Follow-up to revised allergy model

Dear Bruce,

Now I have found some time for the allergy model.   

> There has been a fair amount of discussion around the allergy model,

> stemming from the project to remodel 419199007 |Allergy to substance

> (disorder)|and it’s descendants using proximal primitive parents and

> update the FSNs using a consistent naming pattern. I have been thinking

> of some enhancements to the model but in doing so I feel I need some

> clarification regarding our previous approach. As you may recall, we

> defined the top level concept, “allergic condition” with a pathological

> process relationship, pathological process = allergic process. This

> results in in the pathological process relationship being inherited by

> 609328004 |Allergic disposition (disorder)|. Is this correct? Can a SNOMED

> disorder which is being interpreted as a disposition contain a pathological

> process attribute as part of its definition ? I am sure we considered this

> at some point but I can’t recall the justification.

As far as I remember we agreed that we have the following kinds of entities:

1. allergic dispositions (the same as propensities) –
    subclass of bfo:Disposition or btl2:Disposition

2. allergically modified body parts (such as a nasal mucosa under allergy)
    subclasses of bfo:MaterialEntity or btl2:MaterialObject

3. allergic processes
    subclasses of bfo:Process or btl2:Process

- Common parents of disposition, object and processes are named "condition (disorder) ".

4. clinical life phases (CLPs), defined using 1, 2, o 3
    subclasses of bfo:Process or btl2:Situation

    CLPs are phases of an individual's life in which some defining 

    condition is fully present

All findings and disorders are CLPs  (but so far they are not named as such). 

CLPs  are related to conditions via the RoleGroup relation. 

But correctly, what is named "condition (disorder)", means CLP.  

This is the current model. I do not think it's problematic that the domain of the relation "pathological process" may have, according to the subclasses of allergic condition, different subtypes, like allergic disposition disorders  or allergic process disorders, because both are SNOMED disorders = CLPs, and therefore ontologically homogeneous.

The only problem is the implied operator "some": Literally interpreted there must not be any instance of allergic condition that is not related to any instance of Allergic process. OWL-EL does not provide any alternative. Nevertheless we had agreed to use this pattern.

What the ontological nature of "Allergic process (qualifier value)" really is, remains unclear anyway. "Qualifier value" is not an ontologically sensible category. This means, as long as nobody can tell us what  "Allergic process (qualifier value)" really means (a process class, or some kind of universal instance ?), the question of whether it is correct or not that the pathological process relationship is inherited down the hierarchy cannot be answered. The same applies to whether the clause

         "Pathological-Process – Allergic process"

should be enclosed in a role group or not.

The inheritance of "Pathological-Process – Allergic process"  is certainly not counter-intuitive, because all descendants of allergic condition are defined by some real or hypothetical allergic process. 

> A second question involves my suggestion to replace the

> association of the allergen with the disposition, currently

> represented by a role group containing causative agent

> substance + after allergic sensitization by using a has

> realization relationship as illustrated below.

> Role group { has realization allergic reaction + causative agent substance}

> which states that an allergic disposition is realized by an allergic reaction

> which is caused by an allergenic substance. This approach would

> lead to a consistent model for both allergy and pseudoallergy. The

> has realization role is currently being considered to support the observables

> model. Any concerns/questions regarding this proposal?

I assume you mean the following:

I am in favor of using the relation "has-realization", but again, this is only 100% correct with the "only" (universal) quantifier  and not with "some". But like in the observable model, I consider this an acceptable impropriety. Therefore I support your proposal.  

Some general remarks, regarding the ECE process on allergies, which took quite a lot of time, and requires, again and again, that we get familiar with the formalisms and the names. I see the following complicating factors, mostly depending on the SNOMED CT legacy, but partly also on our group:

• Naming is confusing: Concepts are named "Disorder" but mean "Clinical life phase"; "propensity" and "disposition" are used in the same sense, "Condition" is in the range of the RoleGroup but is never named as such, whereas "Condition (Disorder)" appears in the range of the RoleGroup relation, but means  "CLP with condition". The term "Allergy" is used without modifier, where the meaning is "Allergic disposition" (fully specified names should not be ambiguous)

• The role group relation remains anonymous; its fundamental semantic import remains hidden, there is the risk to see it just as a syntactic element for "grouping" (which is at odds with the description logics interpretation)

• Also in the graphical notation the RoleGroup relation appears as a syntactic element. Both the graphical notation and the SNOMED CT coordination grammar syntax lack clarity compared to the much more intuitive OWL Manchester Syntax

• Too much in SNOMED CT is just ontologically unclear, such as, e.g., Processes categorized as "Qualifier values" 

Best regards,

Stefan

Thus going forward, my impression is that even with the limitations of this approach we should:

1. Continue to define the top level nodes (hypersensitivity condition, allergic condition and pseudoallergic condition) using pathological process hypersensitivity/allergic/pseudoallergic process (qualifier value).

2. Replace role group {after allergic sensitization + causative agent substance} with role group {has realization allergic reaction process (qualifier value) + causative agent substance}.

A few comments.

1. In our model, we did not specifically call out allergic structures as there is currently limited representation of such entities in SNOMED.

2. Allergic condition currently subsumes allergic disposition, allergic reaction (process) and various specific diseases such as allergic rhinitis that could be interpreted as active processes or dispositions. In my clinical experience, the latter is more common. My plan for dealing with these is to move the existing allergic diseases under allergic reaction and to create parallel concepts with FSNs of the type Propensity to allergic disease X e.g. Propensity to allergic rhinitis under allergic disposition

3. Perhaps we can rename allergy to x to propensity to allergic reaction to x and retain allergy to x as the preferred term. I need to obtain approval from the group for this change.

And…just to emphasize, the advantage of this revised model is to be able to apply a consistent approach to the modeling of dispositions associated with material entities simply by creating new pathological process qualifier values. Although I understand the issues this raises, I was very unhappy with having to define e.g. pseudoallergic disposition with a completely different and even less precise model than for allergic disposition.