isolation. Usually "associated procedures" are implicit and in line with the open world view.
103715008|Removal of catheter| is obviously following an 45211000|Insertion of catheter|, but doesn't need to be said.

426877004|Removal of releasable suture following glaucoma surgery (procedure)| - the "glaucoma surgery" provides the specific site, but also perhaps the complexity of the removal compared to other sites?

I can't see much benefit (for the cost) to end-users in deprecating this content. Analytics should be much the same wether "one or two codes" is used
If the rationale is that the information model should support it, that could also be said for keeping "Occurrence:<Period of life between birth and death" out of modelling... (Not that I'm suggesting so).

Matt Cordell its perhaps unfortunate that this discussion - about the potential scope of codes that might be inactivated and ascribed PARTIALLY_EQUIVALENT_TO historical associations - is appearing here, on the page about codes for which the reason for the inactivation is recorded to be "they were ambiguous".

The issue with the codes potentially in scope for PARTIALLY_EQUIVALENT_TO isn't that their meaning was ever ambiguous, but rather that its an excessively precoordinated proxy for information model statement composition and whose originally meaning is therefore only fully encompassed by the union of the many codes pointed at rather than by only one of them, as in the POSSIBLY_EQUIVALENT_TO flavour  of historical association. If any of the kind of content outlined in the discussion so far does get inactivated, I would have thought that the reason for that inactivation would never be "Ambiguous" but rather something like either Non-Conformance to Editorial Policy or Classification Derived Concept or, potentially, some new and additional flavour of reason yet to be defined.

But like a lot of clinical informatics, pulling on this particular piece of the spaghetti soon leads to shark infested waters. As ever, we end up caught on the two horns:

1) We have not agreed what the most principled and scalable approach for a "green field" setting would have been on theoretical grounds, in respect of how and where clinical statement composition should be represented between information and terminology models, given what we know about

a) how unreliable clinicians are at remembering to explicitly record either some of the more complex "a clinically interesting narrative has completed" codes already available or, equally, some significant individual elements of that story as they occurred, such as that the patient was pregnant or a newborn or that they DON'T have obstruction, because this information is e.g. "obviously" implicit to the wetware

b) how difficult it can be for analysts to write and/or technomages to execute the queries that we'd need to run over the information model in order to infer/abstract when the same narrative phenomena have in fact completed (or co-occurred), but from the evidence of more "piecemeal" coding entered over the course of the narrative rather than only at its conclusion. Your statement that "Analytics should be much the same whether one or two codes are used" requires that both representational choices will be mutually transformable and thus provably equivalent. But this is of course empirically most definitely not the current reality in the general case, and it would be a costly equivalence to author and maintain.

c) that even if we knew how to do (b) and could resource maintenance of the high quality and comprehensive assertions required to compute the equivalence transform, it is still useful to have an agreed and shareable symbol to represent, cache and communicate that machine and/or wetware inference/abstraction result, even if we no longer expect it to be entered directly by humans.

2) There is a very considerable pregacy burden and clinical tradition, much of which predates by centuries even the vaguest possibility of representing any of this within any kind of information model. Accordingly, we have huge quantities of historical data already captured against two competing and largely non-interoperating design choices, and any clean new design for the future becomes hobbled by the requirement that it must support entirely seamless, invisible and lossless migration of all existing data, customs and practice from wherever we are now, to wherever we decide we should have been all along.

In the continued absence of exhaustively populated formal and computable statements of equivalence, we can predict that the current situation should not work in practice in the general case, and we further have grounds to suspect that it does not work in practice. But we're largely blind to how bad the situation really is, because of 1(b).

The current approach to spoofing statement composition within the active codes of the terminology is mostly to hijack the subclass relationship, so that we end up with e.g. 

16932000 Nausea and vomiting (disorder) IS-A 422587007 Nausea (finding), IS-A 422400008 Vomiting (disorder)

But this is a brittle and very far from completely implemented mechanism: not only are we now reliant upon clinicians (ideally) remembering to make life easier for the analyst by using the combined code when both symptoms are co-occurrent, we're also relying upon the Authoring Monks to remember to use this subclassing mechanism in order to encode the equivalence for the benefit of those clinicians who forgot. But the monks manifestly forget almost as often as do the clinicians. See e.g. 

16932000 Nausea and vomiting (disorder)
vs. 723975009 Nausea and vomiting following administration of anesthetic agent (disorder)

249519007 Diarrhea and vomiting (finding)
vs. 288199005 Diarrhea and vomiting after gastrointestinal tract surgery (disorder)
and 2919008 Nausea, vomiting and diarrhea (disorder).

17369002 Miscarriage (disorder)
vs. 275425008 Retained products after miscarriage (disorder)
and 10812001000119100 Sepsis due to incomplete miscarriage (disorder)

609588000 Total knee replacement (procedure)
vs. 441600007 Arthrolysis after total prosthetic replacement of knee joint (procedure)

235595009 Gastroesophageal reflux disease (disorder)
vs. 266433003 Gastroesophageal reflux disease with esophagitis (disorder)

However, in the event that one of these codes does get inactivated, those subclass relationships vanish and, by convention, the preferred mechanism for pointing them back into the active taxonomy is by means of historical associations. So, what flavour of historical association should they have, given that WAS_A has been discontinued and (as demonstrated above) the set of erstwhile declared superclasses prior to inactivation may not be complete?

Just a few points of clarification that may be of assistance:

1. Dealing with "AND and "WITH" originally cropped up in the ambiguous section in response to resolving the ambiguous concepts that contained |A and/or B| where the |A and B| required further resolution. As Jeremy has indicated, this now more correctly sits with the inactivation reason "Classification Derived Concept" but may also need to be used elsewhere.
2. There is no intention to systematically inactivate all concepts that have an "AND" or "WITH".
3. Some clinicians have expressed the view that they would prefer to record each disorder separately even if there exists a concept that combines both as they wish to ensure that the management of each disorder can be appropriately attributed.
4. We do not anticipate that there will be high usage of this historical association but where it is used it was felt important to provide a historical association that gave end users and system developers a clear indication of what action is required to maintain the integrity of their historical data.
5. The objective of the Concept Inactivation Project was to improve upon the existing process by identifying new inactivation reasons and historical associations and improving on the definition of usage. This will hopefully improve consistency in the choice of inactivation reason/historical association and how it should be interpreted and implemented by users/vendors. It is our belief that over time the process will continue to be reviewed and evolve to meet the needs of our users and the feedback you are all providing is key to ensuring that happens.