Basic Formal Ontology Discussion
Description
This document summarizes the early stages of a dicsussion on harmonizing SNOMED CT with the Basic Foraml Ontology (BFO), initiated by some talks between SNOMED CT representative and people from the OBO Foundry.
This document summarized initial e-mail discussions.
There is a Google document draft
SNOMED CT and Basic Formal Ontology - Convergence or Contradiction?
https://docs.google.com/document/d/1HcBj5bVIg8lB_uyORZU9A_FWKFsw0sxmB6Xg4UYKygk/edit#
on this topic, to which all MAG members are invited to contribute
Discussion
Peter Hendler
BFO wants terms to be about "things in the real world", not "thoughts". BFO feels most terms should be fully defined. https://bmcmedinformdecismak.biomedcentral.com/articles/10.1186/s12911-018-0651
Some of the objections are:
1 SNOMED should be about things in the real world and not about "concepts"
2 SNOMED should have most of it's terms fully defined
2 SNOMED should be based on BFO and there should be Continuants and Occurents and they should be mutually exclusive
3 SNOMED should have single inheritance.
I address these with counter arguments.
-----------------------------------
Single inheritance is not an advantage to SNOMED. Viral Pneumonia should be both an Infectous Disease and a Respiratory Disease. For Logical retrieval this is an advantage not a problem that needs fixing.
Viral Pneumonia is only a Respiratory Disease in ICD and that is a big disadvantage for concept retrieval in ICD. SNOMED can not be an ontology that is not about "concepts"
but is about "things in the real world".
Past, current, and future medical science can never know "things in the real world". All of the historical "diseases" that no longer exist because of subsequent scientific studies prove that.
A good example today is the "concept" of "Systemic Lupus Erythematosis". This term is quite useful and necessary today. But is is only a concept. It is NOT something in the real world.
Once lupus is understood what exact combinations of genetics and environmental triggers cause "lupus", it will be discovered that either lupus is a term for many different diseases "in reality" or depending on where we draw the line, we might in the future discover that Sjogrens, and perhaps other diseases should in fact all be given a new concept name and grouped together.
This also is related to the criticism that most terms in an ontology should be fully defined. In the real world of clinical medicine, many diseases cannot be fully defined.
Rheumatoid Arthritis is a good example. We don't know what it is. We have somewhat arbitrary clinical criteria that will certainly later be thrown away once we know the exact genetic and environmental mechanisms. Then we may even throw away the whole idea of Rheumatoid Arthritis. We may divide it into many sub diseases, or we may group it with another disease under a more broad heading. So the idea that SNOMED needs to fully define more terms, and the other idea that SNOMED should not be "concepts" but represent things in reality is already a contradiction.
Another problem in BFO is the disjointness between Continuants and Occurents. BFO would have one term for Asthma that is a Dependent Continuant for the propensity for Asthma Attacks. Then Asthma attacks would be Occurants. But for a real patient who has Asthma, they probably always also have some small component of the Occurant going on. Sometimes it's bad enough to go to the Emergency Room, or to take rescue inhalers, but there is no particular arbitarary severity when all of a sudden the Occurent
is turned on and becomes present.
It is far more useful in clinical medicine to think of Chronic and Acute Asthma. And all clinicians understand that the line for Acute Asthma is subjective and partly dependent on the subjective experience of the patient. (Some go to the Emergency room when others would not).
When reading criticism of SNOMED, it appears that many of the critics dont understand what clinical medicine really is. It is not so much about "things in the real world" where everything is either a Continuant or and Occurent, but as discussed above, it is a temporary, always changing world of ideas "concepts" that are very useful for deciding treatments. It is understood that these concepts always are in a state of flux. They come and go, and as they do the usefulness of the concepts progress and become closer to the real world, and the concepts become more useful in making treatment decisions, and in research.
Daniel Karlsson
About single inheritance, what I remember from discussions with Barry and Werner during the Semantic Mining project (2004-8) was that they had objections to having multiple stated parents (multiple genera) (see Buidling ontologies with BFO, page 79). 75570004 |Viral pneumonia (disorder)| fulfills this criteria, although sometimes by necessity multiple stated primitive parents are needed to account for multiple aspects of meaning not defined through attributes.
