[Dcm] terminfo meeting room requirements/Planning for Phoenix

[Stef_Verlinden1 (Stef Verlinden)]

Dear all,

The way I see it DCM could be very useful is when it truly separates clinical domain knowledge from technical issues such as the modelling to a certain reference model and/or standard. Which doesn’t mean that these technical aspects shouldn’t be addressed as well. Only it should be a separate discussion.

From the health care professional (HCP) perspective it’s important to have an, as broad as possible (and independent of technique), agreement on:

• what is the relevant information per clinical domain that needs to be stored and shared between all HCP involved

• the cardinality for each information item that could be acquired

• the constraints for that information

• which coding/ value sets are to be used?

• If and if yes, which codes (Snomed, ICD, IPC, OID, etc.) are attached to each information item

In order to avoid “double entrance”(i.e. similar data is captured in more than one archetype/ (clinical) template/ etc.) guidelines should be present to determine how large/small each clinical domain should be. With regard to this issue, one should clearly separate the separate findings, from the protocols or methods used to interpret these findings (see my second question below).

Another important issue is whether these clinical domains could be used for decision support and/or to support workflow management. In order to do so it’s essential that free text entrance is avoided as much as possible.

This than could lead to a method and tools to verify, validate. asses the quality and/or perform quality control.

Separately agreement should be reached on (and supersets should be created which can be used for) harmonising the standards/ reference models, which could be used to model the ‘independent’ clinical domains.

First question: do others share this view?

My second point is about the optimal ‘size of a clinical domain. My feeling is that we tend to create clinical domains that are ‘too large’ which can lead to more or less similar information that is stored in two different archetypes/(clinical)templates/ DCM’s.

Let me try to explain this with an example.

While looking at the ZIM website (which is unfortunately only in Dutch), the following struck me. There are 2 separate ZIM’s which are used to record more or less similar information. One is the Barthel index the other one the ADL (activities of daily life) index. (So far I only could find an observation.barthel archetype and here one can see the same as in the Barthel ZIM but since there seems to be no ADL AT the double recording issue is less clear).

The Barthel index ZIM or AT is used to register what the capabilities of a patient/client are. The ADL ZIM is used if one wants to register in more detail what these capabilities are.

For example in ADL incontinence can be registered (yes/no), in Barthel there are separate entrances for bladder and bowel, for which on a 3 point scale can be scored whether there is incontinency and how severe it is, if present.

Another example is dressing and undressing. In ADL undressing the upper and lower body is scored separately as well as dressing of upper and lower body, with an extra entrance for socks/ stockings, shoes and zippers, all on a 5-point scale.
In Barthel there is one item dressing/ undressing which is scored on a 3-point scale.
If one chooses to use these 2 ZIM’s side be side data about the same observation will be scored in 2 different locations and what’s worse: healthcare providers type A would use f.i. the ADL ZIM, because it suits their needs, and healthcare provider type B would use the Barthel ZIM and both wouldn’t know from each other that valuable information is already present in another location.

My point is that both the Barthel index as well as the ADL index are ‘protocols’ which can be used to assess a complex clinical situation and as such are too large to see as one clinical domain.

Ideally one should have one observation for urine incontinence, one for bowel/stool incontinence, and one for dressing and one for undressing. This separate observation should be defined in such a manner that they both can be used for a Barthel index and an ADL assessment, which in themselves would be (clinical) templates. Only then all relevant aspects for each specific observation are present at one location where they can be re-used and shared independent of the ‘protocol (ADL of Barthel in this case), that one choose to interpret/ assess the situation.

So my second question is, are there published methods/guidelines to determine how to deal with this issue?

Just out of curiosity a last question about the Barthel ZIM. For each separate ‘observation’ there is a well-defined 5-point scale to score that observation. Still, as an extra option, it is possible to score that observation as free text as well. Is there a specific reason to choose for this ‘dual strategy’?

Cheers,

Stef

[system]

Hi Stef

I believe that your response is universal on first confronting this issue - we need to get things modelled in a generic environment. The problem is that to model things usefully there has to be a high level of coherence between the models. The same thing has to be represented in the same way. Further, a lot of things need to be modelled over and over to get it right - such as how do we represent the average of 3 readings, the maximum, the change in a value etc. Otherwise we end up with thousands and millions of archetypes. If these things are not done the same way each time they are modelled then we will find we are in a mess very quickly.

