HL7 and negation

Thomas Beale thomas.beale at openehr.org
Sat Jun 11 06:09:23 EDT 2016

Let's state the problem as one of associating some sort of absence, 
presence, or in-between indicator with some X in a patient, or some 
other subject...

There is some discussion about this topic in SPECPR-118 
<https://openehr.atlassian.net/browse/SPECPR-118>, in which Ian McNicoll 
made the comment:

I think we need to do a bit more thinking about the pros and cons and 
the somewhat different requirements of e.g negating a diagnosis vs. 
negating a symptom.

This is exactly right - there are different kinds of Xs. There are 
/observable Xs/, including things we use instruments to observe (which 
we generally assume to be reliable in a practical sense). So 
presence/absence claims can reasonably be made in cases where the X 
being observed (e.g. pregnancy, MRSA, being a smoker) is the same thing 
the claim is made about (is pregnant, neg MRSA, non-smoker). In this 
case, the presence / absence can reasonably be said to be part of the 
reported reality of the X in question.

However, if the claim is about a C (some 'condition') where C /cannot be 
directly observed /(or is not, in the current situation) then we are 
looking at an /epistemic claim about knowledge of C/, based on observed 
X, what X means in the context of patient type P, and so on. There is a 
range of /epistemic claims /that could be made about Cs, e.g. the following:

  * doesn't exist - 100% sure C not present in patient - e.g. diabetes
    type I, based on negative oral glucose test
  * may exist - C is effectively one branch of a differential diagnosis
    or other assessment
  * does exist - 100% sure C present in patient - e.g. diabetes type I,
    based on +ve oral glucose test
  * no risk of C in future - 100% sure C will not occur, e.g. BRCA1 or 2
    breast cancer, based on genetic test (we assume the latter is
  * risk of C in future - some likelihood of C occurring
  * guarantee of C occurring in future - future reaction to exposure to
    bee venom in a patient known to be hyper-allergic to been venom

This basically boils down to:

  * it may be reasonable to allow presence or absence of true
    'observables' to be encoded in a binary way (what we think of as
    Observations in openEHR)
  * claims regarding any kind of assessment, opinion, diagnosis, etc of
    something we don't directly observe as such are epistemic claims,
    i.e. claims about type of knowledge we have of some C, and are not
    encodable as a Boolean 'existence' idea, but only as a level of
    certainty or similar. (what we think of as Evaluations in openEHR)

To make things somewhat annoying, there is probably a grey area between 
the two. For example, 'pregnant' could arguably be regarded as a direct 
observation or an assessment. But I think for 95% of cases things are 

So my conclusion is that the way to record presence / absence of true 
observables could reasonably be done in a simple way, while any type of 
assessment has to be recorded in a way that a) allows some range of 
certainty, b) can include the temporal aspect (now, future etc) and c) 
can reflect the current state of the investigative process.

Another annoyance that may prevent simple modelling is that EHRs often 
include statements like 'is diabetic', e.g. reported by an obviously 
diabetic patient about her diagnosis from 20y ago. Such statements are 
not in themselves assessments, they are reports of the outcome of an 
earlier process. As such, it may be reasonable to report such things in 
a more or less binary way, e.g. is / is not diabetic.

I'm not a fan of negation or any other variety of presence, absence, 
risk of etc being part of terminology, at least not pre-coordinated with 
the ontological part (doing so is a total confusion about what the 
terminology expresses). A typology of negation / epistemic claims could 
potentially exist in some separate part of a terminology e.g. SNOMED, to 
be used to code information model property like 'epistemic_status', or 

Aside: apparently the FHIR approach to representing things like 'no 
known allergies' is to infer it by seeing if an allergies list is empty 
or not. That sounds like a bad idea to me. If 'no known allergies' is 
understood as a clinically meaningful statement made by e.g. a GP (based 
on reliable knowledge about the patient), checking for a list being 
empty in some EMR system isn't at all the same thing. All that latter 
does is establish that no allergies have been recorded on this 
particular system.

- thomas

On 08/06/2016 07:54, GF wrote:
> Dear Colleagues,
> HL7 is thinking about the problem of negation.
> http://wiki.hl7.org/index.php?title=Negation_Requirements
> The group discussing it created a document with negation use cases.
> My questions are:
> - Can you let us know your reaction to this list of use cases?
> And
> - How should ‘negation’ be handled the best in respect of semantic 
> interpretability?
> My personal opinions:
> - the boolean should not be used
> - try to translate the ‘negation’ problem into ‘presence and absence’. 
> A concept is or is not present, a numeric result is of is not present.
> - do not use pre- and post co-ordinated concepts using SNOMED but use 
> the SNOMED primitives.
> I’m curious to learn what your opinion is.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openehr.org/pipermail/openehr-clinical_lists.openehr.org/attachments/20160611/d9539cd1/attachment-0002.html>

More information about the openEHR-clinical mailing list