The attached document summariezes the changes that are proposed for bypass and shunt procedures. The contetn can be reviewed in the TS browser under the task ID listed in the document. Please provide you comments relative to this proposal in this forum post.
Bypass and shunt procedures Changes for review by the EAG.docx.pdf (50.8 KB)
The changes as described in the document sound good, but the content visible and for EQG review on the QICONST-16 task of the QICONST project doesn’t appear to always match that document. For example, whereas the document says:
…the browser today is showing this:
I still have significant concerns that a single postcoordinated “Anastomosis procedures” grouper still can not be defined to elegantly subsume all of: primary procedures to construct an anastomosis, subsequent redo, revision or maintenance procedures on an already existing anastomosis, or later take-down procedures to reverse an anastomosis.
In any procedure that fiddles with an anastomosis or shunt, these structures could only ever be reified as anatomical structures and not also (or instead) compiled into the semantics of the method itself.
This would I think be preferable and more flexible that what we have now, which is essentially:
260686004|Method (attribute)|=257741005|Anastomosis - action (qualifier value)|
means same as
(260686004|Method (attribute)|=129376004|Construction - action (qualifier value)|,
405816004|Procedure morphology (attribute)|=41796003|Anastomosis (morphologic abnormality)|)
Indeed, this latter modelling pattern is exactly what we already see in e.g.
1287774005|Attention to anastomosis (procedure)|
47585001|Surgical closure of anastomosis (procedure)|
443407008|Revision of anastomosis of intestine (procedure)|
440070000|Repair of ileoanal pouch (procedure)|
This is a point I’ve raised before: why are we still combining the method with the target only when we’re talking about the primary construction of an anastomosis|shunt|bypass ??
@jrogers I see from your post that you now have access to the QICONST project / task in the TS browser? Because although I can access the task in MS, I cannot open a TS browser on it. This makes it impossible to get an overal view of the new modelling. @jcase
@fhielkema it took a while to work out how to get access: if I follow the SNOMED International Browser link in the endoscopy review document direct to the browser, I can not get past the login.
Instead I must first login at https://ims.ihtsdotools.org, and then click on the icon of six squares in a circle at top right, followed by the Pen icon for “International Authoring”. This opens a new browser window showing a personal dashboard. At the top of the menu at left of the screen is the option “TS Browser”…and if I open that then I can finally set the project and tasks needed to carry out the review.
Looking, perforce, at details, I’m confused by the contrast between 116360008 |Arterial bypass graft (procedure)| and 450683007 |Axillary artery bypass (procedure)|.
From the modelling of arterial bypass graft we can deduce that some arterial structure is bypassed. In axillary artery bypass, it is not stated which structure is bypassed: the axillary artery is the proximal anastomosis. Thus, 116360008 |Arterial bypass graft (procedure)| does not subsume 450683007 |Axillary artery bypass (procedure)|. Why? What in the FSN suggests that the axillary artery is not bypassed but the proximal anastomosis?
I14752007 |Axillary-axillary artery bypass (procedure)| is modelled with proximal and distal anastomosis - which fits the proposal and sheds light on the choice made for 450683007 |Axillary artery bypass (procedure)|. But I still think that subsumption relation between 450683007 |Axillary artery bypass (procedure)| and 116360008 |Arterial bypass graft (procedure)| ought to exist.
388997005 |Axillo-femoral artery bypass (procedure)| also does not state which structure is bypassed; yet 389007008 |Superficial axillo-femoral artery bypass (procedure)| is modelled with a bypass of the aorta. Why the discrepancy? If the aorta is bypassed, should that not be represented in the FSN?
On the whole, I think we risk adding inconsistency to the modelling when we add the structure that is bypassed. @vmedina Does the list of concepts you remodelled in this task contain one or more examples whose proper subsumption relies on modelling the bypassed structure, or is it something we could leave out?
Minor point: the proposal states: When the use of a graft (Substance or device) was stated in the FSN, it was added to the first (Bypass/Shunt RG) using the 424226004 |Using device (attribute)| or the 424361007 |Using substance (attribute)|. This confused me a little because a number of concepts, .e.g 116360008 |Arterial bypass graft (procedure)|, 233384002 |Popliteal-distal bypass (procedure)| and 388997005 |Axillary-femoral arterial bypass (procedure)| have synonyms with and without the word ‘graft’. Reword the proposal to ‘when the use of a particular type of graft’?
Alas, no such luck for me… I followed your instructions to the letter, but clicking ‘international authoring’ takes me to the same overview as a normal Managed Service authoring, although the link is different. When I open the TS browser, I do see projects AUTHORTEST and INTCAVEND but not project QICONST…
Hi @jrogers The issue on classification of Bypass - action and Shunt - action is fixed. Thanks.