The openEHR reference model is designed to model the things that are not of concern to clinicians but are essential for processing. The section does not alter the meaning of an entry, the source of the information, who is the information about and in what way are they related to the subject of the record, who committed the record and when, was it part of this record at the time or at another location, when was it visible in this record etc, etc.

We find pretty quickly that the models get about as useful as word documents - they have lots of text about them, the boundaries of the modelling are not understood clearly and we have a problem. More over, it is a problem that has been experienced by health authorities around the world - especially those preparing for or using HL7 v3 - how to get these models written? No, how to get these models formally expressed.

UML is a modelling language, but it does not solve the problems any more than paper does in this space.

My point is, you have to commit to a reference model in order to make the clinical modelling work. openEHR has been developed over 15 years of experience to get to the point where the models can be expressed AND the software can be implemented independently. A generic approach cannot deliver this. Further, if the environment is sufficiently expressive, you should be able to express the combination of the constraints in the openEHR RM and the archetypes and templates in a single statement in that formalism.

Just want to head off what has taken the DCM group a long time to discover - generic clinical modelling cannot be done except as a feeder to openEHR or HL7 - you have to have a formal outcome and I argue that the word document is as useful.

It is not so intuitive but it is correct. If people are not convinced, think about why it has not been done to this point, and look at the outcomes of efforts to cope with the complexity of health information - they both have reference models as starting points.

Cheers, Sam

Stef Verlinden wrote:

(attachments)

[williamtfgoossen]

Stef, Sam,

I just heard that the DCM proposal was accepted by HL7 steering devision. This means we can now formally work on this.

I followed the thread a little and want to pinpoint to one particular misunderstanding in your comments.

I copied in the particular line:

In a message dated 6-6-2008 16:40:10 W. Europe Daylight Time,
stef verlinden in copied message by sam heard writes:

Ideally one should have one observation for urine incontinence, one for bowel/stool incontinence, and one for dressing and one for undressing. This separate observation should be defined in such a manner that they both can be used for a Barthel index and an ADL assessment, which in themselves would be (clinical) templates. Only then all relevant aspects for each specific observation are present at one location where they can be re-used and shared independent of the ‘protocol (ADL of Barthel in this case), that one choose to interpret/ assess the situation.

Stef, this would do for a generic observation. However, the value sets applied render it different observations. Perhaps some could be done via defining different options for Likert Scales. Now we have the harmonised ISO datatypes almost worldwide available, the key CO (Coded Ordinal) especially for likert scales is available with the explicit option to calculate with it. So a starting point.
However, the reference to Barthel score items and comparing them to a single observations is not comparing apples with pears, which still could be a nice fruit salad, but it is comparing nails with silk. No relevance to each other, and silk could be destroyed by a nail so to speak.
The simple reason is that the Barthel index individual item has only meaning in the whole context of the 10 other items, belonging together and allowing the explicit summation of the mandatory ten scores (not nine, nor six, that would be meaningless). The simple reason for this is that the Barthel index (and the thousands of other scales, indexes of measurement instruments) are tested for validity and reliability in particular populations under specific circumstances, with the specified wording of total, variables and valuesets.

I have had a student analysing the 4 different scores for pressure ulcer risk in order to go the path you describe ( Helleman J, Goossen WT. Modeling nursing care in health level 7 reference information model. Comput Inform Nurs. 2003 Jan-Feb;21(1):37-45.). (With hindsight the titel should have been detailed clinical modelling nursing care for use in HL7 messaging).

We found that it could be doable on the observation level (similar as you suggest), but that it is impossible to go the value set harmonization path.
Therefore I believe we can do what you suggest for a single observation of e.g. urine incontinence type stress incontinence, or type urge incontinence, or functional incontinence etc.
But we cannot do as you suggest for a ‘normal’ single observation of clothing as one part of the clinical template for ADL (activities of daily living, not archetype def. language).
But we cannot include the Barthel, because due to the strict meaning, purpose and context it is completely different and MUST be stored separately.
these two are not comparable entities! Also I believe the Barthel should mandatory be the smalles atomic archetype possible (but still some complexity in it), where the ADL observations can be less complex archetypes that allow the many reuses as you suggest and constrain them on clinical template level.

I also would like to point you to this publication, that handles the same issue on terminologicial side to include scales into LOINC.

White TM, Hauan MJ.

Extending the LOINC conceptual schema to support standardized assessment instruments.
J Am Med Inform Assoc. 2002 Nov-Dec;9(6):586-99.

Hope to have clarified the difference.