About real world vs. thoughts (i.e. mind-dependent entities such as ideas, plans, suggestions, ...), first, thoughts are also part of the real world and not accepting that is being relativistic about ontology (Barry, the relativist!), second, mind-dependent entities are clearly relevant in the health care domain, so at least a health care ontology should comprise both mind-dependent and mind-independent entities. Does SNOMED always clearly separate the two? Probably not, but that requires an analysis beyond the analysis of the definition of disease (...and "a disease is a concept" seems to have been removed from the editorial guide) in SNOMED. Further, the second-order (instances-to-types) relationship between mind-dependent and -independent makes a faithful representation in OWL (and EL in particular) a challenge to say the least (see Schulz et al An Ontological Analysis of Reference in Health Record Statements). In e.g. the Observables model this issue has been discussed but deemed not worth the effort and complexity (e.g. does the alcohol concentration of non-alcoholic beer exist? and then what are the implications for use of existential restrictions?)
Regarding the definition status of some pre-theoretical classes of disease I think Flier and de Vries-Robbé has a good paper here: https://www.ncbi.nlm.nih.gov/pubmed/10765492
Peter Hendler
SNOMED can not be an ontology that is not about "concepts" but is about "things in the real world". Past, current, and future medical science can never know "things in the real world".
Stefan Schulz
TO KNOW → epistemology
TO BE → "ontology.
Things in the real world exist, even if we classify them differently. Clinician make right and wrong diagnoses, but this is not a contradiction to grounding SNOMED CT in an upper-level ontology. Neither is it contradictory that entities in biology are continuous rather than discrete. We cannot exactly delineate anatomical entities like we can disassemble pieces of a machine. We often cannot draw a crisp border between what is normal and what is abnormal. BFO and other upper level ontologies are not adverse to the existence of "fiat entities" (i.e. entities delineated of lumped together by humans).
Peter Hendler
"Systemic Lupus Erythematosis". This term is quite useful and necessary
today. But is is only a concept. It is NOT something in the real world. Once lupus is understood what exact combinations of genetics and environmental triggers cause "lupus", it will be discovered that either lupus is a term for many different diseases "in reality" or depending
on where we draw the line, we might in the future discover that Sjogrens, and perhaps other diseases should in fact all be given a new concept name and grouped together.
Stefan Schulz
Whether concepts, universals, etc. are in the real world or not is a matter of philosophical debate, which is mostly not relevant here. What's out there:
there are objects in the real world we want to categorise
this includes associations of objects (e.g. of medical conditions: syndromes)
there are names we give to categories (concepts, classes …)
there are definitions that help us classify things correctly into these categories (concepts, classes …)
science advances, therefore we
create new categories
reclassify things
Regarding "Systemic Lupus Erythematosus" (SLE): according to the current diagnostic criteria, there are things (disease processes) that are members of this class (concept). Therefore I disagree that SLE is "only" a concept (like, e.g. unicorn). There is an
intension (all the properties that are required to classify a particular thing as an instance of SLE)
extension (all the individual things that are member of the class SLE )
This view is underlying the semantics of description logics. Both BFO and SNOMED CT are committed to description logics.
Assuming there is a new definition of SLE, sanctioned by some international authority, a possible way to deal with it would be to rename the "old" SLE to, e.g., SLE_2019 and the new one SLE_2020. If both are formally defined then you may (ideally) even classify new clinical cases into the old SLE concept and old ones into the new one, which could be good for epidemiological analyses.
Assuming that in 2022 SLE is subdivided into SLE_A, SLE_B, and SLE_C, all of which are subclasses of SLE. An SLE_A patient would still a SLE patient, but this wouldn't make SLE "only a concept".
Peter Hendler
Rheumatoid Arthritis is a good example. We don't know what it is. We have somewhat arbitrary clinical criteria that will certainly later be thrown away once we know the exact genetic and environmental mechanisms. Then we may even throw away the whole idea of Rheumatoid Arthritis. We may divide it into many sub diseases, or we may group it with another disease under a more broad heading.