Hi @fhielkema You are absolutely right, all Arterial bypass should be subsumed by 116360008 |Arterial bypass graft (procedure)|. If we follow the model for Arterial bypass graft with Procedure site - direct = 51114001 |Arterial structure (body structure)|, then every arterial bypass procedure should be modeled with Procedure site - direct = << 51114001 |Arterial structure (body structure)| for consistency and subsumption. When the bypassed structure is stated in the FSN we can add it as target for the Procedure site - direct attribute, when the bypassed structure is not stated in the FSN the model would be Procedure site - direct = 51114001 |Arterial structure (body structure)|. We didn’t see any benefit to adding the bypassed structure unless it is stated in the FSN. In the case of 175276005 |Emergency bypass of aorta by anastomosis of axillary artery to femoral artery (procedure)| the FSN states the bypassed structure, but parents would be modeled with the generic 51114001 |Arterial structure (body structure)| as in 389007008 |Superficial axillo-femoral artery bypass (procedure)|. Will test this solution and come back to you. Synonyms using the term “graft” were left for historical reference. But your proposal on changing the wording in the proposal would work. Will test.
I will let Victor expand on this, but in the initial discussions, it was proposed that the direction of flow was an important consideration for both bypass and shunts. The addition of proximal and distal anastomosis actions made that explicit in the model. The recent addition of the bypassed anatomical structure (where named) was also important to locate the defect. The difference between the method and the morphology is that the former represents the creation of the anastomosis whereas the latter represents the result of the action. Other than the original creation of the anastomosis, it is the morphology that is the target structure for revisions, closures, etc.
Thanks @jcase.
@jrogers The need to represent the direction of flow for both Bypass and Shunt procedures has been discussed in multiple occasions. In the last meeting and after a new discussion on the topic we agreed that it should be represented in the model.
Two arguments to do so are:
Some Bypass/Shunt procedure does not state the Bypassed structure but only the Proximal or origin structure from where the flow is diverted, in this case using Proximal anastomosis - action will ensure that only the procedures represented that way would be subsumed or will subsume the same type of procedure, otherwise any procedure modeled with Anastomosis - action will subsume or be subsumed by ANY procedure of the same type no matter if the anastomosed structure is the “origin” structure, the Bypassed structure and/or the “destination “ structure.
Other procedures are very clear stating the origin and the destination… e.g. 112828007 |Femoral-popliteal artery bypass graft (procedure)|, if we don’t describe where the diversion starts and where it ends we would assume that the flow would go from the popliteal artery to the femoral artery. If that happens we would be facing a complication of the procedure. Same happens with shunts, where the flow is forced by the higher pressure in the source and lower pressure in the destination structure, so, the procedure is aimed to force the flow of the contain, if it is reversed, we will be facing a complication as well.
Please let me know if you have additional comments or questions.
Victor
Thanks @vmedina
For your entertainment, attached is a short technical document I wrote back in 1997 as part of the GALEN-IN-USE project, on the subject of how Bypasses, shunts and fistulae might be modelled. It was motivated by the empirical discovery that, in trials, terminology and classification experts from different centres had devised several entirely different and mutually incompatible ontological patterns for substantially the same representational task. The range of solutions owed much to differences in local surface language and clinical terming conventions.
VUM10 Modelling of Bypasses.doc (91.5 KB)
That ancient document explains my overarching concern here : if some procedures involving a bypass represent that structure as a noun (a bypass structure) whilst others embed the same semantics but as a verb (a bypassing procedure or method) then there are two problems:
Firstly, you can’t create a single grouper under which all procedures involving a bypass can autoclassify
Secondly, even if the precoordinated content applies the idiosyncratic convention systematically, you’ve introduced semantic redundancy into the wider postcoordinating ontology: some users will reasonably but mistakenly assume that the correct way to model the creation of a bypass or anastomosis is as per the construction of many other surgical structures including pouches and stomas (<<410614008|Construction|), and the natural mirror of the pattern used in procedures to remove or revise an existing bypass|anastomosis|shunt|fistula|stoma (e.g. 173597007 Removal of bypass of esophagus), none of which currently involve a bypassing|shunting|fistulating|stomatising “method” (verb).
At the time, I therefore came to the conclusion that because you couldn’t have an ontology that didn’t include both
a) the generalised pattern:
{260686004|Method|=129376004|Construction - action|,
405813007|Procedure site - Direct|= <<123037004|Body structure|}
b) 41796003 Anastomosis (morphologic abnormality) for the purpose of modelling lesions at such surgically created sites
…then sadly you couldn’t have the following in the same ontology:
c) an 257741005 Anastomosis - action (qualifier value).
In a computable and postcoordinating ontology, consistency of representation and ruthless removal of compositional semantic redundancy is essential.