Sincerely yours,

dr. William TF Goossen
director
Results 4 Care b.v
De Stinse 15
3823 VM Amersfoort
email: Results4Care@cs.com
phone + 31654614458
fax +3133 2570169
www.results4care.nl
Dutch Chamber of Commerce number: 32133713

[system]

Hi,

Ordinals are peculiar things.
I agree that (perhaps) there are ordinals at the data types level.
But ordinals at the archetype (concept) level are different kinds.
(‘Fikkie is a dog but not all Fikkies are dogs’, as a Dutch proverb goes)

Ordinals at the concept level act as modifiers to a noun or verb.
A severe headache. A mild hypertension. No digestion. High level of exertion.
Ordinals at the concept level are preferably a code from an external coding system.
Defined for a specific ever changing context.

Ordinals at the data type level are mostly unneeded when we accept my view.

Gerard Freriks

Now we have the harmonised ISO datatypes almost worldwide available, the key CO (Coded Ordinal) especially for likert scales is available with the explicit option to calculate with it. So a starting point.

– –
Gerard Freriks, MD
Huigsloterdijk 378
2158 LR Buitenkaag
The Netherlands

T: +31 252544896
M: +31 620347088
E: gfrer@luna.nl

Those who would give up essential Liberty, to purchase a little temporary
Safety, deserve neither Liberty nor Safety. Benjamin Franklin 11 Nov 1755

[thomas.beale]

with apologies for daring to tread the waters of the medical
professional .....

Williamtfgoossen@cs.com wrote:

...
However, the reference to Barthel score items and comparing them to a
single observations is not comparing apples with pears, which still
could be a nice fruit salad, but it is comparing nails with silk. No
relevance to each other, and silk could be destroyed by a nail so to
speak.
The simple reason is that the Barthel index individual item has only
meaning in the whole context of the 10 other items, belonging together
and allowing the explicit summation of the mandatory ten scores (not
nine, nor six, that would be meaningless). The simple reason for this
is that the Barthel index (and the thousands of other scales, indexes
of measurement instruments) are tested for validity and reliability in
particular populations under specific circumstances, with the
specified wording of total, variables and valuesets.

I believe this is correct as well - there are many such evidence-based
scores that I think we have to consider as 'pre-fabricated' content
molecules in our system of content definition. If we accept this, then
we have a modelling environment that could be thought of as follows:

- all archetypes
    + archetypes that are molecules of atoms constructed by archetype
modellers as a way of standardising some aspect of medical data recording
    + archetypes that wrap pre-fabricated molecules already standardised
by medicine

There may be a score for urinary incontinence in an ADL score 'molecule'
and also a separate observation (described by an archetype of the first
kind) that describes urinary incontinence in some detail. Both of these
possibilities should clearly exist. The interesting design problem that
this throws up is how querying should react.

- thomas beale

[Stef_Verlinden1 (Stef Verlinden)]

Stef, Sam,

I just heard that the DCM proposal was accepted by HL7 steering devision. This means we can now formally work on this.

Congratulations

I followed the thread a little and want to pinpoint to one particular misunderstanding in your comments.

I copied in the particular line:

In a message dated 6-6-2008 16:40:10 W. Europe Daylight Time,
stef verlinden in copied message by sam heard writes:

Ideally one should have one observation for urine incontinence, one for bowel/stool incontinence, and one for dressing and one for undressing. This separate observation should be defined in such a manner that they both can be used for a Barthel index and an ADL assessment, which in themselves would be (clinical) templates. Only then all relevant aspects for each specific observation are present at one location where they can be re-used and shared independent of the ‘protocol (ADL of Barthel in this case), that one choose to interpret/ assess the situation.

Stef, this would do for a generic observation. However, the value sets applied render it different observations. Perhaps some could be done via defining different options for Likert Scales. Now we have the harmonised ISO datatypes almost worldwide available, the key CO (Coded Ordinal) especially for likert scales is available with the explicit option to calculate with it. So a starting point.
However, the reference to Barthel score items and comparing them to a single observations is not comparing apples with pears, which still could be a nice fruit salad, but it is comparing nails with silk. No relevance to each other, and silk could be destroyed by a nail so to speak.
The simple reason is that the Barthel index individual item has only meaning in the whole context of the 10 other items, belonging together and allowing the explicit summation of the mandatory ten scores (not nine, nor six, that would be meaningless). The simple reason for this is that the Barthel index (and the thousands of other scales, indexes of measurement instruments) are tested for validity and reliability in particular populations under specific circumstances, with the specified wording of total, variables and valuesets.