Stefan Schulz
I do agree, this is in line with what I wrote before. But these arbitrary criteria can actually be expressed as full definitions. They only approximate the "real" disease (which probably represents a continuum from normal over subclinical to clinical and can only be diagnosed post mortem). But they are operational disease classes that do have members.
Peter Hendler
So the idea that SNOMED needs to fully define more terms, and the other idea that SNOMED should not be "concepts" but represent things in reality is already a contradiction.
Stefan Schulz
I don't see a contradiction here. There are – often temporary – disease concepts, which are created along very pragmatic criteria, but they do represent things in reality, otherwise they were useless.
Peter Hendler
Another problem in BFO is the disjointness between Continuants and Occurents. BFO would have one term for Asthma that is a Dependent Continuant for the propensity for Asthma Attacks. Then Asthma attacks would be Occurants. But for a real patient who has Asthma, they probably always also have some small component of the Occurant going on. Sometimes it's bad enough to go to the Emergency Room, or to take rescue inhalers, but there is no particular arbitarary severity when all of a sudden the Occurent is turned on and becomes present.
Stefan Schulz
According to what I wrote before, it's just both. Not as a logical conjunction (class overlap), which would violate the disjointness principle, but just two entities: There is both an instance of asthma propensity and an instance of asthma attack.
If you need to interpret just the term "Asthma" without any further information then you could just instantiate a class Asthma you define as equivalent to AsthmaPropensity OR AsthmaAttack.
SNOMED CT does not have the operator OR, but you could still do the following:
AsthmaAttack subclassOf Asthma
AsthmaPropensity subclassOf Asthma
Trying to reconcile ideas with the real world:
There are observations of real phenomena in real patients
There are results of biomedical research on real tissues, cells, populations etc.
There are scientific hypotheses, which are information objects, i.e. in BFO generically dependent continuants
Medical communities decide whether to consolidate such hypotheses or "ideas" into disease definitions and names.
These are often fully defined, not in terms of pathology and aetiology, but in terms of diagnostic criteria
The underlying pathological mechanism is often unclear, which entails that re-definitions occur
Confusingly, the name is kept but the underlying definition is changed
A standard like SNOMED CT should elucidate the relation between name (at a certain time, according to a certain community) and the related class definitions.
Daniel Karlsson
About single inheritance, what I remember from discussions with Barry and Werner during the Semantic Mining project (2004-8) was that they had objections to having multiple stated parents (multiple genera) (see Buidling ontologies with BFO, page 79). 75570004 |Viral pneumonia (disorder)| fulfills this criteria, although sometimes by necessity multiple stated primitive parents are needed to account for multiple aspects of meaning not defined through attributes.
Stefan Schulz
This is also what I remember. There is also often a confusion between an ontology and an OWL model. According to Barry and Werner, these are two separate things. An ontology must not have a class "unicorn" (because unicorns don't exist), but OWL models could, using classes like horn and horse, together with logical constructors.
Daniel Karlsson
About real world vs. thoughts (i.e. mind-dependent entities such as ideas, plans, suggestions, ...), first, thoughts are also part of the real world and not accepting that is being relativistic about ontology (Barry, the relativist!),
Stefan Schulz
Those things are compatible with BFO (see Werner Ceuster's Mental Functioning Ontology), but it seems they think (and misinterpret, partly due to unclear documentation and communication) that disorders are mind-dependent entities, which they are not: disorders themselves exist without anybody naming or classifying them, but the classification criteria, names and definitions for them are mind-dependent. But these are two completely different things.
Daniel Karlsson
Second, mind-dependent entities are clearly relevant in the health care domain, so at least a health care ontology should comprise both mind-dependent and mind-independent entities. Does SNOMED always clearly separate the two? Probably not, but that requires an analysis beyond the analysis of the definition of disease (...and "a disease is a concept" seems to have been removed from the editorial guide) in SNOMED.