As has been said many times, however, bypasses, shunts and anastomoses have further complexity because their proximal and distal points of origin and sometimes also the expected direction of flow all need to be independently (but optionally) representable.
When an ontology within which anastomoses|bypasses|shunts|fistulae can ONLY exist and be referenced as structures is combined with an OWL formalism in which modelling can be indefinitely nested, the general pattern I arrived at for representing the construction of one of these structures was (using an intermediate representation):
MAIN constructing
ACTS_ON (anastomosis|bypass|shunt|fistula structure
CONNECTS_PROXIMALLY body structure
CONNECTS_DISTALLY body structure)
…which is therefore also a pattern that can be dropped seamlessly into these related kinds of procedures:
RUBRIC: ‘Removal of thrombus from fem-pop bypass’
MAIN removing
ACTS_ON (thrombus lesion
HAS_LOCATION (bypassing structure
CONNECTS_PROXIMALLY femoral artery
CONNECTS_DISTALLY popliteal artery))
RUBRIC "Repair after previous bypass from femoral or popliteal artery to infrapopliteal arteries"
MAIN repairing
ACTS_ON (bypass structure
CONNECTS_PROXIMALLY deep femoral artery
CONNECTS_DISTALLY infrapopliteal artery)
The initial problem with the current solution before us therefore is that, in the context of the above regularly structured decompositions of the semantics, it essentially attempts to declare a defined attribute:
'245854003 |Proximal anastomosis (qualifier value) |'
== constructing ACTS_ON (anatomical structure CONNECTS_PROXIMALLY
…but using a logical equivalence construct that is intrinsically problematic (aka non-computable) because it isn’t supported by OWL 2 DL.
A still more challenging problem arises however if we also need to be able to represent the structure being bypassed, for example in:
RUBRIC "Allograft bypass of coronary artery narrowing"
MAIN constructing
USING allograft vein
ACTS_ON (bypassing structure
AROUND coronary artery narrowing)
In the same way as above, you might attempt to solve this in a similar way, perhaps by declaring:
'bypassing - action'
== constructing ACTS_ON (bypassing structure
…but now, by introducing something that looks like an action verb|method, we’ve got something that linguistically looks dangerously like it could also be used as either of the following:
MAIN bypassing - action
ACTS_ON internal iliac artery thrombus
CONNECTS_PROXIMALLY common iliac artery
CONNECTS_DISTALLY femoral artery
MAIN bypassing - action
AROUND internal iliac artery thrombus
CONNECTS_PROXIMALLY common iliac artery
CONNECTS_DISTALLY femoral artery
…but these would then not be transformable, either between each other or (critically) to and from the canonical design of:
MAIN constructing
ACTS_ON (bypassing structure
CONNECTS_PROXIMALLY femoral artery
CONNECTS_DISTALLY popliteal artery
AROUND internal iliac artery thrombus)
because, if we were to expand “bypassing” as per its definition, we would get:
IF
'bypassing - action'
== constructing ACTS_ON (bypassing structure
THEN
MAIN bypassing - action
ACTS_ON internal iliac artery thrombus
CONNECTS_PROXIMALLY common iliac artery
CONNECTS_DISTALLY femoral artery
becomes
MAIN constructing
ACTS_ON (bypassing structure
ACTS_ON internal iliac artery thrombus)
CONNECTS_PROXIMALLY femoral artery
CONNECTS_DISTALLY popliteal artery
AND
MAIN bypassing - action
AROUND internal iliac artery thrombus
CONNECTS_PROXIMALLY common iliac artery
CONNECTS_DISTALLY femoral artery
becomes
MAIN constructing
ACTS_ON (bypassing structure
AROUND internal iliac artery thrombus)
CONNECTS_PROXIMALLY femoral artery
CONNECTS_DISTALLY popliteal artery
…which, though lexically similar to the human reader, are both structurally and semantically entirely different and so won’t/can’t autoclassify as equivalent.
And that’s my fundamental problem here : although the suggested solution puts “the words” into a structure, it does so only by increasing precoordinated and ECL query design idiosyncrasy in the overall ontology design whilst also increasing the level of non-transformable semantic duplication/redundancy in postcoordinating (and ECL design) settings.
Thanks Victor, and Jeremy for the background. I want to reiterate my concern from previous meetings about modelling proximal/distal anastomosis and direction of flow. I don’t dispute that flow direction matters clinically. My question is whether it does any definitional work in an ontology, and what introducing it actually solves.
Direction isn’t actually in the current model, even where the term implies it. Consider 447950005 |Creation of anastomosis from ascending aorta to left pulmonary artery|. It names a from-to direction, but both vessels sit symmetrically on Procedure site - Direct with 257741005 |Anastomosis - action| as the method. Nothing in the logic says which vessel is origin and which is destination, and the definition would be identical if the term read “between” the two, or indeed if it were reversed to read from the left pulmonary artery to the ascending aorta. The intended flow direction is determined by the anatomy and the pressure gradient, so it adds no discriminating power.