I have had a student analysing the 4 different scores for pressure ulcer risk in order to go the path you describe ( Helleman J, Goossen WT. Modeling nursing care in health level 7 reference information model. Comput Inform Nurs. 2003 Jan-Feb;21(1):37-45.). (With hindsight the titel should have been detailed clinical modelling nursing care for use in HL7 messaging).

Thanks, unfortunately I don’t have access to the whole article. Is it possible to send a copy of it? That would be great.

I think I see your point partially but I’m still struggling with it. From my point of view there is still something missing.

Let’s try if I can explain it in another way: The underlying observations (in this example for both Barthel and ADL) are generic: for example the (in)ability to control urine excretion. In order to score if there is, and if present what the severity of urine incontinence is, one should create a dedicated urine excretion (observation) archetype/ DCM/ etc. This one would probably have a lot of extra parameters that are not necessary for the Barthel of ADL assesment.

Scoring a Barthel or an ADL index is actually an interpretation of those specific observations and attach those interpretations to different scales. This is where I agree partially with you. The difference is that from your viewpoint the whole Barthel score is one observation as such and therefore all underlying observations should be stored as one and separately form other observations. From my viewpoint those are different interpretations of (more or less) the same set of observations. For example, the Barthel looks at both urine and faecal incontinence separately. In the ADL it is scored whether there is either urine or faecal incontinence. Nevertheless the underlying ‘observations’ are identical and that why I think those should only be registered once in a specified archetype/ DCM per generic observation.

From this line of reason (if I understand it correctly) the Barthel index en the ADL index should be separate archetypes, where the underlying generic observation archetypes serve as source for the evaluation/interpretation of the situation leading to a (calculated) score. In that case both indexes can remain their own scales and their own rules to calculate the score.

This also provides the opportunity to solve another issue: the Barthel index provides an evaluation of the actual situation. According to which protocol? From literature I can find at least 3 versions of the Barthel index (the original from 1965, a modified version from 1985 and an extended version from 1979). It seems important to know which version is used for the evaluation. This is something you would like to add with the protocol section of your evaluation archetype. In that manner it would be very well possible that there will be 2 different Barthel archetypes both using a different protocol for the evaluation/ calculation of the Barthel score.

In this respect I’m also in doubt whether such an archetype should be an observation or an evaluation AT. At one hand it doesn’t seem to fit the definition of an evaluation AT on the other hand it seems to me that it is the evaluation of a multifactorial situation.

How do others see this?

[Stef_Verlinden1 (Stef Verlinden)]

with apologies for daring to tread the waters of the medical
professional .....

Well the other way around, for one or another reason I seem to get
caught up in discussions that are far to technical for me:-)
I guess we do need both sides of the medal....

...
However, the reference to Barthel score items and comparing them to a
single observations is not comparing apples with pears, which still
could be a nice fruit salad, but it is comparing nails with silk. No
relevance to each other, and silk could be destroyed by a nail so to
speak.
The simple reason is that the Barthel index individual item has only
meaning in the whole context of the 10 other items, belonging
together
and allowing the explicit summation of the mandatory ten scores (not
nine, nor six, that would be meaningless). The simple reason for this
is that the Barthel index (and the thousands of other scales, indexes
of measurement instruments) are tested for validity and
reliability in
particular populations under specific circumstances, with the
specified wording of total, variables and valuesets.

I believe this is correct as well - there are many such evidence-based
scores that I think we have to consider as 'pre-fabricated' content
molecules in our system of content definition. If we accept this, then
we have a modelling environment that could be thought of as follows:

- all archetypes
    + archetypes that are molecules of atoms constructed by archetype
modellers as a way of standardising some aspect of medical data
recording
    + archetypes that wrap pre-fabricated molecules already
standardised
by medicine

I'm not sure but I have the feeling that we're referring to the same
principle (see my previous response to Williams answer). Is that
correct?

There may be a score for urinary incontinence in an ADL score
'molecule'
and also a separate observation (described by an archetype of the
first
kind) that describes urinary incontinence in some detail. Both of
these
possibilities should clearly exist. The interesting design problem
that
this throws up is how querying should react.

The interesting question is how should we solve that?

Cheers,

Stef

[Stef_Verlinden1 (Stef Verlinden)]

Since this response bounced because it was too large, I resend a
shorter version

Hi Stef

Hi Sam,

Thanks for your response.

My point is, you have to commit to a reference model in order to
make the clinical modelling work. openEHR has been developed over
15 years of experience to get to the point where the models can be
expressed AND the software can be implemented independently. A
generic approach cannot deliver this. Further, if the environment
is sufficiently expressive, you should be able to express the
combination of the constraints in the openEHR RM and the archetypes
and templates in a single statement in that formalism.