Stefan Schulz
There is a tendency to ignore the difference between classes and meta-classes:
My headache is an instance of the SNOMED-concept Headache
The SNOMED-concept Headache is an instance of SNOMED-concept
From which we canNOT infer that my headache is an instance of SNOMED-concept
Daniel Karlsson
Further, the second-order (instances-to-types) relationship between mind-dependent and -independent makes a faithful representation in OWL (and EL in particular) a challenge to say the least (see Schulz et al An Ontological Analysis of Reference in Health Record Statements).
In general I think that research helping the community to find and address issues in SNOMED CT is a good thing. Smith, Ceusters, Rector (and Schulz, Bodenreider, Cornet, ...) etc. often have relevant criticism which we should address or at least discuss and relate to in our work. However, not all such research is of high quality, and there seem to be a category of research which reference those relevant papers, and then over- and mis-interpret the findings with misleading conclusions as outcome. One recent (counter-)example being El-Sappagh et al. BMC Medical Informatics and Decision Making (2018) 18:76. Apart from misusing references, the main idea of the paper, mapping SNOMED CT and OGMS, is a gross over-simplification avoiding the difficult problems SNOMED has to address.
https://www.ncbi.nlm.nih.gov/pubmed/30170591
Peter Hendler
Sounds like Jim Case and Kelly Kuru had a good experience in Buffalo. There is an informal verbal agreement about working together but Jim did stress the arguments against having two forms of each disease which would be a major turn off for clinicians. I'll let him tell what happened, but the result seems to be that Barry Smith, Peter Elkin and the BFO crew are now on friendlier terms with SNOMED. I think having Kelly there was a great thing too. Hopefully Jim will comment on this page.
who authored this...
What may contribute to misconceptions about SNOMED CT is its top class 138875005. Its FSN "SNOMED CT Concept (SNOMED RT+CTV3)" is a misnomer, because simple logics infers that, e.g., Queen Elizabeth is a SNOMED CT Concept, just as the aspirin tablet she took yesterday. For the use of SNOMED CT in practice, this may be hairsplitting, but not when communicating with philosophers or logicians.
> Not really, because “SNOMED CT Concept” is just the same as “OWL Thing” or “Top”.
> How does it contribute to misconceptions. It doesn’t imply at all that Queen Elizabeth
> and Aspirin are the same, it only says they are both “things in the world”.
This is an unrealistic desideratum. It has not ever been met by any OBO ontology. Regarding SNOMED CT, this desideratum is unrealistic, because large parts of SNOMED cover areas where
a simple logic is not sufficient to completely define most classes, such as in chemistry
(full definitions are outside of the scope of SNOMED Intl's standardization, e.g. organisms, human anatomy, physical objects, occupations, religions, lifestyles, geographic entities, or certain qualities.
Another reason is diseases like Lupus and Rheumatoid Arthritis. Medical science does not know if these clinically useful terms are “in the real world” at all. It may be found that they are each a group of more than one entity that look similar from the outside but are quite different. Alternatively they might both be different manifestations of one disease. The reality is probably a bit of both. They probably share some common mechanisms and differ in others, just as within each term there may be many differences that will later be sub divided. The reality is not black and white, and it does not lend itself to being fully defined.
3 SNOMED should be based on BFO and there should be Continuants and Occurrents and they should be mutually exclusive
Somewhere in this section we must indicate that clinicians, who already complain that SNOMED is too complex will never agree to separate Asthma the continuant from Asthma the occurrent. They would just look for an alternative terminology that did not force them to make this distinction. And in most cases of most diseases, both the occurrent and the continuant are always both present in the same subject to some degree. For example, in Asthma the continuant, if you were to do pulmonary function tests, you would find some small degree of the occurrent present. Distinguishing them is not only a burden, but would have practically no use for most clinicians in the way they use the terms.
Status | DRAFT |
|---|
Team
@Former user (Deleted)
@stefan.schulz (Unlicensed)
@Daniel Karlsson (Unlicensed)
@Yongsheng Gao
Relevant Documents
Copyright © 2025, SNOMED International