The proximal and distal anastomosis methods are themselves an attempt to assert that flow. In a bypass the proximal anastomosis is the inflow end and the distal the outflow end, so applying those labels is already a directional claim, not a separate question from it. The anatomy fixes which end is which, so the methods only restate something the structures already determine.
An ontology should carry what is necessary and sufficient to define each concept, which is a narrower task than a procedure manual describing how the operation is carried out. Direction of flow is clinically real, but where the anatomy fixes it, it isn’t definitional, and asking authors to assert and maintain it adds cost for no matching gain. Are there a pair of procedures performed in both directions on the same structures, where the direction is a real choice rather than one fixed by the anatomy? Short of that I don’t think it earns a place in the model.
I think Victor was arguing earlier that the need to capture the direction arose from two angles:
Firstly that there are some anatomical sites that can function as either the inflow or the outflow, depending on the procedures. The Iliac and femoral arteries, for example, in aortoiliac|iliofemoral or aortofemoral | femoropopliteal grafts. The argument is that in order to support retrieval of only those bypasses that start from the iliac artery as opposed to more broadly those that are between the iliac artery and “somewhere else”, you need a notion of proximal or distal anastomosis inside the ontology.
His other rationale was I think in the case of modelling the shunt that results - the actual flow, Whilst it is true that the flow is predictable if you know the anatomy, from the normative pressure gradient, there are complications where the normative gradient is inverted and so the direction of flow is actually reversed.
More broadly there are a fair number of other procedures and disorders that involve the notion of “retrograde flow”, so I think there are other areas of medicine where modelling will be difficult without a notion of “direction of flow”. Though, then again, it could be argued that the “proximal|distal anastomosis” pattern is still compiling the direction of flow into the normative anatomy, rather than separating it out cleanly.
Thanks for your comments.
For most bypass procedures, the proximal and distal ends are indeed anatomically fixed by the pressure gradient and the named structures. However, for both bypass and shunt procedures there is a real class of exceptions. The iliac and femoral arteries function as either inflow or outflow depending on the level of occlusion: an aortofemoral bypass originates from the aorta and terminates at the femoral, while a femoropopliteal bypass originates at the femoral. Without a proximal/distal distinction in the model, ECL cannot distinguish a bypass starting from the femoral from one ending at the femoral — both return as “bypass procedure involving femoral artery structure.” For shunt procedures, the same problem is sharper: 20767001 |Shunt of portal vein to vena cava| and a hypothetical reverse shunt involve the same two structures; without directional encoding, they are definitionally identical. These are not edge cases — they are precisely the retrieval scenarios that clinical decision support and surgical audit depend on.
Matt asked: “Are there procedures performed in both directions on the same structures, where direction is a real choice rather than one fixed by anatomy?”
The answer is yes — and it is not a matter of choice but of diagnosis. The pulmonary artery appears as destination in Waterston/Potts (receiving high-pressure aortic blood to increase pulmonary flow) and equally as destination in the Glenn shunt (receiving low-pressure venous blood to bypass the right ventricle). Both the ascending aorta and the superior vena cava connect to the pulmonary artery — but the two procedures are clinically antithetical. An ECL query for “shunt procedures involving the pulmonary artery” without directionality retrieves both families indiscriminately.
However, while modeling in the example task we’ve found a difficult to model the body structures being bypassed unless they are stated in the FSN. Some concepts stating the origin and destination structures can “bypass” a number of structures and modeling them would require to imply each one.
Also, after discussion, other questions arose regarding the authoring burden associated: 1) is there a clinical requirement to record these procedures to the level of detail proposed in the model and 2) what is the threat to retrieval if the more parsimonious modeling approach is used?
All these issues are worth to be discussed.
We are working in a new document aimed to review (and clarify) these issues and bring a fresh proposal for discussion.
Victor
Thanks @jrogers, that example is helpful and a use case I can get my head around. Otherwise the flow seems implied by the normative anatomy, and I just want to be sure there is a value proposition for the added complexity of modelling direction.
That isn’t wrong though. They are both a bypass involving the femoral artery. All three are valid retrieval questions: starts from, ends at, and involves the femoral artery. The first two are really asking which of the two vessels is the more proximal, which the anatomy already settles, so the open question is whether that is best captured on the structures or asserted as a direction on the method.
I absolutely recognise retrograde flow is a clinical concern, but for these bypass and shunt procedures my understanding is that it is an undesired outcome rather than an inherent, modelled goal of the procedure.
A Briefing note addressing all the concerns expressed by the comitte on the last proposal for review by the EAG members is being drafted. It will be circulated soon and will include specific questions that may lead us to final greements to advance further on the resolution of this issue.