Just want to head off what has taken the DCM group a long time to
discover - generic clinical modelling cannot be done except as a
feeder to openEHR or HL7 -

That was the idea behind the initial mail. To discuss if it's useful
to have an generic clinical modelling which is 'independent' of the
underlying reference model (openEHR or HL7) which stays away (if
possible) from the technical discussions especially wrt the
harmonization of those 2 reference models.

harmonized reference models and widely accepted clinical models
(archetypes/ templates/ DCM's) and these should be available as soon
as possible. By separating those two discussions we can work in
parallel and gain a lot of time. Now it seems (to me:-)) that solving
the technical issues has a higher priority (which is of course
essential) but in the meantime al lot of technique independent
clinical modelling issues (F.I. is the Barthel index a molecule build
form atoms or is it an atom in itself) can be discussed and agreed upon.

Since these agreements should have nothing to do with the RM used in
the end I wondered if this DCM platform could be a place to share out
knowlegde and experience and boost the harmonization in clinical
modelling. Also there is a lot of work to be done and the sooner we
can create clinical models that are widely accepted the better.

Cheers,

Stef

[system]

-1-
There is the question.
Many clinical models in the Dutch ZIM’s and the topics of the DCM’s, are they Evaluations based on observations or Observations?
It can be argued that they are Observations that are transformed via an algorithm into something else that is an abstraction of reality: the Evaluation.
Evaluations are the same, almost. But in this case there is no algorithm, but only the experience and knowledge in the mind of the documenter.

-2-Without any doubt we get archetypes that document the same concept but are different internally.
Because other codes are used, other ways of measurements and units, other ways to express (quantitatively, semi-quantitatively, qualitatively), etc, etc.
But all document the same concept.
This situation calls for a type of ‘Archetype Ontology’ so we can document that these archetypes documenting the same concept are ‘synonyms’.
And that when querying we know what archetypes to use in order to collect all documented information about a clinical topic.

Gerard

• all archetypes

• archetypes that are molecules of atoms constructed by archetype
  modellers as a way of standardising some aspect of medical data recording
• archetypes that wrap pre-fabricated molecules already standardised
  by medicine

There may be a score for urinary incontinence in an ADL score ‘molecule’
and also a separate observation (described by an archetype of the first
kind) that describes urinary incontinence in some detail. Both of these
possibilities should clearly exist. The interesting design problem that
this throws up is how querying should react.

– –
Gerard Freriks, MD
Huigsloterdijk 378
2158 LR Buitenkaag
The Netherlands

T: +31 252544896
M: +31 620347088
E: gfrer@luna.nl

Those who would give up essential Liberty, to purchase a little temporary
Safety, deserve neither Liberty nor Safety. Benjamin Franklin 11 Nov 1755

[thomas.beale]

Gerard Freriks wrote:

Hi,

Ordinals are peculiar things.
I agree that (perhaps) there are ordinals at the data types level.
But ordinals at the archetype (concept) level are different kinds.
('Fikkie is a dog but not all Fikkies are dogs', as a Dutch proverb goes)

Ordinals at the concept level act as modifiers to a noun or verb.
A severe headache. A mild hypertension. No digestion. High level of
exertion.
Ordinals at the concept level are preferably a code from an external
coding system.
Defined for a specific ever changing context.
Ordinals at the data type level are mostly unneeded when we accept my
view.

*Gerard,

what would we do with urinalysis ordinals then: trace, +, ++, +++, etc?

- thomas beale

[williamtfgoossen]

In a message dated 7-6-2008 7:22:07 W. Europe Daylight Time, gfrer@luna.nl writes:

Ordinals at the data type level are mostly unneeded when we accept my view.

Gerard,

here you sure have a point: we would not need them in the ideal world.

however, medical and psychological and nursing scientist in health care have been making such scales for decades. They are just there as part of our domain reality and therefore we need to handle them as intended.

William

Sincerely yours,

dr. William TF Goossen
director
Results 4 Care b.v.
De Stinse 15
3823 VM Amersfoort
email: Results4Care@cs.com
phone + 31654614458
fax +3133 2570169
www.results4care.nl
Dutch Chamber of Commerce number: 32133713

[system]

Thomas,

There is an attribute in Observations where we can ‘store’ the result.
Attached to it there is a Modifier Attribute that indicates whether it is a Quantitative result or a Semi-Quantitative or Qualitative result.
Semi-Quantitative is a link to an accepted published code list with (trace, +, ++, +++, ++++) As indicated by the urine analysis strip.
Quantitative indicates the present or not-present state.

Using this way of thinking we are able to express:

• Urine stick measurement xyz: Yes or No
• Urine stick measurement xyz: (trace, +, ++, +++, ++++)
• Urine stick measurement xyz: 7.4 mmol

In an Evaluation we could express:

• It is my opinion that: Urine stick measurement xyz: is (low, moderate, high) Low is defined as …, moderate is defined as …, high is defined as …

Problem to solve:
How can we standardize these Modifier Code Lists
Is SNOMED enough?

what would we do with urinalysis ordinals then: trace, +, ++, +++, etc?

• thomas beale

– –
Gerard Freriks, MD
Huigsloterdijk 378
2158 LR Buitenkaag
The Netherlands

T: +31 252544896
M: +31 620347088
E: gfrer@luna.nl

Those who would give up essential Liberty, to purchase a little temporary
Safety, deserve neither Liberty nor Safety. Benjamin Franklin 11 Nov 1755

[thomas.beale]

Gerard Freriks wrote:

Thomas,

There is an attribute in Observations where we can 'store' the result.
Attached to it there is a Modifier Attribute that indicates whether it
is a Quantitative result or a Semi-Quantitative or Qualitative result.

I am not sure what attribute you are talking about here - do you mean
somehting in the OBSERVATION, HISTORY, or EVENT classes? Or do you mean
DV_QUANTIFIED.magnitude_status?

Semi-Quantitative is a link to an accepted published code list with
(trace, +, ++, +++, ++++) As indicated by the urine analysis strip.
Quantitative indicates the present or not-present state.

Using this way of thinking we are able to express:
- Urine stick measurement xyz: Yes or No
- Urine stick measurement xyz: (trace, +, ++, +++, ++++)
- Urine stick measurement xyz: 7.4 mmol

In an Evaluation we could express:
- It is my opinion that: Urine stick measurement xyz: is (low,
moderate, high) Low is defined as ..., moderate is defined as ....,
high is defined as ...

we could but this isn't a clinical opinion in the normal sense of the
term - it is a 'cognitive' opinion, like trying to say what small
writing says in an eye test. Any such measurement should be objectively
reportable, regardless of whether the values come in bands or not....

[heather.leslie]

Hi Stef,

I’ve been travelling. My belated comments inline…

Cheers

Heather

[system]

Dear all,

Some detailed thoughts.

Reading recent e-mails about Terminfo, DCM and knowing the discussion topic ‘severity’ here my mental ruminations.

• Always there will be various ways to express the same meaning.
  How much one tries to limit this there will be good and bad reasons to have these separate forms of expression.
  The only solution is a management Tool to deal with it.

When making queries via the tool ‘synonyms’ (archetypes with different names and internal models, but expressing the same topic) must be found.
This implies an ontology about Archetypes so we can record and find those synonyms and include them in queries.

This is a way to deal with the ‘grey zone’ between ways to express things as code or structure.

• When we try to express things there are variable levels of abstraction/aggregation we use.
  E.g.
  Observation: Ability to put on socks: No/Yes
  Observation: Ability to put on socks: +, ++, +++, ++++, +++++
  Observation: Ability to put on socks: 30 seconds
  Observation: ability to put on left sock: 30 seconds; right sock: 5 minutes

Observation: Systolic pressure: No, Yes
Observation: Systolic pressure: 0, +, ++, +++, ++++, +++++
Observation: Systolic pressure: no, Low, Normal, High, very high
Observation: Systolic pressure: 120
Observation: Systolic pressure: 120 ± 1
Observation: Systolic pressure: 119.9 ± 0.3

All (most) observations can be transformed from very precise to generic/aggregated.
The are all about the same topic (ability or measurement)

Are all examples of the same nature?

• A measurement by a device is it the same as an observation by a human?

• A liver that is palpable by a human is this an observation or an subjective interpretation and therefor an Evaluation?
  So far any documentation about the patient system as a whole is an observation. I think this is correct.

• But ‘Systolic: High’ isn’t this an Evaluation with implicit or explicit classification criteria?
  The Evaluation is an expression of a professional opinion.
  To document Blood pressure as low or high is this no longer an observation but an Evaluation or both?

• When we document Systolic pressure: +++,
  don’t we need to have in the cluster archetype (pattern) a facility to express what we mean by +++ or high in order to interpret correctly the information?
  In other words is an ordinal at the archetype level not a pattern in itself?
  A pattern that defines a semi-quantitative observation or evaluation?

• When there are patterns for qualitative, semi-quantitatve and quantitative observations and evaluations,
  does this mean that these cluster archetypes as patterns will be expressed no longer as ordinal data types but a collection of quantities and codes?

Gerard

[Heather Leslie] Additionally, and very importantly, once these archetypes
are in the archetype repository, then searching breathing should display
both the Apgar plus the ‘breathing-related’ archetypes; and searching for
mobility should display both the generic mobility archetype plus the Barthel
– then the user can use which archetypes suit their purpose, or query
appropriately. The Repository will play an important role here due to the
ontology basis behind it.

– –
Gerard Freriks, MD
Huigsloterdijk 378
2158 LR Buitenkaag
The Netherlands

T: +31 252544896
M: +31 620347088
E: gfrer@luna.nl

Those who would give up essential Liberty, to purchase a little temporary
Safety, deserve neither Liberty nor Safety. Benjamin Franklin 11 Nov 1755

[thomas.beale]

Gerard Freriks wrote:

Observation: Systolic pressure: No, Yes
Observation: Systolic pressure: 0, +, ++, +++, ++++, +++++

I want to meet the GP who measures BP like this

Are all examples of the same nature?

- A measurement by a device is it the same as an observation by a human?

yes: just a question of the statistical accuracy band.

- A liver that is palpable by a human is this an observation or an
subjective interpretation and therefor an Evaluation?
So far any documentation about the patient system as a whole is an
observation. I think this is correct.

observation: it is a procedure, taught to professionals to be repeatable
and reliable, without which they would need some other observational
method to establish problems in the liver. The fact that it is in use
shows that it is an adequate and useful observational method, and
generates results that are statistically acceptable.

- But 'Systolic: High' isn't this an Evaluation with implicit or
explicit classification criteria?
The Evaluation is an expression of a professional opinion.
To document Blood pressure as low or high is this no longer an
observation but an Evaluation or both?

Here we get into linguistic challenges. If the understanding of 'high'
is derived from the following:

low: < 90 mmHg
normal: 90mmHg - 140mmHg
high: > 140 mmHg

then assuming there is an instrument available to generate the value
'high' it is an observation (at a very coarse precision), since it is a
repeatable and objective. If the 'instrument' is in fact a human being
reading from a normal digital or analog device, then it may questionable
as to whether the more precise measurement should not be recorded.

If however, the word 'high' is being used by the clinician to mean 'this
patient is hypertensive', i.e. the BP is too high, then it is an
evaluation. And that's why words like 'hypertensive' exist. In this
second case, the band of values that correspond to 'high' may vary with
the patient, e.g. sex, weight, age, pregnancy, diabetic, or just plain
personal variation. So the use of the word 'high' in this circumstance
indicates an assessment by the clinician that the observed blood
pressure is _higher than normal_, i.e. _too high_ for the given patient.

- When we document Systolic pressure: +++,
don't we need to have in the cluster archetype (pattern) a facility to
express what we mean by +++ or high in order to interpret correctly
the information?
In other words is an ordinal at the archetype level not a pattern in
itself?
A pattern that defines a semi-quantitative observation or evaluation?

I think we only need to do such things if a physician is realistically
going to use them. I would have thought that 'high' would never be used
as a primary observation, but only in the evaluation sense, and I did
not know that +++ was used at all for BP.

- When there are patterns for qualitative, semi-quantitatve and
quantitative observations and evaluations,
does this mean that these cluster archetypes as patterns will be
expressed no longer as ordinal data types but a collection of
quantities and codes?

well, if we want to use 'high' according to the first sense I describe
above, and +++ likewise, then we would use ordinals, because the latter
supply a means of defining the bands of terms like 'low', 'normal',
'high' etc, but if you want to use them as diagnostic terms, then they
need to be coded and bound to things like Snomed::hypertension etc.

We have to remain ever-vigilant to the intricacies of natural language!

- thomas beale

[Tim_Cook2 (Tim Cook)]

What GREAT reply. I have been thinking all day about how to properly,
respectfully reply to Gerard's email.

I only wish I could have assembled the thoughts and put them into
writing the way you did.

Cheers,
Tim

[Karsten_Hilbert]

Gerard Freriks wrote:
>
> Observation: Systolic pressure: No, Yes
> Observation: Systolic pressure: 0, +, ++, +++, ++++, +++++
I want to meet the GP who measures BP like this

Not sure about the +, ++ but No/Yes may well occur in a
reanimation situation.

Karsten

[Andrew_Patterson]

> Observation: Systolic pressure: No, Yes
> Observation: Systolic pressure: 0, +, ++, +++, ++++, +++++
I want to meet the GP who measures BP like this

Not sure about the +, ++ but No/Yes may well occur in a
reanimation situation.

They can do this now? I, for one, welcome our new zombie overlords..

Andrew

[Stef_Verlinden1 (Stef Verlinden)]

Hi Heather,

Thanks for your extensive reply, this is very helpful. Just some remarks in between your reply.

[Heather Leslie] I both agree and disagree with you, Stef!!!

[Heather Leslie] I always try to model each archetype as a discrete, single

clinical content specification – designed as a maximal data set for

universal use case. This is what I would describe as ‘PURE’ modelling, just

as you describe.

However there are definitely situations where there are groups of concepts

that are only meaningful when grouped together – such as Barthel score.

Apgar (breathing, heart rate, color etc as an assessment of an infant at

birth) and SF-36 questionnaires are other examples which are in common usage

but only useful and valid when cohesive and used as a whole.

I guess the underlying question is, is a Barthel index score always an observation as such, or can it also be constructed from other (smaller/ atomar) observations.
If for a Barthel index score means that a health care providers goes to see the patient and scores for all items on the list and that observation is only used for that purpose, I agree it could be a separate AT.

If however a Barthel index could also be constructed from (previously) recorded (smaller/atom like) observations I would stay with my previous suggestion that a barthel index should be a ‘superarchetype’ (a molecule as Gerard nicely put it) which can be constructed from ‘atoms’ (see also the reply to William’s remarks)

[Heather Leslie] From my point of view the clinical descriptions of

‘Mobility’ as defined in Barthel are only a very limited subset when

describing mobility as a maximal data set archetype. We could try to be

inclusive and have these attributes in part of the ‘pure’ mobility

archetype, but I don’t think that this fragment about ‘mobility’ is

particularly useful, except as part of the Bartel context. I would describe

mobility quite differently in a ‘purely designed’ maximal archetype.

I agree that mobility as defined in Barthel is a very limited subset. Even more I would argue that the mobility score in the Barthel index is an interpretation of the actual observation of the ability to be mobile, which is captured in a ‘pure’ mobility AT.

In Barthel you have a 5 point scale. For example a score ‘2’ means ‘(needs) little help (to be mobile)’. This is not a very objective observation. Little help: According to whose/ which standard? So one would like to have be able to record the underlying ‘objective’ observations. These would be captured in a purely designed maximal archetype.

I’m aware that this is very detailed but to give you an example. Here in the Netherlands the Barthel score is used to asses how much care one gets and this translates into money. So people could have an interest in ‘lowering’ the score (those who receive the care) or to make the score ‘bettter’ (those who have to pay for it). Only if (more or less) objective criteria can be used to generate this score a ‘fair’ assesment can be provided.

[Heather Leslie] These scores or indexes are well used as screening or

assessment scores and, as William has indicated, some such as Barthel have

been validated only when used as a whole. So my opinion is that there is

good reason for simple and ‘PRAGMATIC’ modelling of these scores as a whole,

BUT I would also like the generic components each modeled ‘purely’ as well

ie a maximal and generic ‘mobility’ archetype modeled as well plus the

concept of breathing, as found in Apgar, modeled using all its component

concepts as separate archetype components ie breathing = respiratory rate,

chest expansion, inspection of trachea/chest, auscultation, etc etc.

[Heather Leslie] I generally push back if there is a real or perceived

overlap in content – but I think that in this situation the overlap is

minimal and the intent quite different.

[Heather Leslie] Additionally, and very importantly, once these archetypes

are in the archetype repository, then searching breathing should display

both the Apgar plus the ‘breathing-related’ archetypes; and searching for

mobility should display both the generic mobility archetype plus the Barthel

– then the user can use which archetypes suit their purpose, or query

appropriately. The Repository will play an important role here due to the

ontology basis behind it.

I’m not sure if I agree with that, because that means every time one wants to create a query one have to do extensive research in the AT repository to find out of for every search item an alternative entry exists. Theoretically and technically that will work, but since it involves extra work/ somebody who loves to search for those alternatives, practically it could mean that data that is present in another location is overlooked. Although this probably can’t be prevented entirely one should try to reduce this overlap as much as possible.

Cheers,

Stef